]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
149f577a | 2 | * Copyright (c) 2009, 2010 Nicira Networks. |
e9e28be3 BP |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
8b61709d | 18 | #include <assert.h> |
e9e28be3 | 19 | #include <errno.h> |
8b61709d BP |
20 | #include <fcntl.h> |
21 | #include <arpa/inet.h> | |
22 | #include <inttypes.h> | |
c1c9c9c4 | 23 | #include <linux/gen_stats.h> |
8b61709d | 24 | #include <linux/if_tun.h> |
a740f0de | 25 | #include <linux/ip.h> |
8b61709d BP |
26 | #include <linux/types.h> |
27 | #include <linux/ethtool.h> | |
6f42c8ea | 28 | #include <linux/pkt_sched.h> |
e9e28be3 | 29 | #include <linux/rtnetlink.h> |
8b61709d BP |
30 | #include <linux/sockios.h> |
31 | #include <linux/version.h> | |
32 | #include <sys/types.h> | |
33 | #include <sys/ioctl.h> | |
34 | #include <sys/socket.h> | |
35 | #include <netpacket/packet.h> | |
36 | #include <net/ethernet.h> | |
37 | #include <net/if.h> | |
a740f0de | 38 | #include <linux/if_tunnel.h> |
8b61709d BP |
39 | #include <net/if_arp.h> |
40 | #include <net/if_packet.h> | |
41 | #include <net/route.h> | |
42 | #include <netinet/in.h> | |
e9e28be3 | 43 | #include <poll.h> |
8b61709d BP |
44 | #include <stdlib.h> |
45 | #include <string.h> | |
46 | #include <unistd.h> | |
e9e28be3 BP |
47 | |
48 | #include "coverage.h" | |
8b61709d BP |
49 | #include "dynamic-string.h" |
50 | #include "fatal-signal.h" | |
93b13be8 BP |
51 | #include "hash.h" |
52 | #include "hmap.h" | |
8b61709d | 53 | #include "netdev-provider.h" |
7fbef77a | 54 | #include "netdev-vport.h" |
e9e28be3 BP |
55 | #include "netlink.h" |
56 | #include "ofpbuf.h" | |
8b61709d BP |
57 | #include "openflow/openflow.h" |
58 | #include "packets.h" | |
59 | #include "poll-loop.h" | |
559843ed | 60 | #include "rtnetlink.h" |
8b61709d BP |
61 | #include "socket-util.h" |
62 | #include "shash.h" | |
63 | #include "svec.h" | |
e9e28be3 | 64 | #include "vlog.h" |
5136ce49 BP |
65 | |
66 | VLOG_DEFINE_THIS_MODULE(netdev_linux) | |
8b61709d BP |
67 | \f |
68 | /* These were introduced in Linux 2.6.14, so they might be missing if we have | |
69 | * old headers. */ | |
70 | #ifndef ADVERTISED_Pause | |
71 | #define ADVERTISED_Pause (1 << 13) | |
72 | #endif | |
73 | #ifndef ADVERTISED_Asym_Pause | |
74 | #define ADVERTISED_Asym_Pause (1 << 14) | |
75 | #endif | |
76 | ||
c1c9c9c4 BP |
77 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
78 | * headers. */ | |
79 | #ifndef TC_RTAB_SIZE | |
80 | #define TC_RTAB_SIZE 1024 | |
81 | #endif | |
82 | ||
149f577a | 83 | static struct rtnetlink_notifier netdev_linux_cache_notifier; |
46415c90 | 84 | static int cache_notifier_refcount; |
8b61709d BP |
85 | |
86 | enum { | |
7fbef77a JG |
87 | VALID_IFINDEX = 1 << 0, |
88 | VALID_ETHERADDR = 1 << 1, | |
89 | VALID_IN4 = 1 << 2, | |
90 | VALID_IN6 = 1 << 3, | |
91 | VALID_MTU = 1 << 4, | |
92 | VALID_CARRIER = 1 << 5, | |
93 | VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */ | |
94 | VALID_POLICING = 1 << 7, | |
95 | VALID_HAVE_VPORT_STATS = 1 << 8 | |
8b61709d BP |
96 | }; |
97 | ||
149f577a JG |
98 | struct tap_state { |
99 | int fd; | |
61b999dd | 100 | bool opened; |
149f577a | 101 | }; |
c1c9c9c4 BP |
102 | \f |
103 | /* Traffic control. */ | |
104 | ||
105 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
106 | * network device. |
107 | * | |
108 | * Each TC implementation subclasses this with whatever additional data it | |
109 | * needs. */ | |
c1c9c9c4 BP |
110 | struct tc { |
111 | const struct tc_ops *ops; | |
93b13be8 BP |
112 | struct hmap queues; /* Contains "struct tc_queue"s. |
113 | * Read by generic TC layer. | |
114 | * Written only by TC implementation. */ | |
115 | }; | |
c1c9c9c4 | 116 | |
93b13be8 BP |
117 | /* One traffic control queue. |
118 | * | |
119 | * Each TC implementation subclasses this with whatever additional data it | |
120 | * needs. */ | |
121 | struct tc_queue { | |
122 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
123 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
c1c9c9c4 BP |
124 | }; |
125 | ||
126 | /* A particular kind of traffic control. Each implementation generally maps to | |
127 | * one particular Linux qdisc class. | |
128 | * | |
129 | * The functions below return 0 if successful or a positive errno value on | |
130 | * failure, except where otherwise noted. All of them must be provided, except | |
131 | * where otherwise noted. */ | |
132 | struct tc_ops { | |
133 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
134 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
135 | * appropriate values. */ | |
136 | const char *linux_name; | |
137 | ||
138 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
139 | const char *ovs_name; | |
140 | ||
141 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
142 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
143 | unsigned int n_queues; | |
144 | ||
145 | /* Called to install this TC class on 'netdev'. The implementation should | |
146 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
147 | * and configure it according to 'details'. The implementation may assume | |
148 | * that the current qdisc is the default; that is, there is no need for it | |
149 | * to delete the current qdisc before installing itself. | |
150 | * | |
151 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
152 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
153 | * (which is built as ovs-vswitchd.conf.db(8)). | |
154 | * | |
155 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
156 | * initialized 'struct tc'. | |
157 | * | |
158 | * (This function is null for tc_ops_other, which cannot be installed. For | |
159 | * other TC classes it should always be nonnull.) */ | |
160 | int (*tc_install)(struct netdev *netdev, const struct shash *details); | |
161 | ||
162 | /* Called when the netdev code determines (through a Netlink query) that | |
163 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
164 | * it ourselves and so don't know any of the details. | |
165 | * | |
166 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
167 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
168 | * implementation should parse the other attributes of 'nlmsg' as | |
169 | * necessary to determine its configuration. If necessary it should also | |
170 | * use Netlink queries to determine the configuration of queues on | |
171 | * 'netdev'. | |
172 | * | |
173 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
174 | * initialized 'struct tc'. */ | |
175 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
176 | ||
177 | /* Destroys the data structures allocated by the implementation as part of | |
178 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
179 | * tc_destroy(tc). | |
180 | * | |
181 | * The implementation should not need to perform any Netlink calls. If | |
182 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
183 | * (But it may not be desirable.) | |
184 | * | |
185 | * This function may be null if 'tc' is trivial. */ | |
186 | void (*tc_destroy)(struct tc *tc); | |
187 | ||
188 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
189 | * | |
190 | * The implementation should not need to perform any Netlink calls, because | |
191 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
192 | * cached the configuration. | |
193 | * | |
194 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
195 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
196 | * (which is built as ovs-vswitchd.conf.db(8)). | |
197 | * | |
198 | * This function may be null if 'tc' is not configurable. | |
199 | */ | |
200 | int (*qdisc_get)(const struct netdev *netdev, struct shash *details); | |
201 | ||
202 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
203 | * required Netlink calls to complete the reconfiguration. | |
204 | * | |
205 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
206 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
207 | * (which is built as ovs-vswitchd.conf.db(8)). | |
208 | * | |
209 | * This function may be null if 'tc' is not configurable. | |
210 | */ | |
211 | int (*qdisc_set)(struct netdev *, const struct shash *details); | |
212 | ||
93b13be8 BP |
213 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
214 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
215 | * |
216 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
217 | * in the "other_config" column in the "Queue" table in | |
218 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
219 | * | |
220 | * The implementation should not need to perform any Netlink calls, because | |
221 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
222 | * cached the queue configuration. | |
223 | * | |
224 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
225 | * 0). */ | |
93b13be8 | 226 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
227 | struct shash *details); |
228 | ||
229 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
230 | * 'details', perfoming any required Netlink calls to complete the | |
231 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
232 | * 'n_queues'. | |
233 | * | |
234 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
235 | * in the "other_config" column in the "Queue" table in | |
236 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
237 | * | |
238 | * This function may be null if 'tc' does not have queues or its queues are | |
239 | * not configurable. */ | |
240 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
241 | const struct shash *details); | |
242 | ||
93b13be8 BP |
243 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
244 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
245 | * |
246 | * This function may be null if 'tc' does not have queues or its queues | |
247 | * cannot be deleted. */ | |
93b13be8 | 248 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 249 | |
93b13be8 BP |
250 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
251 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
252 | * |
253 | * On success, initializes '*stats'. | |
254 | * | |
255 | * This function may be null if 'tc' does not have queues or if it cannot | |
256 | * report queue statistics. */ | |
93b13be8 BP |
257 | int (*class_get_stats)(const struct netdev *netdev, |
258 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
259 | struct netdev_queue_stats *stats); |
260 | ||
261 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
262 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
263 | * | |
264 | * This function may be null if 'tc' does not have queues or if it cannot | |
265 | * report queue statistics. */ | |
266 | int (*class_dump_stats)(const struct netdev *netdev, | |
267 | const struct ofpbuf *nlmsg, | |
268 | netdev_dump_queue_stats_cb *cb, void *aux); | |
269 | }; | |
270 | ||
271 | static void | |
272 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
273 | { | |
274 | tc->ops = ops; | |
93b13be8 | 275 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
276 | } |
277 | ||
278 | static void | |
279 | tc_destroy(struct tc *tc) | |
280 | { | |
93b13be8 | 281 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
282 | } |
283 | ||
284 | static const struct tc_ops tc_ops_htb; | |
285 | static const struct tc_ops tc_ops_default; | |
286 | static const struct tc_ops tc_ops_other; | |
287 | ||
288 | static const struct tc_ops *tcs[] = { | |
289 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ | |
290 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ | |
291 | &tc_ops_other, /* Some other qdisc. */ | |
292 | NULL | |
293 | }; | |
149f577a | 294 | |
c1c9c9c4 BP |
295 | static unsigned int tc_make_handle(unsigned int major, unsigned int minor); |
296 | static unsigned int tc_get_major(unsigned int handle); | |
297 | static unsigned int tc_get_minor(unsigned int handle); | |
298 | ||
299 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); | |
300 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
301 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
302 | ||
303 | static struct tcmsg *tc_make_request(const struct netdev *, int type, | |
304 | unsigned int flags, struct ofpbuf *); | |
305 | static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); | |
306 | ||
307 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
308 | struct nlattr **options); | |
309 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
310 | struct nlattr **options, | |
311 | struct netdev_queue_stats *); | |
312 | static int tc_query_class(const struct netdev *, | |
313 | unsigned int handle, unsigned int parent, | |
314 | struct ofpbuf **replyp); | |
315 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
316 | ||
317 | static int tc_del_qdisc(struct netdev *netdev); | |
318 | static int tc_query_qdisc(const struct netdev *netdev); | |
319 | ||
320 | static int tc_calc_cell_log(unsigned int mtu); | |
321 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
322 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
323 | const struct tc_ratespec *rate); | |
324 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
325 | \f | |
149f577a JG |
326 | struct netdev_dev_linux { |
327 | struct netdev_dev netdev_dev; | |
328 | ||
8b61709d | 329 | struct shash_node *shash_node; |
149f577a | 330 | unsigned int cache_valid; |
8b61709d | 331 | |
8722022c BP |
332 | /* The following are figured out "on demand" only. They are only valid |
333 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d BP |
334 | int ifindex; |
335 | uint8_t etheraddr[ETH_ADDR_LEN]; | |
f1acd62b | 336 | struct in_addr address, netmask; |
8b61709d BP |
337 | struct in6_addr in6; |
338 | int mtu; | |
339 | int carrier; | |
8722022c BP |
340 | bool is_internal; /* Is this an openvswitch internal device? */ |
341 | bool is_tap; /* Is this a tuntap device? */ | |
80a86fbe BP |
342 | uint32_t kbits_rate; /* Policing data. */ |
343 | uint32_t kbits_burst; | |
7fbef77a | 344 | bool have_vport_stats; |
c1c9c9c4 | 345 | struct tc *tc; |
149f577a JG |
346 | |
347 | union { | |
348 | struct tap_state tap; | |
349 | } state; | |
8b61709d BP |
350 | }; |
351 | ||
149f577a JG |
352 | struct netdev_linux { |
353 | struct netdev netdev; | |
5b7448ed | 354 | int fd; |
149f577a | 355 | }; |
8b61709d | 356 | |
8b61709d BP |
357 | /* An AF_INET socket (used for ioctl operations). */ |
358 | static int af_inet_sock = -1; | |
359 | ||
ff4ed3c9 BP |
360 | /* A Netlink routing socket that is not subscribed to any multicast groups. */ |
361 | static struct nl_sock *rtnl_sock; | |
362 | ||
8b61709d BP |
363 | struct netdev_linux_notifier { |
364 | struct netdev_notifier notifier; | |
365 | struct list node; | |
366 | }; | |
367 | ||
368 | static struct shash netdev_linux_notifiers = | |
369 | SHASH_INITIALIZER(&netdev_linux_notifiers); | |
46097491 | 370 | static struct rtnetlink_notifier netdev_linux_poll_notifier; |
8b61709d BP |
371 | |
372 | /* This is set pretty low because we probably won't learn anything from the | |
373 | * additional log messages. */ | |
374 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
375 | ||
15b3596a | 376 | static int netdev_linux_init(void); |
6f643e49 | 377 | |
0b0544d7 | 378 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 379 | int cmd, const char *cmd_name); |
149f577a JG |
380 | static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, |
381 | const char *cmd_name); | |
f1acd62b BP |
382 | static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, |
383 | int cmd, const char *cmd_name); | |
8b61709d BP |
384 | static int get_flags(const struct netdev *, int *flagsp); |
385 | static int set_flags(struct netdev *, int flags); | |
386 | static int do_get_ifindex(const char *netdev_name); | |
387 | static int get_ifindex(const struct netdev *, int *ifindexp); | |
388 | static int do_set_addr(struct netdev *netdev, | |
389 | int ioctl_nr, const char *ioctl_name, | |
390 | struct in_addr addr); | |
391 | static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); | |
392 | static int set_etheraddr(const char *netdev_name, int hwaddr_family, | |
393 | const uint8_t[ETH_ADDR_LEN]); | |
394 | static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); | |
395 | static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); | |
396 | ||
15b3596a JG |
397 | static bool |
398 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
399 | { | |
400 | return netdev_class->init == netdev_linux_init; | |
401 | } | |
402 | ||
149f577a JG |
403 | static struct netdev_dev_linux * |
404 | netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) | |
6c88d577 | 405 | { |
15b3596a JG |
406 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); |
407 | assert(is_netdev_linux_class(netdev_class)); | |
408 | ||
149f577a | 409 | return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev); |
6c88d577 JP |
410 | } |
411 | ||
8b61709d BP |
412 | static struct netdev_linux * |
413 | netdev_linux_cast(const struct netdev *netdev) | |
414 | { | |
15b3596a JG |
415 | struct netdev_dev *netdev_dev = netdev_get_dev(netdev); |
416 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); | |
417 | assert(is_netdev_linux_class(netdev_class)); | |
418 | ||
8b61709d BP |
419 | return CONTAINER_OF(netdev, struct netdev_linux, netdev); |
420 | } | |
ff4ed3c9 | 421 | \f |
8b61709d BP |
422 | static int |
423 | netdev_linux_init(void) | |
424 | { | |
425 | static int status = -1; | |
426 | if (status < 0) { | |
ff4ed3c9 | 427 | /* Create AF_INET socket. */ |
8b61709d BP |
428 | af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); |
429 | status = af_inet_sock >= 0 ? 0 : errno; | |
430 | if (status) { | |
431 | VLOG_ERR("failed to create inet socket: %s", strerror(status)); | |
432 | } | |
ff4ed3c9 BP |
433 | |
434 | /* Create rtnetlink socket. */ | |
435 | if (!status) { | |
436 | status = nl_sock_create(NETLINK_ROUTE, 0, 0, 0, &rtnl_sock); | |
437 | if (status) { | |
438 | VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", | |
439 | strerror(status)); | |
440 | } | |
441 | } | |
8b61709d BP |
442 | } |
443 | return status; | |
444 | } | |
445 | ||
446 | static void | |
447 | netdev_linux_run(void) | |
448 | { | |
46097491 | 449 | rtnetlink_notifier_run(); |
8b61709d BP |
450 | } |
451 | ||
452 | static void | |
453 | netdev_linux_wait(void) | |
454 | { | |
46097491 | 455 | rtnetlink_notifier_wait(); |
8b61709d BP |
456 | } |
457 | ||
458 | static void | |
46097491 | 459 | netdev_linux_cache_cb(const struct rtnetlink_change *change, |
67a4917b | 460 | void *aux OVS_UNUSED) |
8b61709d | 461 | { |
149f577a | 462 | struct netdev_dev_linux *dev; |
8b61709d | 463 | if (change) { |
46415c90 JG |
464 | struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname); |
465 | if (base_dev) { | |
15b3596a JG |
466 | const struct netdev_class *netdev_class = |
467 | netdev_dev_get_class(base_dev); | |
468 | ||
469 | if (is_netdev_linux_class(netdev_class)) { | |
470 | dev = netdev_dev_linux_cast(base_dev); | |
471 | dev->cache_valid = 0; | |
472 | } | |
8b61709d BP |
473 | } |
474 | } else { | |
46415c90 | 475 | struct shash device_shash; |
8b61709d | 476 | struct shash_node *node; |
46415c90 JG |
477 | |
478 | shash_init(&device_shash); | |
479 | netdev_dev_get_devices(&netdev_linux_class, &device_shash); | |
480 | SHASH_FOR_EACH (node, &device_shash) { | |
149f577a JG |
481 | dev = node->data; |
482 | dev->cache_valid = 0; | |
8b61709d | 483 | } |
46415c90 | 484 | shash_destroy(&device_shash); |
8b61709d BP |
485 | } |
486 | } | |
487 | ||
149f577a | 488 | /* Creates the netdev device of 'type' with 'name'. */ |
8b61709d | 489 | static int |
b8dcf5e9 BP |
490 | netdev_linux_create_system(const struct netdev_class *class OVS_UNUSED, |
491 | const char *name, const struct shash *args, | |
492 | struct netdev_dev **netdev_devp) | |
6c88d577 | 493 | { |
149f577a JG |
494 | struct netdev_dev_linux *netdev_dev; |
495 | int error; | |
6c88d577 JP |
496 | |
497 | if (!shash_is_empty(args)) { | |
149f577a | 498 | VLOG_WARN("%s: arguments for system devices should be empty", name); |
6c88d577 JP |
499 | } |
500 | ||
46415c90 | 501 | if (!cache_notifier_refcount) { |
149f577a JG |
502 | error = rtnetlink_notifier_register(&netdev_linux_cache_notifier, |
503 | netdev_linux_cache_cb, NULL); | |
504 | if (error) { | |
505 | return error; | |
506 | } | |
507 | } | |
46415c90 | 508 | cache_notifier_refcount++; |
6c88d577 | 509 | |
149f577a | 510 | netdev_dev = xzalloc(sizeof *netdev_dev); |
149f577a | 511 | netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_linux_class); |
46415c90 | 512 | |
149f577a | 513 | *netdev_devp = &netdev_dev->netdev_dev; |
a740f0de JG |
514 | return 0; |
515 | } | |
516 | ||
5b7448ed JG |
517 | /* For most types of netdevs we open the device for each call of |
518 | * netdev_open(). However, this is not the case with tap devices, | |
519 | * since it is only possible to open the device once. In this | |
520 | * situation we share a single file descriptor, and consequently | |
521 | * buffers, across all readers. Therefore once data is read it will | |
522 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 523 | static int |
b8dcf5e9 BP |
524 | netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, |
525 | const char *name, const struct shash *args, | |
526 | struct netdev_dev **netdev_devp) | |
a740f0de | 527 | { |
149f577a | 528 | struct netdev_dev_linux *netdev_dev; |
a740f0de JG |
529 | struct tap_state *state; |
530 | static const char tap_dev[] = "/dev/net/tun"; | |
531 | struct ifreq ifr; | |
532 | int error; | |
533 | ||
534 | if (!shash_is_empty(args)) { | |
149f577a | 535 | VLOG_WARN("%s: arguments for TAP devices should be empty", name); |
6c88d577 JP |
536 | } |
537 | ||
149f577a JG |
538 | netdev_dev = xzalloc(sizeof *netdev_dev); |
539 | state = &netdev_dev->state.tap; | |
a740f0de | 540 | |
6c88d577 | 541 | /* Open tap device. */ |
149f577a JG |
542 | state->fd = open(tap_dev, O_RDWR); |
543 | if (state->fd < 0) { | |
6c88d577 JP |
544 | error = errno; |
545 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); | |
546 | goto error; | |
547 | } | |
548 | ||
549 | /* Create tap device. */ | |
550 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
551 | strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); | |
149f577a | 552 | if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 JP |
553 | VLOG_WARN("%s: creating tap device failed: %s", name, |
554 | strerror(errno)); | |
555 | error = errno; | |
556 | goto error; | |
557 | } | |
558 | ||
559 | /* Make non-blocking. */ | |
149f577a | 560 | error = set_nonblocking(state->fd); |
a740f0de JG |
561 | if (error) { |
562 | goto error; | |
563 | } | |
564 | ||
149f577a JG |
565 | netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); |
566 | *netdev_devp = &netdev_dev->netdev_dev; | |
a740f0de JG |
567 | return 0; |
568 | ||
569 | error: | |
149f577a | 570 | free(netdev_dev); |
a740f0de JG |
571 | return error; |
572 | } | |
573 | ||
a740f0de | 574 | static void |
149f577a | 575 | destroy_tap(struct netdev_dev_linux *netdev_dev) |
a740f0de | 576 | { |
149f577a JG |
577 | struct tap_state *state = &netdev_dev->state.tap; |
578 | ||
579 | if (state->fd >= 0) { | |
580 | close(state->fd); | |
a740f0de JG |
581 | } |
582 | } | |
583 | ||
149f577a | 584 | /* Destroys the netdev device 'netdev_dev_'. */ |
6c88d577 | 585 | static void |
149f577a | 586 | netdev_linux_destroy(struct netdev_dev *netdev_dev_) |
6c88d577 | 587 | { |
149f577a JG |
588 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
589 | const char *type = netdev_dev_get_type(netdev_dev_); | |
6c88d577 | 590 | |
c1c9c9c4 BP |
591 | if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) { |
592 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
593 | } | |
594 | ||
149f577a | 595 | if (!strcmp(type, "system")) { |
46415c90 | 596 | cache_notifier_refcount--; |
149f577a | 597 | |
46415c90 | 598 | if (!cache_notifier_refcount) { |
149f577a JG |
599 | rtnetlink_notifier_unregister(&netdev_linux_cache_notifier); |
600 | } | |
601 | } else if (!strcmp(type, "tap")) { | |
602 | destroy_tap(netdev_dev); | |
6c88d577 | 603 | } |
149f577a | 604 | |
658797c8 | 605 | free(netdev_dev); |
6c88d577 JP |
606 | } |
607 | ||
8b61709d | 608 | static int |
5b7448ed | 609 | netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, |
149f577a | 610 | struct netdev **netdevp) |
8b61709d | 611 | { |
5b7448ed | 612 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
8b61709d BP |
613 | struct netdev_linux *netdev; |
614 | enum netdev_flags flags; | |
615 | int error; | |
616 | ||
617 | /* Allocate network device. */ | |
ec6fde61 | 618 | netdev = xzalloc(sizeof *netdev); |
49a6a163 | 619 | netdev->fd = -1; |
5b7448ed | 620 | netdev_init(&netdev->netdev, netdev_dev_); |
8b61709d BP |
621 | |
622 | error = netdev_get_flags(&netdev->netdev, &flags); | |
623 | if (error == ENODEV) { | |
624 | goto error; | |
625 | } | |
626 | ||
61b999dd JG |
627 | if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") && |
628 | !netdev_dev->state.tap.opened) { | |
629 | ||
630 | /* We assume that the first user of the tap device is the primary user | |
631 | * and give them the tap FD. Subsequent users probably just expect | |
632 | * this to be a system device so open it normally to avoid send/receive | |
633 | * directions appearing to be reversed. */ | |
5b7448ed | 634 | netdev->fd = netdev_dev->state.tap.fd; |
61b999dd | 635 | netdev_dev->state.tap.opened = true; |
5b7448ed | 636 | } else if (ethertype != NETDEV_ETH_TYPE_NONE) { |
8b61709d BP |
637 | struct sockaddr_ll sll; |
638 | int protocol; | |
639 | int ifindex; | |
640 | ||
641 | /* Create file descriptor. */ | |
642 | protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL | |
643 | : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 | |
644 | : ethertype); | |
5b7448ed JG |
645 | netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); |
646 | if (netdev->fd < 0) { | |
8b61709d BP |
647 | error = errno; |
648 | goto error; | |
649 | } | |
8b61709d BP |
650 | |
651 | /* Set non-blocking mode. */ | |
5b7448ed | 652 | error = set_nonblocking(netdev->fd); |
8b61709d BP |
653 | if (error) { |
654 | goto error; | |
655 | } | |
656 | ||
657 | /* Get ethernet device index. */ | |
658 | error = get_ifindex(&netdev->netdev, &ifindex); | |
659 | if (error) { | |
660 | goto error; | |
661 | } | |
662 | ||
663 | /* Bind to specific ethernet device. */ | |
664 | memset(&sll, 0, sizeof sll); | |
665 | sll.sll_family = AF_PACKET; | |
666 | sll.sll_ifindex = ifindex; | |
5b7448ed | 667 | if (bind(netdev->fd, |
8b61709d BP |
668 | (struct sockaddr *) &sll, sizeof sll) < 0) { |
669 | error = errno; | |
5b7448ed | 670 | VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), |
149f577a | 671 | strerror(error)); |
8b61709d BP |
672 | goto error; |
673 | } | |
674 | ||
675 | /* Between the socket() and bind() calls above, the socket receives all | |
676 | * packets of the requested type on all system interfaces. We do not | |
677 | * want to receive that data, but there is no way to avoid it. So we | |
678 | * must now drain out the receive queue. */ | |
5b7448ed | 679 | error = drain_rcvbuf(netdev->fd); |
8b61709d BP |
680 | if (error) { |
681 | goto error; | |
682 | } | |
683 | } | |
684 | ||
685 | *netdevp = &netdev->netdev; | |
686 | return 0; | |
687 | ||
688 | error: | |
149f577a | 689 | netdev_uninit(&netdev->netdev, true); |
8b61709d BP |
690 | return error; |
691 | } | |
692 | ||
693 | /* Closes and destroys 'netdev'. */ | |
694 | static void | |
695 | netdev_linux_close(struct netdev *netdev_) | |
696 | { | |
697 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
698 | ||
49a6a163 | 699 | if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { |
5b7448ed | 700 | close(netdev->fd); |
8b61709d BP |
701 | } |
702 | free(netdev); | |
703 | } | |
e9e28be3 | 704 | |
8b61709d BP |
705 | /* Initializes 'svec' with a list of the names of all known network devices. */ |
706 | static int | |
707 | netdev_linux_enumerate(struct svec *svec) | |
708 | { | |
709 | struct if_nameindex *names; | |
710 | ||
711 | names = if_nameindex(); | |
712 | if (names) { | |
713 | size_t i; | |
714 | ||
715 | for (i = 0; names[i].if_name != NULL; i++) { | |
716 | svec_add(svec, names[i].if_name); | |
717 | } | |
718 | if_freenameindex(names); | |
719 | return 0; | |
720 | } else { | |
721 | VLOG_WARN("could not obtain list of network device names: %s", | |
722 | strerror(errno)); | |
723 | return errno; | |
724 | } | |
725 | } | |
726 | ||
727 | static int | |
728 | netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) | |
729 | { | |
730 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
731 | ||
5b7448ed | 732 | if (netdev->fd < 0) { |
8b61709d | 733 | /* Device was opened with NETDEV_ETH_TYPE_NONE. */ |
c0e5f6ca | 734 | return -EAGAIN; |
8b61709d BP |
735 | } |
736 | ||
737 | for (;;) { | |
5b7448ed | 738 | ssize_t retval = read(netdev->fd, data, size); |
8b61709d BP |
739 | if (retval >= 0) { |
740 | return retval; | |
741 | } else if (errno != EINTR) { | |
742 | if (errno != EAGAIN) { | |
743 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
744 | strerror(errno), netdev_get_name(netdev_)); | |
745 | } | |
c0e5f6ca | 746 | return -errno; |
8b61709d BP |
747 | } |
748 | } | |
749 | } | |
750 | ||
751 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
752 | * when a packet is ready to be received with netdev_recv() on 'netdev'. */ | |
753 | static void | |
754 | netdev_linux_recv_wait(struct netdev *netdev_) | |
755 | { | |
756 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed JG |
757 | if (netdev->fd >= 0) { |
758 | poll_fd_wait(netdev->fd, POLLIN); | |
8b61709d BP |
759 | } |
760 | } | |
761 | ||
762 | /* Discards all packets waiting to be received from 'netdev'. */ | |
763 | static int | |
764 | netdev_linux_drain(struct netdev *netdev_) | |
765 | { | |
766 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 767 | if (netdev->fd < 0) { |
8b61709d | 768 | return 0; |
5b7448ed | 769 | } else if (!strcmp(netdev_get_type(netdev_), "tap")) { |
8b61709d | 770 | struct ifreq ifr; |
149f577a | 771 | int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
8b61709d BP |
772 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
773 | if (error) { | |
774 | return error; | |
775 | } | |
5b7448ed | 776 | drain_fd(netdev->fd, ifr.ifr_qlen); |
8b61709d BP |
777 | return 0; |
778 | } else { | |
5b7448ed | 779 | return drain_rcvbuf(netdev->fd); |
8b61709d BP |
780 | } |
781 | } | |
782 | ||
783 | /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive | |
784 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued | |
785 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
786 | * the packet is too big or too small to transmit on the device. | |
787 | * | |
788 | * The caller retains ownership of 'buffer' in all cases. | |
789 | * | |
790 | * The kernel maintains a packet transmission queue, so the caller is not | |
791 | * expected to do additional queuing of packets. */ | |
792 | static int | |
793 | netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) | |
794 | { | |
795 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
796 | ||
797 | /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. | |
798 | */ | |
5b7448ed | 799 | if (netdev->fd < 0) { |
8b61709d BP |
800 | return EPIPE; |
801 | } | |
802 | ||
803 | for (;;) { | |
5b7448ed | 804 | ssize_t retval = write(netdev->fd, data, size); |
8b61709d BP |
805 | if (retval < 0) { |
806 | /* The Linux AF_PACKET implementation never blocks waiting for room | |
807 | * for packets, instead returning ENOBUFS. Translate this into | |
808 | * EAGAIN for the caller. */ | |
809 | if (errno == ENOBUFS) { | |
810 | return EAGAIN; | |
811 | } else if (errno == EINTR) { | |
812 | continue; | |
813 | } else if (errno != EAGAIN) { | |
814 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", | |
815 | netdev_get_name(netdev_), strerror(errno)); | |
816 | } | |
817 | return errno; | |
818 | } else if (retval != size) { | |
819 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " | |
820 | "%zu) on %s", retval, size, netdev_get_name(netdev_)); | |
821 | return EMSGSIZE; | |
822 | } else { | |
823 | return 0; | |
824 | } | |
825 | } | |
826 | } | |
827 | ||
828 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
829 | * when the packet transmission queue has sufficient room to transmit a packet | |
830 | * with netdev_send(). | |
831 | * | |
832 | * The kernel maintains a packet transmission queue, so the client is not | |
833 | * expected to do additional queuing of packets. Thus, this function is | |
834 | * unlikely to ever be used. It is included for completeness. */ | |
835 | static void | |
836 | netdev_linux_send_wait(struct netdev *netdev_) | |
837 | { | |
838 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 839 | if (netdev->fd < 0) { |
8b61709d | 840 | /* Nothing to do. */ |
5b7448ed JG |
841 | } else if (strcmp(netdev_get_type(netdev_), "tap")) { |
842 | poll_fd_wait(netdev->fd, POLLOUT); | |
8b61709d BP |
843 | } else { |
844 | /* TAP device always accepts packets.*/ | |
845 | poll_immediate_wake(); | |
846 | } | |
847 | } | |
848 | ||
849 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
850 | * otherwise a positive errno value. */ | |
851 | static int | |
852 | netdev_linux_set_etheraddr(struct netdev *netdev_, | |
853 | const uint8_t mac[ETH_ADDR_LEN]) | |
854 | { | |
149f577a JG |
855 | struct netdev_dev_linux *netdev_dev = |
856 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
eb395f2e BP |
857 | int error; |
858 | ||
149f577a JG |
859 | if (!(netdev_dev->cache_valid & VALID_ETHERADDR) |
860 | || !eth_addr_equals(netdev_dev->etheraddr, mac)) { | |
eb395f2e BP |
861 | error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); |
862 | if (!error) { | |
149f577a JG |
863 | netdev_dev->cache_valid |= VALID_ETHERADDR; |
864 | memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); | |
eb395f2e BP |
865 | } |
866 | } else { | |
867 | error = 0; | |
8b61709d BP |
868 | } |
869 | return error; | |
870 | } | |
871 | ||
872 | /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or | |
873 | * free the returned buffer. */ | |
874 | static int | |
875 | netdev_linux_get_etheraddr(const struct netdev *netdev_, | |
876 | uint8_t mac[ETH_ADDR_LEN]) | |
877 | { | |
149f577a JG |
878 | struct netdev_dev_linux *netdev_dev = |
879 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
880 | if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) { | |
8b61709d | 881 | int error = get_etheraddr(netdev_get_name(netdev_), |
149f577a | 882 | netdev_dev->etheraddr); |
8b61709d BP |
883 | if (error) { |
884 | return error; | |
885 | } | |
149f577a | 886 | netdev_dev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 887 | } |
149f577a | 888 | memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); |
8b61709d BP |
889 | return 0; |
890 | } | |
891 | ||
892 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
893 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
894 | * bytes for Ethernet devices. */ | |
895 | static int | |
896 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
897 | { | |
149f577a JG |
898 | struct netdev_dev_linux *netdev_dev = |
899 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
900 | if (!(netdev_dev->cache_valid & VALID_MTU)) { | |
8b61709d BP |
901 | struct ifreq ifr; |
902 | int error; | |
903 | ||
149f577a JG |
904 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
905 | SIOCGIFMTU, "SIOCGIFMTU"); | |
8b61709d BP |
906 | if (error) { |
907 | return error; | |
908 | } | |
149f577a JG |
909 | netdev_dev->mtu = ifr.ifr_mtu; |
910 | netdev_dev->cache_valid |= VALID_MTU; | |
8b61709d | 911 | } |
149f577a | 912 | *mtup = netdev_dev->mtu; |
8b61709d BP |
913 | return 0; |
914 | } | |
915 | ||
9ab3d9a3 BP |
916 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
917 | * On failure, returns a negative errno value. */ | |
918 | static int | |
919 | netdev_linux_get_ifindex(const struct netdev *netdev) | |
920 | { | |
921 | int ifindex, error; | |
922 | ||
923 | error = get_ifindex(netdev, &ifindex); | |
924 | return error ? -error : ifindex; | |
925 | } | |
926 | ||
8b61709d BP |
927 | static int |
928 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
929 | { | |
149f577a JG |
930 | struct netdev_dev_linux *netdev_dev = |
931 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
932 | int error = 0; |
933 | char *fn = NULL; | |
934 | int fd = -1; | |
935 | ||
149f577a | 936 | if (!(netdev_dev->cache_valid & VALID_CARRIER)) { |
8b61709d BP |
937 | char line[8]; |
938 | int retval; | |
939 | ||
149f577a JG |
940 | fn = xasprintf("/sys/class/net/%s/carrier", |
941 | netdev_get_name(netdev_)); | |
8b61709d BP |
942 | fd = open(fn, O_RDONLY); |
943 | if (fd < 0) { | |
944 | error = errno; | |
945 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); | |
946 | goto exit; | |
947 | } | |
948 | ||
949 | retval = read(fd, line, sizeof line); | |
950 | if (retval < 0) { | |
951 | error = errno; | |
952 | if (error == EINVAL) { | |
953 | /* This is the normal return value when we try to check carrier | |
954 | * if the network device is not up. */ | |
955 | } else { | |
956 | VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); | |
957 | } | |
958 | goto exit; | |
959 | } else if (retval == 0) { | |
960 | error = EPROTO; | |
961 | VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); | |
962 | goto exit; | |
963 | } | |
964 | ||
965 | if (line[0] != '0' && line[0] != '1') { | |
966 | error = EPROTO; | |
967 | VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", | |
968 | fn, line[0]); | |
969 | goto exit; | |
970 | } | |
149f577a JG |
971 | netdev_dev->carrier = line[0] != '0'; |
972 | netdev_dev->cache_valid |= VALID_CARRIER; | |
8b61709d | 973 | } |
149f577a | 974 | *carrier = netdev_dev->carrier; |
8b61709d BP |
975 | error = 0; |
976 | ||
977 | exit: | |
978 | if (fd >= 0) { | |
979 | close(fd); | |
980 | } | |
981 | free(fn); | |
982 | return error; | |
983 | } | |
984 | ||
985 | /* Check whether we can we use RTM_GETLINK to get network device statistics. | |
986 | * In pre-2.6.19 kernels, this was only available if wireless extensions were | |
987 | * enabled. */ | |
988 | static bool | |
989 | check_for_working_netlink_stats(void) | |
990 | { | |
991 | /* Decide on the netdev_get_stats() implementation to use. Netlink is | |
992 | * preferable, so if that works, we'll use it. */ | |
993 | int ifindex = do_get_ifindex("lo"); | |
994 | if (ifindex < 0) { | |
995 | VLOG_WARN("failed to get ifindex for lo, " | |
996 | "obtaining netdev stats from proc"); | |
997 | return false; | |
998 | } else { | |
999 | struct netdev_stats stats; | |
1000 | int error = get_stats_via_netlink(ifindex, &stats); | |
1001 | if (!error) { | |
1002 | VLOG_DBG("obtaining netdev stats via rtnetlink"); | |
1003 | return true; | |
1004 | } else { | |
1005 | VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " | |
1006 | "via proc (you are probably running a pre-2.6.19 " | |
1007 | "kernel)", strerror(error)); | |
1008 | return false; | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | ||
8722022c BP |
1013 | /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */ |
1014 | static void | |
1015 | netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev) | |
1016 | { | |
1017 | if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) { | |
1018 | const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev); | |
1019 | const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev); | |
d295e8e9 | 1020 | |
8722022c BP |
1021 | netdev_dev->is_tap = !strcmp(type, "tap"); |
1022 | netdev_dev->is_internal = false; | |
1023 | if (!netdev_dev->is_tap) { | |
1024 | struct ethtool_drvinfo drvinfo; | |
1025 | int error; | |
1026 | ||
1027 | memset(&drvinfo, 0, sizeof drvinfo); | |
1028 | error = netdev_linux_do_ethtool(name, | |
1029 | (struct ethtool_cmd *)&drvinfo, | |
1030 | ETHTOOL_GDRVINFO, | |
1031 | "ETHTOOL_GDRVINFO"); | |
1032 | ||
1033 | if (!error && !strcmp(drvinfo.driver, "openvswitch")) { | |
1034 | netdev_dev->is_internal = true; | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | netdev_dev->cache_valid |= VALID_IS_PSEUDO; | |
1039 | } | |
1040 | } | |
1041 | ||
92df599c JG |
1042 | static void |
1043 | swap_uint64(uint64_t *a, uint64_t *b) | |
1044 | { | |
1045 | *a ^= *b; | |
1046 | *b ^= *a; | |
1047 | *a ^= *b; | |
1048 | } | |
1049 | ||
7fbef77a | 1050 | /* Retrieves current device stats for 'netdev'. */ |
8b61709d | 1051 | static int |
149f577a JG |
1052 | netdev_linux_get_stats(const struct netdev *netdev_, |
1053 | struct netdev_stats *stats) | |
8b61709d | 1054 | { |
149f577a JG |
1055 | struct netdev_dev_linux *netdev_dev = |
1056 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
1057 | static int use_netlink_stats = -1; |
1058 | int error; | |
1059 | ||
1060 | COVERAGE_INC(netdev_get_stats); | |
fe6b0e03 | 1061 | |
7fbef77a JG |
1062 | if (netdev_dev->have_vport_stats || |
1063 | !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { | |
1064 | ||
1065 | error = netdev_vport_get_stats(netdev_, stats); | |
1066 | netdev_dev->have_vport_stats = !error; | |
1067 | netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS; | |
8b61709d | 1068 | } |
8b61709d | 1069 | |
7fbef77a JG |
1070 | if (!netdev_dev->have_vport_stats) { |
1071 | if (use_netlink_stats < 0) { | |
1072 | use_netlink_stats = check_for_working_netlink_stats(); | |
1073 | } | |
1074 | if (use_netlink_stats) { | |
1075 | int ifindex; | |
1076 | ||
1077 | error = get_ifindex(netdev_, &ifindex); | |
1078 | if (!error) { | |
1079 | error = get_stats_via_netlink(ifindex, stats); | |
1080 | } | |
1081 | } else { | |
1082 | error = get_stats_via_proc(netdev_get_name(netdev_), stats); | |
8b61709d | 1083 | } |
8b61709d | 1084 | } |
fe6b0e03 JG |
1085 | |
1086 | /* If this port is an internal port then the transmit and receive stats | |
1087 | * will appear to be swapped relative to the other ports since we are the | |
1088 | * one sending the data, not a remote computer. For consistency, we swap | |
7fbef77a JG |
1089 | * them back here. This does not apply if we are getting stats from the |
1090 | * vport layer because it always tracks stats from the perspective of the | |
1091 | * switch. */ | |
92df599c | 1092 | netdev_linux_update_is_pseudo(netdev_dev); |
7fbef77a JG |
1093 | if (!error && !netdev_dev->have_vport_stats && |
1094 | (netdev_dev->is_internal || netdev_dev->is_tap)) { | |
92df599c JG |
1095 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1096 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1097 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1098 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1099 | stats->rx_length_errors = 0; |
1100 | stats->rx_over_errors = 0; | |
1101 | stats->rx_crc_errors = 0; | |
1102 | stats->rx_frame_errors = 0; | |
1103 | stats->rx_fifo_errors = 0; | |
1104 | stats->rx_missed_errors = 0; | |
1105 | stats->tx_aborted_errors = 0; | |
1106 | stats->tx_carrier_errors = 0; | |
1107 | stats->tx_fifo_errors = 0; | |
1108 | stats->tx_heartbeat_errors = 0; | |
1109 | stats->tx_window_errors = 0; | |
1110 | } | |
1111 | ||
8b61709d BP |
1112 | return error; |
1113 | } | |
1114 | ||
1115 | /* Stores the features supported by 'netdev' into each of '*current', | |
1116 | * '*advertised', '*supported', and '*peer' that are non-null. Each value is a | |
1117 | * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if | |
7671589a | 1118 | * successful, otherwise a positive errno value. */ |
8b61709d BP |
1119 | static int |
1120 | netdev_linux_get_features(struct netdev *netdev, | |
1121 | uint32_t *current, uint32_t *advertised, | |
1122 | uint32_t *supported, uint32_t *peer) | |
1123 | { | |
1124 | struct ethtool_cmd ecmd; | |
1125 | int error; | |
1126 | ||
1127 | memset(&ecmd, 0, sizeof ecmd); | |
0b0544d7 | 1128 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1129 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1130 | if (error) { | |
1131 | return error; | |
1132 | } | |
1133 | ||
1134 | /* Supported features. */ | |
1135 | *supported = 0; | |
1136 | if (ecmd.supported & SUPPORTED_10baseT_Half) { | |
1137 | *supported |= OFPPF_10MB_HD; | |
1138 | } | |
1139 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
1140 | *supported |= OFPPF_10MB_FD; | |
1141 | } | |
1142 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
1143 | *supported |= OFPPF_100MB_HD; | |
1144 | } | |
1145 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
1146 | *supported |= OFPPF_100MB_FD; | |
1147 | } | |
1148 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
1149 | *supported |= OFPPF_1GB_HD; | |
1150 | } | |
1151 | if (ecmd.supported & SUPPORTED_1000baseT_Full) { | |
1152 | *supported |= OFPPF_1GB_FD; | |
1153 | } | |
1154 | if (ecmd.supported & SUPPORTED_10000baseT_Full) { | |
1155 | *supported |= OFPPF_10GB_FD; | |
1156 | } | |
1157 | if (ecmd.supported & SUPPORTED_TP) { | |
1158 | *supported |= OFPPF_COPPER; | |
1159 | } | |
1160 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
1161 | *supported |= OFPPF_FIBER; | |
1162 | } | |
1163 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
1164 | *supported |= OFPPF_AUTONEG; | |
1165 | } | |
1166 | if (ecmd.supported & SUPPORTED_Pause) { | |
1167 | *supported |= OFPPF_PAUSE; | |
1168 | } | |
1169 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
1170 | *supported |= OFPPF_PAUSE_ASYM; | |
1171 | } | |
1172 | ||
1173 | /* Advertised features. */ | |
1174 | *advertised = 0; | |
1175 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { | |
1176 | *advertised |= OFPPF_10MB_HD; | |
1177 | } | |
1178 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
1179 | *advertised |= OFPPF_10MB_FD; | |
1180 | } | |
1181 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
1182 | *advertised |= OFPPF_100MB_HD; | |
1183 | } | |
1184 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
1185 | *advertised |= OFPPF_100MB_FD; | |
1186 | } | |
1187 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
1188 | *advertised |= OFPPF_1GB_HD; | |
1189 | } | |
1190 | if (ecmd.advertising & ADVERTISED_1000baseT_Full) { | |
1191 | *advertised |= OFPPF_1GB_FD; | |
1192 | } | |
1193 | if (ecmd.advertising & ADVERTISED_10000baseT_Full) { | |
1194 | *advertised |= OFPPF_10GB_FD; | |
1195 | } | |
1196 | if (ecmd.advertising & ADVERTISED_TP) { | |
1197 | *advertised |= OFPPF_COPPER; | |
1198 | } | |
1199 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
1200 | *advertised |= OFPPF_FIBER; | |
1201 | } | |
1202 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
1203 | *advertised |= OFPPF_AUTONEG; | |
1204 | } | |
1205 | if (ecmd.advertising & ADVERTISED_Pause) { | |
1206 | *advertised |= OFPPF_PAUSE; | |
1207 | } | |
1208 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
1209 | *advertised |= OFPPF_PAUSE_ASYM; | |
1210 | } | |
1211 | ||
1212 | /* Current settings. */ | |
1213 | if (ecmd.speed == SPEED_10) { | |
1214 | *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; | |
1215 | } else if (ecmd.speed == SPEED_100) { | |
1216 | *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; | |
1217 | } else if (ecmd.speed == SPEED_1000) { | |
1218 | *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; | |
1219 | } else if (ecmd.speed == SPEED_10000) { | |
1220 | *current = OFPPF_10GB_FD; | |
1221 | } else { | |
1222 | *current = 0; | |
1223 | } | |
1224 | ||
1225 | if (ecmd.port == PORT_TP) { | |
1226 | *current |= OFPPF_COPPER; | |
1227 | } else if (ecmd.port == PORT_FIBRE) { | |
1228 | *current |= OFPPF_FIBER; | |
1229 | } | |
1230 | ||
1231 | if (ecmd.autoneg) { | |
1232 | *current |= OFPPF_AUTONEG; | |
1233 | } | |
1234 | ||
1235 | /* Peer advertisements. */ | |
1236 | *peer = 0; /* XXX */ | |
1237 | ||
1238 | return 0; | |
1239 | } | |
1240 | ||
1241 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
1242 | static int | |
1243 | netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) | |
1244 | { | |
1245 | struct ethtool_cmd ecmd; | |
1246 | int error; | |
1247 | ||
1248 | memset(&ecmd, 0, sizeof ecmd); | |
0b0544d7 | 1249 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1250 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1251 | if (error) { | |
1252 | return error; | |
1253 | } | |
1254 | ||
1255 | ecmd.advertising = 0; | |
1256 | if (advertise & OFPPF_10MB_HD) { | |
1257 | ecmd.advertising |= ADVERTISED_10baseT_Half; | |
1258 | } | |
1259 | if (advertise & OFPPF_10MB_FD) { | |
1260 | ecmd.advertising |= ADVERTISED_10baseT_Full; | |
1261 | } | |
1262 | if (advertise & OFPPF_100MB_HD) { | |
1263 | ecmd.advertising |= ADVERTISED_100baseT_Half; | |
1264 | } | |
1265 | if (advertise & OFPPF_100MB_FD) { | |
1266 | ecmd.advertising |= ADVERTISED_100baseT_Full; | |
1267 | } | |
1268 | if (advertise & OFPPF_1GB_HD) { | |
1269 | ecmd.advertising |= ADVERTISED_1000baseT_Half; | |
1270 | } | |
1271 | if (advertise & OFPPF_1GB_FD) { | |
1272 | ecmd.advertising |= ADVERTISED_1000baseT_Full; | |
1273 | } | |
1274 | if (advertise & OFPPF_10GB_FD) { | |
1275 | ecmd.advertising |= ADVERTISED_10000baseT_Full; | |
1276 | } | |
1277 | if (advertise & OFPPF_COPPER) { | |
1278 | ecmd.advertising |= ADVERTISED_TP; | |
1279 | } | |
1280 | if (advertise & OFPPF_FIBER) { | |
1281 | ecmd.advertising |= ADVERTISED_FIBRE; | |
1282 | } | |
1283 | if (advertise & OFPPF_AUTONEG) { | |
1284 | ecmd.advertising |= ADVERTISED_Autoneg; | |
1285 | } | |
1286 | if (advertise & OFPPF_PAUSE) { | |
1287 | ecmd.advertising |= ADVERTISED_Pause; | |
1288 | } | |
1289 | if (advertise & OFPPF_PAUSE_ASYM) { | |
1290 | ecmd.advertising |= ADVERTISED_Asym_Pause; | |
1291 | } | |
0b0544d7 | 1292 | return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1293 | ETHTOOL_SSET, "ETHTOOL_SSET"); |
1294 | } | |
1295 | ||
1296 | /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with | |
1297 | * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device | |
1298 | * and returns 0. Otherwise returns a errno value (specifically ENOENT if | |
1299 | * 'netdev_name' is the name of a network device that is not a VLAN device) and | |
1300 | * sets '*vlan_vid' to -1. */ | |
1301 | static int | |
1302 | netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) | |
1303 | { | |
1304 | const char *netdev_name = netdev_get_name(netdev); | |
1305 | struct ds line = DS_EMPTY_INITIALIZER; | |
1306 | FILE *stream = NULL; | |
1307 | int error; | |
1308 | char *fn; | |
1309 | ||
1310 | COVERAGE_INC(netdev_get_vlan_vid); | |
1311 | fn = xasprintf("/proc/net/vlan/%s", netdev_name); | |
1312 | stream = fopen(fn, "r"); | |
1313 | if (!stream) { | |
1314 | error = errno; | |
1315 | goto done; | |
1316 | } | |
1317 | ||
1318 | if (ds_get_line(&line, stream)) { | |
1319 | if (ferror(stream)) { | |
1320 | error = errno; | |
1321 | VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno)); | |
1322 | } else { | |
1323 | error = EPROTO; | |
1324 | VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn); | |
1325 | } | |
1326 | goto done; | |
1327 | } | |
1328 | ||
1329 | if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) { | |
1330 | error = EPROTO; | |
1331 | VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"", | |
1332 | fn, ds_cstr(&line)); | |
1333 | goto done; | |
1334 | } | |
1335 | ||
1336 | error = 0; | |
1337 | ||
1338 | done: | |
1339 | free(fn); | |
1340 | if (stream) { | |
1341 | fclose(stream); | |
1342 | } | |
1343 | ds_destroy(&line); | |
1344 | if (error) { | |
1345 | *vlan_vid = -1; | |
1346 | } | |
1347 | return error; | |
1348 | } | |
1349 | ||
1350 | #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress" | |
1351 | #define POLICE_CONFIG_CMD "/sbin/tc filter add dev %s parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate %dkbit burst %dk mtu 65535 drop flowid :1" | |
8b61709d | 1352 | |
8e460221 | 1353 | /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a |
6f42c8ea BP |
1354 | * positive errno value. |
1355 | * | |
1356 | * This function is equivalent to running | |
1357 | * /sbin/tc qdisc del dev %s handle ffff: ingress | |
1358 | * but it is much, much faster. | |
1359 | */ | |
8e460221 BP |
1360 | static int |
1361 | netdev_linux_remove_policing(struct netdev *netdev) | |
1362 | { | |
80a86fbe BP |
1363 | struct netdev_dev_linux *netdev_dev = |
1364 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
8e460221 | 1365 | const char *netdev_name = netdev_get_name(netdev); |
8e460221 | 1366 | |
6f42c8ea | 1367 | struct ofpbuf request; |
6f42c8ea | 1368 | struct tcmsg *tcmsg; |
6f42c8ea BP |
1369 | int error; |
1370 | ||
c1c9c9c4 BP |
1371 | tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); |
1372 | tcmsg->tcm_handle = tc_make_handle(0xffff, 0); | |
6f42c8ea BP |
1373 | tcmsg->tcm_parent = TC_H_INGRESS; |
1374 | nl_msg_put_string(&request, TCA_KIND, "ingress"); | |
1375 | nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); | |
c1c9c9c4 BP |
1376 | |
1377 | error = tc_transact(&request, NULL); | |
4d10512c | 1378 | if (error && error != ENOENT && error != EINVAL) { |
6f42c8ea BP |
1379 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", |
1380 | netdev_name, strerror(error)); | |
1381 | return error; | |
1382 | } | |
1383 | ||
80a86fbe BP |
1384 | netdev_dev->kbits_rate = 0; |
1385 | netdev_dev->kbits_burst = 0; | |
1386 | netdev_dev->cache_valid |= VALID_POLICING; | |
8e460221 BP |
1387 | return 0; |
1388 | } | |
1389 | ||
8b61709d BP |
1390 | /* Attempts to set input rate limiting (policing) policy. */ |
1391 | static int | |
1392 | netdev_linux_set_policing(struct netdev *netdev, | |
1393 | uint32_t kbits_rate, uint32_t kbits_burst) | |
1394 | { | |
80a86fbe BP |
1395 | struct netdev_dev_linux *netdev_dev = |
1396 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
8b61709d BP |
1397 | const char *netdev_name = netdev_get_name(netdev); |
1398 | char command[1024]; | |
1399 | ||
1400 | COVERAGE_INC(netdev_set_policing); | |
8e460221 | 1401 | |
80a86fbe BP |
1402 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
1403 | : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ | |
1404 | : kbits_burst); /* Stick with user-specified value. */ | |
1405 | ||
1406 | if (netdev_dev->cache_valid & VALID_POLICING | |
1407 | && netdev_dev->kbits_rate == kbits_rate | |
1408 | && netdev_dev->kbits_burst == kbits_burst) { | |
1409 | /* Assume that settings haven't changed since we last set them. */ | |
1410 | return 0; | |
1411 | } | |
1412 | ||
8e460221 | 1413 | netdev_linux_remove_policing(netdev); |
8b61709d | 1414 | if (kbits_rate) { |
8b61709d BP |
1415 | snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name); |
1416 | if (system(command) != 0) { | |
1417 | VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name); | |
1418 | return -1; | |
1419 | } | |
1420 | ||
1421 | snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name, | |
1422 | kbits_rate, kbits_burst); | |
1423 | if (system(command) != 0) { | |
1424 | VLOG_WARN_RL(&rl, "%s: problem configuring policing", | |
1425 | netdev_name); | |
1426 | return -1; | |
1427 | } | |
80a86fbe BP |
1428 | |
1429 | netdev_dev->kbits_rate = kbits_rate; | |
1430 | netdev_dev->kbits_burst = kbits_burst; | |
1431 | netdev_dev->cache_valid |= VALID_POLICING; | |
8b61709d BP |
1432 | } |
1433 | ||
1434 | return 0; | |
1435 | } | |
1436 | ||
c1c9c9c4 BP |
1437 | static int |
1438 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
1439 | struct svec *types) | |
1440 | { | |
1441 | const struct tc_ops **opsp; | |
1442 | ||
1443 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1444 | const struct tc_ops *ops = *opsp; | |
1445 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
1446 | svec_add(types, ops->ovs_name); | |
1447 | } | |
1448 | } | |
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | static const struct tc_ops * | |
1453 | tc_lookup_ovs_name(const char *name) | |
1454 | { | |
1455 | const struct tc_ops **opsp; | |
1456 | ||
1457 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1458 | const struct tc_ops *ops = *opsp; | |
1459 | if (!strcmp(name, ops->ovs_name)) { | |
1460 | return ops; | |
1461 | } | |
1462 | } | |
1463 | return NULL; | |
1464 | } | |
1465 | ||
1466 | static const struct tc_ops * | |
1467 | tc_lookup_linux_name(const char *name) | |
1468 | { | |
1469 | const struct tc_ops **opsp; | |
1470 | ||
1471 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1472 | const struct tc_ops *ops = *opsp; | |
1473 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
1474 | return ops; | |
1475 | } | |
1476 | } | |
1477 | return NULL; | |
1478 | } | |
1479 | ||
93b13be8 BP |
1480 | static struct tc_queue * |
1481 | tc_find_queue__(const struct netdev *netdev, unsigned int queue_id, | |
1482 | size_t hash) | |
1483 | { | |
1484 | struct netdev_dev_linux *netdev_dev = | |
1485 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1486 | struct tc_queue *queue; | |
1487 | ||
4e8e4213 | 1488 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) { |
93b13be8 BP |
1489 | if (queue->queue_id == queue_id) { |
1490 | return queue; | |
1491 | } | |
1492 | } | |
1493 | return NULL; | |
1494 | } | |
1495 | ||
1496 | static struct tc_queue * | |
1497 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
1498 | { | |
1499 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
1500 | } | |
1501 | ||
c1c9c9c4 BP |
1502 | static int |
1503 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
1504 | const char *type, | |
1505 | struct netdev_qos_capabilities *caps) | |
1506 | { | |
1507 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
1508 | if (!ops) { | |
1509 | return EOPNOTSUPP; | |
1510 | } | |
1511 | caps->n_queues = ops->n_queues; | |
1512 | return 0; | |
1513 | } | |
1514 | ||
1515 | static int | |
1516 | netdev_linux_get_qos(const struct netdev *netdev, | |
1517 | const char **typep, struct shash *details) | |
1518 | { | |
1519 | struct netdev_dev_linux *netdev_dev = | |
1520 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1521 | int error; | |
1522 | ||
1523 | error = tc_query_qdisc(netdev); | |
1524 | if (error) { | |
1525 | return error; | |
1526 | } | |
1527 | ||
1528 | *typep = netdev_dev->tc->ops->ovs_name; | |
1529 | return (netdev_dev->tc->ops->qdisc_get | |
1530 | ? netdev_dev->tc->ops->qdisc_get(netdev, details) | |
1531 | : 0); | |
1532 | } | |
1533 | ||
1534 | static int | |
1535 | netdev_linux_set_qos(struct netdev *netdev, | |
1536 | const char *type, const struct shash *details) | |
1537 | { | |
1538 | struct netdev_dev_linux *netdev_dev = | |
1539 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1540 | const struct tc_ops *new_ops; | |
1541 | int error; | |
1542 | ||
1543 | new_ops = tc_lookup_ovs_name(type); | |
1544 | if (!new_ops || !new_ops->tc_install) { | |
1545 | return EOPNOTSUPP; | |
1546 | } | |
1547 | ||
1548 | error = tc_query_qdisc(netdev); | |
1549 | if (error) { | |
1550 | return error; | |
1551 | } | |
1552 | ||
1553 | if (new_ops == netdev_dev->tc->ops) { | |
1554 | return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0; | |
1555 | } else { | |
1556 | /* Delete existing qdisc. */ | |
1557 | error = tc_del_qdisc(netdev); | |
1558 | if (error) { | |
1559 | return error; | |
1560 | } | |
1561 | assert(netdev_dev->tc == NULL); | |
1562 | ||
1563 | /* Install new qdisc. */ | |
1564 | error = new_ops->tc_install(netdev, details); | |
1565 | assert((error == 0) == (netdev_dev->tc != NULL)); | |
1566 | ||
1567 | return error; | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | static int | |
1572 | netdev_linux_get_queue(const struct netdev *netdev, | |
1573 | unsigned int queue_id, struct shash *details) | |
1574 | { | |
1575 | struct netdev_dev_linux *netdev_dev = | |
1576 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1577 | int error; | |
1578 | ||
1579 | error = tc_query_qdisc(netdev); | |
1580 | if (error) { | |
1581 | return error; | |
93b13be8 BP |
1582 | } else { |
1583 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1584 | return (queue | |
1585 | ? netdev_dev->tc->ops->class_get(netdev, queue, details) | |
1586 | : ENOENT); | |
c1c9c9c4 | 1587 | } |
c1c9c9c4 BP |
1588 | } |
1589 | ||
1590 | static int | |
1591 | netdev_linux_set_queue(struct netdev *netdev, | |
1592 | unsigned int queue_id, const struct shash *details) | |
1593 | { | |
1594 | struct netdev_dev_linux *netdev_dev = | |
1595 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1596 | int error; | |
1597 | ||
1598 | error = tc_query_qdisc(netdev); | |
1599 | if (error) { | |
1600 | return error; | |
1601 | } else if (queue_id >= netdev_dev->tc->ops->n_queues | |
1602 | || !netdev_dev->tc->ops->class_set) { | |
1603 | return EINVAL; | |
1604 | } | |
1605 | ||
1606 | return netdev_dev->tc->ops->class_set(netdev, queue_id, details); | |
1607 | } | |
1608 | ||
1609 | static int | |
1610 | netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id) | |
1611 | { | |
1612 | struct netdev_dev_linux *netdev_dev = | |
1613 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1614 | int error; | |
1615 | ||
1616 | error = tc_query_qdisc(netdev); | |
1617 | if (error) { | |
1618 | return error; | |
1619 | } else if (!netdev_dev->tc->ops->class_delete) { | |
1620 | return EINVAL; | |
93b13be8 BP |
1621 | } else { |
1622 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1623 | return (queue | |
1624 | ? netdev_dev->tc->ops->class_delete(netdev, queue) | |
1625 | : ENOENT); | |
c1c9c9c4 | 1626 | } |
c1c9c9c4 BP |
1627 | } |
1628 | ||
1629 | static int | |
1630 | netdev_linux_get_queue_stats(const struct netdev *netdev, | |
1631 | unsigned int queue_id, | |
1632 | struct netdev_queue_stats *stats) | |
1633 | { | |
1634 | struct netdev_dev_linux *netdev_dev = | |
1635 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1636 | int error; | |
1637 | ||
1638 | error = tc_query_qdisc(netdev); | |
1639 | if (error) { | |
1640 | return error; | |
c1c9c9c4 BP |
1641 | } else if (!netdev_dev->tc->ops->class_get_stats) { |
1642 | return EOPNOTSUPP; | |
93b13be8 BP |
1643 | } else { |
1644 | const struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1645 | return (queue | |
1646 | ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats) | |
1647 | : ENOENT); | |
c1c9c9c4 | 1648 | } |
c1c9c9c4 BP |
1649 | } |
1650 | ||
1651 | static void | |
1652 | start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) | |
1653 | { | |
1654 | struct ofpbuf request; | |
1655 | struct tcmsg *tcmsg; | |
1656 | ||
1657 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); | |
3c4de644 | 1658 | tcmsg->tcm_parent = 0; |
c1c9c9c4 BP |
1659 | nl_dump_start(dump, rtnl_sock, &request); |
1660 | ofpbuf_uninit(&request); | |
1661 | } | |
1662 | ||
1663 | static int | |
1664 | netdev_linux_dump_queues(const struct netdev *netdev, | |
1665 | netdev_dump_queues_cb *cb, void *aux) | |
1666 | { | |
1667 | struct netdev_dev_linux *netdev_dev = | |
1668 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
93b13be8 | 1669 | struct tc_queue *queue; |
c1c9c9c4 BP |
1670 | struct shash details; |
1671 | int last_error; | |
c1c9c9c4 BP |
1672 | int error; |
1673 | ||
1674 | error = tc_query_qdisc(netdev); | |
1675 | if (error) { | |
1676 | return error; | |
1677 | } else if (!netdev_dev->tc->ops->class_get) { | |
1678 | return EOPNOTSUPP; | |
1679 | } | |
1680 | ||
1681 | last_error = 0; | |
1682 | shash_init(&details); | |
4e8e4213 | 1683 | HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) { |
c1c9c9c4 BP |
1684 | shash_clear(&details); |
1685 | ||
93b13be8 | 1686 | error = netdev_dev->tc->ops->class_get(netdev, queue, &details); |
c1c9c9c4 | 1687 | if (!error) { |
93b13be8 | 1688 | (*cb)(queue->queue_id, &details, aux); |
c1c9c9c4 BP |
1689 | } else { |
1690 | last_error = error; | |
1691 | } | |
1692 | } | |
1693 | shash_destroy(&details); | |
1694 | ||
1695 | return last_error; | |
1696 | } | |
1697 | ||
1698 | static int | |
1699 | netdev_linux_dump_queue_stats(const struct netdev *netdev, | |
1700 | netdev_dump_queue_stats_cb *cb, void *aux) | |
1701 | { | |
1702 | struct netdev_dev_linux *netdev_dev = | |
1703 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1704 | struct nl_dump dump; | |
1705 | struct ofpbuf msg; | |
1706 | int last_error; | |
1707 | int error; | |
1708 | ||
1709 | error = tc_query_qdisc(netdev); | |
1710 | if (error) { | |
1711 | return error; | |
1712 | } else if (!netdev_dev->tc->ops->class_dump_stats) { | |
1713 | return EOPNOTSUPP; | |
1714 | } | |
1715 | ||
1716 | last_error = 0; | |
1717 | start_queue_dump(netdev, &dump); | |
1718 | while (nl_dump_next(&dump, &msg)) { | |
1719 | error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux); | |
1720 | if (error) { | |
1721 | last_error = error; | |
1722 | } | |
1723 | } | |
1724 | ||
1725 | error = nl_dump_done(&dump); | |
1726 | return error ? error : last_error; | |
1727 | } | |
1728 | ||
8b61709d | 1729 | static int |
f1acd62b BP |
1730 | netdev_linux_get_in4(const struct netdev *netdev_, |
1731 | struct in_addr *address, struct in_addr *netmask) | |
8b61709d | 1732 | { |
149f577a JG |
1733 | struct netdev_dev_linux *netdev_dev = |
1734 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1735 | ||
1736 | if (!(netdev_dev->cache_valid & VALID_IN4)) { | |
8b61709d BP |
1737 | int error; |
1738 | ||
149f577a | 1739 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, |
8b61709d BP |
1740 | SIOCGIFADDR, "SIOCGIFADDR"); |
1741 | if (error) { | |
1742 | return error; | |
1743 | } | |
1744 | ||
149f577a | 1745 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, |
f1acd62b BP |
1746 | SIOCGIFNETMASK, "SIOCGIFNETMASK"); |
1747 | if (error) { | |
1748 | return error; | |
1749 | } | |
1750 | ||
149f577a | 1751 | netdev_dev->cache_valid |= VALID_IN4; |
8b61709d | 1752 | } |
149f577a JG |
1753 | *address = netdev_dev->address; |
1754 | *netmask = netdev_dev->netmask; | |
f1acd62b | 1755 | return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; |
8b61709d BP |
1756 | } |
1757 | ||
8b61709d | 1758 | static int |
f1acd62b BP |
1759 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
1760 | struct in_addr netmask) | |
8b61709d | 1761 | { |
149f577a JG |
1762 | struct netdev_dev_linux *netdev_dev = |
1763 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
1764 | int error; |
1765 | ||
f1acd62b | 1766 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 1767 | if (!error) { |
149f577a JG |
1768 | netdev_dev->cache_valid |= VALID_IN4; |
1769 | netdev_dev->address = address; | |
1770 | netdev_dev->netmask = netmask; | |
f1acd62b | 1771 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 1772 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 1773 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
1774 | } |
1775 | } | |
1776 | return error; | |
1777 | } | |
1778 | ||
1779 | static bool | |
1780 | parse_if_inet6_line(const char *line, | |
1781 | struct in6_addr *in6, char ifname[16 + 1]) | |
1782 | { | |
1783 | uint8_t *s6 = in6->s6_addr; | |
1784 | #define X8 "%2"SCNx8 | |
1785 | return sscanf(line, | |
1786 | " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 | |
1787 | "%*x %*x %*x %*x %16s\n", | |
1788 | &s6[0], &s6[1], &s6[2], &s6[3], | |
1789 | &s6[4], &s6[5], &s6[6], &s6[7], | |
1790 | &s6[8], &s6[9], &s6[10], &s6[11], | |
1791 | &s6[12], &s6[13], &s6[14], &s6[15], | |
1792 | ifname) == 17; | |
1793 | } | |
1794 | ||
1795 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if | |
1796 | * 'in6' is non-null) and returns true. Otherwise, returns false. */ | |
1797 | static int | |
1798 | netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) | |
1799 | { | |
149f577a JG |
1800 | struct netdev_dev_linux *netdev_dev = |
1801 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1802 | if (!(netdev_dev->cache_valid & VALID_IN6)) { | |
8b61709d BP |
1803 | FILE *file; |
1804 | char line[128]; | |
1805 | ||
149f577a | 1806 | netdev_dev->in6 = in6addr_any; |
8b61709d BP |
1807 | |
1808 | file = fopen("/proc/net/if_inet6", "r"); | |
1809 | if (file != NULL) { | |
1810 | const char *name = netdev_get_name(netdev_); | |
1811 | while (fgets(line, sizeof line, file)) { | |
2a022368 | 1812 | struct in6_addr in6_tmp; |
8b61709d | 1813 | char ifname[16 + 1]; |
2a022368 | 1814 | if (parse_if_inet6_line(line, &in6_tmp, ifname) |
8b61709d BP |
1815 | && !strcmp(name, ifname)) |
1816 | { | |
2a022368 | 1817 | netdev_dev->in6 = in6_tmp; |
8b61709d BP |
1818 | break; |
1819 | } | |
1820 | } | |
1821 | fclose(file); | |
1822 | } | |
149f577a | 1823 | netdev_dev->cache_valid |= VALID_IN6; |
8b61709d | 1824 | } |
149f577a | 1825 | *in6 = netdev_dev->in6; |
8b61709d BP |
1826 | return 0; |
1827 | } | |
1828 | ||
1829 | static void | |
1830 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
1831 | { | |
1832 | struct sockaddr_in sin; | |
1833 | memset(&sin, 0, sizeof sin); | |
1834 | sin.sin_family = AF_INET; | |
1835 | sin.sin_addr = addr; | |
1836 | sin.sin_port = 0; | |
1837 | ||
1838 | memset(sa, 0, sizeof *sa); | |
1839 | memcpy(sa, &sin, sizeof sin); | |
1840 | } | |
1841 | ||
1842 | static int | |
1843 | do_set_addr(struct netdev *netdev, | |
1844 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
1845 | { | |
1846 | struct ifreq ifr; | |
149f577a | 1847 | strncpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); |
8b61709d | 1848 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
149f577a JG |
1849 | |
1850 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
1851 | ioctl_name); | |
8b61709d BP |
1852 | } |
1853 | ||
1854 | /* Adds 'router' as a default IP gateway. */ | |
1855 | static int | |
67a4917b | 1856 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
1857 | { |
1858 | struct in_addr any = { INADDR_ANY }; | |
1859 | struct rtentry rt; | |
1860 | int error; | |
1861 | ||
1862 | memset(&rt, 0, sizeof rt); | |
1863 | make_in4_sockaddr(&rt.rt_dst, any); | |
1864 | make_in4_sockaddr(&rt.rt_gateway, router); | |
1865 | make_in4_sockaddr(&rt.rt_genmask, any); | |
1866 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
1867 | COVERAGE_INC(netdev_add_router); | |
1868 | error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; | |
1869 | if (error) { | |
1870 | VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); | |
1871 | } | |
1872 | return error; | |
1873 | } | |
1874 | ||
f1acd62b BP |
1875 | static int |
1876 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
1877 | char **netdev_name) | |
1878 | { | |
1879 | static const char fn[] = "/proc/net/route"; | |
1880 | FILE *stream; | |
1881 | char line[256]; | |
1882 | int ln; | |
1883 | ||
1884 | *netdev_name = NULL; | |
1885 | stream = fopen(fn, "r"); | |
1886 | if (stream == NULL) { | |
1887 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
1888 | return errno; | |
1889 | } | |
1890 | ||
1891 | ln = 0; | |
1892 | while (fgets(line, sizeof line, stream)) { | |
1893 | if (++ln >= 2) { | |
1894 | char iface[17]; | |
1895 | uint32_t dest, gateway, mask; | |
1896 | int refcnt, metric, mtu; | |
1897 | unsigned int flags, use, window, irtt; | |
1898 | ||
1899 | if (sscanf(line, | |
1900 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
1901 | " %d %u %u\n", | |
1902 | iface, &dest, &gateway, &flags, &refcnt, | |
1903 | &use, &metric, &mask, &mtu, &window, &irtt) != 11) { | |
1904 | ||
d295e8e9 | 1905 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
1906 | fn, ln, line); |
1907 | continue; | |
1908 | } | |
1909 | if (!(flags & RTF_UP)) { | |
1910 | /* Skip routes that aren't up. */ | |
1911 | continue; | |
1912 | } | |
1913 | ||
1914 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 1915 | * network byte order, so we don't need need any endian |
f1acd62b BP |
1916 | * conversions here. */ |
1917 | if ((dest & mask) == (host->s_addr & mask)) { | |
1918 | if (!gateway) { | |
1919 | /* The host is directly reachable. */ | |
1920 | next_hop->s_addr = 0; | |
1921 | } else { | |
1922 | /* To reach the host, we must go through a gateway. */ | |
1923 | next_hop->s_addr = gateway; | |
1924 | } | |
1925 | *netdev_name = xstrdup(iface); | |
1926 | fclose(stream); | |
1927 | return 0; | |
1928 | } | |
1929 | } | |
1930 | } | |
1931 | ||
1932 | fclose(stream); | |
1933 | return ENXIO; | |
1934 | } | |
1935 | ||
8b61709d BP |
1936 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
1937 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
1938 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
1939 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
1940 | static int | |
1941 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
1942 | uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) | |
1943 | { | |
1944 | struct arpreq r; | |
c100e025 | 1945 | struct sockaddr_in sin; |
8b61709d BP |
1946 | int retval; |
1947 | ||
1948 | memset(&r, 0, sizeof r); | |
c100e025 BP |
1949 | sin.sin_family = AF_INET; |
1950 | sin.sin_addr.s_addr = ip; | |
1951 | sin.sin_port = 0; | |
1952 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
1953 | r.arp_ha.sa_family = ARPHRD_ETHER; |
1954 | r.arp_flags = 0; | |
149f577a | 1955 | strncpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d BP |
1956 | COVERAGE_INC(netdev_arp_lookup); |
1957 | retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; | |
1958 | if (!retval) { | |
1959 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
1960 | } else if (retval != ENXIO) { | |
1961 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
149f577a | 1962 | netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval)); |
8b61709d BP |
1963 | } |
1964 | return retval; | |
1965 | } | |
1966 | ||
1967 | static int | |
1968 | nd_to_iff_flags(enum netdev_flags nd) | |
1969 | { | |
1970 | int iff = 0; | |
1971 | if (nd & NETDEV_UP) { | |
1972 | iff |= IFF_UP; | |
1973 | } | |
1974 | if (nd & NETDEV_PROMISC) { | |
1975 | iff |= IFF_PROMISC; | |
1976 | } | |
1977 | return iff; | |
1978 | } | |
1979 | ||
1980 | static int | |
1981 | iff_to_nd_flags(int iff) | |
1982 | { | |
1983 | enum netdev_flags nd = 0; | |
1984 | if (iff & IFF_UP) { | |
1985 | nd |= NETDEV_UP; | |
1986 | } | |
1987 | if (iff & IFF_PROMISC) { | |
1988 | nd |= NETDEV_PROMISC; | |
1989 | } | |
1990 | return nd; | |
1991 | } | |
1992 | ||
1993 | static int | |
1994 | netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, | |
1995 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
1996 | { | |
1997 | int old_flags, new_flags; | |
1998 | int error; | |
1999 | ||
2000 | error = get_flags(netdev, &old_flags); | |
2001 | if (!error) { | |
2002 | *old_flagsp = iff_to_nd_flags(old_flags); | |
2003 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2004 | if (new_flags != old_flags) { | |
2005 | error = set_flags(netdev, new_flags); | |
2006 | } | |
2007 | } | |
2008 | return error; | |
2009 | } | |
2010 | ||
2011 | static void | |
2012 | poll_notify(struct list *list) | |
2013 | { | |
2014 | struct netdev_linux_notifier *notifier; | |
4e8e4213 | 2015 | LIST_FOR_EACH (notifier, node, list) { |
8b61709d BP |
2016 | struct netdev_notifier *n = ¬ifier->notifier; |
2017 | n->cb(n); | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | static void | |
46097491 | 2022 | netdev_linux_poll_cb(const struct rtnetlink_change *change, |
67a4917b | 2023 | void *aux OVS_UNUSED) |
8b61709d BP |
2024 | { |
2025 | if (change) { | |
2026 | struct list *list = shash_find_data(&netdev_linux_notifiers, | |
2027 | change->ifname); | |
2028 | if (list) { | |
2029 | poll_notify(list); | |
2030 | } | |
2031 | } else { | |
2032 | struct shash_node *node; | |
2033 | SHASH_FOR_EACH (node, &netdev_linux_notifiers) { | |
2034 | poll_notify(node->data); | |
2035 | } | |
2036 | } | |
2037 | } | |
2038 | ||
2039 | static int | |
2040 | netdev_linux_poll_add(struct netdev *netdev, | |
2041 | void (*cb)(struct netdev_notifier *), void *aux, | |
2042 | struct netdev_notifier **notifierp) | |
2043 | { | |
2044 | const char *netdev_name = netdev_get_name(netdev); | |
2045 | struct netdev_linux_notifier *notifier; | |
2046 | struct list *list; | |
2047 | ||
2048 | if (shash_is_empty(&netdev_linux_notifiers)) { | |
46097491 | 2049 | int error = rtnetlink_notifier_register(&netdev_linux_poll_notifier, |
8b61709d BP |
2050 | netdev_linux_poll_cb, NULL); |
2051 | if (error) { | |
2052 | return error; | |
2053 | } | |
2054 | } | |
2055 | ||
2056 | list = shash_find_data(&netdev_linux_notifiers, netdev_name); | |
2057 | if (!list) { | |
2058 | list = xmalloc(sizeof *list); | |
2059 | list_init(list); | |
2060 | shash_add(&netdev_linux_notifiers, netdev_name, list); | |
2061 | } | |
2062 | ||
2063 | notifier = xmalloc(sizeof *notifier); | |
2064 | netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); | |
2065 | list_push_back(list, ¬ifier->node); | |
2066 | *notifierp = ¬ifier->notifier; | |
2067 | return 0; | |
2068 | } | |
2069 | ||
2070 | static void | |
2071 | netdev_linux_poll_remove(struct netdev_notifier *notifier_) | |
2072 | { | |
2073 | struct netdev_linux_notifier *notifier = | |
2074 | CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); | |
2075 | struct list *list; | |
2076 | ||
2077 | /* Remove 'notifier' from its list. */ | |
2078 | list = list_remove(¬ifier->node); | |
2079 | if (list_is_empty(list)) { | |
2080 | /* The list is now empty. Remove it from the hash and free it. */ | |
2081 | const char *netdev_name = netdev_get_name(notifier->notifier.netdev); | |
2082 | shash_delete(&netdev_linux_notifiers, | |
2083 | shash_find(&netdev_linux_notifiers, netdev_name)); | |
2084 | free(list); | |
2085 | } | |
2086 | free(notifier); | |
2087 | ||
2088 | /* If that was the last notifier, unregister. */ | |
2089 | if (shash_is_empty(&netdev_linux_notifiers)) { | |
46097491 | 2090 | rtnetlink_notifier_unregister(&netdev_linux_poll_notifier); |
8b61709d BP |
2091 | } |
2092 | } | |
2093 | ||
2094 | const struct netdev_class netdev_linux_class = { | |
149f577a | 2095 | "system", |
8b61709d BP |
2096 | |
2097 | netdev_linux_init, | |
2098 | netdev_linux_run, | |
2099 | netdev_linux_wait, | |
2100 | ||
a740f0de | 2101 | netdev_linux_create_system, |
6c88d577 JP |
2102 | netdev_linux_destroy, |
2103 | NULL, /* reconfigure */ | |
2104 | ||
8b61709d BP |
2105 | netdev_linux_open, |
2106 | netdev_linux_close, | |
2107 | ||
2108 | netdev_linux_enumerate, | |
2109 | ||
2110 | netdev_linux_recv, | |
2111 | netdev_linux_recv_wait, | |
2112 | netdev_linux_drain, | |
2113 | ||
2114 | netdev_linux_send, | |
2115 | netdev_linux_send_wait, | |
2116 | ||
2117 | netdev_linux_set_etheraddr, | |
2118 | netdev_linux_get_etheraddr, | |
2119 | netdev_linux_get_mtu, | |
9ab3d9a3 | 2120 | netdev_linux_get_ifindex, |
8b61709d BP |
2121 | netdev_linux_get_carrier, |
2122 | netdev_linux_get_stats, | |
f4b6076a | 2123 | netdev_vport_set_stats, |
8b61709d BP |
2124 | |
2125 | netdev_linux_get_features, | |
2126 | netdev_linux_set_advertisements, | |
2127 | netdev_linux_get_vlan_vid, | |
c1c9c9c4 | 2128 | |
8b61709d | 2129 | netdev_linux_set_policing, |
c1c9c9c4 BP |
2130 | netdev_linux_get_qos_types, |
2131 | netdev_linux_get_qos_capabilities, | |
2132 | netdev_linux_get_qos, | |
2133 | netdev_linux_set_qos, | |
2134 | netdev_linux_get_queue, | |
2135 | netdev_linux_set_queue, | |
2136 | netdev_linux_delete_queue, | |
2137 | netdev_linux_get_queue_stats, | |
2138 | netdev_linux_dump_queues, | |
2139 | netdev_linux_dump_queue_stats, | |
8b61709d BP |
2140 | |
2141 | netdev_linux_get_in4, | |
2142 | netdev_linux_set_in4, | |
2143 | netdev_linux_get_in6, | |
2144 | netdev_linux_add_router, | |
f1acd62b | 2145 | netdev_linux_get_next_hop, |
8b61709d BP |
2146 | netdev_linux_arp_lookup, |
2147 | ||
2148 | netdev_linux_update_flags, | |
2149 | ||
2150 | netdev_linux_poll_add, | |
2151 | netdev_linux_poll_remove, | |
2152 | }; | |
2153 | ||
2154 | const struct netdev_class netdev_tap_class = { | |
149f577a | 2155 | "tap", |
8b61709d BP |
2156 | |
2157 | netdev_linux_init, | |
149f577a JG |
2158 | netdev_linux_run, |
2159 | netdev_linux_wait, | |
8b61709d | 2160 | |
a740f0de | 2161 | netdev_linux_create_tap, |
6c88d577 JP |
2162 | netdev_linux_destroy, |
2163 | NULL, /* reconfigure */ | |
2164 | ||
8b61709d BP |
2165 | netdev_linux_open, |
2166 | netdev_linux_close, | |
2167 | ||
149f577a | 2168 | NULL, /* enumerate */ |
8b61709d BP |
2169 | |
2170 | netdev_linux_recv, | |
2171 | netdev_linux_recv_wait, | |
2172 | netdev_linux_drain, | |
2173 | ||
2174 | netdev_linux_send, | |
2175 | netdev_linux_send_wait, | |
2176 | ||
2177 | netdev_linux_set_etheraddr, | |
2178 | netdev_linux_get_etheraddr, | |
2179 | netdev_linux_get_mtu, | |
9ab3d9a3 | 2180 | netdev_linux_get_ifindex, |
8b61709d BP |
2181 | netdev_linux_get_carrier, |
2182 | netdev_linux_get_stats, | |
8722022c | 2183 | NULL, /* set_stats */ |
8b61709d BP |
2184 | |
2185 | netdev_linux_get_features, | |
2186 | netdev_linux_set_advertisements, | |
a740f0de | 2187 | netdev_linux_get_vlan_vid, |
c1c9c9c4 | 2188 | |
a740f0de | 2189 | netdev_linux_set_policing, |
c1c9c9c4 BP |
2190 | netdev_linux_get_qos_types, |
2191 | netdev_linux_get_qos_capabilities, | |
2192 | netdev_linux_get_qos, | |
2193 | netdev_linux_set_qos, | |
2194 | netdev_linux_get_queue, | |
2195 | netdev_linux_set_queue, | |
2196 | netdev_linux_delete_queue, | |
2197 | netdev_linux_get_queue_stats, | |
2198 | netdev_linux_dump_queues, | |
2199 | netdev_linux_dump_queue_stats, | |
a740f0de JG |
2200 | |
2201 | netdev_linux_get_in4, | |
2202 | netdev_linux_set_in4, | |
2203 | netdev_linux_get_in6, | |
2204 | netdev_linux_add_router, | |
2205 | netdev_linux_get_next_hop, | |
2206 | netdev_linux_arp_lookup, | |
2207 | ||
2208 | netdev_linux_update_flags, | |
2209 | ||
2210 | netdev_linux_poll_add, | |
2211 | netdev_linux_poll_remove, | |
2212 | }; | |
8b61709d | 2213 | \f |
c1c9c9c4 | 2214 | /* HTB traffic control class. */ |
559843ed | 2215 | |
c1c9c9c4 | 2216 | #define HTB_N_QUEUES 0xf000 |
8b61709d | 2217 | |
c1c9c9c4 BP |
2218 | struct htb { |
2219 | struct tc tc; | |
2220 | unsigned int max_rate; /* In bytes/s. */ | |
2221 | }; | |
8b61709d | 2222 | |
c1c9c9c4 | 2223 | struct htb_class { |
93b13be8 | 2224 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
2225 | unsigned int min_rate; /* In bytes/s. */ |
2226 | unsigned int max_rate; /* In bytes/s. */ | |
2227 | unsigned int burst; /* In bytes. */ | |
2228 | unsigned int priority; /* Lower values are higher priorities. */ | |
2229 | }; | |
8b61709d | 2230 | |
c1c9c9c4 BP |
2231 | static struct htb * |
2232 | htb_get__(const struct netdev *netdev) | |
2233 | { | |
2234 | struct netdev_dev_linux *netdev_dev = | |
2235 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2236 | return CONTAINER_OF(netdev_dev->tc, struct htb, tc); | |
2237 | } | |
2238 | ||
2239 | static struct htb * | |
2240 | htb_install__(struct netdev *netdev, uint64_t max_rate) | |
2241 | { | |
2242 | struct netdev_dev_linux *netdev_dev = | |
2243 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2244 | struct htb *htb; | |
2245 | ||
2246 | htb = xmalloc(sizeof *htb); | |
2247 | tc_init(&htb->tc, &tc_ops_htb); | |
2248 | htb->max_rate = max_rate; | |
2249 | ||
2250 | netdev_dev->tc = &htb->tc; | |
2251 | ||
2252 | return htb; | |
2253 | } | |
2254 | ||
2255 | /* Create an HTB qdisc. | |
2256 | * | |
2257 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default | |
2258 | * 0". */ | |
2259 | static int | |
2260 | htb_setup_qdisc__(struct netdev *netdev) | |
2261 | { | |
2262 | size_t opt_offset; | |
2263 | struct tc_htb_glob opt; | |
2264 | struct ofpbuf request; | |
2265 | struct tcmsg *tcmsg; | |
2266 | ||
2267 | tc_del_qdisc(netdev); | |
2268 | ||
2269 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
2270 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
2271 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
2272 | tcmsg->tcm_parent = TC_H_ROOT; | |
2273 | ||
2274 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2275 | ||
2276 | memset(&opt, 0, sizeof opt); | |
2277 | opt.rate2quantum = 10; | |
2278 | opt.version = 3; | |
2279 | opt.defcls = 0; | |
2280 | ||
2281 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2282 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
2283 | nl_msg_end_nested(&request, opt_offset); | |
2284 | ||
2285 | return tc_transact(&request, NULL); | |
2286 | } | |
2287 | ||
2288 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
2289 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
2290 | static int | |
2291 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
2292 | unsigned int parent, struct htb_class *class) | |
2293 | { | |
2294 | size_t opt_offset; | |
2295 | struct tc_htb_opt opt; | |
2296 | struct ofpbuf request; | |
2297 | struct tcmsg *tcmsg; | |
2298 | int error; | |
2299 | int mtu; | |
2300 | ||
2301 | netdev_get_mtu(netdev, &mtu); | |
2302 | ||
2303 | memset(&opt, 0, sizeof opt); | |
2304 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
2305 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
2306 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); | |
2307 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
2308 | opt.prio = class->priority; | |
2309 | ||
2310 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
2311 | tcmsg->tcm_handle = handle; | |
2312 | tcmsg->tcm_parent = parent; | |
2313 | ||
2314 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2315 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2316 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
2317 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
2318 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
2319 | nl_msg_end_nested(&request, opt_offset); | |
2320 | ||
2321 | error = tc_transact(&request, NULL); | |
2322 | if (error) { | |
2323 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
2324 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
2325 | netdev_get_name(netdev), | |
2326 | tc_get_major(handle), tc_get_minor(handle), | |
2327 | tc_get_major(parent), tc_get_minor(parent), | |
2328 | class->min_rate, class->max_rate, | |
2329 | class->burst, class->priority, strerror(error)); | |
2330 | } | |
2331 | return error; | |
2332 | } | |
2333 | ||
2334 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
2335 | * description of them into 'details'. The description complies with the | |
2336 | * specification given in the vswitch database documentation for linux-htb | |
2337 | * queue details. */ | |
2338 | static int | |
2339 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
2340 | { | |
2341 | static const struct nl_policy tca_htb_policy[] = { | |
2342 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
2343 | .min_len = sizeof(struct tc_htb_opt) }, | |
2344 | }; | |
2345 | ||
2346 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
2347 | const struct tc_htb_opt *htb; | |
2348 | ||
2349 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
2350 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
2351 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
2352 | return EPROTO; | |
2353 | } | |
2354 | ||
2355 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
2356 | class->min_rate = htb->rate.rate; | |
2357 | class->max_rate = htb->ceil.rate; | |
2358 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
2359 | class->priority = htb->prio; | |
2360 | return 0; | |
2361 | } | |
2362 | ||
2363 | static int | |
2364 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
2365 | struct htb_class *options, | |
2366 | struct netdev_queue_stats *stats) | |
2367 | { | |
2368 | struct nlattr *nl_options; | |
2369 | unsigned int handle; | |
2370 | int error; | |
2371 | ||
2372 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
2373 | if (!error && queue_id) { | |
17ee3c1f BP |
2374 | unsigned int major = tc_get_major(handle); |
2375 | unsigned int minor = tc_get_minor(handle); | |
2376 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
2377 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
2378 | } else { |
2379 | error = EPROTO; | |
2380 | } | |
2381 | } | |
2382 | if (!error && options) { | |
2383 | error = htb_parse_tca_options__(nl_options, options); | |
2384 | } | |
2385 | return error; | |
2386 | } | |
2387 | ||
2388 | static void | |
2389 | htb_parse_qdisc_details__(struct netdev *netdev, | |
2390 | const struct shash *details, struct htb_class *hc) | |
2391 | { | |
2392 | const char *max_rate_s; | |
2393 | ||
2394 | max_rate_s = shash_find_data(details, "max-rate"); | |
2395 | hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; | |
2396 | if (!hc->max_rate) { | |
2397 | uint32_t current; | |
2398 | ||
2399 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
2400 | hc->max_rate = netdev_features_to_bps(current) / 8; | |
2401 | } | |
2402 | hc->min_rate = hc->max_rate; | |
2403 | hc->burst = 0; | |
2404 | hc->priority = 0; | |
2405 | } | |
2406 | ||
2407 | static int | |
2408 | htb_parse_class_details__(struct netdev *netdev, | |
2409 | const struct shash *details, struct htb_class *hc) | |
2410 | { | |
2411 | const struct htb *htb = htb_get__(netdev); | |
2412 | const char *min_rate_s = shash_find_data(details, "min-rate"); | |
2413 | const char *max_rate_s = shash_find_data(details, "max-rate"); | |
2414 | const char *burst_s = shash_find_data(details, "burst"); | |
2415 | const char *priority_s = shash_find_data(details, "priority"); | |
2416 | int mtu; | |
2417 | ||
2418 | /* min-rate */ | |
2419 | if (!min_rate_s) { | |
2420 | /* min-rate is required. */ | |
2421 | return EINVAL; | |
2422 | } | |
2423 | hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8; | |
2424 | hc->min_rate = MAX(hc->min_rate, 0); | |
2425 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); | |
2426 | ||
2427 | /* max-rate */ | |
2428 | hc->max_rate = (max_rate_s | |
2429 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
2430 | : htb->max_rate); | |
2431 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); | |
2432 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
2433 | ||
2434 | /* burst | |
2435 | * | |
2436 | * According to hints in the documentation that I've read, it is important | |
2437 | * that 'burst' be at least as big as the largest frame that might be | |
2438 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
2439 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
2440 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
2441 | * the MTU. We actually add 64, instead of 14, as a guard against | |
2442 | * additional headers get tacked on somewhere that we're not aware of. */ | |
2443 | netdev_get_mtu(netdev, &mtu); | |
2444 | hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; | |
2445 | hc->burst = MAX(hc->burst, mtu + 64); | |
2446 | ||
2447 | /* priority */ | |
2448 | hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; | |
2449 | ||
2450 | return 0; | |
2451 | } | |
2452 | ||
2453 | static int | |
2454 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
2455 | unsigned int parent, struct htb_class *options, | |
2456 | struct netdev_queue_stats *stats) | |
2457 | { | |
2458 | struct ofpbuf *reply; | |
2459 | int error; | |
2460 | ||
2461 | error = tc_query_class(netdev, handle, parent, &reply); | |
2462 | if (!error) { | |
2463 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
2464 | ofpbuf_delete(reply); | |
2465 | } | |
2466 | return error; | |
2467 | } | |
2468 | ||
2469 | static int | |
2470 | htb_tc_install(struct netdev *netdev, const struct shash *details) | |
2471 | { | |
2472 | int error; | |
2473 | ||
2474 | error = htb_setup_qdisc__(netdev); | |
2475 | if (!error) { | |
2476 | struct htb_class hc; | |
2477 | ||
2478 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2479 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2480 | tc_make_handle(1, 0), &hc); | |
2481 | if (!error) { | |
2482 | htb_install__(netdev, hc.max_rate); | |
2483 | } | |
2484 | } | |
2485 | return error; | |
2486 | } | |
2487 | ||
93b13be8 BP |
2488 | static struct htb_class * |
2489 | htb_class_cast__(const struct tc_queue *queue) | |
2490 | { | |
2491 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
2492 | } | |
2493 | ||
c1c9c9c4 BP |
2494 | static void |
2495 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
2496 | const struct htb_class *hc) | |
2497 | { | |
2498 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
2499 | size_t hash = hash_int(queue_id, 0); |
2500 | struct tc_queue *queue; | |
c1c9c9c4 BP |
2501 | struct htb_class *hcp; |
2502 | ||
93b13be8 BP |
2503 | queue = tc_find_queue__(netdev, queue_id, hash); |
2504 | if (queue) { | |
2505 | hcp = htb_class_cast__(queue); | |
2506 | } else { | |
c1c9c9c4 | 2507 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
2508 | queue = &hcp->tc_queue; |
2509 | queue->queue_id = queue_id; | |
2510 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); | |
c1c9c9c4 | 2511 | } |
93b13be8 BP |
2512 | |
2513 | hcp->min_rate = hc->min_rate; | |
2514 | hcp->max_rate = hc->max_rate; | |
2515 | hcp->burst = hc->burst; | |
2516 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
2517 | } |
2518 | ||
2519 | static int | |
2520 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2521 | { | |
2522 | struct shash details = SHASH_INITIALIZER(&details); | |
2523 | struct ofpbuf msg; | |
2524 | struct nl_dump dump; | |
2525 | struct htb_class hc; | |
2526 | struct htb *htb; | |
2527 | ||
2528 | /* Get qdisc options. */ | |
2529 | hc.max_rate = 0; | |
2530 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
2531 | htb = htb_install__(netdev, hc.max_rate); | |
2532 | ||
2533 | /* Get queues. */ | |
2534 | start_queue_dump(netdev, &dump); | |
2535 | shash_init(&details); | |
2536 | while (nl_dump_next(&dump, &msg)) { | |
2537 | unsigned int queue_id; | |
2538 | ||
2539 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
2540 | htb_update_queue__(netdev, queue_id, &hc); | |
2541 | } | |
2542 | } | |
2543 | nl_dump_done(&dump); | |
2544 | ||
2545 | return 0; | |
2546 | } | |
2547 | ||
2548 | static void | |
2549 | htb_tc_destroy(struct tc *tc) | |
2550 | { | |
2551 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
93b13be8 | 2552 | struct htb_class *hc, *next; |
c1c9c9c4 | 2553 | |
4e8e4213 | 2554 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { |
93b13be8 | 2555 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 BP |
2556 | free(hc); |
2557 | } | |
2558 | tc_destroy(tc); | |
2559 | free(htb); | |
2560 | } | |
2561 | ||
2562 | static int | |
2563 | htb_qdisc_get(const struct netdev *netdev, struct shash *details) | |
2564 | { | |
2565 | const struct htb *htb = htb_get__(netdev); | |
2566 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate)); | |
2567 | return 0; | |
2568 | } | |
2569 | ||
2570 | static int | |
2571 | htb_qdisc_set(struct netdev *netdev, const struct shash *details) | |
2572 | { | |
2573 | struct htb_class hc; | |
2574 | int error; | |
2575 | ||
2576 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2577 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2578 | tc_make_handle(1, 0), &hc); | |
2579 | if (!error) { | |
2580 | htb_get__(netdev)->max_rate = hc.max_rate; | |
2581 | } | |
2582 | return error; | |
2583 | } | |
2584 | ||
2585 | static int | |
93b13be8 BP |
2586 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
2587 | const struct tc_queue *queue, struct shash *details) | |
c1c9c9c4 | 2588 | { |
93b13be8 | 2589 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 BP |
2590 | |
2591 | shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); | |
2592 | if (hc->min_rate != hc->max_rate) { | |
2593 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); | |
2594 | } | |
2595 | shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst)); | |
2596 | if (hc->priority) { | |
2597 | shash_add(details, "priority", xasprintf("%u", hc->priority)); | |
2598 | } | |
2599 | return 0; | |
2600 | } | |
2601 | ||
2602 | static int | |
2603 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
2604 | const struct shash *details) | |
2605 | { | |
2606 | struct htb_class hc; | |
2607 | int error; | |
2608 | ||
2609 | error = htb_parse_class_details__(netdev, details, &hc); | |
2610 | if (error) { | |
2611 | return error; | |
2612 | } | |
2613 | ||
17ee3c1f | 2614 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
2615 | tc_make_handle(1, 0xfffe), &hc); |
2616 | if (error) { | |
2617 | return error; | |
2618 | } | |
2619 | ||
2620 | htb_update_queue__(netdev, queue_id, &hc); | |
2621 | return 0; | |
2622 | } | |
2623 | ||
2624 | static int | |
93b13be8 | 2625 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 2626 | { |
93b13be8 | 2627 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2628 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
2629 | int error; |
2630 | ||
93b13be8 | 2631 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 2632 | if (!error) { |
93b13be8 | 2633 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 2634 | free(hc); |
c1c9c9c4 BP |
2635 | } |
2636 | return error; | |
2637 | } | |
2638 | ||
2639 | static int | |
93b13be8 | 2640 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
2641 | struct netdev_queue_stats *stats) |
2642 | { | |
93b13be8 | 2643 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
2644 | tc_make_handle(1, 0xfffe), NULL, stats); |
2645 | } | |
2646 | ||
2647 | static int | |
2648 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
2649 | const struct ofpbuf *nlmsg, | |
2650 | netdev_dump_queue_stats_cb *cb, void *aux) | |
2651 | { | |
2652 | struct netdev_queue_stats stats; | |
17ee3c1f | 2653 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
2654 | int error; |
2655 | ||
2656 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
2657 | if (error) { | |
2658 | return error; | |
2659 | } | |
2660 | ||
17ee3c1f BP |
2661 | major = tc_get_major(handle); |
2662 | minor = tc_get_minor(handle); | |
2663 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 2664 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
2665 | } |
2666 | return 0; | |
2667 | } | |
2668 | ||
2669 | static const struct tc_ops tc_ops_htb = { | |
2670 | "htb", /* linux_name */ | |
2671 | "linux-htb", /* ovs_name */ | |
2672 | HTB_N_QUEUES, /* n_queues */ | |
2673 | htb_tc_install, | |
2674 | htb_tc_load, | |
2675 | htb_tc_destroy, | |
2676 | htb_qdisc_get, | |
2677 | htb_qdisc_set, | |
2678 | htb_class_get, | |
2679 | htb_class_set, | |
2680 | htb_class_delete, | |
2681 | htb_class_get_stats, | |
2682 | htb_class_dump_stats | |
2683 | }; | |
2684 | \f | |
2685 | /* "linux-default" traffic control class. | |
2686 | * | |
2687 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
2688 | * the "" (empty string) QoS type in the OVS database. */ | |
2689 | ||
2690 | static void | |
2691 | default_install__(struct netdev *netdev) | |
2692 | { | |
2693 | struct netdev_dev_linux *netdev_dev = | |
2694 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2695 | static struct tc *tc; | |
2696 | ||
2697 | if (!tc) { | |
2698 | tc = xmalloc(sizeof *tc); | |
2699 | tc_init(tc, &tc_ops_default); | |
2700 | } | |
2701 | netdev_dev->tc = tc; | |
2702 | } | |
2703 | ||
2704 | static int | |
2705 | default_tc_install(struct netdev *netdev, | |
2706 | const struct shash *details OVS_UNUSED) | |
2707 | { | |
2708 | default_install__(netdev); | |
2709 | return 0; | |
2710 | } | |
2711 | ||
2712 | static int | |
2713 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2714 | { | |
2715 | default_install__(netdev); | |
2716 | return 0; | |
2717 | } | |
2718 | ||
2719 | static const struct tc_ops tc_ops_default = { | |
2720 | NULL, /* linux_name */ | |
2721 | "", /* ovs_name */ | |
2722 | 0, /* n_queues */ | |
2723 | default_tc_install, | |
2724 | default_tc_load, | |
2725 | NULL, /* tc_destroy */ | |
2726 | NULL, /* qdisc_get */ | |
2727 | NULL, /* qdisc_set */ | |
2728 | NULL, /* class_get */ | |
2729 | NULL, /* class_set */ | |
2730 | NULL, /* class_delete */ | |
2731 | NULL, /* class_get_stats */ | |
2732 | NULL /* class_dump_stats */ | |
2733 | }; | |
2734 | \f | |
2735 | /* "linux-other" traffic control class. | |
2736 | * | |
2737 | * */ | |
2738 | ||
2739 | static int | |
2740 | other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2741 | { | |
2742 | struct netdev_dev_linux *netdev_dev = | |
2743 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2744 | static struct tc *tc; | |
2745 | ||
2746 | if (!tc) { | |
2747 | tc = xmalloc(sizeof *tc); | |
2748 | tc_init(tc, &tc_ops_other); | |
2749 | } | |
2750 | netdev_dev->tc = tc; | |
2751 | return 0; | |
2752 | } | |
2753 | ||
2754 | static const struct tc_ops tc_ops_other = { | |
2755 | NULL, /* linux_name */ | |
2756 | "linux-other", /* ovs_name */ | |
2757 | 0, /* n_queues */ | |
2758 | NULL, /* tc_install */ | |
2759 | other_tc_load, | |
2760 | NULL, /* tc_destroy */ | |
2761 | NULL, /* qdisc_get */ | |
2762 | NULL, /* qdisc_set */ | |
2763 | NULL, /* class_get */ | |
2764 | NULL, /* class_set */ | |
2765 | NULL, /* class_delete */ | |
2766 | NULL, /* class_get_stats */ | |
2767 | NULL /* class_dump_stats */ | |
2768 | }; | |
2769 | \f | |
2770 | /* Traffic control. */ | |
2771 | ||
2772 | /* Number of kernel "tc" ticks per second. */ | |
2773 | static double ticks_per_s; | |
2774 | ||
2775 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
2776 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
2777 | * one jiffy's worth of data. | |
2778 | * | |
2779 | * There are two possibilities here: | |
2780 | * | |
2781 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
2782 | * approximate range of 100 to 1024. That means that we really need to | |
2783 | * make sure that the qdisc can buffer that much data. | |
2784 | * | |
2785 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
2786 | * has finely granular timers and there's no need to fudge additional room | |
2787 | * for buffers. (There's no extra effort needed to implement that: the | |
2788 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
2789 | * come out as 0 in the division. Small integer results in the case of | |
2790 | * really high dividends won't have any real effect anyhow.) | |
2791 | */ | |
2792 | static unsigned int buffer_hz; | |
2793 | ||
2794 | /* Returns tc handle 'major':'minor'. */ | |
2795 | static unsigned int | |
2796 | tc_make_handle(unsigned int major, unsigned int minor) | |
2797 | { | |
2798 | return TC_H_MAKE(major << 16, minor); | |
2799 | } | |
2800 | ||
2801 | /* Returns the major number from 'handle'. */ | |
2802 | static unsigned int | |
2803 | tc_get_major(unsigned int handle) | |
2804 | { | |
2805 | return TC_H_MAJ(handle) >> 16; | |
2806 | } | |
2807 | ||
2808 | /* Returns the minor number from 'handle'. */ | |
2809 | static unsigned int | |
2810 | tc_get_minor(unsigned int handle) | |
2811 | { | |
2812 | return TC_H_MIN(handle); | |
2813 | } | |
2814 | ||
2815 | static struct tcmsg * | |
2816 | tc_make_request(const struct netdev *netdev, int type, unsigned int flags, | |
2817 | struct ofpbuf *request) | |
2818 | { | |
2819 | struct tcmsg *tcmsg; | |
2820 | int ifindex; | |
2821 | int error; | |
2822 | ||
2823 | error = get_ifindex(netdev, &ifindex); | |
2824 | if (error) { | |
2825 | return NULL; | |
2826 | } | |
2827 | ||
2828 | ofpbuf_init(request, 512); | |
2829 | nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); | |
2830 | tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); | |
2831 | tcmsg->tcm_family = AF_UNSPEC; | |
2832 | tcmsg->tcm_ifindex = ifindex; | |
2833 | /* Caller should fill in tcmsg->tcm_handle. */ | |
2834 | /* Caller should fill in tcmsg->tcm_parent. */ | |
2835 | ||
2836 | return tcmsg; | |
2837 | } | |
2838 | ||
2839 | static int | |
2840 | tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) | |
2841 | { | |
2842 | int error = nl_sock_transact(rtnl_sock, request, replyp); | |
2843 | ofpbuf_uninit(request); | |
2844 | return error; | |
2845 | } | |
2846 | ||
2847 | static void | |
2848 | read_psched(void) | |
2849 | { | |
2850 | /* The values in psched are not individually very meaningful, but they are | |
2851 | * important. The tables below show some values seen in the wild. | |
2852 | * | |
2853 | * Some notes: | |
2854 | * | |
2855 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
2856 | * (Before that, there are hints that it was 1000000000.) | |
2857 | * | |
2858 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
2859 | * above. | |
2860 | * | |
2861 | * /proc/net/psched | |
2862 | * ----------------------------------- | |
2863 | * [1] 000c8000 000f4240 000f4240 00000064 | |
2864 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
2865 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
2866 | * [4] 000003e8 00000400 000f4240 00000064 | |
2867 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
2868 | * [6] 000003e8 00000040 000f4240 000000f9 | |
2869 | * | |
2870 | * a b c d ticks_per_s buffer_hz | |
2871 | * ------- --------- ---------- ------------- ----------- ------------- | |
2872 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
2873 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
2874 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
2875 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
2876 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
2877 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
2878 | * | |
2879 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
2880 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
2881 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
2882 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
2883 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
2884 | * [6] 2.6.34 from kernel.org on KVM | |
2885 | */ | |
2886 | static const char fn[] = "/proc/net/psched"; | |
2887 | unsigned int a, b, c, d; | |
2888 | FILE *stream; | |
2889 | ||
2890 | ticks_per_s = 1.0; | |
2891 | buffer_hz = 100; | |
2892 | ||
2893 | stream = fopen(fn, "r"); | |
2894 | if (!stream) { | |
2895 | VLOG_WARN("%s: open failed: %s", fn, strerror(errno)); | |
2896 | return; | |
2897 | } | |
2898 | ||
2899 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
2900 | VLOG_WARN("%s: read failed", fn); | |
2901 | fclose(stream); | |
2902 | return; | |
2903 | } | |
2904 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
2905 | fclose(stream); | |
2906 | ||
2907 | if (!a || !c) { | |
2908 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
2909 | return; | |
2910 | } | |
2911 | ||
2912 | ticks_per_s = (double) a * c / b; | |
2913 | if (c == 1000000) { | |
2914 | buffer_hz = d; | |
2915 | } else { | |
2916 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
2917 | fn, a, b, c, d); | |
2918 | } | |
2919 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
2920 | } | |
2921 | ||
2922 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
2923 | * rate of 'rate' bytes per second. */ | |
2924 | static unsigned int | |
2925 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
2926 | { | |
2927 | if (!buffer_hz) { | |
2928 | read_psched(); | |
2929 | } | |
2930 | return (rate * ticks) / ticks_per_s; | |
2931 | } | |
2932 | ||
2933 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
2934 | * rate of 'rate' bytes per second. */ | |
2935 | static unsigned int | |
2936 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
2937 | { | |
2938 | if (!buffer_hz) { | |
2939 | read_psched(); | |
2940 | } | |
015c93a4 | 2941 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
2942 | } |
2943 | ||
2944 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
2945 | * a transmission rate of 'rate' bytes per second. */ | |
2946 | static unsigned int | |
2947 | tc_buffer_per_jiffy(unsigned int rate) | |
2948 | { | |
2949 | if (!buffer_hz) { | |
2950 | read_psched(); | |
2951 | } | |
2952 | return rate / buffer_hz; | |
2953 | } | |
2954 | ||
2955 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
2956 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
2957 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
2958 | * stores NULL into it if it is absent. | |
2959 | * | |
2960 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
2961 | * 'msg'. | |
2962 | * | |
2963 | * Returns 0 if successful, otherwise a positive errno value. */ | |
2964 | static int | |
2965 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
2966 | struct nlattr **options) | |
2967 | { | |
2968 | static const struct nl_policy tca_policy[] = { | |
2969 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
2970 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
2971 | }; | |
2972 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
2973 | ||
2974 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
2975 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
2976 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
2977 | goto error; | |
2978 | } | |
2979 | ||
2980 | if (kind) { | |
2981 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
2982 | } | |
2983 | ||
2984 | if (options) { | |
2985 | *options = ta[TCA_OPTIONS]; | |
2986 | } | |
2987 | ||
2988 | return 0; | |
2989 | ||
2990 | error: | |
2991 | if (kind) { | |
2992 | *kind = NULL; | |
2993 | } | |
2994 | if (options) { | |
2995 | *options = NULL; | |
2996 | } | |
2997 | return EPROTO; | |
2998 | } | |
2999 | ||
3000 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
3001 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
3002 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
3003 | * arguments may be null. | |
3004 | * | |
3005 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3006 | static int | |
3007 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
3008 | struct nlattr **options, struct netdev_queue_stats *stats) | |
3009 | { | |
3010 | static const struct nl_policy tca_policy[] = { | |
3011 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
3012 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
3013 | }; | |
3014 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3015 | ||
3016 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3017 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3018 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
3019 | goto error; | |
3020 | } | |
3021 | ||
3022 | if (handlep) { | |
3023 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
3024 | *handlep = tc->tcm_handle; | |
3025 | } | |
3026 | ||
3027 | if (options) { | |
3028 | *options = ta[TCA_OPTIONS]; | |
3029 | } | |
3030 | ||
3031 | if (stats) { | |
3032 | const struct gnet_stats_queue *gsq; | |
3033 | struct gnet_stats_basic gsb; | |
3034 | ||
3035 | static const struct nl_policy stats_policy[] = { | |
3036 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
3037 | .min_len = sizeof gsb }, | |
3038 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
3039 | .min_len = sizeof *gsq }, | |
3040 | }; | |
3041 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
3042 | ||
3043 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
3044 | sa, ARRAY_SIZE(sa))) { | |
3045 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
3046 | goto error; | |
3047 | } | |
3048 | ||
3049 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
3050 | * some arch/bitsize combinations. Newer versions of Linux have a | |
3051 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
3052 | * easiest thing to do is just to make a copy. */ | |
3053 | memset(&gsb, 0, sizeof gsb); | |
3054 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
3055 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
3056 | stats->tx_bytes = gsb.bytes; | |
3057 | stats->tx_packets = gsb.packets; | |
3058 | ||
3059 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
3060 | stats->tx_errors = gsq->drops; | |
3061 | } | |
3062 | ||
3063 | return 0; | |
3064 | ||
3065 | error: | |
3066 | if (options) { | |
3067 | *options = NULL; | |
3068 | } | |
3069 | if (stats) { | |
3070 | memset(stats, 0, sizeof *stats); | |
3071 | } | |
3072 | return EPROTO; | |
3073 | } | |
3074 | ||
3075 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
3076 | * on 'netdev'. */ | |
3077 | static int | |
3078 | tc_query_class(const struct netdev *netdev, | |
3079 | unsigned int handle, unsigned int parent, | |
3080 | struct ofpbuf **replyp) | |
3081 | { | |
3082 | struct ofpbuf request; | |
3083 | struct tcmsg *tcmsg; | |
3084 | int error; | |
3085 | ||
3086 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); | |
3087 | tcmsg->tcm_handle = handle; | |
3088 | tcmsg->tcm_parent = parent; | |
3089 | ||
3090 | error = tc_transact(&request, replyp); | |
3091 | if (error) { | |
3092 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
3093 | netdev_get_name(netdev), | |
3094 | tc_get_major(handle), tc_get_minor(handle), | |
3095 | tc_get_major(parent), tc_get_minor(parent), | |
3096 | strerror(error)); | |
3097 | } | |
3098 | return error; | |
3099 | } | |
3100 | ||
3101 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
3102 | static int | |
3103 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
3104 | { | |
3105 | struct ofpbuf request; | |
3106 | struct tcmsg *tcmsg; | |
3107 | int error; | |
3108 | ||
3109 | tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); | |
3110 | tcmsg->tcm_handle = handle; | |
3111 | tcmsg->tcm_parent = 0; | |
3112 | ||
3113 | error = tc_transact(&request, NULL); | |
3114 | if (error) { | |
3115 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
3116 | netdev_get_name(netdev), | |
3117 | tc_get_major(handle), tc_get_minor(handle), | |
3118 | strerror(error)); | |
3119 | } | |
3120 | return error; | |
3121 | } | |
3122 | ||
3123 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
3124 | static int | |
3125 | tc_del_qdisc(struct netdev *netdev) | |
3126 | { | |
3127 | struct netdev_dev_linux *netdev_dev = | |
3128 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3129 | struct ofpbuf request; | |
3130 | struct tcmsg *tcmsg; | |
3131 | int error; | |
3132 | ||
3133 | tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); | |
3134 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3135 | tcmsg->tcm_parent = TC_H_ROOT; | |
3136 | ||
3137 | error = tc_transact(&request, NULL); | |
3138 | if (error == EINVAL) { | |
3139 | /* EINVAL probably means that the default qdisc was in use, in which | |
3140 | * case we've accomplished our purpose. */ | |
3141 | error = 0; | |
3142 | } | |
3143 | if (!error && netdev_dev->tc) { | |
3144 | if (netdev_dev->tc->ops->tc_destroy) { | |
3145 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
3146 | } | |
3147 | netdev_dev->tc = NULL; | |
3148 | } | |
3149 | return error; | |
3150 | } | |
3151 | ||
3152 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the | |
3153 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
3154 | * positive errno value. */ | |
3155 | static int | |
3156 | tc_query_qdisc(const struct netdev *netdev) | |
3157 | { | |
3158 | struct netdev_dev_linux *netdev_dev = | |
3159 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3160 | struct ofpbuf request, *qdisc; | |
3161 | const struct tc_ops *ops; | |
3162 | struct tcmsg *tcmsg; | |
3163 | int load_error; | |
3164 | int error; | |
3165 | ||
3166 | if (netdev_dev->tc) { | |
3167 | return 0; | |
3168 | } | |
3169 | ||
3170 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
3171 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
3172 | * 2.6.35 without that fix backported to it. | |
3173 | * | |
3174 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
3175 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
3176 | * few others. There are a few ways that I can see to do this, but most of | |
3177 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
3178 | * technique chosen here is to assume that any non-default qdisc that we | |
3179 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
3180 | * a class with handle 0:0. | |
3181 | * | |
3182 | * We could check for Linux 2.6.35+ and use a more straightforward method | |
3183 | * there. */ | |
3184 | tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request); | |
3185 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3186 | tcmsg->tcm_parent = 0; | |
3187 | ||
3188 | /* Figure out what tc class to instantiate. */ | |
3189 | error = tc_transact(&request, &qdisc); | |
3190 | if (!error) { | |
3191 | const char *kind; | |
3192 | ||
3193 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
3194 | if (error) { | |
3195 | ops = &tc_ops_other; | |
3196 | } else { | |
3197 | ops = tc_lookup_linux_name(kind); | |
3198 | if (!ops) { | |
3199 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
3200 | VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind); | |
3201 | ||
3202 | ops = &tc_ops_other; | |
3203 | } | |
3204 | } | |
3205 | } else if (error == ENOENT) { | |
3206 | /* Either it's a built-in qdisc, or it's a qdisc set up by some | |
3207 | * other entity that doesn't have a handle 1:0. We will assume | |
3208 | * that it's the system default qdisc. */ | |
3209 | ops = &tc_ops_default; | |
3210 | error = 0; | |
3211 | } else { | |
3212 | /* Who knows? Maybe the device got deleted. */ | |
3213 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
3214 | netdev_get_name(netdev), strerror(error)); | |
3215 | ops = &tc_ops_other; | |
3216 | } | |
3217 | ||
3218 | /* Instantiate it. */ | |
3219 | load_error = ops->tc_load((struct netdev *) netdev, qdisc); | |
3220 | assert((load_error == 0) == (netdev_dev->tc != NULL)); | |
3221 | ofpbuf_delete(qdisc); | |
3222 | ||
3223 | return error ? error : load_error; | |
3224 | } | |
3225 | ||
3226 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
3227 | approximate the time to transmit packets of various lengths. For an MTU of | |
3228 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
3229 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
3230 | possible lengths; and so on. | |
3231 | ||
3232 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
3233 | need to be shifted right to fit within such a 256-entry table. */ | |
3234 | static int | |
3235 | tc_calc_cell_log(unsigned int mtu) | |
3236 | { | |
3237 | int cell_log; | |
3238 | ||
3239 | if (!mtu) { | |
3240 | mtu = ETH_PAYLOAD_MAX; | |
3241 | } | |
3242 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
3243 | ||
3244 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
3245 | mtu >>= 1; | |
3246 | } | |
3247 | ||
3248 | return cell_log; | |
3249 | } | |
3250 | ||
3251 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
3252 | * of 'mtu'. */ | |
3253 | static void | |
3254 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
3255 | { | |
3256 | memset(rate, 0, sizeof *rate); | |
3257 | rate->cell_log = tc_calc_cell_log(mtu); | |
3258 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
3259 | /* rate->cell_align = 0; */ /* distro headers. */ | |
3260 | rate->mpu = ETH_TOTAL_MIN; | |
3261 | rate->rate = Bps; | |
3262 | } | |
3263 | ||
3264 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
3265 | * attribute of the specified "type". | |
3266 | * | |
3267 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
3268 | static void | |
3269 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
3270 | { | |
3271 | uint32_t *rtab; | |
3272 | unsigned int i; | |
3273 | ||
3274 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
3275 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
3276 | unsigned packet_size = (i + 1) << rate->cell_log; | |
3277 | if (packet_size < rate->mpu) { | |
3278 | packet_size = rate->mpu; | |
3279 | } | |
3280 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
3281 | } | |
3282 | } | |
3283 | ||
3284 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
3285 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
3286 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 3287 | * 0 is fine.) */ |
c1c9c9c4 BP |
3288 | static int |
3289 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
3290 | { | |
3291 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
3292 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
3293 | } | |
3294 | ||
3295 | \f | |
3296 | /* Utility functions. */ | |
3297 | ||
3298 | static int | |
3299 | get_stats_via_netlink(int ifindex, struct netdev_stats *stats) | |
3300 | { | |
3301 | /* Policy for RTNLGRP_LINK messages. | |
3302 | * | |
3303 | * There are *many* more fields in these messages, but currently we only | |
3304 | * care about these fields. */ | |
3305 | static const struct nl_policy rtnlgrp_link_policy[] = { | |
3306 | [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, | |
3307 | [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, | |
3308 | .min_len = sizeof(struct rtnl_link_stats) }, | |
3309 | }; | |
3310 | ||
3311 | struct ofpbuf request; | |
3312 | struct ofpbuf *reply; | |
3313 | struct ifinfomsg *ifi; | |
3314 | const struct rtnl_link_stats *rtnl_stats; | |
3315 | struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; | |
3316 | int error; | |
3317 | ||
3318 | ofpbuf_init(&request, 0); | |
3319 | nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST); | |
3320 | ifi = ofpbuf_put_zeros(&request, sizeof *ifi); | |
3321 | ifi->ifi_family = PF_UNSPEC; | |
3322 | ifi->ifi_index = ifindex; | |
3323 | error = nl_sock_transact(rtnl_sock, &request, &reply); | |
3324 | ofpbuf_uninit(&request); | |
3325 | if (error) { | |
3326 | return error; | |
3327 | } | |
3328 | ||
3329 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), | |
3330 | rtnlgrp_link_policy, | |
3331 | attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { | |
3332 | ofpbuf_delete(reply); | |
3333 | return EPROTO; | |
3334 | } | |
3335 | ||
3336 | if (!attrs[IFLA_STATS]) { | |
3337 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
3338 | ofpbuf_delete(reply); | |
3339 | return EPROTO; | |
3340 | } | |
8b61709d BP |
3341 | |
3342 | rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); | |
3343 | stats->rx_packets = rtnl_stats->rx_packets; | |
3344 | stats->tx_packets = rtnl_stats->tx_packets; | |
3345 | stats->rx_bytes = rtnl_stats->rx_bytes; | |
3346 | stats->tx_bytes = rtnl_stats->tx_bytes; | |
3347 | stats->rx_errors = rtnl_stats->rx_errors; | |
3348 | stats->tx_errors = rtnl_stats->tx_errors; | |
3349 | stats->rx_dropped = rtnl_stats->rx_dropped; | |
3350 | stats->tx_dropped = rtnl_stats->tx_dropped; | |
3351 | stats->multicast = rtnl_stats->multicast; | |
3352 | stats->collisions = rtnl_stats->collisions; | |
3353 | stats->rx_length_errors = rtnl_stats->rx_length_errors; | |
3354 | stats->rx_over_errors = rtnl_stats->rx_over_errors; | |
3355 | stats->rx_crc_errors = rtnl_stats->rx_crc_errors; | |
3356 | stats->rx_frame_errors = rtnl_stats->rx_frame_errors; | |
3357 | stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; | |
3358 | stats->rx_missed_errors = rtnl_stats->rx_missed_errors; | |
3359 | stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; | |
3360 | stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; | |
3361 | stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; | |
3362 | stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; | |
3363 | stats->tx_window_errors = rtnl_stats->tx_window_errors; | |
3364 | ||
576e26d7 BP |
3365 | ofpbuf_delete(reply); |
3366 | ||
8b61709d BP |
3367 | return 0; |
3368 | } | |
3369 | ||
3370 | static int | |
3371 | get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) | |
3372 | { | |
3373 | static const char fn[] = "/proc/net/dev"; | |
3374 | char line[1024]; | |
3375 | FILE *stream; | |
3376 | int ln; | |
3377 | ||
3378 | stream = fopen(fn, "r"); | |
3379 | if (!stream) { | |
3380 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
3381 | return errno; | |
3382 | } | |
3383 | ||
3384 | ln = 0; | |
3385 | while (fgets(line, sizeof line, stream)) { | |
3386 | if (++ln >= 3) { | |
3387 | char devname[16]; | |
3388 | #define X64 "%"SCNu64 | |
3389 | if (sscanf(line, | |
3390 | " %15[^:]:" | |
3391 | X64 X64 X64 X64 X64 X64 X64 "%*u" | |
3392 | X64 X64 X64 X64 X64 X64 X64 "%*u", | |
3393 | devname, | |
3394 | &stats->rx_bytes, | |
3395 | &stats->rx_packets, | |
3396 | &stats->rx_errors, | |
3397 | &stats->rx_dropped, | |
3398 | &stats->rx_fifo_errors, | |
3399 | &stats->rx_frame_errors, | |
3400 | &stats->multicast, | |
3401 | &stats->tx_bytes, | |
3402 | &stats->tx_packets, | |
3403 | &stats->tx_errors, | |
3404 | &stats->tx_dropped, | |
3405 | &stats->tx_fifo_errors, | |
3406 | &stats->collisions, | |
3407 | &stats->tx_carrier_errors) != 15) { | |
3408 | VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); | |
3409 | } else if (!strcmp(devname, netdev_name)) { | |
3410 | stats->rx_length_errors = UINT64_MAX; | |
3411 | stats->rx_over_errors = UINT64_MAX; | |
3412 | stats->rx_crc_errors = UINT64_MAX; | |
3413 | stats->rx_missed_errors = UINT64_MAX; | |
3414 | stats->tx_aborted_errors = UINT64_MAX; | |
3415 | stats->tx_heartbeat_errors = UINT64_MAX; | |
3416 | stats->tx_window_errors = UINT64_MAX; | |
3417 | fclose(stream); | |
3418 | return 0; | |
3419 | } | |
3420 | } | |
3421 | } | |
3422 | VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); | |
3423 | fclose(stream); | |
3424 | return ENODEV; | |
3425 | } | |
c1c9c9c4 | 3426 | |
8b61709d BP |
3427 | static int |
3428 | get_flags(const struct netdev *netdev, int *flags) | |
3429 | { | |
3430 | struct ifreq ifr; | |
3431 | int error; | |
3432 | ||
149f577a JG |
3433 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, |
3434 | "SIOCGIFFLAGS"); | |
8b61709d BP |
3435 | *flags = ifr.ifr_flags; |
3436 | return error; | |
3437 | } | |
3438 | ||
3439 | static int | |
3440 | set_flags(struct netdev *netdev, int flags) | |
3441 | { | |
3442 | struct ifreq ifr; | |
3443 | ||
3444 | ifr.ifr_flags = flags; | |
149f577a JG |
3445 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, |
3446 | "SIOCSIFFLAGS"); | |
8b61709d BP |
3447 | } |
3448 | ||
3449 | static int | |
3450 | do_get_ifindex(const char *netdev_name) | |
3451 | { | |
3452 | struct ifreq ifr; | |
3453 | ||
3454 | strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); | |
3455 | COVERAGE_INC(netdev_get_ifindex); | |
3456 | if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { | |
3457 | VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
3458 | netdev_name, strerror(errno)); | |
3459 | return -errno; | |
3460 | } | |
3461 | return ifr.ifr_ifindex; | |
3462 | } | |
3463 | ||
3464 | static int | |
3465 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
3466 | { | |
149f577a JG |
3467 | struct netdev_dev_linux *netdev_dev = |
3468 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d | 3469 | *ifindexp = 0; |
149f577a | 3470 | if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { |
8b61709d BP |
3471 | int ifindex = do_get_ifindex(netdev_get_name(netdev_)); |
3472 | if (ifindex < 0) { | |
3473 | return -ifindex; | |
3474 | } | |
149f577a JG |
3475 | netdev_dev->cache_valid |= VALID_IFINDEX; |
3476 | netdev_dev->ifindex = ifindex; | |
8b61709d | 3477 | } |
149f577a | 3478 | *ifindexp = netdev_dev->ifindex; |
8b61709d BP |
3479 | return 0; |
3480 | } | |
3481 | ||
3482 | static int | |
3483 | get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) | |
3484 | { | |
3485 | struct ifreq ifr; | |
3486 | int hwaddr_family; | |
3487 | ||
3488 | memset(&ifr, 0, sizeof ifr); | |
3489 | strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); | |
3490 | COVERAGE_INC(netdev_get_hwaddr); | |
3491 | if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { | |
3492 | VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", | |
3493 | netdev_name, strerror(errno)); | |
3494 | return errno; | |
3495 | } | |
3496 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
3497 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { | |
3498 | VLOG_WARN("%s device has unknown hardware address family %d", | |
3499 | netdev_name, hwaddr_family); | |
3500 | } | |
3501 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
3502 | return 0; | |
3503 | } | |
3504 | ||
3505 | static int | |
3506 | set_etheraddr(const char *netdev_name, int hwaddr_family, | |
3507 | const uint8_t mac[ETH_ADDR_LEN]) | |
3508 | { | |
3509 | struct ifreq ifr; | |
3510 | ||
3511 | memset(&ifr, 0, sizeof ifr); | |
3512 | strncpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); | |
3513 | ifr.ifr_hwaddr.sa_family = hwaddr_family; | |
3514 | memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); | |
3515 | COVERAGE_INC(netdev_set_hwaddr); | |
3516 | if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { | |
3517 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", | |
3518 | netdev_name, strerror(errno)); | |
3519 | return errno; | |
3520 | } | |
3521 | return 0; | |
3522 | } | |
3523 | ||
3524 | static int | |
0b0544d7 | 3525 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
3526 | int cmd, const char *cmd_name) |
3527 | { | |
3528 | struct ifreq ifr; | |
3529 | ||
3530 | memset(&ifr, 0, sizeof ifr); | |
0b0544d7 | 3531 | strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
3532 | ifr.ifr_data = (caddr_t) ecmd; |
3533 | ||
3534 | ecmd->cmd = cmd; | |
3535 | COVERAGE_INC(netdev_ethtool); | |
3536 | if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { | |
3537 | return 0; | |
3538 | } else { | |
3539 | if (errno != EOPNOTSUPP) { | |
3540 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " | |
0b0544d7 | 3541 | "failed: %s", cmd_name, name, strerror(errno)); |
8b61709d BP |
3542 | } else { |
3543 | /* The device doesn't support this operation. That's pretty | |
3544 | * common, so there's no point in logging anything. */ | |
3545 | } | |
3546 | return errno; | |
3547 | } | |
3548 | } | |
3549 | ||
3550 | static int | |
149f577a JG |
3551 | netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, |
3552 | const char *cmd_name) | |
8b61709d | 3553 | { |
149f577a | 3554 | strncpy(ifr->ifr_name, name, sizeof ifr->ifr_name); |
8b61709d | 3555 | if (ioctl(af_inet_sock, cmd, ifr) == -1) { |
149f577a JG |
3556 | VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, |
3557 | strerror(errno)); | |
8b61709d BP |
3558 | return errno; |
3559 | } | |
3560 | return 0; | |
3561 | } | |
f1acd62b BP |
3562 | |
3563 | static int | |
3564 | netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, | |
3565 | int cmd, const char *cmd_name) | |
3566 | { | |
3567 | struct ifreq ifr; | |
3568 | int error; | |
3569 | ||
3570 | ifr.ifr_addr.sa_family = AF_INET; | |
149f577a | 3571 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); |
f1acd62b BP |
3572 | if (!error) { |
3573 | const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; | |
3574 | *ip = sin->sin_addr; | |
3575 | } | |
3576 | return error; | |
3577 | } |