]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
782e6111 | 2 | * Copyright (c) 2009, 2010, 2011 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> | |
63331829 | 28 | #include <linux/mii.h> |
6f42c8ea | 29 | #include <linux/pkt_sched.h> |
e9e28be3 | 30 | #include <linux/rtnetlink.h> |
8b61709d BP |
31 | #include <linux/sockios.h> |
32 | #include <linux/version.h> | |
33 | #include <sys/types.h> | |
34 | #include <sys/ioctl.h> | |
35 | #include <sys/socket.h> | |
36 | #include <netpacket/packet.h> | |
37 | #include <net/ethernet.h> | |
38 | #include <net/if.h> | |
a740f0de | 39 | #include <linux/if_tunnel.h> |
8b61709d BP |
40 | #include <net/if_arp.h> |
41 | #include <net/if_packet.h> | |
42 | #include <net/route.h> | |
43 | #include <netinet/in.h> | |
e9e28be3 | 44 | #include <poll.h> |
8b61709d BP |
45 | #include <stdlib.h> |
46 | #include <string.h> | |
47 | #include <unistd.h> | |
e9e28be3 BP |
48 | |
49 | #include "coverage.h" | |
9fe3b9a2 | 50 | #include "dpif-linux.h" |
8b61709d BP |
51 | #include "dynamic-string.h" |
52 | #include "fatal-signal.h" | |
93b13be8 BP |
53 | #include "hash.h" |
54 | #include "hmap.h" | |
8b61709d | 55 | #include "netdev-provider.h" |
7fbef77a | 56 | #include "netdev-vport.h" |
e9e28be3 | 57 | #include "netlink.h" |
2fe27d5a | 58 | #include "netlink-socket.h" |
e9e28be3 | 59 | #include "ofpbuf.h" |
8b61709d BP |
60 | #include "openflow/openflow.h" |
61 | #include "packets.h" | |
62 | #include "poll-loop.h" | |
559843ed | 63 | #include "rtnetlink.h" |
21d6e22e | 64 | #include "rtnetlink-link.h" |
8b61709d BP |
65 | #include "socket-util.h" |
66 | #include "shash.h" | |
67 | #include "svec.h" | |
e9e28be3 | 68 | #include "vlog.h" |
5136ce49 | 69 | |
d98e6007 | 70 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea BP |
71 | |
72 | COVERAGE_DEFINE(netdev_get_vlan_vid); | |
73 | COVERAGE_DEFINE(netdev_set_policing); | |
74 | COVERAGE_DEFINE(netdev_arp_lookup); | |
75 | COVERAGE_DEFINE(netdev_get_ifindex); | |
76 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
77 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
78 | COVERAGE_DEFINE(netdev_ethtool); | |
8b61709d BP |
79 | \f |
80 | /* These were introduced in Linux 2.6.14, so they might be missing if we have | |
81 | * old headers. */ | |
82 | #ifndef ADVERTISED_Pause | |
83 | #define ADVERTISED_Pause (1 << 13) | |
84 | #endif | |
85 | #ifndef ADVERTISED_Asym_Pause | |
86 | #define ADVERTISED_Asym_Pause (1 << 14) | |
87 | #endif | |
88 | ||
c1c9c9c4 BP |
89 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
90 | * headers. */ | |
91 | #ifndef TC_RTAB_SIZE | |
92 | #define TC_RTAB_SIZE 1024 | |
93 | #endif | |
94 | ||
149f577a | 95 | static struct rtnetlink_notifier netdev_linux_cache_notifier; |
46415c90 | 96 | static int cache_notifier_refcount; |
8b61709d BP |
97 | |
98 | enum { | |
7fbef77a JG |
99 | VALID_IFINDEX = 1 << 0, |
100 | VALID_ETHERADDR = 1 << 1, | |
101 | VALID_IN4 = 1 << 2, | |
102 | VALID_IN6 = 1 << 3, | |
103 | VALID_MTU = 1 << 4, | |
104 | VALID_CARRIER = 1 << 5, | |
105 | VALID_IS_PSEUDO = 1 << 6, /* Represents is_internal and is_tap. */ | |
106 | VALID_POLICING = 1 << 7, | |
107 | VALID_HAVE_VPORT_STATS = 1 << 8 | |
8b61709d BP |
108 | }; |
109 | ||
149f577a JG |
110 | struct tap_state { |
111 | int fd; | |
61b999dd | 112 | bool opened; |
149f577a | 113 | }; |
c1c9c9c4 BP |
114 | \f |
115 | /* Traffic control. */ | |
116 | ||
117 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
118 | * network device. |
119 | * | |
120 | * Each TC implementation subclasses this with whatever additional data it | |
121 | * needs. */ | |
c1c9c9c4 BP |
122 | struct tc { |
123 | const struct tc_ops *ops; | |
93b13be8 BP |
124 | struct hmap queues; /* Contains "struct tc_queue"s. |
125 | * Read by generic TC layer. | |
126 | * Written only by TC implementation. */ | |
127 | }; | |
c1c9c9c4 | 128 | |
93b13be8 BP |
129 | /* One traffic control queue. |
130 | * | |
131 | * Each TC implementation subclasses this with whatever additional data it | |
132 | * needs. */ | |
133 | struct tc_queue { | |
134 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
135 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
c1c9c9c4 BP |
136 | }; |
137 | ||
138 | /* A particular kind of traffic control. Each implementation generally maps to | |
139 | * one particular Linux qdisc class. | |
140 | * | |
141 | * The functions below return 0 if successful or a positive errno value on | |
142 | * failure, except where otherwise noted. All of them must be provided, except | |
143 | * where otherwise noted. */ | |
144 | struct tc_ops { | |
145 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
146 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
147 | * appropriate values. */ | |
148 | const char *linux_name; | |
149 | ||
150 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
151 | const char *ovs_name; | |
152 | ||
153 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
154 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
155 | unsigned int n_queues; | |
156 | ||
157 | /* Called to install this TC class on 'netdev'. The implementation should | |
158 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
159 | * and configure it according to 'details'. The implementation may assume | |
160 | * that the current qdisc is the default; that is, there is no need for it | |
161 | * to delete the current qdisc before installing itself. | |
162 | * | |
163 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
164 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
165 | * (which is built as ovs-vswitchd.conf.db(8)). | |
166 | * | |
167 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
168 | * initialized 'struct tc'. | |
169 | * | |
170 | * (This function is null for tc_ops_other, which cannot be installed. For | |
171 | * other TC classes it should always be nonnull.) */ | |
172 | int (*tc_install)(struct netdev *netdev, const struct shash *details); | |
173 | ||
174 | /* Called when the netdev code determines (through a Netlink query) that | |
175 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
176 | * it ourselves and so don't know any of the details. | |
177 | * | |
178 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
179 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
180 | * implementation should parse the other attributes of 'nlmsg' as | |
181 | * necessary to determine its configuration. If necessary it should also | |
182 | * use Netlink queries to determine the configuration of queues on | |
183 | * 'netdev'. | |
184 | * | |
185 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
186 | * initialized 'struct tc'. */ | |
187 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
188 | ||
189 | /* Destroys the data structures allocated by the implementation as part of | |
190 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
191 | * tc_destroy(tc). | |
192 | * | |
193 | * The implementation should not need to perform any Netlink calls. If | |
194 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
195 | * (But it may not be desirable.) | |
196 | * | |
197 | * This function may be null if 'tc' is trivial. */ | |
198 | void (*tc_destroy)(struct tc *tc); | |
199 | ||
200 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
201 | * | |
202 | * The implementation should not need to perform any Netlink calls, because | |
203 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
204 | * cached the configuration. | |
205 | * | |
206 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
207 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
208 | * (which is built as ovs-vswitchd.conf.db(8)). | |
209 | * | |
210 | * This function may be null if 'tc' is not configurable. | |
211 | */ | |
212 | int (*qdisc_get)(const struct netdev *netdev, struct shash *details); | |
213 | ||
214 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
215 | * required Netlink calls to complete the reconfiguration. | |
216 | * | |
217 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
218 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
219 | * (which is built as ovs-vswitchd.conf.db(8)). | |
220 | * | |
221 | * This function may be null if 'tc' is not configurable. | |
222 | */ | |
223 | int (*qdisc_set)(struct netdev *, const struct shash *details); | |
224 | ||
93b13be8 BP |
225 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
226 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
227 | * |
228 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
229 | * in the "other_config" column in the "Queue" table in | |
230 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
231 | * | |
232 | * The implementation should not need to perform any Netlink calls, because | |
233 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
234 | * cached the queue configuration. | |
235 | * | |
236 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
237 | * 0). */ | |
93b13be8 | 238 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
239 | struct shash *details); |
240 | ||
241 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
242 | * 'details', perfoming any required Netlink calls to complete the | |
243 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
244 | * 'n_queues'. | |
245 | * | |
246 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
247 | * in the "other_config" column in the "Queue" table in | |
248 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
249 | * | |
250 | * This function may be null if 'tc' does not have queues or its queues are | |
251 | * not configurable. */ | |
252 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
253 | const struct shash *details); | |
254 | ||
93b13be8 BP |
255 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
256 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
257 | * |
258 | * This function may be null if 'tc' does not have queues or its queues | |
259 | * cannot be deleted. */ | |
93b13be8 | 260 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 261 | |
93b13be8 BP |
262 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
263 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
264 | * |
265 | * On success, initializes '*stats'. | |
266 | * | |
267 | * This function may be null if 'tc' does not have queues or if it cannot | |
268 | * report queue statistics. */ | |
93b13be8 BP |
269 | int (*class_get_stats)(const struct netdev *netdev, |
270 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
271 | struct netdev_queue_stats *stats); |
272 | ||
273 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
274 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
275 | * | |
276 | * This function may be null if 'tc' does not have queues or if it cannot | |
277 | * report queue statistics. */ | |
278 | int (*class_dump_stats)(const struct netdev *netdev, | |
279 | const struct ofpbuf *nlmsg, | |
280 | netdev_dump_queue_stats_cb *cb, void *aux); | |
281 | }; | |
282 | ||
283 | static void | |
284 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
285 | { | |
286 | tc->ops = ops; | |
93b13be8 | 287 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
288 | } |
289 | ||
290 | static void | |
291 | tc_destroy(struct tc *tc) | |
292 | { | |
93b13be8 | 293 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
294 | } |
295 | ||
296 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 297 | static const struct tc_ops tc_ops_hfsc; |
c1c9c9c4 BP |
298 | static const struct tc_ops tc_ops_default; |
299 | static const struct tc_ops tc_ops_other; | |
300 | ||
301 | static const struct tc_ops *tcs[] = { | |
302 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ | |
a339aa81 | 303 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
c1c9c9c4 BP |
304 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
305 | &tc_ops_other, /* Some other qdisc. */ | |
306 | NULL | |
307 | }; | |
149f577a | 308 | |
c1c9c9c4 BP |
309 | static unsigned int tc_make_handle(unsigned int major, unsigned int minor); |
310 | static unsigned int tc_get_major(unsigned int handle); | |
311 | static unsigned int tc_get_minor(unsigned int handle); | |
312 | ||
313 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); | |
314 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
315 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
316 | ||
317 | static struct tcmsg *tc_make_request(const struct netdev *, int type, | |
318 | unsigned int flags, struct ofpbuf *); | |
319 | static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); | |
320 | ||
321 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
322 | struct nlattr **options); | |
323 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
324 | struct nlattr **options, | |
325 | struct netdev_queue_stats *); | |
326 | static int tc_query_class(const struct netdev *, | |
327 | unsigned int handle, unsigned int parent, | |
328 | struct ofpbuf **replyp); | |
329 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
330 | ||
331 | static int tc_del_qdisc(struct netdev *netdev); | |
332 | static int tc_query_qdisc(const struct netdev *netdev); | |
333 | ||
334 | static int tc_calc_cell_log(unsigned int mtu); | |
335 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
336 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
337 | const struct tc_ratespec *rate); | |
338 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
339 | \f | |
149f577a JG |
340 | struct netdev_dev_linux { |
341 | struct netdev_dev netdev_dev; | |
342 | ||
8b61709d | 343 | struct shash_node *shash_node; |
149f577a | 344 | unsigned int cache_valid; |
8b61709d | 345 | |
8722022c BP |
346 | /* The following are figured out "on demand" only. They are only valid |
347 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d BP |
348 | int ifindex; |
349 | uint8_t etheraddr[ETH_ADDR_LEN]; | |
f1acd62b | 350 | struct in_addr address, netmask; |
8b61709d BP |
351 | struct in6_addr in6; |
352 | int mtu; | |
353 | int carrier; | |
8722022c BP |
354 | bool is_internal; /* Is this an openvswitch internal device? */ |
355 | bool is_tap; /* Is this a tuntap device? */ | |
80a86fbe BP |
356 | uint32_t kbits_rate; /* Policing data. */ |
357 | uint32_t kbits_burst; | |
7fbef77a | 358 | bool have_vport_stats; |
c1c9c9c4 | 359 | struct tc *tc; |
149f577a JG |
360 | |
361 | union { | |
362 | struct tap_state tap; | |
363 | } state; | |
8b61709d BP |
364 | }; |
365 | ||
149f577a JG |
366 | struct netdev_linux { |
367 | struct netdev netdev; | |
5b7448ed | 368 | int fd; |
149f577a | 369 | }; |
8b61709d | 370 | |
8b61709d BP |
371 | /* An AF_INET socket (used for ioctl operations). */ |
372 | static int af_inet_sock = -1; | |
373 | ||
ff4ed3c9 BP |
374 | /* A Netlink routing socket that is not subscribed to any multicast groups. */ |
375 | static struct nl_sock *rtnl_sock; | |
376 | ||
8b61709d BP |
377 | struct netdev_linux_notifier { |
378 | struct netdev_notifier notifier; | |
379 | struct list node; | |
380 | }; | |
381 | ||
382 | static struct shash netdev_linux_notifiers = | |
383 | SHASH_INITIALIZER(&netdev_linux_notifiers); | |
46097491 | 384 | static struct rtnetlink_notifier netdev_linux_poll_notifier; |
8b61709d BP |
385 | |
386 | /* This is set pretty low because we probably won't learn anything from the | |
387 | * additional log messages. */ | |
388 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
389 | ||
15b3596a | 390 | static int netdev_linux_init(void); |
6f643e49 | 391 | |
0b0544d7 | 392 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 393 | int cmd, const char *cmd_name); |
149f577a JG |
394 | static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, |
395 | const char *cmd_name); | |
f1acd62b BP |
396 | static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, |
397 | int cmd, const char *cmd_name); | |
8b61709d BP |
398 | static int get_flags(const struct netdev *, int *flagsp); |
399 | static int set_flags(struct netdev *, int flags); | |
400 | static int do_get_ifindex(const char *netdev_name); | |
401 | static int get_ifindex(const struct netdev *, int *ifindexp); | |
402 | static int do_set_addr(struct netdev *netdev, | |
403 | int ioctl_nr, const char *ioctl_name, | |
404 | struct in_addr addr); | |
405 | static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); | |
406 | static int set_etheraddr(const char *netdev_name, int hwaddr_family, | |
407 | const uint8_t[ETH_ADDR_LEN]); | |
408 | static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); | |
409 | static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); | |
410 | ||
15b3596a JG |
411 | static bool |
412 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
413 | { | |
414 | return netdev_class->init == netdev_linux_init; | |
415 | } | |
416 | ||
149f577a JG |
417 | static struct netdev_dev_linux * |
418 | netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) | |
6c88d577 | 419 | { |
15b3596a JG |
420 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); |
421 | assert(is_netdev_linux_class(netdev_class)); | |
422 | ||
149f577a | 423 | return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev); |
6c88d577 JP |
424 | } |
425 | ||
8b61709d BP |
426 | static struct netdev_linux * |
427 | netdev_linux_cast(const struct netdev *netdev) | |
428 | { | |
15b3596a JG |
429 | struct netdev_dev *netdev_dev = netdev_get_dev(netdev); |
430 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); | |
431 | assert(is_netdev_linux_class(netdev_class)); | |
432 | ||
8b61709d BP |
433 | return CONTAINER_OF(netdev, struct netdev_linux, netdev); |
434 | } | |
ff4ed3c9 | 435 | \f |
8b61709d BP |
436 | static int |
437 | netdev_linux_init(void) | |
438 | { | |
439 | static int status = -1; | |
440 | if (status < 0) { | |
ff4ed3c9 | 441 | /* Create AF_INET socket. */ |
8b61709d BP |
442 | af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); |
443 | status = af_inet_sock >= 0 ? 0 : errno; | |
444 | if (status) { | |
445 | VLOG_ERR("failed to create inet socket: %s", strerror(status)); | |
446 | } | |
ff4ed3c9 BP |
447 | |
448 | /* Create rtnetlink socket. */ | |
449 | if (!status) { | |
cceb11f5 | 450 | status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); |
ff4ed3c9 BP |
451 | if (status) { |
452 | VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", | |
453 | strerror(status)); | |
454 | } | |
455 | } | |
8b61709d BP |
456 | } |
457 | return status; | |
458 | } | |
459 | ||
460 | static void | |
461 | netdev_linux_run(void) | |
462 | { | |
21d6e22e | 463 | rtnetlink_link_notifier_run(); |
8b61709d BP |
464 | } |
465 | ||
466 | static void | |
467 | netdev_linux_wait(void) | |
468 | { | |
21d6e22e | 469 | rtnetlink_link_notifier_wait(); |
8b61709d BP |
470 | } |
471 | ||
472 | static void | |
21d6e22e | 473 | netdev_linux_cache_cb(const struct rtnetlink_link_change *change, |
67a4917b | 474 | void *aux OVS_UNUSED) |
8b61709d | 475 | { |
149f577a | 476 | struct netdev_dev_linux *dev; |
8b61709d | 477 | if (change) { |
46415c90 JG |
478 | struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname); |
479 | if (base_dev) { | |
15b3596a JG |
480 | const struct netdev_class *netdev_class = |
481 | netdev_dev_get_class(base_dev); | |
482 | ||
483 | if (is_netdev_linux_class(netdev_class)) { | |
484 | dev = netdev_dev_linux_cast(base_dev); | |
485 | dev->cache_valid = 0; | |
486 | } | |
8b61709d BP |
487 | } |
488 | } else { | |
46415c90 | 489 | struct shash device_shash; |
8b61709d | 490 | struct shash_node *node; |
46415c90 JG |
491 | |
492 | shash_init(&device_shash); | |
493 | netdev_dev_get_devices(&netdev_linux_class, &device_shash); | |
494 | SHASH_FOR_EACH (node, &device_shash) { | |
149f577a JG |
495 | dev = node->data; |
496 | dev->cache_valid = 0; | |
8b61709d | 497 | } |
46415c90 | 498 | shash_destroy(&device_shash); |
8b61709d BP |
499 | } |
500 | } | |
501 | ||
c3827f61 | 502 | /* Creates system and internal devices. */ |
8b61709d | 503 | static int |
c3827f61 | 504 | netdev_linux_create(const struct netdev_class *class, |
b8dcf5e9 BP |
505 | const char *name, const struct shash *args, |
506 | struct netdev_dev **netdev_devp) | |
6c88d577 | 507 | { |
149f577a JG |
508 | struct netdev_dev_linux *netdev_dev; |
509 | int error; | |
6c88d577 JP |
510 | |
511 | if (!shash_is_empty(args)) { | |
c3827f61 BP |
512 | VLOG_WARN("%s: arguments for %s devices should be empty", |
513 | name, class->type); | |
6c88d577 JP |
514 | } |
515 | ||
46415c90 | 516 | if (!cache_notifier_refcount) { |
21d6e22e EJ |
517 | error = rtnetlink_link_notifier_register(&netdev_linux_cache_notifier, |
518 | netdev_linux_cache_cb, NULL); | |
149f577a JG |
519 | if (error) { |
520 | return error; | |
521 | } | |
522 | } | |
46415c90 | 523 | cache_notifier_refcount++; |
6c88d577 | 524 | |
149f577a | 525 | netdev_dev = xzalloc(sizeof *netdev_dev); |
6d9e6eb4 | 526 | netdev_dev_init(&netdev_dev->netdev_dev, name, args, class); |
46415c90 | 527 | |
149f577a | 528 | *netdev_devp = &netdev_dev->netdev_dev; |
a740f0de JG |
529 | return 0; |
530 | } | |
531 | ||
5b7448ed JG |
532 | /* For most types of netdevs we open the device for each call of |
533 | * netdev_open(). However, this is not the case with tap devices, | |
534 | * since it is only possible to open the device once. In this | |
535 | * situation we share a single file descriptor, and consequently | |
536 | * buffers, across all readers. Therefore once data is read it will | |
537 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 538 | static int |
b8dcf5e9 BP |
539 | netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, |
540 | const char *name, const struct shash *args, | |
541 | struct netdev_dev **netdev_devp) | |
a740f0de | 542 | { |
149f577a | 543 | struct netdev_dev_linux *netdev_dev; |
a740f0de JG |
544 | struct tap_state *state; |
545 | static const char tap_dev[] = "/dev/net/tun"; | |
546 | struct ifreq ifr; | |
547 | int error; | |
548 | ||
549 | if (!shash_is_empty(args)) { | |
149f577a | 550 | VLOG_WARN("%s: arguments for TAP devices should be empty", name); |
6c88d577 JP |
551 | } |
552 | ||
149f577a JG |
553 | netdev_dev = xzalloc(sizeof *netdev_dev); |
554 | state = &netdev_dev->state.tap; | |
a740f0de | 555 | |
6c88d577 | 556 | /* Open tap device. */ |
149f577a JG |
557 | state->fd = open(tap_dev, O_RDWR); |
558 | if (state->fd < 0) { | |
6c88d577 JP |
559 | error = errno; |
560 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); | |
561 | goto error; | |
562 | } | |
563 | ||
564 | /* Create tap device. */ | |
565 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
71d7c22f | 566 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
149f577a | 567 | if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 JP |
568 | VLOG_WARN("%s: creating tap device failed: %s", name, |
569 | strerror(errno)); | |
570 | error = errno; | |
571 | goto error; | |
572 | } | |
573 | ||
574 | /* Make non-blocking. */ | |
149f577a | 575 | error = set_nonblocking(state->fd); |
a740f0de JG |
576 | if (error) { |
577 | goto error; | |
578 | } | |
579 | ||
6d9e6eb4 | 580 | netdev_dev_init(&netdev_dev->netdev_dev, name, args, &netdev_tap_class); |
149f577a | 581 | *netdev_devp = &netdev_dev->netdev_dev; |
a740f0de JG |
582 | return 0; |
583 | ||
584 | error: | |
149f577a | 585 | free(netdev_dev); |
a740f0de JG |
586 | return error; |
587 | } | |
588 | ||
a740f0de | 589 | static void |
149f577a | 590 | destroy_tap(struct netdev_dev_linux *netdev_dev) |
a740f0de | 591 | { |
149f577a JG |
592 | struct tap_state *state = &netdev_dev->state.tap; |
593 | ||
594 | if (state->fd >= 0) { | |
595 | close(state->fd); | |
a740f0de JG |
596 | } |
597 | } | |
598 | ||
149f577a | 599 | /* Destroys the netdev device 'netdev_dev_'. */ |
6c88d577 | 600 | static void |
149f577a | 601 | netdev_linux_destroy(struct netdev_dev *netdev_dev_) |
6c88d577 | 602 | { |
149f577a | 603 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
d2bb2799 | 604 | const struct netdev_class *class = netdev_dev_get_class(netdev_dev_); |
6c88d577 | 605 | |
c1c9c9c4 BP |
606 | if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) { |
607 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
608 | } | |
609 | ||
d2bb2799 | 610 | if (class == &netdev_linux_class || class == &netdev_internal_class) { |
46415c90 | 611 | cache_notifier_refcount--; |
149f577a | 612 | |
46415c90 | 613 | if (!cache_notifier_refcount) { |
21d6e22e | 614 | rtnetlink_link_notifier_unregister(&netdev_linux_cache_notifier); |
149f577a | 615 | } |
d2bb2799 | 616 | } else if (class == &netdev_tap_class) { |
149f577a | 617 | destroy_tap(netdev_dev); |
d2bb2799 BP |
618 | } else { |
619 | NOT_REACHED(); | |
6c88d577 | 620 | } |
149f577a | 621 | |
658797c8 | 622 | free(netdev_dev); |
6c88d577 JP |
623 | } |
624 | ||
8b61709d | 625 | static int |
5b7448ed | 626 | netdev_linux_open(struct netdev_dev *netdev_dev_, int ethertype, |
149f577a | 627 | struct netdev **netdevp) |
8b61709d | 628 | { |
5b7448ed | 629 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
8b61709d BP |
630 | struct netdev_linux *netdev; |
631 | enum netdev_flags flags; | |
632 | int error; | |
633 | ||
634 | /* Allocate network device. */ | |
ec6fde61 | 635 | netdev = xzalloc(sizeof *netdev); |
49a6a163 | 636 | netdev->fd = -1; |
5b7448ed | 637 | netdev_init(&netdev->netdev, netdev_dev_); |
8b61709d | 638 | |
c3827f61 BP |
639 | /* Verify that the device really exists, by attempting to read its flags. |
640 | * (The flags might be cached, in which case this won't actually do an | |
641 | * ioctl.) | |
642 | * | |
643 | * Don't do this for "internal" netdevs, though, because those have to be | |
644 | * created as netdev objects before they exist in the kernel, because | |
645 | * creating them in the kernel happens by passing a netdev object to | |
646 | * dpif_port_add(). */ | |
647 | if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) { | |
648 | error = netdev_get_flags(&netdev->netdev, &flags); | |
649 | if (error == ENODEV) { | |
650 | goto error; | |
651 | } | |
8b61709d BP |
652 | } |
653 | ||
61b999dd JG |
654 | if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") && |
655 | !netdev_dev->state.tap.opened) { | |
656 | ||
657 | /* We assume that the first user of the tap device is the primary user | |
658 | * and give them the tap FD. Subsequent users probably just expect | |
659 | * this to be a system device so open it normally to avoid send/receive | |
660 | * directions appearing to be reversed. */ | |
5b7448ed | 661 | netdev->fd = netdev_dev->state.tap.fd; |
61b999dd | 662 | netdev_dev->state.tap.opened = true; |
5b7448ed | 663 | } else if (ethertype != NETDEV_ETH_TYPE_NONE) { |
8b61709d BP |
664 | struct sockaddr_ll sll; |
665 | int protocol; | |
666 | int ifindex; | |
667 | ||
668 | /* Create file descriptor. */ | |
669 | protocol = (ethertype == NETDEV_ETH_TYPE_ANY ? ETH_P_ALL | |
670 | : ethertype == NETDEV_ETH_TYPE_802_2 ? ETH_P_802_2 | |
671 | : ethertype); | |
5b7448ed JG |
672 | netdev->fd = socket(PF_PACKET, SOCK_RAW, htons(protocol)); |
673 | if (netdev->fd < 0) { | |
8b61709d BP |
674 | error = errno; |
675 | goto error; | |
676 | } | |
8b61709d BP |
677 | |
678 | /* Set non-blocking mode. */ | |
5b7448ed | 679 | error = set_nonblocking(netdev->fd); |
8b61709d BP |
680 | if (error) { |
681 | goto error; | |
682 | } | |
683 | ||
684 | /* Get ethernet device index. */ | |
685 | error = get_ifindex(&netdev->netdev, &ifindex); | |
686 | if (error) { | |
687 | goto error; | |
688 | } | |
689 | ||
690 | /* Bind to specific ethernet device. */ | |
691 | memset(&sll, 0, sizeof sll); | |
692 | sll.sll_family = AF_PACKET; | |
693 | sll.sll_ifindex = ifindex; | |
5b7448ed | 694 | if (bind(netdev->fd, |
8b61709d BP |
695 | (struct sockaddr *) &sll, sizeof sll) < 0) { |
696 | error = errno; | |
5b7448ed | 697 | VLOG_ERR("bind to %s failed: %s", netdev_dev_get_name(netdev_dev_), |
149f577a | 698 | strerror(error)); |
8b61709d BP |
699 | goto error; |
700 | } | |
701 | ||
702 | /* Between the socket() and bind() calls above, the socket receives all | |
703 | * packets of the requested type on all system interfaces. We do not | |
704 | * want to receive that data, but there is no way to avoid it. So we | |
705 | * must now drain out the receive queue. */ | |
5b7448ed | 706 | error = drain_rcvbuf(netdev->fd); |
8b61709d BP |
707 | if (error) { |
708 | goto error; | |
709 | } | |
710 | } | |
711 | ||
712 | *netdevp = &netdev->netdev; | |
713 | return 0; | |
714 | ||
715 | error: | |
149f577a | 716 | netdev_uninit(&netdev->netdev, true); |
8b61709d BP |
717 | return error; |
718 | } | |
719 | ||
720 | /* Closes and destroys 'netdev'. */ | |
721 | static void | |
722 | netdev_linux_close(struct netdev *netdev_) | |
723 | { | |
724 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
725 | ||
49a6a163 | 726 | if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { |
5b7448ed | 727 | close(netdev->fd); |
8b61709d BP |
728 | } |
729 | free(netdev); | |
730 | } | |
e9e28be3 | 731 | |
8b61709d BP |
732 | /* Initializes 'svec' with a list of the names of all known network devices. */ |
733 | static int | |
734 | netdev_linux_enumerate(struct svec *svec) | |
735 | { | |
736 | struct if_nameindex *names; | |
737 | ||
738 | names = if_nameindex(); | |
739 | if (names) { | |
740 | size_t i; | |
741 | ||
742 | for (i = 0; names[i].if_name != NULL; i++) { | |
743 | svec_add(svec, names[i].if_name); | |
744 | } | |
745 | if_freenameindex(names); | |
746 | return 0; | |
747 | } else { | |
748 | VLOG_WARN("could not obtain list of network device names: %s", | |
749 | strerror(errno)); | |
750 | return errno; | |
751 | } | |
752 | } | |
753 | ||
754 | static int | |
755 | netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) | |
756 | { | |
757 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
758 | ||
5b7448ed | 759 | if (netdev->fd < 0) { |
8b61709d | 760 | /* Device was opened with NETDEV_ETH_TYPE_NONE. */ |
c0e5f6ca | 761 | return -EAGAIN; |
8b61709d BP |
762 | } |
763 | ||
764 | for (;;) { | |
5b7448ed | 765 | ssize_t retval = read(netdev->fd, data, size); |
8b61709d BP |
766 | if (retval >= 0) { |
767 | return retval; | |
768 | } else if (errno != EINTR) { | |
769 | if (errno != EAGAIN) { | |
770 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
771 | strerror(errno), netdev_get_name(netdev_)); | |
772 | } | |
c0e5f6ca | 773 | return -errno; |
8b61709d BP |
774 | } |
775 | } | |
776 | } | |
777 | ||
778 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
779 | * when a packet is ready to be received with netdev_recv() on 'netdev'. */ | |
780 | static void | |
781 | netdev_linux_recv_wait(struct netdev *netdev_) | |
782 | { | |
783 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed JG |
784 | if (netdev->fd >= 0) { |
785 | poll_fd_wait(netdev->fd, POLLIN); | |
8b61709d BP |
786 | } |
787 | } | |
788 | ||
789 | /* Discards all packets waiting to be received from 'netdev'. */ | |
790 | static int | |
791 | netdev_linux_drain(struct netdev *netdev_) | |
792 | { | |
793 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 794 | if (netdev->fd < 0) { |
8b61709d | 795 | return 0; |
5b7448ed | 796 | } else if (!strcmp(netdev_get_type(netdev_), "tap")) { |
8b61709d | 797 | struct ifreq ifr; |
149f577a | 798 | int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
8b61709d BP |
799 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
800 | if (error) { | |
801 | return error; | |
802 | } | |
5b7448ed | 803 | drain_fd(netdev->fd, ifr.ifr_qlen); |
8b61709d BP |
804 | return 0; |
805 | } else { | |
5b7448ed | 806 | return drain_rcvbuf(netdev->fd); |
8b61709d BP |
807 | } |
808 | } | |
809 | ||
810 | /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive | |
811 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued | |
812 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
813 | * the packet is too big or too small to transmit on the device. | |
814 | * | |
815 | * The caller retains ownership of 'buffer' in all cases. | |
816 | * | |
817 | * The kernel maintains a packet transmission queue, so the caller is not | |
818 | * expected to do additional queuing of packets. */ | |
819 | static int | |
820 | netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) | |
821 | { | |
822 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
823 | ||
824 | /* XXX should support sending even if 'ethertype' was NETDEV_ETH_TYPE_NONE. | |
825 | */ | |
5b7448ed | 826 | if (netdev->fd < 0) { |
8b61709d BP |
827 | return EPIPE; |
828 | } | |
829 | ||
830 | for (;;) { | |
5b7448ed | 831 | ssize_t retval = write(netdev->fd, data, size); |
8b61709d BP |
832 | if (retval < 0) { |
833 | /* The Linux AF_PACKET implementation never blocks waiting for room | |
834 | * for packets, instead returning ENOBUFS. Translate this into | |
835 | * EAGAIN for the caller. */ | |
836 | if (errno == ENOBUFS) { | |
837 | return EAGAIN; | |
838 | } else if (errno == EINTR) { | |
839 | continue; | |
840 | } else if (errno != EAGAIN) { | |
841 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", | |
842 | netdev_get_name(netdev_), strerror(errno)); | |
843 | } | |
844 | return errno; | |
845 | } else if (retval != size) { | |
846 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " | |
847 | "%zu) on %s", retval, size, netdev_get_name(netdev_)); | |
848 | return EMSGSIZE; | |
849 | } else { | |
850 | return 0; | |
851 | } | |
852 | } | |
853 | } | |
854 | ||
855 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
856 | * when the packet transmission queue has sufficient room to transmit a packet | |
857 | * with netdev_send(). | |
858 | * | |
859 | * The kernel maintains a packet transmission queue, so the client is not | |
860 | * expected to do additional queuing of packets. Thus, this function is | |
861 | * unlikely to ever be used. It is included for completeness. */ | |
862 | static void | |
863 | netdev_linux_send_wait(struct netdev *netdev_) | |
864 | { | |
865 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 866 | if (netdev->fd < 0) { |
8b61709d | 867 | /* Nothing to do. */ |
5b7448ed JG |
868 | } else if (strcmp(netdev_get_type(netdev_), "tap")) { |
869 | poll_fd_wait(netdev->fd, POLLOUT); | |
8b61709d BP |
870 | } else { |
871 | /* TAP device always accepts packets.*/ | |
872 | poll_immediate_wake(); | |
873 | } | |
874 | } | |
875 | ||
876 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
877 | * otherwise a positive errno value. */ | |
878 | static int | |
879 | netdev_linux_set_etheraddr(struct netdev *netdev_, | |
880 | const uint8_t mac[ETH_ADDR_LEN]) | |
881 | { | |
149f577a JG |
882 | struct netdev_dev_linux *netdev_dev = |
883 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
eb395f2e BP |
884 | int error; |
885 | ||
149f577a JG |
886 | if (!(netdev_dev->cache_valid & VALID_ETHERADDR) |
887 | || !eth_addr_equals(netdev_dev->etheraddr, mac)) { | |
eb395f2e BP |
888 | error = set_etheraddr(netdev_get_name(netdev_), ARPHRD_ETHER, mac); |
889 | if (!error) { | |
149f577a JG |
890 | netdev_dev->cache_valid |= VALID_ETHERADDR; |
891 | memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); | |
eb395f2e BP |
892 | } |
893 | } else { | |
894 | error = 0; | |
8b61709d BP |
895 | } |
896 | return error; | |
897 | } | |
898 | ||
899 | /* Returns a pointer to 'netdev''s MAC address. The caller must not modify or | |
900 | * free the returned buffer. */ | |
901 | static int | |
902 | netdev_linux_get_etheraddr(const struct netdev *netdev_, | |
903 | uint8_t mac[ETH_ADDR_LEN]) | |
904 | { | |
149f577a JG |
905 | struct netdev_dev_linux *netdev_dev = |
906 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
907 | if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) { | |
8b61709d | 908 | int error = get_etheraddr(netdev_get_name(netdev_), |
149f577a | 909 | netdev_dev->etheraddr); |
8b61709d BP |
910 | if (error) { |
911 | return error; | |
912 | } | |
149f577a | 913 | netdev_dev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 914 | } |
149f577a | 915 | memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); |
8b61709d BP |
916 | return 0; |
917 | } | |
918 | ||
919 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
920 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
921 | * bytes for Ethernet devices. */ | |
922 | static int | |
923 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
924 | { | |
149f577a JG |
925 | struct netdev_dev_linux *netdev_dev = |
926 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
927 | if (!(netdev_dev->cache_valid & VALID_MTU)) { | |
8b61709d BP |
928 | struct ifreq ifr; |
929 | int error; | |
930 | ||
149f577a JG |
931 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
932 | SIOCGIFMTU, "SIOCGIFMTU"); | |
8b61709d BP |
933 | if (error) { |
934 | return error; | |
935 | } | |
149f577a JG |
936 | netdev_dev->mtu = ifr.ifr_mtu; |
937 | netdev_dev->cache_valid |= VALID_MTU; | |
8b61709d | 938 | } |
149f577a | 939 | *mtup = netdev_dev->mtu; |
8b61709d BP |
940 | return 0; |
941 | } | |
942 | ||
9ab3d9a3 BP |
943 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
944 | * On failure, returns a negative errno value. */ | |
945 | static int | |
946 | netdev_linux_get_ifindex(const struct netdev *netdev) | |
947 | { | |
948 | int ifindex, error; | |
949 | ||
950 | error = get_ifindex(netdev, &ifindex); | |
951 | return error ? -error : ifindex; | |
952 | } | |
953 | ||
8b61709d BP |
954 | static int |
955 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
956 | { | |
149f577a JG |
957 | struct netdev_dev_linux *netdev_dev = |
958 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
959 | int error = 0; |
960 | char *fn = NULL; | |
961 | int fd = -1; | |
962 | ||
149f577a | 963 | if (!(netdev_dev->cache_valid & VALID_CARRIER)) { |
8b61709d BP |
964 | char line[8]; |
965 | int retval; | |
966 | ||
149f577a JG |
967 | fn = xasprintf("/sys/class/net/%s/carrier", |
968 | netdev_get_name(netdev_)); | |
8b61709d BP |
969 | fd = open(fn, O_RDONLY); |
970 | if (fd < 0) { | |
971 | error = errno; | |
972 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(error)); | |
973 | goto exit; | |
974 | } | |
975 | ||
976 | retval = read(fd, line, sizeof line); | |
977 | if (retval < 0) { | |
978 | error = errno; | |
979 | if (error == EINVAL) { | |
980 | /* This is the normal return value when we try to check carrier | |
981 | * if the network device is not up. */ | |
982 | } else { | |
983 | VLOG_WARN_RL(&rl, "%s: read failed: %s", fn, strerror(error)); | |
984 | } | |
985 | goto exit; | |
986 | } else if (retval == 0) { | |
987 | error = EPROTO; | |
988 | VLOG_WARN_RL(&rl, "%s: unexpected end of file", fn); | |
989 | goto exit; | |
990 | } | |
991 | ||
992 | if (line[0] != '0' && line[0] != '1') { | |
993 | error = EPROTO; | |
994 | VLOG_WARN_RL(&rl, "%s: value is %c (expected 0 or 1)", | |
995 | fn, line[0]); | |
996 | goto exit; | |
997 | } | |
149f577a JG |
998 | netdev_dev->carrier = line[0] != '0'; |
999 | netdev_dev->cache_valid |= VALID_CARRIER; | |
8b61709d | 1000 | } |
149f577a | 1001 | *carrier = netdev_dev->carrier; |
8b61709d BP |
1002 | error = 0; |
1003 | ||
1004 | exit: | |
1005 | if (fd >= 0) { | |
1006 | close(fd); | |
1007 | } | |
1008 | free(fn); | |
1009 | return error; | |
1010 | } | |
1011 | ||
63331829 | 1012 | static int |
782e6111 EJ |
1013 | netdev_linux_do_miimon(const struct netdev *netdev, int cmd, |
1014 | const char *cmd_name, struct mii_ioctl_data *data) | |
63331829 | 1015 | { |
63331829 | 1016 | struct ifreq ifr; |
782e6111 | 1017 | int error; |
63331829 | 1018 | |
63331829 | 1019 | memset(&ifr, 0, sizeof ifr); |
782e6111 EJ |
1020 | memcpy(&ifr.ifr_data, data, sizeof *data); |
1021 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), | |
1022 | &ifr, cmd, cmd_name); | |
1023 | memcpy(data, &ifr.ifr_data, sizeof *data); | |
63331829 | 1024 | |
782e6111 EJ |
1025 | return error; |
1026 | } | |
1027 | ||
1028 | static int | |
1029 | netdev_linux_get_miimon(const struct netdev *netdev, bool *miimon) | |
1030 | { | |
1031 | const char *name = netdev_get_name(netdev); | |
1032 | struct mii_ioctl_data data; | |
1033 | int error; | |
63331829 | 1034 | |
782e6111 EJ |
1035 | *miimon = false; |
1036 | ||
1037 | memset(&data, 0, sizeof data); | |
1038 | error = netdev_linux_do_miimon(netdev, SIOCGMIIPHY, "SIOCGMIIPHY", &data); | |
1039 | if (!error) { | |
1040 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1041 | data.reg_num = MII_BMSR; | |
1042 | error = netdev_linux_do_miimon(netdev, SIOCGMIIREG, "SIOCGMIIREG", | |
1043 | &data); | |
63331829 EJ |
1044 | |
1045 | if (!error) { | |
782e6111 | 1046 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 EJ |
1047 | } else { |
1048 | VLOG_WARN_RL(&rl, "%s: failed to query MII", name); | |
1049 | } | |
1050 | } else { | |
1051 | struct ethtool_cmd ecmd; | |
63331829 EJ |
1052 | |
1053 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1054 | name); | |
1055 | ||
1056 | memset(&ecmd, 0, sizeof ecmd); | |
1057 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1058 | "ETHTOOL_GLINK"); | |
1059 | if (!error) { | |
782e6111 EJ |
1060 | struct ethtool_value eval; |
1061 | ||
1062 | memcpy(&eval, &ecmd, sizeof eval); | |
1063 | *miimon = !!eval.data; | |
63331829 EJ |
1064 | } else { |
1065 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | return error; | |
1070 | } | |
1071 | ||
8b61709d BP |
1072 | /* Check whether we can we use RTM_GETLINK to get network device statistics. |
1073 | * In pre-2.6.19 kernels, this was only available if wireless extensions were | |
1074 | * enabled. */ | |
1075 | static bool | |
1076 | check_for_working_netlink_stats(void) | |
1077 | { | |
1078 | /* Decide on the netdev_get_stats() implementation to use. Netlink is | |
1079 | * preferable, so if that works, we'll use it. */ | |
1080 | int ifindex = do_get_ifindex("lo"); | |
1081 | if (ifindex < 0) { | |
1082 | VLOG_WARN("failed to get ifindex for lo, " | |
1083 | "obtaining netdev stats from proc"); | |
1084 | return false; | |
1085 | } else { | |
1086 | struct netdev_stats stats; | |
1087 | int error = get_stats_via_netlink(ifindex, &stats); | |
1088 | if (!error) { | |
1089 | VLOG_DBG("obtaining netdev stats via rtnetlink"); | |
1090 | return true; | |
1091 | } else { | |
1092 | VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " | |
1093 | "via proc (you are probably running a pre-2.6.19 " | |
1094 | "kernel)", strerror(error)); | |
1095 | return false; | |
1096 | } | |
1097 | } | |
1098 | } | |
1099 | ||
8722022c BP |
1100 | /* Brings the 'is_internal' and 'is_tap' members of 'netdev_dev' up-to-date. */ |
1101 | static void | |
1102 | netdev_linux_update_is_pseudo(struct netdev_dev_linux *netdev_dev) | |
1103 | { | |
1104 | if (!(netdev_dev->cache_valid & VALID_IS_PSEUDO)) { | |
1105 | const char *name = netdev_dev_get_name(&netdev_dev->netdev_dev); | |
1106 | const char *type = netdev_dev_get_type(&netdev_dev->netdev_dev); | |
d295e8e9 | 1107 | |
8722022c | 1108 | netdev_dev->is_tap = !strcmp(type, "tap"); |
9fe3b9a2 BP |
1109 | netdev_dev->is_internal = (!netdev_dev->is_tap |
1110 | && dpif_linux_is_internal_device(name)); | |
8722022c BP |
1111 | netdev_dev->cache_valid |= VALID_IS_PSEUDO; |
1112 | } | |
1113 | } | |
1114 | ||
92df599c JG |
1115 | static void |
1116 | swap_uint64(uint64_t *a, uint64_t *b) | |
1117 | { | |
1118 | *a ^= *b; | |
1119 | *b ^= *a; | |
1120 | *a ^= *b; | |
1121 | } | |
1122 | ||
7fbef77a | 1123 | /* Retrieves current device stats for 'netdev'. */ |
8b61709d | 1124 | static int |
149f577a JG |
1125 | netdev_linux_get_stats(const struct netdev *netdev_, |
1126 | struct netdev_stats *stats) | |
8b61709d | 1127 | { |
149f577a JG |
1128 | struct netdev_dev_linux *netdev_dev = |
1129 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
1130 | static int use_netlink_stats = -1; |
1131 | int error; | |
1132 | ||
7fbef77a JG |
1133 | if (netdev_dev->have_vport_stats || |
1134 | !(netdev_dev->cache_valid & VALID_HAVE_VPORT_STATS)) { | |
1135 | ||
1136 | error = netdev_vport_get_stats(netdev_, stats); | |
1137 | netdev_dev->have_vport_stats = !error; | |
1138 | netdev_dev->cache_valid |= VALID_HAVE_VPORT_STATS; | |
8b61709d | 1139 | } |
8b61709d | 1140 | |
7fbef77a JG |
1141 | if (!netdev_dev->have_vport_stats) { |
1142 | if (use_netlink_stats < 0) { | |
1143 | use_netlink_stats = check_for_working_netlink_stats(); | |
1144 | } | |
1145 | if (use_netlink_stats) { | |
1146 | int ifindex; | |
1147 | ||
1148 | error = get_ifindex(netdev_, &ifindex); | |
1149 | if (!error) { | |
1150 | error = get_stats_via_netlink(ifindex, stats); | |
1151 | } | |
1152 | } else { | |
1153 | error = get_stats_via_proc(netdev_get_name(netdev_), stats); | |
8b61709d | 1154 | } |
8b61709d | 1155 | } |
fe6b0e03 JG |
1156 | |
1157 | /* If this port is an internal port then the transmit and receive stats | |
1158 | * will appear to be swapped relative to the other ports since we are the | |
1159 | * one sending the data, not a remote computer. For consistency, we swap | |
7fbef77a JG |
1160 | * them back here. This does not apply if we are getting stats from the |
1161 | * vport layer because it always tracks stats from the perspective of the | |
1162 | * switch. */ | |
92df599c | 1163 | netdev_linux_update_is_pseudo(netdev_dev); |
7fbef77a JG |
1164 | if (!error && !netdev_dev->have_vport_stats && |
1165 | (netdev_dev->is_internal || netdev_dev->is_tap)) { | |
92df599c JG |
1166 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1167 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1168 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1169 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1170 | stats->rx_length_errors = 0; |
1171 | stats->rx_over_errors = 0; | |
1172 | stats->rx_crc_errors = 0; | |
1173 | stats->rx_frame_errors = 0; | |
1174 | stats->rx_fifo_errors = 0; | |
1175 | stats->rx_missed_errors = 0; | |
1176 | stats->tx_aborted_errors = 0; | |
1177 | stats->tx_carrier_errors = 0; | |
1178 | stats->tx_fifo_errors = 0; | |
1179 | stats->tx_heartbeat_errors = 0; | |
1180 | stats->tx_window_errors = 0; | |
1181 | } | |
1182 | ||
8b61709d BP |
1183 | return error; |
1184 | } | |
1185 | ||
1186 | /* Stores the features supported by 'netdev' into each of '*current', | |
1187 | * '*advertised', '*supported', and '*peer' that are non-null. Each value is a | |
1188 | * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if | |
7671589a | 1189 | * successful, otherwise a positive errno value. */ |
8b61709d | 1190 | static int |
6f2f5cce | 1191 | netdev_linux_get_features(const struct netdev *netdev, |
8b61709d BP |
1192 | uint32_t *current, uint32_t *advertised, |
1193 | uint32_t *supported, uint32_t *peer) | |
1194 | { | |
1195 | struct ethtool_cmd ecmd; | |
1196 | int error; | |
1197 | ||
1198 | memset(&ecmd, 0, sizeof ecmd); | |
0b0544d7 | 1199 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1200 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1201 | if (error) { | |
1202 | return error; | |
1203 | } | |
1204 | ||
1205 | /* Supported features. */ | |
1206 | *supported = 0; | |
1207 | if (ecmd.supported & SUPPORTED_10baseT_Half) { | |
1208 | *supported |= OFPPF_10MB_HD; | |
1209 | } | |
1210 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
1211 | *supported |= OFPPF_10MB_FD; | |
1212 | } | |
1213 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
1214 | *supported |= OFPPF_100MB_HD; | |
1215 | } | |
1216 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
1217 | *supported |= OFPPF_100MB_FD; | |
1218 | } | |
1219 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
1220 | *supported |= OFPPF_1GB_HD; | |
1221 | } | |
1222 | if (ecmd.supported & SUPPORTED_1000baseT_Full) { | |
1223 | *supported |= OFPPF_1GB_FD; | |
1224 | } | |
1225 | if (ecmd.supported & SUPPORTED_10000baseT_Full) { | |
1226 | *supported |= OFPPF_10GB_FD; | |
1227 | } | |
1228 | if (ecmd.supported & SUPPORTED_TP) { | |
1229 | *supported |= OFPPF_COPPER; | |
1230 | } | |
1231 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
1232 | *supported |= OFPPF_FIBER; | |
1233 | } | |
1234 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
1235 | *supported |= OFPPF_AUTONEG; | |
1236 | } | |
1237 | if (ecmd.supported & SUPPORTED_Pause) { | |
1238 | *supported |= OFPPF_PAUSE; | |
1239 | } | |
1240 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
1241 | *supported |= OFPPF_PAUSE_ASYM; | |
1242 | } | |
1243 | ||
1244 | /* Advertised features. */ | |
1245 | *advertised = 0; | |
1246 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { | |
1247 | *advertised |= OFPPF_10MB_HD; | |
1248 | } | |
1249 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
1250 | *advertised |= OFPPF_10MB_FD; | |
1251 | } | |
1252 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
1253 | *advertised |= OFPPF_100MB_HD; | |
1254 | } | |
1255 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
1256 | *advertised |= OFPPF_100MB_FD; | |
1257 | } | |
1258 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
1259 | *advertised |= OFPPF_1GB_HD; | |
1260 | } | |
1261 | if (ecmd.advertising & ADVERTISED_1000baseT_Full) { | |
1262 | *advertised |= OFPPF_1GB_FD; | |
1263 | } | |
1264 | if (ecmd.advertising & ADVERTISED_10000baseT_Full) { | |
1265 | *advertised |= OFPPF_10GB_FD; | |
1266 | } | |
1267 | if (ecmd.advertising & ADVERTISED_TP) { | |
1268 | *advertised |= OFPPF_COPPER; | |
1269 | } | |
1270 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
1271 | *advertised |= OFPPF_FIBER; | |
1272 | } | |
1273 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
1274 | *advertised |= OFPPF_AUTONEG; | |
1275 | } | |
1276 | if (ecmd.advertising & ADVERTISED_Pause) { | |
1277 | *advertised |= OFPPF_PAUSE; | |
1278 | } | |
1279 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
1280 | *advertised |= OFPPF_PAUSE_ASYM; | |
1281 | } | |
1282 | ||
1283 | /* Current settings. */ | |
1284 | if (ecmd.speed == SPEED_10) { | |
1285 | *current = ecmd.duplex ? OFPPF_10MB_FD : OFPPF_10MB_HD; | |
1286 | } else if (ecmd.speed == SPEED_100) { | |
1287 | *current = ecmd.duplex ? OFPPF_100MB_FD : OFPPF_100MB_HD; | |
1288 | } else if (ecmd.speed == SPEED_1000) { | |
1289 | *current = ecmd.duplex ? OFPPF_1GB_FD : OFPPF_1GB_HD; | |
1290 | } else if (ecmd.speed == SPEED_10000) { | |
1291 | *current = OFPPF_10GB_FD; | |
1292 | } else { | |
1293 | *current = 0; | |
1294 | } | |
1295 | ||
1296 | if (ecmd.port == PORT_TP) { | |
1297 | *current |= OFPPF_COPPER; | |
1298 | } else if (ecmd.port == PORT_FIBRE) { | |
1299 | *current |= OFPPF_FIBER; | |
1300 | } | |
1301 | ||
1302 | if (ecmd.autoneg) { | |
1303 | *current |= OFPPF_AUTONEG; | |
1304 | } | |
1305 | ||
1306 | /* Peer advertisements. */ | |
1307 | *peer = 0; /* XXX */ | |
1308 | ||
1309 | return 0; | |
1310 | } | |
1311 | ||
1312 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
1313 | static int | |
1314 | netdev_linux_set_advertisements(struct netdev *netdev, uint32_t advertise) | |
1315 | { | |
1316 | struct ethtool_cmd ecmd; | |
1317 | int error; | |
1318 | ||
1319 | memset(&ecmd, 0, sizeof ecmd); | |
0b0544d7 | 1320 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1321 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1322 | if (error) { | |
1323 | return error; | |
1324 | } | |
1325 | ||
1326 | ecmd.advertising = 0; | |
1327 | if (advertise & OFPPF_10MB_HD) { | |
1328 | ecmd.advertising |= ADVERTISED_10baseT_Half; | |
1329 | } | |
1330 | if (advertise & OFPPF_10MB_FD) { | |
1331 | ecmd.advertising |= ADVERTISED_10baseT_Full; | |
1332 | } | |
1333 | if (advertise & OFPPF_100MB_HD) { | |
1334 | ecmd.advertising |= ADVERTISED_100baseT_Half; | |
1335 | } | |
1336 | if (advertise & OFPPF_100MB_FD) { | |
1337 | ecmd.advertising |= ADVERTISED_100baseT_Full; | |
1338 | } | |
1339 | if (advertise & OFPPF_1GB_HD) { | |
1340 | ecmd.advertising |= ADVERTISED_1000baseT_Half; | |
1341 | } | |
1342 | if (advertise & OFPPF_1GB_FD) { | |
1343 | ecmd.advertising |= ADVERTISED_1000baseT_Full; | |
1344 | } | |
1345 | if (advertise & OFPPF_10GB_FD) { | |
1346 | ecmd.advertising |= ADVERTISED_10000baseT_Full; | |
1347 | } | |
1348 | if (advertise & OFPPF_COPPER) { | |
1349 | ecmd.advertising |= ADVERTISED_TP; | |
1350 | } | |
1351 | if (advertise & OFPPF_FIBER) { | |
1352 | ecmd.advertising |= ADVERTISED_FIBRE; | |
1353 | } | |
1354 | if (advertise & OFPPF_AUTONEG) { | |
1355 | ecmd.advertising |= ADVERTISED_Autoneg; | |
1356 | } | |
1357 | if (advertise & OFPPF_PAUSE) { | |
1358 | ecmd.advertising |= ADVERTISED_Pause; | |
1359 | } | |
1360 | if (advertise & OFPPF_PAUSE_ASYM) { | |
1361 | ecmd.advertising |= ADVERTISED_Asym_Pause; | |
1362 | } | |
0b0544d7 | 1363 | return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1364 | ETHTOOL_SSET, "ETHTOOL_SSET"); |
1365 | } | |
1366 | ||
1367 | /* If 'netdev_name' is the name of a VLAN network device (e.g. one created with | |
1368 | * vconfig(8)), sets '*vlan_vid' to the VLAN VID associated with that device | |
1369 | * and returns 0. Otherwise returns a errno value (specifically ENOENT if | |
1370 | * 'netdev_name' is the name of a network device that is not a VLAN device) and | |
1371 | * sets '*vlan_vid' to -1. */ | |
1372 | static int | |
1373 | netdev_linux_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) | |
1374 | { | |
1375 | const char *netdev_name = netdev_get_name(netdev); | |
1376 | struct ds line = DS_EMPTY_INITIALIZER; | |
1377 | FILE *stream = NULL; | |
1378 | int error; | |
1379 | char *fn; | |
1380 | ||
1381 | COVERAGE_INC(netdev_get_vlan_vid); | |
1382 | fn = xasprintf("/proc/net/vlan/%s", netdev_name); | |
1383 | stream = fopen(fn, "r"); | |
1384 | if (!stream) { | |
1385 | error = errno; | |
1386 | goto done; | |
1387 | } | |
1388 | ||
1389 | if (ds_get_line(&line, stream)) { | |
1390 | if (ferror(stream)) { | |
1391 | error = errno; | |
1392 | VLOG_ERR_RL(&rl, "error reading \"%s\": %s", fn, strerror(errno)); | |
1393 | } else { | |
1394 | error = EPROTO; | |
1395 | VLOG_ERR_RL(&rl, "unexpected end of file reading \"%s\"", fn); | |
1396 | } | |
1397 | goto done; | |
1398 | } | |
1399 | ||
1400 | if (!sscanf(ds_cstr(&line), "%*s VID: %d", vlan_vid)) { | |
1401 | error = EPROTO; | |
1402 | VLOG_ERR_RL(&rl, "parse error reading \"%s\" line 1: \"%s\"", | |
1403 | fn, ds_cstr(&line)); | |
1404 | goto done; | |
1405 | } | |
1406 | ||
1407 | error = 0; | |
1408 | ||
1409 | done: | |
1410 | free(fn); | |
1411 | if (stream) { | |
1412 | fclose(stream); | |
1413 | } | |
1414 | ds_destroy(&line); | |
1415 | if (error) { | |
1416 | *vlan_vid = -1; | |
1417 | } | |
1418 | return error; | |
1419 | } | |
1420 | ||
1421 | #define POLICE_ADD_CMD "/sbin/tc qdisc add dev %s handle ffff: ingress" | |
1422 | #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 | 1423 | |
8e460221 | 1424 | /* Remove ingress policing from 'netdev'. Returns 0 if successful, otherwise a |
6f42c8ea BP |
1425 | * positive errno value. |
1426 | * | |
1427 | * This function is equivalent to running | |
1428 | * /sbin/tc qdisc del dev %s handle ffff: ingress | |
1429 | * but it is much, much faster. | |
1430 | */ | |
8e460221 BP |
1431 | static int |
1432 | netdev_linux_remove_policing(struct netdev *netdev) | |
1433 | { | |
80a86fbe BP |
1434 | struct netdev_dev_linux *netdev_dev = |
1435 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
8e460221 | 1436 | const char *netdev_name = netdev_get_name(netdev); |
8e460221 | 1437 | |
6f42c8ea | 1438 | struct ofpbuf request; |
6f42c8ea | 1439 | struct tcmsg *tcmsg; |
6f42c8ea BP |
1440 | int error; |
1441 | ||
c1c9c9c4 | 1442 | tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
1443 | if (!tcmsg) { |
1444 | return ENODEV; | |
1445 | } | |
c1c9c9c4 | 1446 | tcmsg->tcm_handle = tc_make_handle(0xffff, 0); |
6f42c8ea BP |
1447 | tcmsg->tcm_parent = TC_H_INGRESS; |
1448 | nl_msg_put_string(&request, TCA_KIND, "ingress"); | |
1449 | nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); | |
c1c9c9c4 BP |
1450 | |
1451 | error = tc_transact(&request, NULL); | |
4d10512c | 1452 | if (error && error != ENOENT && error != EINVAL) { |
6f42c8ea BP |
1453 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", |
1454 | netdev_name, strerror(error)); | |
1455 | return error; | |
1456 | } | |
1457 | ||
80a86fbe BP |
1458 | netdev_dev->kbits_rate = 0; |
1459 | netdev_dev->kbits_burst = 0; | |
1460 | netdev_dev->cache_valid |= VALID_POLICING; | |
8e460221 BP |
1461 | return 0; |
1462 | } | |
1463 | ||
8b61709d BP |
1464 | /* Attempts to set input rate limiting (policing) policy. */ |
1465 | static int | |
1466 | netdev_linux_set_policing(struct netdev *netdev, | |
1467 | uint32_t kbits_rate, uint32_t kbits_burst) | |
1468 | { | |
80a86fbe BP |
1469 | struct netdev_dev_linux *netdev_dev = |
1470 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
8b61709d BP |
1471 | const char *netdev_name = netdev_get_name(netdev); |
1472 | char command[1024]; | |
1473 | ||
1474 | COVERAGE_INC(netdev_set_policing); | |
8e460221 | 1475 | |
80a86fbe BP |
1476 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
1477 | : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ | |
1478 | : kbits_burst); /* Stick with user-specified value. */ | |
1479 | ||
1480 | if (netdev_dev->cache_valid & VALID_POLICING | |
1481 | && netdev_dev->kbits_rate == kbits_rate | |
1482 | && netdev_dev->kbits_burst == kbits_burst) { | |
1483 | /* Assume that settings haven't changed since we last set them. */ | |
1484 | return 0; | |
1485 | } | |
1486 | ||
8e460221 | 1487 | netdev_linux_remove_policing(netdev); |
8b61709d | 1488 | if (kbits_rate) { |
8b61709d BP |
1489 | snprintf(command, sizeof(command), POLICE_ADD_CMD, netdev_name); |
1490 | if (system(command) != 0) { | |
1491 | VLOG_WARN_RL(&rl, "%s: problem adding policing", netdev_name); | |
1492 | return -1; | |
1493 | } | |
1494 | ||
1495 | snprintf(command, sizeof(command), POLICE_CONFIG_CMD, netdev_name, | |
1496 | kbits_rate, kbits_burst); | |
1497 | if (system(command) != 0) { | |
1498 | VLOG_WARN_RL(&rl, "%s: problem configuring policing", | |
1499 | netdev_name); | |
1500 | return -1; | |
1501 | } | |
80a86fbe BP |
1502 | |
1503 | netdev_dev->kbits_rate = kbits_rate; | |
1504 | netdev_dev->kbits_burst = kbits_burst; | |
1505 | netdev_dev->cache_valid |= VALID_POLICING; | |
8b61709d BP |
1506 | } |
1507 | ||
1508 | return 0; | |
1509 | } | |
1510 | ||
c1c9c9c4 BP |
1511 | static int |
1512 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
1513 | struct svec *types) | |
1514 | { | |
1515 | const struct tc_ops **opsp; | |
1516 | ||
1517 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1518 | const struct tc_ops *ops = *opsp; | |
1519 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
1520 | svec_add(types, ops->ovs_name); | |
1521 | } | |
1522 | } | |
1523 | return 0; | |
1524 | } | |
1525 | ||
1526 | static const struct tc_ops * | |
1527 | tc_lookup_ovs_name(const char *name) | |
1528 | { | |
1529 | const struct tc_ops **opsp; | |
1530 | ||
1531 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1532 | const struct tc_ops *ops = *opsp; | |
1533 | if (!strcmp(name, ops->ovs_name)) { | |
1534 | return ops; | |
1535 | } | |
1536 | } | |
1537 | return NULL; | |
1538 | } | |
1539 | ||
1540 | static const struct tc_ops * | |
1541 | tc_lookup_linux_name(const char *name) | |
1542 | { | |
1543 | const struct tc_ops **opsp; | |
1544 | ||
1545 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1546 | const struct tc_ops *ops = *opsp; | |
1547 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
1548 | return ops; | |
1549 | } | |
1550 | } | |
1551 | return NULL; | |
1552 | } | |
1553 | ||
93b13be8 BP |
1554 | static struct tc_queue * |
1555 | tc_find_queue__(const struct netdev *netdev, unsigned int queue_id, | |
1556 | size_t hash) | |
1557 | { | |
1558 | struct netdev_dev_linux *netdev_dev = | |
1559 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1560 | struct tc_queue *queue; | |
1561 | ||
4e8e4213 | 1562 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) { |
93b13be8 BP |
1563 | if (queue->queue_id == queue_id) { |
1564 | return queue; | |
1565 | } | |
1566 | } | |
1567 | return NULL; | |
1568 | } | |
1569 | ||
1570 | static struct tc_queue * | |
1571 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
1572 | { | |
1573 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
1574 | } | |
1575 | ||
c1c9c9c4 BP |
1576 | static int |
1577 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
1578 | const char *type, | |
1579 | struct netdev_qos_capabilities *caps) | |
1580 | { | |
1581 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
1582 | if (!ops) { | |
1583 | return EOPNOTSUPP; | |
1584 | } | |
1585 | caps->n_queues = ops->n_queues; | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | static int | |
1590 | netdev_linux_get_qos(const struct netdev *netdev, | |
1591 | const char **typep, struct shash *details) | |
1592 | { | |
1593 | struct netdev_dev_linux *netdev_dev = | |
1594 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1595 | int error; | |
1596 | ||
1597 | error = tc_query_qdisc(netdev); | |
1598 | if (error) { | |
1599 | return error; | |
1600 | } | |
1601 | ||
1602 | *typep = netdev_dev->tc->ops->ovs_name; | |
1603 | return (netdev_dev->tc->ops->qdisc_get | |
1604 | ? netdev_dev->tc->ops->qdisc_get(netdev, details) | |
1605 | : 0); | |
1606 | } | |
1607 | ||
1608 | static int | |
1609 | netdev_linux_set_qos(struct netdev *netdev, | |
1610 | const char *type, const struct shash *details) | |
1611 | { | |
1612 | struct netdev_dev_linux *netdev_dev = | |
1613 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1614 | const struct tc_ops *new_ops; | |
1615 | int error; | |
1616 | ||
1617 | new_ops = tc_lookup_ovs_name(type); | |
1618 | if (!new_ops || !new_ops->tc_install) { | |
1619 | return EOPNOTSUPP; | |
1620 | } | |
1621 | ||
1622 | error = tc_query_qdisc(netdev); | |
1623 | if (error) { | |
1624 | return error; | |
1625 | } | |
1626 | ||
1627 | if (new_ops == netdev_dev->tc->ops) { | |
1628 | return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0; | |
1629 | } else { | |
1630 | /* Delete existing qdisc. */ | |
1631 | error = tc_del_qdisc(netdev); | |
1632 | if (error) { | |
1633 | return error; | |
1634 | } | |
1635 | assert(netdev_dev->tc == NULL); | |
1636 | ||
1637 | /* Install new qdisc. */ | |
1638 | error = new_ops->tc_install(netdev, details); | |
1639 | assert((error == 0) == (netdev_dev->tc != NULL)); | |
1640 | ||
1641 | return error; | |
1642 | } | |
1643 | } | |
1644 | ||
1645 | static int | |
1646 | netdev_linux_get_queue(const struct netdev *netdev, | |
1647 | unsigned int queue_id, struct shash *details) | |
1648 | { | |
1649 | struct netdev_dev_linux *netdev_dev = | |
1650 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1651 | int error; | |
1652 | ||
1653 | error = tc_query_qdisc(netdev); | |
1654 | if (error) { | |
1655 | return error; | |
93b13be8 BP |
1656 | } else { |
1657 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1658 | return (queue | |
1659 | ? netdev_dev->tc->ops->class_get(netdev, queue, details) | |
1660 | : ENOENT); | |
c1c9c9c4 | 1661 | } |
c1c9c9c4 BP |
1662 | } |
1663 | ||
1664 | static int | |
1665 | netdev_linux_set_queue(struct netdev *netdev, | |
1666 | unsigned int queue_id, const struct shash *details) | |
1667 | { | |
1668 | struct netdev_dev_linux *netdev_dev = | |
1669 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1670 | int error; | |
1671 | ||
1672 | error = tc_query_qdisc(netdev); | |
1673 | if (error) { | |
1674 | return error; | |
1675 | } else if (queue_id >= netdev_dev->tc->ops->n_queues | |
1676 | || !netdev_dev->tc->ops->class_set) { | |
1677 | return EINVAL; | |
1678 | } | |
1679 | ||
1680 | return netdev_dev->tc->ops->class_set(netdev, queue_id, details); | |
1681 | } | |
1682 | ||
1683 | static int | |
1684 | netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id) | |
1685 | { | |
1686 | struct netdev_dev_linux *netdev_dev = | |
1687 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1688 | int error; | |
1689 | ||
1690 | error = tc_query_qdisc(netdev); | |
1691 | if (error) { | |
1692 | return error; | |
1693 | } else if (!netdev_dev->tc->ops->class_delete) { | |
1694 | return EINVAL; | |
93b13be8 BP |
1695 | } else { |
1696 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1697 | return (queue | |
1698 | ? netdev_dev->tc->ops->class_delete(netdev, queue) | |
1699 | : ENOENT); | |
c1c9c9c4 | 1700 | } |
c1c9c9c4 BP |
1701 | } |
1702 | ||
1703 | static int | |
1704 | netdev_linux_get_queue_stats(const struct netdev *netdev, | |
1705 | unsigned int queue_id, | |
1706 | struct netdev_queue_stats *stats) | |
1707 | { | |
1708 | struct netdev_dev_linux *netdev_dev = | |
1709 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1710 | int error; | |
1711 | ||
1712 | error = tc_query_qdisc(netdev); | |
1713 | if (error) { | |
1714 | return error; | |
c1c9c9c4 BP |
1715 | } else if (!netdev_dev->tc->ops->class_get_stats) { |
1716 | return EOPNOTSUPP; | |
93b13be8 BP |
1717 | } else { |
1718 | const struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1719 | return (queue | |
1720 | ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats) | |
1721 | : ENOENT); | |
c1c9c9c4 | 1722 | } |
c1c9c9c4 BP |
1723 | } |
1724 | ||
23a98ffe | 1725 | static bool |
c1c9c9c4 BP |
1726 | start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) |
1727 | { | |
1728 | struct ofpbuf request; | |
1729 | struct tcmsg *tcmsg; | |
1730 | ||
1731 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); | |
23a98ffe BP |
1732 | if (!tcmsg) { |
1733 | return false; | |
1734 | } | |
3c4de644 | 1735 | tcmsg->tcm_parent = 0; |
c1c9c9c4 BP |
1736 | nl_dump_start(dump, rtnl_sock, &request); |
1737 | ofpbuf_uninit(&request); | |
23a98ffe | 1738 | return true; |
c1c9c9c4 BP |
1739 | } |
1740 | ||
1741 | static int | |
1742 | netdev_linux_dump_queues(const struct netdev *netdev, | |
1743 | netdev_dump_queues_cb *cb, void *aux) | |
1744 | { | |
1745 | struct netdev_dev_linux *netdev_dev = | |
1746 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
93b13be8 | 1747 | struct tc_queue *queue; |
c1c9c9c4 BP |
1748 | struct shash details; |
1749 | int last_error; | |
c1c9c9c4 BP |
1750 | int error; |
1751 | ||
1752 | error = tc_query_qdisc(netdev); | |
1753 | if (error) { | |
1754 | return error; | |
1755 | } else if (!netdev_dev->tc->ops->class_get) { | |
1756 | return EOPNOTSUPP; | |
1757 | } | |
1758 | ||
1759 | last_error = 0; | |
1760 | shash_init(&details); | |
4e8e4213 | 1761 | HMAP_FOR_EACH (queue, hmap_node, &netdev_dev->tc->queues) { |
c1c9c9c4 BP |
1762 | shash_clear(&details); |
1763 | ||
93b13be8 | 1764 | error = netdev_dev->tc->ops->class_get(netdev, queue, &details); |
c1c9c9c4 | 1765 | if (!error) { |
93b13be8 | 1766 | (*cb)(queue->queue_id, &details, aux); |
c1c9c9c4 BP |
1767 | } else { |
1768 | last_error = error; | |
1769 | } | |
1770 | } | |
1771 | shash_destroy(&details); | |
1772 | ||
1773 | return last_error; | |
1774 | } | |
1775 | ||
1776 | static int | |
1777 | netdev_linux_dump_queue_stats(const struct netdev *netdev, | |
1778 | netdev_dump_queue_stats_cb *cb, void *aux) | |
1779 | { | |
1780 | struct netdev_dev_linux *netdev_dev = | |
1781 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1782 | struct nl_dump dump; | |
1783 | struct ofpbuf msg; | |
1784 | int last_error; | |
1785 | int error; | |
1786 | ||
1787 | error = tc_query_qdisc(netdev); | |
1788 | if (error) { | |
1789 | return error; | |
1790 | } else if (!netdev_dev->tc->ops->class_dump_stats) { | |
1791 | return EOPNOTSUPP; | |
1792 | } | |
1793 | ||
1794 | last_error = 0; | |
23a98ffe BP |
1795 | if (!start_queue_dump(netdev, &dump)) { |
1796 | return ENODEV; | |
1797 | } | |
c1c9c9c4 BP |
1798 | while (nl_dump_next(&dump, &msg)) { |
1799 | error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux); | |
1800 | if (error) { | |
1801 | last_error = error; | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | error = nl_dump_done(&dump); | |
1806 | return error ? error : last_error; | |
1807 | } | |
1808 | ||
8b61709d | 1809 | static int |
f1acd62b BP |
1810 | netdev_linux_get_in4(const struct netdev *netdev_, |
1811 | struct in_addr *address, struct in_addr *netmask) | |
8b61709d | 1812 | { |
149f577a JG |
1813 | struct netdev_dev_linux *netdev_dev = |
1814 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1815 | ||
1816 | if (!(netdev_dev->cache_valid & VALID_IN4)) { | |
8b61709d BP |
1817 | int error; |
1818 | ||
149f577a | 1819 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, |
8b61709d BP |
1820 | SIOCGIFADDR, "SIOCGIFADDR"); |
1821 | if (error) { | |
1822 | return error; | |
1823 | } | |
1824 | ||
149f577a | 1825 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, |
f1acd62b BP |
1826 | SIOCGIFNETMASK, "SIOCGIFNETMASK"); |
1827 | if (error) { | |
1828 | return error; | |
1829 | } | |
1830 | ||
149f577a | 1831 | netdev_dev->cache_valid |= VALID_IN4; |
8b61709d | 1832 | } |
149f577a JG |
1833 | *address = netdev_dev->address; |
1834 | *netmask = netdev_dev->netmask; | |
f1acd62b | 1835 | return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; |
8b61709d BP |
1836 | } |
1837 | ||
8b61709d | 1838 | static int |
f1acd62b BP |
1839 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
1840 | struct in_addr netmask) | |
8b61709d | 1841 | { |
149f577a JG |
1842 | struct netdev_dev_linux *netdev_dev = |
1843 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
1844 | int error; |
1845 | ||
f1acd62b | 1846 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 1847 | if (!error) { |
149f577a JG |
1848 | netdev_dev->cache_valid |= VALID_IN4; |
1849 | netdev_dev->address = address; | |
1850 | netdev_dev->netmask = netmask; | |
f1acd62b | 1851 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 1852 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 1853 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
1854 | } |
1855 | } | |
1856 | return error; | |
1857 | } | |
1858 | ||
1859 | static bool | |
1860 | parse_if_inet6_line(const char *line, | |
1861 | struct in6_addr *in6, char ifname[16 + 1]) | |
1862 | { | |
1863 | uint8_t *s6 = in6->s6_addr; | |
1864 | #define X8 "%2"SCNx8 | |
1865 | return sscanf(line, | |
1866 | " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 | |
1867 | "%*x %*x %*x %*x %16s\n", | |
1868 | &s6[0], &s6[1], &s6[2], &s6[3], | |
1869 | &s6[4], &s6[5], &s6[6], &s6[7], | |
1870 | &s6[8], &s6[9], &s6[10], &s6[11], | |
1871 | &s6[12], &s6[13], &s6[14], &s6[15], | |
1872 | ifname) == 17; | |
1873 | } | |
1874 | ||
1875 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if | |
1876 | * 'in6' is non-null) and returns true. Otherwise, returns false. */ | |
1877 | static int | |
1878 | netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) | |
1879 | { | |
149f577a JG |
1880 | struct netdev_dev_linux *netdev_dev = |
1881 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1882 | if (!(netdev_dev->cache_valid & VALID_IN6)) { | |
8b61709d BP |
1883 | FILE *file; |
1884 | char line[128]; | |
1885 | ||
149f577a | 1886 | netdev_dev->in6 = in6addr_any; |
8b61709d BP |
1887 | |
1888 | file = fopen("/proc/net/if_inet6", "r"); | |
1889 | if (file != NULL) { | |
1890 | const char *name = netdev_get_name(netdev_); | |
1891 | while (fgets(line, sizeof line, file)) { | |
2a022368 | 1892 | struct in6_addr in6_tmp; |
8b61709d | 1893 | char ifname[16 + 1]; |
2a022368 | 1894 | if (parse_if_inet6_line(line, &in6_tmp, ifname) |
8b61709d BP |
1895 | && !strcmp(name, ifname)) |
1896 | { | |
2a022368 | 1897 | netdev_dev->in6 = in6_tmp; |
8b61709d BP |
1898 | break; |
1899 | } | |
1900 | } | |
1901 | fclose(file); | |
1902 | } | |
149f577a | 1903 | netdev_dev->cache_valid |= VALID_IN6; |
8b61709d | 1904 | } |
149f577a | 1905 | *in6 = netdev_dev->in6; |
8b61709d BP |
1906 | return 0; |
1907 | } | |
1908 | ||
1909 | static void | |
1910 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
1911 | { | |
1912 | struct sockaddr_in sin; | |
1913 | memset(&sin, 0, sizeof sin); | |
1914 | sin.sin_family = AF_INET; | |
1915 | sin.sin_addr = addr; | |
1916 | sin.sin_port = 0; | |
1917 | ||
1918 | memset(sa, 0, sizeof *sa); | |
1919 | memcpy(sa, &sin, sizeof sin); | |
1920 | } | |
1921 | ||
1922 | static int | |
1923 | do_set_addr(struct netdev *netdev, | |
1924 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
1925 | { | |
1926 | struct ifreq ifr; | |
71d7c22f | 1927 | ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); |
8b61709d | 1928 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
149f577a JG |
1929 | |
1930 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
1931 | ioctl_name); | |
8b61709d BP |
1932 | } |
1933 | ||
1934 | /* Adds 'router' as a default IP gateway. */ | |
1935 | static int | |
67a4917b | 1936 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
1937 | { |
1938 | struct in_addr any = { INADDR_ANY }; | |
1939 | struct rtentry rt; | |
1940 | int error; | |
1941 | ||
1942 | memset(&rt, 0, sizeof rt); | |
1943 | make_in4_sockaddr(&rt.rt_dst, any); | |
1944 | make_in4_sockaddr(&rt.rt_gateway, router); | |
1945 | make_in4_sockaddr(&rt.rt_genmask, any); | |
1946 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
8b61709d BP |
1947 | error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; |
1948 | if (error) { | |
1949 | VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); | |
1950 | } | |
1951 | return error; | |
1952 | } | |
1953 | ||
f1acd62b BP |
1954 | static int |
1955 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
1956 | char **netdev_name) | |
1957 | { | |
1958 | static const char fn[] = "/proc/net/route"; | |
1959 | FILE *stream; | |
1960 | char line[256]; | |
1961 | int ln; | |
1962 | ||
1963 | *netdev_name = NULL; | |
1964 | stream = fopen(fn, "r"); | |
1965 | if (stream == NULL) { | |
1966 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
1967 | return errno; | |
1968 | } | |
1969 | ||
1970 | ln = 0; | |
1971 | while (fgets(line, sizeof line, stream)) { | |
1972 | if (++ln >= 2) { | |
1973 | char iface[17]; | |
1974 | uint32_t dest, gateway, mask; | |
1975 | int refcnt, metric, mtu; | |
1976 | unsigned int flags, use, window, irtt; | |
1977 | ||
1978 | if (sscanf(line, | |
1979 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
1980 | " %d %u %u\n", | |
1981 | iface, &dest, &gateway, &flags, &refcnt, | |
1982 | &use, &metric, &mask, &mtu, &window, &irtt) != 11) { | |
1983 | ||
d295e8e9 | 1984 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
1985 | fn, ln, line); |
1986 | continue; | |
1987 | } | |
1988 | if (!(flags & RTF_UP)) { | |
1989 | /* Skip routes that aren't up. */ | |
1990 | continue; | |
1991 | } | |
1992 | ||
1993 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 1994 | * network byte order, so we don't need need any endian |
f1acd62b BP |
1995 | * conversions here. */ |
1996 | if ((dest & mask) == (host->s_addr & mask)) { | |
1997 | if (!gateway) { | |
1998 | /* The host is directly reachable. */ | |
1999 | next_hop->s_addr = 0; | |
2000 | } else { | |
2001 | /* To reach the host, we must go through a gateway. */ | |
2002 | next_hop->s_addr = gateway; | |
2003 | } | |
2004 | *netdev_name = xstrdup(iface); | |
2005 | fclose(stream); | |
2006 | return 0; | |
2007 | } | |
2008 | } | |
2009 | } | |
2010 | ||
2011 | fclose(stream); | |
2012 | return ENXIO; | |
2013 | } | |
2014 | ||
e210037e AE |
2015 | static int |
2016 | netdev_linux_get_status(const struct netdev *netdev, struct shash *sh) | |
2017 | { | |
2018 | struct ethtool_drvinfo drvinfo; | |
2019 | int error; | |
2020 | ||
2021 | memset(&drvinfo, 0, sizeof drvinfo); | |
2022 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), | |
2023 | (struct ethtool_cmd *)&drvinfo, | |
2024 | ETHTOOL_GDRVINFO, | |
2025 | "ETHTOOL_GDRVINFO"); | |
2026 | if (!error) { | |
2027 | shash_add(sh, "driver_name", xstrdup(drvinfo.driver)); | |
2028 | shash_add(sh, "driver_version", xstrdup(drvinfo.version)); | |
2029 | shash_add(sh, "firmware_version", xstrdup(drvinfo.fw_version)); | |
2030 | } | |
2031 | ||
2032 | return error; | |
2033 | } | |
2034 | ||
8b61709d BP |
2035 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2036 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2037 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2038 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2039 | static int | |
2040 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
2041 | uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) | |
2042 | { | |
2043 | struct arpreq r; | |
c100e025 | 2044 | struct sockaddr_in sin; |
8b61709d BP |
2045 | int retval; |
2046 | ||
2047 | memset(&r, 0, sizeof r); | |
f2cc621b | 2048 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2049 | sin.sin_family = AF_INET; |
2050 | sin.sin_addr.s_addr = ip; | |
2051 | sin.sin_port = 0; | |
2052 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2053 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2054 | r.arp_flags = 0; | |
71d7c22f | 2055 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d BP |
2056 | COVERAGE_INC(netdev_arp_lookup); |
2057 | retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; | |
2058 | if (!retval) { | |
2059 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2060 | } else if (retval != ENXIO) { | |
2061 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
149f577a | 2062 | netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval)); |
8b61709d BP |
2063 | } |
2064 | return retval; | |
2065 | } | |
2066 | ||
2067 | static int | |
2068 | nd_to_iff_flags(enum netdev_flags nd) | |
2069 | { | |
2070 | int iff = 0; | |
2071 | if (nd & NETDEV_UP) { | |
2072 | iff |= IFF_UP; | |
2073 | } | |
2074 | if (nd & NETDEV_PROMISC) { | |
2075 | iff |= IFF_PROMISC; | |
2076 | } | |
2077 | return iff; | |
2078 | } | |
2079 | ||
2080 | static int | |
2081 | iff_to_nd_flags(int iff) | |
2082 | { | |
2083 | enum netdev_flags nd = 0; | |
2084 | if (iff & IFF_UP) { | |
2085 | nd |= NETDEV_UP; | |
2086 | } | |
2087 | if (iff & IFF_PROMISC) { | |
2088 | nd |= NETDEV_PROMISC; | |
2089 | } | |
2090 | return nd; | |
2091 | } | |
2092 | ||
2093 | static int | |
2094 | netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, | |
2095 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2096 | { | |
2097 | int old_flags, new_flags; | |
2098 | int error; | |
2099 | ||
2100 | error = get_flags(netdev, &old_flags); | |
2101 | if (!error) { | |
2102 | *old_flagsp = iff_to_nd_flags(old_flags); | |
2103 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2104 | if (new_flags != old_flags) { | |
2105 | error = set_flags(netdev, new_flags); | |
2106 | } | |
2107 | } | |
2108 | return error; | |
2109 | } | |
2110 | ||
2111 | static void | |
2112 | poll_notify(struct list *list) | |
2113 | { | |
2114 | struct netdev_linux_notifier *notifier; | |
4e8e4213 | 2115 | LIST_FOR_EACH (notifier, node, list) { |
8b61709d BP |
2116 | struct netdev_notifier *n = ¬ifier->notifier; |
2117 | n->cb(n); | |
2118 | } | |
2119 | } | |
2120 | ||
2121 | static void | |
21d6e22e | 2122 | netdev_linux_poll_cb(const struct rtnetlink_link_change *change, |
67a4917b | 2123 | void *aux OVS_UNUSED) |
8b61709d BP |
2124 | { |
2125 | if (change) { | |
2126 | struct list *list = shash_find_data(&netdev_linux_notifiers, | |
2127 | change->ifname); | |
2128 | if (list) { | |
2129 | poll_notify(list); | |
2130 | } | |
2131 | } else { | |
2132 | struct shash_node *node; | |
2133 | SHASH_FOR_EACH (node, &netdev_linux_notifiers) { | |
2134 | poll_notify(node->data); | |
2135 | } | |
2136 | } | |
2137 | } | |
2138 | ||
2139 | static int | |
2140 | netdev_linux_poll_add(struct netdev *netdev, | |
2141 | void (*cb)(struct netdev_notifier *), void *aux, | |
2142 | struct netdev_notifier **notifierp) | |
2143 | { | |
2144 | const char *netdev_name = netdev_get_name(netdev); | |
2145 | struct netdev_linux_notifier *notifier; | |
2146 | struct list *list; | |
2147 | ||
2148 | if (shash_is_empty(&netdev_linux_notifiers)) { | |
21d6e22e EJ |
2149 | int error; |
2150 | error = rtnetlink_link_notifier_register(&netdev_linux_poll_notifier, | |
2151 | netdev_linux_poll_cb, NULL); | |
8b61709d BP |
2152 | if (error) { |
2153 | return error; | |
2154 | } | |
2155 | } | |
2156 | ||
2157 | list = shash_find_data(&netdev_linux_notifiers, netdev_name); | |
2158 | if (!list) { | |
2159 | list = xmalloc(sizeof *list); | |
2160 | list_init(list); | |
2161 | shash_add(&netdev_linux_notifiers, netdev_name, list); | |
2162 | } | |
2163 | ||
2164 | notifier = xmalloc(sizeof *notifier); | |
2165 | netdev_notifier_init(¬ifier->notifier, netdev, cb, aux); | |
2166 | list_push_back(list, ¬ifier->node); | |
2167 | *notifierp = ¬ifier->notifier; | |
2168 | return 0; | |
2169 | } | |
2170 | ||
2171 | static void | |
2172 | netdev_linux_poll_remove(struct netdev_notifier *notifier_) | |
2173 | { | |
2174 | struct netdev_linux_notifier *notifier = | |
2175 | CONTAINER_OF(notifier_, struct netdev_linux_notifier, notifier); | |
2176 | struct list *list; | |
2177 | ||
2178 | /* Remove 'notifier' from its list. */ | |
2179 | list = list_remove(¬ifier->node); | |
2180 | if (list_is_empty(list)) { | |
2181 | /* The list is now empty. Remove it from the hash and free it. */ | |
2182 | const char *netdev_name = netdev_get_name(notifier->notifier.netdev); | |
2183 | shash_delete(&netdev_linux_notifiers, | |
2184 | shash_find(&netdev_linux_notifiers, netdev_name)); | |
2185 | free(list); | |
2186 | } | |
2187 | free(notifier); | |
2188 | ||
2189 | /* If that was the last notifier, unregister. */ | |
2190 | if (shash_is_empty(&netdev_linux_notifiers)) { | |
21d6e22e | 2191 | rtnetlink_link_notifier_unregister(&netdev_linux_poll_notifier); |
8b61709d BP |
2192 | } |
2193 | } | |
2194 | ||
c3827f61 BP |
2195 | #define NETDEV_LINUX_CLASS(NAME, CREATE, ENUMERATE, SET_STATS) \ |
2196 | { \ | |
2197 | NAME, \ | |
2198 | \ | |
2199 | netdev_linux_init, \ | |
2200 | netdev_linux_run, \ | |
2201 | netdev_linux_wait, \ | |
2202 | \ | |
2203 | CREATE, \ | |
2204 | netdev_linux_destroy, \ | |
6d9e6eb4 | 2205 | NULL, /* set_config */ \ |
c3827f61 BP |
2206 | \ |
2207 | netdev_linux_open, \ | |
2208 | netdev_linux_close, \ | |
2209 | \ | |
2210 | ENUMERATE, \ | |
2211 | \ | |
2212 | netdev_linux_recv, \ | |
2213 | netdev_linux_recv_wait, \ | |
2214 | netdev_linux_drain, \ | |
2215 | \ | |
2216 | netdev_linux_send, \ | |
2217 | netdev_linux_send_wait, \ | |
2218 | \ | |
2219 | netdev_linux_set_etheraddr, \ | |
2220 | netdev_linux_get_etheraddr, \ | |
2221 | netdev_linux_get_mtu, \ | |
2222 | netdev_linux_get_ifindex, \ | |
2223 | netdev_linux_get_carrier, \ | |
63331829 | 2224 | netdev_linux_get_miimon, \ |
c3827f61 BP |
2225 | netdev_linux_get_stats, \ |
2226 | SET_STATS, \ | |
2227 | \ | |
2228 | netdev_linux_get_features, \ | |
2229 | netdev_linux_set_advertisements, \ | |
2230 | netdev_linux_get_vlan_vid, \ | |
2231 | \ | |
2232 | netdev_linux_set_policing, \ | |
2233 | netdev_linux_get_qos_types, \ | |
2234 | netdev_linux_get_qos_capabilities, \ | |
2235 | netdev_linux_get_qos, \ | |
2236 | netdev_linux_set_qos, \ | |
2237 | netdev_linux_get_queue, \ | |
2238 | netdev_linux_set_queue, \ | |
2239 | netdev_linux_delete_queue, \ | |
2240 | netdev_linux_get_queue_stats, \ | |
2241 | netdev_linux_dump_queues, \ | |
2242 | netdev_linux_dump_queue_stats, \ | |
2243 | \ | |
2244 | netdev_linux_get_in4, \ | |
2245 | netdev_linux_set_in4, \ | |
2246 | netdev_linux_get_in6, \ | |
2247 | netdev_linux_add_router, \ | |
2248 | netdev_linux_get_next_hop, \ | |
e210037e | 2249 | netdev_linux_get_status, \ |
c3827f61 BP |
2250 | netdev_linux_arp_lookup, \ |
2251 | \ | |
2252 | netdev_linux_update_flags, \ | |
2253 | \ | |
2254 | netdev_linux_poll_add, \ | |
2255 | netdev_linux_poll_remove \ | |
2256 | } | |
2257 | ||
2258 | const struct netdev_class netdev_linux_class = | |
2259 | NETDEV_LINUX_CLASS( | |
2260 | "system", | |
2261 | netdev_linux_create, | |
2262 | netdev_linux_enumerate, | |
98563392 | 2263 | NULL); /* set_stats */ |
c3827f61 BP |
2264 | |
2265 | const struct netdev_class netdev_tap_class = | |
2266 | NETDEV_LINUX_CLASS( | |
2267 | "tap", | |
2268 | netdev_linux_create_tap, | |
2269 | NULL, /* enumerate */ | |
2270 | NULL); /* set_stats */ | |
2271 | ||
2272 | const struct netdev_class netdev_internal_class = | |
2273 | NETDEV_LINUX_CLASS( | |
2274 | "internal", | |
2275 | netdev_linux_create, | |
2276 | NULL, /* enumerate */ | |
2277 | netdev_vport_set_stats); | |
8b61709d | 2278 | \f |
c1c9c9c4 | 2279 | /* HTB traffic control class. */ |
559843ed | 2280 | |
c1c9c9c4 | 2281 | #define HTB_N_QUEUES 0xf000 |
8b61709d | 2282 | |
c1c9c9c4 BP |
2283 | struct htb { |
2284 | struct tc tc; | |
2285 | unsigned int max_rate; /* In bytes/s. */ | |
2286 | }; | |
8b61709d | 2287 | |
c1c9c9c4 | 2288 | struct htb_class { |
93b13be8 | 2289 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
2290 | unsigned int min_rate; /* In bytes/s. */ |
2291 | unsigned int max_rate; /* In bytes/s. */ | |
2292 | unsigned int burst; /* In bytes. */ | |
2293 | unsigned int priority; /* Lower values are higher priorities. */ | |
2294 | }; | |
8b61709d | 2295 | |
c1c9c9c4 BP |
2296 | static struct htb * |
2297 | htb_get__(const struct netdev *netdev) | |
2298 | { | |
2299 | struct netdev_dev_linux *netdev_dev = | |
2300 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2301 | return CONTAINER_OF(netdev_dev->tc, struct htb, tc); | |
2302 | } | |
2303 | ||
2304 | static struct htb * | |
2305 | htb_install__(struct netdev *netdev, uint64_t max_rate) | |
2306 | { | |
2307 | struct netdev_dev_linux *netdev_dev = | |
2308 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2309 | struct htb *htb; | |
2310 | ||
2311 | htb = xmalloc(sizeof *htb); | |
2312 | tc_init(&htb->tc, &tc_ops_htb); | |
2313 | htb->max_rate = max_rate; | |
2314 | ||
2315 | netdev_dev->tc = &htb->tc; | |
2316 | ||
2317 | return htb; | |
2318 | } | |
2319 | ||
2320 | /* Create an HTB qdisc. | |
2321 | * | |
a339aa81 | 2322 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
2323 | static int |
2324 | htb_setup_qdisc__(struct netdev *netdev) | |
2325 | { | |
2326 | size_t opt_offset; | |
2327 | struct tc_htb_glob opt; | |
2328 | struct ofpbuf request; | |
2329 | struct tcmsg *tcmsg; | |
2330 | ||
2331 | tc_del_qdisc(netdev); | |
2332 | ||
2333 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
2334 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
2335 | if (!tcmsg) { |
2336 | return ENODEV; | |
2337 | } | |
c1c9c9c4 BP |
2338 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
2339 | tcmsg->tcm_parent = TC_H_ROOT; | |
2340 | ||
2341 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2342 | ||
2343 | memset(&opt, 0, sizeof opt); | |
2344 | opt.rate2quantum = 10; | |
2345 | opt.version = 3; | |
4ecf12d5 | 2346 | opt.defcls = 1; |
c1c9c9c4 BP |
2347 | |
2348 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2349 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
2350 | nl_msg_end_nested(&request, opt_offset); | |
2351 | ||
2352 | return tc_transact(&request, NULL); | |
2353 | } | |
2354 | ||
2355 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
2356 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
2357 | static int | |
2358 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
2359 | unsigned int parent, struct htb_class *class) | |
2360 | { | |
2361 | size_t opt_offset; | |
2362 | struct tc_htb_opt opt; | |
2363 | struct ofpbuf request; | |
2364 | struct tcmsg *tcmsg; | |
2365 | int error; | |
2366 | int mtu; | |
2367 | ||
2368 | netdev_get_mtu(netdev, &mtu); | |
f915f1a8 BP |
2369 | if (mtu == INT_MAX) { |
2370 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", | |
2371 | netdev_get_name(netdev)); | |
2372 | return EINVAL; | |
2373 | } | |
c1c9c9c4 BP |
2374 | |
2375 | memset(&opt, 0, sizeof opt); | |
2376 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
2377 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
2378 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); | |
2379 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
2380 | opt.prio = class->priority; | |
2381 | ||
2382 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
23a98ffe BP |
2383 | if (!tcmsg) { |
2384 | return ENODEV; | |
2385 | } | |
c1c9c9c4 BP |
2386 | tcmsg->tcm_handle = handle; |
2387 | tcmsg->tcm_parent = parent; | |
2388 | ||
2389 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2390 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2391 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
2392 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
2393 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
2394 | nl_msg_end_nested(&request, opt_offset); | |
2395 | ||
2396 | error = tc_transact(&request, NULL); | |
2397 | if (error) { | |
2398 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
2399 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
2400 | netdev_get_name(netdev), | |
2401 | tc_get_major(handle), tc_get_minor(handle), | |
2402 | tc_get_major(parent), tc_get_minor(parent), | |
2403 | class->min_rate, class->max_rate, | |
2404 | class->burst, class->priority, strerror(error)); | |
2405 | } | |
2406 | return error; | |
2407 | } | |
2408 | ||
2409 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
2410 | * description of them into 'details'. The description complies with the | |
2411 | * specification given in the vswitch database documentation for linux-htb | |
2412 | * queue details. */ | |
2413 | static int | |
2414 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
2415 | { | |
2416 | static const struct nl_policy tca_htb_policy[] = { | |
2417 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
2418 | .min_len = sizeof(struct tc_htb_opt) }, | |
2419 | }; | |
2420 | ||
2421 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
2422 | const struct tc_htb_opt *htb; | |
2423 | ||
2424 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
2425 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
2426 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
2427 | return EPROTO; | |
2428 | } | |
2429 | ||
2430 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
2431 | class->min_rate = htb->rate.rate; | |
2432 | class->max_rate = htb->ceil.rate; | |
2433 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
2434 | class->priority = htb->prio; | |
2435 | return 0; | |
2436 | } | |
2437 | ||
2438 | static int | |
2439 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
2440 | struct htb_class *options, | |
2441 | struct netdev_queue_stats *stats) | |
2442 | { | |
2443 | struct nlattr *nl_options; | |
2444 | unsigned int handle; | |
2445 | int error; | |
2446 | ||
2447 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
2448 | if (!error && queue_id) { | |
17ee3c1f BP |
2449 | unsigned int major = tc_get_major(handle); |
2450 | unsigned int minor = tc_get_minor(handle); | |
2451 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
2452 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
2453 | } else { |
2454 | error = EPROTO; | |
2455 | } | |
2456 | } | |
2457 | if (!error && options) { | |
2458 | error = htb_parse_tca_options__(nl_options, options); | |
2459 | } | |
2460 | return error; | |
2461 | } | |
2462 | ||
2463 | static void | |
2464 | htb_parse_qdisc_details__(struct netdev *netdev, | |
2465 | const struct shash *details, struct htb_class *hc) | |
2466 | { | |
2467 | const char *max_rate_s; | |
2468 | ||
2469 | max_rate_s = shash_find_data(details, "max-rate"); | |
2470 | hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; | |
2471 | if (!hc->max_rate) { | |
2472 | uint32_t current; | |
2473 | ||
2474 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
2475 | hc->max_rate = netdev_features_to_bps(current) / 8; | |
2476 | } | |
2477 | hc->min_rate = hc->max_rate; | |
2478 | hc->burst = 0; | |
2479 | hc->priority = 0; | |
2480 | } | |
2481 | ||
2482 | static int | |
2483 | htb_parse_class_details__(struct netdev *netdev, | |
2484 | const struct shash *details, struct htb_class *hc) | |
2485 | { | |
2486 | const struct htb *htb = htb_get__(netdev); | |
2487 | const char *min_rate_s = shash_find_data(details, "min-rate"); | |
2488 | const char *max_rate_s = shash_find_data(details, "max-rate"); | |
2489 | const char *burst_s = shash_find_data(details, "burst"); | |
2490 | const char *priority_s = shash_find_data(details, "priority"); | |
2491 | int mtu; | |
2492 | ||
f915f1a8 BP |
2493 | netdev_get_mtu(netdev, &mtu); |
2494 | if (mtu == INT_MAX) { | |
2495 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", | |
2496 | netdev_get_name(netdev)); | |
2497 | return EINVAL; | |
2498 | } | |
2499 | ||
da3827b5 | 2500 | /* min-rate. Don't allow a min-rate below 1500 bytes/s. */ |
c1c9c9c4 BP |
2501 | if (!min_rate_s) { |
2502 | /* min-rate is required. */ | |
2503 | return EINVAL; | |
2504 | } | |
2505 | hc->min_rate = strtoull(min_rate_s, NULL, 10) / 8; | |
da3827b5 | 2506 | hc->min_rate = MAX(hc->min_rate, 1500); |
c1c9c9c4 BP |
2507 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
2508 | ||
2509 | /* max-rate */ | |
2510 | hc->max_rate = (max_rate_s | |
2511 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
2512 | : htb->max_rate); | |
2513 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); | |
2514 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
2515 | ||
2516 | /* burst | |
2517 | * | |
2518 | * According to hints in the documentation that I've read, it is important | |
2519 | * that 'burst' be at least as big as the largest frame that might be | |
2520 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
2521 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
2522 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
2523 | * the MTU. We actually add 64, instead of 14, as a guard against | |
2524 | * additional headers get tacked on somewhere that we're not aware of. */ | |
c1c9c9c4 BP |
2525 | hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; |
2526 | hc->burst = MAX(hc->burst, mtu + 64); | |
2527 | ||
2528 | /* priority */ | |
2529 | hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; | |
2530 | ||
2531 | return 0; | |
2532 | } | |
2533 | ||
2534 | static int | |
2535 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
2536 | unsigned int parent, struct htb_class *options, | |
2537 | struct netdev_queue_stats *stats) | |
2538 | { | |
2539 | struct ofpbuf *reply; | |
2540 | int error; | |
2541 | ||
2542 | error = tc_query_class(netdev, handle, parent, &reply); | |
2543 | if (!error) { | |
2544 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
2545 | ofpbuf_delete(reply); | |
2546 | } | |
2547 | return error; | |
2548 | } | |
2549 | ||
2550 | static int | |
2551 | htb_tc_install(struct netdev *netdev, const struct shash *details) | |
2552 | { | |
2553 | int error; | |
2554 | ||
2555 | error = htb_setup_qdisc__(netdev); | |
2556 | if (!error) { | |
2557 | struct htb_class hc; | |
2558 | ||
2559 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2560 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2561 | tc_make_handle(1, 0), &hc); | |
2562 | if (!error) { | |
2563 | htb_install__(netdev, hc.max_rate); | |
2564 | } | |
2565 | } | |
2566 | return error; | |
2567 | } | |
2568 | ||
93b13be8 BP |
2569 | static struct htb_class * |
2570 | htb_class_cast__(const struct tc_queue *queue) | |
2571 | { | |
2572 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
2573 | } | |
2574 | ||
c1c9c9c4 BP |
2575 | static void |
2576 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
2577 | const struct htb_class *hc) | |
2578 | { | |
2579 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
2580 | size_t hash = hash_int(queue_id, 0); |
2581 | struct tc_queue *queue; | |
c1c9c9c4 BP |
2582 | struct htb_class *hcp; |
2583 | ||
93b13be8 BP |
2584 | queue = tc_find_queue__(netdev, queue_id, hash); |
2585 | if (queue) { | |
2586 | hcp = htb_class_cast__(queue); | |
2587 | } else { | |
c1c9c9c4 | 2588 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
2589 | queue = &hcp->tc_queue; |
2590 | queue->queue_id = queue_id; | |
2591 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); | |
c1c9c9c4 | 2592 | } |
93b13be8 BP |
2593 | |
2594 | hcp->min_rate = hc->min_rate; | |
2595 | hcp->max_rate = hc->max_rate; | |
2596 | hcp->burst = hc->burst; | |
2597 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
2598 | } |
2599 | ||
2600 | static int | |
2601 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2602 | { | |
c1c9c9c4 BP |
2603 | struct ofpbuf msg; |
2604 | struct nl_dump dump; | |
2605 | struct htb_class hc; | |
2606 | struct htb *htb; | |
2607 | ||
2608 | /* Get qdisc options. */ | |
2609 | hc.max_rate = 0; | |
2610 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
2611 | htb = htb_install__(netdev, hc.max_rate); | |
2612 | ||
2613 | /* Get queues. */ | |
23a98ffe BP |
2614 | if (!start_queue_dump(netdev, &dump)) { |
2615 | return ENODEV; | |
2616 | } | |
c1c9c9c4 BP |
2617 | while (nl_dump_next(&dump, &msg)) { |
2618 | unsigned int queue_id; | |
2619 | ||
2620 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
2621 | htb_update_queue__(netdev, queue_id, &hc); | |
2622 | } | |
2623 | } | |
2624 | nl_dump_done(&dump); | |
2625 | ||
2626 | return 0; | |
2627 | } | |
2628 | ||
2629 | static void | |
2630 | htb_tc_destroy(struct tc *tc) | |
2631 | { | |
2632 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
93b13be8 | 2633 | struct htb_class *hc, *next; |
c1c9c9c4 | 2634 | |
4e8e4213 | 2635 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { |
93b13be8 | 2636 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 BP |
2637 | free(hc); |
2638 | } | |
2639 | tc_destroy(tc); | |
2640 | free(htb); | |
2641 | } | |
2642 | ||
2643 | static int | |
2644 | htb_qdisc_get(const struct netdev *netdev, struct shash *details) | |
2645 | { | |
2646 | const struct htb *htb = htb_get__(netdev); | |
2647 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * htb->max_rate)); | |
2648 | return 0; | |
2649 | } | |
2650 | ||
2651 | static int | |
2652 | htb_qdisc_set(struct netdev *netdev, const struct shash *details) | |
2653 | { | |
2654 | struct htb_class hc; | |
2655 | int error; | |
2656 | ||
2657 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2658 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2659 | tc_make_handle(1, 0), &hc); | |
2660 | if (!error) { | |
2661 | htb_get__(netdev)->max_rate = hc.max_rate; | |
2662 | } | |
2663 | return error; | |
2664 | } | |
2665 | ||
2666 | static int | |
93b13be8 BP |
2667 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
2668 | const struct tc_queue *queue, struct shash *details) | |
c1c9c9c4 | 2669 | { |
93b13be8 | 2670 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 BP |
2671 | |
2672 | shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); | |
2673 | if (hc->min_rate != hc->max_rate) { | |
2674 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); | |
2675 | } | |
2676 | shash_add(details, "burst", xasprintf("%llu", 8ULL * hc->burst)); | |
2677 | if (hc->priority) { | |
2678 | shash_add(details, "priority", xasprintf("%u", hc->priority)); | |
2679 | } | |
2680 | return 0; | |
2681 | } | |
2682 | ||
2683 | static int | |
2684 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
2685 | const struct shash *details) | |
2686 | { | |
2687 | struct htb_class hc; | |
2688 | int error; | |
2689 | ||
2690 | error = htb_parse_class_details__(netdev, details, &hc); | |
2691 | if (error) { | |
2692 | return error; | |
2693 | } | |
2694 | ||
17ee3c1f | 2695 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
2696 | tc_make_handle(1, 0xfffe), &hc); |
2697 | if (error) { | |
2698 | return error; | |
2699 | } | |
2700 | ||
2701 | htb_update_queue__(netdev, queue_id, &hc); | |
2702 | return 0; | |
2703 | } | |
2704 | ||
2705 | static int | |
93b13be8 | 2706 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 2707 | { |
93b13be8 | 2708 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2709 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
2710 | int error; |
2711 | ||
93b13be8 | 2712 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 2713 | if (!error) { |
93b13be8 | 2714 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 2715 | free(hc); |
c1c9c9c4 BP |
2716 | } |
2717 | return error; | |
2718 | } | |
2719 | ||
2720 | static int | |
93b13be8 | 2721 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
2722 | struct netdev_queue_stats *stats) |
2723 | { | |
93b13be8 | 2724 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
2725 | tc_make_handle(1, 0xfffe), NULL, stats); |
2726 | } | |
2727 | ||
2728 | static int | |
2729 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
2730 | const struct ofpbuf *nlmsg, | |
2731 | netdev_dump_queue_stats_cb *cb, void *aux) | |
2732 | { | |
2733 | struct netdev_queue_stats stats; | |
17ee3c1f | 2734 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
2735 | int error; |
2736 | ||
2737 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
2738 | if (error) { | |
2739 | return error; | |
2740 | } | |
2741 | ||
17ee3c1f BP |
2742 | major = tc_get_major(handle); |
2743 | minor = tc_get_minor(handle); | |
2744 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 2745 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
2746 | } |
2747 | return 0; | |
2748 | } | |
2749 | ||
2750 | static const struct tc_ops tc_ops_htb = { | |
2751 | "htb", /* linux_name */ | |
2752 | "linux-htb", /* ovs_name */ | |
2753 | HTB_N_QUEUES, /* n_queues */ | |
2754 | htb_tc_install, | |
2755 | htb_tc_load, | |
2756 | htb_tc_destroy, | |
2757 | htb_qdisc_get, | |
2758 | htb_qdisc_set, | |
2759 | htb_class_get, | |
2760 | htb_class_set, | |
2761 | htb_class_delete, | |
2762 | htb_class_get_stats, | |
2763 | htb_class_dump_stats | |
2764 | }; | |
2765 | \f | |
a339aa81 EJ |
2766 | /* "linux-hfsc" traffic control class. */ |
2767 | ||
2768 | #define HFSC_N_QUEUES 0xf000 | |
2769 | ||
2770 | struct hfsc { | |
2771 | struct tc tc; | |
2772 | uint32_t max_rate; | |
2773 | }; | |
2774 | ||
2775 | struct hfsc_class { | |
2776 | struct tc_queue tc_queue; | |
2777 | uint32_t min_rate; | |
2778 | uint32_t max_rate; | |
2779 | }; | |
2780 | ||
2781 | static struct hfsc * | |
2782 | hfsc_get__(const struct netdev *netdev) | |
2783 | { | |
2784 | struct netdev_dev_linux *netdev_dev; | |
2785 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2786 | return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc); | |
2787 | } | |
2788 | ||
2789 | static struct hfsc_class * | |
2790 | hfsc_class_cast__(const struct tc_queue *queue) | |
2791 | { | |
2792 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
2793 | } | |
2794 | ||
2795 | static struct hfsc * | |
2796 | hfsc_install__(struct netdev *netdev, uint32_t max_rate) | |
2797 | { | |
2798 | struct netdev_dev_linux * netdev_dev; | |
2799 | struct hfsc *hfsc; | |
2800 | ||
2801 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2802 | hfsc = xmalloc(sizeof *hfsc); | |
2803 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
2804 | hfsc->max_rate = max_rate; | |
2805 | netdev_dev->tc = &hfsc->tc; | |
2806 | ||
2807 | return hfsc; | |
2808 | } | |
2809 | ||
2810 | static void | |
2811 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
2812 | const struct hfsc_class *hc) | |
2813 | { | |
2814 | size_t hash; | |
2815 | struct hfsc *hfsc; | |
2816 | struct hfsc_class *hcp; | |
2817 | struct tc_queue *queue; | |
2818 | ||
2819 | hfsc = hfsc_get__(netdev); | |
2820 | hash = hash_int(queue_id, 0); | |
2821 | ||
2822 | queue = tc_find_queue__(netdev, queue_id, hash); | |
2823 | if (queue) { | |
2824 | hcp = hfsc_class_cast__(queue); | |
2825 | } else { | |
2826 | hcp = xmalloc(sizeof *hcp); | |
2827 | queue = &hcp->tc_queue; | |
2828 | queue->queue_id = queue_id; | |
2829 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); | |
2830 | } | |
2831 | ||
2832 | hcp->min_rate = hc->min_rate; | |
2833 | hcp->max_rate = hc->max_rate; | |
2834 | } | |
2835 | ||
2836 | static int | |
2837 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
2838 | { | |
2839 | const struct tc_service_curve *rsc, *fsc, *usc; | |
2840 | static const struct nl_policy tca_hfsc_policy[] = { | |
2841 | [TCA_HFSC_RSC] = { | |
2842 | .type = NL_A_UNSPEC, | |
2843 | .optional = false, | |
2844 | .min_len = sizeof(struct tc_service_curve), | |
2845 | }, | |
2846 | [TCA_HFSC_FSC] = { | |
2847 | .type = NL_A_UNSPEC, | |
2848 | .optional = false, | |
2849 | .min_len = sizeof(struct tc_service_curve), | |
2850 | }, | |
2851 | [TCA_HFSC_USC] = { | |
2852 | .type = NL_A_UNSPEC, | |
2853 | .optional = false, | |
2854 | .min_len = sizeof(struct tc_service_curve), | |
2855 | }, | |
2856 | }; | |
2857 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
2858 | ||
2859 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
2860 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
2861 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
2862 | return EPROTO; | |
2863 | } | |
2864 | ||
2865 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
2866 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
2867 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
2868 | ||
2869 | if (rsc->m1 != 0 || rsc->d != 0 || | |
2870 | fsc->m1 != 0 || fsc->d != 0 || | |
2871 | usc->m1 != 0 || usc->d != 0) { | |
2872 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
2873 | "Non-linear service curves are not supported."); | |
2874 | return EPROTO; | |
2875 | } | |
2876 | ||
2877 | if (rsc->m2 != fsc->m2) { | |
2878 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
2879 | "Real-time service curves are not supported "); | |
2880 | return EPROTO; | |
2881 | } | |
2882 | ||
2883 | if (rsc->m2 > usc->m2) { | |
2884 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
2885 | "Min-rate service curve is greater than " | |
2886 | "the max-rate service curve."); | |
2887 | return EPROTO; | |
2888 | } | |
2889 | ||
2890 | class->min_rate = fsc->m2; | |
2891 | class->max_rate = usc->m2; | |
2892 | return 0; | |
2893 | } | |
2894 | ||
2895 | static int | |
2896 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
2897 | struct hfsc_class *options, | |
2898 | struct netdev_queue_stats *stats) | |
2899 | { | |
2900 | int error; | |
2901 | unsigned int handle; | |
2902 | struct nlattr *nl_options; | |
2903 | ||
2904 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
2905 | if (error) { | |
2906 | return error; | |
2907 | } | |
2908 | ||
2909 | if (queue_id) { | |
2910 | unsigned int major, minor; | |
2911 | ||
2912 | major = tc_get_major(handle); | |
2913 | minor = tc_get_minor(handle); | |
2914 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
2915 | *queue_id = minor - 1; | |
2916 | } else { | |
2917 | return EPROTO; | |
2918 | } | |
2919 | } | |
2920 | ||
2921 | if (options) { | |
2922 | error = hfsc_parse_tca_options__(nl_options, options); | |
2923 | } | |
2924 | ||
2925 | return error; | |
2926 | } | |
2927 | ||
2928 | static int | |
2929 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
2930 | unsigned int parent, struct hfsc_class *options, | |
2931 | struct netdev_queue_stats *stats) | |
2932 | { | |
2933 | int error; | |
2934 | struct ofpbuf *reply; | |
2935 | ||
2936 | error = tc_query_class(netdev, handle, parent, &reply); | |
2937 | if (error) { | |
2938 | return error; | |
2939 | } | |
2940 | ||
2941 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
2942 | ofpbuf_delete(reply); | |
2943 | return error; | |
2944 | } | |
2945 | ||
2946 | static void | |
2947 | hfsc_parse_qdisc_details__(struct netdev *netdev, const struct shash *details, | |
2948 | struct hfsc_class *class) | |
2949 | { | |
2950 | uint32_t max_rate; | |
2951 | const char *max_rate_s; | |
2952 | ||
2953 | max_rate_s = shash_find_data(details, "max-rate"); | |
2954 | max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; | |
2955 | ||
2956 | if (!max_rate) { | |
2957 | uint32_t current; | |
2958 | ||
2959 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
2960 | max_rate = netdev_features_to_bps(current) / 8; | |
2961 | } | |
2962 | ||
2963 | class->min_rate = max_rate; | |
2964 | class->max_rate = max_rate; | |
2965 | } | |
2966 | ||
2967 | static int | |
2968 | hfsc_parse_class_details__(struct netdev *netdev, | |
2969 | const struct shash *details, | |
2970 | struct hfsc_class * class) | |
2971 | { | |
2972 | const struct hfsc *hfsc; | |
2973 | uint32_t min_rate, max_rate; | |
2974 | const char *min_rate_s, *max_rate_s; | |
2975 | ||
2976 | hfsc = hfsc_get__(netdev); | |
2977 | min_rate_s = shash_find_data(details, "min-rate"); | |
2978 | max_rate_s = shash_find_data(details, "max-rate"); | |
2979 | ||
2980 | if (!min_rate_s) { | |
2981 | return EINVAL; | |
2982 | } | |
2983 | ||
2984 | min_rate = strtoull(min_rate_s, NULL, 10) / 8; | |
2985 | min_rate = MAX(min_rate, 1500); | |
2986 | min_rate = MIN(min_rate, hfsc->max_rate); | |
2987 | ||
2988 | max_rate = (max_rate_s | |
2989 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
2990 | : hfsc->max_rate); | |
2991 | max_rate = MAX(max_rate, min_rate); | |
2992 | max_rate = MIN(max_rate, hfsc->max_rate); | |
2993 | ||
2994 | class->min_rate = min_rate; | |
2995 | class->max_rate = max_rate; | |
2996 | ||
2997 | return 0; | |
2998 | } | |
2999 | ||
3000 | /* Create an HFSC qdisc. | |
3001 | * | |
3002 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
3003 | static int | |
3004 | hfsc_setup_qdisc__(struct netdev * netdev) | |
3005 | { | |
3006 | struct tcmsg *tcmsg; | |
3007 | struct ofpbuf request; | |
3008 | struct tc_hfsc_qopt opt; | |
3009 | ||
3010 | tc_del_qdisc(netdev); | |
3011 | ||
3012 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3013 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3014 | ||
3015 | if (!tcmsg) { | |
3016 | return ENODEV; | |
3017 | } | |
3018 | ||
3019 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3020 | tcmsg->tcm_parent = TC_H_ROOT; | |
3021 | ||
3022 | memset(&opt, 0, sizeof opt); | |
3023 | opt.defcls = 1; | |
3024 | ||
3025 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3026 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3027 | ||
3028 | return tc_transact(&request, NULL); | |
3029 | } | |
3030 | ||
3031 | /* Create an HFSC class. | |
3032 | * | |
3033 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
3034 | * sc rate <min_rate> ul rate <max_rate>" */ | |
3035 | static int | |
3036 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
3037 | unsigned int parent, struct hfsc_class *class) | |
3038 | { | |
3039 | int error; | |
3040 | size_t opt_offset; | |
3041 | struct tcmsg *tcmsg; | |
3042 | struct ofpbuf request; | |
3043 | struct tc_service_curve min, max; | |
3044 | ||
3045 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
3046 | ||
3047 | if (!tcmsg) { | |
3048 | return ENODEV; | |
3049 | } | |
3050 | ||
3051 | tcmsg->tcm_handle = handle; | |
3052 | tcmsg->tcm_parent = parent; | |
3053 | ||
3054 | min.m1 = 0; | |
3055 | min.d = 0; | |
3056 | min.m2 = class->min_rate; | |
3057 | ||
3058 | max.m1 = 0; | |
3059 | max.d = 0; | |
3060 | max.m2 = class->max_rate; | |
3061 | ||
3062 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3063 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3064 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
3065 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
3066 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
3067 | nl_msg_end_nested(&request, opt_offset); | |
3068 | ||
3069 | error = tc_transact(&request, NULL); | |
3070 | if (error) { | |
3071 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3072 | "min-rate %ubps, max-rate %ubps (%s)", | |
3073 | netdev_get_name(netdev), | |
3074 | tc_get_major(handle), tc_get_minor(handle), | |
3075 | tc_get_major(parent), tc_get_minor(parent), | |
3076 | class->min_rate, class->max_rate, strerror(error)); | |
3077 | } | |
3078 | ||
3079 | return error; | |
3080 | } | |
3081 | ||
3082 | static int | |
3083 | hfsc_tc_install(struct netdev *netdev, const struct shash *details) | |
3084 | { | |
3085 | int error; | |
3086 | struct hfsc_class class; | |
3087 | ||
3088 | error = hfsc_setup_qdisc__(netdev); | |
3089 | ||
3090 | if (error) { | |
3091 | return error; | |
3092 | } | |
3093 | ||
3094 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3095 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3096 | tc_make_handle(1, 0), &class); | |
3097 | ||
3098 | if (error) { | |
3099 | return error; | |
3100 | } | |
3101 | ||
3102 | hfsc_install__(netdev, class.max_rate); | |
3103 | return 0; | |
3104 | } | |
3105 | ||
3106 | static int | |
3107 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3108 | { | |
3109 | struct ofpbuf msg; | |
3110 | struct hfsc *hfsc; | |
3111 | struct nl_dump dump; | |
3112 | struct hfsc_class hc; | |
3113 | ||
3114 | hc.max_rate = 0; | |
3115 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
3116 | hfsc = hfsc_install__(netdev, hc.max_rate); | |
3117 | ||
3118 | if (!start_queue_dump(netdev, &dump)) { | |
3119 | return ENODEV; | |
3120 | } | |
3121 | ||
3122 | while (nl_dump_next(&dump, &msg)) { | |
3123 | unsigned int queue_id; | |
3124 | ||
3125 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
3126 | hfsc_update_queue__(netdev, queue_id, &hc); | |
3127 | } | |
3128 | } | |
3129 | ||
3130 | nl_dump_done(&dump); | |
3131 | return 0; | |
3132 | } | |
3133 | ||
3134 | static void | |
3135 | hfsc_tc_destroy(struct tc *tc) | |
3136 | { | |
3137 | struct hfsc *hfsc; | |
3138 | struct hfsc_class *hc, *next; | |
3139 | ||
3140 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
3141 | ||
3142 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
3143 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3144 | free(hc); | |
3145 | } | |
3146 | ||
3147 | tc_destroy(tc); | |
3148 | free(hfsc); | |
3149 | } | |
3150 | ||
3151 | static int | |
3152 | hfsc_qdisc_get(const struct netdev *netdev, struct shash *details) | |
3153 | { | |
3154 | const struct hfsc *hfsc; | |
3155 | hfsc = hfsc_get__(netdev); | |
3156 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hfsc->max_rate)); | |
3157 | return 0; | |
3158 | } | |
3159 | ||
3160 | static int | |
3161 | hfsc_qdisc_set(struct netdev *netdev, const struct shash *details) | |
3162 | { | |
3163 | int error; | |
3164 | struct hfsc_class class; | |
3165 | ||
3166 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3167 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3168 | tc_make_handle(1, 0), &class); | |
3169 | ||
3170 | if (!error) { | |
3171 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
3172 | } | |
3173 | ||
3174 | return error; | |
3175 | } | |
3176 | ||
3177 | static int | |
3178 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
3179 | const struct tc_queue *queue, struct shash *details) | |
3180 | { | |
3181 | const struct hfsc_class *hc; | |
3182 | ||
3183 | hc = hfsc_class_cast__(queue); | |
3184 | shash_add(details, "min-rate", xasprintf("%llu", 8ULL * hc->min_rate)); | |
3185 | if (hc->min_rate != hc->max_rate) { | |
3186 | shash_add(details, "max-rate", xasprintf("%llu", 8ULL * hc->max_rate)); | |
3187 | } | |
3188 | return 0; | |
3189 | } | |
3190 | ||
3191 | static int | |
3192 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
3193 | const struct shash *details) | |
3194 | { | |
3195 | int error; | |
3196 | struct hfsc_class class; | |
3197 | ||
3198 | error = hfsc_parse_class_details__(netdev, details, &class); | |
3199 | if (error) { | |
3200 | return error; | |
3201 | } | |
3202 | ||
3203 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
3204 | tc_make_handle(1, 0xfffe), &class); | |
3205 | if (error) { | |
3206 | return error; | |
3207 | } | |
3208 | ||
3209 | hfsc_update_queue__(netdev, queue_id, &class); | |
3210 | return 0; | |
3211 | } | |
3212 | ||
3213 | static int | |
3214 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
3215 | { | |
3216 | int error; | |
3217 | struct hfsc *hfsc; | |
3218 | struct hfsc_class *hc; | |
3219 | ||
3220 | hc = hfsc_class_cast__(queue); | |
3221 | hfsc = hfsc_get__(netdev); | |
3222 | ||
3223 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
3224 | if (!error) { | |
3225 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3226 | free(hc); | |
3227 | } | |
3228 | return error; | |
3229 | } | |
3230 | ||
3231 | static int | |
3232 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
3233 | struct netdev_queue_stats *stats) | |
3234 | { | |
3235 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
3236 | tc_make_handle(1, 0xfffe), NULL, stats); | |
3237 | } | |
3238 | ||
3239 | static int | |
3240 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
3241 | const struct ofpbuf *nlmsg, | |
3242 | netdev_dump_queue_stats_cb *cb, void *aux) | |
3243 | { | |
3244 | struct netdev_queue_stats stats; | |
3245 | unsigned int handle, major, minor; | |
3246 | int error; | |
3247 | ||
3248 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
3249 | if (error) { | |
3250 | return error; | |
3251 | } | |
3252 | ||
3253 | major = tc_get_major(handle); | |
3254 | minor = tc_get_minor(handle); | |
3255 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
3256 | (*cb)(minor - 1, &stats, aux); | |
3257 | } | |
3258 | return 0; | |
3259 | } | |
3260 | ||
3261 | static const struct tc_ops tc_ops_hfsc = { | |
3262 | "hfsc", /* linux_name */ | |
3263 | "linux-hfsc", /* ovs_name */ | |
3264 | HFSC_N_QUEUES, /* n_queues */ | |
3265 | hfsc_tc_install, /* tc_install */ | |
3266 | hfsc_tc_load, /* tc_load */ | |
3267 | hfsc_tc_destroy, /* tc_destroy */ | |
3268 | hfsc_qdisc_get, /* qdisc_get */ | |
3269 | hfsc_qdisc_set, /* qdisc_set */ | |
3270 | hfsc_class_get, /* class_get */ | |
3271 | hfsc_class_set, /* class_set */ | |
3272 | hfsc_class_delete, /* class_delete */ | |
3273 | hfsc_class_get_stats, /* class_get_stats */ | |
3274 | hfsc_class_dump_stats /* class_dump_stats */ | |
3275 | }; | |
3276 | \f | |
c1c9c9c4 BP |
3277 | /* "linux-default" traffic control class. |
3278 | * | |
3279 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
3280 | * the "" (empty string) QoS type in the OVS database. */ | |
3281 | ||
3282 | static void | |
3283 | default_install__(struct netdev *netdev) | |
3284 | { | |
3285 | struct netdev_dev_linux *netdev_dev = | |
3286 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3287 | static struct tc *tc; | |
3288 | ||
3289 | if (!tc) { | |
3290 | tc = xmalloc(sizeof *tc); | |
3291 | tc_init(tc, &tc_ops_default); | |
3292 | } | |
3293 | netdev_dev->tc = tc; | |
3294 | } | |
3295 | ||
3296 | static int | |
3297 | default_tc_install(struct netdev *netdev, | |
3298 | const struct shash *details OVS_UNUSED) | |
3299 | { | |
3300 | default_install__(netdev); | |
3301 | return 0; | |
3302 | } | |
3303 | ||
3304 | static int | |
3305 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3306 | { | |
3307 | default_install__(netdev); | |
3308 | return 0; | |
3309 | } | |
3310 | ||
3311 | static const struct tc_ops tc_ops_default = { | |
3312 | NULL, /* linux_name */ | |
3313 | "", /* ovs_name */ | |
3314 | 0, /* n_queues */ | |
3315 | default_tc_install, | |
3316 | default_tc_load, | |
3317 | NULL, /* tc_destroy */ | |
3318 | NULL, /* qdisc_get */ | |
3319 | NULL, /* qdisc_set */ | |
3320 | NULL, /* class_get */ | |
3321 | NULL, /* class_set */ | |
3322 | NULL, /* class_delete */ | |
3323 | NULL, /* class_get_stats */ | |
3324 | NULL /* class_dump_stats */ | |
3325 | }; | |
3326 | \f | |
3327 | /* "linux-other" traffic control class. | |
3328 | * | |
3329 | * */ | |
3330 | ||
3331 | static int | |
3332 | other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3333 | { | |
3334 | struct netdev_dev_linux *netdev_dev = | |
3335 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3336 | static struct tc *tc; | |
3337 | ||
3338 | if (!tc) { | |
3339 | tc = xmalloc(sizeof *tc); | |
3340 | tc_init(tc, &tc_ops_other); | |
3341 | } | |
3342 | netdev_dev->tc = tc; | |
3343 | return 0; | |
3344 | } | |
3345 | ||
3346 | static const struct tc_ops tc_ops_other = { | |
3347 | NULL, /* linux_name */ | |
3348 | "linux-other", /* ovs_name */ | |
3349 | 0, /* n_queues */ | |
3350 | NULL, /* tc_install */ | |
3351 | other_tc_load, | |
3352 | NULL, /* tc_destroy */ | |
3353 | NULL, /* qdisc_get */ | |
3354 | NULL, /* qdisc_set */ | |
3355 | NULL, /* class_get */ | |
3356 | NULL, /* class_set */ | |
3357 | NULL, /* class_delete */ | |
3358 | NULL, /* class_get_stats */ | |
3359 | NULL /* class_dump_stats */ | |
3360 | }; | |
3361 | \f | |
3362 | /* Traffic control. */ | |
3363 | ||
3364 | /* Number of kernel "tc" ticks per second. */ | |
3365 | static double ticks_per_s; | |
3366 | ||
3367 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
3368 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
3369 | * one jiffy's worth of data. | |
3370 | * | |
3371 | * There are two possibilities here: | |
3372 | * | |
3373 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
3374 | * approximate range of 100 to 1024. That means that we really need to | |
3375 | * make sure that the qdisc can buffer that much data. | |
3376 | * | |
3377 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
3378 | * has finely granular timers and there's no need to fudge additional room | |
3379 | * for buffers. (There's no extra effort needed to implement that: the | |
3380 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
3381 | * come out as 0 in the division. Small integer results in the case of | |
3382 | * really high dividends won't have any real effect anyhow.) | |
3383 | */ | |
3384 | static unsigned int buffer_hz; | |
3385 | ||
3386 | /* Returns tc handle 'major':'minor'. */ | |
3387 | static unsigned int | |
3388 | tc_make_handle(unsigned int major, unsigned int minor) | |
3389 | { | |
3390 | return TC_H_MAKE(major << 16, minor); | |
3391 | } | |
3392 | ||
3393 | /* Returns the major number from 'handle'. */ | |
3394 | static unsigned int | |
3395 | tc_get_major(unsigned int handle) | |
3396 | { | |
3397 | return TC_H_MAJ(handle) >> 16; | |
3398 | } | |
3399 | ||
3400 | /* Returns the minor number from 'handle'. */ | |
3401 | static unsigned int | |
3402 | tc_get_minor(unsigned int handle) | |
3403 | { | |
3404 | return TC_H_MIN(handle); | |
3405 | } | |
3406 | ||
3407 | static struct tcmsg * | |
3408 | tc_make_request(const struct netdev *netdev, int type, unsigned int flags, | |
3409 | struct ofpbuf *request) | |
3410 | { | |
3411 | struct tcmsg *tcmsg; | |
3412 | int ifindex; | |
3413 | int error; | |
3414 | ||
3415 | error = get_ifindex(netdev, &ifindex); | |
3416 | if (error) { | |
3417 | return NULL; | |
3418 | } | |
3419 | ||
3420 | ofpbuf_init(request, 512); | |
3421 | nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); | |
3422 | tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); | |
3423 | tcmsg->tcm_family = AF_UNSPEC; | |
3424 | tcmsg->tcm_ifindex = ifindex; | |
3425 | /* Caller should fill in tcmsg->tcm_handle. */ | |
3426 | /* Caller should fill in tcmsg->tcm_parent. */ | |
3427 | ||
3428 | return tcmsg; | |
3429 | } | |
3430 | ||
3431 | static int | |
3432 | tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) | |
3433 | { | |
3434 | int error = nl_sock_transact(rtnl_sock, request, replyp); | |
3435 | ofpbuf_uninit(request); | |
3436 | return error; | |
3437 | } | |
3438 | ||
3439 | static void | |
3440 | read_psched(void) | |
3441 | { | |
3442 | /* The values in psched are not individually very meaningful, but they are | |
3443 | * important. The tables below show some values seen in the wild. | |
3444 | * | |
3445 | * Some notes: | |
3446 | * | |
3447 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
3448 | * (Before that, there are hints that it was 1000000000.) | |
3449 | * | |
3450 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
3451 | * above. | |
3452 | * | |
3453 | * /proc/net/psched | |
3454 | * ----------------------------------- | |
3455 | * [1] 000c8000 000f4240 000f4240 00000064 | |
3456 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
3457 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
3458 | * [4] 000003e8 00000400 000f4240 00000064 | |
3459 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
3460 | * [6] 000003e8 00000040 000f4240 000000f9 | |
3461 | * | |
3462 | * a b c d ticks_per_s buffer_hz | |
3463 | * ------- --------- ---------- ------------- ----------- ------------- | |
3464 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
3465 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3466 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3467 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
3468 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
3469 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
3470 | * | |
3471 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
3472 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
3473 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
3474 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
3475 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
3476 | * [6] 2.6.34 from kernel.org on KVM | |
3477 | */ | |
3478 | static const char fn[] = "/proc/net/psched"; | |
3479 | unsigned int a, b, c, d; | |
3480 | FILE *stream; | |
3481 | ||
3482 | ticks_per_s = 1.0; | |
3483 | buffer_hz = 100; | |
3484 | ||
3485 | stream = fopen(fn, "r"); | |
3486 | if (!stream) { | |
3487 | VLOG_WARN("%s: open failed: %s", fn, strerror(errno)); | |
3488 | return; | |
3489 | } | |
3490 | ||
3491 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
3492 | VLOG_WARN("%s: read failed", fn); | |
3493 | fclose(stream); | |
3494 | return; | |
3495 | } | |
3496 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
3497 | fclose(stream); | |
3498 | ||
3499 | if (!a || !c) { | |
3500 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
3501 | return; | |
3502 | } | |
3503 | ||
3504 | ticks_per_s = (double) a * c / b; | |
3505 | if (c == 1000000) { | |
3506 | buffer_hz = d; | |
3507 | } else { | |
3508 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
3509 | fn, a, b, c, d); | |
3510 | } | |
3511 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
3512 | } | |
3513 | ||
3514 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
3515 | * rate of 'rate' bytes per second. */ | |
3516 | static unsigned int | |
3517 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
3518 | { | |
3519 | if (!buffer_hz) { | |
3520 | read_psched(); | |
3521 | } | |
3522 | return (rate * ticks) / ticks_per_s; | |
3523 | } | |
3524 | ||
3525 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
3526 | * rate of 'rate' bytes per second. */ | |
3527 | static unsigned int | |
3528 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
3529 | { | |
3530 | if (!buffer_hz) { | |
3531 | read_psched(); | |
3532 | } | |
015c93a4 | 3533 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
3534 | } |
3535 | ||
3536 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
3537 | * a transmission rate of 'rate' bytes per second. */ | |
3538 | static unsigned int | |
3539 | tc_buffer_per_jiffy(unsigned int rate) | |
3540 | { | |
3541 | if (!buffer_hz) { | |
3542 | read_psched(); | |
3543 | } | |
3544 | return rate / buffer_hz; | |
3545 | } | |
3546 | ||
3547 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
3548 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
3549 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
3550 | * stores NULL into it if it is absent. | |
3551 | * | |
3552 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
3553 | * 'msg'. | |
3554 | * | |
3555 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3556 | static int | |
3557 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
3558 | struct nlattr **options) | |
3559 | { | |
3560 | static const struct nl_policy tca_policy[] = { | |
3561 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
3562 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
3563 | }; | |
3564 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3565 | ||
3566 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3567 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3568 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
3569 | goto error; | |
3570 | } | |
3571 | ||
3572 | if (kind) { | |
3573 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
3574 | } | |
3575 | ||
3576 | if (options) { | |
3577 | *options = ta[TCA_OPTIONS]; | |
3578 | } | |
3579 | ||
3580 | return 0; | |
3581 | ||
3582 | error: | |
3583 | if (kind) { | |
3584 | *kind = NULL; | |
3585 | } | |
3586 | if (options) { | |
3587 | *options = NULL; | |
3588 | } | |
3589 | return EPROTO; | |
3590 | } | |
3591 | ||
3592 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
3593 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
3594 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
3595 | * arguments may be null. | |
3596 | * | |
3597 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3598 | static int | |
3599 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
3600 | struct nlattr **options, struct netdev_queue_stats *stats) | |
3601 | { | |
3602 | static const struct nl_policy tca_policy[] = { | |
3603 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
3604 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
3605 | }; | |
3606 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3607 | ||
3608 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3609 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3610 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
3611 | goto error; | |
3612 | } | |
3613 | ||
3614 | if (handlep) { | |
3615 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
3616 | *handlep = tc->tcm_handle; | |
3617 | } | |
3618 | ||
3619 | if (options) { | |
3620 | *options = ta[TCA_OPTIONS]; | |
3621 | } | |
3622 | ||
3623 | if (stats) { | |
3624 | const struct gnet_stats_queue *gsq; | |
3625 | struct gnet_stats_basic gsb; | |
3626 | ||
3627 | static const struct nl_policy stats_policy[] = { | |
3628 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
3629 | .min_len = sizeof gsb }, | |
3630 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
3631 | .min_len = sizeof *gsq }, | |
3632 | }; | |
3633 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
3634 | ||
3635 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
3636 | sa, ARRAY_SIZE(sa))) { | |
3637 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
3638 | goto error; | |
3639 | } | |
3640 | ||
3641 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
3642 | * some arch/bitsize combinations. Newer versions of Linux have a | |
3643 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
3644 | * easiest thing to do is just to make a copy. */ | |
3645 | memset(&gsb, 0, sizeof gsb); | |
3646 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
3647 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
3648 | stats->tx_bytes = gsb.bytes; | |
3649 | stats->tx_packets = gsb.packets; | |
3650 | ||
3651 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
3652 | stats->tx_errors = gsq->drops; | |
3653 | } | |
3654 | ||
3655 | return 0; | |
3656 | ||
3657 | error: | |
3658 | if (options) { | |
3659 | *options = NULL; | |
3660 | } | |
3661 | if (stats) { | |
3662 | memset(stats, 0, sizeof *stats); | |
3663 | } | |
3664 | return EPROTO; | |
3665 | } | |
3666 | ||
3667 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
3668 | * on 'netdev'. */ | |
3669 | static int | |
3670 | tc_query_class(const struct netdev *netdev, | |
3671 | unsigned int handle, unsigned int parent, | |
3672 | struct ofpbuf **replyp) | |
3673 | { | |
3674 | struct ofpbuf request; | |
3675 | struct tcmsg *tcmsg; | |
3676 | int error; | |
3677 | ||
3678 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); | |
23a98ffe BP |
3679 | if (!tcmsg) { |
3680 | return ENODEV; | |
3681 | } | |
c1c9c9c4 BP |
3682 | tcmsg->tcm_handle = handle; |
3683 | tcmsg->tcm_parent = parent; | |
3684 | ||
3685 | error = tc_transact(&request, replyp); | |
3686 | if (error) { | |
3687 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
3688 | netdev_get_name(netdev), | |
3689 | tc_get_major(handle), tc_get_minor(handle), | |
3690 | tc_get_major(parent), tc_get_minor(parent), | |
3691 | strerror(error)); | |
3692 | } | |
3693 | return error; | |
3694 | } | |
3695 | ||
3696 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
3697 | static int | |
3698 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
3699 | { | |
3700 | struct ofpbuf request; | |
3701 | struct tcmsg *tcmsg; | |
3702 | int error; | |
3703 | ||
3704 | tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); | |
23a98ffe BP |
3705 | if (!tcmsg) { |
3706 | return ENODEV; | |
3707 | } | |
c1c9c9c4 BP |
3708 | tcmsg->tcm_handle = handle; |
3709 | tcmsg->tcm_parent = 0; | |
3710 | ||
3711 | error = tc_transact(&request, NULL); | |
3712 | if (error) { | |
3713 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
3714 | netdev_get_name(netdev), | |
3715 | tc_get_major(handle), tc_get_minor(handle), | |
3716 | strerror(error)); | |
3717 | } | |
3718 | return error; | |
3719 | } | |
3720 | ||
3721 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
3722 | static int | |
3723 | tc_del_qdisc(struct netdev *netdev) | |
3724 | { | |
3725 | struct netdev_dev_linux *netdev_dev = | |
3726 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3727 | struct ofpbuf request; | |
3728 | struct tcmsg *tcmsg; | |
3729 | int error; | |
3730 | ||
3731 | tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); | |
23a98ffe BP |
3732 | if (!tcmsg) { |
3733 | return ENODEV; | |
3734 | } | |
c1c9c9c4 BP |
3735 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
3736 | tcmsg->tcm_parent = TC_H_ROOT; | |
3737 | ||
3738 | error = tc_transact(&request, NULL); | |
3739 | if (error == EINVAL) { | |
3740 | /* EINVAL probably means that the default qdisc was in use, in which | |
3741 | * case we've accomplished our purpose. */ | |
3742 | error = 0; | |
3743 | } | |
3744 | if (!error && netdev_dev->tc) { | |
3745 | if (netdev_dev->tc->ops->tc_destroy) { | |
3746 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
3747 | } | |
3748 | netdev_dev->tc = NULL; | |
3749 | } | |
3750 | return error; | |
3751 | } | |
3752 | ||
3753 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the | |
3754 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
3755 | * positive errno value. */ | |
3756 | static int | |
3757 | tc_query_qdisc(const struct netdev *netdev) | |
3758 | { | |
3759 | struct netdev_dev_linux *netdev_dev = | |
3760 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3761 | struct ofpbuf request, *qdisc; | |
3762 | const struct tc_ops *ops; | |
3763 | struct tcmsg *tcmsg; | |
3764 | int load_error; | |
3765 | int error; | |
3766 | ||
3767 | if (netdev_dev->tc) { | |
3768 | return 0; | |
3769 | } | |
3770 | ||
3771 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
3772 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
3773 | * 2.6.35 without that fix backported to it. | |
3774 | * | |
3775 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
3776 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
3777 | * few others. There are a few ways that I can see to do this, but most of | |
3778 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
3779 | * technique chosen here is to assume that any non-default qdisc that we | |
3780 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
3781 | * a class with handle 0:0. | |
3782 | * | |
3783 | * We could check for Linux 2.6.35+ and use a more straightforward method | |
3784 | * there. */ | |
3785 | tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request); | |
23a98ffe BP |
3786 | if (!tcmsg) { |
3787 | return ENODEV; | |
3788 | } | |
c1c9c9c4 BP |
3789 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
3790 | tcmsg->tcm_parent = 0; | |
3791 | ||
3792 | /* Figure out what tc class to instantiate. */ | |
3793 | error = tc_transact(&request, &qdisc); | |
3794 | if (!error) { | |
3795 | const char *kind; | |
3796 | ||
3797 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
3798 | if (error) { | |
3799 | ops = &tc_ops_other; | |
3800 | } else { | |
3801 | ops = tc_lookup_linux_name(kind); | |
3802 | if (!ops) { | |
3803 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
3804 | VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind); | |
3805 | ||
3806 | ops = &tc_ops_other; | |
3807 | } | |
3808 | } | |
3809 | } else if (error == ENOENT) { | |
3810 | /* Either it's a built-in qdisc, or it's a qdisc set up by some | |
3811 | * other entity that doesn't have a handle 1:0. We will assume | |
3812 | * that it's the system default qdisc. */ | |
3813 | ops = &tc_ops_default; | |
3814 | error = 0; | |
3815 | } else { | |
3816 | /* Who knows? Maybe the device got deleted. */ | |
3817 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
3818 | netdev_get_name(netdev), strerror(error)); | |
3819 | ops = &tc_ops_other; | |
3820 | } | |
3821 | ||
3822 | /* Instantiate it. */ | |
3823 | load_error = ops->tc_load((struct netdev *) netdev, qdisc); | |
3824 | assert((load_error == 0) == (netdev_dev->tc != NULL)); | |
3825 | ofpbuf_delete(qdisc); | |
3826 | ||
3827 | return error ? error : load_error; | |
3828 | } | |
3829 | ||
3830 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
3831 | approximate the time to transmit packets of various lengths. For an MTU of | |
3832 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
3833 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
3834 | possible lengths; and so on. | |
3835 | ||
3836 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
3837 | need to be shifted right to fit within such a 256-entry table. */ | |
3838 | static int | |
3839 | tc_calc_cell_log(unsigned int mtu) | |
3840 | { | |
3841 | int cell_log; | |
3842 | ||
3843 | if (!mtu) { | |
3844 | mtu = ETH_PAYLOAD_MAX; | |
3845 | } | |
3846 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
3847 | ||
3848 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
3849 | mtu >>= 1; | |
3850 | } | |
3851 | ||
3852 | return cell_log; | |
3853 | } | |
3854 | ||
3855 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
3856 | * of 'mtu'. */ | |
3857 | static void | |
3858 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
3859 | { | |
3860 | memset(rate, 0, sizeof *rate); | |
3861 | rate->cell_log = tc_calc_cell_log(mtu); | |
3862 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
3863 | /* rate->cell_align = 0; */ /* distro headers. */ | |
3864 | rate->mpu = ETH_TOTAL_MIN; | |
3865 | rate->rate = Bps; | |
3866 | } | |
3867 | ||
3868 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
3869 | * attribute of the specified "type". | |
3870 | * | |
3871 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
3872 | static void | |
3873 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
3874 | { | |
3875 | uint32_t *rtab; | |
3876 | unsigned int i; | |
3877 | ||
3878 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
3879 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
3880 | unsigned packet_size = (i + 1) << rate->cell_log; | |
3881 | if (packet_size < rate->mpu) { | |
3882 | packet_size = rate->mpu; | |
3883 | } | |
3884 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
3885 | } | |
3886 | } | |
3887 | ||
3888 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
3889 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
3890 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 3891 | * 0 is fine.) */ |
c1c9c9c4 BP |
3892 | static int |
3893 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
3894 | { | |
3895 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
3896 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
3897 | } | |
3898 | ||
3899 | \f | |
3900 | /* Utility functions. */ | |
3901 | ||
3902 | static int | |
3903 | get_stats_via_netlink(int ifindex, struct netdev_stats *stats) | |
3904 | { | |
3905 | /* Policy for RTNLGRP_LINK messages. | |
3906 | * | |
3907 | * There are *many* more fields in these messages, but currently we only | |
3908 | * care about these fields. */ | |
3909 | static const struct nl_policy rtnlgrp_link_policy[] = { | |
3910 | [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, | |
3911 | [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, | |
3912 | .min_len = sizeof(struct rtnl_link_stats) }, | |
3913 | }; | |
3914 | ||
3915 | struct ofpbuf request; | |
3916 | struct ofpbuf *reply; | |
3917 | struct ifinfomsg *ifi; | |
3918 | const struct rtnl_link_stats *rtnl_stats; | |
3919 | struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; | |
3920 | int error; | |
3921 | ||
3922 | ofpbuf_init(&request, 0); | |
3923 | nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST); | |
3924 | ifi = ofpbuf_put_zeros(&request, sizeof *ifi); | |
3925 | ifi->ifi_family = PF_UNSPEC; | |
3926 | ifi->ifi_index = ifindex; | |
3927 | error = nl_sock_transact(rtnl_sock, &request, &reply); | |
3928 | ofpbuf_uninit(&request); | |
3929 | if (error) { | |
3930 | return error; | |
3931 | } | |
3932 | ||
3933 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), | |
3934 | rtnlgrp_link_policy, | |
3935 | attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { | |
3936 | ofpbuf_delete(reply); | |
3937 | return EPROTO; | |
3938 | } | |
3939 | ||
3940 | if (!attrs[IFLA_STATS]) { | |
3941 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
3942 | ofpbuf_delete(reply); | |
3943 | return EPROTO; | |
3944 | } | |
8b61709d BP |
3945 | |
3946 | rtnl_stats = nl_attr_get(attrs[IFLA_STATS]); | |
3947 | stats->rx_packets = rtnl_stats->rx_packets; | |
3948 | stats->tx_packets = rtnl_stats->tx_packets; | |
3949 | stats->rx_bytes = rtnl_stats->rx_bytes; | |
3950 | stats->tx_bytes = rtnl_stats->tx_bytes; | |
3951 | stats->rx_errors = rtnl_stats->rx_errors; | |
3952 | stats->tx_errors = rtnl_stats->tx_errors; | |
3953 | stats->rx_dropped = rtnl_stats->rx_dropped; | |
3954 | stats->tx_dropped = rtnl_stats->tx_dropped; | |
3955 | stats->multicast = rtnl_stats->multicast; | |
3956 | stats->collisions = rtnl_stats->collisions; | |
3957 | stats->rx_length_errors = rtnl_stats->rx_length_errors; | |
3958 | stats->rx_over_errors = rtnl_stats->rx_over_errors; | |
3959 | stats->rx_crc_errors = rtnl_stats->rx_crc_errors; | |
3960 | stats->rx_frame_errors = rtnl_stats->rx_frame_errors; | |
3961 | stats->rx_fifo_errors = rtnl_stats->rx_fifo_errors; | |
3962 | stats->rx_missed_errors = rtnl_stats->rx_missed_errors; | |
3963 | stats->tx_aborted_errors = rtnl_stats->tx_aborted_errors; | |
3964 | stats->tx_carrier_errors = rtnl_stats->tx_carrier_errors; | |
3965 | stats->tx_fifo_errors = rtnl_stats->tx_fifo_errors; | |
3966 | stats->tx_heartbeat_errors = rtnl_stats->tx_heartbeat_errors; | |
3967 | stats->tx_window_errors = rtnl_stats->tx_window_errors; | |
3968 | ||
576e26d7 BP |
3969 | ofpbuf_delete(reply); |
3970 | ||
8b61709d BP |
3971 | return 0; |
3972 | } | |
3973 | ||
3974 | static int | |
3975 | get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) | |
3976 | { | |
3977 | static const char fn[] = "/proc/net/dev"; | |
3978 | char line[1024]; | |
3979 | FILE *stream; | |
3980 | int ln; | |
3981 | ||
3982 | stream = fopen(fn, "r"); | |
3983 | if (!stream) { | |
3984 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
3985 | return errno; | |
3986 | } | |
3987 | ||
3988 | ln = 0; | |
3989 | while (fgets(line, sizeof line, stream)) { | |
3990 | if (++ln >= 3) { | |
3991 | char devname[16]; | |
3992 | #define X64 "%"SCNu64 | |
3993 | if (sscanf(line, | |
3994 | " %15[^:]:" | |
3995 | X64 X64 X64 X64 X64 X64 X64 "%*u" | |
3996 | X64 X64 X64 X64 X64 X64 X64 "%*u", | |
3997 | devname, | |
3998 | &stats->rx_bytes, | |
3999 | &stats->rx_packets, | |
4000 | &stats->rx_errors, | |
4001 | &stats->rx_dropped, | |
4002 | &stats->rx_fifo_errors, | |
4003 | &stats->rx_frame_errors, | |
4004 | &stats->multicast, | |
4005 | &stats->tx_bytes, | |
4006 | &stats->tx_packets, | |
4007 | &stats->tx_errors, | |
4008 | &stats->tx_dropped, | |
4009 | &stats->tx_fifo_errors, | |
4010 | &stats->collisions, | |
4011 | &stats->tx_carrier_errors) != 15) { | |
4012 | VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); | |
4013 | } else if (!strcmp(devname, netdev_name)) { | |
4014 | stats->rx_length_errors = UINT64_MAX; | |
4015 | stats->rx_over_errors = UINT64_MAX; | |
4016 | stats->rx_crc_errors = UINT64_MAX; | |
4017 | stats->rx_missed_errors = UINT64_MAX; | |
4018 | stats->tx_aborted_errors = UINT64_MAX; | |
4019 | stats->tx_heartbeat_errors = UINT64_MAX; | |
4020 | stats->tx_window_errors = UINT64_MAX; | |
4021 | fclose(stream); | |
4022 | return 0; | |
4023 | } | |
4024 | } | |
4025 | } | |
4026 | VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); | |
4027 | fclose(stream); | |
4028 | return ENODEV; | |
4029 | } | |
c1c9c9c4 | 4030 | |
8b61709d BP |
4031 | static int |
4032 | get_flags(const struct netdev *netdev, int *flags) | |
4033 | { | |
4034 | struct ifreq ifr; | |
4035 | int error; | |
4036 | ||
149f577a JG |
4037 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCGIFFLAGS, |
4038 | "SIOCGIFFLAGS"); | |
8b61709d BP |
4039 | *flags = ifr.ifr_flags; |
4040 | return error; | |
4041 | } | |
4042 | ||
4043 | static int | |
4044 | set_flags(struct netdev *netdev, int flags) | |
4045 | { | |
4046 | struct ifreq ifr; | |
4047 | ||
4048 | ifr.ifr_flags = flags; | |
149f577a JG |
4049 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, |
4050 | "SIOCSIFFLAGS"); | |
8b61709d BP |
4051 | } |
4052 | ||
4053 | static int | |
4054 | do_get_ifindex(const char *netdev_name) | |
4055 | { | |
4056 | struct ifreq ifr; | |
4057 | ||
71d7c22f | 4058 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4059 | COVERAGE_INC(netdev_get_ifindex); |
4060 | if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { | |
4061 | VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
4062 | netdev_name, strerror(errno)); | |
4063 | return -errno; | |
4064 | } | |
4065 | return ifr.ifr_ifindex; | |
4066 | } | |
4067 | ||
4068 | static int | |
4069 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
4070 | { | |
149f577a JG |
4071 | struct netdev_dev_linux *netdev_dev = |
4072 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d | 4073 | *ifindexp = 0; |
149f577a | 4074 | if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { |
8b61709d BP |
4075 | int ifindex = do_get_ifindex(netdev_get_name(netdev_)); |
4076 | if (ifindex < 0) { | |
4077 | return -ifindex; | |
4078 | } | |
149f577a JG |
4079 | netdev_dev->cache_valid |= VALID_IFINDEX; |
4080 | netdev_dev->ifindex = ifindex; | |
8b61709d | 4081 | } |
149f577a | 4082 | *ifindexp = netdev_dev->ifindex; |
8b61709d BP |
4083 | return 0; |
4084 | } | |
4085 | ||
4086 | static int | |
4087 | get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) | |
4088 | { | |
4089 | struct ifreq ifr; | |
4090 | int hwaddr_family; | |
4091 | ||
4092 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4093 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4094 | COVERAGE_INC(netdev_get_hwaddr); |
4095 | if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { | |
4096 | VLOG_ERR("ioctl(SIOCGIFHWADDR) on %s device failed: %s", | |
4097 | netdev_name, strerror(errno)); | |
4098 | return errno; | |
4099 | } | |
4100 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
4101 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { | |
4102 | VLOG_WARN("%s device has unknown hardware address family %d", | |
4103 | netdev_name, hwaddr_family); | |
4104 | } | |
4105 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
4106 | return 0; | |
4107 | } | |
4108 | ||
4109 | static int | |
4110 | set_etheraddr(const char *netdev_name, int hwaddr_family, | |
4111 | const uint8_t mac[ETH_ADDR_LEN]) | |
4112 | { | |
4113 | struct ifreq ifr; | |
4114 | ||
4115 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4116 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4117 | ifr.ifr_hwaddr.sa_family = hwaddr_family; |
4118 | memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); | |
4119 | COVERAGE_INC(netdev_set_hwaddr); | |
4120 | if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { | |
4121 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", | |
4122 | netdev_name, strerror(errno)); | |
4123 | return errno; | |
4124 | } | |
4125 | return 0; | |
4126 | } | |
4127 | ||
4128 | static int | |
0b0544d7 | 4129 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
4130 | int cmd, const char *cmd_name) |
4131 | { | |
4132 | struct ifreq ifr; | |
4133 | ||
4134 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4135 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
4136 | ifr.ifr_data = (caddr_t) ecmd; |
4137 | ||
4138 | ecmd->cmd = cmd; | |
4139 | COVERAGE_INC(netdev_ethtool); | |
4140 | if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { | |
4141 | return 0; | |
4142 | } else { | |
4143 | if (errno != EOPNOTSUPP) { | |
4144 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " | |
0b0544d7 | 4145 | "failed: %s", cmd_name, name, strerror(errno)); |
8b61709d BP |
4146 | } else { |
4147 | /* The device doesn't support this operation. That's pretty | |
4148 | * common, so there's no point in logging anything. */ | |
4149 | } | |
4150 | return errno; | |
4151 | } | |
4152 | } | |
4153 | ||
4154 | static int | |
149f577a JG |
4155 | netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, |
4156 | const char *cmd_name) | |
8b61709d | 4157 | { |
71d7c22f | 4158 | ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name); |
8b61709d | 4159 | if (ioctl(af_inet_sock, cmd, ifr) == -1) { |
149f577a JG |
4160 | VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, |
4161 | strerror(errno)); | |
8b61709d BP |
4162 | return errno; |
4163 | } | |
4164 | return 0; | |
4165 | } | |
f1acd62b BP |
4166 | |
4167 | static int | |
4168 | netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, | |
4169 | int cmd, const char *cmd_name) | |
4170 | { | |
4171 | struct ifreq ifr; | |
4172 | int error; | |
4173 | ||
4174 | ifr.ifr_addr.sa_family = AF_INET; | |
149f577a | 4175 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); |
f1acd62b BP |
4176 | if (!error) { |
4177 | const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; | |
4178 | *ip = sin->sin_addr; | |
4179 | } | |
4180 | return error; | |
4181 | } |