]>
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" | |
064af421 BP |
33 | #include "coverage.h" |
34 | #include "dirs.h" | |
35 | #include "dpif.h" | |
36 | #include "dynamic-string.h" | |
37 | #include "flow.h" | |
38 | #include "hash.h" | |
39 | #include "list.h" | |
40 | #include "mac-learning.h" | |
41 | #include "netdev.h" | |
42 | #include "odp-util.h" | |
43 | #include "ofp-print.h" | |
44 | #include "ofpbuf.h" | |
d65349ea | 45 | #include "ofproto/netflow.h" |
8cd4882f | 46 | #include "ofproto/ofproto.h" |
da285df4 | 47 | #include "packets.h" |
064af421 BP |
48 | #include "poll-loop.h" |
49 | #include "port-array.h" | |
50 | #include "proc-net-compat.h" | |
51 | #include "process.h" | |
76343538 | 52 | #include "sha1.h" |
6c88d577 | 53 | #include "shash.h" |
064af421 | 54 | #include "socket-util.h" |
064af421 BP |
55 | #include "svec.h" |
56 | #include "timeval.h" | |
57 | #include "util.h" | |
da285df4 | 58 | #include "unixctl.h" |
064af421 BP |
59 | #include "vconn.h" |
60 | #include "vconn-ssl.h" | |
b36682d8 | 61 | #include "vswitchd/vswitch-idl.h" |
064af421 BP |
62 | #include "xenserver.h" |
63 | #include "xtoxll.h" | |
64 | ||
65 | #define THIS_MODULE VLM_bridge | |
66 | #include "vlog.h" | |
67 | ||
68 | struct dst { | |
69 | uint16_t vlan; | |
70 | uint16_t dp_ifidx; | |
71 | }; | |
72 | ||
064af421 | 73 | struct iface { |
0c6aea3f | 74 | /* These members are always valid. */ |
064af421 BP |
75 | struct port *port; /* Containing port. */ |
76 | size_t port_ifidx; /* Index within containing port. */ | |
064af421 | 77 | char *name; /* Host network device name. */ |
064af421 | 78 | tag_type tag; /* Tag associated with this interface. */ |
064af421 | 79 | long long delay_expires; /* Time after which 'enabled' may change. */ |
064af421 | 80 | |
0c6aea3f BP |
81 | /* These members are valid only after bridge_reconfigure() causes them to |
82 | * be initialized.*/ | |
83 | int dp_ifidx; /* Index within kernel datapath. */ | |
84 | struct netdev *netdev; /* Network device. */ | |
064af421 | 85 | bool enabled; /* May be chosen for flows? */ |
76343538 BP |
86 | |
87 | /* This member is only valid *during* bridge_reconfigure(). */ | |
88 | const struct ovsrec_interface *cfg; | |
064af421 BP |
89 | }; |
90 | ||
91 | #define BOND_MASK 0xff | |
92 | struct bond_entry { | |
93 | int iface_idx; /* Index of assigned iface, or -1 if none. */ | |
94 | uint64_t tx_bytes; /* Count of bytes recently transmitted. */ | |
95 | tag_type iface_tag; /* Tag associated with iface_idx. */ | |
96 | }; | |
97 | ||
98 | #define MAX_MIRRORS 32 | |
99 | typedef uint32_t mirror_mask_t; | |
100 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
101 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
102 | struct mirror { | |
103 | struct bridge *bridge; | |
104 | size_t idx; | |
105 | char *name; | |
106 | ||
107 | /* Selection criteria. */ | |
108 | struct svec src_ports; | |
109 | struct svec dst_ports; | |
110 | int *vlans; | |
111 | size_t n_vlans; | |
112 | ||
113 | /* Output. */ | |
114 | struct port *out_port; | |
115 | int out_vlan; | |
116 | }; | |
117 | ||
118 | #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ | |
119 | struct port { | |
120 | struct bridge *bridge; | |
121 | size_t port_idx; | |
122 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
123 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. */ | |
124 | char *name; | |
125 | ||
126 | /* An ordinary bridge port has 1 interface. | |
127 | * A bridge port for bonding has at least 2 interfaces. */ | |
128 | struct iface **ifaces; | |
129 | size_t n_ifaces, allocated_ifaces; | |
130 | ||
131 | /* Bonding info. */ | |
132 | struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ | |
133 | int active_iface; /* Ifidx on which bcasts accepted, or -1. */ | |
134 | tag_type active_iface_tag; /* Tag for bcast flows. */ | |
135 | tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ | |
136 | int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ | |
85c74638 | 137 | bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ |
76343538 | 138 | bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ |
064af421 BP |
139 | |
140 | /* Port mirroring info. */ | |
141 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
142 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
143 | bool is_mirror_output_port; /* Does port mirroring send frames here? */ | |
76343538 BP |
144 | |
145 | /* This member is only valid *during* bridge_reconfigure(). */ | |
146 | const struct ovsrec_port *cfg; | |
064af421 BP |
147 | }; |
148 | ||
149 | #define DP_MAX_PORTS 255 | |
150 | struct bridge { | |
151 | struct list node; /* Node in global list of bridges. */ | |
152 | char *name; /* User-specified arbitrary name. */ | |
93dfc067 | 153 | struct mac_learning *ml; /* MAC learning table. */ |
064af421 BP |
154 | bool sent_config_request; /* Successfully sent config request? */ |
155 | uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ | |
156 | ||
157 | /* Support for remote controllers. */ | |
158 | char *controller; /* NULL if there is no remote controller; | |
159 | * "discover" to do controller discovery; | |
160 | * otherwise a vconn name. */ | |
161 | ||
162 | /* OpenFlow switch processing. */ | |
163 | struct ofproto *ofproto; /* OpenFlow switch. */ | |
164 | ||
165 | /* Kernel datapath information. */ | |
c228a364 | 166 | struct dpif *dpif; /* Datapath. */ |
064af421 BP |
167 | struct port_array ifaces; /* Indexed by kernel datapath port number. */ |
168 | ||
169 | /* Bridge ports. */ | |
170 | struct port **ports; | |
171 | size_t n_ports, allocated_ports; | |
172 | ||
173 | /* Bonding. */ | |
174 | bool has_bonded_ports; | |
175 | long long int bond_next_rebalance; | |
176 | ||
177 | /* Flow tracking. */ | |
178 | bool flush; | |
179 | ||
180 | /* Flow statistics gathering. */ | |
181 | time_t next_stats_request; | |
182 | ||
183 | /* Port mirroring. */ | |
184 | struct mirror *mirrors[MAX_MIRRORS]; | |
76343538 BP |
185 | |
186 | /* This member is only valid *during* bridge_reconfigure(). */ | |
187 | const struct ovsrec_bridge *cfg; | |
064af421 BP |
188 | }; |
189 | ||
190 | /* List of all bridges. */ | |
191 | static struct list all_bridges = LIST_INITIALIZER(&all_bridges); | |
192 | ||
193 | /* Maximum number of datapaths. */ | |
194 | enum { DP_MAX = 256 }; | |
195 | ||
196 | static struct bridge *bridge_create(const char *name); | |
197 | static void bridge_destroy(struct bridge *); | |
198 | static struct bridge *bridge_lookup(const char *name); | |
8ca79daa | 199 | static unixctl_cb_func bridge_unixctl_dump_flows; |
064af421 | 200 | static int bridge_run_one(struct bridge *); |
76343538 BP |
201 | static void bridge_reconfigure_one(const struct ovsrec_open_vswitch *, |
202 | struct bridge *); | |
203 | static void bridge_reconfigure_controller(const struct ovsrec_open_vswitch *, | |
204 | struct bridge *); | |
205 | static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces); | |
064af421 BP |
206 | static void bridge_fetch_dp_ifaces(struct bridge *); |
207 | static void bridge_flush(struct bridge *); | |
208 | static void bridge_pick_local_hw_addr(struct bridge *, | |
209 | uint8_t ea[ETH_ADDR_LEN], | |
07c318f4 | 210 | struct iface **hw_addr_iface); |
064af421 BP |
211 | static uint64_t bridge_pick_datapath_id(struct bridge *, |
212 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
07c318f4 BP |
213 | struct iface *hw_addr_iface); |
214 | static struct iface *bridge_get_local_iface(struct bridge *); | |
064af421 BP |
215 | static uint64_t dpid_from_hash(const void *, size_t nbytes); |
216 | ||
8ca79daa | 217 | static unixctl_cb_func bridge_unixctl_fdb_show; |
8c4c1387 | 218 | |
da285df4 | 219 | static void bond_init(void); |
064af421 BP |
220 | static void bond_run(struct bridge *); |
221 | static void bond_wait(struct bridge *); | |
222 | static void bond_rebalance_port(struct port *); | |
2303f3b2 | 223 | static void bond_send_learning_packets(struct port *); |
8b2a2f4a | 224 | static void bond_enable_slave(struct iface *iface, bool enable); |
064af421 | 225 | |
76343538 BP |
226 | static struct port *port_create(struct bridge *, const char *name); |
227 | static void port_reconfigure(struct port *, const struct ovsrec_port *); | |
064af421 BP |
228 | static void port_destroy(struct port *); |
229 | static struct port *port_lookup(const struct bridge *, const char *name); | |
da285df4 | 230 | static struct iface *port_lookup_iface(const struct port *, const char *name); |
064af421 BP |
231 | static struct port *port_from_dp_ifidx(const struct bridge *, |
232 | uint16_t dp_ifidx); | |
233 | static void port_update_bond_compat(struct port *); | |
234 | static void port_update_vlan_compat(struct port *); | |
0c6aea3f | 235 | static void port_update_bonding(struct port *); |
064af421 | 236 | |
76343538 | 237 | #if 0 |
064af421 BP |
238 | static void mirror_create(struct bridge *, const char *name); |
239 | static void mirror_destroy(struct mirror *); | |
240 | static void mirror_reconfigure(struct bridge *); | |
241 | static void mirror_reconfigure_one(struct mirror *); | |
242 | static bool vlan_is_mirrored(const struct mirror *, int vlan); | |
76343538 BP |
243 | #else |
244 | static bool vlan_is_mirrored(const struct mirror *m UNUSED, int vlan UNUSED) | |
245 | { | |
246 | return false; | |
247 | } | |
248 | #endif | |
064af421 | 249 | |
a740f0de JG |
250 | static struct iface *iface_create(struct port *port, |
251 | const struct ovsrec_interface *if_cfg); | |
064af421 BP |
252 | static void iface_destroy(struct iface *); |
253 | static struct iface *iface_lookup(const struct bridge *, const char *name); | |
254 | static struct iface *iface_from_dp_ifidx(const struct bridge *, | |
255 | uint16_t dp_ifidx); | |
557d8e6c | 256 | static bool iface_is_internal(const struct bridge *, const char *name); |
52df17e7 | 257 | static void iface_set_mac(struct iface *); |
064af421 BP |
258 | |
259 | /* Hooks into ofproto processing. */ | |
260 | static struct ofhooks bridge_ofhooks; | |
261 | \f | |
262 | /* Public functions. */ | |
263 | ||
264 | /* Adds the name of each interface used by a bridge, including local and | |
265 | * internal ports, to 'svec'. */ | |
266 | void | |
267 | bridge_get_ifaces(struct svec *svec) | |
268 | { | |
269 | struct bridge *br, *next; | |
270 | size_t i, j; | |
271 | ||
272 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { | |
273 | for (i = 0; i < br->n_ports; i++) { | |
274 | struct port *port = br->ports[i]; | |
275 | ||
276 | for (j = 0; j < port->n_ifaces; j++) { | |
277 | struct iface *iface = port->ifaces[j]; | |
278 | if (iface->dp_ifidx < 0) { | |
b29ba128 | 279 | VLOG_ERR("%s interface not in datapath %s, ignoring", |
c228a364 | 280 | iface->name, dpif_name(br->dpif)); |
064af421 BP |
281 | } else { |
282 | if (iface->dp_ifidx != ODPP_LOCAL) { | |
283 | svec_add(svec, iface->name); | |
284 | } | |
285 | } | |
286 | } | |
287 | } | |
288 | } | |
289 | } | |
290 | ||
064af421 | 291 | void |
76343538 | 292 | bridge_init(const struct ovsrec_open_vswitch *cfg) |
064af421 | 293 | { |
76343538 | 294 | struct svec bridge_names; |
d3d22744 BP |
295 | struct svec dpif_names; |
296 | size_t i; | |
da285df4 | 297 | |
8ca79daa | 298 | unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); |
8c4c1387 | 299 | |
76343538 BP |
300 | svec_init(&bridge_names); |
301 | for (i = 0; i < cfg->n_bridges; i++) { | |
302 | svec_add(&bridge_names, cfg->bridges[i]->name); | |
303 | } | |
304 | svec_sort(&bridge_names); | |
305 | ||
d2a34569 | 306 | svec_init(&dpif_names); |
d3d22744 BP |
307 | dp_enumerate(&dpif_names); |
308 | for (i = 0; i < dpif_names.n; i++) { | |
309 | const char *dpif_name = dpif_names.names[i]; | |
c228a364 | 310 | struct dpif *dpif; |
d3d22744 | 311 | int retval; |
064af421 | 312 | |
d3d22744 | 313 | retval = dpif_open(dpif_name, &dpif); |
064af421 | 314 | if (!retval) { |
d3d22744 BP |
315 | struct svec all_names; |
316 | size_t j; | |
317 | ||
318 | svec_init(&all_names); | |
319 | dpif_get_all_names(dpif, &all_names); | |
320 | for (j = 0; j < all_names.n; j++) { | |
76343538 | 321 | if (svec_contains(&bridge_names, all_names.names[j])) { |
d3d22744 BP |
322 | goto found; |
323 | } | |
064af421 | 324 | } |
d3d22744 BP |
325 | dpif_delete(dpif); |
326 | found: | |
327 | svec_destroy(&all_names); | |
c228a364 | 328 | dpif_close(dpif); |
064af421 BP |
329 | } |
330 | } | |
d0c9d304 | 331 | svec_destroy(&dpif_names); |
064af421 | 332 | |
8ca79daa BP |
333 | unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, |
334 | NULL); | |
4f2cad2c | 335 | |
d3d22744 | 336 | bond_init(); |
76343538 | 337 | bridge_reconfigure(cfg); |
064af421 BP |
338 | } |
339 | ||
340 | #ifdef HAVE_OPENSSL | |
341 | static bool | |
76343538 | 342 | config_string_change(const char *value, char **valuep) |
064af421 | 343 | { |
064af421 BP |
344 | if (value && (!*valuep || strcmp(value, *valuep))) { |
345 | free(*valuep); | |
346 | *valuep = xstrdup(value); | |
347 | return true; | |
348 | } else { | |
349 | return false; | |
350 | } | |
351 | } | |
352 | ||
353 | static void | |
76343538 | 354 | bridge_configure_ssl(const struct ovsrec_ssl *ssl) |
064af421 BP |
355 | { |
356 | /* XXX SSL should be configurable on a per-bridge basis. | |
357 | * XXX should be possible to de-configure SSL. */ | |
358 | static char *private_key_file; | |
359 | static char *certificate_file; | |
360 | static char *cacert_file; | |
7e40e21d | 361 | struct stat s; |
064af421 | 362 | |
76343538 BP |
363 | if (!ssl) { |
364 | /* XXX We can't un-set SSL settings. */ | |
365 | return; | |
366 | } | |
367 | ||
368 | if (config_string_change(ssl->private_key, &private_key_file)) { | |
064af421 BP |
369 | vconn_ssl_set_private_key_file(private_key_file); |
370 | } | |
371 | ||
76343538 | 372 | if (config_string_change(ssl->certificate, &certificate_file)) { |
064af421 BP |
373 | vconn_ssl_set_certificate_file(certificate_file); |
374 | } | |
375 | ||
7e40e21d JP |
376 | /* We assume that even if the filename hasn't changed, if the CA cert |
377 | * file has been removed, that we want to move back into | |
378 | * boot-strapping mode. This opens a small security hole, because | |
379 | * the old certificate will still be trusted until vSwitch is | |
380 | * restarted. We may want to address this in vconn's SSL library. */ | |
76343538 | 381 | if (config_string_change(ssl->ca_cert, &cacert_file) |
de2047c5 | 382 | || (cacert_file && stat(cacert_file, &s) && errno == ENOENT)) { |
76343538 | 383 | vconn_ssl_set_ca_cert_file(cacert_file, ssl->bootstrap_ca_cert); |
064af421 BP |
384 | } |
385 | } | |
386 | #endif | |
387 | ||
6c88d577 JP |
388 | /* Attempt to create the network device 'iface_name' through the netdev |
389 | * library. */ | |
390 | static int | |
76343538 | 391 | set_up_iface(const struct ovsrec_interface *iface_cfg, bool create) |
6c88d577 | 392 | { |
76343538 BP |
393 | struct shash_node *node; |
394 | struct shash options; | |
6c88d577 JP |
395 | int error; |
396 | size_t i; | |
397 | ||
398 | /* If a type is not explicitly declared, then assume it's an existing | |
399 | * "system" device. */ | |
76343538 | 400 | if (iface_cfg->type[0] == '\0' || !strcmp(iface_cfg->type, "system")) { |
6c88d577 JP |
401 | return 0; |
402 | } | |
403 | ||
76343538 BP |
404 | shash_init(&options); |
405 | for (i = 0; i < iface_cfg->n_options; i++) { | |
406 | shash_add(&options, iface_cfg->key_options[i], | |
407 | xstrdup(iface_cfg->value_options[i])); | |
6c88d577 JP |
408 | } |
409 | ||
410 | if (create) { | |
76343538 | 411 | error = netdev_create(iface_cfg->name, iface_cfg->type, &options); |
6c88d577 JP |
412 | } else { |
413 | /* xxx Check to make sure that the type hasn't changed. */ | |
76343538 | 414 | error = netdev_reconfigure(iface_cfg->name, &options); |
6c88d577 JP |
415 | } |
416 | ||
76343538 BP |
417 | SHASH_FOR_EACH (node, &options) { |
418 | free(node->data); | |
419 | } | |
420 | shash_destroy(&options); | |
6c88d577 JP |
421 | |
422 | return error; | |
423 | } | |
424 | ||
6c88d577 | 425 | static int |
76343538 | 426 | reconfigure_iface(const struct ovsrec_interface *iface_cfg) |
6c88d577 | 427 | { |
76343538 | 428 | return set_up_iface(iface_cfg, false); |
6c88d577 JP |
429 | } |
430 | ||
6c88d577 | 431 | |
0c6aea3f BP |
432 | /* iterate_and_prune_ifaces() callback function that opens the network device |
433 | * for 'iface', if it is not already open, and retrieves the interface's MAC | |
434 | * address and carrier status. */ | |
435 | static bool | |
436 | init_iface_netdev(struct bridge *br UNUSED, struct iface *iface, | |
437 | void *aux UNUSED) | |
438 | { | |
439 | if (iface->netdev) { | |
440 | return true; | |
441 | } else if (!netdev_open(iface->name, NETDEV_ETH_TYPE_NONE, | |
442 | &iface->netdev)) { | |
0c6aea3f BP |
443 | netdev_get_carrier(iface->netdev, &iface->enabled); |
444 | return true; | |
445 | } else { | |
446 | /* If the network device can't be opened, then we're not going to try | |
447 | * to do anything with this interface. */ | |
448 | return false; | |
449 | } | |
450 | } | |
451 | ||
6ae39834 | 452 | static bool |
07c318f4 | 453 | check_iface_dp_ifidx(struct bridge *br, struct iface *iface, void *aux UNUSED) |
6ae39834 | 454 | { |
6ae39834 | 455 | if (iface->dp_ifidx >= 0) { |
6ae39834 BP |
456 | VLOG_DBG("%s has interface %s on port %d", |
457 | dpif_name(br->dpif), | |
458 | iface->name, iface->dp_ifidx); | |
459 | return true; | |
460 | } else { | |
461 | VLOG_ERR("%s interface not in %s, dropping", | |
462 | iface->name, dpif_name(br->dpif)); | |
463 | return false; | |
464 | } | |
465 | } | |
466 | ||
795fe1fb | 467 | static bool |
4d678233 | 468 | set_iface_properties(struct bridge *br UNUSED, struct iface *iface, |
795fe1fb BP |
469 | void *aux UNUSED) |
470 | { | |
4d678233 | 471 | /* Set policing attributes. */ |
76343538 BP |
472 | netdev_set_policing(iface->netdev, |
473 | iface->cfg->ingress_policing_rate, | |
474 | iface->cfg->ingress_policing_burst); | |
4d678233 BP |
475 | |
476 | /* Set MAC address of internal interfaces other than the local | |
477 | * interface. */ | |
478 | if (iface->dp_ifidx != ODPP_LOCAL | |
479 | && iface_is_internal(br, iface->name)) { | |
480 | iface_set_mac(iface); | |
481 | } | |
482 | ||
795fe1fb BP |
483 | return true; |
484 | } | |
485 | ||
6ae39834 BP |
486 | /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument. |
487 | * Deletes from 'br' all the interfaces for which 'cb' returns false, and then | |
488 | * deletes from 'br' any ports that no longer have any interfaces. */ | |
489 | static void | |
490 | iterate_and_prune_ifaces(struct bridge *br, | |
491 | bool (*cb)(struct bridge *, struct iface *, | |
492 | void *aux), | |
493 | void *aux) | |
494 | { | |
495 | size_t i, j; | |
496 | ||
497 | for (i = 0; i < br->n_ports; ) { | |
498 | struct port *port = br->ports[i]; | |
499 | for (j = 0; j < port->n_ifaces; ) { | |
500 | struct iface *iface = port->ifaces[j]; | |
501 | if (cb(br, iface, aux)) { | |
502 | j++; | |
503 | } else { | |
504 | iface_destroy(iface); | |
505 | } | |
506 | } | |
507 | ||
508 | if (port->n_ifaces) { | |
509 | i++; | |
510 | } else { | |
511 | VLOG_ERR("%s port has no interfaces, dropping", port->name); | |
512 | port_destroy(port); | |
513 | } | |
514 | } | |
515 | } | |
516 | ||
064af421 | 517 | void |
76343538 | 518 | bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) |
064af421 | 519 | { |
093e47f4 | 520 | struct ovsdb_idl_txn *txn; |
76343538 BP |
521 | struct shash old_br, new_br; |
522 | struct shash_node *node; | |
064af421 | 523 | struct bridge *br, *next; |
6ae39834 | 524 | size_t i; |
064af421 BP |
525 | |
526 | COVERAGE_INC(bridge_reconfigure); | |
527 | ||
093e47f4 BP |
528 | txn = ovsdb_idl_txn_create(ovs_cfg->header_.table->idl); |
529 | ||
632d136c | 530 | /* Collect old and new bridges. */ |
76343538 BP |
531 | shash_init(&old_br); |
532 | shash_init(&new_br); | |
064af421 | 533 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { |
76343538 BP |
534 | shash_add(&old_br, br->name, br); |
535 | } | |
536 | for (i = 0; i < ovs_cfg->n_bridges; i++) { | |
537 | const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i]; | |
538 | if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) { | |
539 | VLOG_WARN("more than one bridge named %s", br_cfg->name); | |
540 | } | |
064af421 | 541 | } |
064af421 BP |
542 | |
543 | /* Get rid of deleted bridges and add new bridges. */ | |
064af421 | 544 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { |
76343538 BP |
545 | struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name); |
546 | if (br_cfg) { | |
547 | br->cfg = br_cfg; | |
548 | } else { | |
064af421 BP |
549 | bridge_destroy(br); |
550 | } | |
551 | } | |
76343538 BP |
552 | SHASH_FOR_EACH (node, &new_br) { |
553 | const char *br_name = node->name; | |
554 | const struct ovsrec_bridge *br_cfg = node->data; | |
555 | if (!shash_find_data(&old_br, br_name)) { | |
556 | br = bridge_create(br_name); | |
fa33d64a BP |
557 | if (br) { |
558 | br->cfg = br_cfg; | |
559 | } | |
064af421 BP |
560 | } |
561 | } | |
76343538 BP |
562 | shash_destroy(&old_br); |
563 | shash_destroy(&new_br); | |
064af421 BP |
564 | |
565 | #ifdef HAVE_OPENSSL | |
566 | /* Configure SSL. */ | |
76343538 | 567 | bridge_configure_ssl(ovs_cfg->ssl); |
064af421 BP |
568 | #endif |
569 | ||
570 | /* Reconfigure all bridges. */ | |
571 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
76343538 | 572 | bridge_reconfigure_one(ovs_cfg, br); |
064af421 BP |
573 | } |
574 | ||
575 | /* Add and delete ports on all datapaths. | |
576 | * | |
577 | * The kernel will reject any attempt to add a given port to a datapath if | |
578 | * that port already belongs to a different datapath, so we must do all | |
579 | * port deletions before any port additions. */ | |
580 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
581 | struct odp_port *dpif_ports; | |
582 | size_t n_dpif_ports; | |
76343538 | 583 | struct shash want_ifaces; |
064af421 | 584 | |
c228a364 | 585 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
064af421 BP |
586 | bridge_get_all_ifaces(br, &want_ifaces); |
587 | for (i = 0; i < n_dpif_ports; i++) { | |
588 | const struct odp_port *p = &dpif_ports[i]; | |
76343538 | 589 | if (!shash_find(&want_ifaces, p->devname) |
064af421 | 590 | && strcmp(p->devname, br->name)) { |
c228a364 | 591 | int retval = dpif_port_del(br->dpif, p->port); |
064af421 | 592 | if (retval) { |
b29ba128 | 593 | VLOG_ERR("failed to remove %s interface from %s: %s", |
c228a364 | 594 | p->devname, dpif_name(br->dpif), |
b29ba128 | 595 | strerror(retval)); |
064af421 BP |
596 | } |
597 | } | |
598 | } | |
76343538 | 599 | shash_destroy(&want_ifaces); |
064af421 BP |
600 | free(dpif_ports); |
601 | } | |
602 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
603 | struct odp_port *dpif_ports; | |
604 | size_t n_dpif_ports; | |
76343538 BP |
605 | struct shash cur_ifaces, want_ifaces; |
606 | struct shash_node *node; | |
064af421 | 607 | |
76343538 | 608 | /* Get the set of interfaces currently in this datapath. */ |
c228a364 | 609 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
76343538 | 610 | shash_init(&cur_ifaces); |
064af421 | 611 | for (i = 0; i < n_dpif_ports; i++) { |
76343538 BP |
612 | const char *name = dpif_ports[i].devname; |
613 | if (!shash_find(&cur_ifaces, name)) { | |
614 | shash_add(&cur_ifaces, name, NULL); | |
615 | } | |
064af421 BP |
616 | } |
617 | free(dpif_ports); | |
064af421 | 618 | |
76343538 BP |
619 | /* Get the set of interfaces we want on this datapath. */ |
620 | bridge_get_all_ifaces(br, &want_ifaces); | |
6c88d577 | 621 | |
76343538 BP |
622 | SHASH_FOR_EACH (node, &want_ifaces) { |
623 | const char *if_name = node->name; | |
624 | struct iface *iface = node->data; | |
35c979bf | 625 | |
76343538 BP |
626 | if (shash_find(&cur_ifaces, if_name)) { |
627 | /* Already exists, just reconfigure it. */ | |
628 | if (iface) { | |
629 | reconfigure_iface(iface->cfg); | |
630 | } | |
631 | } else { | |
632 | /* Need to add to datapath. */ | |
633 | bool internal; | |
634 | int error; | |
635 | ||
76343538 | 636 | /* Add to datapath. */ |
a740f0de | 637 | internal = iface_is_internal(br, if_name); |
76343538 BP |
638 | error = dpif_port_add(br->dpif, if_name, |
639 | internal ? ODP_PORT_INTERNAL : 0, NULL); | |
640 | if (error == EFBIG) { | |
641 | VLOG_ERR("ran out of valid port numbers on %s", | |
642 | dpif_name(br->dpif)); | |
643 | break; | |
644 | } else if (error) { | |
645 | VLOG_ERR("failed to add %s interface to %s: %s", | |
646 | if_name, dpif_name(br->dpif), strerror(error)); | |
647 | } | |
064af421 BP |
648 | } |
649 | } | |
76343538 BP |
650 | shash_destroy(&cur_ifaces); |
651 | shash_destroy(&want_ifaces); | |
064af421 BP |
652 | } |
653 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
654 | uint8_t ea[8]; | |
655 | uint64_t dpid; | |
07c318f4 BP |
656 | struct iface *local_iface; |
657 | struct iface *hw_addr_iface; | |
093e47f4 | 658 | char *dpid_string; |
064af421 | 659 | |
064af421 | 660 | bridge_fetch_dp_ifaces(br); |
0c6aea3f | 661 | iterate_and_prune_ifaces(br, init_iface_netdev, NULL); |
064af421 | 662 | |
07c318f4 | 663 | iterate_and_prune_ifaces(br, check_iface_dp_ifidx, NULL); |
064af421 BP |
664 | |
665 | /* Pick local port hardware address, datapath ID. */ | |
07c318f4 BP |
666 | bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); |
667 | local_iface = bridge_get_local_iface(br); | |
064af421 | 668 | if (local_iface) { |
07c318f4 | 669 | int error = netdev_set_etheraddr(local_iface->netdev, ea); |
064af421 BP |
670 | if (error) { |
671 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
672 | VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " | |
673 | "Ethernet address: %s", | |
674 | br->name, strerror(error)); | |
675 | } | |
676 | } | |
677 | ||
07c318f4 | 678 | dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); |
064af421 BP |
679 | ofproto_set_datapath_id(br->ofproto, dpid); |
680 | ||
093e47f4 BP |
681 | dpid_string = xasprintf("%012"PRIx64, dpid); |
682 | ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); | |
683 | free(dpid_string); | |
684 | ||
064af421 | 685 | /* Set NetFlow configuration on this bridge. */ |
76343538 BP |
686 | if (br->cfg->netflow) { |
687 | struct ovsrec_netflow *nf_cfg = br->cfg->netflow; | |
688 | struct netflow_options opts; | |
689 | ||
690 | memset(&opts, 0, sizeof opts); | |
691 | ||
692 | dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); | |
693 | if (nf_cfg->engine_type) { | |
694 | opts.engine_type = nf_cfg->engine_type; | |
695 | } | |
696 | if (nf_cfg->engine_id) { | |
697 | opts.engine_id = nf_cfg->engine_id; | |
698 | } | |
699 | ||
700 | opts.active_timeout = nf_cfg->active_timeout; | |
701 | if (!opts.active_timeout) { | |
702 | opts.active_timeout = -1; | |
703 | } else if (opts.active_timeout < 0) { | |
e9e2856e JG |
704 | VLOG_WARN("bridge %s: active timeout interval set to negative " |
705 | "value, using default instead (%d seconds)", br->name, | |
706 | NF_ACTIVE_TIMEOUT_DEFAULT); | |
707 | opts.active_timeout = -1; | |
76343538 BP |
708 | } |
709 | ||
710 | opts.add_id_to_iface = nf_cfg->add_id_to_interface; | |
711 | if (opts.add_id_to_iface) { | |
712 | if (opts.engine_id > 0x7f) { | |
713 | VLOG_WARN("bridge %s: netflow port mangling may conflict " | |
714 | "with another vswitch, choose an engine id less " | |
715 | "than 128", br->name); | |
716 | } | |
717 | if (br->n_ports > 508) { | |
718 | VLOG_WARN("bridge %s: netflow port mangling will conflict " | |
719 | "with another port when more than 508 ports are " | |
720 | "used", br->name); | |
721 | } | |
722 | } | |
723 | ||
724 | opts.collectors.n = nf_cfg->n_targets; | |
725 | opts.collectors.names = nf_cfg->targets; | |
726 | if (ofproto_set_netflow(br->ofproto, &opts)) { | |
727 | VLOG_ERR("bridge %s: problem setting netflow collectors", | |
728 | br->name); | |
729 | } | |
730 | } else { | |
731 | ofproto_set_netflow(br->ofproto, NULL); | |
732 | } | |
064af421 BP |
733 | |
734 | /* Update the controller and related settings. It would be more | |
735 | * straightforward to call this from bridge_reconfigure_one(), but we | |
736 | * can't do it there for two reasons. First, and most importantly, at | |
737 | * that point we don't know the dp_ifidx of any interfaces that have | |
738 | * been added to the bridge (because we haven't actually added them to | |
739 | * the datapath). Second, at that point we haven't set the datapath ID | |
740 | * yet; when a controller is configured, resetting the datapath ID will | |
741 | * immediately disconnect from the controller, so it's better to set | |
742 | * the datapath ID before the controller. */ | |
76343538 | 743 | bridge_reconfigure_controller(ovs_cfg, br); |
064af421 BP |
744 | } |
745 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
746 | for (i = 0; i < br->n_ports; i++) { | |
747 | struct port *port = br->ports[i]; | |
52df17e7 | 748 | |
064af421 | 749 | port_update_vlan_compat(port); |
0c6aea3f | 750 | port_update_bonding(port); |
064af421 BP |
751 | } |
752 | } | |
753 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
4d678233 | 754 | iterate_and_prune_ifaces(br, set_iface_properties, NULL); |
064af421 | 755 | } |
093e47f4 BP |
756 | |
757 | ovsdb_idl_txn_commit(txn); | |
758 | ovsdb_idl_txn_destroy(txn); /* XXX */ | |
759 | } | |
760 | ||
761 | static const char * | |
762 | bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) | |
763 | { | |
764 | size_t i; | |
765 | ||
766 | for (i = 0; i < br_cfg->n_other_config; i++) { | |
767 | if (!strcmp(br_cfg->key_other_config[i], key)) { | |
768 | return br_cfg->value_other_config[i]; | |
769 | } | |
770 | } | |
771 | return NULL; | |
064af421 BP |
772 | } |
773 | ||
774 | static void | |
775 | bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], | |
07c318f4 | 776 | struct iface **hw_addr_iface) |
064af421 | 777 | { |
093e47f4 | 778 | const char *hwaddr; |
064af421 BP |
779 | size_t i, j; |
780 | int error; | |
781 | ||
07c318f4 | 782 | *hw_addr_iface = NULL; |
064af421 BP |
783 | |
784 | /* Did the user request a particular MAC? */ | |
093e47f4 BP |
785 | hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); |
786 | if (hwaddr && eth_addr_from_string(hwaddr, ea)) { | |
064af421 BP |
787 | if (eth_addr_is_multicast(ea)) { |
788 | VLOG_ERR("bridge %s: cannot set MAC address to multicast " | |
789 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
790 | } else if (eth_addr_is_zero(ea)) { | |
791 | VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name); | |
792 | } else { | |
793 | return; | |
794 | } | |
795 | } | |
796 | ||
141f4942 BP |
797 | /* Otherwise choose the minimum non-local MAC address among all of the |
798 | * interfaces. */ | |
064af421 BP |
799 | memset(ea, 0xff, sizeof ea); |
800 | for (i = 0; i < br->n_ports; i++) { | |
801 | struct port *port = br->ports[i]; | |
58b7527e | 802 | uint8_t iface_ea[ETH_ADDR_LEN]; |
58b7527e BP |
803 | struct iface *iface; |
804 | ||
805 | /* Mirror output ports don't participate. */ | |
064af421 BP |
806 | if (port->is_mirror_output_port) { |
807 | continue; | |
808 | } | |
58b7527e BP |
809 | |
810 | /* Choose the MAC address to represent the port. */ | |
76343538 | 811 | if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { |
ba09980a BP |
812 | /* Find the interface with this Ethernet address (if any) so that |
813 | * we can provide the correct devname to the caller. */ | |
814 | iface = NULL; | |
815 | for (j = 0; j < port->n_ifaces; j++) { | |
816 | struct iface *candidate = port->ifaces[j]; | |
817 | uint8_t candidate_ea[ETH_ADDR_LEN]; | |
8fef8c71 | 818 | if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) |
ba09980a BP |
819 | && eth_addr_equals(iface_ea, candidate_ea)) { |
820 | iface = candidate; | |
821 | } | |
822 | } | |
58b7527e BP |
823 | } else { |
824 | /* Choose the interface whose MAC address will represent the port. | |
825 | * The Linux kernel bonding code always chooses the MAC address of | |
826 | * the first slave added to a bond, and the Fedora networking | |
827 | * scripts always add slaves to a bond in alphabetical order, so | |
828 | * for compatibility we choose the interface with the name that is | |
829 | * first in alphabetical order. */ | |
830 | iface = port->ifaces[0]; | |
831 | for (j = 1; j < port->n_ifaces; j++) { | |
832 | struct iface *candidate = port->ifaces[j]; | |
833 | if (strcmp(candidate->name, iface->name) < 0) { | |
834 | iface = candidate; | |
835 | } | |
836 | } | |
837 | ||
838 | /* The local port doesn't count (since we're trying to choose its | |
141f4942 BP |
839 | * MAC address anyway). */ |
840 | if (iface->dp_ifidx == ODPP_LOCAL) { | |
064af421 BP |
841 | continue; |
842 | } | |
58b7527e BP |
843 | |
844 | /* Grab MAC. */ | |
07c318f4 | 845 | error = netdev_get_etheraddr(iface->netdev, iface_ea); |
58b7527e | 846 | if (error) { |
064af421 BP |
847 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
848 | VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s", | |
849 | iface->name, strerror(error)); | |
58b7527e | 850 | continue; |
064af421 BP |
851 | } |
852 | } | |
58b7527e BP |
853 | |
854 | /* Compare against our current choice. */ | |
855 | if (!eth_addr_is_multicast(iface_ea) && | |
141f4942 | 856 | !eth_addr_is_local(iface_ea) && |
58b7527e BP |
857 | !eth_addr_is_reserved(iface_ea) && |
858 | !eth_addr_is_zero(iface_ea) && | |
859 | memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) | |
860 | { | |
0babc06f | 861 | memcpy(ea, iface_ea, ETH_ADDR_LEN); |
8fef8c71 | 862 | *hw_addr_iface = iface; |
58b7527e | 863 | } |
064af421 | 864 | } |
141f4942 | 865 | if (eth_addr_is_multicast(ea)) { |
064af421 | 866 | memcpy(ea, br->default_ea, ETH_ADDR_LEN); |
07c318f4 | 867 | *hw_addr_iface = NULL; |
064af421 BP |
868 | VLOG_WARN("bridge %s: using default bridge Ethernet " |
869 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
870 | } else { | |
871 | VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT, | |
872 | br->name, ETH_ADDR_ARGS(ea)); | |
873 | } | |
874 | } | |
875 | ||
876 | /* Choose and returns the datapath ID for bridge 'br' given that the bridge | |
877 | * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of | |
07c318f4 BP |
878 | * an interface on 'br', then that interface must be passed in as |
879 | * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then | |
880 | * 'hw_addr_iface' must be passed in as a null pointer. */ | |
064af421 BP |
881 | static uint64_t |
882 | bridge_pick_datapath_id(struct bridge *br, | |
883 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
07c318f4 | 884 | struct iface *hw_addr_iface) |
064af421 BP |
885 | { |
886 | /* | |
887 | * The procedure for choosing a bridge MAC address will, in the most | |
888 | * ordinary case, also choose a unique MAC that we can use as a datapath | |
889 | * ID. In some special cases, though, multiple bridges will end up with | |
890 | * the same MAC address. This is OK for the bridges, but it will confuse | |
891 | * the OpenFlow controller, because each datapath needs a unique datapath | |
892 | * ID. | |
893 | * | |
894 | * Datapath IDs must be unique. It is also very desirable that they be | |
895 | * stable from one run to the next, so that policy set on a datapath | |
896 | * "sticks". | |
897 | */ | |
093e47f4 | 898 | const char *datapath_id; |
064af421 BP |
899 | uint64_t dpid; |
900 | ||
093e47f4 BP |
901 | datapath_id = bridge_get_other_config(br->cfg, "datapath-id"); |
902 | if (datapath_id && dpid_from_string(datapath_id, &dpid)) { | |
064af421 BP |
903 | return dpid; |
904 | } | |
905 | ||
07c318f4 | 906 | if (hw_addr_iface) { |
064af421 | 907 | int vlan; |
b2f17b15 | 908 | if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) { |
064af421 BP |
909 | /* |
910 | * A bridge whose MAC address is taken from a VLAN network device | |
911 | * (that is, a network device created with vconfig(8) or similar | |
912 | * tool) will have the same MAC address as a bridge on the VLAN | |
913 | * device's physical network device. | |
914 | * | |
915 | * Handle this case by hashing the physical network device MAC | |
916 | * along with the VLAN identifier. | |
917 | */ | |
918 | uint8_t buf[ETH_ADDR_LEN + 2]; | |
919 | memcpy(buf, bridge_ea, ETH_ADDR_LEN); | |
920 | buf[ETH_ADDR_LEN] = vlan >> 8; | |
921 | buf[ETH_ADDR_LEN + 1] = vlan; | |
922 | return dpid_from_hash(buf, sizeof buf); | |
923 | } else { | |
924 | /* | |
925 | * Assume that this bridge's MAC address is unique, since it | |
926 | * doesn't fit any of the cases we handle specially. | |
927 | */ | |
928 | } | |
929 | } else { | |
930 | /* | |
931 | * A purely internal bridge, that is, one that has no non-virtual | |
932 | * network devices on it at all, is more difficult because it has no | |
933 | * natural unique identifier at all. | |
934 | * | |
935 | * When the host is a XenServer, we handle this case by hashing the | |
936 | * host's UUID with the name of the bridge. Names of bridges are | |
937 | * persistent across XenServer reboots, although they can be reused if | |
938 | * an internal network is destroyed and then a new one is later | |
939 | * created, so this is fairly effective. | |
940 | * | |
941 | * When the host is not a XenServer, we punt by using a random MAC | |
942 | * address on each run. | |
943 | */ | |
944 | const char *host_uuid = xenserver_get_host_uuid(); | |
945 | if (host_uuid) { | |
946 | char *combined = xasprintf("%s,%s", host_uuid, br->name); | |
947 | dpid = dpid_from_hash(combined, strlen(combined)); | |
948 | free(combined); | |
949 | return dpid; | |
950 | } | |
951 | } | |
952 | ||
953 | return eth_addr_to_uint64(bridge_ea); | |
954 | } | |
955 | ||
956 | static uint64_t | |
957 | dpid_from_hash(const void *data, size_t n) | |
958 | { | |
5eccf359 | 959 | uint8_t hash[SHA1_DIGEST_SIZE]; |
064af421 BP |
960 | |
961 | BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN); | |
5eccf359 | 962 | sha1_bytes(data, n, hash); |
064af421 BP |
963 | eth_addr_mark_random(hash); |
964 | return eth_addr_to_uint64(hash); | |
965 | } | |
966 | ||
967 | int | |
968 | bridge_run(void) | |
969 | { | |
970 | struct bridge *br, *next; | |
971 | int retval; | |
972 | ||
973 | retval = 0; | |
974 | LIST_FOR_EACH_SAFE (br, next, struct bridge, node, &all_bridges) { | |
975 | int error = bridge_run_one(br); | |
976 | if (error) { | |
977 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
978 | VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " | |
979 | "forcing reconfiguration", br->name); | |
980 | if (!retval) { | |
981 | retval = error; | |
982 | } | |
983 | } | |
984 | } | |
985 | return retval; | |
986 | } | |
987 | ||
988 | void | |
989 | bridge_wait(void) | |
990 | { | |
991 | struct bridge *br; | |
992 | ||
993 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
994 | ofproto_wait(br->ofproto); | |
995 | if (br->controller) { | |
996 | continue; | |
997 | } | |
998 | ||
93dfc067 | 999 | mac_learning_wait(br->ml); |
064af421 | 1000 | bond_wait(br); |
064af421 BP |
1001 | } |
1002 | } | |
1003 | ||
1004 | /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s | |
1005 | * configuration changes. */ | |
1006 | static void | |
1007 | bridge_flush(struct bridge *br) | |
1008 | { | |
1009 | COVERAGE_INC(bridge_flush); | |
1010 | br->flush = true; | |
93dfc067 | 1011 | mac_learning_flush(br->ml); |
064af421 | 1012 | } |
07c318f4 BP |
1013 | |
1014 | /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no | |
1015 | * such interface. */ | |
1016 | static struct iface * | |
1017 | bridge_get_local_iface(struct bridge *br) | |
1018 | { | |
1019 | size_t i, j; | |
1020 | ||
1021 | for (i = 0; i < br->n_ports; i++) { | |
1022 | struct port *port = br->ports[i]; | |
1023 | for (j = 0; j < port->n_ifaces; j++) { | |
1024 | struct iface *iface = port->ifaces[j]; | |
1025 | if (iface->dp_ifidx == ODPP_LOCAL) { | |
1026 | return iface; | |
1027 | } | |
1028 | } | |
1029 | } | |
1030 | ||
1031 | return NULL; | |
1032 | } | |
064af421 | 1033 | \f |
8c4c1387 BP |
1034 | /* Bridge unixctl user interface functions. */ |
1035 | static void | |
8ca79daa BP |
1036 | bridge_unixctl_fdb_show(struct unixctl_conn *conn, |
1037 | const char *args, void *aux UNUSED) | |
8c4c1387 BP |
1038 | { |
1039 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1040 | const struct bridge *br; | |
93dfc067 | 1041 | const struct mac_entry *e; |
8c4c1387 BP |
1042 | |
1043 | br = bridge_lookup(args); | |
1044 | if (!br) { | |
1045 | unixctl_command_reply(conn, 501, "no such bridge"); | |
1046 | return; | |
1047 | } | |
1048 | ||
1049 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
93dfc067 JG |
1050 | LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { |
1051 | if (e->port < 0 || e->port >= br->n_ports) { | |
1052 | continue; | |
8c4c1387 | 1053 | } |
93dfc067 JG |
1054 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", |
1055 | br->ports[e->port]->ifaces[0]->dp_ifidx, | |
1056 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
8c4c1387 BP |
1057 | } |
1058 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
1059 | ds_destroy(&ds); | |
1060 | } | |
1061 | \f | |
064af421 BP |
1062 | /* Bridge reconfiguration functions. */ |
1063 | ||
1064 | static struct bridge * | |
1065 | bridge_create(const char *name) | |
1066 | { | |
1067 | struct bridge *br; | |
1068 | int error; | |
1069 | ||
1070 | assert(!bridge_lookup(name)); | |
ec6fde61 | 1071 | br = xzalloc(sizeof *br); |
064af421 | 1072 | |
efacbce6 BP |
1073 | error = dpif_create_and_open(name, &br->dpif); |
1074 | if (error) { | |
064af421 BP |
1075 | free(br); |
1076 | return NULL; | |
1077 | } | |
efacbce6 | 1078 | dpif_flow_flush(br->dpif); |
064af421 BP |
1079 | |
1080 | error = ofproto_create(name, &bridge_ofhooks, br, &br->ofproto); | |
1081 | if (error) { | |
1082 | VLOG_ERR("failed to create switch %s: %s", name, strerror(error)); | |
c228a364 BP |
1083 | dpif_delete(br->dpif); |
1084 | dpif_close(br->dpif); | |
064af421 BP |
1085 | free(br); |
1086 | return NULL; | |
1087 | } | |
1088 | ||
1089 | br->name = xstrdup(name); | |
1090 | br->ml = mac_learning_create(); | |
1091 | br->sent_config_request = false; | |
1092 | eth_addr_random(br->default_ea); | |
1093 | ||
1094 | port_array_init(&br->ifaces); | |
1095 | ||
1096 | br->flush = false; | |
1097 | br->bond_next_rebalance = time_msec() + 10000; | |
1098 | ||
1099 | list_push_back(&all_bridges, &br->node); | |
1100 | ||
c228a364 | 1101 | VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif)); |
064af421 BP |
1102 | |
1103 | return br; | |
1104 | } | |
1105 | ||
1106 | static void | |
1107 | bridge_destroy(struct bridge *br) | |
1108 | { | |
1109 | if (br) { | |
1110 | int error; | |
1111 | ||
1112 | while (br->n_ports > 0) { | |
1113 | port_destroy(br->ports[br->n_ports - 1]); | |
1114 | } | |
1115 | list_remove(&br->node); | |
c228a364 | 1116 | error = dpif_delete(br->dpif); |
064af421 | 1117 | if (error && error != ENOENT) { |
b29ba128 | 1118 | VLOG_ERR("failed to delete %s: %s", |
c228a364 | 1119 | dpif_name(br->dpif), strerror(error)); |
064af421 | 1120 | } |
c228a364 | 1121 | dpif_close(br->dpif); |
064af421 BP |
1122 | ofproto_destroy(br->ofproto); |
1123 | free(br->controller); | |
1124 | mac_learning_destroy(br->ml); | |
1125 | port_array_destroy(&br->ifaces); | |
1126 | free(br->ports); | |
1127 | free(br->name); | |
1128 | free(br); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | static struct bridge * | |
1133 | bridge_lookup(const char *name) | |
1134 | { | |
1135 | struct bridge *br; | |
1136 | ||
1137 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
1138 | if (!strcmp(br->name, name)) { | |
1139 | return br; | |
1140 | } | |
1141 | } | |
1142 | return NULL; | |
1143 | } | |
1144 | ||
1145 | bool | |
1146 | bridge_exists(const char *name) | |
1147 | { | |
1148 | return bridge_lookup(name) ? true : false; | |
1149 | } | |
1150 | ||
1151 | uint64_t | |
1152 | bridge_get_datapathid(const char *name) | |
1153 | { | |
1154 | struct bridge *br = bridge_lookup(name); | |
1155 | return br ? ofproto_get_datapath_id(br->ofproto) : 0; | |
1156 | } | |
1157 | ||
4f2cad2c JP |
1158 | /* Handle requests for a listing of all flows known by the OpenFlow |
1159 | * stack, including those normally hidden. */ | |
1160 | static void | |
8ca79daa BP |
1161 | bridge_unixctl_dump_flows(struct unixctl_conn *conn, |
1162 | const char *args, void *aux UNUSED) | |
4f2cad2c JP |
1163 | { |
1164 | struct bridge *br; | |
1165 | struct ds results; | |
1166 | ||
1167 | br = bridge_lookup(args); | |
1168 | if (!br) { | |
1169 | unixctl_command_reply(conn, 501, "Unknown bridge"); | |
1170 | return; | |
1171 | } | |
1172 | ||
1173 | ds_init(&results); | |
1174 | ofproto_get_all_flows(br->ofproto, &results); | |
1175 | ||
1176 | unixctl_command_reply(conn, 200, ds_cstr(&results)); | |
1177 | ds_destroy(&results); | |
1178 | } | |
1179 | ||
064af421 BP |
1180 | static int |
1181 | bridge_run_one(struct bridge *br) | |
1182 | { | |
1183 | int error; | |
1184 | ||
1185 | error = ofproto_run1(br->ofproto); | |
1186 | if (error) { | |
1187 | return error; | |
1188 | } | |
1189 | ||
93dfc067 | 1190 | mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); |
064af421 | 1191 | bond_run(br); |
064af421 BP |
1192 | |
1193 | error = ofproto_run2(br->ofproto, br->flush); | |
1194 | br->flush = false; | |
1195 | ||
1196 | return error; | |
1197 | } | |
1198 | ||
76343538 BP |
1199 | static const struct ovsrec_controller * |
1200 | bridge_get_controller(const struct ovsrec_open_vswitch *ovs_cfg, | |
1201 | const struct bridge *br) | |
064af421 | 1202 | { |
76343538 | 1203 | const struct ovsrec_controller *controller; |
064af421 | 1204 | |
76343538 BP |
1205 | controller = (br->cfg->controller ? br->cfg->controller |
1206 | : ovs_cfg->controller ? ovs_cfg->controller | |
1207 | : NULL); | |
1208 | ||
1209 | if (controller && !strcmp(controller->target, "none")) { | |
1210 | return NULL; | |
064af421 | 1211 | } |
76343538 BP |
1212 | |
1213 | return controller; | |
064af421 BP |
1214 | } |
1215 | ||
6ae39834 BP |
1216 | static bool |
1217 | check_duplicate_ifaces(struct bridge *br, struct iface *iface, void *ifaces_) | |
1218 | { | |
1219 | struct svec *ifaces = ifaces_; | |
1220 | if (!svec_contains(ifaces, iface->name)) { | |
1221 | svec_add(ifaces, iface->name); | |
1222 | svec_sort(ifaces); | |
1223 | return true; | |
1224 | } else { | |
1225 | VLOG_ERR("bridge %s: %s interface is on multiple ports, " | |
1226 | "removing from %s", | |
1227 | br->name, iface->name, iface->port->name); | |
1228 | return false; | |
1229 | } | |
1230 | } | |
1231 | ||
064af421 | 1232 | static void |
76343538 BP |
1233 | bridge_reconfigure_one(const struct ovsrec_open_vswitch *ovs_cfg, |
1234 | struct bridge *br) | |
064af421 | 1235 | { |
76343538 BP |
1236 | struct shash old_ports, new_ports; |
1237 | struct svec ifaces; | |
064af421 BP |
1238 | struct svec listeners, old_listeners; |
1239 | struct svec snoops, old_snoops; | |
76343538 BP |
1240 | struct shash_node *node; |
1241 | uint64_t mgmt_id; | |
6ae39834 | 1242 | size_t i; |
064af421 BP |
1243 | |
1244 | /* Collect old ports. */ | |
76343538 | 1245 | shash_init(&old_ports); |
064af421 | 1246 | for (i = 0; i < br->n_ports; i++) { |
76343538 | 1247 | shash_add(&old_ports, br->ports[i]->name, br->ports[i]); |
064af421 | 1248 | } |
064af421 BP |
1249 | |
1250 | /* Collect new ports. */ | |
76343538 BP |
1251 | shash_init(&new_ports); |
1252 | for (i = 0; i < br->cfg->n_ports; i++) { | |
1253 | const char *name = br->cfg->ports[i]->name; | |
1254 | if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { | |
1255 | VLOG_WARN("bridge %s: %s specified twice as bridge port", | |
1256 | br->name, name); | |
1257 | } | |
1258 | } | |
e073f944 BP |
1259 | |
1260 | /* If we have a controller, then we need a local port. Complain if the | |
1261 | * user didn't specify one. | |
1262 | * | |
1263 | * XXX perhaps we should synthesize a port ourselves in this case. */ | |
76343538 | 1264 | if (bridge_get_controller(ovs_cfg, br)) { |
72865317 BP |
1265 | char local_name[IF_NAMESIZE]; |
1266 | int error; | |
1267 | ||
1268 | error = dpif_port_get_name(br->dpif, ODPP_LOCAL, | |
1269 | local_name, sizeof local_name); | |
e073f944 BP |
1270 | if (!error && !shash_find(&new_ports, local_name)) { |
1271 | VLOG_WARN("bridge %s: controller specified but no local port " | |
1272 | "(port named %s) defined", | |
1273 | br->name, local_name); | |
72865317 | 1274 | } |
064af421 | 1275 | } |
064af421 | 1276 | |
76343538 | 1277 | dpid_from_string(ovs_cfg->management_id, &mgmt_id); |
064af421 BP |
1278 | ofproto_set_mgmt_id(br->ofproto, mgmt_id); |
1279 | ||
1280 | /* Get rid of deleted ports and add new ports. */ | |
76343538 BP |
1281 | SHASH_FOR_EACH (node, &old_ports) { |
1282 | if (!shash_find(&new_ports, node->name)) { | |
1283 | port_destroy(node->data); | |
064af421 BP |
1284 | } |
1285 | } | |
76343538 BP |
1286 | SHASH_FOR_EACH (node, &new_ports) { |
1287 | struct port *port = shash_find_data(&old_ports, node->name); | |
1288 | if (!port) { | |
1289 | port = port_create(br, node->name); | |
064af421 | 1290 | } |
76343538 | 1291 | port_reconfigure(port, node->data); |
064af421 | 1292 | } |
76343538 BP |
1293 | shash_destroy(&old_ports); |
1294 | shash_destroy(&new_ports); | |
064af421 BP |
1295 | |
1296 | /* Check and delete duplicate interfaces. */ | |
1297 | svec_init(&ifaces); | |
6ae39834 | 1298 | iterate_and_prune_ifaces(br, check_duplicate_ifaces, &ifaces); |
064af421 BP |
1299 | svec_destroy(&ifaces); |
1300 | ||
1301 | /* Delete all flows if we're switching from connected to standalone or vice | |
1302 | * versa. (XXX Should we delete all flows if we are switching from one | |
1303 | * controller to another?) */ | |
1304 | ||
76343538 | 1305 | #if 0 |
064af421 BP |
1306 | /* Configure OpenFlow management listeners. */ |
1307 | svec_init(&listeners); | |
1308 | cfg_get_all_strings(&listeners, "bridge.%s.openflow.listeners", br->name); | |
1309 | if (!listeners.n) { | |
1310 | svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", | |
1311 | ovs_rundir, br->name)); | |
1312 | } else if (listeners.n == 1 && !strcmp(listeners.names[0], "none")) { | |
1313 | svec_clear(&listeners); | |
1314 | } | |
1315 | svec_sort_unique(&listeners); | |
1316 | ||
1317 | svec_init(&old_listeners); | |
1318 | ofproto_get_listeners(br->ofproto, &old_listeners); | |
1319 | svec_sort_unique(&old_listeners); | |
1320 | ||
1321 | if (!svec_equal(&listeners, &old_listeners)) { | |
1322 | ofproto_set_listeners(br->ofproto, &listeners); | |
1323 | } | |
1324 | svec_destroy(&listeners); | |
1325 | svec_destroy(&old_listeners); | |
1326 | ||
1327 | /* Configure OpenFlow controller connection snooping. */ | |
1328 | svec_init(&snoops); | |
1329 | cfg_get_all_strings(&snoops, "bridge.%s.openflow.snoops", br->name); | |
1330 | if (!snoops.n) { | |
1331 | svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", | |
1332 | ovs_rundir, br->name)); | |
1333 | } else if (snoops.n == 1 && !strcmp(snoops.names[0], "none")) { | |
1334 | svec_clear(&snoops); | |
1335 | } | |
1336 | svec_sort_unique(&snoops); | |
1337 | ||
1338 | svec_init(&old_snoops); | |
1339 | ofproto_get_snoops(br->ofproto, &old_snoops); | |
1340 | svec_sort_unique(&old_snoops); | |
1341 | ||
1342 | if (!svec_equal(&snoops, &old_snoops)) { | |
1343 | ofproto_set_snoops(br->ofproto, &snoops); | |
1344 | } | |
1345 | svec_destroy(&snoops); | |
1346 | svec_destroy(&old_snoops); | |
76343538 BP |
1347 | #else |
1348 | /* Default listener. */ | |
1349 | svec_init(&listeners); | |
1350 | svec_add_nocopy(&listeners, xasprintf("punix:%s/%s.mgmt", | |
1351 | ovs_rundir, br->name)); | |
1352 | svec_init(&old_listeners); | |
1353 | ofproto_get_listeners(br->ofproto, &old_listeners); | |
1354 | if (!svec_equal(&listeners, &old_listeners)) { | |
1355 | ofproto_set_listeners(br->ofproto, &listeners); | |
1356 | } | |
1357 | svec_destroy(&listeners); | |
1358 | svec_destroy(&old_listeners); | |
064af421 | 1359 | |
76343538 BP |
1360 | /* Default snoop. */ |
1361 | svec_init(&snoops); | |
1362 | svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", | |
1363 | ovs_rundir, br->name)); | |
1364 | svec_init(&old_snoops); | |
1365 | ofproto_get_snoops(br->ofproto, &old_snoops); | |
1366 | if (!svec_equal(&snoops, &old_snoops)) { | |
1367 | ofproto_set_snoops(br->ofproto, &snoops); | |
1368 | } | |
1369 | svec_destroy(&snoops); | |
1370 | svec_destroy(&old_snoops); | |
1371 | #endif | |
1372 | ||
1373 | #if 0 | |
064af421 | 1374 | mirror_reconfigure(br); |
76343538 | 1375 | #endif |
064af421 BP |
1376 | } |
1377 | ||
1378 | static void | |
76343538 BP |
1379 | bridge_reconfigure_controller(const struct ovsrec_open_vswitch *ovs_cfg, |
1380 | struct bridge *br) | |
064af421 BP |
1381 | { |
1382 | char *pfx = xasprintf("bridge.%s.controller", br->name); | |
76343538 | 1383 | const struct ovsrec_controller *c; |
064af421 | 1384 | |
76343538 BP |
1385 | c = bridge_get_controller(ovs_cfg, br); |
1386 | if ((br->controller != NULL) != (c != NULL)) { | |
064af421 BP |
1387 | ofproto_flush_flows(br->ofproto); |
1388 | } | |
1389 | free(br->controller); | |
76343538 | 1390 | br->controller = c ? xstrdup(c->target) : NULL; |
064af421 | 1391 | |
76343538 | 1392 | if (c) { |
064af421 BP |
1393 | int max_backoff, probe; |
1394 | int rate_limit, burst_limit; | |
1395 | ||
76343538 | 1396 | if (!strcmp(c->target, "discover")) { |
064af421 | 1397 | ofproto_set_discovery(br->ofproto, true, |
76343538 BP |
1398 | c->discover_accept_regex, |
1399 | c->discover_update_resolv_conf); | |
064af421 | 1400 | } else { |
07c318f4 | 1401 | struct iface *local_iface; |
76343538 | 1402 | struct in_addr ip; |
064af421 | 1403 | bool in_band; |
064af421 | 1404 | |
76343538 BP |
1405 | in_band = (!c->connection_mode |
1406 | || !strcmp(c->connection_mode, "out-of-band")); | |
064af421 BP |
1407 | ofproto_set_discovery(br->ofproto, false, NULL, NULL); |
1408 | ofproto_set_in_band(br->ofproto, in_band); | |
1409 | ||
07c318f4 | 1410 | local_iface = bridge_get_local_iface(br); |
76343538 | 1411 | if (local_iface && c->local_ip && inet_aton(c->local_ip, &ip)) { |
07c318f4 BP |
1412 | struct netdev *netdev = local_iface->netdev; |
1413 | struct in_addr ip, mask, gateway; | |
76343538 BP |
1414 | |
1415 | if (!c->local_netmask || !inet_aton(c->local_netmask, &mask)) { | |
1416 | mask.s_addr = 0; | |
1417 | } | |
1418 | if (!c->local_gateway | |
1419 | || !inet_aton(c->local_gateway, &gateway)) { | |
1420 | gateway.s_addr = 0; | |
1421 | } | |
07c318f4 BP |
1422 | |
1423 | netdev_turn_flags_on(netdev, NETDEV_UP, true); | |
1424 | if (!mask.s_addr) { | |
1425 | mask.s_addr = guess_netmask(ip.s_addr); | |
1426 | } | |
1427 | if (!netdev_set_in4(netdev, ip, mask)) { | |
1428 | VLOG_INFO("bridge %s: configured IP address "IP_FMT", " | |
1429 | "netmask "IP_FMT, | |
1430 | br->name, IP_ARGS(&ip.s_addr), | |
1431 | IP_ARGS(&mask.s_addr)); | |
1432 | } | |
064af421 | 1433 | |
07c318f4 BP |
1434 | if (gateway.s_addr) { |
1435 | if (!netdev_add_router(netdev, gateway)) { | |
1436 | VLOG_INFO("bridge %s: configured gateway "IP_FMT, | |
1437 | br->name, IP_ARGS(&gateway.s_addr)); | |
064af421 BP |
1438 | } |
1439 | } | |
064af421 BP |
1440 | } |
1441 | } | |
1442 | ||
064af421 | 1443 | ofproto_set_failure(br->ofproto, |
76343538 BP |
1444 | (!c->fail_mode |
1445 | || !strcmp(c->fail_mode, "standalone") | |
1446 | || !strcmp(c->fail_mode, "open"))); | |
1447 | ||
1448 | probe = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; | |
952efc48 | 1449 | ofproto_set_probe_interval(br->ofproto, probe); |
064af421 | 1450 | |
76343538 | 1451 | max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; |
064af421 BP |
1452 | ofproto_set_max_backoff(br->ofproto, max_backoff); |
1453 | ||
76343538 BP |
1454 | rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; |
1455 | burst_limit = c->controller_burst_limit ? *c->controller_burst_limit : 0; | |
064af421 BP |
1456 | ofproto_set_rate_limit(br->ofproto, rate_limit, burst_limit); |
1457 | ||
76343538 | 1458 | ofproto_set_remote_execution(br->ofproto, NULL, NULL); /* XXX */ |
064af421 BP |
1459 | } else { |
1460 | union ofp_action action; | |
1461 | flow_t flow; | |
1462 | ||
1463 | /* Set up a flow that matches every packet and directs them to | |
1464 | * OFPP_NORMAL (which goes to us). */ | |
1465 | memset(&action, 0, sizeof action); | |
1466 | action.type = htons(OFPAT_OUTPUT); | |
1467 | action.output.len = htons(sizeof action); | |
1468 | action.output.port = htons(OFPP_NORMAL); | |
1469 | memset(&flow, 0, sizeof flow); | |
1470 | ofproto_add_flow(br->ofproto, &flow, OFPFW_ALL, 0, | |
1471 | &action, 1, 0); | |
1472 | ||
1473 | ofproto_set_in_band(br->ofproto, false); | |
1474 | ofproto_set_max_backoff(br->ofproto, 1); | |
1475 | ofproto_set_probe_interval(br->ofproto, 5); | |
1476 | ofproto_set_failure(br->ofproto, false); | |
064af421 BP |
1477 | } |
1478 | free(pfx); | |
1479 | ||
1480 | ofproto_set_controller(br->ofproto, br->controller); | |
1481 | } | |
1482 | ||
1483 | static void | |
76343538 | 1484 | bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) |
064af421 BP |
1485 | { |
1486 | size_t i, j; | |
1487 | ||
76343538 | 1488 | shash_init(ifaces); |
064af421 BP |
1489 | for (i = 0; i < br->n_ports; i++) { |
1490 | struct port *port = br->ports[i]; | |
1491 | for (j = 0; j < port->n_ifaces; j++) { | |
1492 | struct iface *iface = port->ifaces[j]; | |
76343538 | 1493 | shash_add_once(ifaces, iface->name, iface); |
064af421 | 1494 | } |
76343538 BP |
1495 | if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { |
1496 | shash_add_once(ifaces, port->name, NULL); | |
35c979bf | 1497 | } |
064af421 | 1498 | } |
064af421 BP |
1499 | } |
1500 | ||
1501 | /* For robustness, in case the administrator moves around datapath ports behind | |
1502 | * our back, we re-check all the datapath port numbers here. | |
1503 | * | |
1504 | * This function will set the 'dp_ifidx' members of interfaces that have | |
1505 | * disappeared to -1, so only call this function from a context where those | |
1506 | * 'struct iface's will be removed from the bridge. Otherwise, the -1 | |
1507 | * 'dp_ifidx'es will cause trouble later when we try to send them to the | |
1508 | * datapath, which doesn't support UINT16_MAX+1 ports. */ | |
1509 | static void | |
1510 | bridge_fetch_dp_ifaces(struct bridge *br) | |
1511 | { | |
1512 | struct odp_port *dpif_ports; | |
1513 | size_t n_dpif_ports; | |
1514 | size_t i, j; | |
1515 | ||
1516 | /* Reset all interface numbers. */ | |
1517 | for (i = 0; i < br->n_ports; i++) { | |
1518 | struct port *port = br->ports[i]; | |
1519 | for (j = 0; j < port->n_ifaces; j++) { | |
1520 | struct iface *iface = port->ifaces[j]; | |
1521 | iface->dp_ifidx = -1; | |
1522 | } | |
1523 | } | |
1524 | port_array_clear(&br->ifaces); | |
1525 | ||
c228a364 | 1526 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
064af421 BP |
1527 | for (i = 0; i < n_dpif_ports; i++) { |
1528 | struct odp_port *p = &dpif_ports[i]; | |
1529 | struct iface *iface = iface_lookup(br, p->devname); | |
1530 | if (iface) { | |
1531 | if (iface->dp_ifidx >= 0) { | |
b29ba128 | 1532 | VLOG_WARN("%s reported interface %s twice", |
c228a364 | 1533 | dpif_name(br->dpif), p->devname); |
064af421 | 1534 | } else if (iface_from_dp_ifidx(br, p->port)) { |
b29ba128 | 1535 | VLOG_WARN("%s reported interface %"PRIu16" twice", |
c228a364 | 1536 | dpif_name(br->dpif), p->port); |
064af421 BP |
1537 | } else { |
1538 | port_array_set(&br->ifaces, p->port, iface); | |
1539 | iface->dp_ifidx = p->port; | |
1540 | } | |
093e47f4 BP |
1541 | |
1542 | if (iface->cfg) { | |
1543 | int64_t ofport = (iface->dp_ifidx >= 0 | |
1544 | ? odp_port_to_ofp_port(iface->dp_ifidx) | |
1545 | : -1); | |
1546 | ovsrec_interface_set_ofport(iface->cfg, &ofport, 1); | |
1547 | } | |
064af421 BP |
1548 | } |
1549 | } | |
1550 | free(dpif_ports); | |
1551 | } | |
1552 | \f | |
1553 | /* Bridge packet processing functions. */ | |
1554 | ||
da285df4 BP |
1555 | static int |
1556 | bond_hash(const uint8_t mac[ETH_ADDR_LEN]) | |
1557 | { | |
1558 | return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; | |
1559 | } | |
1560 | ||
064af421 BP |
1561 | static struct bond_entry * |
1562 | lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) | |
1563 | { | |
da285df4 | 1564 | return &port->bond_hash[bond_hash(mac)]; |
064af421 BP |
1565 | } |
1566 | ||
1567 | static int | |
1568 | bond_choose_iface(const struct port *port) | |
1569 | { | |
8b2a2f4a JG |
1570 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
1571 | size_t i, best_down_slave = -1; | |
1572 | long long next_delay_expiration = LLONG_MAX; | |
1573 | ||
064af421 | 1574 | for (i = 0; i < port->n_ifaces; i++) { |
8b2a2f4a JG |
1575 | struct iface *iface = port->ifaces[i]; |
1576 | ||
1577 | if (iface->enabled) { | |
064af421 | 1578 | return i; |
8b2a2f4a JG |
1579 | } else if (iface->delay_expires < next_delay_expiration) { |
1580 | best_down_slave = i; | |
1581 | next_delay_expiration = iface->delay_expires; | |
064af421 BP |
1582 | } |
1583 | } | |
8b2a2f4a JG |
1584 | |
1585 | if (best_down_slave != -1) { | |
1586 | struct iface *iface = port->ifaces[best_down_slave]; | |
1587 | ||
1588 | VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " | |
1589 | "since no other interface is up", iface->name, | |
1590 | iface->delay_expires - time_msec()); | |
1591 | bond_enable_slave(iface, true); | |
1592 | } | |
1593 | ||
1594 | return best_down_slave; | |
064af421 BP |
1595 | } |
1596 | ||
1597 | static bool | |
2303f3b2 | 1598 | choose_output_iface(const struct port *port, const uint8_t *dl_src, |
064af421 BP |
1599 | uint16_t *dp_ifidx, tag_type *tags) |
1600 | { | |
1601 | struct iface *iface; | |
1602 | ||
1603 | assert(port->n_ifaces); | |
1604 | if (port->n_ifaces == 1) { | |
1605 | iface = port->ifaces[0]; | |
1606 | } else { | |
2303f3b2 | 1607 | struct bond_entry *e = lookup_bond_entry(port, dl_src); |
064af421 BP |
1608 | if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces |
1609 | || !port->ifaces[e->iface_idx]->enabled) { | |
1610 | /* XXX select interface properly. The current interface selection | |
1611 | * is only good for testing the rebalancing code. */ | |
1612 | e->iface_idx = bond_choose_iface(port); | |
1613 | if (e->iface_idx < 0) { | |
1614 | *tags |= port->no_ifaces_tag; | |
1615 | return false; | |
1616 | } | |
1617 | e->iface_tag = tag_create_random(); | |
85c74638 | 1618 | ((struct port *) port)->bond_compat_is_stale = true; |
064af421 BP |
1619 | } |
1620 | *tags |= e->iface_tag; | |
1621 | iface = port->ifaces[e->iface_idx]; | |
1622 | } | |
1623 | *dp_ifidx = iface->dp_ifidx; | |
1624 | *tags |= iface->tag; /* Currently only used for bonding. */ | |
1625 | return true; | |
1626 | } | |
1627 | ||
1628 | static void | |
1629 | bond_link_status_update(struct iface *iface, bool carrier) | |
1630 | { | |
1631 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1632 | struct port *port = iface->port; | |
1633 | ||
1634 | if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { | |
1635 | /* Nothing to do. */ | |
1636 | return; | |
1637 | } | |
1638 | VLOG_INFO_RL(&rl, "interface %s: carrier %s", | |
1639 | iface->name, carrier ? "detected" : "dropped"); | |
1640 | if (carrier == iface->enabled) { | |
1641 | iface->delay_expires = LLONG_MAX; | |
1642 | VLOG_INFO_RL(&rl, "interface %s: will not be %s", | |
1643 | iface->name, carrier ? "disabled" : "enabled"); | |
8b2a2f4a JG |
1644 | } else if (carrier && port->active_iface < 0) { |
1645 | bond_enable_slave(iface, true); | |
1646 | if (port->updelay) { | |
1647 | VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " | |
1648 | "other interface is up", iface->name, port->updelay); | |
1649 | } | |
064af421 BP |
1650 | } else { |
1651 | int delay = carrier ? port->updelay : port->downdelay; | |
1652 | iface->delay_expires = time_msec() + delay; | |
1653 | if (delay) { | |
1654 | VLOG_INFO_RL(&rl, | |
1655 | "interface %s: will be %s if it stays %s for %d ms", | |
1656 | iface->name, | |
1657 | carrier ? "enabled" : "disabled", | |
1658 | carrier ? "up" : "down", | |
1659 | delay); | |
1660 | } | |
1661 | } | |
1662 | } | |
1663 | ||
1664 | static void | |
1665 | bond_choose_active_iface(struct port *port) | |
1666 | { | |
1667 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1668 | ||
1669 | port->active_iface = bond_choose_iface(port); | |
1670 | port->active_iface_tag = tag_create_random(); | |
1671 | if (port->active_iface >= 0) { | |
1672 | VLOG_INFO_RL(&rl, "port %s: active interface is now %s", | |
1673 | port->name, port->ifaces[port->active_iface]->name); | |
1674 | } else { | |
1675 | VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", | |
1676 | port->name); | |
1677 | } | |
1678 | } | |
1679 | ||
da285df4 BP |
1680 | static void |
1681 | bond_enable_slave(struct iface *iface, bool enable) | |
1682 | { | |
1683 | struct port *port = iface->port; | |
1684 | struct bridge *br = port->bridge; | |
1685 | ||
8b2a2f4a JG |
1686 | /* This acts as a recursion check. If the act of disabling a slave |
1687 | * causes a different slave to be enabled, the flag will allow us to | |
1688 | * skip redundant work when we reenter this function. It must be | |
1689 | * cleared on exit to keep things safe with multiple bonds. */ | |
1690 | static bool moving_active_iface = false; | |
1691 | ||
da285df4 BP |
1692 | iface->delay_expires = LLONG_MAX; |
1693 | if (enable == iface->enabled) { | |
1694 | return; | |
1695 | } | |
1696 | ||
1697 | iface->enabled = enable; | |
1698 | if (!iface->enabled) { | |
17ea75b2 | 1699 | VLOG_WARN("interface %s: disabled", iface->name); |
da285df4 BP |
1700 | ofproto_revalidate(br->ofproto, iface->tag); |
1701 | if (iface->port_ifidx == port->active_iface) { | |
1702 | ofproto_revalidate(br->ofproto, | |
1703 | port->active_iface_tag); | |
8b2a2f4a JG |
1704 | |
1705 | /* Disabling a slave can lead to another slave being immediately | |
1706 | * enabled if there will be no active slaves but one is waiting | |
1707 | * on an updelay. In this case we do not need to run most of the | |
1708 | * code for the newly enabled slave since there was no period | |
1709 | * without an active slave and it is redundant with the disabling | |
1710 | * path. */ | |
1711 | moving_active_iface = true; | |
da285df4 BP |
1712 | bond_choose_active_iface(port); |
1713 | } | |
1714 | bond_send_learning_packets(port); | |
1715 | } else { | |
17ea75b2 | 1716 | VLOG_WARN("interface %s: enabled", iface->name); |
8b2a2f4a | 1717 | if (port->active_iface < 0 && !moving_active_iface) { |
da285df4 BP |
1718 | ofproto_revalidate(br->ofproto, port->no_ifaces_tag); |
1719 | bond_choose_active_iface(port); | |
1720 | bond_send_learning_packets(port); | |
1721 | } | |
1722 | iface->tag = tag_create_random(); | |
1723 | } | |
8b2a2f4a JG |
1724 | |
1725 | moving_active_iface = false; | |
1726 | port->bond_compat_is_stale = true; | |
da285df4 BP |
1727 | } |
1728 | ||
064af421 BP |
1729 | static void |
1730 | bond_run(struct bridge *br) | |
1731 | { | |
1732 | size_t i, j; | |
1733 | ||
1734 | for (i = 0; i < br->n_ports; i++) { | |
1735 | struct port *port = br->ports[i]; | |
85c74638 | 1736 | |
8b2a2f4a JG |
1737 | if (port->n_ifaces >= 2) { |
1738 | for (j = 0; j < port->n_ifaces; j++) { | |
1739 | struct iface *iface = port->ifaces[j]; | |
1740 | if (time_msec() >= iface->delay_expires) { | |
1741 | bond_enable_slave(iface, !iface->enabled); | |
1742 | } | |
1743 | } | |
1744 | } | |
1745 | ||
85c74638 BP |
1746 | if (port->bond_compat_is_stale) { |
1747 | port->bond_compat_is_stale = false; | |
1748 | port_update_bond_compat(port); | |
1749 | } | |
064af421 BP |
1750 | } |
1751 | } | |
1752 | ||
1753 | static void | |
1754 | bond_wait(struct bridge *br) | |
1755 | { | |
1756 | size_t i, j; | |
1757 | ||
1758 | for (i = 0; i < br->n_ports; i++) { | |
1759 | struct port *port = br->ports[i]; | |
1760 | if (port->n_ifaces < 2) { | |
1761 | continue; | |
1762 | } | |
1763 | for (j = 0; j < port->n_ifaces; j++) { | |
1764 | struct iface *iface = port->ifaces[j]; | |
1765 | if (iface->delay_expires != LLONG_MAX) { | |
1766 | poll_timer_wait(iface->delay_expires - time_msec()); | |
1767 | } | |
1768 | } | |
1769 | } | |
1770 | } | |
1771 | ||
1772 | static bool | |
1773 | set_dst(struct dst *p, const flow_t *flow, | |
1774 | const struct port *in_port, const struct port *out_port, | |
1775 | tag_type *tags) | |
1776 | { | |
064af421 BP |
1777 | p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE |
1778 | : in_port->vlan >= 0 ? in_port->vlan | |
1779 | : ntohs(flow->dl_vlan)); | |
2303f3b2 | 1780 | return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); |
064af421 BP |
1781 | } |
1782 | ||
1783 | static void | |
1784 | swap_dst(struct dst *p, struct dst *q) | |
1785 | { | |
1786 | struct dst tmp = *p; | |
1787 | *p = *q; | |
1788 | *q = tmp; | |
1789 | } | |
1790 | ||
1791 | /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in | |
1792 | * 'dsts'. (This may help performance by reducing the number of VLAN changes | |
1793 | * that we push to the datapath. We could in fact fully sort the array by | |
1794 | * vlan, but in most cases there are at most two different vlan tags so that's | |
1795 | * possibly overkill.) */ | |
1796 | static void | |
1797 | partition_dsts(struct dst *dsts, size_t n_dsts, int vlan) | |
1798 | { | |
1799 | struct dst *first = dsts; | |
1800 | struct dst *last = dsts + n_dsts; | |
1801 | ||
1802 | while (first != last) { | |
1803 | /* Invariants: | |
1804 | * - All dsts < first have vlan == 'vlan'. | |
1805 | * - All dsts >= last have vlan != 'vlan'. | |
1806 | * - first < last. */ | |
1807 | while (first->vlan == vlan) { | |
1808 | if (++first == last) { | |
1809 | return; | |
1810 | } | |
1811 | } | |
1812 | ||
1813 | /* Same invariants, plus one additional: | |
1814 | * - first->vlan != vlan. | |
1815 | */ | |
1816 | while (last[-1].vlan != vlan) { | |
1817 | if (--last == first) { | |
1818 | return; | |
1819 | } | |
1820 | } | |
1821 | ||
1822 | /* Same invariants, plus one additional: | |
1823 | * - last[-1].vlan == vlan.*/ | |
1824 | swap_dst(first++, --last); | |
1825 | } | |
1826 | } | |
1827 | ||
1828 | static int | |
1829 | mirror_mask_ffs(mirror_mask_t mask) | |
1830 | { | |
1831 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
1832 | return ffs(mask); | |
1833 | } | |
1834 | ||
1835 | static bool | |
1836 | dst_is_duplicate(const struct dst *dsts, size_t n_dsts, | |
1837 | const struct dst *test) | |
1838 | { | |
1839 | size_t i; | |
1840 | for (i = 0; i < n_dsts; i++) { | |
1841 | if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) { | |
1842 | return true; | |
1843 | } | |
1844 | } | |
1845 | return false; | |
1846 | } | |
1847 | ||
1848 | static bool | |
1849 | port_trunks_vlan(const struct port *port, uint16_t vlan) | |
1850 | { | |
1851 | return port->vlan < 0 && bitmap_is_set(port->trunks, vlan); | |
1852 | } | |
1853 | ||
1854 | static bool | |
1855 | port_includes_vlan(const struct port *port, uint16_t vlan) | |
1856 | { | |
1857 | return vlan == port->vlan || port_trunks_vlan(port, vlan); | |
1858 | } | |
1859 | ||
1860 | static size_t | |
1861 | compose_dsts(const struct bridge *br, const flow_t *flow, uint16_t vlan, | |
1862 | const struct port *in_port, const struct port *out_port, | |
6a07af36 | 1863 | struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) |
064af421 BP |
1864 | { |
1865 | mirror_mask_t mirrors = in_port->src_mirrors; | |
1866 | struct dst *dst = dsts; | |
1867 | size_t i; | |
1868 | ||
064af421 BP |
1869 | if (out_port == FLOOD_PORT) { |
1870 | /* XXX use ODP_FLOOD if no vlans or bonding. */ | |
1871 | /* XXX even better, define each VLAN as a datapath port group */ | |
1872 | for (i = 0; i < br->n_ports; i++) { | |
1873 | struct port *port = br->ports[i]; | |
1874 | if (port != in_port && port_includes_vlan(port, vlan) | |
1875 | && !port->is_mirror_output_port | |
1876 | && set_dst(dst, flow, in_port, port, tags)) { | |
1877 | mirrors |= port->dst_mirrors; | |
1878 | dst++; | |
1879 | } | |
1880 | } | |
6a07af36 | 1881 | *nf_output_iface = NF_OUT_FLOOD; |
064af421 | 1882 | } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { |
6a07af36 | 1883 | *nf_output_iface = dst->dp_ifidx; |
064af421 BP |
1884 | mirrors |= out_port->dst_mirrors; |
1885 | dst++; | |
1886 | } | |
1887 | ||
1888 | while (mirrors) { | |
1889 | struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
1890 | if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { | |
1891 | if (m->out_port) { | |
1892 | if (set_dst(dst, flow, in_port, m->out_port, tags) | |
1893 | && !dst_is_duplicate(dsts, dst - dsts, dst)) { | |
1894 | dst++; | |
1895 | } | |
1896 | } else { | |
1897 | for (i = 0; i < br->n_ports; i++) { | |
1898 | struct port *port = br->ports[i]; | |
1899 | if (port_includes_vlan(port, m->out_vlan) | |
274de4d2 | 1900 | && set_dst(dst, flow, in_port, port, tags)) |
064af421 | 1901 | { |
43aa5f47 JG |
1902 | int flow_vlan; |
1903 | ||
064af421 BP |
1904 | if (port->vlan < 0) { |
1905 | dst->vlan = m->out_vlan; | |
1906 | } | |
274de4d2 BP |
1907 | if (dst_is_duplicate(dsts, dst - dsts, dst)) { |
1908 | continue; | |
1909 | } | |
43aa5f47 JG |
1910 | |
1911 | /* Use the vlan tag on the original flow instead of | |
1912 | * the one passed in the vlan parameter. This ensures | |
1913 | * that we compare the vlan from before any implicit | |
1914 | * tagging tags place. This is necessary because | |
1915 | * dst->vlan is the final vlan, after removing implicit | |
1916 | * tags. */ | |
1917 | flow_vlan = ntohs(flow->dl_vlan); | |
1918 | if (flow_vlan == 0) { | |
1919 | flow_vlan = OFP_VLAN_NONE; | |
1920 | } | |
1921 | if (port == in_port && dst->vlan == flow_vlan) { | |
064af421 BP |
1922 | /* Don't send out input port on same VLAN. */ |
1923 | continue; | |
1924 | } | |
1925 | dst++; | |
1926 | } | |
1927 | } | |
1928 | } | |
1929 | } | |
1930 | mirrors &= mirrors - 1; | |
1931 | } | |
1932 | ||
1933 | partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); | |
1934 | return dst - dsts; | |
1935 | } | |
1936 | ||
1937 | static void UNUSED | |
1938 | print_dsts(const struct dst *dsts, size_t n) | |
1939 | { | |
1940 | for (; n--; dsts++) { | |
1941 | printf(">p%"PRIu16, dsts->dp_ifidx); | |
1942 | if (dsts->vlan != OFP_VLAN_NONE) { | |
1943 | printf("v%"PRIu16, dsts->vlan); | |
1944 | } | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | static void | |
1949 | compose_actions(struct bridge *br, const flow_t *flow, uint16_t vlan, | |
1950 | const struct port *in_port, const struct port *out_port, | |
6a07af36 JG |
1951 | tag_type *tags, struct odp_actions *actions, |
1952 | uint16_t *nf_output_iface) | |
064af421 BP |
1953 | { |
1954 | struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; | |
1955 | size_t n_dsts; | |
1956 | const struct dst *p; | |
1957 | uint16_t cur_vlan; | |
1958 | ||
6a07af36 JG |
1959 | n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, |
1960 | nf_output_iface); | |
064af421 BP |
1961 | |
1962 | cur_vlan = ntohs(flow->dl_vlan); | |
1963 | for (p = dsts; p < &dsts[n_dsts]; p++) { | |
1964 | union odp_action *a; | |
1965 | if (p->vlan != cur_vlan) { | |
1966 | if (p->vlan == OFP_VLAN_NONE) { | |
1967 | odp_actions_add(actions, ODPAT_STRIP_VLAN); | |
1968 | } else { | |
1969 | a = odp_actions_add(actions, ODPAT_SET_VLAN_VID); | |
1970 | a->vlan_vid.vlan_vid = htons(p->vlan); | |
1971 | } | |
1972 | cur_vlan = p->vlan; | |
1973 | } | |
1974 | a = odp_actions_add(actions, ODPAT_OUTPUT); | |
1975 | a->output.port = p->dp_ifidx; | |
1976 | } | |
1977 | } | |
1978 | ||
e96a4d80 JG |
1979 | /* Returns the effective vlan of a packet, taking into account both the |
1980 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
1981 | * the packet is untagged and -1 indicates it has an invalid header and | |
1982 | * should be dropped. */ | |
1983 | static int flow_get_vlan(struct bridge *br, const flow_t *flow, | |
1984 | struct port *in_port, bool have_packet) | |
1985 | { | |
1986 | /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet | |
1987 | * belongs to VLAN 0, so we should treat both cases identically. (In the | |
1988 | * former case, the packet has an 802.1Q header that specifies VLAN 0, | |
1989 | * presumably to allow a priority to be specified. In the latter case, the | |
1990 | * packet does not have any 802.1Q header.) */ | |
1991 | int vlan = ntohs(flow->dl_vlan); | |
1992 | if (vlan == OFP_VLAN_NONE) { | |
1993 | vlan = 0; | |
1994 | } | |
1995 | if (in_port->vlan >= 0) { | |
1996 | if (vlan) { | |
1997 | /* XXX support double tagging? */ | |
1998 | if (have_packet) { | |
1999 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2000 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
2001 | "packet received on port %s configured with " | |
2002 | "implicit VLAN %"PRIu16, | |
2003 | br->name, ntohs(flow->dl_vlan), | |
2004 | in_port->name, in_port->vlan); | |
2005 | } | |
2006 | return -1; | |
2007 | } | |
2008 | vlan = in_port->vlan; | |
2009 | } else { | |
2010 | if (!port_includes_vlan(in_port, vlan)) { | |
2011 | if (have_packet) { | |
2012 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2013 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
2014 | "packet received on port %s not configured for " | |
2015 | "trunking VLAN %d", | |
2016 | br->name, vlan, in_port->name, vlan); | |
2017 | } | |
2018 | return -1; | |
2019 | } | |
2020 | } | |
2021 | ||
2022 | return vlan; | |
2023 | } | |
2024 | ||
2025 | static void | |
2026 | update_learning_table(struct bridge *br, const flow_t *flow, int vlan, | |
2027 | struct port *in_port) | |
2028 | { | |
2029 | tag_type rev_tag = mac_learning_learn(br->ml, flow->dl_src, | |
2030 | vlan, in_port->port_idx); | |
2031 | if (rev_tag) { | |
2032 | /* The log messages here could actually be useful in debugging, | |
2033 | * so keep the rate limit relatively high. */ | |
2034 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, | |
2035 | 300); | |
2036 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
2037 | "on port %s in VLAN %d", | |
2038 | br->name, ETH_ADDR_ARGS(flow->dl_src), | |
2039 | in_port->name, vlan); | |
2040 | ofproto_revalidate(br->ofproto, rev_tag); | |
2041 | } | |
2042 | } | |
2043 | ||
064af421 | 2044 | static bool |
69d60f9f | 2045 | is_bcast_arp_reply(const flow_t *flow) |
064af421 | 2046 | { |
064af421 | 2047 | return (flow->dl_type == htons(ETH_TYPE_ARP) |
69d60f9f JG |
2048 | && flow->nw_proto == ARP_OP_REPLY |
2049 | && eth_addr_is_broadcast(flow->dl_dst)); | |
064af421 BP |
2050 | } |
2051 | ||
2052 | /* If the composed actions may be applied to any packet in the given 'flow', | |
2053 | * returns true. Otherwise, the actions should only be applied to 'packet', or | |
2054 | * not at all, if 'packet' was NULL. */ | |
2055 | static bool | |
2056 | process_flow(struct bridge *br, const flow_t *flow, | |
2057 | const struct ofpbuf *packet, struct odp_actions *actions, | |
6a07af36 | 2058 | tag_type *tags, uint16_t *nf_output_iface) |
064af421 BP |
2059 | { |
2060 | struct iface *in_iface; | |
2061 | struct port *in_port; | |
2062 | struct port *out_port = NULL; /* By default, drop the packet/flow. */ | |
2063 | int vlan; | |
93dfc067 | 2064 | int out_port_idx; |
064af421 BP |
2065 | |
2066 | /* Find the interface and port structure for the received packet. */ | |
2067 | in_iface = iface_from_dp_ifidx(br, flow->in_port); | |
2068 | if (!in_iface) { | |
2069 | /* No interface? Something fishy... */ | |
2070 | if (packet != NULL) { | |
2071 | /* Odd. A few possible reasons here: | |
2072 | * | |
2073 | * - We deleted an interface but there are still a few packets | |
2074 | * queued up from it. | |
2075 | * | |
2076 | * - Someone externally added an interface (e.g. with "ovs-dpctl | |
2077 | * add-if") that we don't know about. | |
2078 | * | |
2079 | * - Packet arrived on the local port but the local port is not | |
2080 | * one of our bridge ports. | |
2081 | */ | |
2082 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2083 | ||
2084 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
2085 | "interface %"PRIu16, br->name, flow->in_port); | |
2086 | } | |
2087 | ||
2088 | /* Return without adding any actions, to drop packets on this flow. */ | |
2089 | return true; | |
2090 | } | |
2091 | in_port = in_iface->port; | |
e96a4d80 JG |
2092 | vlan = flow_get_vlan(br, flow, in_port, !!packet); |
2093 | if (vlan < 0) { | |
2094 | goto done; | |
064af421 BP |
2095 | } |
2096 | ||
064af421 BP |
2097 | /* Drop frames for reserved multicast addresses. */ |
2098 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
2099 | goto done; | |
2100 | } | |
2101 | ||
2102 | /* Drop frames on ports reserved for mirroring. */ | |
2103 | if (in_port->is_mirror_output_port) { | |
2104 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2105 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port %s, " | |
2106 | "which is reserved exclusively for mirroring", | |
2107 | br->name, in_port->name); | |
2108 | goto done; | |
2109 | } | |
2110 | ||
3a55ef14 JG |
2111 | /* Packets received on bonds need special attention to avoid duplicates. */ |
2112 | if (in_port->n_ifaces > 1) { | |
2113 | int src_idx; | |
2114 | ||
2115 | if (eth_addr_is_multicast(flow->dl_dst)) { | |
2116 | *tags |= in_port->active_iface_tag; | |
2117 | if (in_port->active_iface != in_iface->port_ifidx) { | |
2118 | /* Drop all multicast packets on inactive slaves. */ | |
2119 | goto done; | |
c93b1d6a | 2120 | } |
064af421 | 2121 | } |
3a55ef14 JG |
2122 | |
2123 | /* Drop all packets for which we have learned a different input | |
2124 | * port, because we probably sent the packet on one slave and got | |
2125 | * it back on the other. Broadcast ARP replies are an exception | |
2126 | * to this rule: the host has moved to another switch. */ | |
2127 | src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan); | |
2128 | if (src_idx != -1 && src_idx != in_port->port_idx && | |
69d60f9f | 2129 | !is_bcast_arp_reply(flow)) { |
3a55ef14 JG |
2130 | goto done; |
2131 | } | |
064af421 BP |
2132 | } |
2133 | ||
2134 | /* MAC learning. */ | |
2135 | out_port = FLOOD_PORT; | |
93dfc067 JG |
2136 | /* Learn source MAC (but don't try to learn from revalidation). */ |
2137 | if (packet) { | |
e96a4d80 | 2138 | update_learning_table(br, flow, vlan, in_port); |
93dfc067 JG |
2139 | } |
2140 | ||
2141 | /* Determine output port. */ | |
2142 | out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, | |
2143 | tags); | |
2144 | if (out_port_idx >= 0 && out_port_idx < br->n_ports) { | |
2145 | out_port = br->ports[out_port_idx]; | |
e96a4d80 | 2146 | } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { |
93dfc067 | 2147 | /* If we are revalidating but don't have a learning entry then |
e96a4d80 JG |
2148 | * eject the flow. Installing a flow that floods packets opens |
2149 | * up a window of time where we could learn from a packet reflected | |
2150 | * on a bond and blackhole packets before the learning table is | |
2151 | * updated to reflect the correct port. */ | |
93dfc067 | 2152 | return false; |
064af421 BP |
2153 | } |
2154 | ||
ba54bf4f BP |
2155 | /* Don't send packets out their input ports. */ |
2156 | if (in_port == out_port) { | |
064af421 BP |
2157 | out_port = NULL; |
2158 | } | |
2159 | ||
2160 | done: | |
6a07af36 JG |
2161 | compose_actions(br, flow, vlan, in_port, out_port, tags, actions, |
2162 | nf_output_iface); | |
064af421 | 2163 | |
69d60f9f | 2164 | return true; |
064af421 BP |
2165 | } |
2166 | ||
2167 | /* Careful: 'opp' is in host byte order and opp->port_no is an OFP port | |
2168 | * number. */ | |
2169 | static void | |
2170 | bridge_port_changed_ofhook_cb(enum ofp_port_reason reason, | |
2171 | const struct ofp_phy_port *opp, | |
2172 | void *br_) | |
2173 | { | |
2174 | struct bridge *br = br_; | |
2175 | struct iface *iface; | |
2176 | struct port *port; | |
2177 | ||
2178 | iface = iface_from_dp_ifidx(br, ofp_port_to_odp_port(opp->port_no)); | |
2179 | if (!iface) { | |
2180 | return; | |
2181 | } | |
2182 | port = iface->port; | |
2183 | ||
2184 | if (reason == OFPPR_DELETE) { | |
2185 | VLOG_WARN("bridge %s: interface %s deleted unexpectedly", | |
2186 | br->name, iface->name); | |
2187 | iface_destroy(iface); | |
2188 | if (!port->n_ifaces) { | |
2189 | VLOG_WARN("bridge %s: port %s has no interfaces, dropping", | |
2190 | br->name, port->name); | |
2191 | port_destroy(port); | |
2192 | } | |
2193 | ||
2194 | bridge_flush(br); | |
2195 | } else { | |
064af421 BP |
2196 | if (port->n_ifaces > 1) { |
2197 | bool up = !(opp->state & OFPPS_LINK_DOWN); | |
2198 | bond_link_status_update(iface, up); | |
2199 | port_update_bond_compat(port); | |
2200 | } | |
2201 | } | |
2202 | } | |
2203 | ||
2204 | static bool | |
2205 | bridge_normal_ofhook_cb(const flow_t *flow, const struct ofpbuf *packet, | |
6a07af36 JG |
2206 | struct odp_actions *actions, tag_type *tags, |
2207 | uint16_t *nf_output_iface, void *br_) | |
064af421 BP |
2208 | { |
2209 | struct bridge *br = br_; | |
2210 | ||
064af421 | 2211 | COVERAGE_INC(bridge_process_flow); |
6a07af36 | 2212 | return process_flow(br, flow, packet, actions, tags, nf_output_iface); |
064af421 BP |
2213 | } |
2214 | ||
2215 | static void | |
2216 | bridge_account_flow_ofhook_cb(const flow_t *flow, | |
2217 | const union odp_action *actions, | |
2218 | size_t n_actions, unsigned long long int n_bytes, | |
2219 | void *br_) | |
2220 | { | |
2221 | struct bridge *br = br_; | |
e96a4d80 | 2222 | struct port *in_port; |
064af421 BP |
2223 | const union odp_action *a; |
2224 | ||
e96a4d80 JG |
2225 | /* Feed information from the active flows back into the learning table |
2226 | * to ensure that table is always in sync with what is actually flowing | |
2227 | * through the datapath. */ | |
2228 | in_port = port_from_dp_ifidx(br, flow->in_port); | |
f1bd68ab JG |
2229 | if (in_port) { |
2230 | int vlan = flow_get_vlan(br, flow, in_port, false); | |
2231 | if (vlan >= 0) { | |
2232 | update_learning_table(br, flow, vlan, in_port); | |
2233 | } | |
e96a4d80 JG |
2234 | } |
2235 | ||
064af421 BP |
2236 | if (!br->has_bonded_ports) { |
2237 | return; | |
2238 | } | |
2239 | ||
2240 | for (a = actions; a < &actions[n_actions]; a++) { | |
2241 | if (a->type == ODPAT_OUTPUT) { | |
e96a4d80 JG |
2242 | struct port *out_port = port_from_dp_ifidx(br, a->output.port); |
2243 | if (out_port && out_port->n_ifaces >= 2) { | |
2244 | struct bond_entry *e = lookup_bond_entry(out_port, | |
2245 | flow->dl_src); | |
064af421 BP |
2246 | e->tx_bytes += n_bytes; |
2247 | } | |
2248 | } | |
2249 | } | |
2250 | } | |
2251 | ||
2252 | static void | |
2253 | bridge_account_checkpoint_ofhook_cb(void *br_) | |
2254 | { | |
2255 | struct bridge *br = br_; | |
2256 | size_t i; | |
2257 | ||
2258 | if (!br->has_bonded_ports) { | |
2259 | return; | |
2260 | } | |
2261 | ||
2262 | /* The current ofproto implementation calls this callback at least once a | |
2263 | * second, so this timer implementation is sufficient. */ | |
2264 | if (time_msec() < br->bond_next_rebalance) { | |
2265 | return; | |
2266 | } | |
2267 | br->bond_next_rebalance = time_msec() + 10000; | |
2268 | ||
2269 | for (i = 0; i < br->n_ports; i++) { | |
2270 | struct port *port = br->ports[i]; | |
2271 | if (port->n_ifaces > 1) { | |
2272 | bond_rebalance_port(port); | |
2273 | } | |
2274 | } | |
2275 | } | |
2276 | ||
2277 | static struct ofhooks bridge_ofhooks = { | |
2278 | bridge_port_changed_ofhook_cb, | |
2279 | bridge_normal_ofhook_cb, | |
2280 | bridge_account_flow_ofhook_cb, | |
2281 | bridge_account_checkpoint_ofhook_cb, | |
2282 | }; | |
2283 | \f | |
2303f3b2 BP |
2284 | /* Bonding functions. */ |
2285 | ||
064af421 BP |
2286 | /* Statistics for a single interface on a bonded port, used for load-based |
2287 | * bond rebalancing. */ | |
2288 | struct slave_balance { | |
2289 | struct iface *iface; /* The interface. */ | |
2290 | uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */ | |
2291 | ||
2292 | /* All the "bond_entry"s that are assigned to this interface, in order of | |
2293 | * increasing tx_bytes. */ | |
2294 | struct bond_entry **hashes; | |
2295 | size_t n_hashes; | |
2296 | }; | |
2297 | ||
2298 | /* Sorts pointers to pointers to bond_entries in ascending order by the | |
2299 | * interface to which they are assigned, and within a single interface in | |
2300 | * ascending order of bytes transmitted. */ | |
2301 | static int | |
2302 | compare_bond_entries(const void *a_, const void *b_) | |
2303 | { | |
2304 | const struct bond_entry *const *ap = a_; | |
2305 | const struct bond_entry *const *bp = b_; | |
2306 | const struct bond_entry *a = *ap; | |
2307 | const struct bond_entry *b = *bp; | |
2308 | if (a->iface_idx != b->iface_idx) { | |
2309 | return a->iface_idx > b->iface_idx ? 1 : -1; | |
2310 | } else if (a->tx_bytes != b->tx_bytes) { | |
2311 | return a->tx_bytes > b->tx_bytes ? 1 : -1; | |
2312 | } else { | |
2313 | return 0; | |
2314 | } | |
2315 | } | |
2316 | ||
2317 | /* Sorts slave_balances so that enabled ports come first, and otherwise in | |
2318 | * *descending* order by number of bytes transmitted. */ | |
2319 | static int | |
2320 | compare_slave_balance(const void *a_, const void *b_) | |
2321 | { | |
2322 | const struct slave_balance *a = a_; | |
2323 | const struct slave_balance *b = b_; | |
2324 | if (a->iface->enabled != b->iface->enabled) { | |
2325 | return a->iface->enabled ? -1 : 1; | |
2326 | } else if (a->tx_bytes != b->tx_bytes) { | |
2327 | return a->tx_bytes > b->tx_bytes ? -1 : 1; | |
2328 | } else { | |
2329 | return 0; | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | static void | |
2334 | swap_bals(struct slave_balance *a, struct slave_balance *b) | |
2335 | { | |
2336 | struct slave_balance tmp = *a; | |
2337 | *a = *b; | |
2338 | *b = tmp; | |
2339 | } | |
2340 | ||
2341 | /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order | |
2342 | * given that 'p' (and only 'p') might be in the wrong location. | |
2343 | * | |
2344 | * This function invalidates 'p', since it might now be in a different memory | |
2345 | * location. */ | |
2346 | static void | |
2347 | resort_bals(struct slave_balance *p, | |
2348 | struct slave_balance bals[], size_t n_bals) | |
2349 | { | |
2350 | if (n_bals > 1) { | |
2351 | for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { | |
2352 | swap_bals(p, p - 1); | |
2353 | } | |
2354 | for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { | |
2355 | swap_bals(p, p + 1); | |
2356 | } | |
2357 | } | |
2358 | } | |
2359 | ||
2360 | static void | |
2361 | log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) | |
2362 | { | |
2363 | if (VLOG_IS_DBG_ENABLED()) { | |
2364 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2365 | const struct slave_balance *b; | |
2366 | ||
2367 | for (b = bals; b < bals + n_bals; b++) { | |
2368 | size_t i; | |
2369 | ||
2370 | if (b > bals) { | |
2371 | ds_put_char(&ds, ','); | |
2372 | } | |
2373 | ds_put_format(&ds, " %s %"PRIu64"kB", | |
2374 | b->iface->name, b->tx_bytes / 1024); | |
2375 | ||
2376 | if (!b->iface->enabled) { | |
2377 | ds_put_cstr(&ds, " (disabled)"); | |
2378 | } | |
2379 | if (b->n_hashes > 0) { | |
2380 | ds_put_cstr(&ds, " ("); | |
2381 | for (i = 0; i < b->n_hashes; i++) { | |
2382 | const struct bond_entry *e = b->hashes[i]; | |
2383 | if (i > 0) { | |
2384 | ds_put_cstr(&ds, " + "); | |
2385 | } | |
2386 | ds_put_format(&ds, "h%td: %"PRIu64"kB", | |
2387 | e - port->bond_hash, e->tx_bytes / 1024); | |
2388 | } | |
2389 | ds_put_cstr(&ds, ")"); | |
2390 | } | |
2391 | } | |
2392 | VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); | |
2393 | ds_destroy(&ds); | |
2394 | } | |
2395 | } | |
2396 | ||
2397 | /* Shifts 'hash' from 'from' to 'to' within 'port'. */ | |
2398 | static void | |
2399 | bond_shift_load(struct slave_balance *from, struct slave_balance *to, | |
5422a9e1 | 2400 | int hash_idx) |
064af421 | 2401 | { |
5422a9e1 | 2402 | struct bond_entry *hash = from->hashes[hash_idx]; |
064af421 BP |
2403 | struct port *port = from->iface->port; |
2404 | uint64_t delta = hash->tx_bytes; | |
2405 | ||
2406 | VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " | |
2407 | "from %s to %s (now carrying %"PRIu64"kB and " | |
2408 | "%"PRIu64"kB load, respectively)", | |
2409 | port->name, delta / 1024, hash - port->bond_hash, | |
2410 | from->iface->name, to->iface->name, | |
2411 | (from->tx_bytes - delta) / 1024, | |
2412 | (to->tx_bytes + delta) / 1024); | |
2413 | ||
2414 | /* Delete element from from->hashes. | |
2415 | * | |
2416 | * We don't bother to add the element to to->hashes because not only would | |
2417 | * it require more work, the only purpose it would be to allow that hash to | |
2418 | * be migrated to another slave in this rebalancing run, and there is no | |
2419 | * point in doing that. */ | |
5422a9e1 | 2420 | if (hash_idx == 0) { |
064af421 BP |
2421 | from->hashes++; |
2422 | } else { | |
5422a9e1 JG |
2423 | memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, |
2424 | (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); | |
064af421 BP |
2425 | } |
2426 | from->n_hashes--; | |
2427 | ||
2428 | /* Shift load away from 'from' to 'to'. */ | |
2429 | from->tx_bytes -= delta; | |
2430 | to->tx_bytes += delta; | |
2431 | ||
2432 | /* Arrange for flows to be revalidated. */ | |
2433 | ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); | |
2434 | hash->iface_idx = to->iface->port_ifidx; | |
2435 | hash->iface_tag = tag_create_random(); | |
064af421 BP |
2436 | } |
2437 | ||
2438 | static void | |
2439 | bond_rebalance_port(struct port *port) | |
2440 | { | |
2441 | struct slave_balance bals[DP_MAX_PORTS]; | |
2442 | size_t n_bals; | |
2443 | struct bond_entry *hashes[BOND_MASK + 1]; | |
2444 | struct slave_balance *b, *from, *to; | |
2445 | struct bond_entry *e; | |
2446 | size_t i; | |
2447 | ||
2448 | /* Sets up 'bals' to describe each of the port's interfaces, sorted in | |
2449 | * descending order of tx_bytes, so that bals[0] represents the most | |
2450 | * heavily loaded slave and bals[n_bals - 1] represents the least heavily | |
2451 | * loaded slave. | |
2452 | * | |
2453 | * The code is a bit tricky: to avoid dynamically allocating a 'hashes' | |
2454 | * array for each slave_balance structure, we sort our local array of | |
2455 | * hashes in order by slave, so that all of the hashes for a given slave | |
2456 | * become contiguous in memory, and then we point each 'hashes' members of | |
2457 | * a slave_balance structure to the start of a contiguous group. */ | |
2458 | n_bals = port->n_ifaces; | |
2459 | for (b = bals; b < &bals[n_bals]; b++) { | |
2460 | b->iface = port->ifaces[b - bals]; | |
2461 | b->tx_bytes = 0; | |
2462 | b->hashes = NULL; | |
2463 | b->n_hashes = 0; | |
2464 | } | |
2465 | for (i = 0; i <= BOND_MASK; i++) { | |
2466 | hashes[i] = &port->bond_hash[i]; | |
2467 | } | |
2468 | qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); | |
2469 | for (i = 0; i <= BOND_MASK; i++) { | |
2470 | e = hashes[i]; | |
2471 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
2472 | b = &bals[e->iface_idx]; | |
2473 | b->tx_bytes += e->tx_bytes; | |
2474 | if (!b->hashes) { | |
2475 | b->hashes = &hashes[i]; | |
2476 | } | |
2477 | b->n_hashes++; | |
2478 | } | |
2479 | } | |
2480 | qsort(bals, n_bals, sizeof *bals, compare_slave_balance); | |
2481 | log_bals(bals, n_bals, port); | |
2482 | ||
2483 | /* Discard slaves that aren't enabled (which were sorted to the back of the | |
2484 | * array earlier). */ | |
2485 | while (!bals[n_bals - 1].iface->enabled) { | |
2486 | n_bals--; | |
2487 | if (!n_bals) { | |
2488 | return; | |
2489 | } | |
2490 | } | |
2491 | ||
2492 | /* Shift load from the most-loaded slaves to the least-loaded slaves. */ | |
2493 | to = &bals[n_bals - 1]; | |
2494 | for (from = bals; from < to; ) { | |
2495 | uint64_t overload = from->tx_bytes - to->tx_bytes; | |
2496 | if (overload < to->tx_bytes >> 5 || overload < 100000) { | |
2497 | /* The extra load on 'from' (and all less-loaded slaves), compared | |
2498 | * to that of 'to' (the least-loaded slave), is less than ~3%, or | |
2499 | * it is less than ~1Mbps. No point in rebalancing. */ | |
2500 | break; | |
2501 | } else if (from->n_hashes == 1) { | |
2502 | /* 'from' only carries a single MAC hash, so we can't shift any | |
2503 | * load away from it, even though we want to. */ | |
2504 | from++; | |
2505 | } else { | |
2506 | /* 'from' is carrying significantly more load than 'to', and that | |
2507 | * load is split across at least two different hashes. Pick a hash | |
2508 | * to migrate to 'to' (the least-loaded slave), given that doing so | |
5422a9e1 JG |
2509 | * must decrease the ratio of the load on the two slaves by at |
2510 | * least 0.1. | |
064af421 BP |
2511 | * |
2512 | * The sort order we use means that we prefer to shift away the | |
2513 | * smallest hashes instead of the biggest ones. There is little | |
2514 | * reason behind this decision; we could use the opposite sort | |
2515 | * order to shift away big hashes ahead of small ones. */ | |
2516 | size_t i; | |
5422a9e1 | 2517 | bool order_swapped; |
064af421 BP |
2518 | |
2519 | for (i = 0; i < from->n_hashes; i++) { | |
5422a9e1 | 2520 | double old_ratio, new_ratio; |
064af421 | 2521 | uint64_t delta = from->hashes[i]->tx_bytes; |
5422a9e1 JG |
2522 | |
2523 | if (delta == 0 || from->tx_bytes - delta == 0) { | |
2524 | /* Pointless move. */ | |
2525 | continue; | |
2526 | } | |
2527 | ||
2528 | order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; | |
2529 | ||
2530 | if (to->tx_bytes == 0) { | |
2531 | /* Nothing on the new slave, move it. */ | |
2532 | break; | |
2533 | } | |
2534 | ||
2535 | old_ratio = (double)from->tx_bytes / to->tx_bytes; | |
2536 | new_ratio = (double)(from->tx_bytes - delta) / | |
2537 | (to->tx_bytes + delta); | |
2538 | ||
2539 | if (new_ratio == 0) { | |
2540 | /* Should already be covered but check to prevent division | |
2541 | * by zero. */ | |
2542 | continue; | |
2543 | } | |
2544 | ||
2545 | if (new_ratio < 1) { | |
2546 | new_ratio = 1 / new_ratio; | |
2547 | } | |
2548 | ||
2549 | if (old_ratio - new_ratio > 0.1) { | |
2550 | /* Would decrease the ratio, move it. */ | |
064af421 BP |
2551 | break; |
2552 | } | |
2553 | } | |
2554 | if (i < from->n_hashes) { | |
5422a9e1 JG |
2555 | bond_shift_load(from, to, i); |
2556 | port->bond_compat_is_stale = true; | |
2557 | ||
2558 | /* If the result of the migration changed the relative order of | |
2559 | * 'from' and 'to' swap them back to maintain invariants. */ | |
2560 | if (order_swapped) { | |
2561 | swap_bals(from, to); | |
2562 | } | |
064af421 BP |
2563 | |
2564 | /* Re-sort 'bals'. Note that this may make 'from' and 'to' | |
2565 | * point to different slave_balance structures. It is only | |
2566 | * valid to do these two operations in a row at all because we | |
2567 | * know that 'from' will not move past 'to' and vice versa. */ | |
2568 | resort_bals(from, bals, n_bals); | |
2569 | resort_bals(to, bals, n_bals); | |
2570 | } else { | |
2571 | from++; | |
2572 | } | |
2573 | } | |
2574 | } | |
2575 | ||
2576 | /* Implement exponentially weighted moving average. A weight of 1/2 causes | |
2577 | * historical data to decay to <1% in 7 rebalancing runs. */ | |
2578 | for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { | |
2579 | e->tx_bytes /= 2; | |
2580 | } | |
2581 | } | |
2303f3b2 BP |
2582 | |
2583 | static void | |
2584 | bond_send_learning_packets(struct port *port) | |
2585 | { | |
2586 | struct bridge *br = port->bridge; | |
2587 | struct mac_entry *e; | |
2588 | struct ofpbuf packet; | |
2589 | int error, n_packets, n_errors; | |
2590 | ||
93dfc067 | 2591 | if (!port->n_ifaces || port->active_iface < 0) { |
2303f3b2 BP |
2592 | return; |
2593 | } | |
2594 | ||
2595 | ofpbuf_init(&packet, 128); | |
2596 | error = n_packets = n_errors = 0; | |
2597 | LIST_FOR_EACH (e, struct mac_entry, lru_node, &br->ml->lrus) { | |
2303f3b2 | 2598 | union ofp_action actions[2], *a; |
2303f3b2 BP |
2599 | uint16_t dp_ifidx; |
2600 | tag_type tags = 0; | |
2601 | flow_t flow; | |
2602 | int retval; | |
2603 | ||
2604 | if (e->port == port->port_idx | |
2605 | || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { | |
2606 | continue; | |
2607 | } | |
2608 | ||
2303f3b2 BP |
2609 | /* Compose actions. */ |
2610 | memset(actions, 0, sizeof actions); | |
2611 | a = actions; | |
2612 | if (e->vlan) { | |
2613 | a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); | |
2614 | a->vlan_vid.len = htons(sizeof *a); | |
2615 | a->vlan_vid.vlan_vid = htons(e->vlan); | |
2616 | a++; | |
2617 | } | |
2618 | a->output.type = htons(OFPAT_OUTPUT); | |
2619 | a->output.len = htons(sizeof *a); | |
2620 | a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); | |
2621 | a++; | |
2622 | ||
2623 | /* Send packet. */ | |
2624 | n_packets++; | |
b9e8b45a BP |
2625 | compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, |
2626 | e->mac); | |
2303f3b2 BP |
2627 | flow_extract(&packet, ODPP_NONE, &flow); |
2628 | retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, | |
2629 | &packet); | |
2630 | if (retval) { | |
2631 | error = retval; | |
2632 | n_errors++; | |
2633 | } | |
2634 | } | |
2635 | ofpbuf_uninit(&packet); | |
2636 | ||
2637 | if (n_errors) { | |
2638 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2639 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
2640 | "packets, last error was: %s", | |
2641 | port->name, n_errors, n_packets, strerror(error)); | |
2642 | } else { | |
2643 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
2644 | port->name, n_packets); | |
2645 | } | |
2646 | } | |
064af421 | 2647 | \f |
da285df4 BP |
2648 | /* Bonding unixctl user interface functions. */ |
2649 | ||
2650 | static void | |
8ca79daa BP |
2651 | bond_unixctl_list(struct unixctl_conn *conn, |
2652 | const char *args UNUSED, void *aux UNUSED) | |
da285df4 BP |
2653 | { |
2654 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2655 | const struct bridge *br; | |
2656 | ||
2657 | ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); | |
2658 | ||
2659 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
2660 | size_t i; | |
2661 | ||
2662 | for (i = 0; i < br->n_ports; i++) { | |
2663 | const struct port *port = br->ports[i]; | |
2664 | if (port->n_ifaces > 1) { | |
2665 | size_t j; | |
2666 | ||
2667 | ds_put_format(&ds, "%s\t%s\t", br->name, port->name); | |
2668 | for (j = 0; j < port->n_ifaces; j++) { | |
2669 | const struct iface *iface = port->ifaces[j]; | |
2670 | if (j) { | |
2671 | ds_put_cstr(&ds, ", "); | |
2672 | } | |
2673 | ds_put_cstr(&ds, iface->name); | |
2674 | } | |
2675 | ds_put_char(&ds, '\n'); | |
2676 | } | |
2677 | } | |
2678 | } | |
2679 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
2680 | ds_destroy(&ds); | |
2681 | } | |
2682 | ||
2683 | static struct port * | |
2684 | bond_find(const char *name) | |
2685 | { | |
2686 | const struct bridge *br; | |
2687 | ||
2688 | LIST_FOR_EACH (br, struct bridge, node, &all_bridges) { | |
2689 | size_t i; | |
2690 | ||
2691 | for (i = 0; i < br->n_ports; i++) { | |
2692 | struct port *port = br->ports[i]; | |
2693 | if (!strcmp(port->name, name) && port->n_ifaces > 1) { | |
2694 | return port; | |
2695 | } | |
2696 | } | |
2697 | } | |
2698 | return NULL; | |
2699 | } | |
2700 | ||
2701 | static void | |
8ca79daa BP |
2702 | bond_unixctl_show(struct unixctl_conn *conn, |
2703 | const char *args, void *aux UNUSED) | |
da285df4 BP |
2704 | { |
2705 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2706 | const struct port *port; | |
2707 | size_t j; | |
2708 | ||
2709 | port = bond_find(args); | |
2710 | if (!port) { | |
2711 | unixctl_command_reply(conn, 501, "no such bond"); | |
2712 | return; | |
2713 | } | |
2714 | ||
2715 | ds_put_format(&ds, "updelay: %d ms\n", port->updelay); | |
2716 | ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); | |
2717 | ds_put_format(&ds, "next rebalance: %lld ms\n", | |
2718 | port->bridge->bond_next_rebalance - time_msec()); | |
2719 | for (j = 0; j < port->n_ifaces; j++) { | |
2720 | const struct iface *iface = port->ifaces[j]; | |
2721 | struct bond_entry *be; | |
2722 | ||
2723 | /* Basic info. */ | |
2724 | ds_put_format(&ds, "slave %s: %s\n", | |
2725 | iface->name, iface->enabled ? "enabled" : "disabled"); | |
2726 | if (j == port->active_iface) { | |
2727 | ds_put_cstr(&ds, "\tactive slave\n"); | |
2728 | } | |
2729 | if (iface->delay_expires != LLONG_MAX) { | |
2730 | ds_put_format(&ds, "\t%s expires in %lld ms\n", | |
2731 | iface->enabled ? "downdelay" : "updelay", | |
2732 | iface->delay_expires - time_msec()); | |
2733 | } | |
2734 | ||
2735 | /* Hashes. */ | |
2736 | for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { | |
2737 | int hash = be - port->bond_hash; | |
2738 | struct mac_entry *me; | |
2739 | ||
2740 | if (be->iface_idx != j) { | |
2741 | continue; | |
2742 | } | |
2743 | ||
2886875a | 2744 | ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", |
da285df4 BP |
2745 | hash, be->tx_bytes / 1024); |
2746 | ||
2747 | /* MACs. */ | |
da285df4 BP |
2748 | LIST_FOR_EACH (me, struct mac_entry, lru_node, |
2749 | &port->bridge->ml->lrus) { | |
2750 | uint16_t dp_ifidx; | |
2751 | tag_type tags = 0; | |
2752 | if (bond_hash(me->mac) == hash | |
2753 | && me->port != port->port_idx | |
2754 | && choose_output_iface(port, me->mac, &dp_ifidx, &tags) | |
2755 | && dp_ifidx == iface->dp_ifidx) | |
2756 | { | |
2757 | ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", | |
2758 | ETH_ADDR_ARGS(me->mac)); | |
2759 | } | |
2760 | } | |
2761 | } | |
2762 | } | |
2763 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
2764 | ds_destroy(&ds); | |
2765 | } | |
2766 | ||
2767 | static void | |
8ca79daa BP |
2768 | bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, |
2769 | void *aux UNUSED) | |
da285df4 BP |
2770 | { |
2771 | char *args = (char *) args_; | |
2772 | char *save_ptr = NULL; | |
2773 | char *bond_s, *hash_s, *slave_s; | |
2774 | uint8_t mac[ETH_ADDR_LEN]; | |
2775 | struct port *port; | |
2776 | struct iface *iface; | |
2777 | struct bond_entry *entry; | |
2778 | int hash; | |
2779 | ||
2780 | bond_s = strtok_r(args, " ", &save_ptr); | |
2781 | hash_s = strtok_r(NULL, " ", &save_ptr); | |
2782 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2783 | if (!slave_s) { | |
2784 | unixctl_command_reply(conn, 501, | |
2785 | "usage: bond/migrate BOND HASH SLAVE"); | |
2786 | return; | |
2787 | } | |
2788 | ||
2789 | port = bond_find(bond_s); | |
2790 | if (!port) { | |
2791 | unixctl_command_reply(conn, 501, "no such bond"); | |
2792 | return; | |
2793 | } | |
2794 | ||
eaa71334 BP |
2795 | if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) |
2796 | == ETH_ADDR_SCAN_COUNT) { | |
da285df4 BP |
2797 | hash = bond_hash(mac); |
2798 | } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) { | |
2799 | hash = atoi(hash_s) & BOND_MASK; | |
2800 | } else { | |
2801 | unixctl_command_reply(conn, 501, "bad hash"); | |
2802 | return; | |
2803 | } | |
2804 | ||
2805 | iface = port_lookup_iface(port, slave_s); | |
2806 | if (!iface) { | |
2807 | unixctl_command_reply(conn, 501, "no such slave"); | |
2808 | return; | |
2809 | } | |
2810 | ||
2811 | if (!iface->enabled) { | |
2812 | unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); | |
2813 | return; | |
2814 | } | |
2815 | ||
2816 | entry = &port->bond_hash[hash]; | |
2817 | ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); | |
2818 | entry->iface_idx = iface->port_ifidx; | |
2819 | entry->iface_tag = tag_create_random(); | |
85c74638 | 2820 | port->bond_compat_is_stale = true; |
da285df4 BP |
2821 | unixctl_command_reply(conn, 200, "migrated"); |
2822 | } | |
2823 | ||
2824 | static void | |
8ca79daa BP |
2825 | bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, |
2826 | void *aux UNUSED) | |
da285df4 BP |
2827 | { |
2828 | char *args = (char *) args_; | |
2829 | char *save_ptr = NULL; | |
2830 | char *bond_s, *slave_s; | |
2831 | struct port *port; | |
2832 | struct iface *iface; | |
2833 | ||
2834 | bond_s = strtok_r(args, " ", &save_ptr); | |
2835 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2836 | if (!slave_s) { | |
2837 | unixctl_command_reply(conn, 501, | |
2838 | "usage: bond/set-active-slave BOND SLAVE"); | |
2839 | return; | |
2840 | } | |
2841 | ||
2842 | port = bond_find(bond_s); | |
2843 | if (!port) { | |
2844 | unixctl_command_reply(conn, 501, "no such bond"); | |
2845 | return; | |
2846 | } | |
2847 | ||
2848 | iface = port_lookup_iface(port, slave_s); | |
2849 | if (!iface) { | |
2850 | unixctl_command_reply(conn, 501, "no such slave"); | |
2851 | return; | |
2852 | } | |
2853 | ||
2854 | if (!iface->enabled) { | |
2855 | unixctl_command_reply(conn, 501, "cannot make disabled slave active"); | |
2856 | return; | |
2857 | } | |
2858 | ||
2859 | if (port->active_iface != iface->port_ifidx) { | |
2860 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
2861 | port->active_iface = iface->port_ifidx; | |
2862 | port->active_iface_tag = tag_create_random(); | |
2863 | VLOG_INFO("port %s: active interface is now %s", | |
2864 | port->name, iface->name); | |
2865 | bond_send_learning_packets(port); | |
2866 | unixctl_command_reply(conn, 200, "done"); | |
2867 | } else { | |
2868 | unixctl_command_reply(conn, 200, "no change"); | |
2869 | } | |
2870 | } | |
2871 | ||
2872 | static void | |
2873 | enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) | |
2874 | { | |
2875 | char *args = (char *) args_; | |
2876 | char *save_ptr = NULL; | |
2877 | char *bond_s, *slave_s; | |
2878 | struct port *port; | |
2879 | struct iface *iface; | |
2880 | ||
2881 | bond_s = strtok_r(args, " ", &save_ptr); | |
2882 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
2883 | if (!slave_s) { | |
2884 | unixctl_command_reply(conn, 501, | |
2885 | "usage: bond/enable/disable-slave BOND SLAVE"); | |
2886 | return; | |
2887 | } | |
2888 | ||
2889 | port = bond_find(bond_s); | |
2890 | if (!port) { | |
2891 | unixctl_command_reply(conn, 501, "no such bond"); | |
2892 | return; | |
2893 | } | |
2894 | ||
2895 | iface = port_lookup_iface(port, slave_s); | |
2896 | if (!iface) { | |
2897 | unixctl_command_reply(conn, 501, "no such slave"); | |
2898 | return; | |
2899 | } | |
2900 | ||
2901 | bond_enable_slave(iface, enable); | |
2902 | unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); | |
2903 | } | |
2904 | ||
2905 | static void | |
8ca79daa BP |
2906 | bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, |
2907 | void *aux UNUSED) | |
da285df4 BP |
2908 | { |
2909 | enable_slave(conn, args, true); | |
2910 | } | |
2911 | ||
2912 | static void | |
8ca79daa BP |
2913 | bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, |
2914 | void *aux UNUSED) | |
da285df4 BP |
2915 | { |
2916 | enable_slave(conn, args, false); | |
2917 | } | |
2918 | ||
e0644b61 | 2919 | static void |
8ca79daa BP |
2920 | bond_unixctl_hash(struct unixctl_conn *conn, const char *args, |
2921 | void *aux UNUSED) | |
e0644b61 IC |
2922 | { |
2923 | uint8_t mac[ETH_ADDR_LEN]; | |
2924 | uint8_t hash; | |
2925 | char *hash_cstr; | |
2926 | ||
2927 | if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) | |
2928 | == ETH_ADDR_SCAN_COUNT) { | |
2929 | hash = bond_hash(mac); | |
2930 | ||
2931 | hash_cstr = xasprintf("%u", hash); | |
2932 | unixctl_command_reply(conn, 200, hash_cstr); | |
2933 | free(hash_cstr); | |
2934 | } else { | |
2935 | unixctl_command_reply(conn, 501, "invalid mac"); | |
2936 | } | |
2937 | } | |
2938 | ||
da285df4 BP |
2939 | static void |
2940 | bond_init(void) | |
2941 | { | |
8ca79daa BP |
2942 | unixctl_command_register("bond/list", bond_unixctl_list, NULL); |
2943 | unixctl_command_register("bond/show", bond_unixctl_show, NULL); | |
2944 | unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); | |
da285df4 | 2945 | unixctl_command_register("bond/set-active-slave", |
8ca79daa BP |
2946 | bond_unixctl_set_active_slave, NULL); |
2947 | unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, | |
2948 | NULL); | |
2949 | unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, | |
2950 | NULL); | |
2951 | unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); | |
da285df4 BP |
2952 | } |
2953 | \f | |
064af421 BP |
2954 | /* Port functions. */ |
2955 | ||
76343538 | 2956 | static struct port * |
064af421 BP |
2957 | port_create(struct bridge *br, const char *name) |
2958 | { | |
2959 | struct port *port; | |
2960 | ||
ec6fde61 | 2961 | port = xzalloc(sizeof *port); |
064af421 BP |
2962 | port->bridge = br; |
2963 | port->port_idx = br->n_ports; | |
2964 | port->vlan = -1; | |
2965 | port->trunks = NULL; | |
2966 | port->name = xstrdup(name); | |
2967 | port->active_iface = -1; | |
064af421 BP |
2968 | |
2969 | if (br->n_ports >= br->allocated_ports) { | |
2970 | br->ports = x2nrealloc(br->ports, &br->allocated_ports, | |
2971 | sizeof *br->ports); | |
2972 | } | |
2973 | br->ports[br->n_ports++] = port; | |
2974 | ||
2975 | VLOG_INFO("created port %s on bridge %s", port->name, br->name); | |
2976 | bridge_flush(br); | |
76343538 BP |
2977 | |
2978 | return port; | |
064af421 BP |
2979 | } |
2980 | ||
2981 | static void | |
76343538 | 2982 | port_reconfigure(struct port *port, const struct ovsrec_port *cfg) |
064af421 | 2983 | { |
76343538 BP |
2984 | struct shash old_ifaces, new_ifaces; |
2985 | struct shash_node *node; | |
064af421 BP |
2986 | unsigned long *trunks; |
2987 | int vlan; | |
2988 | size_t i; | |
2989 | ||
76343538 BP |
2990 | port->cfg = cfg; |
2991 | ||
064af421 | 2992 | /* Collect old and new interfaces. */ |
76343538 BP |
2993 | shash_init(&old_ifaces); |
2994 | shash_init(&new_ifaces); | |
064af421 | 2995 | for (i = 0; i < port->n_ifaces; i++) { |
76343538 | 2996 | shash_add(&old_ifaces, port->ifaces[i]->name, port->ifaces[i]); |
064af421 | 2997 | } |
76343538 BP |
2998 | for (i = 0; i < cfg->n_interfaces; i++) { |
2999 | const char *name = cfg->interfaces[i]->name; | |
3000 | if (!shash_add_once(&new_ifaces, name, cfg->interfaces[i])) { | |
3001 | VLOG_WARN("port %s: %s specified twice as port interface", | |
3002 | port->name, name); | |
064af421 | 3003 | } |
76343538 BP |
3004 | } |
3005 | port->updelay = cfg->bond_updelay; | |
3006 | if (port->updelay < 0) { | |
3007 | port->updelay = 0; | |
3008 | } | |
3009 | port->updelay = cfg->bond_downdelay; | |
3010 | if (port->downdelay < 0) { | |
3011 | port->downdelay = 0; | |
064af421 BP |
3012 | } |
3013 | ||
3014 | /* Get rid of deleted interfaces and add new interfaces. */ | |
76343538 BP |
3015 | SHASH_FOR_EACH (node, &old_ifaces) { |
3016 | if (!shash_find(&new_ifaces, node->name)) { | |
3017 | iface_destroy(node->data); | |
064af421 BP |
3018 | } |
3019 | } | |
76343538 BP |
3020 | SHASH_FOR_EACH (node, &new_ifaces) { |
3021 | const struct ovsrec_interface *if_cfg = node->data; | |
76343538 BP |
3022 | struct iface *iface; |
3023 | ||
a740f0de | 3024 | iface = shash_find_data(&old_ifaces, if_cfg->name); |
76343538 | 3025 | if (!iface) { |
a740f0de | 3026 | iface = iface_create(port, if_cfg); |
064af421 | 3027 | } |
76343538 | 3028 | iface->cfg = if_cfg; |
064af421 BP |
3029 | } |
3030 | ||
3031 | /* Get VLAN tag. */ | |
3032 | vlan = -1; | |
76343538 BP |
3033 | if (cfg->tag) { |
3034 | if (port->n_ifaces < 2) { | |
3035 | vlan = *cfg->tag; | |
064af421 BP |
3036 | if (vlan >= 0 && vlan <= 4095) { |
3037 | VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); | |
76343538 BP |
3038 | } else { |
3039 | vlan = -1; | |
064af421 BP |
3040 | } |
3041 | } else { | |
3042 | /* It's possible that bonded, VLAN-tagged ports make sense. Maybe | |
3043 | * they even work as-is. But they have not been tested. */ | |
3044 | VLOG_WARN("port %s: VLAN tags not supported on bonded ports", | |
3045 | port->name); | |
3046 | } | |
3047 | } | |
3048 | if (port->vlan != vlan) { | |
3049 | port->vlan = vlan; | |
3050 | bridge_flush(port->bridge); | |
3051 | } | |
3052 | ||
3053 | /* Get trunked VLANs. */ | |
3054 | trunks = NULL; | |
3055 | if (vlan < 0) { | |
76343538 | 3056 | size_t n_errors; |
064af421 BP |
3057 | size_t i; |
3058 | ||
3059 | trunks = bitmap_allocate(4096); | |
064af421 | 3060 | n_errors = 0; |
76343538 BP |
3061 | for (i = 0; i < cfg->n_trunks; i++) { |
3062 | int trunk = cfg->trunks[i]; | |
064af421 BP |
3063 | if (trunk >= 0) { |
3064 | bitmap_set1(trunks, trunk); | |
3065 | } else { | |
3066 | n_errors++; | |
3067 | } | |
3068 | } | |
3069 | if (n_errors) { | |
3070 | VLOG_ERR("port %s: invalid values for %zu trunk VLANs", | |
76343538 | 3071 | port->name, cfg->n_trunks); |
064af421 | 3072 | } |
76343538 | 3073 | if (n_errors == cfg->n_trunks) { |
064af421 BP |
3074 | if (n_errors) { |
3075 | VLOG_ERR("port %s: no valid trunks, trunking all VLANs", | |
3076 | port->name); | |
3077 | } | |
3078 | bitmap_set_multiple(trunks, 0, 4096, 1); | |
3079 | } | |
3080 | } else { | |
76343538 BP |
3081 | if (cfg->n_trunks) { |
3082 | VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", | |
3083 | port->name); | |
064af421 BP |
3084 | } |
3085 | } | |
3086 | if (trunks == NULL | |
3087 | ? port->trunks != NULL | |
3088 | : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { | |
3089 | bridge_flush(port->bridge); | |
3090 | } | |
3091 | bitmap_free(port->trunks); | |
3092 | port->trunks = trunks; | |
3093 | ||
76343538 BP |
3094 | shash_destroy(&old_ifaces); |
3095 | shash_destroy(&new_ifaces); | |
064af421 BP |
3096 | } |
3097 | ||
3098 | static void | |
3099 | port_destroy(struct port *port) | |
3100 | { | |
3101 | if (port) { | |
3102 | struct bridge *br = port->bridge; | |
3103 | struct port *del; | |
064af421 BP |
3104 | |
3105 | proc_net_compat_update_vlan(port->name, NULL, 0); | |
85c74638 | 3106 | proc_net_compat_update_bond(port->name, NULL); |
064af421 | 3107 | |
76343538 | 3108 | #if 0 |
064af421 BP |
3109 | for (i = 0; i < MAX_MIRRORS; i++) { |
3110 | struct mirror *m = br->mirrors[i]; | |
3111 | if (m && m->out_port == port) { | |
3112 | mirror_destroy(m); | |
3113 | } | |
3114 | } | |
76343538 | 3115 | #endif |
064af421 BP |
3116 | |
3117 | while (port->n_ifaces > 0) { | |
3118 | iface_destroy(port->ifaces[port->n_ifaces - 1]); | |
3119 | } | |
3120 | ||
3121 | del = br->ports[port->port_idx] = br->ports[--br->n_ports]; | |
3122 | del->port_idx = port->port_idx; | |
3123 | ||
3124 | free(port->ifaces); | |
3125 | bitmap_free(port->trunks); | |
3126 | free(port->name); | |
3127 | free(port); | |
3128 | bridge_flush(br); | |
3129 | } | |
3130 | } | |
3131 | ||
3132 | static struct port * | |
3133 | port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
3134 | { | |
3135 | struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx); | |
3136 | return iface ? iface->port : NULL; | |
3137 | } | |
3138 | ||
3139 | static struct port * | |
3140 | port_lookup(const struct bridge *br, const char *name) | |
3141 | { | |
3142 | size_t i; | |
3143 | ||
3144 | for (i = 0; i < br->n_ports; i++) { | |
3145 | struct port *port = br->ports[i]; | |
3146 | if (!strcmp(port->name, name)) { | |
3147 | return port; | |
3148 | } | |
3149 | } | |
3150 | return NULL; | |
3151 | } | |
3152 | ||
da285df4 BP |
3153 | static struct iface * |
3154 | port_lookup_iface(const struct port *port, const char *name) | |
3155 | { | |
3156 | size_t j; | |
3157 | ||
3158 | for (j = 0; j < port->n_ifaces; j++) { | |
3159 | struct iface *iface = port->ifaces[j]; | |
3160 | if (!strcmp(iface->name, name)) { | |
3161 | return iface; | |
3162 | } | |
3163 | } | |
3164 | return NULL; | |
3165 | } | |
3166 | ||
064af421 BP |
3167 | static void |
3168 | port_update_bonding(struct port *port) | |
3169 | { | |
3170 | if (port->n_ifaces < 2) { | |
3171 | /* Not a bonded port. */ | |
3172 | if (port->bond_hash) { | |
3173 | free(port->bond_hash); | |
3174 | port->bond_hash = NULL; | |
85c74638 | 3175 | port->bond_compat_is_stale = true; |
76343538 | 3176 | port->bond_fake_iface = false; |
064af421 BP |
3177 | } |
3178 | } else { | |
3179 | if (!port->bond_hash) { | |
3180 | size_t i; | |
3181 | ||
3182 | port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); | |
3183 | for (i = 0; i <= BOND_MASK; i++) { | |
3184 | struct bond_entry *e = &port->bond_hash[i]; | |
3185 | e->iface_idx = -1; | |
3186 | e->tx_bytes = 0; | |
3187 | } | |
3188 | port->no_ifaces_tag = tag_create_random(); | |
3189 | bond_choose_active_iface(port); | |
3190 | } | |
85c74638 | 3191 | port->bond_compat_is_stale = true; |
76343538 | 3192 | port->bond_fake_iface = port->cfg->bond_fake_iface; |
064af421 BP |
3193 | } |
3194 | } | |
3195 | ||
3196 | static void | |
3197 | port_update_bond_compat(struct port *port) | |
3198 | { | |
2aebae83 | 3199 | struct compat_bond_hash compat_hashes[BOND_MASK + 1]; |
064af421 BP |
3200 | struct compat_bond bond; |
3201 | size_t i; | |
3202 | ||
3203 | if (port->n_ifaces < 2) { | |
85c74638 | 3204 | proc_net_compat_update_bond(port->name, NULL); |
064af421 BP |
3205 | return; |
3206 | } | |
3207 | ||
3208 | bond.up = false; | |
3209 | bond.updelay = port->updelay; | |
3210 | bond.downdelay = port->downdelay; | |
2aebae83 BP |
3211 | |
3212 | bond.n_hashes = 0; | |
3213 | bond.hashes = compat_hashes; | |
3214 | if (port->bond_hash) { | |
3215 | const struct bond_entry *e; | |
3216 | for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { | |
3217 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
3218 | struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; | |
3219 | cbh->hash = e - port->bond_hash; | |
3220 | cbh->netdev_name = port->ifaces[e->iface_idx]->name; | |
3221 | } | |
3222 | } | |
3223 | } | |
3224 | ||
064af421 BP |
3225 | bond.n_slaves = port->n_ifaces; |
3226 | bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); | |
3227 | for (i = 0; i < port->n_ifaces; i++) { | |
3228 | struct iface *iface = port->ifaces[i]; | |
3229 | struct compat_bond_slave *slave = &bond.slaves[i]; | |
3230 | slave->name = iface->name; | |
6643c03b JP |
3231 | |
3232 | /* We need to make the same determination as the Linux bonding | |
3233 | * code to determine whether a slave should be consider "up". | |
3234 | * The Linux function bond_miimon_inspect() supports four | |
3235 | * BOND_LINK_* states: | |
3236 | * | |
3237 | * - BOND_LINK_UP: carrier detected, updelay has passed. | |
3238 | * - BOND_LINK_FAIL: carrier lost, downdelay in progress. | |
3239 | * - BOND_LINK_DOWN: carrier lost, downdelay has passed. | |
3240 | * - BOND_LINK_BACK: carrier detected, updelay in progress. | |
3241 | * | |
3242 | * The function bond_info_show_slave() only considers BOND_LINK_UP | |
3243 | * to be "up" and anything else to be "down". | |
3244 | */ | |
3245 | slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; | |
064af421 BP |
3246 | if (slave->up) { |
3247 | bond.up = true; | |
3248 | } | |
96b9b7a9 | 3249 | netdev_get_etheraddr(iface->netdev, slave->mac); |
064af421 | 3250 | } |
2aebae83 | 3251 | |
76343538 | 3252 | if (port->bond_fake_iface) { |
c6303ea1 JG |
3253 | struct netdev *bond_netdev; |
3254 | ||
3255 | if (!netdev_open(port->name, NETDEV_ETH_TYPE_NONE, &bond_netdev)) { | |
3256 | if (bond.up) { | |
3257 | netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); | |
3258 | } else { | |
3259 | netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); | |
3260 | } | |
3261 | netdev_close(bond_netdev); | |
3262 | } | |
3263 | } | |
3264 | ||
064af421 BP |
3265 | proc_net_compat_update_bond(port->name, &bond); |
3266 | free(bond.slaves); | |
3267 | } | |
3268 | ||
3269 | static void | |
3270 | port_update_vlan_compat(struct port *port) | |
3271 | { | |
3272 | struct bridge *br = port->bridge; | |
3273 | char *vlandev_name = NULL; | |
3274 | ||
3275 | if (port->vlan > 0) { | |
3276 | /* Figure out the name that the VLAN device should actually have, if it | |
3277 | * existed. This takes some work because the VLAN device would not | |
3278 | * have port->name in its name; rather, it would have the trunk port's | |
3279 | * name, and 'port' would be attached to a bridge that also had the | |
3280 | * VLAN device one of its ports. So we need to find a trunk port that | |
3281 | * includes port->vlan. | |
3282 | * | |
3283 | * There might be more than one candidate. This doesn't happen on | |
3284 | * XenServer, so if it happens we just pick the first choice in | |
3285 | * alphabetical order instead of creating multiple VLAN devices. */ | |
3286 | size_t i; | |
3287 | for (i = 0; i < br->n_ports; i++) { | |
3288 | struct port *p = br->ports[i]; | |
3289 | if (port_trunks_vlan(p, port->vlan) | |
3290 | && p->n_ifaces | |
3291 | && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) | |
3292 | { | |
96b9b7a9 BP |
3293 | uint8_t ea[ETH_ADDR_LEN]; |
3294 | netdev_get_etheraddr(p->ifaces[0]->netdev, ea); | |
064af421 BP |
3295 | if (!eth_addr_is_multicast(ea) && |
3296 | !eth_addr_is_reserved(ea) && | |
3297 | !eth_addr_is_zero(ea)) { | |
3298 | vlandev_name = p->name; | |
3299 | } | |
3300 | } | |
3301 | } | |
3302 | } | |
3303 | proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); | |
3304 | } | |
3305 | \f | |
3306 | /* Interface functions. */ | |
3307 | ||
76343538 | 3308 | static struct iface * |
a740f0de | 3309 | iface_create(struct port *port, const struct ovsrec_interface *if_cfg) |
064af421 | 3310 | { |
064af421 | 3311 | struct iface *iface; |
a740f0de JG |
3312 | char *name = if_cfg->name; |
3313 | int error; | |
064af421 | 3314 | |
ec6fde61 | 3315 | iface = xzalloc(sizeof *iface); |
064af421 BP |
3316 | iface->port = port; |
3317 | iface->port_ifidx = port->n_ifaces; | |
3318 | iface->name = xstrdup(name); | |
3319 | iface->dp_ifidx = -1; | |
3320 | iface->tag = tag_create_random(); | |
064af421 | 3321 | iface->delay_expires = LLONG_MAX; |
0c6aea3f | 3322 | iface->netdev = NULL; |
064af421 BP |
3323 | |
3324 | if (port->n_ifaces >= port->allocated_ifaces) { | |
3325 | port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, | |
3326 | sizeof *port->ifaces); | |
3327 | } | |
3328 | port->ifaces[port->n_ifaces++] = iface; | |
3329 | if (port->n_ifaces > 1) { | |
3330 | port->bridge->has_bonded_ports = true; | |
3331 | } | |
3332 | ||
a740f0de JG |
3333 | /* Attempt to create the network interface in case it |
3334 | * doesn't exist yet. */ | |
3335 | error = set_up_iface(if_cfg, true); | |
3336 | if (error) { | |
3337 | VLOG_WARN("could not create iface %s: %s\n", iface->name, | |
3338 | strerror(error)); | |
3339 | } | |
3340 | ||
064af421 BP |
3341 | VLOG_DBG("attached network device %s to port %s", iface->name, port->name); |
3342 | ||
064af421 | 3343 | bridge_flush(port->bridge); |
76343538 BP |
3344 | |
3345 | return iface; | |
064af421 BP |
3346 | } |
3347 | ||
3348 | static void | |
3349 | iface_destroy(struct iface *iface) | |
3350 | { | |
3351 | if (iface) { | |
3352 | struct port *port = iface->port; | |
3353 | struct bridge *br = port->bridge; | |
3354 | bool del_active = port->active_iface == iface->port_ifidx; | |
3355 | struct iface *del; | |
3356 | ||
3357 | if (iface->dp_ifidx >= 0) { | |
3358 | port_array_set(&br->ifaces, iface->dp_ifidx, NULL); | |
3359 | } | |
3360 | ||
3361 | del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; | |
3362 | del->port_ifidx = iface->port_ifidx; | |
3363 | ||
0c6aea3f | 3364 | netdev_close(iface->netdev); |
064af421 BP |
3365 | |
3366 | if (del_active) { | |
3367 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
3368 | bond_choose_active_iface(port); | |
2303f3b2 | 3369 | bond_send_learning_packets(port); |
064af421 BP |
3370 | } |
3371 | ||
a740f0de JG |
3372 | netdev_destroy(iface->name); |
3373 | free(iface->name); | |
3374 | free(iface); | |
3375 | ||
064af421 BP |
3376 | bridge_flush(port->bridge); |
3377 | } | |
3378 | } | |
3379 | ||
3380 | static struct iface * | |
3381 | iface_lookup(const struct bridge *br, const char *name) | |
3382 | { | |
3383 | size_t i, j; | |
3384 | ||
3385 | for (i = 0; i < br->n_ports; i++) { | |
3386 | struct port *port = br->ports[i]; | |
3387 | for (j = 0; j < port->n_ifaces; j++) { | |
3388 | struct iface *iface = port->ifaces[j]; | |
3389 | if (!strcmp(iface->name, name)) { | |
3390 | return iface; | |
3391 | } | |
3392 | } | |
3393 | } | |
3394 | return NULL; | |
3395 | } | |
3396 | ||
3397 | static struct iface * | |
3398 | iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
3399 | { | |
3400 | return port_array_get(&br->ifaces, dp_ifidx); | |
3401 | } | |
557d8e6c BP |
3402 | |
3403 | /* Returns true if 'iface' is the name of an "internal" interface on bridge | |
3404 | * 'br', that is, an interface that is entirely simulated within the datapath. | |
3405 | * The local port (ODPP_LOCAL) is always an internal interface. Other local | |
3406 | * interfaces are created by setting "iface.<iface>.internal = true". | |
3407 | * | |
3408 | * In addition, we have a kluge-y feature that creates an internal port with | |
3409 | * the name of a bonded port if "bonding.<bondname>.fake-iface = true" is set. | |
3410 | * This feature needs to go away in the long term. Until then, this is one | |
3411 | * reason why this function takes a name instead of a struct iface: the fake | |
3412 | * interfaces created this way do not have a struct iface. */ | |
3413 | static bool | |
76343538 | 3414 | iface_is_internal(const struct bridge *br, const char *if_name) |
557d8e6c | 3415 | { |
76343538 BP |
3416 | /* XXX wastes time */ |
3417 | struct iface *iface; | |
3418 | struct port *port; | |
3419 | ||
3420 | if (!strcmp(if_name, br->name)) { | |
557d8e6c BP |
3421 | return true; |
3422 | } | |
3423 | ||
76343538 BP |
3424 | iface = iface_lookup(br, if_name); |
3425 | if (iface && !strcmp(iface->cfg->type, "internal")) { | |
3426 | return true; | |
557d8e6c BP |
3427 | } |
3428 | ||
76343538 BP |
3429 | port = port_lookup(br, if_name); |
3430 | if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { | |
3431 | return true; | |
3432 | } | |
557d8e6c BP |
3433 | return false; |
3434 | } | |
52df17e7 BP |
3435 | |
3436 | /* Set Ethernet address of 'iface', if one is specified in the configuration | |
3437 | * file. */ | |
3438 | static void | |
3439 | iface_set_mac(struct iface *iface) | |
3440 | { | |
76343538 | 3441 | uint8_t ea[ETH_ADDR_LEN]; |
52df17e7 | 3442 | |
76343538 | 3443 | if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) { |
52df17e7 BP |
3444 | if (eth_addr_is_multicast(ea)) { |
3445 | VLOG_ERR("interface %s: cannot set MAC to multicast address", | |
3446 | iface->name); | |
3447 | } else if (iface->dp_ifidx == ODPP_LOCAL) { | |
3448 | VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", | |
3449 | iface->name, iface->name); | |
3450 | } else { | |
4d678233 | 3451 | int error = netdev_set_etheraddr(iface->netdev, ea); |
52df17e7 BP |
3452 | if (error) { |
3453 | VLOG_ERR("interface %s: setting MAC failed (%s)", | |
3454 | iface->name, strerror(error)); | |
3455 | } | |
3456 | } | |
3457 | } | |
3458 | } | |
064af421 BP |
3459 | \f |
3460 | /* Port mirroring. */ | |
3461 | ||
76343538 | 3462 | #if 0 |
064af421 | 3463 | static void |
76343538 | 3464 | mirror_reconfigure(struct bridge *br UNUSED) |
064af421 BP |
3465 | { |
3466 | struct svec old_mirrors, new_mirrors; | |
f2d7fd66 JG |
3467 | size_t i, n_rspan_vlans; |
3468 | unsigned long *rspan_vlans; | |
064af421 BP |
3469 | |
3470 | /* Collect old and new mirrors. */ | |
3471 | svec_init(&old_mirrors); | |
3472 | svec_init(&new_mirrors); | |
3473 | cfg_get_subsections(&new_mirrors, "mirror.%s", br->name); | |
3474 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3475 | if (br->mirrors[i]) { | |
3476 | svec_add(&old_mirrors, br->mirrors[i]->name); | |
3477 | } | |
3478 | } | |
3479 | ||
3480 | /* Get rid of deleted mirrors and add new mirrors. */ | |
3481 | svec_sort(&old_mirrors); | |
3482 | assert(svec_is_unique(&old_mirrors)); | |
3483 | svec_sort(&new_mirrors); | |
3484 | assert(svec_is_unique(&new_mirrors)); | |
3485 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3486 | struct mirror *m = br->mirrors[i]; | |
3487 | if (m && !svec_contains(&new_mirrors, m->name)) { | |
3488 | mirror_destroy(m); | |
3489 | } | |
3490 | } | |
3491 | for (i = 0; i < new_mirrors.n; i++) { | |
3492 | const char *name = new_mirrors.names[i]; | |
3493 | if (!svec_contains(&old_mirrors, name)) { | |
3494 | mirror_create(br, name); | |
3495 | } | |
3496 | } | |
3497 | svec_destroy(&old_mirrors); | |
3498 | svec_destroy(&new_mirrors); | |
3499 | ||
3500 | /* Reconfigure all mirrors. */ | |
3501 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3502 | if (br->mirrors[i]) { | |
3503 | mirror_reconfigure_one(br->mirrors[i]); | |
3504 | } | |
3505 | } | |
3506 | ||
3507 | /* Update port reserved status. */ | |
3508 | for (i = 0; i < br->n_ports; i++) { | |
3509 | br->ports[i]->is_mirror_output_port = false; | |
3510 | } | |
3511 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3512 | struct mirror *m = br->mirrors[i]; | |
3513 | if (m && m->out_port) { | |
3514 | m->out_port->is_mirror_output_port = true; | |
3515 | } | |
3516 | } | |
f2d7fd66 JG |
3517 | |
3518 | /* Update learning disabled vlans (for RSPAN). */ | |
3519 | rspan_vlans = NULL; | |
3520 | n_rspan_vlans = cfg_count("vlan.%s.disable-learning", br->name); | |
3521 | if (n_rspan_vlans) { | |
3522 | rspan_vlans = bitmap_allocate(4096); | |
3523 | ||
3524 | for (i = 0; i < n_rspan_vlans; i++) { | |
3525 | int vlan = cfg_get_vlan(i, "vlan.%s.disable-learning", br->name); | |
3526 | if (vlan >= 0) { | |
3527 | bitmap_set1(rspan_vlans, vlan); | |
42061b2a JG |
3528 | VLOG_INFO("bridge %s: disabling learning on vlan %d\n", |
3529 | br->name, vlan); | |
f2d7fd66 JG |
3530 | } else { |
3531 | VLOG_ERR("bridge %s: invalid value '%s' for learning disabled " | |
3532 | "VLAN", br->name, | |
3533 | cfg_get_string(i, "vlan.%s.disable-learning", br->name)); | |
3534 | } | |
3535 | } | |
3536 | } | |
3537 | if (mac_learning_set_disabled_vlans(br->ml, rspan_vlans)) { | |
3538 | bridge_flush(br); | |
3539 | } | |
064af421 BP |
3540 | } |
3541 | ||
3542 | static void | |
3543 | mirror_create(struct bridge *br, const char *name) | |
3544 | { | |
3545 | struct mirror *m; | |
3546 | size_t i; | |
3547 | ||
3548 | for (i = 0; ; i++) { | |
3549 | if (i >= MAX_MIRRORS) { | |
3550 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
3551 | "cannot create %s", br->name, MAX_MIRRORS, name); | |
3552 | return; | |
3553 | } | |
3554 | if (!br->mirrors[i]) { | |
3555 | break; | |
3556 | } | |
3557 | } | |
3558 | ||
3559 | VLOG_INFO("created port mirror %s on bridge %s", name, br->name); | |
3560 | bridge_flush(br); | |
3561 | ||
ec6fde61 | 3562 | br->mirrors[i] = m = xzalloc(sizeof *m); |
064af421 BP |
3563 | m->bridge = br; |
3564 | m->idx = i; | |
3565 | m->name = xstrdup(name); | |
3566 | svec_init(&m->src_ports); | |
3567 | svec_init(&m->dst_ports); | |
3568 | m->vlans = NULL; | |
3569 | m->n_vlans = 0; | |
3570 | m->out_vlan = -1; | |
3571 | m->out_port = NULL; | |
3572 | } | |
3573 | ||
3574 | static void | |
3575 | mirror_destroy(struct mirror *m) | |
3576 | { | |
3577 | if (m) { | |
3578 | struct bridge *br = m->bridge; | |
3579 | size_t i; | |
3580 | ||
3581 | for (i = 0; i < br->n_ports; i++) { | |
3582 | br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
3583 | br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
3584 | } | |
3585 | ||
3586 | svec_destroy(&m->src_ports); | |
3587 | svec_destroy(&m->dst_ports); | |
3588 | free(m->vlans); | |
3589 | ||
3590 | m->bridge->mirrors[m->idx] = NULL; | |
3591 | free(m); | |
3592 | ||
3593 | bridge_flush(br); | |
3594 | } | |
3595 | } | |
3596 | ||
3597 | static void | |
3598 | prune_ports(struct mirror *m, struct svec *ports) | |
3599 | { | |
3600 | struct svec tmp; | |
3601 | size_t i; | |
3602 | ||
3603 | svec_sort_unique(ports); | |
3604 | ||
3605 | svec_init(&tmp); | |
3606 | for (i = 0; i < ports->n; i++) { | |
3607 | const char *name = ports->names[i]; | |
3608 | if (port_lookup(m->bridge, name)) { | |
3609 | svec_add(&tmp, name); | |
3610 | } else { | |
3611 | VLOG_WARN("mirror.%s.%s: cannot match on nonexistent port %s", | |
3612 | m->bridge->name, m->name, name); | |
3613 | } | |
3614 | } | |
3615 | svec_swap(ports, &tmp); | |
3616 | svec_destroy(&tmp); | |
3617 | } | |
3618 | ||
3619 | static size_t | |
3620 | prune_vlans(struct mirror *m, struct svec *vlan_strings, int **vlans) | |
3621 | { | |
3622 | size_t n_vlans, i; | |
3623 | ||
3624 | /* This isn't perfect: it won't combine "0" and "00", and the textual sort | |
3625 | * order won't give us numeric sort order. But that's good enough for what | |
3626 | * we need right now. */ | |
3627 | svec_sort_unique(vlan_strings); | |
3628 | ||
3629 | *vlans = xmalloc(sizeof *vlans * vlan_strings->n); | |
3630 | n_vlans = 0; | |
3631 | for (i = 0; i < vlan_strings->n; i++) { | |
3632 | const char *name = vlan_strings->names[i]; | |
3633 | int vlan; | |
3634 | if (!str_to_int(name, 10, &vlan) || vlan < 0 || vlan > 4095) { | |
3635 | VLOG_WARN("mirror.%s.%s.select.vlan: ignoring invalid VLAN %s", | |
3636 | m->bridge->name, m->name, name); | |
3637 | } else { | |
3638 | (*vlans)[n_vlans++] = vlan; | |
3639 | } | |
3640 | } | |
3641 | return n_vlans; | |
3642 | } | |
3643 | ||
3644 | static bool | |
3645 | vlan_is_mirrored(const struct mirror *m, int vlan) | |
3646 | { | |
3647 | size_t i; | |
3648 | ||
3649 | for (i = 0; i < m->n_vlans; i++) { | |
3650 | if (m->vlans[i] == vlan) { | |
3651 | return true; | |
3652 | } | |
3653 | } | |
3654 | return false; | |
3655 | } | |
3656 | ||
3657 | static bool | |
3658 | port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) | |
3659 | { | |
3660 | size_t i; | |
3661 | ||
3662 | for (i = 0; i < m->n_vlans; i++) { | |
3663 | if (port_trunks_vlan(p, m->vlans[i])) { | |
3664 | return true; | |
3665 | } | |
3666 | } | |
3667 | return false; | |
3668 | } | |
3669 | ||
3670 | static void | |
76343538 | 3671 | mirror_reconfigure_one(struct mirror *m UNUSED) |
064af421 BP |
3672 | { |
3673 | char *pfx = xasprintf("mirror.%s.%s", m->bridge->name, m->name); | |
3674 | struct svec src_ports, dst_ports, ports; | |
3675 | struct svec vlan_strings; | |
3676 | mirror_mask_t mirror_bit; | |
3677 | const char *out_port_name; | |
3678 | struct port *out_port; | |
3679 | int out_vlan; | |
3680 | size_t n_vlans; | |
3681 | int *vlans; | |
3682 | size_t i; | |
3683 | bool mirror_all_ports; | |
e0c27cff | 3684 | bool any_ports_specified; |
064af421 BP |
3685 | |
3686 | /* Get output port. */ | |
3687 | out_port_name = cfg_get_key(0, "mirror.%s.%s.output.port", | |
3688 | m->bridge->name, m->name); | |
3689 | if (out_port_name) { | |
3690 | out_port = port_lookup(m->bridge, out_port_name); | |
3691 | if (!out_port) { | |
3692 | VLOG_ERR("%s.output.port: bridge %s does not have a port " | |
3693 | "named %s", pfx, m->bridge->name, out_port_name); | |
3694 | mirror_destroy(m); | |
3695 | free(pfx); | |
3696 | return; | |
3697 | } | |
3698 | out_vlan = -1; | |
3699 | ||
3700 | if (cfg_has("%s.output.vlan", pfx)) { | |
3701 | VLOG_ERR("%s.output.port and %s.output.vlan both specified; " | |
3702 | "ignoring %s.output.vlan", pfx, pfx, pfx); | |
3703 | } | |
3704 | } else if (cfg_has("%s.output.vlan", pfx)) { | |
3705 | out_port = NULL; | |
3706 | out_vlan = cfg_get_vlan(0, "%s.output.vlan", pfx); | |
3707 | } else { | |
3708 | VLOG_ERR("%s: neither %s.output.port nor %s.output.vlan specified, " | |
3709 | "but exactly one is required; disabling port mirror %s", | |
3710 | pfx, pfx, pfx, pfx); | |
3711 | mirror_destroy(m); | |
3712 | free(pfx); | |
3713 | return; | |
3714 | } | |
3715 | ||
3716 | /* Get all the ports, and drop duplicates and ports that don't exist. */ | |
3717 | svec_init(&src_ports); | |
3718 | svec_init(&dst_ports); | |
3719 | svec_init(&ports); | |
3720 | cfg_get_all_keys(&src_ports, "%s.select.src-port", pfx); | |
3721 | cfg_get_all_keys(&dst_ports, "%s.select.dst-port", pfx); | |
3722 | cfg_get_all_keys(&ports, "%s.select.port", pfx); | |
e0c27cff | 3723 | any_ports_specified = src_ports.n || dst_ports.n || ports.n; |
064af421 BP |
3724 | svec_append(&src_ports, &ports); |
3725 | svec_append(&dst_ports, &ports); | |
3726 | svec_destroy(&ports); | |
3727 | prune_ports(m, &src_ports); | |
3728 | prune_ports(m, &dst_ports); | |
e0c27cff BP |
3729 | if (any_ports_specified && !src_ports.n && !dst_ports.n) { |
3730 | VLOG_ERR("%s: none of the specified ports exist; " | |
3731 | "disabling port mirror %s", pfx, pfx); | |
3732 | mirror_destroy(m); | |
3733 | goto exit; | |
3734 | } | |
064af421 BP |
3735 | |
3736 | /* Get all the vlans, and drop duplicate and invalid vlans. */ | |
3737 | svec_init(&vlan_strings); | |
3738 | cfg_get_all_keys(&vlan_strings, "%s.select.vlan", pfx); | |
3739 | n_vlans = prune_vlans(m, &vlan_strings, &vlans); | |
3740 | svec_destroy(&vlan_strings); | |
3741 | ||
3742 | /* Update mirror data. */ | |
3743 | if (!svec_equal(&m->src_ports, &src_ports) | |
3744 | || !svec_equal(&m->dst_ports, &dst_ports) | |
3745 | || m->n_vlans != n_vlans | |
3746 | || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) | |
3747 | || m->out_port != out_port | |
3748 | || m->out_vlan != out_vlan) { | |
3749 | bridge_flush(m->bridge); | |
3750 | } | |
3751 | svec_swap(&m->src_ports, &src_ports); | |
3752 | svec_swap(&m->dst_ports, &dst_ports); | |
3753 | free(m->vlans); | |
3754 | m->vlans = vlans; | |
3755 | m->n_vlans = n_vlans; | |
3756 | m->out_port = out_port; | |
3757 | m->out_vlan = out_vlan; | |
3758 | ||
3759 | /* If no selection criteria have been given, mirror for all ports. */ | |
3760 | mirror_all_ports = (!m->src_ports.n) && (!m->dst_ports.n) && (!m->n_vlans); | |
3761 | ||
3762 | /* Update ports. */ | |
3763 | mirror_bit = MIRROR_MASK_C(1) << m->idx; | |
3764 | for (i = 0; i < m->bridge->n_ports; i++) { | |
3765 | struct port *port = m->bridge->ports[i]; | |
3766 | ||
3767 | if (mirror_all_ports | |
3768 | || svec_contains(&m->src_ports, port->name) | |
3769 | || (m->n_vlans | |
3770 | && (!port->vlan | |
3771 | ? port_trunks_any_mirrored_vlan(m, port) | |
3772 | : vlan_is_mirrored(m, port->vlan)))) { | |
3773 | port->src_mirrors |= mirror_bit; | |
3774 | } else { | |
3775 | port->src_mirrors &= ~mirror_bit; | |
3776 | } | |
3777 | ||
3778 | if (mirror_all_ports || svec_contains(&m->dst_ports, port->name)) { | |
3779 | port->dst_mirrors |= mirror_bit; | |
3780 | } else { | |
3781 | port->dst_mirrors &= ~mirror_bit; | |
3782 | } | |
3783 | } | |
3784 | ||
3785 | /* Clean up. */ | |
e0c27cff | 3786 | exit: |
064af421 BP |
3787 | svec_destroy(&src_ports); |
3788 | svec_destroy(&dst_ports); | |
3789 | free(pfx); | |
3790 | } | |
76343538 | 3791 | #endif |