]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
e0edde6f | 2 | * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. |
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> | |
d3980822 BP |
18 | |
19 | #include "netdev-linux.h" | |
20 | ||
8b61709d | 21 | #include <assert.h> |
e9e28be3 | 22 | #include <errno.h> |
8b61709d BP |
23 | #include <fcntl.h> |
24 | #include <arpa/inet.h> | |
25 | #include <inttypes.h> | |
c1c9c9c4 | 26 | #include <linux/gen_stats.h> |
bb7d0e22 | 27 | #include <linux/if_ether.h> |
8b61709d BP |
28 | #include <linux/if_tun.h> |
29 | #include <linux/types.h> | |
30 | #include <linux/ethtool.h> | |
63331829 | 31 | #include <linux/mii.h> |
f8500004 | 32 | #include <linux/pkt_cls.h> |
6f42c8ea | 33 | #include <linux/pkt_sched.h> |
e9e28be3 | 34 | #include <linux/rtnetlink.h> |
8b61709d BP |
35 | #include <linux/sockios.h> |
36 | #include <linux/version.h> | |
37 | #include <sys/types.h> | |
38 | #include <sys/ioctl.h> | |
39 | #include <sys/socket.h> | |
40 | #include <netpacket/packet.h> | |
8b61709d BP |
41 | #include <net/if.h> |
42 | #include <net/if_arp.h> | |
43 | #include <net/if_packet.h> | |
44 | #include <net/route.h> | |
45 | #include <netinet/in.h> | |
e9e28be3 | 46 | #include <poll.h> |
8b61709d BP |
47 | #include <stdlib.h> |
48 | #include <string.h> | |
49 | #include <unistd.h> | |
e9e28be3 BP |
50 | |
51 | #include "coverage.h" | |
9fe3b9a2 | 52 | #include "dpif-linux.h" |
8b61709d BP |
53 | #include "dynamic-string.h" |
54 | #include "fatal-signal.h" | |
93b13be8 BP |
55 | #include "hash.h" |
56 | #include "hmap.h" | |
8b61709d | 57 | #include "netdev-provider.h" |
7fbef77a | 58 | #include "netdev-vport.h" |
e9e28be3 | 59 | #include "netlink.h" |
45c8d3a1 | 60 | #include "netlink-notifier.h" |
2fe27d5a | 61 | #include "netlink-socket.h" |
e9e28be3 | 62 | #include "ofpbuf.h" |
8b61709d BP |
63 | #include "openflow/openflow.h" |
64 | #include "packets.h" | |
65 | #include "poll-loop.h" | |
21d6e22e | 66 | #include "rtnetlink-link.h" |
8b61709d BP |
67 | #include "socket-util.h" |
68 | #include "shash.h" | |
19993ef3 | 69 | #include "sset.h" |
1670c579 | 70 | #include "timer.h" |
e9e28be3 | 71 | #include "vlog.h" |
5136ce49 | 72 | |
d98e6007 | 73 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea | 74 | |
d76f09ea BP |
75 | COVERAGE_DEFINE(netdev_set_policing); |
76 | COVERAGE_DEFINE(netdev_arp_lookup); | |
77 | COVERAGE_DEFINE(netdev_get_ifindex); | |
78 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
79 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
ab985a77 BP |
80 | COVERAGE_DEFINE(netdev_get_ethtool); |
81 | COVERAGE_DEFINE(netdev_set_ethtool); | |
4f925bd3 | 82 | |
8b61709d BP |
83 | \f |
84 | /* These were introduced in Linux 2.6.14, so they might be missing if we have | |
85 | * old headers. */ | |
86 | #ifndef ADVERTISED_Pause | |
87 | #define ADVERTISED_Pause (1 << 13) | |
88 | #endif | |
89 | #ifndef ADVERTISED_Asym_Pause | |
90 | #define ADVERTISED_Asym_Pause (1 << 14) | |
91 | #endif | |
92 | ||
e47bd51a JP |
93 | /* These were introduced in Linux 2.6.24, so they might be missing if we |
94 | * have old headers. */ | |
95 | #ifndef ETHTOOL_GFLAGS | |
96 | #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ | |
97 | #endif | |
98 | #ifndef ETHTOOL_SFLAGS | |
99 | #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ | |
100 | #endif | |
101 | ||
c1c9c9c4 BP |
102 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
103 | * headers. */ | |
104 | #ifndef TC_RTAB_SIZE | |
105 | #define TC_RTAB_SIZE 1024 | |
106 | #endif | |
107 | ||
2ee6545f | 108 | static struct nln_notifier *netdev_linux_cache_notifier = NULL; |
46415c90 | 109 | static int cache_notifier_refcount; |
8b61709d BP |
110 | |
111 | enum { | |
7fbef77a JG |
112 | VALID_IFINDEX = 1 << 0, |
113 | VALID_ETHERADDR = 1 << 1, | |
114 | VALID_IN4 = 1 << 2, | |
115 | VALID_IN6 = 1 << 3, | |
116 | VALID_MTU = 1 << 4, | |
3a183124 | 117 | VALID_POLICING = 1 << 5, |
4f925bd3 PS |
118 | VALID_VPORT_STAT_ERROR = 1 << 6, |
119 | VALID_DRVINFO = 1 << 7, | |
51f87458 | 120 | VALID_FEATURES = 1 << 8, |
8b61709d BP |
121 | }; |
122 | ||
149f577a JG |
123 | struct tap_state { |
124 | int fd; | |
61b999dd | 125 | bool opened; |
149f577a | 126 | }; |
c1c9c9c4 BP |
127 | \f |
128 | /* Traffic control. */ | |
129 | ||
130 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
131 | * network device. |
132 | * | |
133 | * Each TC implementation subclasses this with whatever additional data it | |
134 | * needs. */ | |
c1c9c9c4 BP |
135 | struct tc { |
136 | const struct tc_ops *ops; | |
93b13be8 BP |
137 | struct hmap queues; /* Contains "struct tc_queue"s. |
138 | * Read by generic TC layer. | |
139 | * Written only by TC implementation. */ | |
140 | }; | |
c1c9c9c4 | 141 | |
93b13be8 BP |
142 | /* One traffic control queue. |
143 | * | |
144 | * Each TC implementation subclasses this with whatever additional data it | |
145 | * needs. */ | |
146 | struct tc_queue { | |
147 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
148 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
c1c9c9c4 BP |
149 | }; |
150 | ||
151 | /* A particular kind of traffic control. Each implementation generally maps to | |
152 | * one particular Linux qdisc class. | |
153 | * | |
154 | * The functions below return 0 if successful or a positive errno value on | |
155 | * failure, except where otherwise noted. All of them must be provided, except | |
156 | * where otherwise noted. */ | |
157 | struct tc_ops { | |
158 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
159 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
160 | * appropriate values. */ | |
161 | const char *linux_name; | |
162 | ||
163 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
164 | const char *ovs_name; | |
165 | ||
166 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
167 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
168 | unsigned int n_queues; | |
169 | ||
170 | /* Called to install this TC class on 'netdev'. The implementation should | |
171 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
172 | * and configure it according to 'details'. The implementation may assume | |
173 | * that the current qdisc is the default; that is, there is no need for it | |
174 | * to delete the current qdisc before installing itself. | |
175 | * | |
176 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
177 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
178 | * (which is built as ovs-vswitchd.conf.db(8)). | |
179 | * | |
180 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
181 | * initialized 'struct tc'. | |
182 | * | |
183 | * (This function is null for tc_ops_other, which cannot be installed. For | |
184 | * other TC classes it should always be nonnull.) */ | |
79f1cbe9 | 185 | int (*tc_install)(struct netdev *netdev, const struct smap *details); |
c1c9c9c4 BP |
186 | |
187 | /* Called when the netdev code determines (through a Netlink query) that | |
188 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
189 | * it ourselves and so don't know any of the details. | |
190 | * | |
191 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
192 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
193 | * implementation should parse the other attributes of 'nlmsg' as | |
194 | * necessary to determine its configuration. If necessary it should also | |
195 | * use Netlink queries to determine the configuration of queues on | |
196 | * 'netdev'. | |
197 | * | |
198 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
199 | * initialized 'struct tc'. */ | |
200 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
201 | ||
202 | /* Destroys the data structures allocated by the implementation as part of | |
203 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
204 | * tc_destroy(tc). | |
205 | * | |
206 | * The implementation should not need to perform any Netlink calls. If | |
207 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
208 | * (But it may not be desirable.) | |
209 | * | |
210 | * This function may be null if 'tc' is trivial. */ | |
211 | void (*tc_destroy)(struct tc *tc); | |
212 | ||
213 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
214 | * | |
215 | * The implementation should not need to perform any Netlink calls, because | |
216 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
217 | * cached the configuration. | |
218 | * | |
219 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
220 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
221 | * (which is built as ovs-vswitchd.conf.db(8)). | |
222 | * | |
223 | * This function may be null if 'tc' is not configurable. | |
224 | */ | |
79f1cbe9 | 225 | int (*qdisc_get)(const struct netdev *netdev, struct smap *details); |
c1c9c9c4 BP |
226 | |
227 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
228 | * required Netlink calls to complete the reconfiguration. | |
229 | * | |
230 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
231 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
232 | * (which is built as ovs-vswitchd.conf.db(8)). | |
233 | * | |
234 | * This function may be null if 'tc' is not configurable. | |
235 | */ | |
79f1cbe9 | 236 | int (*qdisc_set)(struct netdev *, const struct smap *details); |
c1c9c9c4 | 237 | |
93b13be8 BP |
238 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
239 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
240 | * |
241 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
242 | * in the "other_config" column in the "Queue" table in | |
243 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
244 | * | |
245 | * The implementation should not need to perform any Netlink calls, because | |
246 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
247 | * cached the queue configuration. | |
248 | * | |
249 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
250 | * 0). */ | |
93b13be8 | 251 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
79f1cbe9 | 252 | struct smap *details); |
c1c9c9c4 BP |
253 | |
254 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
255 | * 'details', perfoming any required Netlink calls to complete the | |
256 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
257 | * 'n_queues'. | |
258 | * | |
259 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
260 | * in the "other_config" column in the "Queue" table in | |
261 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
262 | * | |
263 | * This function may be null if 'tc' does not have queues or its queues are | |
264 | * not configurable. */ | |
265 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
79f1cbe9 | 266 | const struct smap *details); |
c1c9c9c4 | 267 | |
93b13be8 BP |
268 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
269 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
270 | * |
271 | * This function may be null if 'tc' does not have queues or its queues | |
272 | * cannot be deleted. */ | |
93b13be8 | 273 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 274 | |
93b13be8 BP |
275 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
276 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
277 | * |
278 | * On success, initializes '*stats'. | |
279 | * | |
280 | * This function may be null if 'tc' does not have queues or if it cannot | |
281 | * report queue statistics. */ | |
93b13be8 BP |
282 | int (*class_get_stats)(const struct netdev *netdev, |
283 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
284 | struct netdev_queue_stats *stats); |
285 | ||
286 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
287 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
288 | * | |
289 | * This function may be null if 'tc' does not have queues or if it cannot | |
290 | * report queue statistics. */ | |
291 | int (*class_dump_stats)(const struct netdev *netdev, | |
292 | const struct ofpbuf *nlmsg, | |
293 | netdev_dump_queue_stats_cb *cb, void *aux); | |
294 | }; | |
295 | ||
296 | static void | |
297 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
298 | { | |
299 | tc->ops = ops; | |
93b13be8 | 300 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
301 | } |
302 | ||
303 | static void | |
304 | tc_destroy(struct tc *tc) | |
305 | { | |
93b13be8 | 306 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
307 | } |
308 | ||
309 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 310 | static const struct tc_ops tc_ops_hfsc; |
c1c9c9c4 BP |
311 | static const struct tc_ops tc_ops_default; |
312 | static const struct tc_ops tc_ops_other; | |
313 | ||
314 | static const struct tc_ops *tcs[] = { | |
315 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ | |
a339aa81 | 316 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
c1c9c9c4 BP |
317 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
318 | &tc_ops_other, /* Some other qdisc. */ | |
319 | NULL | |
320 | }; | |
149f577a | 321 | |
c1c9c9c4 BP |
322 | static unsigned int tc_make_handle(unsigned int major, unsigned int minor); |
323 | static unsigned int tc_get_major(unsigned int handle); | |
324 | static unsigned int tc_get_minor(unsigned int handle); | |
325 | ||
326 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); | |
327 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
328 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
329 | ||
330 | static struct tcmsg *tc_make_request(const struct netdev *, int type, | |
331 | unsigned int flags, struct ofpbuf *); | |
332 | static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); | |
f8500004 JP |
333 | static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add); |
334 | static int tc_add_policer(struct netdev *netdev, int kbits_rate, | |
335 | int kbits_burst); | |
c1c9c9c4 BP |
336 | |
337 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
338 | struct nlattr **options); | |
339 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
340 | struct nlattr **options, | |
341 | struct netdev_queue_stats *); | |
342 | static int tc_query_class(const struct netdev *, | |
343 | unsigned int handle, unsigned int parent, | |
344 | struct ofpbuf **replyp); | |
345 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
346 | ||
347 | static int tc_del_qdisc(struct netdev *netdev); | |
348 | static int tc_query_qdisc(const struct netdev *netdev); | |
349 | ||
350 | static int tc_calc_cell_log(unsigned int mtu); | |
351 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
352 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
353 | const struct tc_ratespec *rate); | |
354 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
355 | \f | |
149f577a JG |
356 | struct netdev_dev_linux { |
357 | struct netdev_dev netdev_dev; | |
358 | ||
8b61709d | 359 | struct shash_node *shash_node; |
149f577a | 360 | unsigned int cache_valid; |
ac4d3bcb | 361 | unsigned int change_seq; |
8b61709d | 362 | |
1670c579 EJ |
363 | bool miimon; /* Link status of last poll. */ |
364 | long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ | |
365 | struct timer miimon_timer; | |
366 | ||
8722022c BP |
367 | /* The following are figured out "on demand" only. They are only valid |
368 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d BP |
369 | int ifindex; |
370 | uint8_t etheraddr[ETH_ADDR_LEN]; | |
f1acd62b | 371 | struct in_addr address, netmask; |
8b61709d BP |
372 | struct in6_addr in6; |
373 | int mtu; | |
059e5f4f | 374 | unsigned int ifi_flags; |
65c3058c | 375 | long long int carrier_resets; |
80a86fbe BP |
376 | uint32_t kbits_rate; /* Policing data. */ |
377 | uint32_t kbits_burst; | |
bba1e6f3 PS |
378 | int vport_stats_error; /* Cached error code from vport_get_stats(). |
379 | 0 or an errno value. */ | |
90a6637d | 380 | int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ |
44445cac | 381 | int ether_addr_error; /* Cached error code from set/get etheraddr. */ |
c9f71668 | 382 | int netdev_policing_error; /* Cached error code from set policing. */ |
51f87458 | 383 | int get_features_error; /* Cached error code from ETHTOOL_GSET. */ |
c7b1b0a5 | 384 | int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */ |
51f87458 | 385 | |
a00ca915 EJ |
386 | enum netdev_features current; /* Cached from ETHTOOL_GSET. */ |
387 | enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ | |
388 | enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ | |
389 | enum netdev_features peer; /* Cached from ETHTOOL_GSET. */ | |
90a6637d | 390 | |
4f925bd3 | 391 | struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ |
c1c9c9c4 | 392 | struct tc *tc; |
149f577a JG |
393 | |
394 | union { | |
395 | struct tap_state tap; | |
396 | } state; | |
8b61709d BP |
397 | }; |
398 | ||
149f577a JG |
399 | struct netdev_linux { |
400 | struct netdev netdev; | |
5b7448ed | 401 | int fd; |
149f577a | 402 | }; |
8b61709d | 403 | |
76c308b5 BP |
404 | /* Sockets used for ioctl operations. */ |
405 | static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ | |
8b61709d | 406 | |
ff4ed3c9 BP |
407 | /* A Netlink routing socket that is not subscribed to any multicast groups. */ |
408 | static struct nl_sock *rtnl_sock; | |
409 | ||
8b61709d BP |
410 | /* This is set pretty low because we probably won't learn anything from the |
411 | * additional log messages. */ | |
412 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
413 | ||
15b3596a | 414 | static int netdev_linux_init(void); |
6f643e49 | 415 | |
0b0544d7 | 416 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 417 | int cmd, const char *cmd_name); |
149f577a JG |
418 | static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, |
419 | const char *cmd_name); | |
f1acd62b BP |
420 | static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, |
421 | int cmd, const char *cmd_name); | |
059e5f4f EJ |
422 | static int get_flags(const struct netdev_dev *, unsigned int *flags); |
423 | static int set_flags(struct netdev *, unsigned int flags); | |
8b61709d BP |
424 | static int do_get_ifindex(const char *netdev_name); |
425 | static int get_ifindex(const struct netdev *, int *ifindexp); | |
426 | static int do_set_addr(struct netdev *netdev, | |
427 | int ioctl_nr, const char *ioctl_name, | |
428 | struct in_addr addr); | |
429 | static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); | |
44445cac | 430 | static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]); |
8b61709d BP |
431 | static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); |
432 | static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); | |
488d734d | 433 | static int af_packet_sock(void); |
1670c579 EJ |
434 | static void netdev_linux_miimon_run(void); |
435 | static void netdev_linux_miimon_wait(void); | |
8b61709d | 436 | |
15b3596a JG |
437 | static bool |
438 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
439 | { | |
440 | return netdev_class->init == netdev_linux_init; | |
441 | } | |
442 | ||
149f577a JG |
443 | static struct netdev_dev_linux * |
444 | netdev_dev_linux_cast(const struct netdev_dev *netdev_dev) | |
6c88d577 | 445 | { |
15b3596a JG |
446 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); |
447 | assert(is_netdev_linux_class(netdev_class)); | |
448 | ||
149f577a | 449 | return CONTAINER_OF(netdev_dev, struct netdev_dev_linux, netdev_dev); |
6c88d577 JP |
450 | } |
451 | ||
8b61709d BP |
452 | static struct netdev_linux * |
453 | netdev_linux_cast(const struct netdev *netdev) | |
454 | { | |
15b3596a JG |
455 | struct netdev_dev *netdev_dev = netdev_get_dev(netdev); |
456 | const struct netdev_class *netdev_class = netdev_dev_get_class(netdev_dev); | |
457 | assert(is_netdev_linux_class(netdev_class)); | |
458 | ||
8b61709d BP |
459 | return CONTAINER_OF(netdev, struct netdev_linux, netdev); |
460 | } | |
ff4ed3c9 | 461 | \f |
8b61709d BP |
462 | static int |
463 | netdev_linux_init(void) | |
464 | { | |
465 | static int status = -1; | |
466 | if (status < 0) { | |
ff4ed3c9 | 467 | /* Create AF_INET socket. */ |
8b61709d BP |
468 | af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); |
469 | status = af_inet_sock >= 0 ? 0 : errno; | |
470 | if (status) { | |
471 | VLOG_ERR("failed to create inet socket: %s", strerror(status)); | |
472 | } | |
ff4ed3c9 BP |
473 | |
474 | /* Create rtnetlink socket. */ | |
475 | if (!status) { | |
cceb11f5 | 476 | status = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); |
ff4ed3c9 BP |
477 | if (status) { |
478 | VLOG_ERR_RL(&rl, "failed to create rtnetlink socket: %s", | |
479 | strerror(status)); | |
480 | } | |
481 | } | |
8b61709d BP |
482 | } |
483 | return status; | |
484 | } | |
485 | ||
486 | static void | |
487 | netdev_linux_run(void) | |
488 | { | |
18a23781 | 489 | rtnetlink_link_run(); |
1670c579 | 490 | netdev_linux_miimon_run(); |
8b61709d BP |
491 | } |
492 | ||
493 | static void | |
494 | netdev_linux_wait(void) | |
495 | { | |
18a23781 | 496 | rtnetlink_link_wait(); |
1670c579 | 497 | netdev_linux_miimon_wait(); |
8b61709d BP |
498 | } |
499 | ||
4f925bd3 PS |
500 | static int |
501 | netdev_linux_get_drvinfo(struct netdev_dev_linux *netdev_dev) | |
502 | { | |
503 | ||
504 | int error; | |
505 | ||
506 | if (netdev_dev->cache_valid & VALID_DRVINFO) { | |
507 | return 0; | |
508 | } | |
509 | ||
ab985a77 | 510 | COVERAGE_INC(netdev_get_ethtool); |
4f925bd3 PS |
511 | memset(&netdev_dev->drvinfo, 0, sizeof netdev_dev->drvinfo); |
512 | error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, | |
513 | (struct ethtool_cmd *)&netdev_dev->drvinfo, | |
514 | ETHTOOL_GDRVINFO, | |
515 | "ETHTOOL_GDRVINFO"); | |
516 | if (!error) { | |
517 | netdev_dev->cache_valid |= VALID_DRVINFO; | |
518 | } | |
519 | return error; | |
520 | } | |
521 | ||
ac4d3bcb | 522 | static void |
4f925bd3 PS |
523 | netdev_dev_linux_changed(struct netdev_dev_linux *dev, |
524 | unsigned int ifi_flags, | |
525 | unsigned int mask) | |
ac4d3bcb EJ |
526 | { |
527 | dev->change_seq++; | |
528 | if (!dev->change_seq) { | |
529 | dev->change_seq++; | |
530 | } | |
8aa77183 BP |
531 | |
532 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
533 | dev->carrier_resets++; | |
534 | } | |
535 | dev->ifi_flags = ifi_flags; | |
536 | ||
4f925bd3 PS |
537 | dev->cache_valid &= mask; |
538 | } | |
539 | ||
540 | static void | |
541 | netdev_dev_linux_update(struct netdev_dev_linux *dev, | |
542 | const struct rtnetlink_link_change *change) | |
543 | { | |
544 | if (change->nlmsg_type == RTM_NEWLINK) { | |
545 | /* Keep drv-info */ | |
546 | netdev_dev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO); | |
90a6637d | 547 | |
c7b1b0a5 | 548 | /* Update netdev from rtnl-change msg. */ |
90a6637d PS |
549 | if (change->mtu) { |
550 | dev->mtu = change->mtu; | |
551 | dev->cache_valid |= VALID_MTU; | |
552 | dev->netdev_mtu_error = 0; | |
553 | } | |
554 | ||
44445cac PS |
555 | if (!eth_addr_is_zero(change->addr)) { |
556 | memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN); | |
557 | dev->cache_valid |= VALID_ETHERADDR; | |
558 | dev->ether_addr_error = 0; | |
559 | } | |
560 | ||
c7b1b0a5 PS |
561 | dev->ifindex = change->ifi_index; |
562 | dev->cache_valid |= VALID_IFINDEX; | |
563 | dev->get_ifindex_error = 0; | |
564 | ||
4f925bd3 PS |
565 | } else { |
566 | netdev_dev_linux_changed(dev, change->ifi_flags, 0); | |
567 | } | |
ac4d3bcb EJ |
568 | } |
569 | ||
8b61709d | 570 | static void |
21d6e22e | 571 | netdev_linux_cache_cb(const struct rtnetlink_link_change *change, |
67a4917b | 572 | void *aux OVS_UNUSED) |
8b61709d | 573 | { |
149f577a | 574 | struct netdev_dev_linux *dev; |
8b61709d | 575 | if (change) { |
46415c90 JG |
576 | struct netdev_dev *base_dev = netdev_dev_from_name(change->ifname); |
577 | if (base_dev) { | |
15b3596a JG |
578 | const struct netdev_class *netdev_class = |
579 | netdev_dev_get_class(base_dev); | |
580 | ||
581 | if (is_netdev_linux_class(netdev_class)) { | |
582 | dev = netdev_dev_linux_cast(base_dev); | |
4f925bd3 | 583 | netdev_dev_linux_update(dev, change); |
15b3596a | 584 | } |
8b61709d BP |
585 | } |
586 | } else { | |
46415c90 | 587 | struct shash device_shash; |
8b61709d | 588 | struct shash_node *node; |
46415c90 JG |
589 | |
590 | shash_init(&device_shash); | |
591 | netdev_dev_get_devices(&netdev_linux_class, &device_shash); | |
592 | SHASH_FOR_EACH (node, &device_shash) { | |
059e5f4f | 593 | unsigned int flags; |
3a183124 | 594 | |
149f577a | 595 | dev = node->data; |
3a183124 | 596 | |
755be9ea | 597 | get_flags(&dev->netdev_dev, &flags); |
4f925bd3 | 598 | netdev_dev_linux_changed(dev, flags, 0); |
8b61709d | 599 | } |
46415c90 | 600 | shash_destroy(&device_shash); |
8b61709d BP |
601 | } |
602 | } | |
603 | ||
604 | static int | |
1f6e0fbd | 605 | cache_notifier_ref(void) |
6c88d577 | 606 | { |
46415c90 | 607 | if (!cache_notifier_refcount) { |
2ee6545f EJ |
608 | assert(!netdev_linux_cache_notifier); |
609 | ||
610 | netdev_linux_cache_notifier = | |
611 | rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL); | |
612 | ||
613 | if (!netdev_linux_cache_notifier) { | |
614 | return EINVAL; | |
149f577a JG |
615 | } |
616 | } | |
46415c90 | 617 | cache_notifier_refcount++; |
6c88d577 | 618 | |
1f6e0fbd BP |
619 | return 0; |
620 | } | |
621 | ||
622 | static void | |
623 | cache_notifier_unref(void) | |
624 | { | |
625 | assert(cache_notifier_refcount > 0); | |
626 | if (!--cache_notifier_refcount) { | |
627 | assert(netdev_linux_cache_notifier); | |
628 | rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier); | |
629 | netdev_linux_cache_notifier = NULL; | |
630 | } | |
631 | } | |
632 | ||
633 | /* Creates system and internal devices. */ | |
634 | static int | |
635 | netdev_linux_create(const struct netdev_class *class, const char *name, | |
636 | struct netdev_dev **netdev_devp) | |
637 | { | |
638 | struct netdev_dev_linux *netdev_dev; | |
639 | int error; | |
640 | ||
641 | error = cache_notifier_ref(); | |
642 | if (error) { | |
643 | return error; | |
644 | } | |
645 | ||
149f577a | 646 | netdev_dev = xzalloc(sizeof *netdev_dev); |
ac4d3bcb | 647 | netdev_dev->change_seq = 1; |
de5cdb90 | 648 | netdev_dev_init(&netdev_dev->netdev_dev, name, class); |
c37d4da4 | 649 | get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags); |
46415c90 | 650 | |
149f577a | 651 | *netdev_devp = &netdev_dev->netdev_dev; |
a740f0de JG |
652 | return 0; |
653 | } | |
654 | ||
5b7448ed JG |
655 | /* For most types of netdevs we open the device for each call of |
656 | * netdev_open(). However, this is not the case with tap devices, | |
657 | * since it is only possible to open the device once. In this | |
658 | * situation we share a single file descriptor, and consequently | |
659 | * buffers, across all readers. Therefore once data is read it will | |
660 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 661 | static int |
b8dcf5e9 | 662 | netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, |
de5cdb90 | 663 | const char *name, struct netdev_dev **netdev_devp) |
a740f0de | 664 | { |
149f577a | 665 | struct netdev_dev_linux *netdev_dev; |
a740f0de JG |
666 | struct tap_state *state; |
667 | static const char tap_dev[] = "/dev/net/tun"; | |
668 | struct ifreq ifr; | |
669 | int error; | |
670 | ||
149f577a JG |
671 | netdev_dev = xzalloc(sizeof *netdev_dev); |
672 | state = &netdev_dev->state.tap; | |
a740f0de | 673 | |
1f6e0fbd BP |
674 | error = cache_notifier_ref(); |
675 | if (error) { | |
676 | goto error; | |
677 | } | |
678 | ||
6c88d577 | 679 | /* Open tap device. */ |
149f577a JG |
680 | state->fd = open(tap_dev, O_RDWR); |
681 | if (state->fd < 0) { | |
6c88d577 JP |
682 | error = errno; |
683 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, strerror(error)); | |
1f6e0fbd | 684 | goto error_unref_notifier; |
6c88d577 JP |
685 | } |
686 | ||
687 | /* Create tap device. */ | |
688 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
71d7c22f | 689 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
149f577a | 690 | if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 JP |
691 | VLOG_WARN("%s: creating tap device failed: %s", name, |
692 | strerror(errno)); | |
693 | error = errno; | |
1f6e0fbd | 694 | goto error_unref_notifier; |
6c88d577 JP |
695 | } |
696 | ||
697 | /* Make non-blocking. */ | |
149f577a | 698 | error = set_nonblocking(state->fd); |
a740f0de | 699 | if (error) { |
1f6e0fbd | 700 | goto error_unref_notifier; |
a740f0de JG |
701 | } |
702 | ||
de5cdb90 | 703 | netdev_dev_init(&netdev_dev->netdev_dev, name, &netdev_tap_class); |
149f577a | 704 | *netdev_devp = &netdev_dev->netdev_dev; |
a740f0de JG |
705 | return 0; |
706 | ||
1f6e0fbd BP |
707 | error_unref_notifier: |
708 | cache_notifier_unref(); | |
a740f0de | 709 | error: |
149f577a | 710 | free(netdev_dev); |
a740f0de JG |
711 | return error; |
712 | } | |
713 | ||
a740f0de | 714 | static void |
149f577a | 715 | destroy_tap(struct netdev_dev_linux *netdev_dev) |
a740f0de | 716 | { |
149f577a JG |
717 | struct tap_state *state = &netdev_dev->state.tap; |
718 | ||
719 | if (state->fd >= 0) { | |
720 | close(state->fd); | |
a740f0de JG |
721 | } |
722 | } | |
723 | ||
149f577a | 724 | /* Destroys the netdev device 'netdev_dev_'. */ |
6c88d577 | 725 | static void |
149f577a | 726 | netdev_linux_destroy(struct netdev_dev *netdev_dev_) |
6c88d577 | 727 | { |
149f577a | 728 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
d2bb2799 | 729 | const struct netdev_class *class = netdev_dev_get_class(netdev_dev_); |
6c88d577 | 730 | |
c1c9c9c4 BP |
731 | if (netdev_dev->tc && netdev_dev->tc->ops->tc_destroy) { |
732 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
733 | } | |
734 | ||
1f6e0fbd | 735 | if (class == &netdev_tap_class) { |
149f577a | 736 | destroy_tap(netdev_dev); |
6c88d577 | 737 | } |
658797c8 | 738 | free(netdev_dev); |
1f6e0fbd BP |
739 | |
740 | cache_notifier_unref(); | |
6c88d577 JP |
741 | } |
742 | ||
8b61709d | 743 | static int |
7b6b0ef4 | 744 | netdev_linux_open(struct netdev_dev *netdev_dev_, struct netdev **netdevp) |
8b61709d | 745 | { |
5b7448ed | 746 | struct netdev_dev_linux *netdev_dev = netdev_dev_linux_cast(netdev_dev_); |
8b61709d BP |
747 | struct netdev_linux *netdev; |
748 | enum netdev_flags flags; | |
749 | int error; | |
750 | ||
751 | /* Allocate network device. */ | |
ec6fde61 | 752 | netdev = xzalloc(sizeof *netdev); |
49a6a163 | 753 | netdev->fd = -1; |
5b7448ed | 754 | netdev_init(&netdev->netdev, netdev_dev_); |
8b61709d | 755 | |
c3827f61 BP |
756 | /* Verify that the device really exists, by attempting to read its flags. |
757 | * (The flags might be cached, in which case this won't actually do an | |
758 | * ioctl.) | |
759 | * | |
760 | * Don't do this for "internal" netdevs, though, because those have to be | |
761 | * created as netdev objects before they exist in the kernel, because | |
762 | * creating them in the kernel happens by passing a netdev object to | |
763 | * dpif_port_add(). */ | |
764 | if (netdev_dev_get_class(netdev_dev_) != &netdev_internal_class) { | |
765 | error = netdev_get_flags(&netdev->netdev, &flags); | |
766 | if (error == ENODEV) { | |
767 | goto error; | |
768 | } | |
8b61709d BP |
769 | } |
770 | ||
61b999dd JG |
771 | if (!strcmp(netdev_dev_get_type(netdev_dev_), "tap") && |
772 | !netdev_dev->state.tap.opened) { | |
773 | ||
774 | /* We assume that the first user of the tap device is the primary user | |
775 | * and give them the tap FD. Subsequent users probably just expect | |
776 | * this to be a system device so open it normally to avoid send/receive | |
777 | * directions appearing to be reversed. */ | |
5b7448ed | 778 | netdev->fd = netdev_dev->state.tap.fd; |
61b999dd | 779 | netdev_dev->state.tap.opened = true; |
8b61709d BP |
780 | } |
781 | ||
782 | *netdevp = &netdev->netdev; | |
783 | return 0; | |
784 | ||
785 | error: | |
149f577a | 786 | netdev_uninit(&netdev->netdev, true); |
8b61709d BP |
787 | return error; |
788 | } | |
789 | ||
790 | /* Closes and destroys 'netdev'. */ | |
791 | static void | |
792 | netdev_linux_close(struct netdev *netdev_) | |
793 | { | |
794 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
795 | ||
49a6a163 | 796 | if (netdev->fd > 0 && strcmp(netdev_get_type(netdev_), "tap")) { |
5b7448ed | 797 | close(netdev->fd); |
8b61709d BP |
798 | } |
799 | free(netdev); | |
800 | } | |
e9e28be3 | 801 | |
7b6b0ef4 BP |
802 | static int |
803 | netdev_linux_listen(struct netdev *netdev_) | |
804 | { | |
805 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
806 | struct sockaddr_ll sll; | |
807 | int ifindex; | |
808 | int error; | |
809 | int fd; | |
810 | ||
811 | if (netdev->fd >= 0) { | |
812 | return 0; | |
813 | } | |
814 | ||
815 | /* Create file descriptor. */ | |
816 | fd = socket(PF_PACKET, SOCK_RAW, 0); | |
817 | if (fd < 0) { | |
818 | error = errno; | |
819 | VLOG_ERR("failed to create raw socket (%s)", strerror(error)); | |
820 | goto error; | |
821 | } | |
822 | ||
823 | /* Set non-blocking mode. */ | |
824 | error = set_nonblocking(fd); | |
825 | if (error) { | |
826 | goto error; | |
827 | } | |
828 | ||
829 | /* Get ethernet device index. */ | |
830 | error = get_ifindex(&netdev->netdev, &ifindex); | |
831 | if (error) { | |
832 | goto error; | |
833 | } | |
834 | ||
835 | /* Bind to specific ethernet device. */ | |
836 | memset(&sll, 0, sizeof sll); | |
837 | sll.sll_family = AF_PACKET; | |
838 | sll.sll_ifindex = ifindex; | |
839 | sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL); | |
840 | if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) { | |
841 | error = errno; | |
842 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
843 | netdev_get_name(netdev_), strerror(error)); | |
844 | goto error; | |
845 | } | |
846 | ||
847 | netdev->fd = fd; | |
848 | return 0; | |
849 | ||
850 | error: | |
851 | if (fd >= 0) { | |
852 | close(fd); | |
853 | } | |
854 | return error; | |
855 | } | |
856 | ||
8b61709d BP |
857 | static int |
858 | netdev_linux_recv(struct netdev *netdev_, void *data, size_t size) | |
859 | { | |
860 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
861 | ||
5b7448ed | 862 | if (netdev->fd < 0) { |
7b6b0ef4 | 863 | /* Device is not listening. */ |
c0e5f6ca | 864 | return -EAGAIN; |
8b61709d BP |
865 | } |
866 | ||
867 | for (;;) { | |
8e8cddf7 BP |
868 | ssize_t retval; |
869 | ||
870 | retval = (netdev_->netdev_dev->netdev_class == &netdev_tap_class | |
871 | ? read(netdev->fd, data, size) | |
872 | : recv(netdev->fd, data, size, MSG_TRUNC)); | |
0e15264f BP |
873 | if (retval >= 0) { |
874 | return retval <= size ? retval : -EMSGSIZE; | |
8b61709d BP |
875 | } else if (errno != EINTR) { |
876 | if (errno != EAGAIN) { | |
877 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
878 | strerror(errno), netdev_get_name(netdev_)); | |
879 | } | |
c0e5f6ca | 880 | return -errno; |
8b61709d BP |
881 | } |
882 | } | |
883 | } | |
884 | ||
885 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
886 | * when a packet is ready to be received with netdev_recv() on 'netdev'. */ | |
887 | static void | |
888 | netdev_linux_recv_wait(struct netdev *netdev_) | |
889 | { | |
890 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed JG |
891 | if (netdev->fd >= 0) { |
892 | poll_fd_wait(netdev->fd, POLLIN); | |
8b61709d BP |
893 | } |
894 | } | |
895 | ||
896 | /* Discards all packets waiting to be received from 'netdev'. */ | |
897 | static int | |
898 | netdev_linux_drain(struct netdev *netdev_) | |
899 | { | |
900 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 901 | if (netdev->fd < 0) { |
8b61709d | 902 | return 0; |
5b7448ed | 903 | } else if (!strcmp(netdev_get_type(netdev_), "tap")) { |
8b61709d | 904 | struct ifreq ifr; |
149f577a | 905 | int error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
8b61709d BP |
906 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
907 | if (error) { | |
908 | return error; | |
909 | } | |
5b7448ed | 910 | drain_fd(netdev->fd, ifr.ifr_qlen); |
8b61709d BP |
911 | return 0; |
912 | } else { | |
5b7448ed | 913 | return drain_rcvbuf(netdev->fd); |
8b61709d BP |
914 | } |
915 | } | |
916 | ||
917 | /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive | |
918 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued | |
919 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
920 | * the packet is too big or too small to transmit on the device. | |
921 | * | |
922 | * The caller retains ownership of 'buffer' in all cases. | |
923 | * | |
924 | * The kernel maintains a packet transmission queue, so the caller is not | |
925 | * expected to do additional queuing of packets. */ | |
926 | static int | |
927 | netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) | |
928 | { | |
f23347ea BP |
929 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
930 | for (;;) { | |
931 | ssize_t retval; | |
8b61709d | 932 | |
f23347ea BP |
933 | if (netdev->fd < 0) { |
934 | /* Use our AF_PACKET socket to send to this device. */ | |
935 | struct sockaddr_ll sll; | |
936 | struct msghdr msg; | |
937 | struct iovec iov; | |
938 | int ifindex; | |
939 | int error; | |
488d734d BP |
940 | int sock; |
941 | ||
942 | sock = af_packet_sock(); | |
943 | if (sock < 0) { | |
944 | return sock; | |
945 | } | |
f23347ea BP |
946 | |
947 | error = get_ifindex(netdev_, &ifindex); | |
948 | if (error) { | |
949 | return error; | |
950 | } | |
8b61709d | 951 | |
f23347ea BP |
952 | /* We don't bother setting most fields in sockaddr_ll because the |
953 | * kernel ignores them for SOCK_RAW. */ | |
954 | memset(&sll, 0, sizeof sll); | |
955 | sll.sll_family = AF_PACKET; | |
956 | sll.sll_ifindex = ifindex; | |
76c308b5 | 957 | |
ebc56baa | 958 | iov.iov_base = CONST_CAST(void *, data); |
f23347ea | 959 | iov.iov_len = size; |
76c308b5 | 960 | |
f23347ea BP |
961 | msg.msg_name = &sll; |
962 | msg.msg_namelen = sizeof sll; | |
963 | msg.msg_iov = &iov; | |
964 | msg.msg_iovlen = 1; | |
965 | msg.msg_control = NULL; | |
966 | msg.msg_controllen = 0; | |
967 | msg.msg_flags = 0; | |
968 | ||
488d734d | 969 | retval = sendmsg(sock, &msg, 0); |
f23347ea BP |
970 | } else { |
971 | /* Use the netdev's own fd to send to this device. This is | |
972 | * essential for tap devices, because packets sent to a tap device | |
973 | * with an AF_PACKET socket will loop back to be *received* again | |
974 | * on the tap device. */ | |
975 | retval = write(netdev->fd, data, size); | |
976 | } | |
76c308b5 | 977 | |
8b61709d BP |
978 | if (retval < 0) { |
979 | /* The Linux AF_PACKET implementation never blocks waiting for room | |
980 | * for packets, instead returning ENOBUFS. Translate this into | |
981 | * EAGAIN for the caller. */ | |
982 | if (errno == ENOBUFS) { | |
983 | return EAGAIN; | |
984 | } else if (errno == EINTR) { | |
985 | continue; | |
986 | } else if (errno != EAGAIN) { | |
987 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", | |
988 | netdev_get_name(netdev_), strerror(errno)); | |
989 | } | |
990 | return errno; | |
991 | } else if (retval != size) { | |
992 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " | |
993 | "%zu) on %s", retval, size, netdev_get_name(netdev_)); | |
994 | return EMSGSIZE; | |
995 | } else { | |
996 | return 0; | |
997 | } | |
998 | } | |
999 | } | |
1000 | ||
1001 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
1002 | * when the packet transmission queue has sufficient room to transmit a packet | |
1003 | * with netdev_send(). | |
1004 | * | |
1005 | * The kernel maintains a packet transmission queue, so the client is not | |
1006 | * expected to do additional queuing of packets. Thus, this function is | |
1007 | * unlikely to ever be used. It is included for completeness. */ | |
1008 | static void | |
1009 | netdev_linux_send_wait(struct netdev *netdev_) | |
1010 | { | |
1011 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5b7448ed | 1012 | if (netdev->fd < 0) { |
8b61709d | 1013 | /* Nothing to do. */ |
5b7448ed JG |
1014 | } else if (strcmp(netdev_get_type(netdev_), "tap")) { |
1015 | poll_fd_wait(netdev->fd, POLLOUT); | |
8b61709d BP |
1016 | } else { |
1017 | /* TAP device always accepts packets.*/ | |
1018 | poll_immediate_wake(); | |
1019 | } | |
1020 | } | |
1021 | ||
1022 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
1023 | * otherwise a positive errno value. */ | |
1024 | static int | |
1025 | netdev_linux_set_etheraddr(struct netdev *netdev_, | |
1026 | const uint8_t mac[ETH_ADDR_LEN]) | |
1027 | { | |
149f577a JG |
1028 | struct netdev_dev_linux *netdev_dev = |
1029 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
eb395f2e BP |
1030 | int error; |
1031 | ||
44445cac PS |
1032 | if (netdev_dev->cache_valid & VALID_ETHERADDR) { |
1033 | if (netdev_dev->ether_addr_error) { | |
1034 | return netdev_dev->ether_addr_error; | |
1035 | } | |
1036 | if (eth_addr_equals(netdev_dev->etheraddr, mac)) { | |
1037 | return 0; | |
1038 | } | |
1039 | netdev_dev->cache_valid &= ~VALID_ETHERADDR; | |
1040 | } | |
1041 | ||
1042 | error = set_etheraddr(netdev_get_name(netdev_), mac); | |
1043 | if (!error || error == ENODEV) { | |
1044 | netdev_dev->ether_addr_error = error; | |
1045 | netdev_dev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 1046 | if (!error) { |
149f577a | 1047 | memcpy(netdev_dev->etheraddr, mac, ETH_ADDR_LEN); |
eb395f2e | 1048 | } |
8b61709d | 1049 | } |
44445cac | 1050 | |
8b61709d BP |
1051 | return error; |
1052 | } | |
1053 | ||
44445cac | 1054 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d BP |
1055 | static int |
1056 | netdev_linux_get_etheraddr(const struct netdev *netdev_, | |
1057 | uint8_t mac[ETH_ADDR_LEN]) | |
1058 | { | |
149f577a JG |
1059 | struct netdev_dev_linux *netdev_dev = |
1060 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
44445cac | 1061 | |
149f577a | 1062 | if (!(netdev_dev->cache_valid & VALID_ETHERADDR)) { |
8b61709d | 1063 | int error = get_etheraddr(netdev_get_name(netdev_), |
149f577a | 1064 | netdev_dev->etheraddr); |
44445cac PS |
1065 | |
1066 | netdev_dev->ether_addr_error = error; | |
149f577a | 1067 | netdev_dev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 1068 | } |
44445cac PS |
1069 | |
1070 | if (!netdev_dev->ether_addr_error) { | |
1071 | memcpy(mac, netdev_dev->etheraddr, ETH_ADDR_LEN); | |
1072 | } | |
1073 | ||
1074 | return netdev_dev->ether_addr_error; | |
8b61709d BP |
1075 | } |
1076 | ||
1077 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1078 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1079 | * bytes for Ethernet devices. */ | |
1080 | static int | |
1081 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1082 | { | |
149f577a JG |
1083 | struct netdev_dev_linux *netdev_dev = |
1084 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1085 | if (!(netdev_dev->cache_valid & VALID_MTU)) { | |
8b61709d BP |
1086 | struct ifreq ifr; |
1087 | int error; | |
1088 | ||
149f577a JG |
1089 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
1090 | SIOCGIFMTU, "SIOCGIFMTU"); | |
90a6637d PS |
1091 | |
1092 | netdev_dev->netdev_mtu_error = error; | |
149f577a JG |
1093 | netdev_dev->mtu = ifr.ifr_mtu; |
1094 | netdev_dev->cache_valid |= VALID_MTU; | |
8b61709d | 1095 | } |
90a6637d PS |
1096 | |
1097 | if (!netdev_dev->netdev_mtu_error) { | |
1098 | *mtup = netdev_dev->mtu; | |
1099 | } | |
1100 | return netdev_dev->netdev_mtu_error; | |
8b61709d BP |
1101 | } |
1102 | ||
9b020780 PS |
1103 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1104 | * networking ioctl interface. | |
1105 | */ | |
1106 | static int | |
1107 | netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) | |
1108 | { | |
1109 | struct netdev_dev_linux *netdev_dev = | |
1110 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1111 | struct ifreq ifr; | |
1112 | int error; | |
1113 | ||
90a6637d PS |
1114 | if (netdev_dev->cache_valid & VALID_MTU) { |
1115 | if (netdev_dev->netdev_mtu_error) { | |
1116 | return netdev_dev->netdev_mtu_error; | |
1117 | } | |
1118 | if (netdev_dev->mtu == mtu) { | |
1119 | return 0; | |
1120 | } | |
1121 | netdev_dev->cache_valid &= ~VALID_MTU; | |
153e5481 | 1122 | } |
9b020780 PS |
1123 | ifr.ifr_mtu = mtu; |
1124 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, | |
1125 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d PS |
1126 | if (!error || error == ENODEV) { |
1127 | netdev_dev->netdev_mtu_error = error; | |
1128 | netdev_dev->mtu = ifr.ifr_mtu; | |
1129 | netdev_dev->cache_valid |= VALID_MTU; | |
9b020780 | 1130 | } |
90a6637d | 1131 | return error; |
9b020780 PS |
1132 | } |
1133 | ||
9ab3d9a3 BP |
1134 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1135 | * On failure, returns a negative errno value. */ | |
1136 | static int | |
1137 | netdev_linux_get_ifindex(const struct netdev *netdev) | |
1138 | { | |
1139 | int ifindex, error; | |
1140 | ||
1141 | error = get_ifindex(netdev, &ifindex); | |
1142 | return error ? -error : ifindex; | |
1143 | } | |
1144 | ||
8b61709d BP |
1145 | static int |
1146 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1147 | { | |
149f577a JG |
1148 | struct netdev_dev_linux *netdev_dev = |
1149 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d | 1150 | |
1670c579 EJ |
1151 | if (netdev_dev->miimon_interval > 0) { |
1152 | *carrier = netdev_dev->miimon; | |
3a183124 | 1153 | } else { |
c37d4da4 | 1154 | *carrier = (netdev_dev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1155 | } |
8b61709d | 1156 | |
3a183124 | 1157 | return 0; |
8b61709d BP |
1158 | } |
1159 | ||
65c3058c EJ |
1160 | static long long int |
1161 | netdev_linux_get_carrier_resets(const struct netdev *netdev) | |
1162 | { | |
1163 | return netdev_dev_linux_cast(netdev_get_dev(netdev))->carrier_resets; | |
1164 | } | |
1165 | ||
63331829 | 1166 | static int |
1670c579 EJ |
1167 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1168 | struct mii_ioctl_data *data) | |
63331829 | 1169 | { |
63331829 | 1170 | struct ifreq ifr; |
782e6111 | 1171 | int error; |
63331829 | 1172 | |
63331829 | 1173 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1174 | memcpy(&ifr.ifr_data, data, sizeof *data); |
1670c579 | 1175 | error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1176 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1177 | |
782e6111 EJ |
1178 | return error; |
1179 | } | |
1180 | ||
1181 | static int | |
1670c579 | 1182 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1183 | { |
782e6111 EJ |
1184 | struct mii_ioctl_data data; |
1185 | int error; | |
63331829 | 1186 | |
782e6111 EJ |
1187 | *miimon = false; |
1188 | ||
1189 | memset(&data, 0, sizeof data); | |
1670c579 | 1190 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
1191 | if (!error) { |
1192 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1193 | data.reg_num = MII_BMSR; | |
1670c579 | 1194 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 1195 | &data); |
63331829 EJ |
1196 | |
1197 | if (!error) { | |
782e6111 | 1198 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 EJ |
1199 | } else { |
1200 | VLOG_WARN_RL(&rl, "%s: failed to query MII", name); | |
1201 | } | |
1202 | } else { | |
1203 | struct ethtool_cmd ecmd; | |
63331829 EJ |
1204 | |
1205 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1206 | name); | |
1207 | ||
ab985a77 | 1208 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
1209 | memset(&ecmd, 0, sizeof ecmd); |
1210 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1211 | "ETHTOOL_GLINK"); | |
1212 | if (!error) { | |
782e6111 EJ |
1213 | struct ethtool_value eval; |
1214 | ||
1215 | memcpy(&eval, &ecmd, sizeof eval); | |
1216 | *miimon = !!eval.data; | |
63331829 EJ |
1217 | } else { |
1218 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1219 | } | |
1220 | } | |
1221 | ||
1222 | return error; | |
1223 | } | |
1224 | ||
1670c579 EJ |
1225 | static int |
1226 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
1227 | long long int interval) | |
1228 | { | |
1229 | struct netdev_dev_linux *netdev_dev; | |
1230 | ||
1231 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1232 | ||
1233 | interval = interval > 0 ? MAX(interval, 100) : 0; | |
1234 | if (netdev_dev->miimon_interval != interval) { | |
1235 | netdev_dev->miimon_interval = interval; | |
1236 | timer_set_expired(&netdev_dev->miimon_timer); | |
1237 | } | |
1238 | ||
1239 | return 0; | |
1240 | } | |
1241 | ||
1242 | static void | |
1243 | netdev_linux_miimon_run(void) | |
1244 | { | |
1245 | struct shash device_shash; | |
1246 | struct shash_node *node; | |
1247 | ||
1248 | shash_init(&device_shash); | |
1249 | netdev_dev_get_devices(&netdev_linux_class, &device_shash); | |
1250 | SHASH_FOR_EACH (node, &device_shash) { | |
1251 | struct netdev_dev_linux *dev = node->data; | |
1252 | bool miimon; | |
1253 | ||
1254 | if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { | |
1255 | continue; | |
1256 | } | |
1257 | ||
1258 | netdev_linux_get_miimon(dev->netdev_dev.name, &miimon); | |
1259 | if (miimon != dev->miimon) { | |
1670c579 | 1260 | dev->miimon = miimon; |
4f925bd3 | 1261 | netdev_dev_linux_changed(dev, dev->ifi_flags, 0); |
1670c579 EJ |
1262 | } |
1263 | ||
1264 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); | |
1265 | } | |
1266 | ||
1267 | shash_destroy(&device_shash); | |
1268 | } | |
1269 | ||
1270 | static void | |
1271 | netdev_linux_miimon_wait(void) | |
1272 | { | |
1273 | struct shash device_shash; | |
1274 | struct shash_node *node; | |
1275 | ||
1276 | shash_init(&device_shash); | |
1277 | netdev_dev_get_devices(&netdev_linux_class, &device_shash); | |
1278 | SHASH_FOR_EACH (node, &device_shash) { | |
1279 | struct netdev_dev_linux *dev = node->data; | |
1280 | ||
1281 | if (dev->miimon_interval > 0) { | |
1282 | timer_wait(&dev->miimon_timer); | |
1283 | } | |
1284 | } | |
1285 | shash_destroy(&device_shash); | |
1286 | } | |
1287 | ||
8b61709d BP |
1288 | /* Check whether we can we use RTM_GETLINK to get network device statistics. |
1289 | * In pre-2.6.19 kernels, this was only available if wireless extensions were | |
1290 | * enabled. */ | |
1291 | static bool | |
1292 | check_for_working_netlink_stats(void) | |
1293 | { | |
1294 | /* Decide on the netdev_get_stats() implementation to use. Netlink is | |
1295 | * preferable, so if that works, we'll use it. */ | |
1296 | int ifindex = do_get_ifindex("lo"); | |
1297 | if (ifindex < 0) { | |
1298 | VLOG_WARN("failed to get ifindex for lo, " | |
1299 | "obtaining netdev stats from proc"); | |
1300 | return false; | |
1301 | } else { | |
1302 | struct netdev_stats stats; | |
1303 | int error = get_stats_via_netlink(ifindex, &stats); | |
1304 | if (!error) { | |
1305 | VLOG_DBG("obtaining netdev stats via rtnetlink"); | |
1306 | return true; | |
1307 | } else { | |
1308 | VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " | |
1309 | "via proc (you are probably running a pre-2.6.19 " | |
1310 | "kernel)", strerror(error)); | |
1311 | return false; | |
1312 | } | |
1313 | } | |
1314 | } | |
1315 | ||
92df599c JG |
1316 | static void |
1317 | swap_uint64(uint64_t *a, uint64_t *b) | |
1318 | { | |
1de0e8ae BP |
1319 | uint64_t tmp = *a; |
1320 | *a = *b; | |
1321 | *b = tmp; | |
92df599c JG |
1322 | } |
1323 | ||
f613a0d7 PS |
1324 | static void |
1325 | get_stats_via_vport(const struct netdev *netdev_, | |
1326 | struct netdev_stats *stats) | |
8b61709d | 1327 | { |
149f577a JG |
1328 | struct netdev_dev_linux *netdev_dev = |
1329 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d | 1330 | |
bba1e6f3 PS |
1331 | if (!netdev_dev->vport_stats_error || |
1332 | !(netdev_dev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 1333 | int error; |
7fbef77a JG |
1334 | |
1335 | error = netdev_vport_get_stats(netdev_, stats); | |
f613a0d7 | 1336 | if (error) { |
a57a8488 BP |
1337 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
1338 | "(%s)", netdev_get_name(netdev_), strerror(error)); | |
f613a0d7 | 1339 | } |
bba1e6f3 PS |
1340 | netdev_dev->vport_stats_error = error; |
1341 | netdev_dev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 1342 | } |
f613a0d7 | 1343 | } |
8b61709d | 1344 | |
f613a0d7 PS |
1345 | static int |
1346 | netdev_linux_sys_get_stats(const struct netdev *netdev_, | |
1347 | struct netdev_stats *stats) | |
1348 | { | |
1349 | static int use_netlink_stats = -1; | |
1350 | int error; | |
1351 | ||
1352 | if (use_netlink_stats < 0) { | |
1353 | use_netlink_stats = check_for_working_netlink_stats(); | |
1354 | } | |
1355 | ||
1356 | if (use_netlink_stats) { | |
1357 | int ifindex; | |
1358 | ||
1359 | error = get_ifindex(netdev_, &ifindex); | |
1360 | if (!error) { | |
1361 | error = get_stats_via_netlink(ifindex, stats); | |
7fbef77a | 1362 | } |
f613a0d7 PS |
1363 | } else { |
1364 | error = get_stats_via_proc(netdev_get_name(netdev_), stats); | |
1365 | } | |
7fbef77a | 1366 | |
f613a0d7 PS |
1367 | if (error) { |
1368 | VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d", | |
1369 | netdev_get_name(netdev_), error); | |
1370 | } | |
1371 | return error; | |
1372 | ||
1373 | } | |
1374 | ||
1375 | /* Retrieves current device stats for 'netdev-linux'. */ | |
1376 | static int | |
1377 | netdev_linux_get_stats(const struct netdev *netdev_, | |
1378 | struct netdev_stats *stats) | |
1379 | { | |
1380 | struct netdev_dev_linux *netdev_dev = | |
1381 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1382 | struct netdev_stats dev_stats; | |
1383 | int error; | |
1384 | ||
1385 | get_stats_via_vport(netdev_, stats); | |
1386 | ||
1387 | error = netdev_linux_sys_get_stats(netdev_, &dev_stats); | |
1388 | ||
1389 | if (error) { | |
bba1e6f3 | 1390 | if (netdev_dev->vport_stats_error) { |
f613a0d7 | 1391 | return error; |
7fbef77a | 1392 | } else { |
f613a0d7 PS |
1393 | return 0; |
1394 | } | |
1395 | } | |
1396 | ||
bba1e6f3 | 1397 | if (netdev_dev->vport_stats_error) { |
f613a0d7 PS |
1398 | /* stats not available from OVS then use ioctl stats. */ |
1399 | *stats = dev_stats; | |
1400 | } else { | |
1401 | stats->rx_errors += dev_stats.rx_errors; | |
1402 | stats->tx_errors += dev_stats.tx_errors; | |
1403 | stats->rx_dropped += dev_stats.rx_dropped; | |
1404 | stats->tx_dropped += dev_stats.tx_dropped; | |
1405 | stats->multicast += dev_stats.multicast; | |
1406 | stats->collisions += dev_stats.collisions; | |
1407 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
1408 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
1409 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
1410 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
1411 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
1412 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
1413 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
1414 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
1415 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
1416 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
1417 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
1418 | } | |
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
1423 | * netdev-internal. */ | |
1424 | static int | |
bba1e6f3 | 1425 | netdev_tap_get_stats(const struct netdev *netdev_, |
f613a0d7 PS |
1426 | struct netdev_stats *stats) |
1427 | { | |
1428 | struct netdev_dev_linux *netdev_dev = | |
1429 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1430 | struct netdev_stats dev_stats; | |
1431 | int error; | |
1432 | ||
1433 | get_stats_via_vport(netdev_, stats); | |
1434 | ||
1435 | error = netdev_linux_sys_get_stats(netdev_, &dev_stats); | |
1436 | if (error) { | |
bba1e6f3 | 1437 | if (netdev_dev->vport_stats_error) { |
f613a0d7 PS |
1438 | return error; |
1439 | } else { | |
1440 | return 0; | |
8b61709d | 1441 | } |
8b61709d | 1442 | } |
fe6b0e03 JG |
1443 | |
1444 | /* If this port is an internal port then the transmit and receive stats | |
1445 | * will appear to be swapped relative to the other ports since we are the | |
1446 | * one sending the data, not a remote computer. For consistency, we swap | |
7fbef77a JG |
1447 | * them back here. This does not apply if we are getting stats from the |
1448 | * vport layer because it always tracks stats from the perspective of the | |
1449 | * switch. */ | |
bba1e6f3 | 1450 | if (netdev_dev->vport_stats_error) { |
f613a0d7 | 1451 | *stats = dev_stats; |
92df599c JG |
1452 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1453 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1454 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1455 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1456 | stats->rx_length_errors = 0; |
1457 | stats->rx_over_errors = 0; | |
1458 | stats->rx_crc_errors = 0; | |
1459 | stats->rx_frame_errors = 0; | |
1460 | stats->rx_fifo_errors = 0; | |
1461 | stats->rx_missed_errors = 0; | |
1462 | stats->tx_aborted_errors = 0; | |
1463 | stats->tx_carrier_errors = 0; | |
1464 | stats->tx_fifo_errors = 0; | |
1465 | stats->tx_heartbeat_errors = 0; | |
1466 | stats->tx_window_errors = 0; | |
f613a0d7 PS |
1467 | } else { |
1468 | stats->rx_dropped += dev_stats.tx_dropped; | |
1469 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 1470 | |
f613a0d7 PS |
1471 | stats->rx_errors += dev_stats.tx_errors; |
1472 | stats->tx_errors += dev_stats.rx_errors; | |
1473 | ||
1474 | stats->multicast += dev_stats.multicast; | |
1475 | stats->collisions += dev_stats.collisions; | |
1476 | } | |
1477 | return 0; | |
8b61709d BP |
1478 | } |
1479 | ||
bba1e6f3 PS |
1480 | static int |
1481 | netdev_internal_get_stats(const struct netdev *netdev_, | |
1482 | struct netdev_stats *stats) | |
1483 | { | |
1484 | struct netdev_dev_linux *netdev_dev = | |
1485 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1486 | ||
1487 | get_stats_via_vport(netdev_, stats); | |
1488 | return netdev_dev->vport_stats_error; | |
1489 | } | |
1490 | ||
51f87458 PS |
1491 | static void |
1492 | netdev_linux_read_features(struct netdev_dev_linux *netdev_dev) | |
8b61709d BP |
1493 | { |
1494 | struct ethtool_cmd ecmd; | |
6c038611 | 1495 | uint32_t speed; |
8b61709d BP |
1496 | int error; |
1497 | ||
51f87458 PS |
1498 | if (netdev_dev->cache_valid & VALID_FEATURES) { |
1499 | return; | |
1500 | } | |
1501 | ||
ab985a77 | 1502 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1503 | memset(&ecmd, 0, sizeof ecmd); |
51f87458 | 1504 | error = netdev_linux_do_ethtool(netdev_dev->netdev_dev.name, &ecmd, |
8b61709d BP |
1505 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1506 | if (error) { | |
51f87458 | 1507 | goto out; |
8b61709d BP |
1508 | } |
1509 | ||
1510 | /* Supported features. */ | |
51f87458 | 1511 | netdev_dev->supported = 0; |
8b61709d | 1512 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
51f87458 | 1513 | netdev_dev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
1514 | } |
1515 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
51f87458 | 1516 | netdev_dev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
1517 | } |
1518 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
51f87458 | 1519 | netdev_dev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
1520 | } |
1521 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
51f87458 | 1522 | netdev_dev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
1523 | } |
1524 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
51f87458 | 1525 | netdev_dev->supported |= NETDEV_F_1GB_HD; |
8b61709d BP |
1526 | } |
1527 | if (ecmd.supported & SUPPORTED_1000baseT_Full) { | |
51f87458 | 1528 | netdev_dev->supported |= NETDEV_F_1GB_FD; |
8b61709d BP |
1529 | } |
1530 | if (ecmd.supported & SUPPORTED_10000baseT_Full) { | |
51f87458 | 1531 | netdev_dev->supported |= NETDEV_F_10GB_FD; |
8b61709d BP |
1532 | } |
1533 | if (ecmd.supported & SUPPORTED_TP) { | |
51f87458 | 1534 | netdev_dev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
1535 | } |
1536 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
51f87458 | 1537 | netdev_dev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
1538 | } |
1539 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
51f87458 | 1540 | netdev_dev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
1541 | } |
1542 | if (ecmd.supported & SUPPORTED_Pause) { | |
51f87458 | 1543 | netdev_dev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
1544 | } |
1545 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
51f87458 | 1546 | netdev_dev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1547 | } |
1548 | ||
1549 | /* Advertised features. */ | |
51f87458 | 1550 | netdev_dev->advertised = 0; |
8b61709d | 1551 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
51f87458 | 1552 | netdev_dev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
1553 | } |
1554 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
51f87458 | 1555 | netdev_dev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
1556 | } |
1557 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
51f87458 | 1558 | netdev_dev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
1559 | } |
1560 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
51f87458 | 1561 | netdev_dev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
1562 | } |
1563 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
51f87458 | 1564 | netdev_dev->advertised |= NETDEV_F_1GB_HD; |
8b61709d BP |
1565 | } |
1566 | if (ecmd.advertising & ADVERTISED_1000baseT_Full) { | |
51f87458 | 1567 | netdev_dev->advertised |= NETDEV_F_1GB_FD; |
8b61709d BP |
1568 | } |
1569 | if (ecmd.advertising & ADVERTISED_10000baseT_Full) { | |
51f87458 | 1570 | netdev_dev->advertised |= NETDEV_F_10GB_FD; |
8b61709d BP |
1571 | } |
1572 | if (ecmd.advertising & ADVERTISED_TP) { | |
51f87458 | 1573 | netdev_dev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
1574 | } |
1575 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
51f87458 | 1576 | netdev_dev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
1577 | } |
1578 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
51f87458 | 1579 | netdev_dev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
1580 | } |
1581 | if (ecmd.advertising & ADVERTISED_Pause) { | |
51f87458 | 1582 | netdev_dev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
1583 | } |
1584 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
51f87458 | 1585 | netdev_dev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1586 | } |
1587 | ||
1588 | /* Current settings. */ | |
2a529ead | 1589 | speed = ecmd.speed; |
6c038611 | 1590 | if (speed == SPEED_10) { |
51f87458 | 1591 | netdev_dev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 1592 | } else if (speed == SPEED_100) { |
51f87458 | 1593 | netdev_dev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 1594 | } else if (speed == SPEED_1000) { |
51f87458 | 1595 | netdev_dev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 1596 | } else if (speed == SPEED_10000) { |
51f87458 | 1597 | netdev_dev->current = NETDEV_F_10GB_FD; |
6c038611 | 1598 | } else if (speed == 40000) { |
51f87458 | 1599 | netdev_dev->current = NETDEV_F_40GB_FD; |
6c038611 | 1600 | } else if (speed == 100000) { |
51f87458 | 1601 | netdev_dev->current = NETDEV_F_100GB_FD; |
6c038611 | 1602 | } else if (speed == 1000000) { |
51f87458 | 1603 | netdev_dev->current = NETDEV_F_1TB_FD; |
8b61709d | 1604 | } else { |
51f87458 | 1605 | netdev_dev->current = 0; |
8b61709d BP |
1606 | } |
1607 | ||
1608 | if (ecmd.port == PORT_TP) { | |
51f87458 | 1609 | netdev_dev->current |= NETDEV_F_COPPER; |
8b61709d | 1610 | } else if (ecmd.port == PORT_FIBRE) { |
51f87458 | 1611 | netdev_dev->current |= NETDEV_F_FIBER; |
8b61709d BP |
1612 | } |
1613 | ||
1614 | if (ecmd.autoneg) { | |
51f87458 | 1615 | netdev_dev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
1616 | } |
1617 | ||
1618 | /* Peer advertisements. */ | |
51f87458 | 1619 | netdev_dev->peer = 0; /* XXX */ |
8b61709d | 1620 | |
51f87458 PS |
1621 | out: |
1622 | netdev_dev->cache_valid |= VALID_FEATURES; | |
1623 | netdev_dev->get_features_error = error; | |
1624 | } | |
1625 | ||
1626 | /* Stores the features supported by 'netdev' into each of '*current', | |
1627 | * '*advertised', '*supported', and '*peer' that are non-null. Each value is a | |
1628 | * bitmap of NETDEV_* bits. Returns 0 if successful, otherwise a positive | |
1629 | * errno value. */ | |
1630 | static int | |
1631 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
1632 | enum netdev_features *current, |
1633 | enum netdev_features *advertised, | |
1634 | enum netdev_features *supported, | |
1635 | enum netdev_features *peer) | |
51f87458 PS |
1636 | { |
1637 | struct netdev_dev_linux *netdev_dev = | |
1638 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
1639 | ||
1640 | netdev_linux_read_features(netdev_dev); | |
1641 | ||
1642 | if (!netdev_dev->get_features_error) { | |
1643 | *current = netdev_dev->current; | |
1644 | *advertised = netdev_dev->advertised; | |
1645 | *supported = netdev_dev->supported; | |
1646 | *peer = netdev_dev->peer; | |
1647 | } | |
1648 | return netdev_dev->get_features_error; | |
8b61709d BP |
1649 | } |
1650 | ||
1651 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
1652 | static int | |
6c038611 BP |
1653 | netdev_linux_set_advertisements(struct netdev *netdev, |
1654 | enum netdev_features advertise) | |
8b61709d BP |
1655 | { |
1656 | struct ethtool_cmd ecmd; | |
1657 | int error; | |
1658 | ||
ab985a77 | 1659 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1660 | memset(&ecmd, 0, sizeof ecmd); |
0b0544d7 | 1661 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1662 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1663 | if (error) { | |
1664 | return error; | |
1665 | } | |
1666 | ||
1667 | ecmd.advertising = 0; | |
6c038611 | 1668 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
1669 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
1670 | } | |
6c038611 | 1671 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
1672 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
1673 | } | |
6c038611 | 1674 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
1675 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
1676 | } | |
6c038611 | 1677 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
1678 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
1679 | } | |
6c038611 | 1680 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
1681 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
1682 | } | |
6c038611 | 1683 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
1684 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
1685 | } | |
6c038611 | 1686 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
1687 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
1688 | } | |
6c038611 | 1689 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
1690 | ecmd.advertising |= ADVERTISED_TP; |
1691 | } | |
6c038611 | 1692 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
1693 | ecmd.advertising |= ADVERTISED_FIBRE; |
1694 | } | |
6c038611 | 1695 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
1696 | ecmd.advertising |= ADVERTISED_Autoneg; |
1697 | } | |
6c038611 | 1698 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
1699 | ecmd.advertising |= ADVERTISED_Pause; |
1700 | } | |
6c038611 | 1701 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
1702 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
1703 | } | |
ab985a77 | 1704 | COVERAGE_INC(netdev_set_ethtool); |
0b0544d7 | 1705 | return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1706 | ETHTOOL_SSET, "ETHTOOL_SSET"); |
1707 | } | |
1708 | ||
f8500004 JP |
1709 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
1710 | * successful, otherwise a positive errno value. */ | |
8b61709d BP |
1711 | static int |
1712 | netdev_linux_set_policing(struct netdev *netdev, | |
1713 | uint32_t kbits_rate, uint32_t kbits_burst) | |
1714 | { | |
80a86fbe BP |
1715 | struct netdev_dev_linux *netdev_dev = |
1716 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
8b61709d | 1717 | const char *netdev_name = netdev_get_name(netdev); |
f8500004 | 1718 | int error; |
8b61709d | 1719 | |
8e460221 | 1720 | |
80a86fbe BP |
1721 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
1722 | : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ | |
1723 | : kbits_burst); /* Stick with user-specified value. */ | |
1724 | ||
c9f71668 PS |
1725 | if (netdev_dev->cache_valid & VALID_POLICING) { |
1726 | if (netdev_dev->netdev_policing_error) { | |
1727 | return netdev_dev->netdev_policing_error; | |
1728 | } | |
1729 | ||
1730 | if (netdev_dev->kbits_rate == kbits_rate && | |
1731 | netdev_dev->kbits_burst == kbits_burst) { | |
1732 | /* Assume that settings haven't changed since we last set them. */ | |
1733 | return 0; | |
1734 | } | |
1735 | netdev_dev->cache_valid &= ~VALID_POLICING; | |
80a86fbe BP |
1736 | } |
1737 | ||
ac8c3412 | 1738 | COVERAGE_INC(netdev_set_policing); |
f8500004 JP |
1739 | /* Remove any existing ingress qdisc. */ |
1740 | error = tc_add_del_ingress_qdisc(netdev, false); | |
1741 | if (error) { | |
1742 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
1743 | netdev_name, strerror(error)); | |
c9f71668 | 1744 | goto out; |
f8500004 JP |
1745 | } |
1746 | ||
8b61709d | 1747 | if (kbits_rate) { |
f8500004 JP |
1748 | error = tc_add_del_ingress_qdisc(netdev, true); |
1749 | if (error) { | |
1750 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
1751 | netdev_name, strerror(error)); | |
c9f71668 | 1752 | goto out; |
8b61709d BP |
1753 | } |
1754 | ||
f8500004 JP |
1755 | error = tc_add_policer(netdev, kbits_rate, kbits_burst); |
1756 | if (error){ | |
1757 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
1758 | netdev_name, strerror(error)); | |
c9f71668 | 1759 | goto out; |
8b61709d | 1760 | } |
8b61709d BP |
1761 | } |
1762 | ||
f8500004 JP |
1763 | netdev_dev->kbits_rate = kbits_rate; |
1764 | netdev_dev->kbits_burst = kbits_burst; | |
f8500004 | 1765 | |
c9f71668 PS |
1766 | out: |
1767 | if (!error || error == ENODEV) { | |
1768 | netdev_dev->netdev_policing_error = error; | |
1769 | netdev_dev->cache_valid |= VALID_POLICING; | |
1770 | } | |
1771 | return error; | |
8b61709d BP |
1772 | } |
1773 | ||
c1c9c9c4 BP |
1774 | static int |
1775 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 1776 | struct sset *types) |
c1c9c9c4 BP |
1777 | { |
1778 | const struct tc_ops **opsp; | |
1779 | ||
1780 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1781 | const struct tc_ops *ops = *opsp; | |
1782 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 1783 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
1784 | } |
1785 | } | |
1786 | return 0; | |
1787 | } | |
1788 | ||
1789 | static const struct tc_ops * | |
1790 | tc_lookup_ovs_name(const char *name) | |
1791 | { | |
1792 | const struct tc_ops **opsp; | |
1793 | ||
1794 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1795 | const struct tc_ops *ops = *opsp; | |
1796 | if (!strcmp(name, ops->ovs_name)) { | |
1797 | return ops; | |
1798 | } | |
1799 | } | |
1800 | return NULL; | |
1801 | } | |
1802 | ||
1803 | static const struct tc_ops * | |
1804 | tc_lookup_linux_name(const char *name) | |
1805 | { | |
1806 | const struct tc_ops **opsp; | |
1807 | ||
1808 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1809 | const struct tc_ops *ops = *opsp; | |
1810 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
1811 | return ops; | |
1812 | } | |
1813 | } | |
1814 | return NULL; | |
1815 | } | |
1816 | ||
93b13be8 BP |
1817 | static struct tc_queue * |
1818 | tc_find_queue__(const struct netdev *netdev, unsigned int queue_id, | |
1819 | size_t hash) | |
1820 | { | |
1821 | struct netdev_dev_linux *netdev_dev = | |
1822 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1823 | struct tc_queue *queue; | |
1824 | ||
4e8e4213 | 1825 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev_dev->tc->queues) { |
93b13be8 BP |
1826 | if (queue->queue_id == queue_id) { |
1827 | return queue; | |
1828 | } | |
1829 | } | |
1830 | return NULL; | |
1831 | } | |
1832 | ||
1833 | static struct tc_queue * | |
1834 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
1835 | { | |
1836 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
1837 | } | |
1838 | ||
c1c9c9c4 BP |
1839 | static int |
1840 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
1841 | const char *type, | |
1842 | struct netdev_qos_capabilities *caps) | |
1843 | { | |
1844 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
1845 | if (!ops) { | |
1846 | return EOPNOTSUPP; | |
1847 | } | |
1848 | caps->n_queues = ops->n_queues; | |
1849 | return 0; | |
1850 | } | |
1851 | ||
1852 | static int | |
1853 | netdev_linux_get_qos(const struct netdev *netdev, | |
79f1cbe9 | 1854 | const char **typep, struct smap *details) |
c1c9c9c4 BP |
1855 | { |
1856 | struct netdev_dev_linux *netdev_dev = | |
1857 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1858 | int error; | |
1859 | ||
1860 | error = tc_query_qdisc(netdev); | |
1861 | if (error) { | |
1862 | return error; | |
1863 | } | |
1864 | ||
1865 | *typep = netdev_dev->tc->ops->ovs_name; | |
1866 | return (netdev_dev->tc->ops->qdisc_get | |
1867 | ? netdev_dev->tc->ops->qdisc_get(netdev, details) | |
1868 | : 0); | |
1869 | } | |
1870 | ||
1871 | static int | |
1872 | netdev_linux_set_qos(struct netdev *netdev, | |
79f1cbe9 | 1873 | const char *type, const struct smap *details) |
c1c9c9c4 BP |
1874 | { |
1875 | struct netdev_dev_linux *netdev_dev = | |
1876 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1877 | const struct tc_ops *new_ops; | |
1878 | int error; | |
1879 | ||
1880 | new_ops = tc_lookup_ovs_name(type); | |
1881 | if (!new_ops || !new_ops->tc_install) { | |
1882 | return EOPNOTSUPP; | |
1883 | } | |
1884 | ||
1885 | error = tc_query_qdisc(netdev); | |
1886 | if (error) { | |
1887 | return error; | |
1888 | } | |
1889 | ||
1890 | if (new_ops == netdev_dev->tc->ops) { | |
1891 | return new_ops->qdisc_set ? new_ops->qdisc_set(netdev, details) : 0; | |
1892 | } else { | |
1893 | /* Delete existing qdisc. */ | |
1894 | error = tc_del_qdisc(netdev); | |
1895 | if (error) { | |
1896 | return error; | |
1897 | } | |
1898 | assert(netdev_dev->tc == NULL); | |
1899 | ||
1900 | /* Install new qdisc. */ | |
1901 | error = new_ops->tc_install(netdev, details); | |
1902 | assert((error == 0) == (netdev_dev->tc != NULL)); | |
1903 | ||
1904 | return error; | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | static int | |
1909 | netdev_linux_get_queue(const struct netdev *netdev, | |
79f1cbe9 | 1910 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 BP |
1911 | { |
1912 | struct netdev_dev_linux *netdev_dev = | |
1913 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1914 | int error; | |
1915 | ||
1916 | error = tc_query_qdisc(netdev); | |
1917 | if (error) { | |
1918 | return error; | |
93b13be8 BP |
1919 | } else { |
1920 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1921 | return (queue | |
1922 | ? netdev_dev->tc->ops->class_get(netdev, queue, details) | |
1923 | : ENOENT); | |
c1c9c9c4 | 1924 | } |
c1c9c9c4 BP |
1925 | } |
1926 | ||
1927 | static int | |
1928 | netdev_linux_set_queue(struct netdev *netdev, | |
79f1cbe9 | 1929 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 BP |
1930 | { |
1931 | struct netdev_dev_linux *netdev_dev = | |
1932 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1933 | int error; | |
1934 | ||
1935 | error = tc_query_qdisc(netdev); | |
1936 | if (error) { | |
1937 | return error; | |
1938 | } else if (queue_id >= netdev_dev->tc->ops->n_queues | |
1939 | || !netdev_dev->tc->ops->class_set) { | |
1940 | return EINVAL; | |
1941 | } | |
1942 | ||
1943 | return netdev_dev->tc->ops->class_set(netdev, queue_id, details); | |
1944 | } | |
1945 | ||
1946 | static int | |
1947 | netdev_linux_delete_queue(struct netdev *netdev, unsigned int queue_id) | |
1948 | { | |
1949 | struct netdev_dev_linux *netdev_dev = | |
1950 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1951 | int error; | |
1952 | ||
1953 | error = tc_query_qdisc(netdev); | |
1954 | if (error) { | |
1955 | return error; | |
1956 | } else if (!netdev_dev->tc->ops->class_delete) { | |
1957 | return EINVAL; | |
93b13be8 BP |
1958 | } else { |
1959 | struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1960 | return (queue | |
1961 | ? netdev_dev->tc->ops->class_delete(netdev, queue) | |
1962 | : ENOENT); | |
c1c9c9c4 | 1963 | } |
c1c9c9c4 BP |
1964 | } |
1965 | ||
1966 | static int | |
1967 | netdev_linux_get_queue_stats(const struct netdev *netdev, | |
1968 | unsigned int queue_id, | |
1969 | struct netdev_queue_stats *stats) | |
1970 | { | |
1971 | struct netdev_dev_linux *netdev_dev = | |
1972 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
1973 | int error; | |
1974 | ||
1975 | error = tc_query_qdisc(netdev); | |
1976 | if (error) { | |
1977 | return error; | |
c1c9c9c4 BP |
1978 | } else if (!netdev_dev->tc->ops->class_get_stats) { |
1979 | return EOPNOTSUPP; | |
93b13be8 BP |
1980 | } else { |
1981 | const struct tc_queue *queue = tc_find_queue(netdev, queue_id); | |
1982 | return (queue | |
1983 | ? netdev_dev->tc->ops->class_get_stats(netdev, queue, stats) | |
1984 | : ENOENT); | |
c1c9c9c4 | 1985 | } |
c1c9c9c4 BP |
1986 | } |
1987 | ||
23a98ffe | 1988 | static bool |
c1c9c9c4 BP |
1989 | start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) |
1990 | { | |
1991 | struct ofpbuf request; | |
1992 | struct tcmsg *tcmsg; | |
1993 | ||
1994 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); | |
23a98ffe BP |
1995 | if (!tcmsg) { |
1996 | return false; | |
1997 | } | |
3c4de644 | 1998 | tcmsg->tcm_parent = 0; |
c1c9c9c4 BP |
1999 | nl_dump_start(dump, rtnl_sock, &request); |
2000 | ofpbuf_uninit(&request); | |
23a98ffe | 2001 | return true; |
c1c9c9c4 BP |
2002 | } |
2003 | ||
2004 | static int | |
2005 | netdev_linux_dump_queues(const struct netdev *netdev, | |
2006 | netdev_dump_queues_cb *cb, void *aux) | |
2007 | { | |
2008 | struct netdev_dev_linux *netdev_dev = | |
2009 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
f486e840 | 2010 | struct tc_queue *queue, *next_queue; |
79f1cbe9 | 2011 | struct smap details; |
c1c9c9c4 | 2012 | int last_error; |
c1c9c9c4 BP |
2013 | int error; |
2014 | ||
2015 | error = tc_query_qdisc(netdev); | |
2016 | if (error) { | |
2017 | return error; | |
2018 | } else if (!netdev_dev->tc->ops->class_get) { | |
2019 | return EOPNOTSUPP; | |
2020 | } | |
2021 | ||
2022 | last_error = 0; | |
79f1cbe9 | 2023 | smap_init(&details); |
f486e840 BP |
2024 | HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, |
2025 | &netdev_dev->tc->queues) { | |
79f1cbe9 | 2026 | smap_clear(&details); |
c1c9c9c4 | 2027 | |
93b13be8 | 2028 | error = netdev_dev->tc->ops->class_get(netdev, queue, &details); |
c1c9c9c4 | 2029 | if (!error) { |
93b13be8 | 2030 | (*cb)(queue->queue_id, &details, aux); |
c1c9c9c4 BP |
2031 | } else { |
2032 | last_error = error; | |
2033 | } | |
2034 | } | |
79f1cbe9 | 2035 | smap_destroy(&details); |
c1c9c9c4 BP |
2036 | |
2037 | return last_error; | |
2038 | } | |
2039 | ||
2040 | static int | |
2041 | netdev_linux_dump_queue_stats(const struct netdev *netdev, | |
2042 | netdev_dump_queue_stats_cb *cb, void *aux) | |
2043 | { | |
2044 | struct netdev_dev_linux *netdev_dev = | |
2045 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2046 | struct nl_dump dump; | |
2047 | struct ofpbuf msg; | |
2048 | int last_error; | |
2049 | int error; | |
2050 | ||
2051 | error = tc_query_qdisc(netdev); | |
2052 | if (error) { | |
2053 | return error; | |
2054 | } else if (!netdev_dev->tc->ops->class_dump_stats) { | |
2055 | return EOPNOTSUPP; | |
2056 | } | |
2057 | ||
2058 | last_error = 0; | |
23a98ffe BP |
2059 | if (!start_queue_dump(netdev, &dump)) { |
2060 | return ENODEV; | |
2061 | } | |
c1c9c9c4 BP |
2062 | while (nl_dump_next(&dump, &msg)) { |
2063 | error = netdev_dev->tc->ops->class_dump_stats(netdev, &msg, cb, aux); | |
2064 | if (error) { | |
2065 | last_error = error; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | error = nl_dump_done(&dump); | |
2070 | return error ? error : last_error; | |
2071 | } | |
2072 | ||
8b61709d | 2073 | static int |
f1acd62b BP |
2074 | netdev_linux_get_in4(const struct netdev *netdev_, |
2075 | struct in_addr *address, struct in_addr *netmask) | |
8b61709d | 2076 | { |
149f577a JG |
2077 | struct netdev_dev_linux *netdev_dev = |
2078 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
2079 | ||
2080 | if (!(netdev_dev->cache_valid & VALID_IN4)) { | |
8b61709d BP |
2081 | int error; |
2082 | ||
149f577a | 2083 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->address, |
8b61709d BP |
2084 | SIOCGIFADDR, "SIOCGIFADDR"); |
2085 | if (error) { | |
2086 | return error; | |
2087 | } | |
2088 | ||
149f577a | 2089 | error = netdev_linux_get_ipv4(netdev_, &netdev_dev->netmask, |
f1acd62b BP |
2090 | SIOCGIFNETMASK, "SIOCGIFNETMASK"); |
2091 | if (error) { | |
2092 | return error; | |
2093 | } | |
2094 | ||
149f577a | 2095 | netdev_dev->cache_valid |= VALID_IN4; |
8b61709d | 2096 | } |
149f577a JG |
2097 | *address = netdev_dev->address; |
2098 | *netmask = netdev_dev->netmask; | |
f1acd62b | 2099 | return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; |
8b61709d BP |
2100 | } |
2101 | ||
8b61709d | 2102 | static int |
f1acd62b BP |
2103 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
2104 | struct in_addr netmask) | |
8b61709d | 2105 | { |
149f577a JG |
2106 | struct netdev_dev_linux *netdev_dev = |
2107 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
8b61709d BP |
2108 | int error; |
2109 | ||
f1acd62b | 2110 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 2111 | if (!error) { |
149f577a JG |
2112 | netdev_dev->cache_valid |= VALID_IN4; |
2113 | netdev_dev->address = address; | |
2114 | netdev_dev->netmask = netmask; | |
f1acd62b | 2115 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 2116 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 2117 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
2118 | } |
2119 | } | |
2120 | return error; | |
2121 | } | |
2122 | ||
2123 | static bool | |
2124 | parse_if_inet6_line(const char *line, | |
2125 | struct in6_addr *in6, char ifname[16 + 1]) | |
2126 | { | |
2127 | uint8_t *s6 = in6->s6_addr; | |
2128 | #define X8 "%2"SCNx8 | |
2129 | return sscanf(line, | |
2130 | " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 | |
2131 | "%*x %*x %*x %*x %16s\n", | |
2132 | &s6[0], &s6[1], &s6[2], &s6[3], | |
2133 | &s6[4], &s6[5], &s6[6], &s6[7], | |
2134 | &s6[8], &s6[9], &s6[10], &s6[11], | |
2135 | &s6[12], &s6[13], &s6[14], &s6[15], | |
2136 | ifname) == 17; | |
2137 | } | |
2138 | ||
2139 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if | |
2140 | * 'in6' is non-null) and returns true. Otherwise, returns false. */ | |
2141 | static int | |
2142 | netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) | |
2143 | { | |
149f577a JG |
2144 | struct netdev_dev_linux *netdev_dev = |
2145 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
2146 | if (!(netdev_dev->cache_valid & VALID_IN6)) { | |
8b61709d BP |
2147 | FILE *file; |
2148 | char line[128]; | |
2149 | ||
149f577a | 2150 | netdev_dev->in6 = in6addr_any; |
8b61709d BP |
2151 | |
2152 | file = fopen("/proc/net/if_inet6", "r"); | |
2153 | if (file != NULL) { | |
2154 | const char *name = netdev_get_name(netdev_); | |
2155 | while (fgets(line, sizeof line, file)) { | |
2a022368 | 2156 | struct in6_addr in6_tmp; |
8b61709d | 2157 | char ifname[16 + 1]; |
2a022368 | 2158 | if (parse_if_inet6_line(line, &in6_tmp, ifname) |
8b61709d BP |
2159 | && !strcmp(name, ifname)) |
2160 | { | |
2a022368 | 2161 | netdev_dev->in6 = in6_tmp; |
8b61709d BP |
2162 | break; |
2163 | } | |
2164 | } | |
2165 | fclose(file); | |
2166 | } | |
149f577a | 2167 | netdev_dev->cache_valid |= VALID_IN6; |
8b61709d | 2168 | } |
149f577a | 2169 | *in6 = netdev_dev->in6; |
8b61709d BP |
2170 | return 0; |
2171 | } | |
2172 | ||
2173 | static void | |
2174 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
2175 | { | |
2176 | struct sockaddr_in sin; | |
2177 | memset(&sin, 0, sizeof sin); | |
2178 | sin.sin_family = AF_INET; | |
2179 | sin.sin_addr = addr; | |
2180 | sin.sin_port = 0; | |
2181 | ||
2182 | memset(sa, 0, sizeof *sa); | |
2183 | memcpy(sa, &sin, sizeof sin); | |
2184 | } | |
2185 | ||
2186 | static int | |
2187 | do_set_addr(struct netdev *netdev, | |
2188 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
2189 | { | |
2190 | struct ifreq ifr; | |
71d7c22f | 2191 | ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); |
8b61709d | 2192 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
149f577a JG |
2193 | |
2194 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
2195 | ioctl_name); | |
8b61709d BP |
2196 | } |
2197 | ||
2198 | /* Adds 'router' as a default IP gateway. */ | |
2199 | static int | |
67a4917b | 2200 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
2201 | { |
2202 | struct in_addr any = { INADDR_ANY }; | |
2203 | struct rtentry rt; | |
2204 | int error; | |
2205 | ||
2206 | memset(&rt, 0, sizeof rt); | |
2207 | make_in4_sockaddr(&rt.rt_dst, any); | |
2208 | make_in4_sockaddr(&rt.rt_gateway, router); | |
2209 | make_in4_sockaddr(&rt.rt_genmask, any); | |
2210 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
8b61709d BP |
2211 | error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; |
2212 | if (error) { | |
2213 | VLOG_WARN("ioctl(SIOCADDRT): %s", strerror(error)); | |
2214 | } | |
2215 | return error; | |
2216 | } | |
2217 | ||
f1acd62b BP |
2218 | static int |
2219 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
2220 | char **netdev_name) | |
2221 | { | |
2222 | static const char fn[] = "/proc/net/route"; | |
2223 | FILE *stream; | |
2224 | char line[256]; | |
2225 | int ln; | |
2226 | ||
2227 | *netdev_name = NULL; | |
2228 | stream = fopen(fn, "r"); | |
2229 | if (stream == NULL) { | |
2230 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
2231 | return errno; | |
2232 | } | |
2233 | ||
2234 | ln = 0; | |
2235 | while (fgets(line, sizeof line, stream)) { | |
2236 | if (++ln >= 2) { | |
2237 | char iface[17]; | |
dbba996b | 2238 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
2239 | int refcnt, metric, mtu; |
2240 | unsigned int flags, use, window, irtt; | |
2241 | ||
2242 | if (sscanf(line, | |
2243 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
2244 | " %d %u %u\n", | |
2245 | iface, &dest, &gateway, &flags, &refcnt, | |
2246 | &use, &metric, &mask, &mtu, &window, &irtt) != 11) { | |
2247 | ||
d295e8e9 | 2248 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
2249 | fn, ln, line); |
2250 | continue; | |
2251 | } | |
2252 | if (!(flags & RTF_UP)) { | |
2253 | /* Skip routes that aren't up. */ | |
2254 | continue; | |
2255 | } | |
2256 | ||
2257 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 2258 | * network byte order, so we don't need need any endian |
f1acd62b BP |
2259 | * conversions here. */ |
2260 | if ((dest & mask) == (host->s_addr & mask)) { | |
2261 | if (!gateway) { | |
2262 | /* The host is directly reachable. */ | |
2263 | next_hop->s_addr = 0; | |
2264 | } else { | |
2265 | /* To reach the host, we must go through a gateway. */ | |
2266 | next_hop->s_addr = gateway; | |
2267 | } | |
2268 | *netdev_name = xstrdup(iface); | |
2269 | fclose(stream); | |
2270 | return 0; | |
2271 | } | |
2272 | } | |
2273 | } | |
2274 | ||
2275 | fclose(stream); | |
2276 | return ENXIO; | |
2277 | } | |
2278 | ||
e210037e | 2279 | static int |
79f1cbe9 | 2280 | netdev_linux_get_drv_info(const struct netdev *netdev, struct smap *smap) |
e210037e | 2281 | { |
e210037e | 2282 | int error; |
4f925bd3 PS |
2283 | struct netdev_dev_linux *netdev_dev = |
2284 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
e210037e | 2285 | |
4f925bd3 | 2286 | error = netdev_linux_get_drvinfo(netdev_dev); |
e210037e | 2287 | if (!error) { |
79f1cbe9 EJ |
2288 | smap_add(smap, "driver_name", netdev_dev->drvinfo.driver); |
2289 | smap_add(smap, "driver_version", netdev_dev->drvinfo.version); | |
2290 | smap_add(smap, "firmware_version", netdev_dev->drvinfo.fw_version); | |
e210037e | 2291 | } |
e210037e AE |
2292 | return error; |
2293 | } | |
2294 | ||
4f925bd3 | 2295 | static int |
79f1cbe9 EJ |
2296 | netdev_internal_get_drv_info(const struct netdev *netdev OVS_UNUSED, |
2297 | struct smap *smap) | |
4f925bd3 | 2298 | { |
79f1cbe9 | 2299 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
2300 | return 0; |
2301 | } | |
2302 | ||
8b61709d BP |
2303 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2304 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2305 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2306 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2307 | static int | |
2308 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
dbba996b | 2309 | ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) |
8b61709d BP |
2310 | { |
2311 | struct arpreq r; | |
c100e025 | 2312 | struct sockaddr_in sin; |
8b61709d BP |
2313 | int retval; |
2314 | ||
2315 | memset(&r, 0, sizeof r); | |
f2cc621b | 2316 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2317 | sin.sin_family = AF_INET; |
2318 | sin.sin_addr.s_addr = ip; | |
2319 | sin.sin_port = 0; | |
2320 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2321 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2322 | r.arp_flags = 0; | |
71d7c22f | 2323 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d BP |
2324 | COVERAGE_INC(netdev_arp_lookup); |
2325 | retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; | |
2326 | if (!retval) { | |
2327 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2328 | } else if (retval != ENXIO) { | |
2329 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
149f577a | 2330 | netdev_get_name(netdev), IP_ARGS(&ip), strerror(retval)); |
8b61709d BP |
2331 | } |
2332 | return retval; | |
2333 | } | |
2334 | ||
2335 | static int | |
2336 | nd_to_iff_flags(enum netdev_flags nd) | |
2337 | { | |
2338 | int iff = 0; | |
2339 | if (nd & NETDEV_UP) { | |
2340 | iff |= IFF_UP; | |
2341 | } | |
2342 | if (nd & NETDEV_PROMISC) { | |
2343 | iff |= IFF_PROMISC; | |
2344 | } | |
2345 | return iff; | |
2346 | } | |
2347 | ||
2348 | static int | |
2349 | iff_to_nd_flags(int iff) | |
2350 | { | |
2351 | enum netdev_flags nd = 0; | |
2352 | if (iff & IFF_UP) { | |
2353 | nd |= NETDEV_UP; | |
2354 | } | |
2355 | if (iff & IFF_PROMISC) { | |
2356 | nd |= NETDEV_PROMISC; | |
2357 | } | |
2358 | return nd; | |
2359 | } | |
2360 | ||
2361 | static int | |
2362 | netdev_linux_update_flags(struct netdev *netdev, enum netdev_flags off, | |
2363 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2364 | { | |
c37d4da4 | 2365 | struct netdev_dev_linux *netdev_dev; |
8b61709d | 2366 | int old_flags, new_flags; |
c37d4da4 EJ |
2367 | int error = 0; |
2368 | ||
2369 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2370 | old_flags = netdev_dev->ifi_flags; | |
2371 | *old_flagsp = iff_to_nd_flags(old_flags); | |
2372 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2373 | if (new_flags != old_flags) { | |
2374 | error = set_flags(netdev, new_flags); | |
2375 | get_flags(&netdev_dev->netdev_dev, &netdev_dev->ifi_flags); | |
8b61709d BP |
2376 | } |
2377 | return error; | |
2378 | } | |
2379 | ||
ac4d3bcb EJ |
2380 | static unsigned int |
2381 | netdev_linux_change_seq(const struct netdev *netdev) | |
2382 | { | |
2383 | return netdev_dev_linux_cast(netdev_get_dev(netdev))->change_seq; | |
2384 | } | |
2385 | ||
4f925bd3 | 2386 | #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \ |
51f87458 | 2387 | GET_FEATURES, GET_STATUS) \ |
c3827f61 BP |
2388 | { \ |
2389 | NAME, \ | |
2390 | \ | |
2391 | netdev_linux_init, \ | |
2392 | netdev_linux_run, \ | |
2393 | netdev_linux_wait, \ | |
2394 | \ | |
2395 | CREATE, \ | |
2396 | netdev_linux_destroy, \ | |
de5cdb90 | 2397 | NULL, /* get_config */ \ |
6d9e6eb4 | 2398 | NULL, /* set_config */ \ |
c3827f61 BP |
2399 | \ |
2400 | netdev_linux_open, \ | |
2401 | netdev_linux_close, \ | |
2402 | \ | |
7b6b0ef4 | 2403 | netdev_linux_listen, \ |
c3827f61 BP |
2404 | netdev_linux_recv, \ |
2405 | netdev_linux_recv_wait, \ | |
2406 | netdev_linux_drain, \ | |
2407 | \ | |
2408 | netdev_linux_send, \ | |
2409 | netdev_linux_send_wait, \ | |
2410 | \ | |
2411 | netdev_linux_set_etheraddr, \ | |
2412 | netdev_linux_get_etheraddr, \ | |
2413 | netdev_linux_get_mtu, \ | |
9b020780 | 2414 | netdev_linux_set_mtu, \ |
c3827f61 BP |
2415 | netdev_linux_get_ifindex, \ |
2416 | netdev_linux_get_carrier, \ | |
65c3058c | 2417 | netdev_linux_get_carrier_resets, \ |
1670c579 | 2418 | netdev_linux_set_miimon_interval, \ |
f613a0d7 | 2419 | GET_STATS, \ |
c3827f61 BP |
2420 | SET_STATS, \ |
2421 | \ | |
51f87458 | 2422 | GET_FEATURES, \ |
c3827f61 | 2423 | netdev_linux_set_advertisements, \ |
c3827f61 BP |
2424 | \ |
2425 | netdev_linux_set_policing, \ | |
2426 | netdev_linux_get_qos_types, \ | |
2427 | netdev_linux_get_qos_capabilities, \ | |
2428 | netdev_linux_get_qos, \ | |
2429 | netdev_linux_set_qos, \ | |
2430 | netdev_linux_get_queue, \ | |
2431 | netdev_linux_set_queue, \ | |
2432 | netdev_linux_delete_queue, \ | |
2433 | netdev_linux_get_queue_stats, \ | |
2434 | netdev_linux_dump_queues, \ | |
2435 | netdev_linux_dump_queue_stats, \ | |
2436 | \ | |
2437 | netdev_linux_get_in4, \ | |
2438 | netdev_linux_set_in4, \ | |
2439 | netdev_linux_get_in6, \ | |
2440 | netdev_linux_add_router, \ | |
2441 | netdev_linux_get_next_hop, \ | |
4f925bd3 | 2442 | GET_STATUS, \ |
c3827f61 BP |
2443 | netdev_linux_arp_lookup, \ |
2444 | \ | |
2445 | netdev_linux_update_flags, \ | |
2446 | \ | |
ac4d3bcb | 2447 | netdev_linux_change_seq \ |
c3827f61 BP |
2448 | } |
2449 | ||
2450 | const struct netdev_class netdev_linux_class = | |
2451 | NETDEV_LINUX_CLASS( | |
2452 | "system", | |
2453 | netdev_linux_create, | |
f613a0d7 | 2454 | netdev_linux_get_stats, |
4f925bd3 | 2455 | NULL, /* set_stats */ |
51f87458 | 2456 | netdev_linux_get_features, |
2c2ea5a8 | 2457 | netdev_linux_get_drv_info); |
c3827f61 BP |
2458 | |
2459 | const struct netdev_class netdev_tap_class = | |
2460 | NETDEV_LINUX_CLASS( | |
2461 | "tap", | |
2462 | netdev_linux_create_tap, | |
bba1e6f3 | 2463 | netdev_tap_get_stats, |
4f925bd3 | 2464 | NULL, /* set_stats */ |
51f87458 | 2465 | netdev_linux_get_features, |
2c2ea5a8 | 2466 | netdev_linux_get_drv_info); |
c3827f61 BP |
2467 | |
2468 | const struct netdev_class netdev_internal_class = | |
2469 | NETDEV_LINUX_CLASS( | |
2470 | "internal", | |
2471 | netdev_linux_create, | |
bba1e6f3 | 2472 | netdev_internal_get_stats, |
4f925bd3 | 2473 | netdev_vport_set_stats, |
51f87458 | 2474 | NULL, /* get_features */ |
2c2ea5a8 | 2475 | netdev_internal_get_drv_info); |
8b61709d | 2476 | \f |
c1c9c9c4 | 2477 | /* HTB traffic control class. */ |
559843ed | 2478 | |
c1c9c9c4 | 2479 | #define HTB_N_QUEUES 0xf000 |
8b61709d | 2480 | |
c1c9c9c4 BP |
2481 | struct htb { |
2482 | struct tc tc; | |
2483 | unsigned int max_rate; /* In bytes/s. */ | |
2484 | }; | |
8b61709d | 2485 | |
c1c9c9c4 | 2486 | struct htb_class { |
93b13be8 | 2487 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
2488 | unsigned int min_rate; /* In bytes/s. */ |
2489 | unsigned int max_rate; /* In bytes/s. */ | |
2490 | unsigned int burst; /* In bytes. */ | |
2491 | unsigned int priority; /* Lower values are higher priorities. */ | |
2492 | }; | |
8b61709d | 2493 | |
c1c9c9c4 BP |
2494 | static struct htb * |
2495 | htb_get__(const struct netdev *netdev) | |
2496 | { | |
2497 | struct netdev_dev_linux *netdev_dev = | |
2498 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2499 | return CONTAINER_OF(netdev_dev->tc, struct htb, tc); | |
2500 | } | |
2501 | ||
24045e35 | 2502 | static void |
c1c9c9c4 BP |
2503 | htb_install__(struct netdev *netdev, uint64_t max_rate) |
2504 | { | |
2505 | struct netdev_dev_linux *netdev_dev = | |
2506 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2507 | struct htb *htb; | |
2508 | ||
2509 | htb = xmalloc(sizeof *htb); | |
2510 | tc_init(&htb->tc, &tc_ops_htb); | |
2511 | htb->max_rate = max_rate; | |
2512 | ||
2513 | netdev_dev->tc = &htb->tc; | |
c1c9c9c4 BP |
2514 | } |
2515 | ||
2516 | /* Create an HTB qdisc. | |
2517 | * | |
a339aa81 | 2518 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
2519 | static int |
2520 | htb_setup_qdisc__(struct netdev *netdev) | |
2521 | { | |
2522 | size_t opt_offset; | |
2523 | struct tc_htb_glob opt; | |
2524 | struct ofpbuf request; | |
2525 | struct tcmsg *tcmsg; | |
2526 | ||
2527 | tc_del_qdisc(netdev); | |
2528 | ||
2529 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
2530 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
2531 | if (!tcmsg) { |
2532 | return ENODEV; | |
2533 | } | |
c1c9c9c4 BP |
2534 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
2535 | tcmsg->tcm_parent = TC_H_ROOT; | |
2536 | ||
2537 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2538 | ||
2539 | memset(&opt, 0, sizeof opt); | |
2540 | opt.rate2quantum = 10; | |
2541 | opt.version = 3; | |
4ecf12d5 | 2542 | opt.defcls = 1; |
c1c9c9c4 BP |
2543 | |
2544 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2545 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
2546 | nl_msg_end_nested(&request, opt_offset); | |
2547 | ||
2548 | return tc_transact(&request, NULL); | |
2549 | } | |
2550 | ||
2551 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
2552 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
2553 | static int | |
2554 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
2555 | unsigned int parent, struct htb_class *class) | |
2556 | { | |
2557 | size_t opt_offset; | |
2558 | struct tc_htb_opt opt; | |
2559 | struct ofpbuf request; | |
2560 | struct tcmsg *tcmsg; | |
2561 | int error; | |
2562 | int mtu; | |
2563 | ||
9b020780 PS |
2564 | error = netdev_get_mtu(netdev, &mtu); |
2565 | if (error) { | |
f915f1a8 BP |
2566 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
2567 | netdev_get_name(netdev)); | |
9b020780 | 2568 | return error; |
f915f1a8 | 2569 | } |
c1c9c9c4 BP |
2570 | |
2571 | memset(&opt, 0, sizeof opt); | |
2572 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
2573 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
2574 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); | |
2575 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
2576 | opt.prio = class->priority; | |
2577 | ||
2578 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
23a98ffe BP |
2579 | if (!tcmsg) { |
2580 | return ENODEV; | |
2581 | } | |
c1c9c9c4 BP |
2582 | tcmsg->tcm_handle = handle; |
2583 | tcmsg->tcm_parent = parent; | |
2584 | ||
2585 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2586 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2587 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
2588 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
2589 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
2590 | nl_msg_end_nested(&request, opt_offset); | |
2591 | ||
2592 | error = tc_transact(&request, NULL); | |
2593 | if (error) { | |
2594 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
2595 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
2596 | netdev_get_name(netdev), | |
2597 | tc_get_major(handle), tc_get_minor(handle), | |
2598 | tc_get_major(parent), tc_get_minor(parent), | |
2599 | class->min_rate, class->max_rate, | |
2600 | class->burst, class->priority, strerror(error)); | |
2601 | } | |
2602 | return error; | |
2603 | } | |
2604 | ||
2605 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
2606 | * description of them into 'details'. The description complies with the | |
2607 | * specification given in the vswitch database documentation for linux-htb | |
2608 | * queue details. */ | |
2609 | static int | |
2610 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
2611 | { | |
2612 | static const struct nl_policy tca_htb_policy[] = { | |
2613 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
2614 | .min_len = sizeof(struct tc_htb_opt) }, | |
2615 | }; | |
2616 | ||
2617 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
2618 | const struct tc_htb_opt *htb; | |
2619 | ||
2620 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
2621 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
2622 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
2623 | return EPROTO; | |
2624 | } | |
2625 | ||
2626 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
2627 | class->min_rate = htb->rate.rate; | |
2628 | class->max_rate = htb->ceil.rate; | |
2629 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
2630 | class->priority = htb->prio; | |
2631 | return 0; | |
2632 | } | |
2633 | ||
2634 | static int | |
2635 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
2636 | struct htb_class *options, | |
2637 | struct netdev_queue_stats *stats) | |
2638 | { | |
2639 | struct nlattr *nl_options; | |
2640 | unsigned int handle; | |
2641 | int error; | |
2642 | ||
2643 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
2644 | if (!error && queue_id) { | |
17ee3c1f BP |
2645 | unsigned int major = tc_get_major(handle); |
2646 | unsigned int minor = tc_get_minor(handle); | |
2647 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
2648 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
2649 | } else { |
2650 | error = EPROTO; | |
2651 | } | |
2652 | } | |
2653 | if (!error && options) { | |
2654 | error = htb_parse_tca_options__(nl_options, options); | |
2655 | } | |
2656 | return error; | |
2657 | } | |
2658 | ||
2659 | static void | |
2660 | htb_parse_qdisc_details__(struct netdev *netdev, | |
79f1cbe9 | 2661 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
2662 | { |
2663 | const char *max_rate_s; | |
2664 | ||
79f1cbe9 | 2665 | max_rate_s = smap_get(details, "max-rate"); |
c1c9c9c4 BP |
2666 | hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
2667 | if (!hc->max_rate) { | |
a00ca915 | 2668 | enum netdev_features current; |
c1c9c9c4 BP |
2669 | |
2670 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
2671 | hc->max_rate = netdev_features_to_bps(current) / 8; | |
2672 | } | |
2673 | hc->min_rate = hc->max_rate; | |
2674 | hc->burst = 0; | |
2675 | hc->priority = 0; | |
2676 | } | |
2677 | ||
2678 | static int | |
2679 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 2680 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
2681 | { |
2682 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 EJ |
2683 | const char *min_rate_s = smap_get(details, "min-rate"); |
2684 | const char *max_rate_s = smap_get(details, "max-rate"); | |
2685 | const char *burst_s = smap_get(details, "burst"); | |
2686 | const char *priority_s = smap_get(details, "priority"); | |
9b020780 | 2687 | int mtu, error; |
c1c9c9c4 | 2688 | |
9b020780 PS |
2689 | error = netdev_get_mtu(netdev, &mtu); |
2690 | if (error) { | |
f915f1a8 BP |
2691 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
2692 | netdev_get_name(netdev)); | |
9b020780 | 2693 | return error; |
f915f1a8 BP |
2694 | } |
2695 | ||
4f104611 EJ |
2696 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
2697 | * on uncongested links. */ | |
c45ab5e9 | 2698 | hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
4f104611 | 2699 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
2700 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
2701 | ||
2702 | /* max-rate */ | |
2703 | hc->max_rate = (max_rate_s | |
2704 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
2705 | : htb->max_rate); | |
2706 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); | |
2707 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
2708 | ||
2709 | /* burst | |
2710 | * | |
2711 | * According to hints in the documentation that I've read, it is important | |
2712 | * that 'burst' be at least as big as the largest frame that might be | |
2713 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
2714 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
2715 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
2716 | * the MTU. We actually add 64, instead of 14, as a guard against | |
2717 | * additional headers get tacked on somewhere that we're not aware of. */ | |
c1c9c9c4 BP |
2718 | hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; |
2719 | hc->burst = MAX(hc->burst, mtu + 64); | |
2720 | ||
2721 | /* priority */ | |
2722 | hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; | |
2723 | ||
2724 | return 0; | |
2725 | } | |
2726 | ||
2727 | static int | |
2728 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
2729 | unsigned int parent, struct htb_class *options, | |
2730 | struct netdev_queue_stats *stats) | |
2731 | { | |
2732 | struct ofpbuf *reply; | |
2733 | int error; | |
2734 | ||
2735 | error = tc_query_class(netdev, handle, parent, &reply); | |
2736 | if (!error) { | |
2737 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
2738 | ofpbuf_delete(reply); | |
2739 | } | |
2740 | return error; | |
2741 | } | |
2742 | ||
2743 | static int | |
79f1cbe9 | 2744 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
2745 | { |
2746 | int error; | |
2747 | ||
2748 | error = htb_setup_qdisc__(netdev); | |
2749 | if (!error) { | |
2750 | struct htb_class hc; | |
2751 | ||
2752 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2753 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2754 | tc_make_handle(1, 0), &hc); | |
2755 | if (!error) { | |
2756 | htb_install__(netdev, hc.max_rate); | |
2757 | } | |
2758 | } | |
2759 | return error; | |
2760 | } | |
2761 | ||
93b13be8 BP |
2762 | static struct htb_class * |
2763 | htb_class_cast__(const struct tc_queue *queue) | |
2764 | { | |
2765 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
2766 | } | |
2767 | ||
c1c9c9c4 BP |
2768 | static void |
2769 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
2770 | const struct htb_class *hc) | |
2771 | { | |
2772 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
2773 | size_t hash = hash_int(queue_id, 0); |
2774 | struct tc_queue *queue; | |
c1c9c9c4 BP |
2775 | struct htb_class *hcp; |
2776 | ||
93b13be8 BP |
2777 | queue = tc_find_queue__(netdev, queue_id, hash); |
2778 | if (queue) { | |
2779 | hcp = htb_class_cast__(queue); | |
2780 | } else { | |
c1c9c9c4 | 2781 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
2782 | queue = &hcp->tc_queue; |
2783 | queue->queue_id = queue_id; | |
2784 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); | |
c1c9c9c4 | 2785 | } |
93b13be8 BP |
2786 | |
2787 | hcp->min_rate = hc->min_rate; | |
2788 | hcp->max_rate = hc->max_rate; | |
2789 | hcp->burst = hc->burst; | |
2790 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
2791 | } |
2792 | ||
2793 | static int | |
2794 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2795 | { | |
c1c9c9c4 BP |
2796 | struct ofpbuf msg; |
2797 | struct nl_dump dump; | |
2798 | struct htb_class hc; | |
c1c9c9c4 BP |
2799 | |
2800 | /* Get qdisc options. */ | |
2801 | hc.max_rate = 0; | |
2802 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 2803 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
2804 | |
2805 | /* Get queues. */ | |
23a98ffe BP |
2806 | if (!start_queue_dump(netdev, &dump)) { |
2807 | return ENODEV; | |
2808 | } | |
c1c9c9c4 BP |
2809 | while (nl_dump_next(&dump, &msg)) { |
2810 | unsigned int queue_id; | |
2811 | ||
2812 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
2813 | htb_update_queue__(netdev, queue_id, &hc); | |
2814 | } | |
2815 | } | |
2816 | nl_dump_done(&dump); | |
2817 | ||
2818 | return 0; | |
2819 | } | |
2820 | ||
2821 | static void | |
2822 | htb_tc_destroy(struct tc *tc) | |
2823 | { | |
2824 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
93b13be8 | 2825 | struct htb_class *hc, *next; |
c1c9c9c4 | 2826 | |
4e8e4213 | 2827 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { |
93b13be8 | 2828 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 BP |
2829 | free(hc); |
2830 | } | |
2831 | tc_destroy(tc); | |
2832 | free(htb); | |
2833 | } | |
2834 | ||
2835 | static int | |
79f1cbe9 | 2836 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
2837 | { |
2838 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 2839 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
2840 | return 0; |
2841 | } | |
2842 | ||
2843 | static int | |
79f1cbe9 | 2844 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
2845 | { |
2846 | struct htb_class hc; | |
2847 | int error; | |
2848 | ||
2849 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2850 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2851 | tc_make_handle(1, 0), &hc); | |
2852 | if (!error) { | |
2853 | htb_get__(netdev)->max_rate = hc.max_rate; | |
2854 | } | |
2855 | return error; | |
2856 | } | |
2857 | ||
2858 | static int | |
93b13be8 | 2859 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 2860 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 2861 | { |
93b13be8 | 2862 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2863 | |
79f1cbe9 | 2864 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 2865 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 2866 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 2867 | } |
79f1cbe9 | 2868 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 2869 | if (hc->priority) { |
79f1cbe9 | 2870 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
2871 | } |
2872 | return 0; | |
2873 | } | |
2874 | ||
2875 | static int | |
2876 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 2877 | const struct smap *details) |
c1c9c9c4 BP |
2878 | { |
2879 | struct htb_class hc; | |
2880 | int error; | |
2881 | ||
2882 | error = htb_parse_class_details__(netdev, details, &hc); | |
2883 | if (error) { | |
2884 | return error; | |
2885 | } | |
2886 | ||
17ee3c1f | 2887 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
2888 | tc_make_handle(1, 0xfffe), &hc); |
2889 | if (error) { | |
2890 | return error; | |
2891 | } | |
2892 | ||
2893 | htb_update_queue__(netdev, queue_id, &hc); | |
2894 | return 0; | |
2895 | } | |
2896 | ||
2897 | static int | |
93b13be8 | 2898 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 2899 | { |
93b13be8 | 2900 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2901 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
2902 | int error; |
2903 | ||
93b13be8 | 2904 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 2905 | if (!error) { |
93b13be8 | 2906 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 2907 | free(hc); |
c1c9c9c4 BP |
2908 | } |
2909 | return error; | |
2910 | } | |
2911 | ||
2912 | static int | |
93b13be8 | 2913 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
2914 | struct netdev_queue_stats *stats) |
2915 | { | |
93b13be8 | 2916 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
2917 | tc_make_handle(1, 0xfffe), NULL, stats); |
2918 | } | |
2919 | ||
2920 | static int | |
2921 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
2922 | const struct ofpbuf *nlmsg, | |
2923 | netdev_dump_queue_stats_cb *cb, void *aux) | |
2924 | { | |
2925 | struct netdev_queue_stats stats; | |
17ee3c1f | 2926 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
2927 | int error; |
2928 | ||
2929 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
2930 | if (error) { | |
2931 | return error; | |
2932 | } | |
2933 | ||
17ee3c1f BP |
2934 | major = tc_get_major(handle); |
2935 | minor = tc_get_minor(handle); | |
2936 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 2937 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
2938 | } |
2939 | return 0; | |
2940 | } | |
2941 | ||
2942 | static const struct tc_ops tc_ops_htb = { | |
2943 | "htb", /* linux_name */ | |
2944 | "linux-htb", /* ovs_name */ | |
2945 | HTB_N_QUEUES, /* n_queues */ | |
2946 | htb_tc_install, | |
2947 | htb_tc_load, | |
2948 | htb_tc_destroy, | |
2949 | htb_qdisc_get, | |
2950 | htb_qdisc_set, | |
2951 | htb_class_get, | |
2952 | htb_class_set, | |
2953 | htb_class_delete, | |
2954 | htb_class_get_stats, | |
2955 | htb_class_dump_stats | |
2956 | }; | |
2957 | \f | |
a339aa81 EJ |
2958 | /* "linux-hfsc" traffic control class. */ |
2959 | ||
2960 | #define HFSC_N_QUEUES 0xf000 | |
2961 | ||
2962 | struct hfsc { | |
2963 | struct tc tc; | |
2964 | uint32_t max_rate; | |
2965 | }; | |
2966 | ||
2967 | struct hfsc_class { | |
2968 | struct tc_queue tc_queue; | |
2969 | uint32_t min_rate; | |
2970 | uint32_t max_rate; | |
2971 | }; | |
2972 | ||
2973 | static struct hfsc * | |
2974 | hfsc_get__(const struct netdev *netdev) | |
2975 | { | |
2976 | struct netdev_dev_linux *netdev_dev; | |
2977 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2978 | return CONTAINER_OF(netdev_dev->tc, struct hfsc, tc); | |
2979 | } | |
2980 | ||
2981 | static struct hfsc_class * | |
2982 | hfsc_class_cast__(const struct tc_queue *queue) | |
2983 | { | |
2984 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
2985 | } | |
2986 | ||
24045e35 | 2987 | static void |
a339aa81 EJ |
2988 | hfsc_install__(struct netdev *netdev, uint32_t max_rate) |
2989 | { | |
2990 | struct netdev_dev_linux * netdev_dev; | |
2991 | struct hfsc *hfsc; | |
2992 | ||
2993 | netdev_dev = netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
2994 | hfsc = xmalloc(sizeof *hfsc); | |
2995 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
2996 | hfsc->max_rate = max_rate; | |
2997 | netdev_dev->tc = &hfsc->tc; | |
a339aa81 EJ |
2998 | } |
2999 | ||
3000 | static void | |
3001 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
3002 | const struct hfsc_class *hc) | |
3003 | { | |
3004 | size_t hash; | |
3005 | struct hfsc *hfsc; | |
3006 | struct hfsc_class *hcp; | |
3007 | struct tc_queue *queue; | |
3008 | ||
3009 | hfsc = hfsc_get__(netdev); | |
3010 | hash = hash_int(queue_id, 0); | |
3011 | ||
3012 | queue = tc_find_queue__(netdev, queue_id, hash); | |
3013 | if (queue) { | |
3014 | hcp = hfsc_class_cast__(queue); | |
3015 | } else { | |
3016 | hcp = xmalloc(sizeof *hcp); | |
3017 | queue = &hcp->tc_queue; | |
3018 | queue->queue_id = queue_id; | |
3019 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); | |
3020 | } | |
3021 | ||
3022 | hcp->min_rate = hc->min_rate; | |
3023 | hcp->max_rate = hc->max_rate; | |
3024 | } | |
3025 | ||
3026 | static int | |
3027 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
3028 | { | |
3029 | const struct tc_service_curve *rsc, *fsc, *usc; | |
3030 | static const struct nl_policy tca_hfsc_policy[] = { | |
3031 | [TCA_HFSC_RSC] = { | |
3032 | .type = NL_A_UNSPEC, | |
3033 | .optional = false, | |
3034 | .min_len = sizeof(struct tc_service_curve), | |
3035 | }, | |
3036 | [TCA_HFSC_FSC] = { | |
3037 | .type = NL_A_UNSPEC, | |
3038 | .optional = false, | |
3039 | .min_len = sizeof(struct tc_service_curve), | |
3040 | }, | |
3041 | [TCA_HFSC_USC] = { | |
3042 | .type = NL_A_UNSPEC, | |
3043 | .optional = false, | |
3044 | .min_len = sizeof(struct tc_service_curve), | |
3045 | }, | |
3046 | }; | |
3047 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
3048 | ||
3049 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
3050 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
3051 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
3052 | return EPROTO; | |
3053 | } | |
3054 | ||
3055 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
3056 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
3057 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
3058 | ||
3059 | if (rsc->m1 != 0 || rsc->d != 0 || | |
3060 | fsc->m1 != 0 || fsc->d != 0 || | |
3061 | usc->m1 != 0 || usc->d != 0) { | |
3062 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3063 | "Non-linear service curves are not supported."); | |
3064 | return EPROTO; | |
3065 | } | |
3066 | ||
3067 | if (rsc->m2 != fsc->m2) { | |
3068 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3069 | "Real-time service curves are not supported "); | |
3070 | return EPROTO; | |
3071 | } | |
3072 | ||
3073 | if (rsc->m2 > usc->m2) { | |
3074 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3075 | "Min-rate service curve is greater than " | |
3076 | "the max-rate service curve."); | |
3077 | return EPROTO; | |
3078 | } | |
3079 | ||
3080 | class->min_rate = fsc->m2; | |
3081 | class->max_rate = usc->m2; | |
3082 | return 0; | |
3083 | } | |
3084 | ||
3085 | static int | |
3086 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
3087 | struct hfsc_class *options, | |
3088 | struct netdev_queue_stats *stats) | |
3089 | { | |
3090 | int error; | |
3091 | unsigned int handle; | |
3092 | struct nlattr *nl_options; | |
3093 | ||
3094 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
3095 | if (error) { | |
3096 | return error; | |
3097 | } | |
3098 | ||
3099 | if (queue_id) { | |
3100 | unsigned int major, minor; | |
3101 | ||
3102 | major = tc_get_major(handle); | |
3103 | minor = tc_get_minor(handle); | |
3104 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
3105 | *queue_id = minor - 1; | |
3106 | } else { | |
3107 | return EPROTO; | |
3108 | } | |
3109 | } | |
3110 | ||
3111 | if (options) { | |
3112 | error = hfsc_parse_tca_options__(nl_options, options); | |
3113 | } | |
3114 | ||
3115 | return error; | |
3116 | } | |
3117 | ||
3118 | static int | |
3119 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
3120 | unsigned int parent, struct hfsc_class *options, | |
3121 | struct netdev_queue_stats *stats) | |
3122 | { | |
3123 | int error; | |
3124 | struct ofpbuf *reply; | |
3125 | ||
3126 | error = tc_query_class(netdev, handle, parent, &reply); | |
3127 | if (error) { | |
3128 | return error; | |
3129 | } | |
3130 | ||
3131 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
3132 | ofpbuf_delete(reply); | |
3133 | return error; | |
3134 | } | |
3135 | ||
3136 | static void | |
79f1cbe9 | 3137 | hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, |
a339aa81 EJ |
3138 | struct hfsc_class *class) |
3139 | { | |
3140 | uint32_t max_rate; | |
3141 | const char *max_rate_s; | |
3142 | ||
79f1cbe9 | 3143 | max_rate_s = smap_get(details, "max-rate"); |
a339aa81 EJ |
3144 | max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
3145 | ||
3146 | if (!max_rate) { | |
a00ca915 | 3147 | enum netdev_features current; |
a339aa81 EJ |
3148 | |
3149 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
3150 | max_rate = netdev_features_to_bps(current) / 8; | |
3151 | } | |
3152 | ||
3153 | class->min_rate = max_rate; | |
3154 | class->max_rate = max_rate; | |
3155 | } | |
3156 | ||
3157 | static int | |
3158 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 3159 | const struct smap *details, |
a339aa81 EJ |
3160 | struct hfsc_class * class) |
3161 | { | |
3162 | const struct hfsc *hfsc; | |
3163 | uint32_t min_rate, max_rate; | |
3164 | const char *min_rate_s, *max_rate_s; | |
3165 | ||
3166 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 EJ |
3167 | min_rate_s = smap_get(details, "min-rate"); |
3168 | max_rate_s = smap_get(details, "max-rate"); | |
a339aa81 | 3169 | |
c45ab5e9 | 3170 | min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
79398bad | 3171 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
3172 | min_rate = MIN(min_rate, hfsc->max_rate); |
3173 | ||
3174 | max_rate = (max_rate_s | |
3175 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
3176 | : hfsc->max_rate); | |
3177 | max_rate = MAX(max_rate, min_rate); | |
3178 | max_rate = MIN(max_rate, hfsc->max_rate); | |
3179 | ||
3180 | class->min_rate = min_rate; | |
3181 | class->max_rate = max_rate; | |
3182 | ||
3183 | return 0; | |
3184 | } | |
3185 | ||
3186 | /* Create an HFSC qdisc. | |
3187 | * | |
3188 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
3189 | static int | |
3190 | hfsc_setup_qdisc__(struct netdev * netdev) | |
3191 | { | |
3192 | struct tcmsg *tcmsg; | |
3193 | struct ofpbuf request; | |
3194 | struct tc_hfsc_qopt opt; | |
3195 | ||
3196 | tc_del_qdisc(netdev); | |
3197 | ||
3198 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3199 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3200 | ||
3201 | if (!tcmsg) { | |
3202 | return ENODEV; | |
3203 | } | |
3204 | ||
3205 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3206 | tcmsg->tcm_parent = TC_H_ROOT; | |
3207 | ||
3208 | memset(&opt, 0, sizeof opt); | |
3209 | opt.defcls = 1; | |
3210 | ||
3211 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3212 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3213 | ||
3214 | return tc_transact(&request, NULL); | |
3215 | } | |
3216 | ||
3217 | /* Create an HFSC class. | |
3218 | * | |
3219 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
3220 | * sc rate <min_rate> ul rate <max_rate>" */ | |
3221 | static int | |
3222 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
3223 | unsigned int parent, struct hfsc_class *class) | |
3224 | { | |
3225 | int error; | |
3226 | size_t opt_offset; | |
3227 | struct tcmsg *tcmsg; | |
3228 | struct ofpbuf request; | |
3229 | struct tc_service_curve min, max; | |
3230 | ||
3231 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
3232 | ||
3233 | if (!tcmsg) { | |
3234 | return ENODEV; | |
3235 | } | |
3236 | ||
3237 | tcmsg->tcm_handle = handle; | |
3238 | tcmsg->tcm_parent = parent; | |
3239 | ||
3240 | min.m1 = 0; | |
3241 | min.d = 0; | |
3242 | min.m2 = class->min_rate; | |
3243 | ||
3244 | max.m1 = 0; | |
3245 | max.d = 0; | |
3246 | max.m2 = class->max_rate; | |
3247 | ||
3248 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3249 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3250 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
3251 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
3252 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
3253 | nl_msg_end_nested(&request, opt_offset); | |
3254 | ||
3255 | error = tc_transact(&request, NULL); | |
3256 | if (error) { | |
3257 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3258 | "min-rate %ubps, max-rate %ubps (%s)", | |
3259 | netdev_get_name(netdev), | |
3260 | tc_get_major(handle), tc_get_minor(handle), | |
3261 | tc_get_major(parent), tc_get_minor(parent), | |
3262 | class->min_rate, class->max_rate, strerror(error)); | |
3263 | } | |
3264 | ||
3265 | return error; | |
3266 | } | |
3267 | ||
3268 | static int | |
79f1cbe9 | 3269 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
3270 | { |
3271 | int error; | |
3272 | struct hfsc_class class; | |
3273 | ||
3274 | error = hfsc_setup_qdisc__(netdev); | |
3275 | ||
3276 | if (error) { | |
3277 | return error; | |
3278 | } | |
3279 | ||
3280 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3281 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3282 | tc_make_handle(1, 0), &class); | |
3283 | ||
3284 | if (error) { | |
3285 | return error; | |
3286 | } | |
3287 | ||
3288 | hfsc_install__(netdev, class.max_rate); | |
3289 | return 0; | |
3290 | } | |
3291 | ||
3292 | static int | |
3293 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3294 | { | |
3295 | struct ofpbuf msg; | |
a339aa81 EJ |
3296 | struct nl_dump dump; |
3297 | struct hfsc_class hc; | |
3298 | ||
3299 | hc.max_rate = 0; | |
3300 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 3301 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 EJ |
3302 | |
3303 | if (!start_queue_dump(netdev, &dump)) { | |
3304 | return ENODEV; | |
3305 | } | |
3306 | ||
3307 | while (nl_dump_next(&dump, &msg)) { | |
3308 | unsigned int queue_id; | |
3309 | ||
3310 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
3311 | hfsc_update_queue__(netdev, queue_id, &hc); | |
3312 | } | |
3313 | } | |
3314 | ||
3315 | nl_dump_done(&dump); | |
3316 | return 0; | |
3317 | } | |
3318 | ||
3319 | static void | |
3320 | hfsc_tc_destroy(struct tc *tc) | |
3321 | { | |
3322 | struct hfsc *hfsc; | |
3323 | struct hfsc_class *hc, *next; | |
3324 | ||
3325 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
3326 | ||
3327 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
3328 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3329 | free(hc); | |
3330 | } | |
3331 | ||
3332 | tc_destroy(tc); | |
3333 | free(hfsc); | |
3334 | } | |
3335 | ||
3336 | static int | |
79f1cbe9 | 3337 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
3338 | { |
3339 | const struct hfsc *hfsc; | |
3340 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 3341 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
3342 | return 0; |
3343 | } | |
3344 | ||
3345 | static int | |
79f1cbe9 | 3346 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
3347 | { |
3348 | int error; | |
3349 | struct hfsc_class class; | |
3350 | ||
3351 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3352 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3353 | tc_make_handle(1, 0), &class); | |
3354 | ||
3355 | if (!error) { | |
3356 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
3357 | } | |
3358 | ||
3359 | return error; | |
3360 | } | |
3361 | ||
3362 | static int | |
3363 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 3364 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
3365 | { |
3366 | const struct hfsc_class *hc; | |
3367 | ||
3368 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 3369 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 3370 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 3371 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
3372 | } |
3373 | return 0; | |
3374 | } | |
3375 | ||
3376 | static int | |
3377 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 3378 | const struct smap *details) |
a339aa81 EJ |
3379 | { |
3380 | int error; | |
3381 | struct hfsc_class class; | |
3382 | ||
3383 | error = hfsc_parse_class_details__(netdev, details, &class); | |
3384 | if (error) { | |
3385 | return error; | |
3386 | } | |
3387 | ||
3388 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
3389 | tc_make_handle(1, 0xfffe), &class); | |
3390 | if (error) { | |
3391 | return error; | |
3392 | } | |
3393 | ||
3394 | hfsc_update_queue__(netdev, queue_id, &class); | |
3395 | return 0; | |
3396 | } | |
3397 | ||
3398 | static int | |
3399 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
3400 | { | |
3401 | int error; | |
3402 | struct hfsc *hfsc; | |
3403 | struct hfsc_class *hc; | |
3404 | ||
3405 | hc = hfsc_class_cast__(queue); | |
3406 | hfsc = hfsc_get__(netdev); | |
3407 | ||
3408 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
3409 | if (!error) { | |
3410 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3411 | free(hc); | |
3412 | } | |
3413 | return error; | |
3414 | } | |
3415 | ||
3416 | static int | |
3417 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
3418 | struct netdev_queue_stats *stats) | |
3419 | { | |
3420 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
3421 | tc_make_handle(1, 0xfffe), NULL, stats); | |
3422 | } | |
3423 | ||
3424 | static int | |
3425 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
3426 | const struct ofpbuf *nlmsg, | |
3427 | netdev_dump_queue_stats_cb *cb, void *aux) | |
3428 | { | |
3429 | struct netdev_queue_stats stats; | |
3430 | unsigned int handle, major, minor; | |
3431 | int error; | |
3432 | ||
3433 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
3434 | if (error) { | |
3435 | return error; | |
3436 | } | |
3437 | ||
3438 | major = tc_get_major(handle); | |
3439 | minor = tc_get_minor(handle); | |
3440 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
3441 | (*cb)(minor - 1, &stats, aux); | |
3442 | } | |
3443 | return 0; | |
3444 | } | |
3445 | ||
3446 | static const struct tc_ops tc_ops_hfsc = { | |
3447 | "hfsc", /* linux_name */ | |
3448 | "linux-hfsc", /* ovs_name */ | |
3449 | HFSC_N_QUEUES, /* n_queues */ | |
3450 | hfsc_tc_install, /* tc_install */ | |
3451 | hfsc_tc_load, /* tc_load */ | |
3452 | hfsc_tc_destroy, /* tc_destroy */ | |
3453 | hfsc_qdisc_get, /* qdisc_get */ | |
3454 | hfsc_qdisc_set, /* qdisc_set */ | |
3455 | hfsc_class_get, /* class_get */ | |
3456 | hfsc_class_set, /* class_set */ | |
3457 | hfsc_class_delete, /* class_delete */ | |
3458 | hfsc_class_get_stats, /* class_get_stats */ | |
3459 | hfsc_class_dump_stats /* class_dump_stats */ | |
3460 | }; | |
3461 | \f | |
c1c9c9c4 BP |
3462 | /* "linux-default" traffic control class. |
3463 | * | |
3464 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
3465 | * the "" (empty string) QoS type in the OVS database. */ | |
3466 | ||
3467 | static void | |
3468 | default_install__(struct netdev *netdev) | |
3469 | { | |
3470 | struct netdev_dev_linux *netdev_dev = | |
3471 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3472 | static struct tc *tc; | |
3473 | ||
3474 | if (!tc) { | |
3475 | tc = xmalloc(sizeof *tc); | |
3476 | tc_init(tc, &tc_ops_default); | |
3477 | } | |
3478 | netdev_dev->tc = tc; | |
3479 | } | |
3480 | ||
3481 | static int | |
3482 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 3483 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
3484 | { |
3485 | default_install__(netdev); | |
3486 | return 0; | |
3487 | } | |
3488 | ||
3489 | static int | |
3490 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3491 | { | |
3492 | default_install__(netdev); | |
3493 | return 0; | |
3494 | } | |
3495 | ||
3496 | static const struct tc_ops tc_ops_default = { | |
3497 | NULL, /* linux_name */ | |
3498 | "", /* ovs_name */ | |
3499 | 0, /* n_queues */ | |
3500 | default_tc_install, | |
3501 | default_tc_load, | |
3502 | NULL, /* tc_destroy */ | |
3503 | NULL, /* qdisc_get */ | |
3504 | NULL, /* qdisc_set */ | |
3505 | NULL, /* class_get */ | |
3506 | NULL, /* class_set */ | |
3507 | NULL, /* class_delete */ | |
3508 | NULL, /* class_get_stats */ | |
3509 | NULL /* class_dump_stats */ | |
3510 | }; | |
3511 | \f | |
3512 | /* "linux-other" traffic control class. | |
3513 | * | |
3514 | * */ | |
3515 | ||
3516 | static int | |
3517 | other_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3518 | { | |
3519 | struct netdev_dev_linux *netdev_dev = | |
3520 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
3521 | static struct tc *tc; | |
3522 | ||
3523 | if (!tc) { | |
3524 | tc = xmalloc(sizeof *tc); | |
3525 | tc_init(tc, &tc_ops_other); | |
3526 | } | |
3527 | netdev_dev->tc = tc; | |
3528 | return 0; | |
3529 | } | |
3530 | ||
3531 | static const struct tc_ops tc_ops_other = { | |
3532 | NULL, /* linux_name */ | |
3533 | "linux-other", /* ovs_name */ | |
3534 | 0, /* n_queues */ | |
3535 | NULL, /* tc_install */ | |
3536 | other_tc_load, | |
3537 | NULL, /* tc_destroy */ | |
3538 | NULL, /* qdisc_get */ | |
3539 | NULL, /* qdisc_set */ | |
3540 | NULL, /* class_get */ | |
3541 | NULL, /* class_set */ | |
3542 | NULL, /* class_delete */ | |
3543 | NULL, /* class_get_stats */ | |
3544 | NULL /* class_dump_stats */ | |
3545 | }; | |
3546 | \f | |
3547 | /* Traffic control. */ | |
3548 | ||
3549 | /* Number of kernel "tc" ticks per second. */ | |
3550 | static double ticks_per_s; | |
3551 | ||
3552 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
3553 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
3554 | * one jiffy's worth of data. | |
3555 | * | |
3556 | * There are two possibilities here: | |
3557 | * | |
3558 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
3559 | * approximate range of 100 to 1024. That means that we really need to | |
3560 | * make sure that the qdisc can buffer that much data. | |
3561 | * | |
3562 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
3563 | * has finely granular timers and there's no need to fudge additional room | |
3564 | * for buffers. (There's no extra effort needed to implement that: the | |
3565 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
3566 | * come out as 0 in the division. Small integer results in the case of | |
3567 | * really high dividends won't have any real effect anyhow.) | |
3568 | */ | |
3569 | static unsigned int buffer_hz; | |
3570 | ||
3571 | /* Returns tc handle 'major':'minor'. */ | |
3572 | static unsigned int | |
3573 | tc_make_handle(unsigned int major, unsigned int minor) | |
3574 | { | |
3575 | return TC_H_MAKE(major << 16, minor); | |
3576 | } | |
3577 | ||
3578 | /* Returns the major number from 'handle'. */ | |
3579 | static unsigned int | |
3580 | tc_get_major(unsigned int handle) | |
3581 | { | |
3582 | return TC_H_MAJ(handle) >> 16; | |
3583 | } | |
3584 | ||
3585 | /* Returns the minor number from 'handle'. */ | |
3586 | static unsigned int | |
3587 | tc_get_minor(unsigned int handle) | |
3588 | { | |
3589 | return TC_H_MIN(handle); | |
3590 | } | |
3591 | ||
3592 | static struct tcmsg * | |
3593 | tc_make_request(const struct netdev *netdev, int type, unsigned int flags, | |
3594 | struct ofpbuf *request) | |
3595 | { | |
3596 | struct tcmsg *tcmsg; | |
3597 | int ifindex; | |
3598 | int error; | |
3599 | ||
3600 | error = get_ifindex(netdev, &ifindex); | |
3601 | if (error) { | |
3602 | return NULL; | |
3603 | } | |
3604 | ||
3605 | ofpbuf_init(request, 512); | |
3606 | nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); | |
3607 | tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); | |
3608 | tcmsg->tcm_family = AF_UNSPEC; | |
3609 | tcmsg->tcm_ifindex = ifindex; | |
3610 | /* Caller should fill in tcmsg->tcm_handle. */ | |
3611 | /* Caller should fill in tcmsg->tcm_parent. */ | |
3612 | ||
3613 | return tcmsg; | |
3614 | } | |
3615 | ||
3616 | static int | |
3617 | tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) | |
3618 | { | |
3619 | int error = nl_sock_transact(rtnl_sock, request, replyp); | |
3620 | ofpbuf_uninit(request); | |
3621 | return error; | |
3622 | } | |
3623 | ||
f8500004 JP |
3624 | /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for |
3625 | * policing configuration. | |
3626 | * | |
3627 | * This function is equivalent to running the following when 'add' is true: | |
3628 | * /sbin/tc qdisc add dev <devname> handle ffff: ingress | |
3629 | * | |
3630 | * This function is equivalent to running the following when 'add' is false: | |
3631 | * /sbin/tc qdisc del dev <devname> handle ffff: ingress | |
3632 | * | |
3633 | * The configuration and stats may be seen with the following command: | |
3634 | * /sbin/tc -s qdisc show dev <devname> | |
3635 | * | |
3636 | * Returns 0 if successful, otherwise a positive errno value. | |
3637 | */ | |
3638 | static int | |
3639 | tc_add_del_ingress_qdisc(struct netdev *netdev, bool add) | |
3640 | { | |
3641 | struct ofpbuf request; | |
3642 | struct tcmsg *tcmsg; | |
3643 | int error; | |
3644 | int type = add ? RTM_NEWQDISC : RTM_DELQDISC; | |
3645 | int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0; | |
3646 | ||
3647 | tcmsg = tc_make_request(netdev, type, flags, &request); | |
3648 | if (!tcmsg) { | |
3649 | return ENODEV; | |
3650 | } | |
3651 | tcmsg->tcm_handle = tc_make_handle(0xffff, 0); | |
3652 | tcmsg->tcm_parent = TC_H_INGRESS; | |
3653 | nl_msg_put_string(&request, TCA_KIND, "ingress"); | |
3654 | nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); | |
3655 | ||
3656 | error = tc_transact(&request, NULL); | |
3657 | if (error) { | |
3658 | /* If we're deleting the qdisc, don't worry about some of the | |
3659 | * error conditions. */ | |
3660 | if (!add && (error == ENOENT || error == EINVAL)) { | |
3661 | return 0; | |
3662 | } | |
3663 | return error; | |
3664 | } | |
3665 | ||
3666 | return 0; | |
3667 | } | |
3668 | ||
3669 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size | |
3670 | * of 'kbits_burst'. | |
3671 | * | |
3672 | * This function is equivalent to running: | |
3673 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
3674 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
3675 | * mtu 65535 drop | |
3676 | * | |
3677 | * The configuration and stats may be seen with the following command: | |
3678 | * /sbin/tc -s filter show <devname> eth0 parent ffff: | |
3679 | * | |
3680 | * Returns 0 if successful, otherwise a positive errno value. | |
3681 | */ | |
3682 | static int | |
3683 | tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst) | |
3684 | { | |
3685 | struct tc_police tc_police; | |
3686 | struct ofpbuf request; | |
3687 | struct tcmsg *tcmsg; | |
3688 | size_t basic_offset; | |
3689 | size_t police_offset; | |
3690 | int error; | |
3691 | int mtu = 65535; | |
3692 | ||
3693 | memset(&tc_police, 0, sizeof tc_police); | |
3694 | tc_police.action = TC_POLICE_SHOT; | |
3695 | tc_police.mtu = mtu; | |
3696 | tc_fill_rate(&tc_police.rate, kbits_rate/8 * 1000, mtu); | |
3697 | tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate, | |
3698 | kbits_burst * 1024); | |
3699 | ||
3700 | tcmsg = tc_make_request(netdev, RTM_NEWTFILTER, | |
3701 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3702 | if (!tcmsg) { | |
3703 | return ENODEV; | |
3704 | } | |
3705 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
3706 | tcmsg->tcm_info = tc_make_handle(49, | |
3707 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
3708 | ||
3709 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
3710 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3711 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
3712 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
3713 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
3714 | nl_msg_end_nested(&request, police_offset); | |
3715 | nl_msg_end_nested(&request, basic_offset); | |
3716 | ||
3717 | error = tc_transact(&request, NULL); | |
3718 | if (error) { | |
3719 | return error; | |
3720 | } | |
3721 | ||
3722 | return 0; | |
3723 | } | |
3724 | ||
c1c9c9c4 BP |
3725 | static void |
3726 | read_psched(void) | |
3727 | { | |
3728 | /* The values in psched are not individually very meaningful, but they are | |
3729 | * important. The tables below show some values seen in the wild. | |
3730 | * | |
3731 | * Some notes: | |
3732 | * | |
3733 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
3734 | * (Before that, there are hints that it was 1000000000.) | |
3735 | * | |
3736 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
3737 | * above. | |
3738 | * | |
3739 | * /proc/net/psched | |
3740 | * ----------------------------------- | |
3741 | * [1] 000c8000 000f4240 000f4240 00000064 | |
3742 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
3743 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
3744 | * [4] 000003e8 00000400 000f4240 00000064 | |
3745 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
3746 | * [6] 000003e8 00000040 000f4240 000000f9 | |
3747 | * | |
3748 | * a b c d ticks_per_s buffer_hz | |
3749 | * ------- --------- ---------- ------------- ----------- ------------- | |
3750 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
3751 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3752 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3753 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
3754 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
3755 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
3756 | * | |
3757 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
3758 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
3759 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
3760 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
3761 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
3762 | * [6] 2.6.34 from kernel.org on KVM | |
3763 | */ | |
3764 | static const char fn[] = "/proc/net/psched"; | |
3765 | unsigned int a, b, c, d; | |
3766 | FILE *stream; | |
3767 | ||
3768 | ticks_per_s = 1.0; | |
3769 | buffer_hz = 100; | |
3770 | ||
3771 | stream = fopen(fn, "r"); | |
3772 | if (!stream) { | |
3773 | VLOG_WARN("%s: open failed: %s", fn, strerror(errno)); | |
3774 | return; | |
3775 | } | |
3776 | ||
3777 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
3778 | VLOG_WARN("%s: read failed", fn); | |
3779 | fclose(stream); | |
3780 | return; | |
3781 | } | |
3782 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
3783 | fclose(stream); | |
3784 | ||
3785 | if (!a || !c) { | |
3786 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
3787 | return; | |
3788 | } | |
3789 | ||
3790 | ticks_per_s = (double) a * c / b; | |
3791 | if (c == 1000000) { | |
3792 | buffer_hz = d; | |
3793 | } else { | |
3794 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
3795 | fn, a, b, c, d); | |
3796 | } | |
3797 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
3798 | } | |
3799 | ||
3800 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
3801 | * rate of 'rate' bytes per second. */ | |
3802 | static unsigned int | |
3803 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
3804 | { | |
3805 | if (!buffer_hz) { | |
3806 | read_psched(); | |
3807 | } | |
3808 | return (rate * ticks) / ticks_per_s; | |
3809 | } | |
3810 | ||
3811 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
3812 | * rate of 'rate' bytes per second. */ | |
3813 | static unsigned int | |
3814 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
3815 | { | |
3816 | if (!buffer_hz) { | |
3817 | read_psched(); | |
3818 | } | |
015c93a4 | 3819 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
3820 | } |
3821 | ||
3822 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
3823 | * a transmission rate of 'rate' bytes per second. */ | |
3824 | static unsigned int | |
3825 | tc_buffer_per_jiffy(unsigned int rate) | |
3826 | { | |
3827 | if (!buffer_hz) { | |
3828 | read_psched(); | |
3829 | } | |
3830 | return rate / buffer_hz; | |
3831 | } | |
3832 | ||
3833 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
3834 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
3835 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
3836 | * stores NULL into it if it is absent. | |
3837 | * | |
3838 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
3839 | * 'msg'. | |
3840 | * | |
3841 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3842 | static int | |
3843 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
3844 | struct nlattr **options) | |
3845 | { | |
3846 | static const struct nl_policy tca_policy[] = { | |
3847 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
3848 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
3849 | }; | |
3850 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3851 | ||
3852 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3853 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3854 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
3855 | goto error; | |
3856 | } | |
3857 | ||
3858 | if (kind) { | |
3859 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
3860 | } | |
3861 | ||
3862 | if (options) { | |
3863 | *options = ta[TCA_OPTIONS]; | |
3864 | } | |
3865 | ||
3866 | return 0; | |
3867 | ||
3868 | error: | |
3869 | if (kind) { | |
3870 | *kind = NULL; | |
3871 | } | |
3872 | if (options) { | |
3873 | *options = NULL; | |
3874 | } | |
3875 | return EPROTO; | |
3876 | } | |
3877 | ||
3878 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
3879 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
3880 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
3881 | * arguments may be null. | |
3882 | * | |
3883 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3884 | static int | |
3885 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
3886 | struct nlattr **options, struct netdev_queue_stats *stats) | |
3887 | { | |
3888 | static const struct nl_policy tca_policy[] = { | |
3889 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
3890 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
3891 | }; | |
3892 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3893 | ||
3894 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3895 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3896 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
3897 | goto error; | |
3898 | } | |
3899 | ||
3900 | if (handlep) { | |
3901 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
3902 | *handlep = tc->tcm_handle; | |
3903 | } | |
3904 | ||
3905 | if (options) { | |
3906 | *options = ta[TCA_OPTIONS]; | |
3907 | } | |
3908 | ||
3909 | if (stats) { | |
3910 | const struct gnet_stats_queue *gsq; | |
3911 | struct gnet_stats_basic gsb; | |
3912 | ||
3913 | static const struct nl_policy stats_policy[] = { | |
3914 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
3915 | .min_len = sizeof gsb }, | |
3916 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
3917 | .min_len = sizeof *gsq }, | |
3918 | }; | |
3919 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
3920 | ||
3921 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
3922 | sa, ARRAY_SIZE(sa))) { | |
3923 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
3924 | goto error; | |
3925 | } | |
3926 | ||
3927 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
3928 | * some arch/bitsize combinations. Newer versions of Linux have a | |
3929 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
3930 | * easiest thing to do is just to make a copy. */ | |
3931 | memset(&gsb, 0, sizeof gsb); | |
3932 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
3933 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
3934 | stats->tx_bytes = gsb.bytes; | |
3935 | stats->tx_packets = gsb.packets; | |
3936 | ||
3937 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
3938 | stats->tx_errors = gsq->drops; | |
3939 | } | |
3940 | ||
3941 | return 0; | |
3942 | ||
3943 | error: | |
3944 | if (options) { | |
3945 | *options = NULL; | |
3946 | } | |
3947 | if (stats) { | |
3948 | memset(stats, 0, sizeof *stats); | |
3949 | } | |
3950 | return EPROTO; | |
3951 | } | |
3952 | ||
3953 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
3954 | * on 'netdev'. */ | |
3955 | static int | |
3956 | tc_query_class(const struct netdev *netdev, | |
3957 | unsigned int handle, unsigned int parent, | |
3958 | struct ofpbuf **replyp) | |
3959 | { | |
3960 | struct ofpbuf request; | |
3961 | struct tcmsg *tcmsg; | |
3962 | int error; | |
3963 | ||
3964 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); | |
23a98ffe BP |
3965 | if (!tcmsg) { |
3966 | return ENODEV; | |
3967 | } | |
c1c9c9c4 BP |
3968 | tcmsg->tcm_handle = handle; |
3969 | tcmsg->tcm_parent = parent; | |
3970 | ||
3971 | error = tc_transact(&request, replyp); | |
3972 | if (error) { | |
3973 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
3974 | netdev_get_name(netdev), | |
3975 | tc_get_major(handle), tc_get_minor(handle), | |
3976 | tc_get_major(parent), tc_get_minor(parent), | |
3977 | strerror(error)); | |
3978 | } | |
3979 | return error; | |
3980 | } | |
3981 | ||
3982 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
3983 | static int | |
3984 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
3985 | { | |
3986 | struct ofpbuf request; | |
3987 | struct tcmsg *tcmsg; | |
3988 | int error; | |
3989 | ||
3990 | tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); | |
23a98ffe BP |
3991 | if (!tcmsg) { |
3992 | return ENODEV; | |
3993 | } | |
c1c9c9c4 BP |
3994 | tcmsg->tcm_handle = handle; |
3995 | tcmsg->tcm_parent = 0; | |
3996 | ||
3997 | error = tc_transact(&request, NULL); | |
3998 | if (error) { | |
3999 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
4000 | netdev_get_name(netdev), | |
4001 | tc_get_major(handle), tc_get_minor(handle), | |
4002 | strerror(error)); | |
4003 | } | |
4004 | return error; | |
4005 | } | |
4006 | ||
4007 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
4008 | static int | |
4009 | tc_del_qdisc(struct netdev *netdev) | |
4010 | { | |
4011 | struct netdev_dev_linux *netdev_dev = | |
4012 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
4013 | struct ofpbuf request; | |
4014 | struct tcmsg *tcmsg; | |
4015 | int error; | |
4016 | ||
4017 | tcmsg = tc_make_request(netdev, RTM_DELQDISC, 0, &request); | |
23a98ffe BP |
4018 | if (!tcmsg) { |
4019 | return ENODEV; | |
4020 | } | |
c1c9c9c4 BP |
4021 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
4022 | tcmsg->tcm_parent = TC_H_ROOT; | |
4023 | ||
4024 | error = tc_transact(&request, NULL); | |
4025 | if (error == EINVAL) { | |
4026 | /* EINVAL probably means that the default qdisc was in use, in which | |
4027 | * case we've accomplished our purpose. */ | |
4028 | error = 0; | |
4029 | } | |
4030 | if (!error && netdev_dev->tc) { | |
4031 | if (netdev_dev->tc->ops->tc_destroy) { | |
4032 | netdev_dev->tc->ops->tc_destroy(netdev_dev->tc); | |
4033 | } | |
4034 | netdev_dev->tc = NULL; | |
4035 | } | |
4036 | return error; | |
4037 | } | |
4038 | ||
4039 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the | |
4040 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
4041 | * positive errno value. */ | |
4042 | static int | |
4043 | tc_query_qdisc(const struct netdev *netdev) | |
4044 | { | |
4045 | struct netdev_dev_linux *netdev_dev = | |
4046 | netdev_dev_linux_cast(netdev_get_dev(netdev)); | |
4047 | struct ofpbuf request, *qdisc; | |
4048 | const struct tc_ops *ops; | |
4049 | struct tcmsg *tcmsg; | |
4050 | int load_error; | |
4051 | int error; | |
4052 | ||
4053 | if (netdev_dev->tc) { | |
4054 | return 0; | |
4055 | } | |
4056 | ||
4057 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
4058 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
4059 | * 2.6.35 without that fix backported to it. | |
4060 | * | |
4061 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
4062 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
4063 | * few others. There are a few ways that I can see to do this, but most of | |
4064 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
4065 | * technique chosen here is to assume that any non-default qdisc that we | |
4066 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
4067 | * a class with handle 0:0. | |
4068 | * | |
4069 | * We could check for Linux 2.6.35+ and use a more straightforward method | |
4070 | * there. */ | |
4071 | tcmsg = tc_make_request(netdev, RTM_GETQDISC, NLM_F_ECHO, &request); | |
23a98ffe BP |
4072 | if (!tcmsg) { |
4073 | return ENODEV; | |
4074 | } | |
c1c9c9c4 BP |
4075 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
4076 | tcmsg->tcm_parent = 0; | |
4077 | ||
4078 | /* Figure out what tc class to instantiate. */ | |
4079 | error = tc_transact(&request, &qdisc); | |
4080 | if (!error) { | |
4081 | const char *kind; | |
4082 | ||
4083 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
4084 | if (error) { | |
4085 | ops = &tc_ops_other; | |
4086 | } else { | |
4087 | ops = tc_lookup_linux_name(kind); | |
4088 | if (!ops) { | |
4089 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
4090 | VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind); | |
4091 | ||
4092 | ops = &tc_ops_other; | |
4093 | } | |
4094 | } | |
4095 | } else if (error == ENOENT) { | |
4096 | /* Either it's a built-in qdisc, or it's a qdisc set up by some | |
4097 | * other entity that doesn't have a handle 1:0. We will assume | |
4098 | * that it's the system default qdisc. */ | |
4099 | ops = &tc_ops_default; | |
4100 | error = 0; | |
4101 | } else { | |
4102 | /* Who knows? Maybe the device got deleted. */ | |
4103 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
4104 | netdev_get_name(netdev), strerror(error)); | |
4105 | ops = &tc_ops_other; | |
4106 | } | |
4107 | ||
4108 | /* Instantiate it. */ | |
ebc56baa | 4109 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev), qdisc); |
c1c9c9c4 BP |
4110 | assert((load_error == 0) == (netdev_dev->tc != NULL)); |
4111 | ofpbuf_delete(qdisc); | |
4112 | ||
4113 | return error ? error : load_error; | |
4114 | } | |
4115 | ||
4116 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
4117 | approximate the time to transmit packets of various lengths. For an MTU of | |
4118 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
4119 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
4120 | possible lengths; and so on. | |
4121 | ||
4122 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
4123 | need to be shifted right to fit within such a 256-entry table. */ | |
4124 | static int | |
4125 | tc_calc_cell_log(unsigned int mtu) | |
4126 | { | |
4127 | int cell_log; | |
4128 | ||
4129 | if (!mtu) { | |
4130 | mtu = ETH_PAYLOAD_MAX; | |
4131 | } | |
4132 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
4133 | ||
4134 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
4135 | mtu >>= 1; | |
4136 | } | |
4137 | ||
4138 | return cell_log; | |
4139 | } | |
4140 | ||
4141 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
4142 | * of 'mtu'. */ | |
4143 | static void | |
4144 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
4145 | { | |
4146 | memset(rate, 0, sizeof *rate); | |
4147 | rate->cell_log = tc_calc_cell_log(mtu); | |
4148 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
4149 | /* rate->cell_align = 0; */ /* distro headers. */ | |
4150 | rate->mpu = ETH_TOTAL_MIN; | |
4151 | rate->rate = Bps; | |
4152 | } | |
4153 | ||
4154 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
4155 | * attribute of the specified "type". | |
4156 | * | |
4157 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
4158 | static void | |
4159 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
4160 | { | |
4161 | uint32_t *rtab; | |
4162 | unsigned int i; | |
4163 | ||
4164 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
4165 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
4166 | unsigned packet_size = (i + 1) << rate->cell_log; | |
4167 | if (packet_size < rate->mpu) { | |
4168 | packet_size = rate->mpu; | |
4169 | } | |
4170 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
4171 | } | |
4172 | } | |
4173 | ||
4174 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
4175 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
4176 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 4177 | * 0 is fine.) */ |
c1c9c9c4 BP |
4178 | static int |
4179 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
4180 | { | |
4181 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
4182 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
4183 | } | |
d3980822 | 4184 | \f |
aaf2fb1a BP |
4185 | /* Linux-only functions declared in netdev-linux.h */ |
4186 | ||
025e874a BP |
4187 | /* Returns a fd for an AF_INET socket or a negative errno value. */ |
4188 | int | |
4189 | netdev_linux_get_af_inet_sock(void) | |
4190 | { | |
4191 | int error = netdev_linux_init(); | |
4192 | return error ? -error : af_inet_sock; | |
4193 | } | |
4194 | ||
aaf2fb1a BP |
4195 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If |
4196 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
4197 | int | |
4198 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
4199 | const char *flag_name, bool enable) | |
4200 | { | |
4201 | const char *netdev_name = netdev_get_name(netdev); | |
4202 | struct ethtool_value evalue; | |
4203 | uint32_t new_flags; | |
4204 | int error; | |
4205 | ||
ab985a77 | 4206 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
4207 | memset(&evalue, 0, sizeof evalue); |
4208 | error = netdev_linux_do_ethtool(netdev_name, | |
4209 | (struct ethtool_cmd *)&evalue, | |
4210 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
4211 | if (error) { | |
4212 | return error; | |
4213 | } | |
4214 | ||
ab985a77 | 4215 | COVERAGE_INC(netdev_set_ethtool); |
aaf2fb1a BP |
4216 | evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
4217 | error = netdev_linux_do_ethtool(netdev_name, | |
4218 | (struct ethtool_cmd *)&evalue, | |
4219 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
4220 | if (error) { | |
4221 | return error; | |
4222 | } | |
4223 | ||
ab985a77 | 4224 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
4225 | memset(&evalue, 0, sizeof evalue); |
4226 | error = netdev_linux_do_ethtool(netdev_name, | |
4227 | (struct ethtool_cmd *)&evalue, | |
4228 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
4229 | if (error) { | |
4230 | return error; | |
4231 | } | |
4232 | ||
4233 | if (new_flags != evalue.data) { | |
4234 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
4235 | "device %s failed", enable ? "enable" : "disable", | |
4236 | flag_name, netdev_name); | |
4237 | return EOPNOTSUPP; | |
4238 | } | |
4239 | ||
4240 | return 0; | |
4241 | } | |
4242 | \f | |
4243 | /* Utility functions. */ | |
4244 | ||
d3980822 | 4245 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 4246 | static void |
d3980822 BP |
4247 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
4248 | const struct rtnl_link_stats *src) | |
4249 | { | |
f613a0d7 PS |
4250 | dst->rx_packets = src->rx_packets; |
4251 | dst->tx_packets = src->tx_packets; | |
4252 | dst->rx_bytes = src->rx_bytes; | |
4253 | dst->tx_bytes = src->tx_bytes; | |
4254 | dst->rx_errors = src->rx_errors; | |
4255 | dst->tx_errors = src->tx_errors; | |
4256 | dst->rx_dropped = src->rx_dropped; | |
4257 | dst->tx_dropped = src->tx_dropped; | |
4258 | dst->multicast = src->multicast; | |
4259 | dst->collisions = src->collisions; | |
4260 | dst->rx_length_errors = src->rx_length_errors; | |
4261 | dst->rx_over_errors = src->rx_over_errors; | |
4262 | dst->rx_crc_errors = src->rx_crc_errors; | |
4263 | dst->rx_frame_errors = src->rx_frame_errors; | |
4264 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
4265 | dst->rx_missed_errors = src->rx_missed_errors; | |
4266 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
4267 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
4268 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
4269 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
4270 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
4271 | } |
4272 | ||
c1c9c9c4 BP |
4273 | static int |
4274 | get_stats_via_netlink(int ifindex, struct netdev_stats *stats) | |
4275 | { | |
4276 | /* Policy for RTNLGRP_LINK messages. | |
4277 | * | |
4278 | * There are *many* more fields in these messages, but currently we only | |
4279 | * care about these fields. */ | |
4280 | static const struct nl_policy rtnlgrp_link_policy[] = { | |
4281 | [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, | |
4282 | [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, | |
4283 | .min_len = sizeof(struct rtnl_link_stats) }, | |
4284 | }; | |
4285 | ||
4286 | struct ofpbuf request; | |
4287 | struct ofpbuf *reply; | |
4288 | struct ifinfomsg *ifi; | |
c1c9c9c4 BP |
4289 | struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; |
4290 | int error; | |
4291 | ||
4292 | ofpbuf_init(&request, 0); | |
4293 | nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST); | |
4294 | ifi = ofpbuf_put_zeros(&request, sizeof *ifi); | |
4295 | ifi->ifi_family = PF_UNSPEC; | |
4296 | ifi->ifi_index = ifindex; | |
4297 | error = nl_sock_transact(rtnl_sock, &request, &reply); | |
4298 | ofpbuf_uninit(&request); | |
4299 | if (error) { | |
4300 | return error; | |
4301 | } | |
4302 | ||
4303 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), | |
4304 | rtnlgrp_link_policy, | |
4305 | attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { | |
4306 | ofpbuf_delete(reply); | |
4307 | return EPROTO; | |
4308 | } | |
4309 | ||
4310 | if (!attrs[IFLA_STATS]) { | |
4311 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
4312 | ofpbuf_delete(reply); | |
4313 | return EPROTO; | |
4314 | } | |
8b61709d | 4315 | |
d3980822 | 4316 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS])); |
8b61709d | 4317 | |
576e26d7 BP |
4318 | ofpbuf_delete(reply); |
4319 | ||
8b61709d BP |
4320 | return 0; |
4321 | } | |
4322 | ||
4323 | static int | |
4324 | get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) | |
4325 | { | |
4326 | static const char fn[] = "/proc/net/dev"; | |
4327 | char line[1024]; | |
4328 | FILE *stream; | |
4329 | int ln; | |
4330 | ||
4331 | stream = fopen(fn, "r"); | |
4332 | if (!stream) { | |
4333 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, strerror(errno)); | |
4334 | return errno; | |
4335 | } | |
4336 | ||
4337 | ln = 0; | |
4338 | while (fgets(line, sizeof line, stream)) { | |
4339 | if (++ln >= 3) { | |
4340 | char devname[16]; | |
4341 | #define X64 "%"SCNu64 | |
4342 | if (sscanf(line, | |
4343 | " %15[^:]:" | |
4344 | X64 X64 X64 X64 X64 X64 X64 "%*u" | |
4345 | X64 X64 X64 X64 X64 X64 X64 "%*u", | |
4346 | devname, | |
4347 | &stats->rx_bytes, | |
4348 | &stats->rx_packets, | |
4349 | &stats->rx_errors, | |
4350 | &stats->rx_dropped, | |
4351 | &stats->rx_fifo_errors, | |
4352 | &stats->rx_frame_errors, | |
4353 | &stats->multicast, | |
4354 | &stats->tx_bytes, | |
4355 | &stats->tx_packets, | |
4356 | &stats->tx_errors, | |
4357 | &stats->tx_dropped, | |
4358 | &stats->tx_fifo_errors, | |
4359 | &stats->collisions, | |
4360 | &stats->tx_carrier_errors) != 15) { | |
4361 | VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); | |
4362 | } else if (!strcmp(devname, netdev_name)) { | |
4363 | stats->rx_length_errors = UINT64_MAX; | |
4364 | stats->rx_over_errors = UINT64_MAX; | |
4365 | stats->rx_crc_errors = UINT64_MAX; | |
4366 | stats->rx_missed_errors = UINT64_MAX; | |
4367 | stats->tx_aborted_errors = UINT64_MAX; | |
4368 | stats->tx_heartbeat_errors = UINT64_MAX; | |
4369 | stats->tx_window_errors = UINT64_MAX; | |
4370 | fclose(stream); | |
4371 | return 0; | |
4372 | } | |
4373 | } | |
4374 | } | |
4375 | VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); | |
4376 | fclose(stream); | |
4377 | return ENODEV; | |
4378 | } | |
c1c9c9c4 | 4379 | |
3a183124 | 4380 | static int |
059e5f4f | 4381 | get_flags(const struct netdev_dev *dev, unsigned int *flags) |
8b61709d BP |
4382 | { |
4383 | struct ifreq ifr; | |
4384 | int error; | |
4385 | ||
755be9ea EJ |
4386 | *flags = 0; |
4387 | error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS, | |
149f577a | 4388 | "SIOCGIFFLAGS"); |
755be9ea EJ |
4389 | if (!error) { |
4390 | *flags = ifr.ifr_flags; | |
4391 | } | |
8b61709d BP |
4392 | return error; |
4393 | } | |
4394 | ||
4395 | static int | |
059e5f4f | 4396 | set_flags(struct netdev *netdev, unsigned int flags) |
8b61709d BP |
4397 | { |
4398 | struct ifreq ifr; | |
4399 | ||
4400 | ifr.ifr_flags = flags; | |
149f577a JG |
4401 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, SIOCSIFFLAGS, |
4402 | "SIOCSIFFLAGS"); | |
8b61709d BP |
4403 | } |
4404 | ||
4405 | static int | |
4406 | do_get_ifindex(const char *netdev_name) | |
4407 | { | |
4408 | struct ifreq ifr; | |
4409 | ||
71d7c22f | 4410 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4411 | COVERAGE_INC(netdev_get_ifindex); |
4412 | if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { | |
4413 | VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
4414 | netdev_name, strerror(errno)); | |
4415 | return -errno; | |
4416 | } | |
4417 | return ifr.ifr_ifindex; | |
4418 | } | |
4419 | ||
4420 | static int | |
4421 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
4422 | { | |
149f577a JG |
4423 | struct netdev_dev_linux *netdev_dev = |
4424 | netdev_dev_linux_cast(netdev_get_dev(netdev_)); | |
c7b1b0a5 | 4425 | |
149f577a | 4426 | if (!(netdev_dev->cache_valid & VALID_IFINDEX)) { |
8b61709d | 4427 | int ifindex = do_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 4428 | |
8b61709d | 4429 | if (ifindex < 0) { |
c7b1b0a5 PS |
4430 | netdev_dev->get_ifindex_error = -ifindex; |
4431 | netdev_dev->ifindex = 0; | |
4432 | } else { | |
4433 | netdev_dev->get_ifindex_error = 0; | |
4434 | netdev_dev->ifindex = ifindex; | |
8b61709d | 4435 | } |
149f577a | 4436 | netdev_dev->cache_valid |= VALID_IFINDEX; |
8b61709d | 4437 | } |
c7b1b0a5 | 4438 | |
149f577a | 4439 | *ifindexp = netdev_dev->ifindex; |
c7b1b0a5 | 4440 | return netdev_dev->get_ifindex_error; |
8b61709d BP |
4441 | } |
4442 | ||
4443 | static int | |
4444 | get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) | |
4445 | { | |
4446 | struct ifreq ifr; | |
4447 | int hwaddr_family; | |
4448 | ||
4449 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4450 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4451 | COVERAGE_INC(netdev_get_hwaddr); |
4452 | if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { | |
78857dfb BP |
4453 | /* ENODEV probably means that a vif disappeared asynchronously and |
4454 | * hasn't been removed from the database yet, so reduce the log level | |
4455 | * to INFO for that case. */ | |
4456 | VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR, | |
4457 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", | |
4458 | netdev_name, strerror(errno)); | |
8b61709d BP |
4459 | return errno; |
4460 | } | |
4461 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
4462 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { | |
4463 | VLOG_WARN("%s device has unknown hardware address family %d", | |
4464 | netdev_name, hwaddr_family); | |
4465 | } | |
4466 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
4467 | return 0; | |
4468 | } | |
4469 | ||
4470 | static int | |
44445cac | 4471 | set_etheraddr(const char *netdev_name, |
8b61709d BP |
4472 | const uint8_t mac[ETH_ADDR_LEN]) |
4473 | { | |
4474 | struct ifreq ifr; | |
4475 | ||
4476 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4477 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 4478 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
8b61709d BP |
4479 | memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); |
4480 | COVERAGE_INC(netdev_set_hwaddr); | |
4481 | if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { | |
4482 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", | |
4483 | netdev_name, strerror(errno)); | |
4484 | return errno; | |
4485 | } | |
4486 | return 0; | |
4487 | } | |
4488 | ||
4489 | static int | |
0b0544d7 | 4490 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
4491 | int cmd, const char *cmd_name) |
4492 | { | |
4493 | struct ifreq ifr; | |
4494 | ||
4495 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4496 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
4497 | ifr.ifr_data = (caddr_t) ecmd; |
4498 | ||
4499 | ecmd->cmd = cmd; | |
8b61709d BP |
4500 | if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { |
4501 | return 0; | |
4502 | } else { | |
4503 | if (errno != EOPNOTSUPP) { | |
4504 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " | |
0b0544d7 | 4505 | "failed: %s", cmd_name, name, strerror(errno)); |
8b61709d BP |
4506 | } else { |
4507 | /* The device doesn't support this operation. That's pretty | |
4508 | * common, so there's no point in logging anything. */ | |
4509 | } | |
4510 | return errno; | |
4511 | } | |
4512 | } | |
4513 | ||
4514 | static int | |
149f577a JG |
4515 | netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, |
4516 | const char *cmd_name) | |
8b61709d | 4517 | { |
71d7c22f | 4518 | ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name); |
8b61709d | 4519 | if (ioctl(af_inet_sock, cmd, ifr) == -1) { |
149f577a JG |
4520 | VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, |
4521 | strerror(errno)); | |
8b61709d BP |
4522 | return errno; |
4523 | } | |
4524 | return 0; | |
4525 | } | |
f1acd62b BP |
4526 | |
4527 | static int | |
4528 | netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, | |
4529 | int cmd, const char *cmd_name) | |
4530 | { | |
4531 | struct ifreq ifr; | |
4532 | int error; | |
4533 | ||
4534 | ifr.ifr_addr.sa_family = AF_INET; | |
149f577a | 4535 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); |
f1acd62b BP |
4536 | if (!error) { |
4537 | const struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr; | |
4538 | *ip = sin->sin_addr; | |
4539 | } | |
4540 | return error; | |
4541 | } | |
488d734d BP |
4542 | |
4543 | /* Returns an AF_PACKET raw socket or a negative errno value. */ | |
4544 | static int | |
4545 | af_packet_sock(void) | |
4546 | { | |
4547 | static int sock = INT_MIN; | |
4548 | ||
4549 | if (sock == INT_MIN) { | |
4550 | sock = socket(AF_PACKET, SOCK_RAW, 0); | |
4551 | if (sock >= 0) { | |
4552 | set_nonblocking(sock); | |
4553 | } else { | |
4554 | sock = -errno; | |
4555 | VLOG_ERR("failed to create packet socket: %s", strerror(errno)); | |
4556 | } | |
4557 | } | |
4558 | ||
4559 | return sock; | |
4560 | } |