]>
Commit | Line | Data |
---|---|---|
2c30e5d1 | 1 | /* Copyright (c) 2008, 2009, 2010 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> | |
9d82ec47 | 23 | #include <sys/socket.h> |
064af421 BP |
24 | #include <net/if.h> |
25 | #include <openflow/openflow.h> | |
26 | #include <signal.h> | |
27 | #include <stdlib.h> | |
28 | #include <strings.h> | |
29 | #include <sys/stat.h> | |
30 | #include <sys/socket.h> | |
7e40e21d | 31 | #include <sys/types.h> |
064af421 BP |
32 | #include <unistd.h> |
33 | #include "bitmap.h" | |
064af421 BP |
34 | #include "coverage.h" |
35 | #include "dirs.h" | |
36 | #include "dpif.h" | |
37 | #include "dynamic-string.h" | |
38 | #include "flow.h" | |
39 | #include "hash.h" | |
d9a8717a | 40 | #include "hmap.h" |
cd11000b | 41 | #include "jsonrpc.h" |
064af421 BP |
42 | #include "list.h" |
43 | #include "mac-learning.h" | |
44 | #include "netdev.h" | |
45 | #include "odp-util.h" | |
46 | #include "ofp-print.h" | |
47 | #include "ofpbuf.h" | |
d65349ea | 48 | #include "ofproto/netflow.h" |
8cd4882f | 49 | #include "ofproto/ofproto.h" |
dd0d105c | 50 | #include "ovsdb-data.h" |
da285df4 | 51 | #include "packets.h" |
064af421 | 52 | #include "poll-loop.h" |
064af421 BP |
53 | #include "proc-net-compat.h" |
54 | #include "process.h" | |
76343538 | 55 | #include "sha1.h" |
6c88d577 | 56 | #include "shash.h" |
064af421 | 57 | #include "socket-util.h" |
fe55ad15 | 58 | #include "stream-ssl.h" |
064af421 | 59 | #include "svec.h" |
ce887677 | 60 | #include "system-stats.h" |
064af421 BP |
61 | #include "timeval.h" |
62 | #include "util.h" | |
da285df4 | 63 | #include "unixctl.h" |
064af421 | 64 | #include "vconn.h" |
b36682d8 | 65 | #include "vswitchd/vswitch-idl.h" |
064af421 | 66 | #include "xenserver.h" |
5136ce49 | 67 | #include "vlog.h" |
064af421 | 68 | #include "xtoxll.h" |
72b06300 | 69 | #include "sflow_api.h" |
064af421 | 70 | |
5136ce49 | 71 | VLOG_DEFINE_THIS_MODULE(bridge) |
064af421 BP |
72 | |
73 | struct dst { | |
74 | uint16_t vlan; | |
75 | uint16_t dp_ifidx; | |
76 | }; | |
77 | ||
064af421 | 78 | struct iface { |
0c6aea3f | 79 | /* These members are always valid. */ |
064af421 BP |
80 | struct port *port; /* Containing port. */ |
81 | size_t port_ifidx; /* Index within containing port. */ | |
064af421 | 82 | char *name; /* Host network device name. */ |
064af421 | 83 | tag_type tag; /* Tag associated with this interface. */ |
064af421 | 84 | long long delay_expires; /* Time after which 'enabled' may change. */ |
064af421 | 85 | |
0c6aea3f | 86 | /* These members are valid only after bridge_reconfigure() causes them to |
1e0b752d | 87 | * be initialized. */ |
d9a8717a | 88 | struct hmap_node dp_ifidx_node; /* In struct bridge's "ifaces" hmap. */ |
0c6aea3f BP |
89 | int dp_ifidx; /* Index within kernel datapath. */ |
90 | struct netdev *netdev; /* Network device. */ | |
064af421 | 91 | bool enabled; /* May be chosen for flows? */ |
6cefe1da | 92 | const char *type; /* Usually same as cfg->type. */ |
76343538 | 93 | const struct ovsrec_interface *cfg; |
064af421 BP |
94 | }; |
95 | ||
96 | #define BOND_MASK 0xff | |
97 | struct bond_entry { | |
98 | int iface_idx; /* Index of assigned iface, or -1 if none. */ | |
99 | uint64_t tx_bytes; /* Count of bytes recently transmitted. */ | |
100 | tag_type iface_tag; /* Tag associated with iface_idx. */ | |
101 | }; | |
102 | ||
103 | #define MAX_MIRRORS 32 | |
104 | typedef uint32_t mirror_mask_t; | |
105 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
106 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
107 | struct mirror { | |
108 | struct bridge *bridge; | |
109 | size_t idx; | |
110 | char *name; | |
dd0d105c | 111 | struct uuid uuid; /* UUID of this "mirror" record in database. */ |
064af421 BP |
112 | |
113 | /* Selection criteria. */ | |
37e7f427 BP |
114 | struct shash src_ports; /* Name is port name; data is always NULL. */ |
115 | struct shash dst_ports; /* Name is port name; data is always NULL. */ | |
064af421 BP |
116 | int *vlans; |
117 | size_t n_vlans; | |
118 | ||
119 | /* Output. */ | |
120 | struct port *out_port; | |
121 | int out_vlan; | |
122 | }; | |
123 | ||
124 | #define FLOOD_PORT ((struct port *) 1) /* The 'flood' output port. */ | |
125 | struct port { | |
126 | struct bridge *bridge; | |
127 | size_t port_idx; | |
128 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
3e9c481c BP |
129 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. |
130 | * NULL if all VLANs are trunked. */ | |
1e0b752d | 131 | const struct ovsrec_port *cfg; |
064af421 BP |
132 | char *name; |
133 | ||
134 | /* An ordinary bridge port has 1 interface. | |
135 | * A bridge port for bonding has at least 2 interfaces. */ | |
136 | struct iface **ifaces; | |
137 | size_t n_ifaces, allocated_ifaces; | |
138 | ||
139 | /* Bonding info. */ | |
140 | struct bond_entry *bond_hash; /* An array of (BOND_MASK + 1) elements. */ | |
141 | int active_iface; /* Ifidx on which bcasts accepted, or -1. */ | |
142 | tag_type active_iface_tag; /* Tag for bcast flows. */ | |
143 | tag_type no_ifaces_tag; /* Tag for flows when all ifaces disabled. */ | |
144 | int updelay, downdelay; /* Delay before iface goes up/down, in ms. */ | |
85c74638 | 145 | bool bond_compat_is_stale; /* Need to call port_update_bond_compat()? */ |
76343538 | 146 | bool bond_fake_iface; /* Fake a bond interface for legacy compat? */ |
e5ae7df8 | 147 | long long int bond_next_fake_iface_update; /* Time of next update. */ |
c8143c88 BP |
148 | int bond_rebalance_interval; /* Interval between rebalances, in ms. */ |
149 | long long int bond_next_rebalance; /* Next rebalancing time. */ | |
0bb5c3ec | 150 | struct netdev_monitor *monitor; /* Tracks carrier up/down status. */ |
064af421 BP |
151 | |
152 | /* Port mirroring info. */ | |
153 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
154 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
155 | bool is_mirror_output_port; /* Does port mirroring send frames here? */ | |
064af421 BP |
156 | }; |
157 | ||
158 | #define DP_MAX_PORTS 255 | |
159 | struct bridge { | |
160 | struct list node; /* Node in global list of bridges. */ | |
161 | char *name; /* User-specified arbitrary name. */ | |
93dfc067 | 162 | struct mac_learning *ml; /* MAC learning table. */ |
064af421 | 163 | uint8_t default_ea[ETH_ADDR_LEN]; /* Default MAC. */ |
1e0b752d | 164 | const struct ovsrec_bridge *cfg; |
064af421 | 165 | |
064af421 BP |
166 | /* OpenFlow switch processing. */ |
167 | struct ofproto *ofproto; /* OpenFlow switch. */ | |
168 | ||
169 | /* Kernel datapath information. */ | |
c228a364 | 170 | struct dpif *dpif; /* Datapath. */ |
d9a8717a | 171 | struct hmap ifaces; /* Contains "struct iface"s. */ |
064af421 BP |
172 | |
173 | /* Bridge ports. */ | |
174 | struct port **ports; | |
175 | size_t n_ports, allocated_ports; | |
4a1ee6ae | 176 | struct shash iface_by_name; /* "struct iface"s indexed by name. */ |
836fad5e | 177 | struct shash port_by_name; /* "struct port"s indexed by name. */ |
064af421 BP |
178 | |
179 | /* Bonding. */ | |
180 | bool has_bonded_ports; | |
064af421 BP |
181 | |
182 | /* Flow tracking. */ | |
183 | bool flush; | |
184 | ||
064af421 BP |
185 | /* Port mirroring. */ |
186 | struct mirror *mirrors[MAX_MIRRORS]; | |
064af421 BP |
187 | }; |
188 | ||
189 | /* List of all bridges. */ | |
190 | static struct list all_bridges = LIST_INITIALIZER(&all_bridges); | |
191 | ||
c5187f17 BP |
192 | /* OVSDB IDL used to obtain configuration. */ |
193 | static struct ovsdb_idl *idl; | |
194 | ||
cd0cd65f BP |
195 | /* Each time this timer expires, the bridge fetches systems and interface |
196 | * statistics and pushes them into the database. */ | |
197 | #define STATS_INTERVAL (5 * 1000) /* In milliseconds. */ | |
198 | static long long int stats_timer = LLONG_MIN; | |
018f1525 | 199 | |
1a6f1e2a | 200 | static struct bridge *bridge_create(const struct ovsrec_bridge *br_cfg); |
064af421 BP |
201 | static void bridge_destroy(struct bridge *); |
202 | static struct bridge *bridge_lookup(const char *name); | |
8ca79daa | 203 | static unixctl_cb_func bridge_unixctl_dump_flows; |
fa05809b | 204 | static unixctl_cb_func bridge_unixctl_reconnect; |
064af421 | 205 | static int bridge_run_one(struct bridge *); |
1a048029 | 206 | static size_t bridge_get_controllers(const struct bridge *br, |
76ce9432 | 207 | struct ovsrec_controller ***controllersp); |
1a048029 JP |
208 | static void bridge_reconfigure_one(struct bridge *); |
209 | static void bridge_reconfigure_remotes(struct bridge *, | |
11bbff1b BP |
210 | const struct sockaddr_in *managers, |
211 | size_t n_managers); | |
76343538 | 212 | static void bridge_get_all_ifaces(const struct bridge *, struct shash *ifaces); |
064af421 BP |
213 | static void bridge_fetch_dp_ifaces(struct bridge *); |
214 | static void bridge_flush(struct bridge *); | |
215 | static void bridge_pick_local_hw_addr(struct bridge *, | |
216 | uint8_t ea[ETH_ADDR_LEN], | |
07c318f4 | 217 | struct iface **hw_addr_iface); |
064af421 BP |
218 | static uint64_t bridge_pick_datapath_id(struct bridge *, |
219 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
07c318f4 BP |
220 | struct iface *hw_addr_iface); |
221 | static struct iface *bridge_get_local_iface(struct bridge *); | |
064af421 BP |
222 | static uint64_t dpid_from_hash(const void *, size_t nbytes); |
223 | ||
8ca79daa | 224 | static unixctl_cb_func bridge_unixctl_fdb_show; |
8c4c1387 | 225 | |
da285df4 | 226 | static void bond_init(void); |
064af421 BP |
227 | static void bond_run(struct bridge *); |
228 | static void bond_wait(struct bridge *); | |
229 | static void bond_rebalance_port(struct port *); | |
2303f3b2 | 230 | static void bond_send_learning_packets(struct port *); |
8b2a2f4a | 231 | static void bond_enable_slave(struct iface *iface, bool enable); |
064af421 | 232 | |
76343538 BP |
233 | static struct port *port_create(struct bridge *, const char *name); |
234 | static void port_reconfigure(struct port *, const struct ovsrec_port *); | |
4a1ee6ae | 235 | static void port_del_ifaces(struct port *, const struct ovsrec_port *); |
064af421 BP |
236 | static void port_destroy(struct port *); |
237 | static struct port *port_lookup(const struct bridge *, const char *name); | |
da285df4 | 238 | static struct iface *port_lookup_iface(const struct port *, const char *name); |
064af421 BP |
239 | static struct port *port_from_dp_ifidx(const struct bridge *, |
240 | uint16_t dp_ifidx); | |
241 | static void port_update_bond_compat(struct port *); | |
242 | static void port_update_vlan_compat(struct port *); | |
0c6aea3f | 243 | static void port_update_bonding(struct port *); |
064af421 | 244 | |
dd0d105c | 245 | static void mirror_create(struct bridge *, struct ovsrec_mirror *); |
064af421 BP |
246 | static void mirror_destroy(struct mirror *); |
247 | static void mirror_reconfigure(struct bridge *); | |
37e7f427 | 248 | static void mirror_reconfigure_one(struct mirror *, struct ovsrec_mirror *); |
064af421 BP |
249 | static bool vlan_is_mirrored(const struct mirror *, int vlan); |
250 | ||
d295e8e9 | 251 | static struct iface *iface_create(struct port *port, |
a740f0de | 252 | const struct ovsrec_interface *if_cfg); |
064af421 BP |
253 | static void iface_destroy(struct iface *); |
254 | static struct iface *iface_lookup(const struct bridge *, const char *name); | |
255 | static struct iface *iface_from_dp_ifidx(const struct bridge *, | |
256 | uint16_t dp_ifidx); | |
52df17e7 | 257 | static void iface_set_mac(struct iface *); |
bcd49a45 | 258 | static void iface_set_ofport(const struct ovsrec_interface *, int64_t ofport); |
c1c9c9c4 | 259 | static void iface_update_qos(struct iface *, const struct ovsrec_qos *); |
064af421 | 260 | |
43776b8f BP |
261 | static void shash_from_ovs_idl_map(char **keys, char **values, size_t n, |
262 | struct shash *); | |
263 | ||
064af421 BP |
264 | /* Hooks into ofproto processing. */ |
265 | static struct ofhooks bridge_ofhooks; | |
266 | \f | |
267 | /* Public functions. */ | |
268 | ||
c5187f17 BP |
269 | /* Initializes the bridge module, configuring it to obtain its configuration |
270 | * from an OVSDB server accessed over 'remote', which should be a string in a | |
271 | * form acceptable to ovsdb_idl_create(). */ | |
064af421 | 272 | void |
c5187f17 BP |
273 | bridge_init(const char *remote) |
274 | { | |
275 | /* Create connection to database. */ | |
276 | idl = ovsdb_idl_create(remote, &ovsrec_idl_class); | |
277 | ||
e85bbd75 BP |
278 | ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_cur_cfg); |
279 | ovsdb_idl_set_write_only(idl, &ovsrec_open_vswitch_col_statistics); | |
280 | ovsdb_idl_omit(idl, &ovsrec_open_vswitch_col_external_ids); | |
281 | ||
282 | ovsdb_idl_omit(idl, &ovsrec_bridge_col_external_ids); | |
283 | ||
284 | ovsdb_idl_omit(idl, &ovsrec_port_col_external_ids); | |
285 | ovsdb_idl_omit(idl, &ovsrec_port_col_fake_bridge); | |
286 | ||
287 | ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_ofport); | |
288 | ovsdb_idl_set_write_only(idl, &ovsrec_interface_col_statistics); | |
289 | ovsdb_idl_omit(idl, &ovsrec_interface_col_external_ids); | |
290 | ||
c5187f17 BP |
291 | /* Register unixctl commands. */ |
292 | unixctl_command_register("fdb/show", bridge_unixctl_fdb_show, NULL); | |
293 | unixctl_command_register("bridge/dump-flows", bridge_unixctl_dump_flows, | |
294 | NULL); | |
fa05809b BP |
295 | unixctl_command_register("bridge/reconnect", bridge_unixctl_reconnect, |
296 | NULL); | |
c5187f17 BP |
297 | bond_init(); |
298 | } | |
299 | ||
300 | /* Performs configuration that is only necessary once at ovs-vswitchd startup, | |
301 | * but for which the ovs-vswitchd configuration 'cfg' is required. */ | |
302 | static void | |
303 | bridge_configure_once(const struct ovsrec_open_vswitch *cfg) | |
064af421 | 304 | { |
c5187f17 | 305 | static bool already_configured_once; |
76343538 | 306 | struct svec bridge_names; |
1a6f1e2a | 307 | struct svec dpif_names, dpif_types; |
d3d22744 | 308 | size_t i; |
da285df4 | 309 | |
c5187f17 BP |
310 | /* Only do this once per ovs-vswitchd run. */ |
311 | if (already_configured_once) { | |
312 | return; | |
313 | } | |
314 | already_configured_once = true; | |
8c4c1387 | 315 | |
cd0cd65f | 316 | stats_timer = time_msec() + STATS_INTERVAL; |
018f1525 | 317 | |
c5187f17 | 318 | /* Get all the configured bridges' names from 'cfg' into 'bridge_names'. */ |
76343538 BP |
319 | svec_init(&bridge_names); |
320 | for (i = 0; i < cfg->n_bridges; i++) { | |
321 | svec_add(&bridge_names, cfg->bridges[i]->name); | |
322 | } | |
323 | svec_sort(&bridge_names); | |
324 | ||
c5187f17 BP |
325 | /* Iterate over all system dpifs and delete any of them that do not appear |
326 | * in 'cfg'. */ | |
d2a34569 | 327 | svec_init(&dpif_names); |
1a6f1e2a JG |
328 | svec_init(&dpif_types); |
329 | dp_enumerate_types(&dpif_types); | |
330 | for (i = 0; i < dpif_types.n; i++) { | |
c228a364 | 331 | struct dpif *dpif; |
d3d22744 | 332 | int retval; |
1a6f1e2a | 333 | size_t j; |
064af421 | 334 | |
1a6f1e2a | 335 | dp_enumerate_names(dpif_types.names[i], &dpif_names); |
d3d22744 | 336 | |
c5187f17 | 337 | /* For each dpif... */ |
1a6f1e2a JG |
338 | for (j = 0; j < dpif_names.n; j++) { |
339 | retval = dpif_open(dpif_names.names[j], dpif_types.names[i], &dpif); | |
340 | if (!retval) { | |
341 | struct svec all_names; | |
342 | size_t k; | |
343 | ||
c5187f17 | 344 | /* ...check whether any of its names is in 'bridge_names'. */ |
1a6f1e2a JG |
345 | svec_init(&all_names); |
346 | dpif_get_all_names(dpif, &all_names); | |
347 | for (k = 0; k < all_names.n; k++) { | |
348 | if (svec_contains(&bridge_names, all_names.names[k])) { | |
349 | goto found; | |
350 | } | |
d3d22744 | 351 | } |
c5187f17 BP |
352 | |
353 | /* No. Delete the dpif. */ | |
1a6f1e2a | 354 | dpif_delete(dpif); |
c5187f17 | 355 | |
1a6f1e2a JG |
356 | found: |
357 | svec_destroy(&all_names); | |
358 | dpif_close(dpif); | |
064af421 | 359 | } |
064af421 BP |
360 | } |
361 | } | |
7128cd6b | 362 | svec_destroy(&bridge_names); |
d0c9d304 | 363 | svec_destroy(&dpif_names); |
1a6f1e2a | 364 | svec_destroy(&dpif_types); |
064af421 BP |
365 | } |
366 | ||
43776b8f BP |
367 | /* Initializes 'options' and fills it with the options for 'if_cfg'. Merges |
368 | * keys from "options" and "other_config", preferring "options" keys over | |
369 | * "other_config" keys. | |
370 | * | |
371 | * The value strings in '*options' are taken directly from if_cfg, not copied, | |
372 | * so the caller should not modify or free them. */ | |
373 | static void | |
374 | iface_get_options(const struct ovsrec_interface *if_cfg, struct shash *options) | |
375 | { | |
376 | size_t i; | |
377 | ||
378 | shash_from_ovs_idl_map(if_cfg->key_options, if_cfg->value_options, | |
379 | if_cfg->n_options, options); | |
380 | ||
381 | for (i = 0; i < if_cfg->n_other_config; i++) { | |
382 | char *key = if_cfg->key_other_config[i]; | |
383 | char *value = if_cfg->value_other_config[i]; | |
384 | ||
385 | if (!shash_find_data(options, key)) { | |
386 | shash_add(options, key, value); | |
387 | } else { | |
388 | VLOG_WARN("%s: ignoring \"other_config\" key %s that conflicts " | |
389 | "with \"options\" key %s", if_cfg->name, key, key); | |
390 | } | |
391 | } | |
392 | } | |
393 | ||
6cefe1da BP |
394 | /* Returns the type of network device that 'iface' should have. (This is |
395 | * ordinarily the same type as the interface, but the network devices for | |
396 | * "internal" ports have type "system".) */ | |
397 | static const char * | |
398 | iface_get_netdev_type(const struct iface *iface) | |
399 | { | |
400 | return !strcmp(iface->type, "internal") ? "system" : iface->type; | |
401 | } | |
402 | ||
43776b8f | 403 | /* Attempt to create the network device for 'iface' through the netdev |
6c88d577 JP |
404 | * library. */ |
405 | static int | |
43776b8f | 406 | create_iface_netdev(struct iface *iface) |
6c88d577 | 407 | { |
43776b8f | 408 | struct netdev_options netdev_options; |
76343538 | 409 | struct shash options; |
43776b8f | 410 | int error; |
6c88d577 | 411 | |
43776b8f BP |
412 | memset(&netdev_options, 0, sizeof netdev_options); |
413 | netdev_options.name = iface->cfg->name; | |
6cefe1da | 414 | netdev_options.type = iface_get_netdev_type(iface); |
43776b8f BP |
415 | netdev_options.args = &options; |
416 | netdev_options.ethertype = NETDEV_ETH_TYPE_NONE; | |
6c88d577 | 417 | |
43776b8f | 418 | iface_get_options(iface->cfg, &options); |
d1984028 | 419 | |
43776b8f | 420 | error = netdev_open(&netdev_options, &iface->netdev); |
149f577a | 421 | |
43776b8f | 422 | if (iface->netdev) { |
85da620e | 423 | iface->enabled = netdev_get_carrier(iface->netdev); |
43776b8f | 424 | } |
149f577a | 425 | |
43776b8f | 426 | shash_destroy(&options); |
149f577a | 427 | |
43776b8f BP |
428 | return error; |
429 | } | |
149f577a | 430 | |
43776b8f BP |
431 | static int |
432 | reconfigure_iface_netdev(struct iface *iface) | |
433 | { | |
434 | const char *netdev_type, *iface_type; | |
435 | struct shash options; | |
436 | int error; | |
d4c2000b | 437 | |
43776b8f BP |
438 | /* Skip reconfiguration if the device has the wrong type. This shouldn't |
439 | * happen, but... */ | |
6cefe1da | 440 | iface_type = iface_get_netdev_type(iface); |
43776b8f BP |
441 | netdev_type = netdev_get_type(iface->netdev); |
442 | if (iface_type && strcmp(netdev_type, iface_type)) { | |
443 | VLOG_WARN("%s: attempting change device type from %s to %s", | |
444 | iface->cfg->name, netdev_type, iface_type); | |
445 | return EINVAL; | |
6c88d577 | 446 | } |
43776b8f BP |
447 | |
448 | /* Reconfigure device. */ | |
449 | iface_get_options(iface->cfg, &options); | |
450 | error = netdev_reconfigure(iface->netdev, &options); | |
451 | shash_destroy(&options); | |
6c88d577 JP |
452 | |
453 | return error; | |
454 | } | |
455 | ||
64d64dd7 | 456 | /* Callback for iterate_and_prune_ifaces(). */ |
0c6aea3f | 457 | static bool |
64d64dd7 | 458 | check_iface(struct bridge *br, struct iface *iface, void *aux OVS_UNUSED) |
0c6aea3f | 459 | { |
149f577a | 460 | if (!iface->netdev) { |
64d64dd7 BP |
461 | /* We already reported a related error, don't bother duplicating it. */ |
462 | return false; | |
0c6aea3f | 463 | } |
149f577a | 464 | |
64d64dd7 | 465 | if (iface->dp_ifidx < 0) { |
6ae39834 BP |
466 | VLOG_ERR("%s interface not in %s, dropping", |
467 | iface->name, dpif_name(br->dpif)); | |
468 | return false; | |
469 | } | |
64d64dd7 BP |
470 | |
471 | VLOG_DBG("%s has interface %s on port %d", dpif_name(br->dpif), | |
472 | iface->name, iface->dp_ifidx); | |
473 | return true; | |
6ae39834 BP |
474 | } |
475 | ||
64d64dd7 | 476 | /* Callback for iterate_and_prune_ifaces(). */ |
795fe1fb | 477 | static bool |
67a4917b BP |
478 | set_iface_properties(struct bridge *br OVS_UNUSED, struct iface *iface, |
479 | void *aux OVS_UNUSED) | |
795fe1fb | 480 | { |
4d678233 | 481 | /* Set policing attributes. */ |
76343538 BP |
482 | netdev_set_policing(iface->netdev, |
483 | iface->cfg->ingress_policing_rate, | |
484 | iface->cfg->ingress_policing_burst); | |
4d678233 BP |
485 | |
486 | /* Set MAC address of internal interfaces other than the local | |
487 | * interface. */ | |
6cefe1da | 488 | if (iface->dp_ifidx != ODPP_LOCAL && !strcmp(iface->type, "internal")) { |
4d678233 BP |
489 | iface_set_mac(iface); |
490 | } | |
491 | ||
795fe1fb BP |
492 | return true; |
493 | } | |
494 | ||
6ae39834 BP |
495 | /* Calls 'cb' for each interfaces in 'br', passing along the 'aux' argument. |
496 | * Deletes from 'br' all the interfaces for which 'cb' returns false, and then | |
497 | * deletes from 'br' any ports that no longer have any interfaces. */ | |
498 | static void | |
499 | iterate_and_prune_ifaces(struct bridge *br, | |
500 | bool (*cb)(struct bridge *, struct iface *, | |
501 | void *aux), | |
502 | void *aux) | |
503 | { | |
504 | size_t i, j; | |
505 | ||
506 | for (i = 0; i < br->n_ports; ) { | |
507 | struct port *port = br->ports[i]; | |
508 | for (j = 0; j < port->n_ifaces; ) { | |
509 | struct iface *iface = port->ifaces[j]; | |
510 | if (cb(br, iface, aux)) { | |
511 | j++; | |
512 | } else { | |
bcd49a45 | 513 | iface_set_ofport(iface->cfg, -1); |
6ae39834 BP |
514 | iface_destroy(iface); |
515 | } | |
516 | } | |
517 | ||
518 | if (port->n_ifaces) { | |
519 | i++; | |
520 | } else { | |
521 | VLOG_ERR("%s port has no interfaces, dropping", port->name); | |
522 | port_destroy(port); | |
523 | } | |
524 | } | |
525 | } | |
526 | ||
cd11000b BP |
527 | /* Looks at the list of managers in 'ovs_cfg' and extracts their remote IP |
528 | * addresses and ports into '*managersp' and '*n_managersp'. The caller is | |
529 | * responsible for freeing '*managersp' (with free()). | |
530 | * | |
531 | * You may be asking yourself "why does ovs-vswitchd care?", because | |
532 | * ovsdb-server is responsible for connecting to the managers, and ovs-vswitchd | |
533 | * should not be and in fact is not directly involved in that. But | |
534 | * ovs-vswitchd needs to make sure that ovsdb-server can reach the managers, so | |
535 | * it has to tell in-band control where the managers are to enable that. | |
536 | */ | |
537 | static void | |
538 | collect_managers(const struct ovsrec_open_vswitch *ovs_cfg, | |
539 | struct sockaddr_in **managersp, size_t *n_managersp) | |
540 | { | |
541 | struct sockaddr_in *managers = NULL; | |
542 | size_t n_managers = 0; | |
543 | ||
544 | if (ovs_cfg->n_managers > 0) { | |
545 | size_t i; | |
546 | ||
547 | managers = xmalloc(ovs_cfg->n_managers * sizeof *managers); | |
548 | for (i = 0; i < ovs_cfg->n_managers; i++) { | |
549 | const char *name = ovs_cfg->managers[i]; | |
550 | struct sockaddr_in *sin = &managers[i]; | |
551 | ||
552 | if ((!strncmp(name, "tcp:", 4) | |
553 | && inet_parse_active(name + 4, JSONRPC_TCP_PORT, sin)) || | |
554 | (!strncmp(name, "ssl:", 4) | |
555 | && inet_parse_active(name + 4, JSONRPC_SSL_PORT, sin))) { | |
556 | n_managers++; | |
557 | } | |
558 | } | |
559 | } | |
560 | ||
561 | *managersp = managers; | |
562 | *n_managersp = n_managers; | |
563 | } | |
564 | ||
c5187f17 | 565 | static void |
76343538 | 566 | bridge_reconfigure(const struct ovsrec_open_vswitch *ovs_cfg) |
064af421 | 567 | { |
76343538 BP |
568 | struct shash old_br, new_br; |
569 | struct shash_node *node; | |
064af421 | 570 | struct bridge *br, *next; |
cd11000b BP |
571 | struct sockaddr_in *managers; |
572 | size_t n_managers; | |
6ae39834 | 573 | size_t i; |
72b06300 | 574 | int sflow_bridge_number; |
064af421 BP |
575 | |
576 | COVERAGE_INC(bridge_reconfigure); | |
577 | ||
cd11000b BP |
578 | collect_managers(ovs_cfg, &managers, &n_managers); |
579 | ||
632d136c | 580 | /* Collect old and new bridges. */ |
76343538 BP |
581 | shash_init(&old_br); |
582 | shash_init(&new_br); | |
4e8e4213 | 583 | LIST_FOR_EACH (br, node, &all_bridges) { |
76343538 BP |
584 | shash_add(&old_br, br->name, br); |
585 | } | |
586 | for (i = 0; i < ovs_cfg->n_bridges; i++) { | |
587 | const struct ovsrec_bridge *br_cfg = ovs_cfg->bridges[i]; | |
588 | if (!shash_add_once(&new_br, br_cfg->name, br_cfg)) { | |
589 | VLOG_WARN("more than one bridge named %s", br_cfg->name); | |
590 | } | |
064af421 | 591 | } |
064af421 BP |
592 | |
593 | /* Get rid of deleted bridges and add new bridges. */ | |
4e8e4213 | 594 | LIST_FOR_EACH_SAFE (br, next, node, &all_bridges) { |
76343538 BP |
595 | struct ovsrec_bridge *br_cfg = shash_find_data(&new_br, br->name); |
596 | if (br_cfg) { | |
597 | br->cfg = br_cfg; | |
598 | } else { | |
064af421 BP |
599 | bridge_destroy(br); |
600 | } | |
601 | } | |
76343538 BP |
602 | SHASH_FOR_EACH (node, &new_br) { |
603 | const char *br_name = node->name; | |
604 | const struct ovsrec_bridge *br_cfg = node->data; | |
1a6f1e2a JG |
605 | br = shash_find_data(&old_br, br_name); |
606 | if (br) { | |
607 | /* If the bridge datapath type has changed, we need to tear it | |
608 | * down and recreate. */ | |
609 | if (strcmp(br->cfg->datapath_type, br_cfg->datapath_type)) { | |
610 | bridge_destroy(br); | |
611 | bridge_create(br_cfg); | |
fa33d64a | 612 | } |
1a6f1e2a JG |
613 | } else { |
614 | bridge_create(br_cfg); | |
064af421 BP |
615 | } |
616 | } | |
76343538 BP |
617 | shash_destroy(&old_br); |
618 | shash_destroy(&new_br); | |
064af421 | 619 | |
064af421 | 620 | /* Reconfigure all bridges. */ |
4e8e4213 | 621 | LIST_FOR_EACH (br, node, &all_bridges) { |
1a048029 | 622 | bridge_reconfigure_one(br); |
064af421 BP |
623 | } |
624 | ||
625 | /* Add and delete ports on all datapaths. | |
626 | * | |
627 | * The kernel will reject any attempt to add a given port to a datapath if | |
628 | * that port already belongs to a different datapath, so we must do all | |
629 | * port deletions before any port additions. */ | |
4e8e4213 | 630 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
631 | struct odp_port *dpif_ports; |
632 | size_t n_dpif_ports; | |
76343538 | 633 | struct shash want_ifaces; |
064af421 | 634 | |
c228a364 | 635 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
064af421 BP |
636 | bridge_get_all_ifaces(br, &want_ifaces); |
637 | for (i = 0; i < n_dpif_ports; i++) { | |
638 | const struct odp_port *p = &dpif_ports[i]; | |
76343538 | 639 | if (!shash_find(&want_ifaces, p->devname) |
064af421 | 640 | && strcmp(p->devname, br->name)) { |
c228a364 | 641 | int retval = dpif_port_del(br->dpif, p->port); |
064af421 | 642 | if (retval) { |
b29ba128 | 643 | VLOG_ERR("failed to remove %s interface from %s: %s", |
c228a364 | 644 | p->devname, dpif_name(br->dpif), |
b29ba128 | 645 | strerror(retval)); |
064af421 BP |
646 | } |
647 | } | |
648 | } | |
76343538 | 649 | shash_destroy(&want_ifaces); |
064af421 BP |
650 | free(dpif_ports); |
651 | } | |
4e8e4213 | 652 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
653 | struct odp_port *dpif_ports; |
654 | size_t n_dpif_ports; | |
76343538 | 655 | struct shash cur_ifaces, want_ifaces; |
064af421 | 656 | |
76343538 | 657 | /* Get the set of interfaces currently in this datapath. */ |
c228a364 | 658 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
76343538 | 659 | shash_init(&cur_ifaces); |
064af421 | 660 | for (i = 0; i < n_dpif_ports; i++) { |
76343538 | 661 | const char *name = dpif_ports[i].devname; |
3a6ccc8c | 662 | shash_add_once(&cur_ifaces, name, &dpif_ports[i]); |
064af421 | 663 | } |
064af421 | 664 | |
76343538 BP |
665 | /* Get the set of interfaces we want on this datapath. */ |
666 | bridge_get_all_ifaces(br, &want_ifaces); | |
6c88d577 | 667 | |
3a6ccc8c | 668 | hmap_clear(&br->ifaces); |
76343538 BP |
669 | SHASH_FOR_EACH (node, &want_ifaces) { |
670 | const char *if_name = node->name; | |
671 | struct iface *iface = node->data; | |
3a6ccc8c BP |
672 | bool internal = !iface || !strcmp(iface->type, "internal"); |
673 | struct odp_port *dpif_port = shash_find_data(&cur_ifaces, if_name); | |
674 | int error; | |
675 | ||
676 | /* If we have a port or a netdev already, and it's not the type we | |
677 | * want, then delete the port (if any) and close the netdev (if | |
678 | * any). */ | |
679 | if (internal | |
680 | ? dpif_port && !(dpif_port->flags & ODP_PORT_INTERNAL) | |
681 | : (iface->netdev | |
682 | && strcmp(iface->type, netdev_get_type(iface->netdev)))) | |
683 | { | |
684 | if (dpif_port) { | |
685 | error = ofproto_port_del(br->ofproto, dpif_port->port); | |
686 | if (error) { | |
687 | continue; | |
688 | } | |
689 | dpif_port = NULL; | |
76343538 | 690 | } |
3a6ccc8c BP |
691 | if (iface) { |
692 | netdev_close(iface->netdev); | |
693 | iface->netdev = NULL; | |
694 | } | |
695 | } | |
76343538 | 696 | |
3a6ccc8c BP |
697 | /* If it's not an internal port, open (possibly create) the |
698 | * netdev. */ | |
699 | if (!internal) { | |
700 | if (!iface->netdev) { | |
82057f51 BP |
701 | error = create_iface_netdev(iface); |
702 | if (error) { | |
703 | VLOG_WARN("could not create iface %s: %s", iface->name, | |
704 | strerror(error)); | |
3a6ccc8c | 705 | continue; |
82057f51 | 706 | } |
3a6ccc8c BP |
707 | } else { |
708 | reconfigure_iface_netdev(iface); | |
82057f51 | 709 | } |
3a6ccc8c | 710 | } |
82057f51 | 711 | |
3a6ccc8c BP |
712 | /* If it's not part of the datapath, add it. */ |
713 | if (!dpif_port) { | |
76343538 BP |
714 | error = dpif_port_add(br->dpif, if_name, |
715 | internal ? ODP_PORT_INTERNAL : 0, NULL); | |
716 | if (error == EFBIG) { | |
717 | VLOG_ERR("ran out of valid port numbers on %s", | |
718 | dpif_name(br->dpif)); | |
719 | break; | |
720 | } else if (error) { | |
721 | VLOG_ERR("failed to add %s interface to %s: %s", | |
722 | if_name, dpif_name(br->dpif), strerror(error)); | |
3a6ccc8c | 723 | continue; |
76343538 | 724 | } |
064af421 | 725 | } |
3a6ccc8c BP |
726 | |
727 | /* If it's an internal port, open the netdev. */ | |
728 | if (internal) { | |
729 | if (iface && !iface->netdev) { | |
730 | error = create_iface_netdev(iface); | |
731 | if (error) { | |
732 | VLOG_WARN("could not create iface %s: %s", iface->name, | |
733 | strerror(error)); | |
734 | continue; | |
735 | } | |
736 | } | |
737 | } else { | |
738 | assert(iface->netdev != NULL); | |
739 | } | |
064af421 | 740 | } |
3a6ccc8c | 741 | free(dpif_ports); |
76343538 BP |
742 | shash_destroy(&cur_ifaces); |
743 | shash_destroy(&want_ifaces); | |
064af421 | 744 | } |
72b06300 | 745 | sflow_bridge_number = 0; |
4e8e4213 | 746 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
747 | uint8_t ea[8]; |
748 | uint64_t dpid; | |
07c318f4 BP |
749 | struct iface *local_iface; |
750 | struct iface *hw_addr_iface; | |
093e47f4 | 751 | char *dpid_string; |
064af421 | 752 | |
064af421 | 753 | bridge_fetch_dp_ifaces(br); |
064af421 | 754 | |
64d64dd7 | 755 | iterate_and_prune_ifaces(br, check_iface, NULL); |
064af421 BP |
756 | |
757 | /* Pick local port hardware address, datapath ID. */ | |
07c318f4 BP |
758 | bridge_pick_local_hw_addr(br, ea, &hw_addr_iface); |
759 | local_iface = bridge_get_local_iface(br); | |
064af421 | 760 | if (local_iface) { |
07c318f4 | 761 | int error = netdev_set_etheraddr(local_iface->netdev, ea); |
064af421 BP |
762 | if (error) { |
763 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
764 | VLOG_ERR_RL(&rl, "bridge %s: failed to set bridge " | |
765 | "Ethernet address: %s", | |
766 | br->name, strerror(error)); | |
767 | } | |
768 | } | |
769 | ||
07c318f4 | 770 | dpid = bridge_pick_datapath_id(br, ea, hw_addr_iface); |
064af421 BP |
771 | ofproto_set_datapath_id(br->ofproto, dpid); |
772 | ||
fc09119a | 773 | dpid_string = xasprintf("%016"PRIx64, dpid); |
093e47f4 BP |
774 | ovsrec_bridge_set_datapath_id(br->cfg, dpid_string); |
775 | free(dpid_string); | |
776 | ||
064af421 | 777 | /* Set NetFlow configuration on this bridge. */ |
76343538 BP |
778 | if (br->cfg->netflow) { |
779 | struct ovsrec_netflow *nf_cfg = br->cfg->netflow; | |
780 | struct netflow_options opts; | |
781 | ||
782 | memset(&opts, 0, sizeof opts); | |
783 | ||
784 | dpif_get_netflow_ids(br->dpif, &opts.engine_type, &opts.engine_id); | |
785 | if (nf_cfg->engine_type) { | |
cd746526 | 786 | opts.engine_type = *nf_cfg->engine_type; |
76343538 BP |
787 | } |
788 | if (nf_cfg->engine_id) { | |
cd746526 | 789 | opts.engine_id = *nf_cfg->engine_id; |
76343538 BP |
790 | } |
791 | ||
792 | opts.active_timeout = nf_cfg->active_timeout; | |
793 | if (!opts.active_timeout) { | |
794 | opts.active_timeout = -1; | |
795 | } else if (opts.active_timeout < 0) { | |
e9e2856e JG |
796 | VLOG_WARN("bridge %s: active timeout interval set to negative " |
797 | "value, using default instead (%d seconds)", br->name, | |
798 | NF_ACTIVE_TIMEOUT_DEFAULT); | |
799 | opts.active_timeout = -1; | |
76343538 BP |
800 | } |
801 | ||
802 | opts.add_id_to_iface = nf_cfg->add_id_to_interface; | |
803 | if (opts.add_id_to_iface) { | |
804 | if (opts.engine_id > 0x7f) { | |
805 | VLOG_WARN("bridge %s: netflow port mangling may conflict " | |
806 | "with another vswitch, choose an engine id less " | |
807 | "than 128", br->name); | |
808 | } | |
809 | if (br->n_ports > 508) { | |
810 | VLOG_WARN("bridge %s: netflow port mangling will conflict " | |
811 | "with another port when more than 508 ports are " | |
812 | "used", br->name); | |
813 | } | |
814 | } | |
815 | ||
816 | opts.collectors.n = nf_cfg->n_targets; | |
817 | opts.collectors.names = nf_cfg->targets; | |
818 | if (ofproto_set_netflow(br->ofproto, &opts)) { | |
d295e8e9 | 819 | VLOG_ERR("bridge %s: problem setting netflow collectors", |
76343538 BP |
820 | br->name); |
821 | } | |
822 | } else { | |
823 | ofproto_set_netflow(br->ofproto, NULL); | |
824 | } | |
064af421 | 825 | |
a4af0040 JP |
826 | /* Set sFlow configuration on this bridge. */ |
827 | if (br->cfg->sflow) { | |
d49d354b | 828 | const struct ovsrec_sflow *sflow_cfg = br->cfg->sflow; |
76ce9432 | 829 | struct ovsrec_controller **controllers; |
72b06300 | 830 | struct ofproto_sflow_options oso; |
76ce9432 | 831 | size_t n_controllers; |
72b06300 | 832 | |
a4af0040 JP |
833 | memset(&oso, 0, sizeof oso); |
834 | ||
835 | oso.targets.n = sflow_cfg->n_targets; | |
836 | oso.targets.names = sflow_cfg->targets; | |
72b06300 BP |
837 | |
838 | oso.sampling_rate = SFL_DEFAULT_SAMPLING_RATE; | |
a4af0040 JP |
839 | if (sflow_cfg->sampling) { |
840 | oso.sampling_rate = *sflow_cfg->sampling; | |
72b06300 BP |
841 | } |
842 | ||
843 | oso.polling_interval = SFL_DEFAULT_POLLING_INTERVAL; | |
a4af0040 JP |
844 | if (sflow_cfg->polling) { |
845 | oso.polling_interval = *sflow_cfg->polling; | |
72b06300 BP |
846 | } |
847 | ||
848 | oso.header_len = SFL_DEFAULT_HEADER_SIZE; | |
a4af0040 JP |
849 | if (sflow_cfg->header) { |
850 | oso.header_len = *sflow_cfg->header; | |
72b06300 BP |
851 | } |
852 | ||
853 | oso.sub_id = sflow_bridge_number++; | |
a4af0040 JP |
854 | oso.agent_device = sflow_cfg->agent; |
855 | ||
76ce9432 | 856 | oso.control_ip = NULL; |
1a048029 | 857 | n_controllers = bridge_get_controllers(br, &controllers); |
76ce9432 BP |
858 | for (i = 0; i < n_controllers; i++) { |
859 | if (controllers[i]->local_ip) { | |
860 | oso.control_ip = controllers[i]->local_ip; | |
861 | break; | |
862 | } | |
863 | } | |
72b06300 BP |
864 | ofproto_set_sflow(br->ofproto, &oso); |
865 | ||
7b99db05 | 866 | /* Do not destroy oso.targets because it is owned by sflow_cfg. */ |
72b06300 BP |
867 | } else { |
868 | ofproto_set_sflow(br->ofproto, NULL); | |
869 | } | |
870 | ||
064af421 BP |
871 | /* Update the controller and related settings. It would be more |
872 | * straightforward to call this from bridge_reconfigure_one(), but we | |
873 | * can't do it there for two reasons. First, and most importantly, at | |
874 | * that point we don't know the dp_ifidx of any interfaces that have | |
875 | * been added to the bridge (because we haven't actually added them to | |
876 | * the datapath). Second, at that point we haven't set the datapath ID | |
877 | * yet; when a controller is configured, resetting the datapath ID will | |
878 | * immediately disconnect from the controller, so it's better to set | |
879 | * the datapath ID before the controller. */ | |
1a048029 | 880 | bridge_reconfigure_remotes(br, managers, n_managers); |
064af421 | 881 | } |
4e8e4213 | 882 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
883 | for (i = 0; i < br->n_ports; i++) { |
884 | struct port *port = br->ports[i]; | |
c1c9c9c4 | 885 | int j; |
52df17e7 | 886 | |
064af421 | 887 | port_update_vlan_compat(port); |
0c6aea3f | 888 | port_update_bonding(port); |
c1c9c9c4 BP |
889 | |
890 | for (j = 0; j < port->n_ifaces; j++) { | |
891 | iface_update_qos(port->ifaces[j], port->cfg->qos); | |
892 | } | |
064af421 BP |
893 | } |
894 | } | |
4e8e4213 | 895 | LIST_FOR_EACH (br, node, &all_bridges) { |
4d678233 | 896 | iterate_and_prune_ifaces(br, set_iface_properties, NULL); |
064af421 | 897 | } |
093e47f4 | 898 | |
cd11000b | 899 | free(managers); |
093e47f4 BP |
900 | } |
901 | ||
902 | static const char * | |
af6278e1 BP |
903 | get_ovsrec_key_value(const struct ovsdb_idl_row *row, |
904 | const struct ovsdb_idl_column *column, | |
905 | const char *key) | |
093e47f4 | 906 | { |
af6278e1 BP |
907 | const struct ovsdb_datum *datum; |
908 | union ovsdb_atom atom; | |
909 | unsigned int idx; | |
093e47f4 | 910 | |
af6278e1 BP |
911 | datum = ovsdb_idl_get(row, column, OVSDB_TYPE_STRING, OVSDB_TYPE_STRING); |
912 | atom.string = (char *) key; | |
913 | idx = ovsdb_datum_find_key(datum, &atom, OVSDB_TYPE_STRING); | |
914 | return idx == UINT_MAX ? NULL : datum->values[idx].string; | |
064af421 BP |
915 | } |
916 | ||
c8143c88 BP |
917 | static const char * |
918 | bridge_get_other_config(const struct ovsrec_bridge *br_cfg, const char *key) | |
919 | { | |
af6278e1 BP |
920 | return get_ovsrec_key_value(&br_cfg->header_, |
921 | &ovsrec_bridge_col_other_config, key); | |
c8143c88 BP |
922 | } |
923 | ||
064af421 BP |
924 | static void |
925 | bridge_pick_local_hw_addr(struct bridge *br, uint8_t ea[ETH_ADDR_LEN], | |
07c318f4 | 926 | struct iface **hw_addr_iface) |
064af421 | 927 | { |
093e47f4 | 928 | const char *hwaddr; |
064af421 BP |
929 | size_t i, j; |
930 | int error; | |
931 | ||
07c318f4 | 932 | *hw_addr_iface = NULL; |
064af421 BP |
933 | |
934 | /* Did the user request a particular MAC? */ | |
093e47f4 BP |
935 | hwaddr = bridge_get_other_config(br->cfg, "hwaddr"); |
936 | if (hwaddr && eth_addr_from_string(hwaddr, ea)) { | |
064af421 BP |
937 | if (eth_addr_is_multicast(ea)) { |
938 | VLOG_ERR("bridge %s: cannot set MAC address to multicast " | |
939 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
940 | } else if (eth_addr_is_zero(ea)) { | |
941 | VLOG_ERR("bridge %s: cannot set MAC address to zero", br->name); | |
942 | } else { | |
943 | return; | |
944 | } | |
945 | } | |
946 | ||
141f4942 BP |
947 | /* Otherwise choose the minimum non-local MAC address among all of the |
948 | * interfaces. */ | |
064af421 BP |
949 | memset(ea, 0xff, sizeof ea); |
950 | for (i = 0; i < br->n_ports; i++) { | |
951 | struct port *port = br->ports[i]; | |
58b7527e | 952 | uint8_t iface_ea[ETH_ADDR_LEN]; |
58b7527e BP |
953 | struct iface *iface; |
954 | ||
955 | /* Mirror output ports don't participate. */ | |
064af421 BP |
956 | if (port->is_mirror_output_port) { |
957 | continue; | |
958 | } | |
58b7527e BP |
959 | |
960 | /* Choose the MAC address to represent the port. */ | |
76343538 | 961 | if (port->cfg->mac && eth_addr_from_string(port->cfg->mac, iface_ea)) { |
ba09980a BP |
962 | /* Find the interface with this Ethernet address (if any) so that |
963 | * we can provide the correct devname to the caller. */ | |
964 | iface = NULL; | |
965 | for (j = 0; j < port->n_ifaces; j++) { | |
966 | struct iface *candidate = port->ifaces[j]; | |
967 | uint8_t candidate_ea[ETH_ADDR_LEN]; | |
8fef8c71 | 968 | if (!netdev_get_etheraddr(candidate->netdev, candidate_ea) |
ba09980a BP |
969 | && eth_addr_equals(iface_ea, candidate_ea)) { |
970 | iface = candidate; | |
971 | } | |
972 | } | |
58b7527e BP |
973 | } else { |
974 | /* Choose the interface whose MAC address will represent the port. | |
975 | * The Linux kernel bonding code always chooses the MAC address of | |
976 | * the first slave added to a bond, and the Fedora networking | |
977 | * scripts always add slaves to a bond in alphabetical order, so | |
978 | * for compatibility we choose the interface with the name that is | |
979 | * first in alphabetical order. */ | |
980 | iface = port->ifaces[0]; | |
981 | for (j = 1; j < port->n_ifaces; j++) { | |
982 | struct iface *candidate = port->ifaces[j]; | |
983 | if (strcmp(candidate->name, iface->name) < 0) { | |
984 | iface = candidate; | |
985 | } | |
986 | } | |
987 | ||
988 | /* The local port doesn't count (since we're trying to choose its | |
141f4942 BP |
989 | * MAC address anyway). */ |
990 | if (iface->dp_ifidx == ODPP_LOCAL) { | |
064af421 BP |
991 | continue; |
992 | } | |
58b7527e BP |
993 | |
994 | /* Grab MAC. */ | |
07c318f4 | 995 | error = netdev_get_etheraddr(iface->netdev, iface_ea); |
58b7527e | 996 | if (error) { |
064af421 BP |
997 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
998 | VLOG_ERR_RL(&rl, "failed to obtain Ethernet address of %s: %s", | |
999 | iface->name, strerror(error)); | |
58b7527e | 1000 | continue; |
064af421 BP |
1001 | } |
1002 | } | |
58b7527e BP |
1003 | |
1004 | /* Compare against our current choice. */ | |
1005 | if (!eth_addr_is_multicast(iface_ea) && | |
141f4942 | 1006 | !eth_addr_is_local(iface_ea) && |
58b7527e BP |
1007 | !eth_addr_is_reserved(iface_ea) && |
1008 | !eth_addr_is_zero(iface_ea) && | |
1009 | memcmp(iface_ea, ea, ETH_ADDR_LEN) < 0) | |
1010 | { | |
0babc06f | 1011 | memcpy(ea, iface_ea, ETH_ADDR_LEN); |
8fef8c71 | 1012 | *hw_addr_iface = iface; |
58b7527e | 1013 | } |
064af421 | 1014 | } |
141f4942 | 1015 | if (eth_addr_is_multicast(ea)) { |
064af421 | 1016 | memcpy(ea, br->default_ea, ETH_ADDR_LEN); |
07c318f4 | 1017 | *hw_addr_iface = NULL; |
064af421 BP |
1018 | VLOG_WARN("bridge %s: using default bridge Ethernet " |
1019 | "address "ETH_ADDR_FMT, br->name, ETH_ADDR_ARGS(ea)); | |
1020 | } else { | |
1021 | VLOG_DBG("bridge %s: using bridge Ethernet address "ETH_ADDR_FMT, | |
1022 | br->name, ETH_ADDR_ARGS(ea)); | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | /* Choose and returns the datapath ID for bridge 'br' given that the bridge | |
1027 | * Ethernet address is 'bridge_ea'. If 'bridge_ea' is the Ethernet address of | |
07c318f4 BP |
1028 | * an interface on 'br', then that interface must be passed in as |
1029 | * 'hw_addr_iface'; if 'bridge_ea' was derived some other way, then | |
1030 | * 'hw_addr_iface' must be passed in as a null pointer. */ | |
064af421 BP |
1031 | static uint64_t |
1032 | bridge_pick_datapath_id(struct bridge *br, | |
1033 | const uint8_t bridge_ea[ETH_ADDR_LEN], | |
07c318f4 | 1034 | struct iface *hw_addr_iface) |
064af421 BP |
1035 | { |
1036 | /* | |
1037 | * The procedure for choosing a bridge MAC address will, in the most | |
1038 | * ordinary case, also choose a unique MAC that we can use as a datapath | |
1039 | * ID. In some special cases, though, multiple bridges will end up with | |
1040 | * the same MAC address. This is OK for the bridges, but it will confuse | |
1041 | * the OpenFlow controller, because each datapath needs a unique datapath | |
1042 | * ID. | |
1043 | * | |
1044 | * Datapath IDs must be unique. It is also very desirable that they be | |
1045 | * stable from one run to the next, so that policy set on a datapath | |
1046 | * "sticks". | |
1047 | */ | |
093e47f4 | 1048 | const char *datapath_id; |
064af421 BP |
1049 | uint64_t dpid; |
1050 | ||
093e47f4 BP |
1051 | datapath_id = bridge_get_other_config(br->cfg, "datapath-id"); |
1052 | if (datapath_id && dpid_from_string(datapath_id, &dpid)) { | |
064af421 BP |
1053 | return dpid; |
1054 | } | |
1055 | ||
07c318f4 | 1056 | if (hw_addr_iface) { |
064af421 | 1057 | int vlan; |
b2f17b15 | 1058 | if (!netdev_get_vlan_vid(hw_addr_iface->netdev, &vlan)) { |
064af421 BP |
1059 | /* |
1060 | * A bridge whose MAC address is taken from a VLAN network device | |
1061 | * (that is, a network device created with vconfig(8) or similar | |
1062 | * tool) will have the same MAC address as a bridge on the VLAN | |
1063 | * device's physical network device. | |
1064 | * | |
1065 | * Handle this case by hashing the physical network device MAC | |
1066 | * along with the VLAN identifier. | |
1067 | */ | |
1068 | uint8_t buf[ETH_ADDR_LEN + 2]; | |
1069 | memcpy(buf, bridge_ea, ETH_ADDR_LEN); | |
1070 | buf[ETH_ADDR_LEN] = vlan >> 8; | |
1071 | buf[ETH_ADDR_LEN + 1] = vlan; | |
1072 | return dpid_from_hash(buf, sizeof buf); | |
1073 | } else { | |
1074 | /* | |
1075 | * Assume that this bridge's MAC address is unique, since it | |
1076 | * doesn't fit any of the cases we handle specially. | |
1077 | */ | |
1078 | } | |
1079 | } else { | |
1080 | /* | |
1081 | * A purely internal bridge, that is, one that has no non-virtual | |
1082 | * network devices on it at all, is more difficult because it has no | |
1083 | * natural unique identifier at all. | |
1084 | * | |
1085 | * When the host is a XenServer, we handle this case by hashing the | |
1086 | * host's UUID with the name of the bridge. Names of bridges are | |
1087 | * persistent across XenServer reboots, although they can be reused if | |
1088 | * an internal network is destroyed and then a new one is later | |
1089 | * created, so this is fairly effective. | |
1090 | * | |
1091 | * When the host is not a XenServer, we punt by using a random MAC | |
1092 | * address on each run. | |
1093 | */ | |
1094 | const char *host_uuid = xenserver_get_host_uuid(); | |
1095 | if (host_uuid) { | |
1096 | char *combined = xasprintf("%s,%s", host_uuid, br->name); | |
1097 | dpid = dpid_from_hash(combined, strlen(combined)); | |
1098 | free(combined); | |
1099 | return dpid; | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | return eth_addr_to_uint64(bridge_ea); | |
1104 | } | |
1105 | ||
1106 | static uint64_t | |
1107 | dpid_from_hash(const void *data, size_t n) | |
1108 | { | |
5eccf359 | 1109 | uint8_t hash[SHA1_DIGEST_SIZE]; |
064af421 BP |
1110 | |
1111 | BUILD_ASSERT_DECL(sizeof hash >= ETH_ADDR_LEN); | |
5eccf359 | 1112 | sha1_bytes(data, n, hash); |
064af421 BP |
1113 | eth_addr_mark_random(hash); |
1114 | return eth_addr_to_uint64(hash); | |
1115 | } | |
1116 | ||
018f1525 BP |
1117 | static void |
1118 | iface_refresh_stats(struct iface *iface) | |
1119 | { | |
1120 | struct iface_stat { | |
1121 | char *name; | |
1122 | int offset; | |
1123 | }; | |
1124 | static const struct iface_stat iface_stats[] = { | |
1125 | { "rx_packets", offsetof(struct netdev_stats, rx_packets) }, | |
1126 | { "tx_packets", offsetof(struct netdev_stats, tx_packets) }, | |
1127 | { "rx_bytes", offsetof(struct netdev_stats, rx_bytes) }, | |
1128 | { "tx_bytes", offsetof(struct netdev_stats, tx_bytes) }, | |
1129 | { "rx_dropped", offsetof(struct netdev_stats, rx_dropped) }, | |
1130 | { "tx_dropped", offsetof(struct netdev_stats, tx_dropped) }, | |
1131 | { "rx_errors", offsetof(struct netdev_stats, rx_errors) }, | |
1132 | { "tx_errors", offsetof(struct netdev_stats, tx_errors) }, | |
1133 | { "rx_frame_err", offsetof(struct netdev_stats, rx_frame_errors) }, | |
1134 | { "rx_over_err", offsetof(struct netdev_stats, rx_over_errors) }, | |
1135 | { "rx_crc_err", offsetof(struct netdev_stats, rx_crc_errors) }, | |
1136 | { "collisions", offsetof(struct netdev_stats, collisions) }, | |
1137 | }; | |
1138 | enum { N_STATS = ARRAY_SIZE(iface_stats) }; | |
1139 | const struct iface_stat *s; | |
1140 | ||
1141 | char *keys[N_STATS]; | |
1142 | int64_t values[N_STATS]; | |
1143 | int n; | |
1144 | ||
1145 | struct netdev_stats stats; | |
1146 | ||
1147 | /* Intentionally ignore return value, since errors will set 'stats' to | |
1148 | * all-1s, and we will deal with that correctly below. */ | |
1149 | netdev_get_stats(iface->netdev, &stats); | |
1150 | ||
1151 | n = 0; | |
1152 | for (s = iface_stats; s < &iface_stats[N_STATS]; s++) { | |
1153 | uint64_t value = *(uint64_t *) (((char *) &stats) + s->offset); | |
1154 | if (value != UINT64_MAX) { | |
1155 | keys[n] = s->name; | |
1156 | values[n] = value; | |
1157 | n++; | |
1158 | } | |
1159 | } | |
1160 | ||
1161 | ovsrec_interface_set_statistics(iface->cfg, keys, values, n); | |
1162 | } | |
1163 | ||
ce887677 BP |
1164 | static void |
1165 | refresh_system_stats(const struct ovsrec_open_vswitch *cfg) | |
1166 | { | |
1167 | struct ovsdb_datum datum; | |
1168 | struct shash stats; | |
1169 | ||
1170 | shash_init(&stats); | |
1171 | get_system_stats(&stats); | |
1172 | ||
1173 | ovsdb_datum_from_shash(&datum, &stats); | |
1174 | ovsdb_idl_txn_write(&cfg->header_, &ovsrec_open_vswitch_col_statistics, | |
1175 | &datum); | |
1176 | } | |
1177 | ||
c5187f17 | 1178 | void |
064af421 BP |
1179 | bridge_run(void) |
1180 | { | |
d54ff998 BP |
1181 | const struct ovsrec_open_vswitch *cfg; |
1182 | ||
c5187f17 | 1183 | bool datapath_destroyed; |
d54ff998 | 1184 | bool database_changed; |
c5187f17 | 1185 | struct bridge *br; |
064af421 | 1186 | |
c5187f17 BP |
1187 | /* Let each bridge do the work that it needs to do. */ |
1188 | datapath_destroyed = false; | |
4e8e4213 | 1189 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
1190 | int error = bridge_run_one(br); |
1191 | if (error) { | |
1192 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1193 | VLOG_ERR_RL(&rl, "bridge %s: datapath was destroyed externally, " | |
1194 | "forcing reconfiguration", br->name); | |
c5187f17 BP |
1195 | datapath_destroyed = true; |
1196 | } | |
1197 | } | |
1198 | ||
1199 | /* (Re)configure if necessary. */ | |
d54ff998 BP |
1200 | database_changed = ovsdb_idl_run(idl); |
1201 | cfg = ovsrec_open_vswitch_first(idl); | |
1202 | if (database_changed || datapath_destroyed) { | |
c5187f17 BP |
1203 | if (cfg) { |
1204 | struct ovsdb_idl_txn *txn = ovsdb_idl_txn_create(idl); | |
1205 | ||
1206 | bridge_configure_once(cfg); | |
1207 | bridge_reconfigure(cfg); | |
1208 | ||
1209 | ovsrec_open_vswitch_set_cur_cfg(cfg, cfg->next_cfg); | |
1210 | ovsdb_idl_txn_commit(txn); | |
1211 | ovsdb_idl_txn_destroy(txn); /* XXX */ | |
1e0b752d BP |
1212 | } else { |
1213 | /* We still need to reconfigure to avoid dangling pointers to | |
1214 | * now-destroyed ovsrec structures inside bridge data. */ | |
1215 | static const struct ovsrec_open_vswitch null_cfg; | |
1216 | ||
1217 | bridge_reconfigure(&null_cfg); | |
064af421 BP |
1218 | } |
1219 | } | |
018f1525 | 1220 | |
d54ff998 BP |
1221 | #ifdef HAVE_OPENSSL |
1222 | /* Re-configure SSL. We do this on every trip through the main loop, | |
1223 | * instead of just when the database changes, because the contents of the | |
1224 | * key and certificate files can change without the database changing. */ | |
1225 | if (cfg && cfg->ssl) { | |
1226 | const struct ovsrec_ssl *ssl = cfg->ssl; | |
1227 | ||
1228 | stream_ssl_set_key_and_cert(ssl->private_key, ssl->certificate); | |
1229 | stream_ssl_set_ca_cert_file(ssl->ca_cert, ssl->bootstrap_ca_cert); | |
1230 | } | |
1231 | #endif | |
1232 | ||
cd0cd65f BP |
1233 | /* Refresh system and interface stats if necessary. */ |
1234 | if (time_msec() >= stats_timer) { | |
ce887677 BP |
1235 | if (cfg) { |
1236 | struct ovsdb_idl_txn *txn; | |
018f1525 | 1237 | |
ce887677 | 1238 | txn = ovsdb_idl_txn_create(idl); |
4e8e4213 | 1239 | LIST_FOR_EACH (br, node, &all_bridges) { |
ce887677 | 1240 | size_t i; |
018f1525 | 1241 | |
ce887677 BP |
1242 | for (i = 0; i < br->n_ports; i++) { |
1243 | struct port *port = br->ports[i]; | |
1244 | size_t j; | |
018f1525 | 1245 | |
ce887677 BP |
1246 | for (j = 0; j < port->n_ifaces; j++) { |
1247 | struct iface *iface = port->ifaces[j]; | |
1248 | iface_refresh_stats(iface); | |
1249 | } | |
018f1525 BP |
1250 | } |
1251 | } | |
ce887677 BP |
1252 | refresh_system_stats(cfg); |
1253 | ovsdb_idl_txn_commit(txn); | |
1254 | ovsdb_idl_txn_destroy(txn); /* XXX */ | |
018f1525 | 1255 | } |
018f1525 | 1256 | |
cd0cd65f | 1257 | stats_timer = time_msec() + STATS_INTERVAL; |
018f1525 | 1258 | } |
064af421 BP |
1259 | } |
1260 | ||
1261 | void | |
1262 | bridge_wait(void) | |
1263 | { | |
1264 | struct bridge *br; | |
1265 | ||
4e8e4213 | 1266 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 | 1267 | ofproto_wait(br->ofproto); |
7d674866 | 1268 | if (ofproto_has_primary_controller(br->ofproto)) { |
064af421 BP |
1269 | continue; |
1270 | } | |
1271 | ||
93dfc067 | 1272 | mac_learning_wait(br->ml); |
064af421 | 1273 | bond_wait(br); |
064af421 | 1274 | } |
c5187f17 | 1275 | ovsdb_idl_wait(idl); |
cd0cd65f | 1276 | poll_timer_wait_until(stats_timer); |
064af421 BP |
1277 | } |
1278 | ||
1279 | /* Forces 'br' to revalidate all of its flows. This is appropriate when 'br''s | |
1280 | * configuration changes. */ | |
1281 | static void | |
1282 | bridge_flush(struct bridge *br) | |
1283 | { | |
1284 | COVERAGE_INC(bridge_flush); | |
1285 | br->flush = true; | |
93dfc067 | 1286 | mac_learning_flush(br->ml); |
064af421 | 1287 | } |
07c318f4 BP |
1288 | |
1289 | /* Returns the 'br' interface for the ODPP_LOCAL port, or null if 'br' has no | |
1290 | * such interface. */ | |
1291 | static struct iface * | |
1292 | bridge_get_local_iface(struct bridge *br) | |
1293 | { | |
1294 | size_t i, j; | |
1295 | ||
1296 | for (i = 0; i < br->n_ports; i++) { | |
1297 | struct port *port = br->ports[i]; | |
1298 | for (j = 0; j < port->n_ifaces; j++) { | |
1299 | struct iface *iface = port->ifaces[j]; | |
1300 | if (iface->dp_ifidx == ODPP_LOCAL) { | |
1301 | return iface; | |
1302 | } | |
1303 | } | |
1304 | } | |
1305 | ||
1306 | return NULL; | |
1307 | } | |
064af421 | 1308 | \f |
8c4c1387 BP |
1309 | /* Bridge unixctl user interface functions. */ |
1310 | static void | |
8ca79daa | 1311 | bridge_unixctl_fdb_show(struct unixctl_conn *conn, |
c69ee87c | 1312 | const char *args, void *aux OVS_UNUSED) |
8c4c1387 BP |
1313 | { |
1314 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1315 | const struct bridge *br; | |
93dfc067 | 1316 | const struct mac_entry *e; |
8c4c1387 BP |
1317 | |
1318 | br = bridge_lookup(args); | |
1319 | if (!br) { | |
1320 | unixctl_command_reply(conn, 501, "no such bridge"); | |
1321 | return; | |
1322 | } | |
1323 | ||
1324 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
4e8e4213 | 1325 | LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { |
93dfc067 JG |
1326 | if (e->port < 0 || e->port >= br->n_ports) { |
1327 | continue; | |
8c4c1387 | 1328 | } |
93dfc067 JG |
1329 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", |
1330 | br->ports[e->port]->ifaces[0]->dp_ifidx, | |
1331 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
8c4c1387 BP |
1332 | } |
1333 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
1334 | ds_destroy(&ds); | |
1335 | } | |
1336 | \f | |
064af421 | 1337 | /* Bridge reconfiguration functions. */ |
064af421 | 1338 | static struct bridge * |
1a6f1e2a | 1339 | bridge_create(const struct ovsrec_bridge *br_cfg) |
064af421 BP |
1340 | { |
1341 | struct bridge *br; | |
1342 | int error; | |
1343 | ||
1a6f1e2a | 1344 | assert(!bridge_lookup(br_cfg->name)); |
ec6fde61 | 1345 | br = xzalloc(sizeof *br); |
064af421 | 1346 | |
1a6f1e2a JG |
1347 | error = dpif_create_and_open(br_cfg->name, br_cfg->datapath_type, |
1348 | &br->dpif); | |
efacbce6 | 1349 | if (error) { |
064af421 BP |
1350 | free(br); |
1351 | return NULL; | |
1352 | } | |
efacbce6 | 1353 | dpif_flow_flush(br->dpif); |
064af421 | 1354 | |
1a6f1e2a JG |
1355 | error = ofproto_create(br_cfg->name, br_cfg->datapath_type, &bridge_ofhooks, |
1356 | br, &br->ofproto); | |
064af421 | 1357 | if (error) { |
1a6f1e2a JG |
1358 | VLOG_ERR("failed to create switch %s: %s", br_cfg->name, |
1359 | strerror(error)); | |
c228a364 BP |
1360 | dpif_delete(br->dpif); |
1361 | dpif_close(br->dpif); | |
064af421 BP |
1362 | free(br); |
1363 | return NULL; | |
1364 | } | |
1365 | ||
1a6f1e2a JG |
1366 | br->name = xstrdup(br_cfg->name); |
1367 | br->cfg = br_cfg; | |
064af421 | 1368 | br->ml = mac_learning_create(); |
70150daf | 1369 | eth_addr_nicira_random(br->default_ea); |
064af421 | 1370 | |
d9a8717a | 1371 | hmap_init(&br->ifaces); |
064af421 | 1372 | |
836fad5e | 1373 | shash_init(&br->port_by_name); |
4a1ee6ae BP |
1374 | shash_init(&br->iface_by_name); |
1375 | ||
064af421 | 1376 | br->flush = false; |
064af421 BP |
1377 | |
1378 | list_push_back(&all_bridges, &br->node); | |
1379 | ||
c228a364 | 1380 | VLOG_INFO("created bridge %s on %s", br->name, dpif_name(br->dpif)); |
064af421 BP |
1381 | |
1382 | return br; | |
1383 | } | |
1384 | ||
1385 | static void | |
1386 | bridge_destroy(struct bridge *br) | |
1387 | { | |
1388 | if (br) { | |
1389 | int error; | |
1390 | ||
1391 | while (br->n_ports > 0) { | |
1392 | port_destroy(br->ports[br->n_ports - 1]); | |
1393 | } | |
1394 | list_remove(&br->node); | |
c228a364 | 1395 | error = dpif_delete(br->dpif); |
064af421 | 1396 | if (error && error != ENOENT) { |
b29ba128 | 1397 | VLOG_ERR("failed to delete %s: %s", |
c228a364 | 1398 | dpif_name(br->dpif), strerror(error)); |
064af421 | 1399 | } |
c228a364 | 1400 | dpif_close(br->dpif); |
064af421 | 1401 | ofproto_destroy(br->ofproto); |
064af421 | 1402 | mac_learning_destroy(br->ml); |
d9a8717a | 1403 | hmap_destroy(&br->ifaces); |
836fad5e | 1404 | shash_destroy(&br->port_by_name); |
4a1ee6ae | 1405 | shash_destroy(&br->iface_by_name); |
064af421 BP |
1406 | free(br->ports); |
1407 | free(br->name); | |
1408 | free(br); | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | static struct bridge * | |
1413 | bridge_lookup(const char *name) | |
1414 | { | |
1415 | struct bridge *br; | |
1416 | ||
4e8e4213 | 1417 | LIST_FOR_EACH (br, node, &all_bridges) { |
064af421 BP |
1418 | if (!strcmp(br->name, name)) { |
1419 | return br; | |
1420 | } | |
1421 | } | |
1422 | return NULL; | |
1423 | } | |
1424 | ||
4f2cad2c JP |
1425 | /* Handle requests for a listing of all flows known by the OpenFlow |
1426 | * stack, including those normally hidden. */ | |
1427 | static void | |
8ca79daa | 1428 | bridge_unixctl_dump_flows(struct unixctl_conn *conn, |
c69ee87c | 1429 | const char *args, void *aux OVS_UNUSED) |
4f2cad2c JP |
1430 | { |
1431 | struct bridge *br; | |
1432 | struct ds results; | |
d295e8e9 | 1433 | |
4f2cad2c JP |
1434 | br = bridge_lookup(args); |
1435 | if (!br) { | |
1436 | unixctl_command_reply(conn, 501, "Unknown bridge"); | |
1437 | return; | |
1438 | } | |
1439 | ||
1440 | ds_init(&results); | |
1441 | ofproto_get_all_flows(br->ofproto, &results); | |
1442 | ||
1443 | unixctl_command_reply(conn, 200, ds_cstr(&results)); | |
1444 | ds_destroy(&results); | |
1445 | } | |
1446 | ||
fa05809b BP |
1447 | /* "bridge/reconnect [BRIDGE]": makes BRIDGE drop all of its controller |
1448 | * connections and reconnect. If BRIDGE is not specified, then all bridges | |
1449 | * drop their controller connections and reconnect. */ | |
1450 | static void | |
1451 | bridge_unixctl_reconnect(struct unixctl_conn *conn, | |
1452 | const char *args, void *aux OVS_UNUSED) | |
1453 | { | |
1454 | struct bridge *br; | |
1455 | if (args[0] != '\0') { | |
1456 | br = bridge_lookup(args); | |
1457 | if (!br) { | |
1458 | unixctl_command_reply(conn, 501, "Unknown bridge"); | |
1459 | return; | |
1460 | } | |
1461 | ofproto_reconnect_controllers(br->ofproto); | |
1462 | } else { | |
4e8e4213 | 1463 | LIST_FOR_EACH (br, node, &all_bridges) { |
fa05809b BP |
1464 | ofproto_reconnect_controllers(br->ofproto); |
1465 | } | |
1466 | } | |
1467 | unixctl_command_reply(conn, 200, NULL); | |
1468 | } | |
1469 | ||
064af421 BP |
1470 | static int |
1471 | bridge_run_one(struct bridge *br) | |
1472 | { | |
1473 | int error; | |
1474 | ||
1475 | error = ofproto_run1(br->ofproto); | |
1476 | if (error) { | |
1477 | return error; | |
1478 | } | |
1479 | ||
93dfc067 | 1480 | mac_learning_run(br->ml, ofproto_get_revalidate_set(br->ofproto)); |
064af421 | 1481 | bond_run(br); |
064af421 BP |
1482 | |
1483 | error = ofproto_run2(br->ofproto, br->flush); | |
1484 | br->flush = false; | |
1485 | ||
1486 | return error; | |
1487 | } | |
1488 | ||
76ce9432 | 1489 | static size_t |
1a048029 | 1490 | bridge_get_controllers(const struct bridge *br, |
76ce9432 | 1491 | struct ovsrec_controller ***controllersp) |
064af421 | 1492 | { |
76ce9432 BP |
1493 | struct ovsrec_controller **controllers; |
1494 | size_t n_controllers; | |
064af421 | 1495 | |
1a048029 JP |
1496 | controllers = br->cfg->controller; |
1497 | n_controllers = br->cfg->n_controller; | |
76343538 | 1498 | |
76ce9432 BP |
1499 | if (n_controllers == 1 && !strcmp(controllers[0]->target, "none")) { |
1500 | controllers = NULL; | |
1501 | n_controllers = 0; | |
064af421 | 1502 | } |
76343538 | 1503 | |
76ce9432 BP |
1504 | if (controllersp) { |
1505 | *controllersp = controllers; | |
1506 | } | |
1507 | return n_controllers; | |
064af421 BP |
1508 | } |
1509 | ||
1510 | static void | |
1a048029 | 1511 | bridge_reconfigure_one(struct bridge *br) |
064af421 | 1512 | { |
76343538 | 1513 | struct shash old_ports, new_ports; |
064af421 | 1514 | struct svec snoops, old_snoops; |
76343538 | 1515 | struct shash_node *node; |
31681a5d | 1516 | enum ofproto_fail_mode fail_mode; |
6ae39834 | 1517 | size_t i; |
064af421 BP |
1518 | |
1519 | /* Collect old ports. */ | |
76343538 | 1520 | shash_init(&old_ports); |
064af421 | 1521 | for (i = 0; i < br->n_ports; i++) { |
76343538 | 1522 | shash_add(&old_ports, br->ports[i]->name, br->ports[i]); |
064af421 | 1523 | } |
064af421 BP |
1524 | |
1525 | /* Collect new ports. */ | |
76343538 BP |
1526 | shash_init(&new_ports); |
1527 | for (i = 0; i < br->cfg->n_ports; i++) { | |
1528 | const char *name = br->cfg->ports[i]->name; | |
1529 | if (!shash_add_once(&new_ports, name, br->cfg->ports[i])) { | |
1530 | VLOG_WARN("bridge %s: %s specified twice as bridge port", | |
1531 | br->name, name); | |
1532 | } | |
1533 | } | |
e073f944 BP |
1534 | |
1535 | /* If we have a controller, then we need a local port. Complain if the | |
1536 | * user didn't specify one. | |
1537 | * | |
1538 | * XXX perhaps we should synthesize a port ourselves in this case. */ | |
1a048029 | 1539 | if (bridge_get_controllers(br, NULL)) { |
72865317 BP |
1540 | char local_name[IF_NAMESIZE]; |
1541 | int error; | |
1542 | ||
1543 | error = dpif_port_get_name(br->dpif, ODPP_LOCAL, | |
1544 | local_name, sizeof local_name); | |
e073f944 BP |
1545 | if (!error && !shash_find(&new_ports, local_name)) { |
1546 | VLOG_WARN("bridge %s: controller specified but no local port " | |
1547 | "(port named %s) defined", | |
1548 | br->name, local_name); | |
72865317 | 1549 | } |
064af421 | 1550 | } |
064af421 | 1551 | |
4a1ee6ae BP |
1552 | /* Get rid of deleted ports. |
1553 | * Get rid of deleted interfaces on ports that still exist. */ | |
76343538 | 1554 | SHASH_FOR_EACH (node, &old_ports) { |
4a1ee6ae BP |
1555 | struct port *port = node->data; |
1556 | const struct ovsrec_port *port_cfg; | |
1557 | ||
1558 | port_cfg = shash_find_data(&new_ports, node->name); | |
1559 | if (!port_cfg) { | |
1560 | port_destroy(port); | |
1561 | } else { | |
1562 | port_del_ifaces(port, port_cfg); | |
064af421 BP |
1563 | } |
1564 | } | |
4a1ee6ae BP |
1565 | |
1566 | /* Create new ports. | |
1567 | * Add new interfaces to existing ports. | |
1568 | * Reconfigure existing ports. */ | |
76343538 BP |
1569 | SHASH_FOR_EACH (node, &new_ports) { |
1570 | struct port *port = shash_find_data(&old_ports, node->name); | |
1571 | if (!port) { | |
1572 | port = port_create(br, node->name); | |
064af421 | 1573 | } |
ceb4559f | 1574 | |
76343538 | 1575 | port_reconfigure(port, node->data); |
ceb4559f JG |
1576 | if (!port->n_ifaces) { |
1577 | VLOG_WARN("bridge %s: port %s has no interfaces, dropping", | |
1578 | br->name, port->name); | |
1579 | port_destroy(port); | |
1580 | } | |
064af421 | 1581 | } |
76343538 BP |
1582 | shash_destroy(&old_ports); |
1583 | shash_destroy(&new_ports); | |
064af421 | 1584 | |
31681a5d JP |
1585 | /* Set the fail-mode */ |
1586 | fail_mode = !br->cfg->fail_mode | |
1587 | || !strcmp(br->cfg->fail_mode, "standalone") | |
1588 | ? OFPROTO_FAIL_STANDALONE | |
1589 | : OFPROTO_FAIL_SECURE; | |
7d674866 BP |
1590 | if (ofproto_get_fail_mode(br->ofproto) != fail_mode |
1591 | && !ofproto_has_primary_controller(br->ofproto)) { | |
abdfe474 JP |
1592 | ofproto_flush_flows(br->ofproto); |
1593 | } | |
31681a5d JP |
1594 | ofproto_set_fail_mode(br->ofproto, fail_mode); |
1595 | ||
064af421 BP |
1596 | /* Delete all flows if we're switching from connected to standalone or vice |
1597 | * versa. (XXX Should we delete all flows if we are switching from one | |
1598 | * controller to another?) */ | |
1599 | ||
5a243150 | 1600 | /* Configure OpenFlow controller connection snooping. */ |
76343538 BP |
1601 | svec_init(&snoops); |
1602 | svec_add_nocopy(&snoops, xasprintf("punix:%s/%s.snoop", | |
1603 | ovs_rundir, br->name)); | |
1604 | svec_init(&old_snoops); | |
1605 | ofproto_get_snoops(br->ofproto, &old_snoops); | |
1606 | if (!svec_equal(&snoops, &old_snoops)) { | |
1607 | ofproto_set_snoops(br->ofproto, &snoops); | |
1608 | } | |
1609 | svec_destroy(&snoops); | |
1610 | svec_destroy(&old_snoops); | |
76343538 | 1611 | |
064af421 BP |
1612 | mirror_reconfigure(br); |
1613 | } | |
1614 | ||
7d674866 BP |
1615 | /* Initializes 'oc' appropriately as a management service controller for |
1616 | * 'br'. | |
1617 | * | |
1618 | * The caller must free oc->target when it is no longer needed. */ | |
1619 | static void | |
1620 | bridge_ofproto_controller_for_mgmt(const struct bridge *br, | |
1621 | struct ofproto_controller *oc) | |
1622 | { | |
1623 | oc->target = xasprintf("punix:%s/%s.mgmt", ovs_rundir, br->name); | |
1624 | oc->max_backoff = 0; | |
1625 | oc->probe_interval = 60; | |
1626 | oc->band = OFPROTO_OUT_OF_BAND; | |
1627 | oc->accept_re = NULL; | |
1628 | oc->update_resolv_conf = false; | |
1629 | oc->rate_limit = 0; | |
1630 | oc->burst_limit = 0; | |
1631 | } | |
1632 | ||
1633 | /* Converts ovsrec_controller 'c' into an ofproto_controller in 'oc'. */ | |
1634 | static void | |
1635 | bridge_ofproto_controller_from_ovsrec(const struct ovsrec_controller *c, | |
1636 | struct ofproto_controller *oc) | |
1637 | { | |
1638 | oc->target = c->target; | |
1639 | oc->max_backoff = c->max_backoff ? *c->max_backoff / 1000 : 8; | |
1640 | oc->probe_interval = c->inactivity_probe ? *c->inactivity_probe / 1000 : 5; | |
1641 | oc->band = (!c->connection_mode || !strcmp(c->connection_mode, "in-band") | |
1642 | ? OFPROTO_IN_BAND : OFPROTO_OUT_OF_BAND); | |
1643 | oc->accept_re = c->discover_accept_regex; | |
1644 | oc->update_resolv_conf = c->discover_update_resolv_conf; | |
1645 | oc->rate_limit = c->controller_rate_limit ? *c->controller_rate_limit : 0; | |
1646 | oc->burst_limit = (c->controller_burst_limit | |
1647 | ? *c->controller_burst_limit : 0); | |
1648 | } | |
1649 | ||
1650 | /* Configures the IP stack for 'br''s local interface properly according to the | |
1651 | * configuration in 'c'. */ | |
1652 | static void | |
1653 | bridge_configure_local_iface_netdev(struct bridge *br, | |
1654 | struct ovsrec_controller *c) | |
1655 | { | |
1656 | struct netdev *netdev; | |
1657 | struct in_addr mask, gateway; | |
1658 | ||
1659 | struct iface *local_iface; | |
1660 | struct in_addr ip; | |
1661 | ||
1662 | /* Controller discovery does its own TCP/IP configuration later. */ | |
1663 | if (strcmp(c->target, "discover")) { | |
1664 | return; | |
1665 | } | |
1666 | ||
1667 | /* If there's no local interface or no IP address, give up. */ | |
1668 | local_iface = bridge_get_local_iface(br); | |
1669 | if (!local_iface || !c->local_ip || !inet_aton(c->local_ip, &ip)) { | |
1670 | return; | |
1671 | } | |
1672 | ||
1673 | /* Bring up the local interface. */ | |
1674 | netdev = local_iface->netdev; | |
1675 | netdev_turn_flags_on(netdev, NETDEV_UP, true); | |
1676 | ||
1677 | /* Configure the IP address and netmask. */ | |
1678 | if (!c->local_netmask | |
1679 | || !inet_aton(c->local_netmask, &mask) | |
1680 | || !mask.s_addr) { | |
1681 | mask.s_addr = guess_netmask(ip.s_addr); | |
1682 | } | |
1683 | if (!netdev_set_in4(netdev, ip, mask)) { | |
1684 | VLOG_INFO("bridge %s: configured IP address "IP_FMT", netmask "IP_FMT, | |
1685 | br->name, IP_ARGS(&ip.s_addr), IP_ARGS(&mask.s_addr)); | |
1686 | } | |
1687 | ||
1688 | /* Configure the default gateway. */ | |
1689 | if (c->local_gateway | |
1690 | && inet_aton(c->local_gateway, &gateway) | |
1691 | && gateway.s_addr) { | |
1692 | if (!netdev_add_router(netdev, gateway)) { | |
1693 | VLOG_INFO("bridge %s: configured gateway "IP_FMT, | |
1694 | br->name, IP_ARGS(&gateway.s_addr)); | |
1695 | } | |
1696 | } | |
1697 | } | |
1698 | ||
064af421 | 1699 | static void |
1a048029 | 1700 | bridge_reconfigure_remotes(struct bridge *br, |
11bbff1b BP |
1701 | const struct sockaddr_in *managers, |
1702 | size_t n_managers) | |
064af421 | 1703 | { |
76ce9432 BP |
1704 | struct ovsrec_controller **controllers; |
1705 | size_t n_controllers; | |
7d674866 | 1706 | bool had_primary; |
8731b2b6 JP |
1707 | const char *disable_ib_str; |
1708 | bool disable_in_band = false; | |
7d674866 BP |
1709 | |
1710 | struct ofproto_controller *ocs; | |
1711 | size_t n_ocs; | |
1712 | size_t i; | |
064af421 | 1713 | |
8731b2b6 JP |
1714 | |
1715 | /* Check if we should disable in-band control on this bridge. */ | |
1716 | disable_ib_str = bridge_get_other_config(br->cfg, "disable-in-band"); | |
1717 | if (disable_ib_str && !strcmp(disable_ib_str, "true")) { | |
1718 | disable_in_band = true; | |
1719 | } | |
1720 | ||
1721 | if (disable_in_band) { | |
1722 | ofproto_set_extra_in_band_remotes(br->ofproto, NULL, 0); | |
1723 | } else { | |
1724 | ofproto_set_extra_in_band_remotes(br->ofproto, managers, n_managers); | |
1725 | } | |
7d674866 | 1726 | had_primary = ofproto_has_primary_controller(br->ofproto); |
cd11000b | 1727 | |
1a048029 | 1728 | n_controllers = bridge_get_controllers(br, &controllers); |
064af421 | 1729 | |
7d674866 BP |
1730 | ocs = xmalloc((n_controllers + 1) * sizeof *ocs); |
1731 | n_ocs = 0; | |
064af421 | 1732 | |
7d674866 BP |
1733 | bridge_ofproto_controller_for_mgmt(br, &ocs[n_ocs++]); |
1734 | for (i = 0; i < n_controllers; i++) { | |
1735 | struct ovsrec_controller *c = controllers[i]; | |
76ce9432 | 1736 | |
7d674866 BP |
1737 | if (!strncmp(c->target, "punix:", 6) |
1738 | || !strncmp(c->target, "unix:", 5)) { | |
1739 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
76ce9432 | 1740 | |
7d674866 BP |
1741 | /* Prevent remote ovsdb-server users from accessing arbitrary Unix |
1742 | * domain sockets and overwriting arbitrary local files. */ | |
1743 | VLOG_ERR_RL(&rl, "%s: not adding Unix domain socket controller " | |
1744 | "\"%s\" due to possibility for remote exploit", | |
1745 | dpif_name(br->dpif), c->target); | |
1746 | continue; | |
1747 | } | |
76ce9432 | 1748 | |
7d674866 | 1749 | bridge_configure_local_iface_netdev(br, c); |
8731b2b6 JP |
1750 | bridge_ofproto_controller_from_ovsrec(c, &ocs[n_ocs]); |
1751 | if (disable_in_band) { | |
1752 | ocs[n_ocs].band = OFPROTO_OUT_OF_BAND; | |
1753 | } | |
1754 | n_ocs++; | |
7d674866 | 1755 | } |
76ce9432 | 1756 | |
7d674866 BP |
1757 | ofproto_set_controllers(br->ofproto, ocs, n_ocs); |
1758 | free(ocs[0].target); /* From bridge_ofproto_controller_for_mgmt(). */ | |
1759 | free(ocs); | |
064af421 | 1760 | |
7d674866 BP |
1761 | if (had_primary != ofproto_has_primary_controller(br->ofproto)) { |
1762 | ofproto_flush_flows(br->ofproto); | |
1763 | } | |
76ce9432 | 1764 | |
7d674866 BP |
1765 | /* If there are no controllers and the bridge is in standalone |
1766 | * mode, set up a flow that matches every packet and directs | |
1767 | * them to OFPP_NORMAL (which goes to us). Otherwise, the | |
1768 | * switch is in secure mode and we won't pass any traffic until | |
1769 | * a controller has been defined and it tells us to do so. */ | |
1770 | if (!n_controllers | |
1771 | && ofproto_get_fail_mode(br->ofproto) == OFPROTO_FAIL_STANDALONE) { | |
1772 | union ofp_action action; | |
ae412e7d | 1773 | struct flow flow; |
76ce9432 | 1774 | |
7d674866 BP |
1775 | memset(&action, 0, sizeof action); |
1776 | action.type = htons(OFPAT_OUTPUT); | |
1777 | action.output.len = htons(sizeof action); | |
1778 | action.output.port = htons(OFPP_NORMAL); | |
1779 | memset(&flow, 0, sizeof flow); | |
1780 | ofproto_add_flow(br->ofproto, &flow, OVSFW_ALL, 0, &action, 1, 0); | |
064af421 | 1781 | } |
064af421 BP |
1782 | } |
1783 | ||
1784 | static void | |
76343538 | 1785 | bridge_get_all_ifaces(const struct bridge *br, struct shash *ifaces) |
064af421 BP |
1786 | { |
1787 | size_t i, j; | |
1788 | ||
76343538 | 1789 | shash_init(ifaces); |
064af421 BP |
1790 | for (i = 0; i < br->n_ports; i++) { |
1791 | struct port *port = br->ports[i]; | |
1792 | for (j = 0; j < port->n_ifaces; j++) { | |
1793 | struct iface *iface = port->ifaces[j]; | |
76343538 | 1794 | shash_add_once(ifaces, iface->name, iface); |
064af421 | 1795 | } |
76343538 BP |
1796 | if (port->n_ifaces > 1 && port->cfg->bond_fake_iface) { |
1797 | shash_add_once(ifaces, port->name, NULL); | |
35c979bf | 1798 | } |
064af421 | 1799 | } |
064af421 BP |
1800 | } |
1801 | ||
1802 | /* For robustness, in case the administrator moves around datapath ports behind | |
1803 | * our back, we re-check all the datapath port numbers here. | |
1804 | * | |
1805 | * This function will set the 'dp_ifidx' members of interfaces that have | |
1806 | * disappeared to -1, so only call this function from a context where those | |
1807 | * 'struct iface's will be removed from the bridge. Otherwise, the -1 | |
1808 | * 'dp_ifidx'es will cause trouble later when we try to send them to the | |
1809 | * datapath, which doesn't support UINT16_MAX+1 ports. */ | |
1810 | static void | |
1811 | bridge_fetch_dp_ifaces(struct bridge *br) | |
1812 | { | |
1813 | struct odp_port *dpif_ports; | |
1814 | size_t n_dpif_ports; | |
1815 | size_t i, j; | |
1816 | ||
1817 | /* Reset all interface numbers. */ | |
1818 | for (i = 0; i < br->n_ports; i++) { | |
1819 | struct port *port = br->ports[i]; | |
1820 | for (j = 0; j < port->n_ifaces; j++) { | |
1821 | struct iface *iface = port->ifaces[j]; | |
1822 | iface->dp_ifidx = -1; | |
1823 | } | |
1824 | } | |
d9a8717a | 1825 | hmap_clear(&br->ifaces); |
064af421 | 1826 | |
c228a364 | 1827 | dpif_port_list(br->dpif, &dpif_ports, &n_dpif_ports); |
064af421 BP |
1828 | for (i = 0; i < n_dpif_ports; i++) { |
1829 | struct odp_port *p = &dpif_ports[i]; | |
1830 | struct iface *iface = iface_lookup(br, p->devname); | |
1831 | if (iface) { | |
1832 | if (iface->dp_ifidx >= 0) { | |
b29ba128 | 1833 | VLOG_WARN("%s reported interface %s twice", |
c228a364 | 1834 | dpif_name(br->dpif), p->devname); |
064af421 | 1835 | } else if (iface_from_dp_ifidx(br, p->port)) { |
b29ba128 | 1836 | VLOG_WARN("%s reported interface %"PRIu16" twice", |
c228a364 | 1837 | dpif_name(br->dpif), p->port); |
064af421 | 1838 | } else { |
064af421 | 1839 | iface->dp_ifidx = p->port; |
d9a8717a BP |
1840 | hmap_insert(&br->ifaces, &iface->dp_ifidx_node, |
1841 | hash_int(iface->dp_ifidx, 0)); | |
064af421 | 1842 | } |
093e47f4 | 1843 | |
bcd49a45 BP |
1844 | iface_set_ofport(iface->cfg, |
1845 | (iface->dp_ifidx >= 0 | |
1846 | ? odp_port_to_ofp_port(iface->dp_ifidx) | |
1847 | : -1)); | |
064af421 BP |
1848 | } |
1849 | } | |
1850 | free(dpif_ports); | |
1851 | } | |
1852 | \f | |
1853 | /* Bridge packet processing functions. */ | |
1854 | ||
da285df4 BP |
1855 | static int |
1856 | bond_hash(const uint8_t mac[ETH_ADDR_LEN]) | |
1857 | { | |
1858 | return hash_bytes(mac, ETH_ADDR_LEN, 0) & BOND_MASK; | |
1859 | } | |
1860 | ||
064af421 BP |
1861 | static struct bond_entry * |
1862 | lookup_bond_entry(const struct port *port, const uint8_t mac[ETH_ADDR_LEN]) | |
1863 | { | |
da285df4 | 1864 | return &port->bond_hash[bond_hash(mac)]; |
064af421 BP |
1865 | } |
1866 | ||
1867 | static int | |
1868 | bond_choose_iface(const struct port *port) | |
1869 | { | |
8b2a2f4a JG |
1870 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
1871 | size_t i, best_down_slave = -1; | |
1872 | long long next_delay_expiration = LLONG_MAX; | |
1873 | ||
064af421 | 1874 | for (i = 0; i < port->n_ifaces; i++) { |
8b2a2f4a JG |
1875 | struct iface *iface = port->ifaces[i]; |
1876 | ||
1877 | if (iface->enabled) { | |
064af421 | 1878 | return i; |
8b2a2f4a JG |
1879 | } else if (iface->delay_expires < next_delay_expiration) { |
1880 | best_down_slave = i; | |
1881 | next_delay_expiration = iface->delay_expires; | |
064af421 BP |
1882 | } |
1883 | } | |
8b2a2f4a JG |
1884 | |
1885 | if (best_down_slave != -1) { | |
1886 | struct iface *iface = port->ifaces[best_down_slave]; | |
1887 | ||
1888 | VLOG_INFO_RL(&rl, "interface %s: skipping remaining %lli ms updelay " | |
1889 | "since no other interface is up", iface->name, | |
1890 | iface->delay_expires - time_msec()); | |
1891 | bond_enable_slave(iface, true); | |
1892 | } | |
1893 | ||
1894 | return best_down_slave; | |
064af421 BP |
1895 | } |
1896 | ||
1897 | static bool | |
2303f3b2 | 1898 | choose_output_iface(const struct port *port, const uint8_t *dl_src, |
064af421 BP |
1899 | uint16_t *dp_ifidx, tag_type *tags) |
1900 | { | |
1901 | struct iface *iface; | |
1902 | ||
1903 | assert(port->n_ifaces); | |
1904 | if (port->n_ifaces == 1) { | |
1905 | iface = port->ifaces[0]; | |
1906 | } else { | |
2303f3b2 | 1907 | struct bond_entry *e = lookup_bond_entry(port, dl_src); |
064af421 BP |
1908 | if (e->iface_idx < 0 || e->iface_idx >= port->n_ifaces |
1909 | || !port->ifaces[e->iface_idx]->enabled) { | |
1910 | /* XXX select interface properly. The current interface selection | |
1911 | * is only good for testing the rebalancing code. */ | |
1912 | e->iface_idx = bond_choose_iface(port); | |
1913 | if (e->iface_idx < 0) { | |
1914 | *tags |= port->no_ifaces_tag; | |
1915 | return false; | |
1916 | } | |
1917 | e->iface_tag = tag_create_random(); | |
85c74638 | 1918 | ((struct port *) port)->bond_compat_is_stale = true; |
064af421 BP |
1919 | } |
1920 | *tags |= e->iface_tag; | |
1921 | iface = port->ifaces[e->iface_idx]; | |
1922 | } | |
1923 | *dp_ifidx = iface->dp_ifidx; | |
1924 | *tags |= iface->tag; /* Currently only used for bonding. */ | |
1925 | return true; | |
1926 | } | |
1927 | ||
1928 | static void | |
1929 | bond_link_status_update(struct iface *iface, bool carrier) | |
1930 | { | |
1931 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1932 | struct port *port = iface->port; | |
1933 | ||
1934 | if ((carrier == iface->enabled) == (iface->delay_expires == LLONG_MAX)) { | |
1935 | /* Nothing to do. */ | |
1936 | return; | |
1937 | } | |
1938 | VLOG_INFO_RL(&rl, "interface %s: carrier %s", | |
1939 | iface->name, carrier ? "detected" : "dropped"); | |
1940 | if (carrier == iface->enabled) { | |
1941 | iface->delay_expires = LLONG_MAX; | |
1942 | VLOG_INFO_RL(&rl, "interface %s: will not be %s", | |
1943 | iface->name, carrier ? "disabled" : "enabled"); | |
8b2a2f4a JG |
1944 | } else if (carrier && port->active_iface < 0) { |
1945 | bond_enable_slave(iface, true); | |
1946 | if (port->updelay) { | |
1947 | VLOG_INFO_RL(&rl, "interface %s: skipping %d ms updelay since no " | |
1948 | "other interface is up", iface->name, port->updelay); | |
1949 | } | |
064af421 BP |
1950 | } else { |
1951 | int delay = carrier ? port->updelay : port->downdelay; | |
1952 | iface->delay_expires = time_msec() + delay; | |
1953 | if (delay) { | |
1954 | VLOG_INFO_RL(&rl, | |
1955 | "interface %s: will be %s if it stays %s for %d ms", | |
1956 | iface->name, | |
1957 | carrier ? "enabled" : "disabled", | |
1958 | carrier ? "up" : "down", | |
1959 | delay); | |
1960 | } | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | static void | |
1965 | bond_choose_active_iface(struct port *port) | |
1966 | { | |
1967 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
1968 | ||
1969 | port->active_iface = bond_choose_iface(port); | |
1970 | port->active_iface_tag = tag_create_random(); | |
1971 | if (port->active_iface >= 0) { | |
1972 | VLOG_INFO_RL(&rl, "port %s: active interface is now %s", | |
1973 | port->name, port->ifaces[port->active_iface]->name); | |
1974 | } else { | |
1975 | VLOG_WARN_RL(&rl, "port %s: all ports disabled, no active interface", | |
1976 | port->name); | |
1977 | } | |
1978 | } | |
1979 | ||
da285df4 BP |
1980 | static void |
1981 | bond_enable_slave(struct iface *iface, bool enable) | |
1982 | { | |
1983 | struct port *port = iface->port; | |
1984 | struct bridge *br = port->bridge; | |
1985 | ||
8b2a2f4a JG |
1986 | /* This acts as a recursion check. If the act of disabling a slave |
1987 | * causes a different slave to be enabled, the flag will allow us to | |
1988 | * skip redundant work when we reenter this function. It must be | |
1989 | * cleared on exit to keep things safe with multiple bonds. */ | |
1990 | static bool moving_active_iface = false; | |
1991 | ||
da285df4 BP |
1992 | iface->delay_expires = LLONG_MAX; |
1993 | if (enable == iface->enabled) { | |
1994 | return; | |
1995 | } | |
1996 | ||
1997 | iface->enabled = enable; | |
1998 | if (!iface->enabled) { | |
17ea75b2 | 1999 | VLOG_WARN("interface %s: disabled", iface->name); |
da285df4 BP |
2000 | ofproto_revalidate(br->ofproto, iface->tag); |
2001 | if (iface->port_ifidx == port->active_iface) { | |
2002 | ofproto_revalidate(br->ofproto, | |
2003 | port->active_iface_tag); | |
8b2a2f4a JG |
2004 | |
2005 | /* Disabling a slave can lead to another slave being immediately | |
2006 | * enabled if there will be no active slaves but one is waiting | |
2007 | * on an updelay. In this case we do not need to run most of the | |
2008 | * code for the newly enabled slave since there was no period | |
2009 | * without an active slave and it is redundant with the disabling | |
2010 | * path. */ | |
2011 | moving_active_iface = true; | |
da285df4 BP |
2012 | bond_choose_active_iface(port); |
2013 | } | |
2014 | bond_send_learning_packets(port); | |
2015 | } else { | |
17ea75b2 | 2016 | VLOG_WARN("interface %s: enabled", iface->name); |
8b2a2f4a | 2017 | if (port->active_iface < 0 && !moving_active_iface) { |
da285df4 BP |
2018 | ofproto_revalidate(br->ofproto, port->no_ifaces_tag); |
2019 | bond_choose_active_iface(port); | |
2020 | bond_send_learning_packets(port); | |
2021 | } | |
2022 | iface->tag = tag_create_random(); | |
2023 | } | |
8b2a2f4a JG |
2024 | |
2025 | moving_active_iface = false; | |
2026 | port->bond_compat_is_stale = true; | |
da285df4 BP |
2027 | } |
2028 | ||
8722022c BP |
2029 | /* Attempts to make the sum of the bond slaves' statistics appear on the fake |
2030 | * bond interface. */ | |
2031 | static void | |
2032 | bond_update_fake_iface_stats(struct port *port) | |
2033 | { | |
2034 | struct netdev_stats bond_stats; | |
2035 | struct netdev *bond_dev; | |
2036 | size_t i; | |
2037 | ||
2038 | memset(&bond_stats, 0, sizeof bond_stats); | |
2039 | ||
2040 | for (i = 0; i < port->n_ifaces; i++) { | |
2041 | struct netdev_stats slave_stats; | |
2042 | ||
2043 | if (!netdev_get_stats(port->ifaces[i]->netdev, &slave_stats)) { | |
f4b6076a JG |
2044 | /* XXX: We swap the stats here because they are swapped back when |
2045 | * reported by the internal device. The reason for this is | |
2046 | * internal devices normally represent packets going into the system | |
2047 | * but when used as fake bond device they represent packets leaving | |
2048 | * the system. We really should do this in the internal device | |
2049 | * itself because changing it here reverses the counts from the | |
2050 | * perspective of the switch. However, the internal device doesn't | |
2051 | * know what type of device it represents so we have to do it here | |
2052 | * for now. */ | |
2053 | bond_stats.tx_packets += slave_stats.rx_packets; | |
2054 | bond_stats.tx_bytes += slave_stats.rx_bytes; | |
2055 | bond_stats.rx_packets += slave_stats.tx_packets; | |
2056 | bond_stats.rx_bytes += slave_stats.tx_bytes; | |
8722022c BP |
2057 | } |
2058 | } | |
2059 | ||
2060 | if (!netdev_open_default(port->name, &bond_dev)) { | |
2061 | netdev_set_stats(bond_dev, &bond_stats); | |
2062 | netdev_close(bond_dev); | |
2063 | } | |
2064 | } | |
2065 | ||
064af421 BP |
2066 | static void |
2067 | bond_run(struct bridge *br) | |
2068 | { | |
2069 | size_t i, j; | |
2070 | ||
2071 | for (i = 0; i < br->n_ports; i++) { | |
2072 | struct port *port = br->ports[i]; | |
85c74638 | 2073 | |
8b2a2f4a | 2074 | if (port->n_ifaces >= 2) { |
0bb5c3ec BP |
2075 | char *devname; |
2076 | ||
2077 | /* Track carrier going up and down on interfaces. */ | |
2078 | while (!netdev_monitor_poll(port->monitor, &devname)) { | |
2079 | struct iface *iface; | |
0bb5c3ec BP |
2080 | |
2081 | iface = port_lookup_iface(port, devname); | |
85da620e JG |
2082 | if (iface) { |
2083 | bool carrier = netdev_get_carrier(iface->netdev); | |
2084 | ||
0bb5c3ec BP |
2085 | bond_link_status_update(iface, carrier); |
2086 | port_update_bond_compat(port); | |
2087 | } | |
2088 | free(devname); | |
2089 | } | |
2090 | ||
8b2a2f4a JG |
2091 | for (j = 0; j < port->n_ifaces; j++) { |
2092 | struct iface *iface = port->ifaces[j]; | |
2093 | if (time_msec() >= iface->delay_expires) { | |
2094 | bond_enable_slave(iface, !iface->enabled); | |
2095 | } | |
2096 | } | |
8722022c BP |
2097 | |
2098 | if (port->bond_fake_iface | |
2099 | && time_msec() >= port->bond_next_fake_iface_update) { | |
2100 | bond_update_fake_iface_stats(port); | |
2101 | port->bond_next_fake_iface_update = time_msec() + 1000; | |
2102 | } | |
8b2a2f4a JG |
2103 | } |
2104 | ||
85c74638 BP |
2105 | if (port->bond_compat_is_stale) { |
2106 | port->bond_compat_is_stale = false; | |
2107 | port_update_bond_compat(port); | |
2108 | } | |
064af421 BP |
2109 | } |
2110 | } | |
2111 | ||
2112 | static void | |
2113 | bond_wait(struct bridge *br) | |
2114 | { | |
2115 | size_t i, j; | |
2116 | ||
2117 | for (i = 0; i < br->n_ports; i++) { | |
2118 | struct port *port = br->ports[i]; | |
2119 | if (port->n_ifaces < 2) { | |
2120 | continue; | |
2121 | } | |
0bb5c3ec | 2122 | netdev_monitor_poll_wait(port->monitor); |
064af421 BP |
2123 | for (j = 0; j < port->n_ifaces; j++) { |
2124 | struct iface *iface = port->ifaces[j]; | |
2125 | if (iface->delay_expires != LLONG_MAX) { | |
7cf8b266 | 2126 | poll_timer_wait_until(iface->delay_expires); |
064af421 BP |
2127 | } |
2128 | } | |
8722022c | 2129 | if (port->bond_fake_iface) { |
7cf8b266 | 2130 | poll_timer_wait_until(port->bond_next_fake_iface_update); |
8722022c | 2131 | } |
064af421 BP |
2132 | } |
2133 | } | |
2134 | ||
2135 | static bool | |
ae412e7d | 2136 | set_dst(struct dst *p, const struct flow *flow, |
064af421 BP |
2137 | const struct port *in_port, const struct port *out_port, |
2138 | tag_type *tags) | |
2139 | { | |
064af421 BP |
2140 | p->vlan = (out_port->vlan >= 0 ? OFP_VLAN_NONE |
2141 | : in_port->vlan >= 0 ? in_port->vlan | |
2142 | : ntohs(flow->dl_vlan)); | |
2303f3b2 | 2143 | return choose_output_iface(out_port, flow->dl_src, &p->dp_ifidx, tags); |
064af421 BP |
2144 | } |
2145 | ||
2146 | static void | |
2147 | swap_dst(struct dst *p, struct dst *q) | |
2148 | { | |
2149 | struct dst tmp = *p; | |
2150 | *p = *q; | |
2151 | *q = tmp; | |
2152 | } | |
2153 | ||
2154 | /* Moves all the dsts with vlan == 'vlan' to the front of the 'n_dsts' in | |
2155 | * 'dsts'. (This may help performance by reducing the number of VLAN changes | |
2156 | * that we push to the datapath. We could in fact fully sort the array by | |
2157 | * vlan, but in most cases there are at most two different vlan tags so that's | |
2158 | * possibly overkill.) */ | |
2159 | static void | |
2160 | partition_dsts(struct dst *dsts, size_t n_dsts, int vlan) | |
2161 | { | |
2162 | struct dst *first = dsts; | |
2163 | struct dst *last = dsts + n_dsts; | |
2164 | ||
2165 | while (first != last) { | |
2166 | /* Invariants: | |
2167 | * - All dsts < first have vlan == 'vlan'. | |
2168 | * - All dsts >= last have vlan != 'vlan'. | |
2169 | * - first < last. */ | |
2170 | while (first->vlan == vlan) { | |
2171 | if (++first == last) { | |
2172 | return; | |
2173 | } | |
2174 | } | |
2175 | ||
2176 | /* Same invariants, plus one additional: | |
2177 | * - first->vlan != vlan. | |
2178 | */ | |
2179 | while (last[-1].vlan != vlan) { | |
2180 | if (--last == first) { | |
2181 | return; | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | /* Same invariants, plus one additional: | |
2186 | * - last[-1].vlan == vlan.*/ | |
2187 | swap_dst(first++, --last); | |
2188 | } | |
2189 | } | |
2190 | ||
2191 | static int | |
2192 | mirror_mask_ffs(mirror_mask_t mask) | |
2193 | { | |
2194 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
2195 | return ffs(mask); | |
2196 | } | |
2197 | ||
2198 | static bool | |
2199 | dst_is_duplicate(const struct dst *dsts, size_t n_dsts, | |
2200 | const struct dst *test) | |
2201 | { | |
2202 | size_t i; | |
2203 | for (i = 0; i < n_dsts; i++) { | |
2204 | if (dsts[i].vlan == test->vlan && dsts[i].dp_ifidx == test->dp_ifidx) { | |
2205 | return true; | |
2206 | } | |
2207 | } | |
2208 | return false; | |
2209 | } | |
2210 | ||
2211 | static bool | |
2212 | port_trunks_vlan(const struct port *port, uint16_t vlan) | |
2213 | { | |
3e9c481c BP |
2214 | return (port->vlan < 0 |
2215 | && (!port->trunks || bitmap_is_set(port->trunks, vlan))); | |
064af421 BP |
2216 | } |
2217 | ||
2218 | static bool | |
2219 | port_includes_vlan(const struct port *port, uint16_t vlan) | |
2220 | { | |
2221 | return vlan == port->vlan || port_trunks_vlan(port, vlan); | |
2222 | } | |
2223 | ||
a4e2e1f2 EJ |
2224 | static bool |
2225 | port_is_floodable(const struct port *port) | |
2226 | { | |
2227 | int i; | |
2228 | ||
2229 | for (i = 0; i < port->n_ifaces; i++) { | |
2230 | if (!ofproto_port_is_floodable(port->bridge->ofproto, | |
2231 | port->ifaces[i]->dp_ifidx)) { | |
2232 | return false; | |
2233 | } | |
2234 | } | |
2235 | return true; | |
2236 | } | |
2237 | ||
064af421 | 2238 | static size_t |
ae412e7d | 2239 | compose_dsts(const struct bridge *br, const struct flow *flow, uint16_t vlan, |
064af421 | 2240 | const struct port *in_port, const struct port *out_port, |
6a07af36 | 2241 | struct dst dsts[], tag_type *tags, uint16_t *nf_output_iface) |
064af421 BP |
2242 | { |
2243 | mirror_mask_t mirrors = in_port->src_mirrors; | |
2244 | struct dst *dst = dsts; | |
2245 | size_t i; | |
2246 | ||
064af421 BP |
2247 | if (out_port == FLOOD_PORT) { |
2248 | /* XXX use ODP_FLOOD if no vlans or bonding. */ | |
2249 | /* XXX even better, define each VLAN as a datapath port group */ | |
2250 | for (i = 0; i < br->n_ports; i++) { | |
2251 | struct port *port = br->ports[i]; | |
a4e2e1f2 EJ |
2252 | if (port != in_port |
2253 | && port_is_floodable(port) | |
2254 | && port_includes_vlan(port, vlan) | |
064af421 BP |
2255 | && !port->is_mirror_output_port |
2256 | && set_dst(dst, flow, in_port, port, tags)) { | |
2257 | mirrors |= port->dst_mirrors; | |
2258 | dst++; | |
2259 | } | |
2260 | } | |
6a07af36 | 2261 | *nf_output_iface = NF_OUT_FLOOD; |
064af421 | 2262 | } else if (out_port && set_dst(dst, flow, in_port, out_port, tags)) { |
6a07af36 | 2263 | *nf_output_iface = dst->dp_ifidx; |
064af421 BP |
2264 | mirrors |= out_port->dst_mirrors; |
2265 | dst++; | |
2266 | } | |
2267 | ||
2268 | while (mirrors) { | |
2269 | struct mirror *m = br->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
2270 | if (!m->n_vlans || vlan_is_mirrored(m, vlan)) { | |
2271 | if (m->out_port) { | |
2272 | if (set_dst(dst, flow, in_port, m->out_port, tags) | |
2273 | && !dst_is_duplicate(dsts, dst - dsts, dst)) { | |
2274 | dst++; | |
2275 | } | |
2276 | } else { | |
2277 | for (i = 0; i < br->n_ports; i++) { | |
2278 | struct port *port = br->ports[i]; | |
2279 | if (port_includes_vlan(port, m->out_vlan) | |
274de4d2 | 2280 | && set_dst(dst, flow, in_port, port, tags)) |
064af421 | 2281 | { |
43aa5f47 JG |
2282 | int flow_vlan; |
2283 | ||
064af421 BP |
2284 | if (port->vlan < 0) { |
2285 | dst->vlan = m->out_vlan; | |
2286 | } | |
274de4d2 BP |
2287 | if (dst_is_duplicate(dsts, dst - dsts, dst)) { |
2288 | continue; | |
2289 | } | |
43aa5f47 JG |
2290 | |
2291 | /* Use the vlan tag on the original flow instead of | |
2292 | * the one passed in the vlan parameter. This ensures | |
2293 | * that we compare the vlan from before any implicit | |
2294 | * tagging tags place. This is necessary because | |
2295 | * dst->vlan is the final vlan, after removing implicit | |
2296 | * tags. */ | |
2297 | flow_vlan = ntohs(flow->dl_vlan); | |
2298 | if (flow_vlan == 0) { | |
2299 | flow_vlan = OFP_VLAN_NONE; | |
2300 | } | |
2301 | if (port == in_port && dst->vlan == flow_vlan) { | |
064af421 BP |
2302 | /* Don't send out input port on same VLAN. */ |
2303 | continue; | |
2304 | } | |
2305 | dst++; | |
2306 | } | |
2307 | } | |
2308 | } | |
2309 | } | |
2310 | mirrors &= mirrors - 1; | |
2311 | } | |
2312 | ||
2313 | partition_dsts(dsts, dst - dsts, ntohs(flow->dl_vlan)); | |
2314 | return dst - dsts; | |
2315 | } | |
2316 | ||
67a4917b | 2317 | static void OVS_UNUSED |
064af421 BP |
2318 | print_dsts(const struct dst *dsts, size_t n) |
2319 | { | |
2320 | for (; n--; dsts++) { | |
2321 | printf(">p%"PRIu16, dsts->dp_ifidx); | |
2322 | if (dsts->vlan != OFP_VLAN_NONE) { | |
2323 | printf("v%"PRIu16, dsts->vlan); | |
2324 | } | |
2325 | } | |
2326 | } | |
2327 | ||
2328 | static void | |
ae412e7d | 2329 | compose_actions(struct bridge *br, const struct flow *flow, uint16_t vlan, |
064af421 | 2330 | const struct port *in_port, const struct port *out_port, |
6a07af36 JG |
2331 | tag_type *tags, struct odp_actions *actions, |
2332 | uint16_t *nf_output_iface) | |
064af421 BP |
2333 | { |
2334 | struct dst dsts[DP_MAX_PORTS * (MAX_MIRRORS + 1)]; | |
2335 | size_t n_dsts; | |
2336 | const struct dst *p; | |
2337 | uint16_t cur_vlan; | |
2338 | ||
6a07af36 JG |
2339 | n_dsts = compose_dsts(br, flow, vlan, in_port, out_port, dsts, tags, |
2340 | nf_output_iface); | |
064af421 BP |
2341 | |
2342 | cur_vlan = ntohs(flow->dl_vlan); | |
2343 | for (p = dsts; p < &dsts[n_dsts]; p++) { | |
2344 | union odp_action *a; | |
2345 | if (p->vlan != cur_vlan) { | |
2346 | if (p->vlan == OFP_VLAN_NONE) { | |
2347 | odp_actions_add(actions, ODPAT_STRIP_VLAN); | |
2348 | } else { | |
26233bb4 BP |
2349 | a = odp_actions_add(actions, ODPAT_SET_DL_TCI); |
2350 | a->dl_tci.tci = htons(p->vlan & VLAN_VID_MASK); | |
27bcf966 | 2351 | a->dl_tci.tci |= htons(flow->dl_vlan_pcp << VLAN_PCP_SHIFT); |
064af421 BP |
2352 | } |
2353 | cur_vlan = p->vlan; | |
2354 | } | |
2355 | a = odp_actions_add(actions, ODPAT_OUTPUT); | |
2356 | a->output.port = p->dp_ifidx; | |
2357 | } | |
2358 | } | |
2359 | ||
e96a4d80 JG |
2360 | /* Returns the effective vlan of a packet, taking into account both the |
2361 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
2362 | * the packet is untagged and -1 indicates it has an invalid header and | |
2363 | * should be dropped. */ | |
ae412e7d | 2364 | static int flow_get_vlan(struct bridge *br, const struct flow *flow, |
e96a4d80 JG |
2365 | struct port *in_port, bool have_packet) |
2366 | { | |
2367 | /* Note that dl_vlan of 0 and of OFP_VLAN_NONE both mean that the packet | |
2368 | * belongs to VLAN 0, so we should treat both cases identically. (In the | |
2369 | * former case, the packet has an 802.1Q header that specifies VLAN 0, | |
2370 | * presumably to allow a priority to be specified. In the latter case, the | |
2371 | * packet does not have any 802.1Q header.) */ | |
2372 | int vlan = ntohs(flow->dl_vlan); | |
2373 | if (vlan == OFP_VLAN_NONE) { | |
2374 | vlan = 0; | |
2375 | } | |
2376 | if (in_port->vlan >= 0) { | |
2377 | if (vlan) { | |
2378 | /* XXX support double tagging? */ | |
2379 | if (have_packet) { | |
2380 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2381 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
2382 | "packet received on port %s configured with " | |
2383 | "implicit VLAN %"PRIu16, | |
2384 | br->name, ntohs(flow->dl_vlan), | |
2385 | in_port->name, in_port->vlan); | |
2386 | } | |
2387 | return -1; | |
2388 | } | |
2389 | vlan = in_port->vlan; | |
2390 | } else { | |
2391 | if (!port_includes_vlan(in_port, vlan)) { | |
2392 | if (have_packet) { | |
2393 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2394 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
2395 | "packet received on port %s not configured for " | |
2396 | "trunking VLAN %d", | |
2397 | br->name, vlan, in_port->name, vlan); | |
2398 | } | |
2399 | return -1; | |
2400 | } | |
2401 | } | |
2402 | ||
2403 | return vlan; | |
2404 | } | |
2405 | ||
7febb910 JG |
2406 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
2407 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
2408 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
2409 | static bool | |
ae412e7d | 2410 | is_gratuitous_arp(const struct flow *flow) |
7febb910 JG |
2411 | { |
2412 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
2413 | && eth_addr_is_broadcast(flow->dl_dst) | |
2414 | && (flow->nw_proto == ARP_OP_REPLY | |
2415 | || (flow->nw_proto == ARP_OP_REQUEST | |
2416 | && flow->nw_src == flow->nw_dst))); | |
2417 | } | |
2418 | ||
e96a4d80 | 2419 | static void |
ae412e7d | 2420 | update_learning_table(struct bridge *br, const struct flow *flow, int vlan, |
e96a4d80 JG |
2421 | struct port *in_port) |
2422 | { | |
7febb910 JG |
2423 | enum grat_arp_lock_type lock_type; |
2424 | tag_type rev_tag; | |
2425 | ||
2426 | /* We don't want to learn from gratuitous ARP packets that are reflected | |
2427 | * back over bond slaves so we lock the learning table. */ | |
2428 | lock_type = !is_gratuitous_arp(flow) ? GRAT_ARP_LOCK_NONE : | |
2429 | (in_port->n_ifaces == 1) ? GRAT_ARP_LOCK_SET : | |
2430 | GRAT_ARP_LOCK_CHECK; | |
2431 | ||
2432 | rev_tag = mac_learning_learn(br->ml, flow->dl_src, vlan, in_port->port_idx, | |
2433 | lock_type); | |
e96a4d80 JG |
2434 | if (rev_tag) { |
2435 | /* The log messages here could actually be useful in debugging, | |
2436 | * so keep the rate limit relatively high. */ | |
2437 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, | |
2438 | 300); | |
2439 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
2440 | "on port %s in VLAN %d", | |
2441 | br->name, ETH_ADDR_ARGS(flow->dl_src), | |
2442 | in_port->name, vlan); | |
2443 | ofproto_revalidate(br->ofproto, rev_tag); | |
2444 | } | |
2445 | } | |
2446 | ||
14a34d00 BP |
2447 | /* Determines whether packets in 'flow' within 'br' should be forwarded or |
2448 | * dropped. Returns true if they may be forwarded, false if they should be | |
2449 | * dropped. | |
2450 | * | |
2451 | * If 'have_packet' is true, it indicates that the caller is processing a | |
2452 | * received packet. If 'have_packet' is false, then the caller is just | |
2453 | * revalidating an existing flow because configuration has changed. Either | |
2454 | * way, 'have_packet' only affects logging (there is no point in logging errors | |
2455 | * during revalidation). | |
2456 | * | |
2457 | * Sets '*in_portp' to the input port. This will be a null pointer if | |
2458 | * flow->in_port does not designate a known input port (in which case | |
2459 | * is_admissible() returns false). | |
2460 | * | |
2461 | * When returning true, sets '*vlanp' to the effective VLAN of the input | |
2462 | * packet, as returned by flow_get_vlan(). | |
2463 | * | |
2464 | * May also add tags to '*tags', although the current implementation only does | |
2465 | * so in one special case. | |
2466 | */ | |
064af421 | 2467 | static bool |
ae412e7d | 2468 | is_admissible(struct bridge *br, const struct flow *flow, bool have_packet, |
14a34d00 | 2469 | tag_type *tags, int *vlanp, struct port **in_portp) |
064af421 BP |
2470 | { |
2471 | struct iface *in_iface; | |
2472 | struct port *in_port; | |
064af421 BP |
2473 | int vlan; |
2474 | ||
2475 | /* Find the interface and port structure for the received packet. */ | |
2476 | in_iface = iface_from_dp_ifidx(br, flow->in_port); | |
2477 | if (!in_iface) { | |
2478 | /* No interface? Something fishy... */ | |
14a34d00 | 2479 | if (have_packet) { |
064af421 BP |
2480 | /* Odd. A few possible reasons here: |
2481 | * | |
2482 | * - We deleted an interface but there are still a few packets | |
2483 | * queued up from it. | |
2484 | * | |
2485 | * - Someone externally added an interface (e.g. with "ovs-dpctl | |
2486 | * add-if") that we don't know about. | |
2487 | * | |
2488 | * - Packet arrived on the local port but the local port is not | |
2489 | * one of our bridge ports. | |
2490 | */ | |
2491 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2492 | ||
2493 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
d295e8e9 | 2494 | "interface %"PRIu16, br->name, flow->in_port); |
064af421 BP |
2495 | } |
2496 | ||
14a34d00 BP |
2497 | *in_portp = NULL; |
2498 | return false; | |
064af421 | 2499 | } |
14a34d00 BP |
2500 | *in_portp = in_port = in_iface->port; |
2501 | *vlanp = vlan = flow_get_vlan(br, flow, in_port, have_packet); | |
e96a4d80 | 2502 | if (vlan < 0) { |
14a34d00 | 2503 | return false; |
064af421 BP |
2504 | } |
2505 | ||
064af421 BP |
2506 | /* Drop frames for reserved multicast addresses. */ |
2507 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
14a34d00 | 2508 | return false; |
064af421 BP |
2509 | } |
2510 | ||
2511 | /* Drop frames on ports reserved for mirroring. */ | |
2512 | if (in_port->is_mirror_output_port) { | |
14a34d00 | 2513 | if (have_packet) { |
f925807d BP |
2514 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); |
2515 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
2516 | "%s, which is reserved exclusively for mirroring", | |
2517 | br->name, in_port->name); | |
2518 | } | |
14a34d00 | 2519 | return false; |
064af421 BP |
2520 | } |
2521 | ||
3a55ef14 JG |
2522 | /* Packets received on bonds need special attention to avoid duplicates. */ |
2523 | if (in_port->n_ifaces > 1) { | |
2524 | int src_idx; | |
7febb910 | 2525 | bool is_grat_arp_locked; |
3a55ef14 JG |
2526 | |
2527 | if (eth_addr_is_multicast(flow->dl_dst)) { | |
2528 | *tags |= in_port->active_iface_tag; | |
2529 | if (in_port->active_iface != in_iface->port_ifidx) { | |
2530 | /* Drop all multicast packets on inactive slaves. */ | |
14a34d00 | 2531 | return false; |
c93b1d6a | 2532 | } |
064af421 | 2533 | } |
3a55ef14 JG |
2534 | |
2535 | /* Drop all packets for which we have learned a different input | |
2536 | * port, because we probably sent the packet on one slave and got | |
5d0ae138 | 2537 | * it back on the other. Gratuitous ARP packets are an exception |
7febb910 JG |
2538 | * to this rule: the host has moved to another switch. The exception |
2539 | * to the exception is if we locked the learning table to avoid | |
2540 | * reflections on bond slaves. If this is the case, just drop the | |
2541 | * packet now. */ | |
2542 | src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan, | |
2543 | &is_grat_arp_locked); | |
3a55ef14 | 2544 | if (src_idx != -1 && src_idx != in_port->port_idx && |
7febb910 | 2545 | (!is_gratuitous_arp(flow) || is_grat_arp_locked)) { |
14a34d00 | 2546 | return false; |
3a55ef14 | 2547 | } |
064af421 BP |
2548 | } |
2549 | ||
14a34d00 BP |
2550 | return true; |
2551 | } | |
2552 | ||
2553 | /* If the composed actions may be applied to any packet in the given 'flow', | |
2554 | * returns true. Otherwise, the actions should only be applied to 'packet', or | |
2555 | * not at all, if 'packet' was NULL. */ | |
2556 | static bool | |
ae412e7d | 2557 | process_flow(struct bridge *br, const struct flow *flow, |
14a34d00 BP |
2558 | const struct ofpbuf *packet, struct odp_actions *actions, |
2559 | tag_type *tags, uint16_t *nf_output_iface) | |
2560 | { | |
2561 | struct port *in_port; | |
2562 | struct port *out_port; | |
2563 | int vlan; | |
2564 | int out_port_idx; | |
2565 | ||
2566 | /* Check whether we should drop packets in this flow. */ | |
2567 | if (!is_admissible(br, flow, packet != NULL, tags, &vlan, &in_port)) { | |
2568 | out_port = NULL; | |
2569 | goto done; | |
2570 | } | |
2571 | ||
93dfc067 JG |
2572 | /* Learn source MAC (but don't try to learn from revalidation). */ |
2573 | if (packet) { | |
e96a4d80 | 2574 | update_learning_table(br, flow, vlan, in_port); |
93dfc067 JG |
2575 | } |
2576 | ||
2577 | /* Determine output port. */ | |
7febb910 JG |
2578 | out_port_idx = mac_learning_lookup_tag(br->ml, flow->dl_dst, vlan, tags, |
2579 | NULL); | |
93dfc067 JG |
2580 | if (out_port_idx >= 0 && out_port_idx < br->n_ports) { |
2581 | out_port = br->ports[out_port_idx]; | |
e96a4d80 | 2582 | } else if (!packet && !eth_addr_is_multicast(flow->dl_dst)) { |
93dfc067 | 2583 | /* If we are revalidating but don't have a learning entry then |
e96a4d80 JG |
2584 | * eject the flow. Installing a flow that floods packets opens |
2585 | * up a window of time where we could learn from a packet reflected | |
2586 | * on a bond and blackhole packets before the learning table is | |
2587 | * updated to reflect the correct port. */ | |
93dfc067 | 2588 | return false; |
1609aa03 BP |
2589 | } else { |
2590 | out_port = FLOOD_PORT; | |
064af421 BP |
2591 | } |
2592 | ||
ba54bf4f BP |
2593 | /* Don't send packets out their input ports. */ |
2594 | if (in_port == out_port) { | |
064af421 BP |
2595 | out_port = NULL; |
2596 | } | |
2597 | ||
2598 | done: | |
14a34d00 BP |
2599 | if (in_port) { |
2600 | compose_actions(br, flow, vlan, in_port, out_port, tags, actions, | |
2601 | nf_output_iface); | |
2602 | } | |
064af421 | 2603 | |
69d60f9f | 2604 | return true; |
064af421 BP |
2605 | } |
2606 | ||
064af421 | 2607 | static bool |
ae412e7d | 2608 | bridge_normal_ofhook_cb(const struct flow *flow, const struct ofpbuf *packet, |
6a07af36 JG |
2609 | struct odp_actions *actions, tag_type *tags, |
2610 | uint16_t *nf_output_iface, void *br_) | |
064af421 BP |
2611 | { |
2612 | struct bridge *br = br_; | |
2613 | ||
064af421 | 2614 | COVERAGE_INC(bridge_process_flow); |
26d79bf2 | 2615 | |
6a07af36 | 2616 | return process_flow(br, flow, packet, actions, tags, nf_output_iface); |
064af421 BP |
2617 | } |
2618 | ||
2619 | static void | |
ae412e7d | 2620 | bridge_account_flow_ofhook_cb(const struct flow *flow, tag_type tags, |
064af421 BP |
2621 | const union odp_action *actions, |
2622 | size_t n_actions, unsigned long long int n_bytes, | |
2623 | void *br_) | |
2624 | { | |
2625 | struct bridge *br = br_; | |
2626 | const union odp_action *a; | |
db0e2ad1 | 2627 | struct port *in_port; |
26d79bf2 | 2628 | tag_type dummy = 0; |
db0e2ad1 | 2629 | int vlan; |
064af421 | 2630 | |
26d79bf2 BP |
2631 | /* Feed information from the active flows back into the learning table to |
2632 | * ensure that table is always in sync with what is actually flowing | |
2633 | * through the datapath. | |
2634 | * | |
2635 | * We test that 'tags' is nonzero to ensure that only flows that include an | |
2636 | * OFPP_NORMAL action are used for learning. This works because | |
2637 | * bridge_normal_ofhook_cb() always sets a nonzero tag value. */ | |
2638 | if (tags && is_admissible(br, flow, false, &dummy, &vlan, &in_port)) { | |
db0e2ad1 | 2639 | update_learning_table(br, flow, vlan, in_port); |
e96a4d80 JG |
2640 | } |
2641 | ||
26d79bf2 | 2642 | /* Account for bond slave utilization. */ |
064af421 BP |
2643 | if (!br->has_bonded_ports) { |
2644 | return; | |
2645 | } | |
064af421 BP |
2646 | for (a = actions; a < &actions[n_actions]; a++) { |
2647 | if (a->type == ODPAT_OUTPUT) { | |
e96a4d80 JG |
2648 | struct port *out_port = port_from_dp_ifidx(br, a->output.port); |
2649 | if (out_port && out_port->n_ifaces >= 2) { | |
2650 | struct bond_entry *e = lookup_bond_entry(out_port, | |
2651 | flow->dl_src); | |
064af421 BP |
2652 | e->tx_bytes += n_bytes; |
2653 | } | |
2654 | } | |
2655 | } | |
2656 | } | |
2657 | ||
2658 | static void | |
2659 | bridge_account_checkpoint_ofhook_cb(void *br_) | |
2660 | { | |
2661 | struct bridge *br = br_; | |
c8143c88 | 2662 | long long int now; |
064af421 BP |
2663 | size_t i; |
2664 | ||
2665 | if (!br->has_bonded_ports) { | |
2666 | return; | |
2667 | } | |
2668 | ||
c8143c88 | 2669 | now = time_msec(); |
064af421 BP |
2670 | for (i = 0; i < br->n_ports; i++) { |
2671 | struct port *port = br->ports[i]; | |
c8143c88 BP |
2672 | if (port->n_ifaces > 1 && now >= port->bond_next_rebalance) { |
2673 | port->bond_next_rebalance = now + port->bond_rebalance_interval; | |
064af421 BP |
2674 | bond_rebalance_port(port); |
2675 | } | |
2676 | } | |
2677 | } | |
2678 | ||
2679 | static struct ofhooks bridge_ofhooks = { | |
064af421 BP |
2680 | bridge_normal_ofhook_cb, |
2681 | bridge_account_flow_ofhook_cb, | |
2682 | bridge_account_checkpoint_ofhook_cb, | |
2683 | }; | |
2684 | \f | |
2303f3b2 BP |
2685 | /* Bonding functions. */ |
2686 | ||
064af421 BP |
2687 | /* Statistics for a single interface on a bonded port, used for load-based |
2688 | * bond rebalancing. */ | |
2689 | struct slave_balance { | |
2690 | struct iface *iface; /* The interface. */ | |
2691 | uint64_t tx_bytes; /* Sum of hashes[*]->tx_bytes. */ | |
2692 | ||
2693 | /* All the "bond_entry"s that are assigned to this interface, in order of | |
2694 | * increasing tx_bytes. */ | |
2695 | struct bond_entry **hashes; | |
2696 | size_t n_hashes; | |
2697 | }; | |
2698 | ||
2699 | /* Sorts pointers to pointers to bond_entries in ascending order by the | |
2700 | * interface to which they are assigned, and within a single interface in | |
2701 | * ascending order of bytes transmitted. */ | |
2702 | static int | |
2703 | compare_bond_entries(const void *a_, const void *b_) | |
2704 | { | |
2705 | const struct bond_entry *const *ap = a_; | |
2706 | const struct bond_entry *const *bp = b_; | |
2707 | const struct bond_entry *a = *ap; | |
2708 | const struct bond_entry *b = *bp; | |
2709 | if (a->iface_idx != b->iface_idx) { | |
2710 | return a->iface_idx > b->iface_idx ? 1 : -1; | |
2711 | } else if (a->tx_bytes != b->tx_bytes) { | |
2712 | return a->tx_bytes > b->tx_bytes ? 1 : -1; | |
2713 | } else { | |
2714 | return 0; | |
2715 | } | |
2716 | } | |
2717 | ||
2718 | /* Sorts slave_balances so that enabled ports come first, and otherwise in | |
2719 | * *descending* order by number of bytes transmitted. */ | |
2720 | static int | |
2721 | compare_slave_balance(const void *a_, const void *b_) | |
2722 | { | |
2723 | const struct slave_balance *a = a_; | |
2724 | const struct slave_balance *b = b_; | |
2725 | if (a->iface->enabled != b->iface->enabled) { | |
2726 | return a->iface->enabled ? -1 : 1; | |
2727 | } else if (a->tx_bytes != b->tx_bytes) { | |
2728 | return a->tx_bytes > b->tx_bytes ? -1 : 1; | |
2729 | } else { | |
2730 | return 0; | |
2731 | } | |
2732 | } | |
2733 | ||
2734 | static void | |
2735 | swap_bals(struct slave_balance *a, struct slave_balance *b) | |
2736 | { | |
2737 | struct slave_balance tmp = *a; | |
2738 | *a = *b; | |
2739 | *b = tmp; | |
2740 | } | |
2741 | ||
2742 | /* Restores the 'n_bals' slave_balance structures in 'bals' to sorted order | |
2743 | * given that 'p' (and only 'p') might be in the wrong location. | |
2744 | * | |
2745 | * This function invalidates 'p', since it might now be in a different memory | |
2746 | * location. */ | |
2747 | static void | |
2748 | resort_bals(struct slave_balance *p, | |
2749 | struct slave_balance bals[], size_t n_bals) | |
2750 | { | |
2751 | if (n_bals > 1) { | |
2752 | for (; p > bals && p->tx_bytes > p[-1].tx_bytes; p--) { | |
2753 | swap_bals(p, p - 1); | |
2754 | } | |
2755 | for (; p < &bals[n_bals - 1] && p->tx_bytes < p[1].tx_bytes; p++) { | |
2756 | swap_bals(p, p + 1); | |
2757 | } | |
2758 | } | |
2759 | } | |
2760 | ||
2761 | static void | |
2762 | log_bals(const struct slave_balance *bals, size_t n_bals, struct port *port) | |
2763 | { | |
2764 | if (VLOG_IS_DBG_ENABLED()) { | |
2765 | struct ds ds = DS_EMPTY_INITIALIZER; | |
2766 | const struct slave_balance *b; | |
2767 | ||
2768 | for (b = bals; b < bals + n_bals; b++) { | |
2769 | size_t i; | |
2770 | ||
2771 | if (b > bals) { | |
2772 | ds_put_char(&ds, ','); | |
2773 | } | |
2774 | ds_put_format(&ds, " %s %"PRIu64"kB", | |
2775 | b->iface->name, b->tx_bytes / 1024); | |
2776 | ||
2777 | if (!b->iface->enabled) { | |
2778 | ds_put_cstr(&ds, " (disabled)"); | |
2779 | } | |
2780 | if (b->n_hashes > 0) { | |
2781 | ds_put_cstr(&ds, " ("); | |
2782 | for (i = 0; i < b->n_hashes; i++) { | |
2783 | const struct bond_entry *e = b->hashes[i]; | |
2784 | if (i > 0) { | |
2785 | ds_put_cstr(&ds, " + "); | |
2786 | } | |
2787 | ds_put_format(&ds, "h%td: %"PRIu64"kB", | |
2788 | e - port->bond_hash, e->tx_bytes / 1024); | |
2789 | } | |
2790 | ds_put_cstr(&ds, ")"); | |
2791 | } | |
2792 | } | |
2793 | VLOG_DBG("bond %s:%s", port->name, ds_cstr(&ds)); | |
2794 | ds_destroy(&ds); | |
2795 | } | |
2796 | } | |
2797 | ||
2798 | /* Shifts 'hash' from 'from' to 'to' within 'port'. */ | |
2799 | static void | |
2800 | bond_shift_load(struct slave_balance *from, struct slave_balance *to, | |
5422a9e1 | 2801 | int hash_idx) |
064af421 | 2802 | { |
5422a9e1 | 2803 | struct bond_entry *hash = from->hashes[hash_idx]; |
064af421 BP |
2804 | struct port *port = from->iface->port; |
2805 | uint64_t delta = hash->tx_bytes; | |
2806 | ||
2807 | VLOG_INFO("bond %s: shift %"PRIu64"kB of load (with hash %td) " | |
2808 | "from %s to %s (now carrying %"PRIu64"kB and " | |
2809 | "%"PRIu64"kB load, respectively)", | |
2810 | port->name, delta / 1024, hash - port->bond_hash, | |
2811 | from->iface->name, to->iface->name, | |
2812 | (from->tx_bytes - delta) / 1024, | |
2813 | (to->tx_bytes + delta) / 1024); | |
2814 | ||
2815 | /* Delete element from from->hashes. | |
2816 | * | |
2817 | * We don't bother to add the element to to->hashes because not only would | |
2818 | * it require more work, the only purpose it would be to allow that hash to | |
2819 | * be migrated to another slave in this rebalancing run, and there is no | |
2820 | * point in doing that. */ | |
5422a9e1 | 2821 | if (hash_idx == 0) { |
064af421 BP |
2822 | from->hashes++; |
2823 | } else { | |
5422a9e1 JG |
2824 | memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1, |
2825 | (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes); | |
064af421 BP |
2826 | } |
2827 | from->n_hashes--; | |
2828 | ||
2829 | /* Shift load away from 'from' to 'to'. */ | |
2830 | from->tx_bytes -= delta; | |
2831 | to->tx_bytes += delta; | |
2832 | ||
2833 | /* Arrange for flows to be revalidated. */ | |
2834 | ofproto_revalidate(port->bridge->ofproto, hash->iface_tag); | |
2835 | hash->iface_idx = to->iface->port_ifidx; | |
2836 | hash->iface_tag = tag_create_random(); | |
064af421 BP |
2837 | } |
2838 | ||
2839 | static void | |
2840 | bond_rebalance_port(struct port *port) | |
2841 | { | |
2842 | struct slave_balance bals[DP_MAX_PORTS]; | |
2843 | size_t n_bals; | |
2844 | struct bond_entry *hashes[BOND_MASK + 1]; | |
2845 | struct slave_balance *b, *from, *to; | |
2846 | struct bond_entry *e; | |
2847 | size_t i; | |
2848 | ||
2849 | /* Sets up 'bals' to describe each of the port's interfaces, sorted in | |
2850 | * descending order of tx_bytes, so that bals[0] represents the most | |
2851 | * heavily loaded slave and bals[n_bals - 1] represents the least heavily | |
2852 | * loaded slave. | |
2853 | * | |
2854 | * The code is a bit tricky: to avoid dynamically allocating a 'hashes' | |
2855 | * array for each slave_balance structure, we sort our local array of | |
2856 | * hashes in order by slave, so that all of the hashes for a given slave | |
2857 | * become contiguous in memory, and then we point each 'hashes' members of | |
2858 | * a slave_balance structure to the start of a contiguous group. */ | |
2859 | n_bals = port->n_ifaces; | |
2860 | for (b = bals; b < &bals[n_bals]; b++) { | |
2861 | b->iface = port->ifaces[b - bals]; | |
2862 | b->tx_bytes = 0; | |
2863 | b->hashes = NULL; | |
2864 | b->n_hashes = 0; | |
2865 | } | |
2866 | for (i = 0; i <= BOND_MASK; i++) { | |
2867 | hashes[i] = &port->bond_hash[i]; | |
2868 | } | |
2869 | qsort(hashes, BOND_MASK + 1, sizeof *hashes, compare_bond_entries); | |
2870 | for (i = 0; i <= BOND_MASK; i++) { | |
2871 | e = hashes[i]; | |
2872 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
2873 | b = &bals[e->iface_idx]; | |
2874 | b->tx_bytes += e->tx_bytes; | |
2875 | if (!b->hashes) { | |
2876 | b->hashes = &hashes[i]; | |
2877 | } | |
2878 | b->n_hashes++; | |
2879 | } | |
2880 | } | |
2881 | qsort(bals, n_bals, sizeof *bals, compare_slave_balance); | |
2882 | log_bals(bals, n_bals, port); | |
2883 | ||
2884 | /* Discard slaves that aren't enabled (which were sorted to the back of the | |
2885 | * array earlier). */ | |
2886 | while (!bals[n_bals - 1].iface->enabled) { | |
2887 | n_bals--; | |
2888 | if (!n_bals) { | |
2889 | return; | |
2890 | } | |
2891 | } | |
2892 | ||
2893 | /* Shift load from the most-loaded slaves to the least-loaded slaves. */ | |
2894 | to = &bals[n_bals - 1]; | |
2895 | for (from = bals; from < to; ) { | |
2896 | uint64_t overload = from->tx_bytes - to->tx_bytes; | |
2897 | if (overload < to->tx_bytes >> 5 || overload < 100000) { | |
2898 | /* The extra load on 'from' (and all less-loaded slaves), compared | |
2899 | * to that of 'to' (the least-loaded slave), is less than ~3%, or | |
2900 | * it is less than ~1Mbps. No point in rebalancing. */ | |
2901 | break; | |
2902 | } else if (from->n_hashes == 1) { | |
2903 | /* 'from' only carries a single MAC hash, so we can't shift any | |
2904 | * load away from it, even though we want to. */ | |
2905 | from++; | |
2906 | } else { | |
2907 | /* 'from' is carrying significantly more load than 'to', and that | |
2908 | * load is split across at least two different hashes. Pick a hash | |
2909 | * to migrate to 'to' (the least-loaded slave), given that doing so | |
5422a9e1 JG |
2910 | * must decrease the ratio of the load on the two slaves by at |
2911 | * least 0.1. | |
064af421 BP |
2912 | * |
2913 | * The sort order we use means that we prefer to shift away the | |
2914 | * smallest hashes instead of the biggest ones. There is little | |
2915 | * reason behind this decision; we could use the opposite sort | |
2916 | * order to shift away big hashes ahead of small ones. */ | |
5422a9e1 | 2917 | bool order_swapped; |
064af421 BP |
2918 | |
2919 | for (i = 0; i < from->n_hashes; i++) { | |
5422a9e1 | 2920 | double old_ratio, new_ratio; |
064af421 | 2921 | uint64_t delta = from->hashes[i]->tx_bytes; |
5422a9e1 JG |
2922 | |
2923 | if (delta == 0 || from->tx_bytes - delta == 0) { | |
2924 | /* Pointless move. */ | |
2925 | continue; | |
2926 | } | |
2927 | ||
2928 | order_swapped = from->tx_bytes - delta < to->tx_bytes + delta; | |
2929 | ||
2930 | if (to->tx_bytes == 0) { | |
2931 | /* Nothing on the new slave, move it. */ | |
2932 | break; | |
2933 | } | |
2934 | ||
2935 | old_ratio = (double)from->tx_bytes / to->tx_bytes; | |
2936 | new_ratio = (double)(from->tx_bytes - delta) / | |
2937 | (to->tx_bytes + delta); | |
2938 | ||
2939 | if (new_ratio == 0) { | |
2940 | /* Should already be covered but check to prevent division | |
2941 | * by zero. */ | |
2942 | continue; | |
2943 | } | |
2944 | ||
2945 | if (new_ratio < 1) { | |
2946 | new_ratio = 1 / new_ratio; | |
2947 | } | |
2948 | ||
2949 | if (old_ratio - new_ratio > 0.1) { | |
2950 | /* Would decrease the ratio, move it. */ | |
064af421 BP |
2951 | break; |
2952 | } | |
2953 | } | |
2954 | if (i < from->n_hashes) { | |
5422a9e1 JG |
2955 | bond_shift_load(from, to, i); |
2956 | port->bond_compat_is_stale = true; | |
2957 | ||
2958 | /* If the result of the migration changed the relative order of | |
2959 | * 'from' and 'to' swap them back to maintain invariants. */ | |
2960 | if (order_swapped) { | |
2961 | swap_bals(from, to); | |
2962 | } | |
064af421 BP |
2963 | |
2964 | /* Re-sort 'bals'. Note that this may make 'from' and 'to' | |
2965 | * point to different slave_balance structures. It is only | |
2966 | * valid to do these two operations in a row at all because we | |
2967 | * know that 'from' will not move past 'to' and vice versa. */ | |
2968 | resort_bals(from, bals, n_bals); | |
2969 | resort_bals(to, bals, n_bals); | |
2970 | } else { | |
2971 | from++; | |
2972 | } | |
2973 | } | |
2974 | } | |
2975 | ||
2976 | /* Implement exponentially weighted moving average. A weight of 1/2 causes | |
2977 | * historical data to decay to <1% in 7 rebalancing runs. */ | |
2978 | for (e = &port->bond_hash[0]; e <= &port->bond_hash[BOND_MASK]; e++) { | |
2979 | e->tx_bytes /= 2; | |
2980 | } | |
2981 | } | |
2303f3b2 BP |
2982 | |
2983 | static void | |
2984 | bond_send_learning_packets(struct port *port) | |
2985 | { | |
2986 | struct bridge *br = port->bridge; | |
2987 | struct mac_entry *e; | |
2988 | struct ofpbuf packet; | |
2989 | int error, n_packets, n_errors; | |
2990 | ||
93dfc067 | 2991 | if (!port->n_ifaces || port->active_iface < 0) { |
2303f3b2 BP |
2992 | return; |
2993 | } | |
2994 | ||
2995 | ofpbuf_init(&packet, 128); | |
2996 | error = n_packets = n_errors = 0; | |
4e8e4213 | 2997 | LIST_FOR_EACH (e, lru_node, &br->ml->lrus) { |
2303f3b2 | 2998 | union ofp_action actions[2], *a; |
2303f3b2 BP |
2999 | uint16_t dp_ifidx; |
3000 | tag_type tags = 0; | |
ae412e7d | 3001 | struct flow flow; |
2303f3b2 BP |
3002 | int retval; |
3003 | ||
3004 | if (e->port == port->port_idx | |
3005 | || !choose_output_iface(port, e->mac, &dp_ifidx, &tags)) { | |
3006 | continue; | |
3007 | } | |
3008 | ||
2303f3b2 BP |
3009 | /* Compose actions. */ |
3010 | memset(actions, 0, sizeof actions); | |
3011 | a = actions; | |
3012 | if (e->vlan) { | |
3013 | a->vlan_vid.type = htons(OFPAT_SET_VLAN_VID); | |
3014 | a->vlan_vid.len = htons(sizeof *a); | |
3015 | a->vlan_vid.vlan_vid = htons(e->vlan); | |
3016 | a++; | |
3017 | } | |
3018 | a->output.type = htons(OFPAT_OUTPUT); | |
3019 | a->output.len = htons(sizeof *a); | |
3020 | a->output.port = htons(odp_port_to_ofp_port(dp_ifidx)); | |
3021 | a++; | |
3022 | ||
3023 | /* Send packet. */ | |
3024 | n_packets++; | |
b9e8b45a BP |
3025 | compose_benign_packet(&packet, "Open vSwitch Bond Failover", 0xf177, |
3026 | e->mac); | |
659586ef | 3027 | flow_extract(&packet, 0, ODPP_NONE, &flow); |
2303f3b2 BP |
3028 | retval = ofproto_send_packet(br->ofproto, &flow, actions, a - actions, |
3029 | &packet); | |
3030 | if (retval) { | |
3031 | error = retval; | |
3032 | n_errors++; | |
3033 | } | |
3034 | } | |
3035 | ofpbuf_uninit(&packet); | |
3036 | ||
3037 | if (n_errors) { | |
3038 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3039 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
3040 | "packets, last error was: %s", | |
3041 | port->name, n_errors, n_packets, strerror(error)); | |
3042 | } else { | |
3043 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
3044 | port->name, n_packets); | |
3045 | } | |
3046 | } | |
064af421 | 3047 | \f |
da285df4 BP |
3048 | /* Bonding unixctl user interface functions. */ |
3049 | ||
3050 | static void | |
8ca79daa | 3051 | bond_unixctl_list(struct unixctl_conn *conn, |
c69ee87c | 3052 | const char *args OVS_UNUSED, void *aux OVS_UNUSED) |
da285df4 BP |
3053 | { |
3054 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3055 | const struct bridge *br; | |
3056 | ||
3057 | ds_put_cstr(&ds, "bridge\tbond\tslaves\n"); | |
3058 | ||
4e8e4213 | 3059 | LIST_FOR_EACH (br, node, &all_bridges) { |
da285df4 BP |
3060 | size_t i; |
3061 | ||
3062 | for (i = 0; i < br->n_ports; i++) { | |
3063 | const struct port *port = br->ports[i]; | |
3064 | if (port->n_ifaces > 1) { | |
3065 | size_t j; | |
3066 | ||
3067 | ds_put_format(&ds, "%s\t%s\t", br->name, port->name); | |
3068 | for (j = 0; j < port->n_ifaces; j++) { | |
3069 | const struct iface *iface = port->ifaces[j]; | |
3070 | if (j) { | |
3071 | ds_put_cstr(&ds, ", "); | |
3072 | } | |
3073 | ds_put_cstr(&ds, iface->name); | |
3074 | } | |
3075 | ds_put_char(&ds, '\n'); | |
3076 | } | |
3077 | } | |
3078 | } | |
3079 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
3080 | ds_destroy(&ds); | |
3081 | } | |
3082 | ||
3083 | static struct port * | |
3084 | bond_find(const char *name) | |
3085 | { | |
3086 | const struct bridge *br; | |
3087 | ||
4e8e4213 | 3088 | LIST_FOR_EACH (br, node, &all_bridges) { |
da285df4 BP |
3089 | size_t i; |
3090 | ||
3091 | for (i = 0; i < br->n_ports; i++) { | |
3092 | struct port *port = br->ports[i]; | |
3093 | if (!strcmp(port->name, name) && port->n_ifaces > 1) { | |
3094 | return port; | |
3095 | } | |
3096 | } | |
3097 | } | |
3098 | return NULL; | |
3099 | } | |
3100 | ||
3101 | static void | |
8ca79daa | 3102 | bond_unixctl_show(struct unixctl_conn *conn, |
c69ee87c | 3103 | const char *args, void *aux OVS_UNUSED) |
da285df4 BP |
3104 | { |
3105 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3106 | const struct port *port; | |
3107 | size_t j; | |
3108 | ||
3109 | port = bond_find(args); | |
3110 | if (!port) { | |
3111 | unixctl_command_reply(conn, 501, "no such bond"); | |
3112 | return; | |
3113 | } | |
3114 | ||
3115 | ds_put_format(&ds, "updelay: %d ms\n", port->updelay); | |
3116 | ds_put_format(&ds, "downdelay: %d ms\n", port->downdelay); | |
3117 | ds_put_format(&ds, "next rebalance: %lld ms\n", | |
c8143c88 | 3118 | port->bond_next_rebalance - time_msec()); |
da285df4 BP |
3119 | for (j = 0; j < port->n_ifaces; j++) { |
3120 | const struct iface *iface = port->ifaces[j]; | |
3121 | struct bond_entry *be; | |
3122 | ||
3123 | /* Basic info. */ | |
3124 | ds_put_format(&ds, "slave %s: %s\n", | |
3125 | iface->name, iface->enabled ? "enabled" : "disabled"); | |
3126 | if (j == port->active_iface) { | |
3127 | ds_put_cstr(&ds, "\tactive slave\n"); | |
3128 | } | |
3129 | if (iface->delay_expires != LLONG_MAX) { | |
3130 | ds_put_format(&ds, "\t%s expires in %lld ms\n", | |
3131 | iface->enabled ? "downdelay" : "updelay", | |
3132 | iface->delay_expires - time_msec()); | |
3133 | } | |
3134 | ||
3135 | /* Hashes. */ | |
3136 | for (be = port->bond_hash; be <= &port->bond_hash[BOND_MASK]; be++) { | |
3137 | int hash = be - port->bond_hash; | |
3138 | struct mac_entry *me; | |
3139 | ||
3140 | if (be->iface_idx != j) { | |
3141 | continue; | |
3142 | } | |
3143 | ||
2886875a | 3144 | ds_put_format(&ds, "\thash %d: %"PRIu64" kB load\n", |
da285df4 BP |
3145 | hash, be->tx_bytes / 1024); |
3146 | ||
3147 | /* MACs. */ | |
4e8e4213 | 3148 | LIST_FOR_EACH (me, lru_node, &port->bridge->ml->lrus) { |
da285df4 BP |
3149 | uint16_t dp_ifidx; |
3150 | tag_type tags = 0; | |
3151 | if (bond_hash(me->mac) == hash | |
3152 | && me->port != port->port_idx | |
3153 | && choose_output_iface(port, me->mac, &dp_ifidx, &tags) | |
3154 | && dp_ifidx == iface->dp_ifidx) | |
3155 | { | |
3156 | ds_put_format(&ds, "\t\t"ETH_ADDR_FMT"\n", | |
3157 | ETH_ADDR_ARGS(me->mac)); | |
3158 | } | |
3159 | } | |
3160 | } | |
3161 | } | |
3162 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
3163 | ds_destroy(&ds); | |
3164 | } | |
3165 | ||
3166 | static void | |
8ca79daa | 3167 | bond_unixctl_migrate(struct unixctl_conn *conn, const char *args_, |
c69ee87c | 3168 | void *aux OVS_UNUSED) |
da285df4 BP |
3169 | { |
3170 | char *args = (char *) args_; | |
3171 | char *save_ptr = NULL; | |
3172 | char *bond_s, *hash_s, *slave_s; | |
3173 | uint8_t mac[ETH_ADDR_LEN]; | |
3174 | struct port *port; | |
3175 | struct iface *iface; | |
3176 | struct bond_entry *entry; | |
3177 | int hash; | |
3178 | ||
3179 | bond_s = strtok_r(args, " ", &save_ptr); | |
3180 | hash_s = strtok_r(NULL, " ", &save_ptr); | |
3181 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
3182 | if (!slave_s) { | |
3183 | unixctl_command_reply(conn, 501, | |
3184 | "usage: bond/migrate BOND HASH SLAVE"); | |
3185 | return; | |
3186 | } | |
3187 | ||
3188 | port = bond_find(bond_s); | |
3189 | if (!port) { | |
3190 | unixctl_command_reply(conn, 501, "no such bond"); | |
3191 | return; | |
3192 | } | |
3193 | ||
eaa71334 BP |
3194 | if (sscanf(hash_s, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) |
3195 | == ETH_ADDR_SCAN_COUNT) { | |
da285df4 BP |
3196 | hash = bond_hash(mac); |
3197 | } else if (strspn(hash_s, "0123456789") == strlen(hash_s)) { | |
3198 | hash = atoi(hash_s) & BOND_MASK; | |
3199 | } else { | |
3200 | unixctl_command_reply(conn, 501, "bad hash"); | |
3201 | return; | |
3202 | } | |
3203 | ||
3204 | iface = port_lookup_iface(port, slave_s); | |
3205 | if (!iface) { | |
3206 | unixctl_command_reply(conn, 501, "no such slave"); | |
3207 | return; | |
3208 | } | |
3209 | ||
3210 | if (!iface->enabled) { | |
3211 | unixctl_command_reply(conn, 501, "cannot migrate to disabled slave"); | |
3212 | return; | |
3213 | } | |
3214 | ||
3215 | entry = &port->bond_hash[hash]; | |
3216 | ofproto_revalidate(port->bridge->ofproto, entry->iface_tag); | |
3217 | entry->iface_idx = iface->port_ifidx; | |
3218 | entry->iface_tag = tag_create_random(); | |
85c74638 | 3219 | port->bond_compat_is_stale = true; |
da285df4 BP |
3220 | unixctl_command_reply(conn, 200, "migrated"); |
3221 | } | |
3222 | ||
3223 | static void | |
8ca79daa | 3224 | bond_unixctl_set_active_slave(struct unixctl_conn *conn, const char *args_, |
c69ee87c | 3225 | void *aux OVS_UNUSED) |
da285df4 BP |
3226 | { |
3227 | char *args = (char *) args_; | |
3228 | char *save_ptr = NULL; | |
3229 | char *bond_s, *slave_s; | |
3230 | struct port *port; | |
3231 | struct iface *iface; | |
3232 | ||
3233 | bond_s = strtok_r(args, " ", &save_ptr); | |
3234 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
3235 | if (!slave_s) { | |
3236 | unixctl_command_reply(conn, 501, | |
3237 | "usage: bond/set-active-slave BOND SLAVE"); | |
3238 | return; | |
3239 | } | |
3240 | ||
3241 | port = bond_find(bond_s); | |
3242 | if (!port) { | |
3243 | unixctl_command_reply(conn, 501, "no such bond"); | |
3244 | return; | |
3245 | } | |
3246 | ||
3247 | iface = port_lookup_iface(port, slave_s); | |
3248 | if (!iface) { | |
3249 | unixctl_command_reply(conn, 501, "no such slave"); | |
3250 | return; | |
3251 | } | |
3252 | ||
3253 | if (!iface->enabled) { | |
3254 | unixctl_command_reply(conn, 501, "cannot make disabled slave active"); | |
3255 | return; | |
3256 | } | |
3257 | ||
3258 | if (port->active_iface != iface->port_ifidx) { | |
3259 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
3260 | port->active_iface = iface->port_ifidx; | |
3261 | port->active_iface_tag = tag_create_random(); | |
3262 | VLOG_INFO("port %s: active interface is now %s", | |
3263 | port->name, iface->name); | |
3264 | bond_send_learning_packets(port); | |
3265 | unixctl_command_reply(conn, 200, "done"); | |
3266 | } else { | |
3267 | unixctl_command_reply(conn, 200, "no change"); | |
3268 | } | |
3269 | } | |
3270 | ||
3271 | static void | |
3272 | enable_slave(struct unixctl_conn *conn, const char *args_, bool enable) | |
3273 | { | |
3274 | char *args = (char *) args_; | |
3275 | char *save_ptr = NULL; | |
3276 | char *bond_s, *slave_s; | |
3277 | struct port *port; | |
3278 | struct iface *iface; | |
3279 | ||
3280 | bond_s = strtok_r(args, " ", &save_ptr); | |
3281 | slave_s = strtok_r(NULL, " ", &save_ptr); | |
3282 | if (!slave_s) { | |
3283 | unixctl_command_reply(conn, 501, | |
3284 | "usage: bond/enable/disable-slave BOND SLAVE"); | |
3285 | return; | |
3286 | } | |
3287 | ||
3288 | port = bond_find(bond_s); | |
3289 | if (!port) { | |
3290 | unixctl_command_reply(conn, 501, "no such bond"); | |
3291 | return; | |
3292 | } | |
3293 | ||
3294 | iface = port_lookup_iface(port, slave_s); | |
3295 | if (!iface) { | |
3296 | unixctl_command_reply(conn, 501, "no such slave"); | |
3297 | return; | |
3298 | } | |
3299 | ||
3300 | bond_enable_slave(iface, enable); | |
3301 | unixctl_command_reply(conn, 501, enable ? "enabled" : "disabled"); | |
3302 | } | |
3303 | ||
3304 | static void | |
8ca79daa | 3305 | bond_unixctl_enable_slave(struct unixctl_conn *conn, const char *args, |
c69ee87c | 3306 | void *aux OVS_UNUSED) |
da285df4 BP |
3307 | { |
3308 | enable_slave(conn, args, true); | |
3309 | } | |
3310 | ||
3311 | static void | |
8ca79daa | 3312 | bond_unixctl_disable_slave(struct unixctl_conn *conn, const char *args, |
c69ee87c | 3313 | void *aux OVS_UNUSED) |
da285df4 BP |
3314 | { |
3315 | enable_slave(conn, args, false); | |
3316 | } | |
3317 | ||
e0644b61 | 3318 | static void |
8ca79daa | 3319 | bond_unixctl_hash(struct unixctl_conn *conn, const char *args, |
c69ee87c | 3320 | void *aux OVS_UNUSED) |
e0644b61 IC |
3321 | { |
3322 | uint8_t mac[ETH_ADDR_LEN]; | |
3323 | uint8_t hash; | |
3324 | char *hash_cstr; | |
3325 | ||
3326 | if (sscanf(args, ETH_ADDR_SCAN_FMT, ETH_ADDR_SCAN_ARGS(mac)) | |
3327 | == ETH_ADDR_SCAN_COUNT) { | |
3328 | hash = bond_hash(mac); | |
3329 | ||
3330 | hash_cstr = xasprintf("%u", hash); | |
3331 | unixctl_command_reply(conn, 200, hash_cstr); | |
3332 | free(hash_cstr); | |
3333 | } else { | |
3334 | unixctl_command_reply(conn, 501, "invalid mac"); | |
3335 | } | |
3336 | } | |
3337 | ||
da285df4 BP |
3338 | static void |
3339 | bond_init(void) | |
3340 | { | |
8ca79daa BP |
3341 | unixctl_command_register("bond/list", bond_unixctl_list, NULL); |
3342 | unixctl_command_register("bond/show", bond_unixctl_show, NULL); | |
3343 | unixctl_command_register("bond/migrate", bond_unixctl_migrate, NULL); | |
da285df4 | 3344 | unixctl_command_register("bond/set-active-slave", |
8ca79daa BP |
3345 | bond_unixctl_set_active_slave, NULL); |
3346 | unixctl_command_register("bond/enable-slave", bond_unixctl_enable_slave, | |
3347 | NULL); | |
3348 | unixctl_command_register("bond/disable-slave", bond_unixctl_disable_slave, | |
3349 | NULL); | |
3350 | unixctl_command_register("bond/hash", bond_unixctl_hash, NULL); | |
da285df4 BP |
3351 | } |
3352 | \f | |
064af421 BP |
3353 | /* Port functions. */ |
3354 | ||
76343538 | 3355 | static struct port * |
064af421 BP |
3356 | port_create(struct bridge *br, const char *name) |
3357 | { | |
3358 | struct port *port; | |
3359 | ||
ec6fde61 | 3360 | port = xzalloc(sizeof *port); |
064af421 BP |
3361 | port->bridge = br; |
3362 | port->port_idx = br->n_ports; | |
3363 | port->vlan = -1; | |
3364 | port->trunks = NULL; | |
3365 | port->name = xstrdup(name); | |
3366 | port->active_iface = -1; | |
064af421 BP |
3367 | |
3368 | if (br->n_ports >= br->allocated_ports) { | |
3369 | br->ports = x2nrealloc(br->ports, &br->allocated_ports, | |
3370 | sizeof *br->ports); | |
3371 | } | |
3372 | br->ports[br->n_ports++] = port; | |
836fad5e | 3373 | shash_add_assert(&br->port_by_name, port->name, port); |
064af421 BP |
3374 | |
3375 | VLOG_INFO("created port %s on bridge %s", port->name, br->name); | |
3376 | bridge_flush(br); | |
76343538 BP |
3377 | |
3378 | return port; | |
064af421 BP |
3379 | } |
3380 | ||
c8143c88 BP |
3381 | static const char * |
3382 | get_port_other_config(const struct ovsrec_port *port, const char *key, | |
3383 | const char *default_value) | |
3384 | { | |
af6278e1 BP |
3385 | const char *value; |
3386 | ||
3387 | value = get_ovsrec_key_value(&port->header_, &ovsrec_port_col_other_config, | |
3388 | key); | |
c8143c88 BP |
3389 | return value ? value : default_value; |
3390 | } | |
3391 | ||
4a1ee6ae BP |
3392 | static void |
3393 | port_del_ifaces(struct port *port, const struct ovsrec_port *cfg) | |
3394 | { | |
3395 | struct shash new_ifaces; | |
3396 | size_t i; | |
3397 | ||
3398 | /* Collect list of new interfaces. */ | |
3399 | shash_init(&new_ifaces); | |
3400 | for (i = 0; i < cfg->n_interfaces; i++) { | |
3401 | const char *name = cfg->interfaces[i]->name; | |
3402 | shash_add_once(&new_ifaces, name, NULL); | |
3403 | } | |
3404 | ||
3405 | /* Get rid of deleted interfaces. */ | |
3406 | for (i = 0; i < port->n_ifaces; ) { | |
3407 | if (!shash_find(&new_ifaces, cfg->interfaces[i]->name)) { | |
3408 | iface_destroy(port->ifaces[i]); | |
3409 | } else { | |
3410 | i++; | |
3411 | } | |
3412 | } | |
3413 | ||
3414 | shash_destroy(&new_ifaces); | |
3415 | } | |
3416 | ||
064af421 | 3417 | static void |
76343538 | 3418 | port_reconfigure(struct port *port, const struct ovsrec_port *cfg) |
064af421 | 3419 | { |
4a1ee6ae | 3420 | struct shash new_ifaces; |
c8143c88 | 3421 | long long int next_rebalance; |
064af421 BP |
3422 | unsigned long *trunks; |
3423 | int vlan; | |
3424 | size_t i; | |
3425 | ||
76343538 BP |
3426 | port->cfg = cfg; |
3427 | ||
4a1ee6ae | 3428 | /* Update settings. */ |
76343538 BP |
3429 | port->updelay = cfg->bond_updelay; |
3430 | if (port->updelay < 0) { | |
3431 | port->updelay = 0; | |
3432 | } | |
7ef6b685 | 3433 | port->downdelay = cfg->bond_downdelay; |
76343538 BP |
3434 | if (port->downdelay < 0) { |
3435 | port->downdelay = 0; | |
064af421 | 3436 | } |
c8143c88 BP |
3437 | port->bond_rebalance_interval = atoi( |
3438 | get_port_other_config(cfg, "bond-rebalance-interval", "10000")); | |
3439 | if (port->bond_rebalance_interval < 1000) { | |
3440 | port->bond_rebalance_interval = 1000; | |
3441 | } | |
3442 | next_rebalance = time_msec() + port->bond_rebalance_interval; | |
3443 | if (port->bond_next_rebalance > next_rebalance) { | |
3444 | port->bond_next_rebalance = next_rebalance; | |
3445 | } | |
064af421 | 3446 | |
4a1ee6ae BP |
3447 | /* Add new interfaces and update 'cfg' member of existing ones. */ |
3448 | shash_init(&new_ifaces); | |
3449 | for (i = 0; i < cfg->n_interfaces; i++) { | |
3450 | const struct ovsrec_interface *if_cfg = cfg->interfaces[i]; | |
76343538 BP |
3451 | struct iface *iface; |
3452 | ||
4a1ee6ae BP |
3453 | if (!shash_add_once(&new_ifaces, if_cfg->name, NULL)) { |
3454 | VLOG_WARN("port %s: %s specified twice as port interface", | |
3455 | port->name, if_cfg->name); | |
bcd49a45 | 3456 | iface_set_ofport(if_cfg, -1); |
4a1ee6ae BP |
3457 | continue; |
3458 | } | |
3459 | ||
3460 | iface = iface_lookup(port->bridge, if_cfg->name); | |
3461 | if (iface) { | |
3462 | if (iface->port != port) { | |
3463 | VLOG_ERR("bridge %s: %s interface is on multiple ports, " | |
3464 | "removing from %s", | |
3465 | port->bridge->name, if_cfg->name, iface->port->name); | |
3466 | continue; | |
3467 | } | |
149f577a | 3468 | iface->cfg = if_cfg; |
4a1ee6ae | 3469 | } else { |
6cefe1da | 3470 | iface = iface_create(port, if_cfg); |
064af421 | 3471 | } |
6cefe1da BP |
3472 | |
3473 | /* Determine interface type. The local port always has type | |
3474 | * "internal". Other ports take their type from the database and | |
3475 | * default to "system" if none is specified. */ | |
3476 | iface->type = (!strcmp(if_cfg->name, port->bridge->name) ? "internal" | |
3477 | : if_cfg->type[0] ? if_cfg->type | |
3478 | : "system"); | |
064af421 | 3479 | } |
4a1ee6ae | 3480 | shash_destroy(&new_ifaces); |
064af421 BP |
3481 | |
3482 | /* Get VLAN tag. */ | |
3483 | vlan = -1; | |
76343538 BP |
3484 | if (cfg->tag) { |
3485 | if (port->n_ifaces < 2) { | |
3486 | vlan = *cfg->tag; | |
064af421 BP |
3487 | if (vlan >= 0 && vlan <= 4095) { |
3488 | VLOG_DBG("port %s: assigning VLAN tag %d", port->name, vlan); | |
76343538 BP |
3489 | } else { |
3490 | vlan = -1; | |
064af421 BP |
3491 | } |
3492 | } else { | |
3493 | /* It's possible that bonded, VLAN-tagged ports make sense. Maybe | |
3494 | * they even work as-is. But they have not been tested. */ | |
3495 | VLOG_WARN("port %s: VLAN tags not supported on bonded ports", | |
3496 | port->name); | |
3497 | } | |
3498 | } | |
3499 | if (port->vlan != vlan) { | |
3500 | port->vlan = vlan; | |
3501 | bridge_flush(port->bridge); | |
3502 | } | |
3503 | ||
3504 | /* Get trunked VLANs. */ | |
3505 | trunks = NULL; | |
3e9c481c | 3506 | if (vlan < 0 && cfg->n_trunks) { |
76343538 | 3507 | size_t n_errors; |
064af421 BP |
3508 | |
3509 | trunks = bitmap_allocate(4096); | |
064af421 | 3510 | n_errors = 0; |
76343538 BP |
3511 | for (i = 0; i < cfg->n_trunks; i++) { |
3512 | int trunk = cfg->trunks[i]; | |
064af421 BP |
3513 | if (trunk >= 0) { |
3514 | bitmap_set1(trunks, trunk); | |
3515 | } else { | |
3516 | n_errors++; | |
3517 | } | |
3518 | } | |
3519 | if (n_errors) { | |
3520 | VLOG_ERR("port %s: invalid values for %zu trunk VLANs", | |
76343538 | 3521 | port->name, cfg->n_trunks); |
064af421 | 3522 | } |
76343538 | 3523 | if (n_errors == cfg->n_trunks) { |
3e9c481c | 3524 | VLOG_ERR("port %s: no valid trunks, trunking all VLANs", |
76343538 | 3525 | port->name); |
3e9c481c BP |
3526 | bitmap_free(trunks); |
3527 | trunks = NULL; | |
064af421 | 3528 | } |
3e9c481c BP |
3529 | } else if (vlan >= 0 && cfg->n_trunks) { |
3530 | VLOG_ERR("port %s: ignoring trunks in favor of implicit vlan", | |
3531 | port->name); | |
064af421 BP |
3532 | } |
3533 | if (trunks == NULL | |
3534 | ? port->trunks != NULL | |
3535 | : port->trunks == NULL || !bitmap_equal(trunks, port->trunks, 4096)) { | |
3536 | bridge_flush(port->bridge); | |
3537 | } | |
3538 | bitmap_free(port->trunks); | |
3539 | port->trunks = trunks; | |
064af421 BP |
3540 | } |
3541 | ||
3542 | static void | |
3543 | port_destroy(struct port *port) | |
3544 | { | |
3545 | if (port) { | |
3546 | struct bridge *br = port->bridge; | |
3547 | struct port *del; | |
37e7f427 | 3548 | int i; |
064af421 BP |
3549 | |
3550 | proc_net_compat_update_vlan(port->name, NULL, 0); | |
85c74638 | 3551 | proc_net_compat_update_bond(port->name, NULL); |
064af421 BP |
3552 | |
3553 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3554 | struct mirror *m = br->mirrors[i]; | |
3555 | if (m && m->out_port == port) { | |
3556 | mirror_destroy(m); | |
3557 | } | |
3558 | } | |
3559 | ||
3560 | while (port->n_ifaces > 0) { | |
3561 | iface_destroy(port->ifaces[port->n_ifaces - 1]); | |
3562 | } | |
3563 | ||
836fad5e BP |
3564 | shash_find_and_delete_assert(&br->port_by_name, port->name); |
3565 | ||
064af421 BP |
3566 | del = br->ports[port->port_idx] = br->ports[--br->n_ports]; |
3567 | del->port_idx = port->port_idx; | |
3568 | ||
99707a7a JP |
3569 | VLOG_INFO("destroyed port %s on bridge %s", port->name, br->name); |
3570 | ||
0bb5c3ec | 3571 | netdev_monitor_destroy(port->monitor); |
064af421 BP |
3572 | free(port->ifaces); |
3573 | bitmap_free(port->trunks); | |
3574 | free(port->name); | |
3575 | free(port); | |
3576 | bridge_flush(br); | |
3577 | } | |
3578 | } | |
3579 | ||
3580 | static struct port * | |
3581 | port_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
3582 | { | |
3583 | struct iface *iface = iface_from_dp_ifidx(br, dp_ifidx); | |
3584 | return iface ? iface->port : NULL; | |
3585 | } | |
3586 | ||
3587 | static struct port * | |
3588 | port_lookup(const struct bridge *br, const char *name) | |
3589 | { | |
836fad5e | 3590 | return shash_find_data(&br->port_by_name, name); |
064af421 BP |
3591 | } |
3592 | ||
da285df4 BP |
3593 | static struct iface * |
3594 | port_lookup_iface(const struct port *port, const char *name) | |
3595 | { | |
4a1ee6ae BP |
3596 | struct iface *iface = iface_lookup(port->bridge, name); |
3597 | return iface && iface->port == port ? iface : NULL; | |
da285df4 BP |
3598 | } |
3599 | ||
064af421 BP |
3600 | static void |
3601 | port_update_bonding(struct port *port) | |
3602 | { | |
0bb5c3ec BP |
3603 | if (port->monitor) { |
3604 | netdev_monitor_destroy(port->monitor); | |
3605 | port->monitor = NULL; | |
3606 | } | |
064af421 BP |
3607 | if (port->n_ifaces < 2) { |
3608 | /* Not a bonded port. */ | |
3609 | if (port->bond_hash) { | |
3610 | free(port->bond_hash); | |
3611 | port->bond_hash = NULL; | |
85c74638 | 3612 | port->bond_compat_is_stale = true; |
76343538 | 3613 | port->bond_fake_iface = false; |
064af421 BP |
3614 | } |
3615 | } else { | |
0bb5c3ec | 3616 | size_t i; |
064af421 | 3617 | |
0bb5c3ec | 3618 | if (!port->bond_hash) { |
064af421 BP |
3619 | port->bond_hash = xcalloc(BOND_MASK + 1, sizeof *port->bond_hash); |
3620 | for (i = 0; i <= BOND_MASK; i++) { | |
3621 | struct bond_entry *e = &port->bond_hash[i]; | |
3622 | e->iface_idx = -1; | |
3623 | e->tx_bytes = 0; | |
3624 | } | |
3625 | port->no_ifaces_tag = tag_create_random(); | |
3626 | bond_choose_active_iface(port); | |
c8143c88 BP |
3627 | port->bond_next_rebalance |
3628 | = time_msec() + port->bond_rebalance_interval; | |
8722022c BP |
3629 | |
3630 | if (port->cfg->bond_fake_iface) { | |
3631 | port->bond_next_fake_iface_update = time_msec(); | |
3632 | } | |
064af421 | 3633 | } |
85c74638 | 3634 | port->bond_compat_is_stale = true; |
76343538 | 3635 | port->bond_fake_iface = port->cfg->bond_fake_iface; |
0bb5c3ec BP |
3636 | |
3637 | port->monitor = netdev_monitor_create(); | |
3638 | for (i = 0; i < port->n_ifaces; i++) { | |
3639 | netdev_monitor_add(port->monitor, port->ifaces[i]->netdev); | |
3640 | } | |
064af421 BP |
3641 | } |
3642 | } | |
3643 | ||
3644 | static void | |
3645 | port_update_bond_compat(struct port *port) | |
3646 | { | |
2aebae83 | 3647 | struct compat_bond_hash compat_hashes[BOND_MASK + 1]; |
064af421 BP |
3648 | struct compat_bond bond; |
3649 | size_t i; | |
3650 | ||
3651 | if (port->n_ifaces < 2) { | |
85c74638 | 3652 | proc_net_compat_update_bond(port->name, NULL); |
064af421 BP |
3653 | return; |
3654 | } | |
3655 | ||
3656 | bond.up = false; | |
3657 | bond.updelay = port->updelay; | |
3658 | bond.downdelay = port->downdelay; | |
2aebae83 BP |
3659 | |
3660 | bond.n_hashes = 0; | |
3661 | bond.hashes = compat_hashes; | |
3662 | if (port->bond_hash) { | |
3663 | const struct bond_entry *e; | |
3664 | for (e = port->bond_hash; e <= &port->bond_hash[BOND_MASK]; e++) { | |
3665 | if (e->iface_idx >= 0 && e->iface_idx < port->n_ifaces) { | |
3666 | struct compat_bond_hash *cbh = &bond.hashes[bond.n_hashes++]; | |
3667 | cbh->hash = e - port->bond_hash; | |
3668 | cbh->netdev_name = port->ifaces[e->iface_idx]->name; | |
3669 | } | |
3670 | } | |
3671 | } | |
3672 | ||
064af421 BP |
3673 | bond.n_slaves = port->n_ifaces; |
3674 | bond.slaves = xmalloc(port->n_ifaces * sizeof *bond.slaves); | |
3675 | for (i = 0; i < port->n_ifaces; i++) { | |
3676 | struct iface *iface = port->ifaces[i]; | |
3677 | struct compat_bond_slave *slave = &bond.slaves[i]; | |
3678 | slave->name = iface->name; | |
6643c03b JP |
3679 | |
3680 | /* We need to make the same determination as the Linux bonding | |
3681 | * code to determine whether a slave should be consider "up". | |
d295e8e9 | 3682 | * The Linux function bond_miimon_inspect() supports four |
6643c03b | 3683 | * BOND_LINK_* states: |
d295e8e9 | 3684 | * |
6643c03b JP |
3685 | * - BOND_LINK_UP: carrier detected, updelay has passed. |
3686 | * - BOND_LINK_FAIL: carrier lost, downdelay in progress. | |
3687 | * - BOND_LINK_DOWN: carrier lost, downdelay has passed. | |
3688 | * - BOND_LINK_BACK: carrier detected, updelay in progress. | |
3689 | * | |
d295e8e9 | 3690 | * The function bond_info_show_slave() only considers BOND_LINK_UP |
6643c03b JP |
3691 | * to be "up" and anything else to be "down". |
3692 | */ | |
3693 | slave->up = iface->enabled && iface->delay_expires == LLONG_MAX; | |
064af421 BP |
3694 | if (slave->up) { |
3695 | bond.up = true; | |
3696 | } | |
96b9b7a9 | 3697 | netdev_get_etheraddr(iface->netdev, slave->mac); |
064af421 | 3698 | } |
2aebae83 | 3699 | |
76343538 | 3700 | if (port->bond_fake_iface) { |
c6303ea1 JG |
3701 | struct netdev *bond_netdev; |
3702 | ||
149f577a | 3703 | if (!netdev_open_default(port->name, &bond_netdev)) { |
c6303ea1 JG |
3704 | if (bond.up) { |
3705 | netdev_turn_flags_on(bond_netdev, NETDEV_UP, true); | |
3706 | } else { | |
3707 | netdev_turn_flags_off(bond_netdev, NETDEV_UP, true); | |
3708 | } | |
3709 | netdev_close(bond_netdev); | |
3710 | } | |
3711 | } | |
3712 | ||
064af421 BP |
3713 | proc_net_compat_update_bond(port->name, &bond); |
3714 | free(bond.slaves); | |
3715 | } | |
3716 | ||
3717 | static void | |
3718 | port_update_vlan_compat(struct port *port) | |
3719 | { | |
3720 | struct bridge *br = port->bridge; | |
3721 | char *vlandev_name = NULL; | |
3722 | ||
3723 | if (port->vlan > 0) { | |
3724 | /* Figure out the name that the VLAN device should actually have, if it | |
3725 | * existed. This takes some work because the VLAN device would not | |
3726 | * have port->name in its name; rather, it would have the trunk port's | |
3727 | * name, and 'port' would be attached to a bridge that also had the | |
3728 | * VLAN device one of its ports. So we need to find a trunk port that | |
3729 | * includes port->vlan. | |
3730 | * | |
3731 | * There might be more than one candidate. This doesn't happen on | |
3732 | * XenServer, so if it happens we just pick the first choice in | |
3733 | * alphabetical order instead of creating multiple VLAN devices. */ | |
3734 | size_t i; | |
3735 | for (i = 0; i < br->n_ports; i++) { | |
3736 | struct port *p = br->ports[i]; | |
3737 | if (port_trunks_vlan(p, port->vlan) | |
3738 | && p->n_ifaces | |
3739 | && (!vlandev_name || strcmp(p->name, vlandev_name) <= 0)) | |
3740 | { | |
96b9b7a9 BP |
3741 | uint8_t ea[ETH_ADDR_LEN]; |
3742 | netdev_get_etheraddr(p->ifaces[0]->netdev, ea); | |
064af421 BP |
3743 | if (!eth_addr_is_multicast(ea) && |
3744 | !eth_addr_is_reserved(ea) && | |
3745 | !eth_addr_is_zero(ea)) { | |
3746 | vlandev_name = p->name; | |
3747 | } | |
3748 | } | |
3749 | } | |
3750 | } | |
3751 | proc_net_compat_update_vlan(port->name, vlandev_name, port->vlan); | |
3752 | } | |
3753 | \f | |
3754 | /* Interface functions. */ | |
3755 | ||
76343538 | 3756 | static struct iface * |
a740f0de | 3757 | iface_create(struct port *port, const struct ovsrec_interface *if_cfg) |
064af421 | 3758 | { |
4a1ee6ae | 3759 | struct bridge *br = port->bridge; |
064af421 | 3760 | struct iface *iface; |
a740f0de | 3761 | char *name = if_cfg->name; |
064af421 | 3762 | |
ec6fde61 | 3763 | iface = xzalloc(sizeof *iface); |
064af421 BP |
3764 | iface->port = port; |
3765 | iface->port_ifidx = port->n_ifaces; | |
3766 | iface->name = xstrdup(name); | |
3767 | iface->dp_ifidx = -1; | |
3768 | iface->tag = tag_create_random(); | |
064af421 | 3769 | iface->delay_expires = LLONG_MAX; |
0c6aea3f | 3770 | iface->netdev = NULL; |
149f577a | 3771 | iface->cfg = if_cfg; |
064af421 | 3772 | |
2457b24f JG |
3773 | shash_add_assert(&br->iface_by_name, iface->name, iface); |
3774 | ||
064af421 BP |
3775 | if (port->n_ifaces >= port->allocated_ifaces) { |
3776 | port->ifaces = x2nrealloc(port->ifaces, &port->allocated_ifaces, | |
3777 | sizeof *port->ifaces); | |
3778 | } | |
3779 | port->ifaces[port->n_ifaces++] = iface; | |
3780 | if (port->n_ifaces > 1) { | |
4a1ee6ae | 3781 | br->has_bonded_ports = true; |
064af421 BP |
3782 | } |
3783 | ||
3784 | VLOG_DBG("attached network device %s to port %s", iface->name, port->name); | |
3785 | ||
4a1ee6ae | 3786 | bridge_flush(br); |
76343538 BP |
3787 | |
3788 | return iface; | |
064af421 BP |
3789 | } |
3790 | ||
3791 | static void | |
3792 | iface_destroy(struct iface *iface) | |
3793 | { | |
3794 | if (iface) { | |
3795 | struct port *port = iface->port; | |
3796 | struct bridge *br = port->bridge; | |
3797 | bool del_active = port->active_iface == iface->port_ifidx; | |
3798 | struct iface *del; | |
3799 | ||
4a1ee6ae BP |
3800 | shash_find_and_delete_assert(&br->iface_by_name, iface->name); |
3801 | ||
064af421 | 3802 | if (iface->dp_ifidx >= 0) { |
d9a8717a | 3803 | hmap_remove(&br->ifaces, &iface->dp_ifidx_node); |
064af421 BP |
3804 | } |
3805 | ||
3806 | del = port->ifaces[iface->port_ifidx] = port->ifaces[--port->n_ifaces]; | |
3807 | del->port_ifidx = iface->port_ifidx; | |
3808 | ||
0c6aea3f | 3809 | netdev_close(iface->netdev); |
064af421 BP |
3810 | |
3811 | if (del_active) { | |
3812 | ofproto_revalidate(port->bridge->ofproto, port->active_iface_tag); | |
3813 | bond_choose_active_iface(port); | |
2303f3b2 | 3814 | bond_send_learning_packets(port); |
064af421 BP |
3815 | } |
3816 | ||
a740f0de JG |
3817 | free(iface->name); |
3818 | free(iface); | |
3819 | ||
064af421 BP |
3820 | bridge_flush(port->bridge); |
3821 | } | |
3822 | } | |
3823 | ||
3824 | static struct iface * | |
3825 | iface_lookup(const struct bridge *br, const char *name) | |
3826 | { | |
4a1ee6ae | 3827 | return shash_find_data(&br->iface_by_name, name); |
064af421 BP |
3828 | } |
3829 | ||
3830 | static struct iface * | |
3831 | iface_from_dp_ifidx(const struct bridge *br, uint16_t dp_ifidx) | |
3832 | { | |
d9a8717a BP |
3833 | struct iface *iface; |
3834 | ||
4e8e4213 | 3835 | HMAP_FOR_EACH_IN_BUCKET (iface, dp_ifidx_node, |
d9a8717a BP |
3836 | hash_int(dp_ifidx, 0), &br->ifaces) { |
3837 | if (iface->dp_ifidx == dp_ifidx) { | |
3838 | return iface; | |
3839 | } | |
3840 | } | |
3841 | return NULL; | |
064af421 | 3842 | } |
557d8e6c | 3843 | |
52df17e7 BP |
3844 | /* Set Ethernet address of 'iface', if one is specified in the configuration |
3845 | * file. */ | |
3846 | static void | |
3847 | iface_set_mac(struct iface *iface) | |
3848 | { | |
76343538 | 3849 | uint8_t ea[ETH_ADDR_LEN]; |
52df17e7 | 3850 | |
76343538 | 3851 | if (iface->cfg->mac && eth_addr_from_string(iface->cfg->mac, ea)) { |
52df17e7 BP |
3852 | if (eth_addr_is_multicast(ea)) { |
3853 | VLOG_ERR("interface %s: cannot set MAC to multicast address", | |
3854 | iface->name); | |
3855 | } else if (iface->dp_ifidx == ODPP_LOCAL) { | |
3856 | VLOG_ERR("ignoring iface.%s.mac; use bridge.%s.mac instead", | |
3857 | iface->name, iface->name); | |
3858 | } else { | |
4d678233 | 3859 | int error = netdev_set_etheraddr(iface->netdev, ea); |
52df17e7 BP |
3860 | if (error) { |
3861 | VLOG_ERR("interface %s: setting MAC failed (%s)", | |
3862 | iface->name, strerror(error)); | |
3863 | } | |
3864 | } | |
3865 | } | |
3866 | } | |
c1c9c9c4 | 3867 | |
bcd49a45 BP |
3868 | /* Sets the ofport column of 'if_cfg' to 'ofport'. */ |
3869 | static void | |
3870 | iface_set_ofport(const struct ovsrec_interface *if_cfg, int64_t ofport) | |
3871 | { | |
3872 | if (if_cfg) { | |
3873 | ovsrec_interface_set_ofport(if_cfg, &ofport, 1); | |
3874 | } | |
3875 | } | |
3876 | ||
43776b8f BP |
3877 | /* Adds the 'n' key-value pairs in 'keys' in 'values' to 'shash'. |
3878 | * | |
3879 | * The value strings in '*shash' are taken directly from values[], not copied, | |
3880 | * so the caller should not modify or free them. */ | |
c1c9c9c4 BP |
3881 | static void |
3882 | shash_from_ovs_idl_map(char **keys, char **values, size_t n, | |
3883 | struct shash *shash) | |
3884 | { | |
3885 | size_t i; | |
3886 | ||
3887 | shash_init(shash); | |
3888 | for (i = 0; i < n; i++) { | |
3889 | shash_add(shash, keys[i], values[i]); | |
3890 | } | |
3891 | } | |
3892 | ||
3893 | struct iface_delete_queues_cbdata { | |
3894 | struct netdev *netdev; | |
44fca7f9 | 3895 | const struct ovsdb_datum *queues; |
c1c9c9c4 BP |
3896 | }; |
3897 | ||
3898 | static bool | |
44fca7f9 | 3899 | queue_ids_include(const struct ovsdb_datum *queues, int64_t target) |
c1c9c9c4 | 3900 | { |
44fca7f9 | 3901 | union ovsdb_atom atom; |
c1c9c9c4 | 3902 | |
44fca7f9 BP |
3903 | atom.integer = target; |
3904 | return ovsdb_datum_find_key(queues, &atom, OVSDB_TYPE_INTEGER) != UINT_MAX; | |
c1c9c9c4 BP |
3905 | } |
3906 | ||
3907 | static void | |
3908 | iface_delete_queues(unsigned int queue_id, | |
3909 | const struct shash *details OVS_UNUSED, void *cbdata_) | |
3910 | { | |
3911 | struct iface_delete_queues_cbdata *cbdata = cbdata_; | |
3912 | ||
44fca7f9 | 3913 | if (!queue_ids_include(cbdata->queues, queue_id)) { |
c1c9c9c4 BP |
3914 | netdev_delete_queue(cbdata->netdev, queue_id); |
3915 | } | |
3916 | } | |
3917 | ||
3918 | static void | |
3919 | iface_update_qos(struct iface *iface, const struct ovsrec_qos *qos) | |
3920 | { | |
3921 | if (!qos || qos->type[0] == '\0') { | |
3922 | netdev_set_qos(iface->netdev, NULL, NULL); | |
3923 | } else { | |
3924 | struct iface_delete_queues_cbdata cbdata; | |
3925 | struct shash details; | |
3926 | size_t i; | |
3927 | ||
3928 | /* Configure top-level Qos for 'iface'. */ | |
3929 | shash_from_ovs_idl_map(qos->key_other_config, qos->value_other_config, | |
3930 | qos->n_other_config, &details); | |
3931 | netdev_set_qos(iface->netdev, qos->type, &details); | |
3932 | shash_destroy(&details); | |
3933 | ||
3934 | /* Deconfigure queues that were deleted. */ | |
3935 | cbdata.netdev = iface->netdev; | |
44fca7f9 BP |
3936 | cbdata.queues = ovsrec_qos_get_queues(qos, OVSDB_TYPE_INTEGER, |
3937 | OVSDB_TYPE_UUID); | |
c1c9c9c4 BP |
3938 | netdev_dump_queues(iface->netdev, iface_delete_queues, &cbdata); |
3939 | ||
3940 | /* Configure queues for 'iface'. */ | |
3941 | for (i = 0; i < qos->n_queues; i++) { | |
3942 | const struct ovsrec_queue *queue = qos->value_queues[i]; | |
3943 | unsigned int queue_id = qos->key_queues[i]; | |
3944 | ||
3945 | shash_from_ovs_idl_map(queue->key_other_config, | |
3946 | queue->value_other_config, | |
3947 | queue->n_other_config, &details); | |
3948 | netdev_set_queue(iface->netdev, queue_id, &details); | |
3949 | shash_destroy(&details); | |
3950 | } | |
3951 | } | |
3952 | } | |
064af421 BP |
3953 | \f |
3954 | /* Port mirroring. */ | |
3955 | ||
dd0d105c BP |
3956 | static struct mirror * |
3957 | mirror_find_by_uuid(struct bridge *br, const struct uuid *uuid) | |
3958 | { | |
3959 | int i; | |
3960 | ||
3961 | for (i = 0; i < MAX_MIRRORS; i++) { | |
3962 | struct mirror *m = br->mirrors[i]; | |
3963 | if (m && uuid_equals(uuid, &m->uuid)) { | |
3964 | return m; | |
3965 | } | |
3966 | } | |
3967 | return NULL; | |
3968 | } | |
3969 | ||
064af421 | 3970 | static void |
37e7f427 | 3971 | mirror_reconfigure(struct bridge *br) |
064af421 | 3972 | { |
f2d7fd66 | 3973 | unsigned long *rspan_vlans; |
37e7f427 | 3974 | int i; |
064af421 | 3975 | |
dd0d105c | 3976 | /* Get rid of deleted mirrors. */ |
064af421 | 3977 | for (i = 0; i < MAX_MIRRORS; i++) { |
dd0d105c BP |
3978 | struct mirror *m = br->mirrors[i]; |
3979 | if (m) { | |
3980 | const struct ovsdb_datum *mc; | |
3981 | union ovsdb_atom atom; | |
3982 | ||
3983 | mc = ovsrec_bridge_get_mirrors(br->cfg, OVSDB_TYPE_UUID); | |
3984 | atom.uuid = br->mirrors[i]->uuid; | |
3985 | if (ovsdb_datum_find_key(mc, &atom, OVSDB_TYPE_UUID) == UINT_MAX) { | |
3986 | mirror_destroy(m); | |
3987 | } | |
064af421 BP |
3988 | } |
3989 | } | |
3990 | ||
dd0d105c | 3991 | /* Add new mirrors and reconfigure existing ones. */ |
37e7f427 BP |
3992 | for (i = 0; i < br->cfg->n_mirrors; i++) { |
3993 | struct ovsrec_mirror *cfg = br->cfg->mirrors[i]; | |
dd0d105c BP |
3994 | struct mirror *m = mirror_find_by_uuid(br, &cfg->header_.uuid); |
3995 | if (m) { | |
3996 | mirror_reconfigure_one(m, cfg); | |
3997 | } else { | |
3998 | mirror_create(br, cfg); | |
064af421 BP |
3999 | } |
4000 | } | |
4001 | ||
4002 | /* Update port reserved status. */ | |
4003 | for (i = 0; i < br->n_ports; i++) { | |
4004 | br->ports[i]->is_mirror_output_port = false; | |
4005 | } | |
4006 | for (i = 0; i < MAX_MIRRORS; i++) { | |
4007 | struct mirror *m = br->mirrors[i]; | |
4008 | if (m && m->out_port) { | |
4009 | m->out_port->is_mirror_output_port = true; | |
4010 | } | |
4011 | } | |
f2d7fd66 | 4012 | |
8f30d09a | 4013 | /* Update flooded vlans (for RSPAN). */ |
f2d7fd66 | 4014 | rspan_vlans = NULL; |
37e7f427 | 4015 | if (br->cfg->n_flood_vlans) { |
f2d7fd66 JG |
4016 | rspan_vlans = bitmap_allocate(4096); |
4017 | ||
37e7f427 BP |
4018 | for (i = 0; i < br->cfg->n_flood_vlans; i++) { |
4019 | int64_t vlan = br->cfg->flood_vlans[i]; | |
4020 | if (vlan >= 0 && vlan < 4096) { | |
f2d7fd66 | 4021 | bitmap_set1(rspan_vlans, vlan); |
37e7f427 | 4022 | VLOG_INFO("bridge %s: disabling learning on vlan %"PRId64, |
42061b2a | 4023 | br->name, vlan); |
f2d7fd66 | 4024 | } else { |
37e7f427 BP |
4025 | VLOG_ERR("bridge %s: invalid value %"PRId64 "for flood VLAN", |
4026 | br->name, vlan); | |
f2d7fd66 JG |
4027 | } |
4028 | } | |
4029 | } | |
8f30d09a | 4030 | if (mac_learning_set_flood_vlans(br->ml, rspan_vlans)) { |
f2d7fd66 JG |
4031 | bridge_flush(br); |
4032 | } | |
064af421 BP |
4033 | } |
4034 | ||
dd0d105c BP |
4035 | static void |
4036 | mirror_create(struct bridge *br, struct ovsrec_mirror *cfg) | |
064af421 BP |
4037 | { |
4038 | struct mirror *m; | |
4039 | size_t i; | |
4040 | ||
4041 | for (i = 0; ; i++) { | |
4042 | if (i >= MAX_MIRRORS) { | |
4043 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
dd0d105c BP |
4044 | "cannot create %s", br->name, MAX_MIRRORS, cfg->name); |
4045 | return; | |
064af421 BP |
4046 | } |
4047 | if (!br->mirrors[i]) { | |
4048 | break; | |
4049 | } | |
4050 | } | |
4051 | ||
dd0d105c | 4052 | VLOG_INFO("created port mirror %s on bridge %s", cfg->name, br->name); |
064af421 BP |
4053 | bridge_flush(br); |
4054 | ||
ec6fde61 | 4055 | br->mirrors[i] = m = xzalloc(sizeof *m); |
064af421 BP |
4056 | m->bridge = br; |
4057 | m->idx = i; | |
dd0d105c | 4058 | m->name = xstrdup(cfg->name); |
37e7f427 BP |
4059 | shash_init(&m->src_ports); |
4060 | shash_init(&m->dst_ports); | |
064af421 BP |
4061 | m->vlans = NULL; |
4062 | m->n_vlans = 0; | |
4063 | m->out_vlan = -1; | |
4064 | m->out_port = NULL; | |
37e7f427 | 4065 | |
dd0d105c | 4066 | mirror_reconfigure_one(m, cfg); |
064af421 BP |
4067 | } |
4068 | ||
4069 | static void | |
4070 | mirror_destroy(struct mirror *m) | |
4071 | { | |
4072 | if (m) { | |
4073 | struct bridge *br = m->bridge; | |
4074 | size_t i; | |
4075 | ||
4076 | for (i = 0; i < br->n_ports; i++) { | |
4077 | br->ports[i]->src_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
4078 | br->ports[i]->dst_mirrors &= ~(MIRROR_MASK_C(1) << m->idx); | |
4079 | } | |
4080 | ||
37e7f427 BP |
4081 | shash_destroy(&m->src_ports); |
4082 | shash_destroy(&m->dst_ports); | |
064af421 BP |
4083 | free(m->vlans); |
4084 | ||
4085 | m->bridge->mirrors[m->idx] = NULL; | |
786880a5 | 4086 | free(m->name); |
064af421 BP |
4087 | free(m); |
4088 | ||
4089 | bridge_flush(br); | |
4090 | } | |
4091 | } | |
4092 | ||
4093 | static void | |
37e7f427 BP |
4094 | mirror_collect_ports(struct mirror *m, struct ovsrec_port **ports, int n_ports, |
4095 | struct shash *names) | |
064af421 | 4096 | { |
064af421 BP |
4097 | size_t i; |
4098 | ||
37e7f427 BP |
4099 | for (i = 0; i < n_ports; i++) { |
4100 | const char *name = ports[i]->name; | |
064af421 | 4101 | if (port_lookup(m->bridge, name)) { |
37e7f427 | 4102 | shash_add_once(names, name, NULL); |
064af421 | 4103 | } else { |
37e7f427 BP |
4104 | VLOG_WARN("bridge %s: mirror %s cannot match on nonexistent " |
4105 | "port %s", m->bridge->name, m->name, name); | |
064af421 BP |
4106 | } |
4107 | } | |
064af421 BP |
4108 | } |
4109 | ||
4110 | static size_t | |
37e7f427 BP |
4111 | mirror_collect_vlans(struct mirror *m, const struct ovsrec_mirror *cfg, |
4112 | int **vlans) | |
064af421 | 4113 | { |
37e7f427 BP |
4114 | size_t n_vlans; |
4115 | size_t i; | |
064af421 | 4116 | |
0b523ea7 | 4117 | *vlans = xmalloc(sizeof **vlans * cfg->n_select_vlan); |
064af421 | 4118 | n_vlans = 0; |
37e7f427 BP |
4119 | for (i = 0; i < cfg->n_select_vlan; i++) { |
4120 | int64_t vlan = cfg->select_vlan[i]; | |
4121 | if (vlan < 0 || vlan > 4095) { | |
4122 | VLOG_WARN("bridge %s: mirror %s selects invalid VLAN %"PRId64, | |
4123 | m->bridge->name, m->name, vlan); | |
064af421 BP |
4124 | } else { |
4125 | (*vlans)[n_vlans++] = vlan; | |
4126 | } | |
4127 | } | |
4128 | return n_vlans; | |
4129 | } | |
4130 | ||
4131 | static bool | |
4132 | vlan_is_mirrored(const struct mirror *m, int vlan) | |
4133 | { | |
4134 | size_t i; | |
4135 | ||
4136 | for (i = 0; i < m->n_vlans; i++) { | |
4137 | if (m->vlans[i] == vlan) { | |
4138 | return true; | |
4139 | } | |
4140 | } | |
4141 | return false; | |
4142 | } | |
4143 | ||
4144 | static bool | |
4145 | port_trunks_any_mirrored_vlan(const struct mirror *m, const struct port *p) | |
4146 | { | |
4147 | size_t i; | |
4148 | ||
4149 | for (i = 0; i < m->n_vlans; i++) { | |
4150 | if (port_trunks_vlan(p, m->vlans[i])) { | |
4151 | return true; | |
4152 | } | |
4153 | } | |
4154 | return false; | |
4155 | } | |
4156 | ||
4157 | static void | |
37e7f427 | 4158 | mirror_reconfigure_one(struct mirror *m, struct ovsrec_mirror *cfg) |
064af421 | 4159 | { |
37e7f427 | 4160 | struct shash src_ports, dst_ports; |
064af421 | 4161 | mirror_mask_t mirror_bit; |
064af421 BP |
4162 | struct port *out_port; |
4163 | int out_vlan; | |
4164 | size_t n_vlans; | |
4165 | int *vlans; | |
4166 | size_t i; | |
064af421 | 4167 | |
dd0d105c BP |
4168 | /* Set name. */ |
4169 | if (strcmp(cfg->name, m->name)) { | |
4170 | free(m->name); | |
4171 | m->name = xstrdup(cfg->name); | |
4172 | } | |
4173 | ||
064af421 | 4174 | /* Get output port. */ |
37e7f427 BP |
4175 | if (cfg->output_port) { |
4176 | out_port = port_lookup(m->bridge, cfg->output_port->name); | |
064af421 | 4177 | if (!out_port) { |
37e7f427 BP |
4178 | VLOG_ERR("bridge %s: mirror %s outputs to port not on bridge", |
4179 | m->bridge->name, m->name); | |
064af421 | 4180 | mirror_destroy(m); |
064af421 BP |
4181 | return; |
4182 | } | |
4183 | out_vlan = -1; | |
4184 | ||
37e7f427 BP |
4185 | if (cfg->output_vlan) { |
4186 | VLOG_ERR("bridge %s: mirror %s specifies both output port and " | |
4187 | "output vlan; ignoring output vlan", | |
4188 | m->bridge->name, m->name); | |
064af421 | 4189 | } |
37e7f427 | 4190 | } else if (cfg->output_vlan) { |
064af421 | 4191 | out_port = NULL; |
37e7f427 | 4192 | out_vlan = *cfg->output_vlan; |
064af421 | 4193 | } else { |
37e7f427 BP |
4194 | VLOG_ERR("bridge %s: mirror %s does not specify output; ignoring", |
4195 | m->bridge->name, m->name); | |
064af421 | 4196 | mirror_destroy(m); |
064af421 BP |
4197 | return; |
4198 | } | |
4199 | ||
37e7f427 BP |
4200 | shash_init(&src_ports); |
4201 | shash_init(&dst_ports); | |
939ff267 BP |
4202 | if (cfg->select_all) { |
4203 | for (i = 0; i < m->bridge->n_ports; i++) { | |
4204 | const char *name = m->bridge->ports[i]->name; | |
4205 | shash_add_once(&src_ports, name, NULL); | |
4206 | shash_add_once(&dst_ports, name, NULL); | |
4207 | } | |
4208 | vlans = NULL; | |
4209 | n_vlans = 0; | |
4210 | } else { | |
4211 | /* Get ports, and drop duplicates and ports that don't exist. */ | |
4212 | mirror_collect_ports(m, cfg->select_src_port, cfg->n_select_src_port, | |
4213 | &src_ports); | |
4214 | mirror_collect_ports(m, cfg->select_dst_port, cfg->n_select_dst_port, | |
4215 | &dst_ports); | |
064af421 | 4216 | |
939ff267 BP |
4217 | /* Get all the vlans, and drop duplicate and invalid vlans. */ |
4218 | n_vlans = mirror_collect_vlans(m, cfg, &vlans); | |
37e7f427 | 4219 | } |
064af421 BP |
4220 | |
4221 | /* Update mirror data. */ | |
37e7f427 BP |
4222 | if (!shash_equal_keys(&m->src_ports, &src_ports) |
4223 | || !shash_equal_keys(&m->dst_ports, &dst_ports) | |
064af421 BP |
4224 | || m->n_vlans != n_vlans |
4225 | || memcmp(m->vlans, vlans, sizeof *vlans * n_vlans) | |
4226 | || m->out_port != out_port | |
4227 | || m->out_vlan != out_vlan) { | |
4228 | bridge_flush(m->bridge); | |
4229 | } | |
37e7f427 BP |
4230 | shash_swap(&m->src_ports, &src_ports); |
4231 | shash_swap(&m->dst_ports, &dst_ports); | |
064af421 BP |
4232 | free(m->vlans); |
4233 | m->vlans = vlans; | |
4234 | m->n_vlans = n_vlans; | |
4235 | m->out_port = out_port; | |
4236 | m->out_vlan = out_vlan; | |
4237 | ||
064af421 BP |
4238 | /* Update ports. */ |
4239 | mirror_bit = MIRROR_MASK_C(1) << m->idx; | |
4240 | for (i = 0; i < m->bridge->n_ports; i++) { | |
4241 | struct port *port = m->bridge->ports[i]; | |
4242 | ||
939ff267 | 4243 | if (shash_find(&m->src_ports, port->name) |
064af421 BP |
4244 | || (m->n_vlans |
4245 | && (!port->vlan | |
4246 | ? port_trunks_any_mirrored_vlan(m, port) | |
4247 | : vlan_is_mirrored(m, port->vlan)))) { | |
4248 | port->src_mirrors |= mirror_bit; | |
4249 | } else { | |
4250 | port->src_mirrors &= ~mirror_bit; | |
4251 | } | |
4252 | ||
939ff267 | 4253 | if (shash_find(&m->dst_ports, port->name)) { |
064af421 BP |
4254 | port->dst_mirrors |= mirror_bit; |
4255 | } else { | |
4256 | port->dst_mirrors &= ~mirror_bit; | |
4257 | } | |
4258 | } | |
4259 | ||
4260 | /* Clean up. */ | |
37e7f427 BP |
4261 | shash_destroy(&src_ports); |
4262 | shash_destroy(&dst_ports); | |
064af421 | 4263 | } |