]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
275707c3 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013 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 | ||
e9e28be3 | 21 | #include <errno.h> |
8b61709d BP |
22 | #include <fcntl.h> |
23 | #include <arpa/inet.h> | |
24 | #include <inttypes.h> | |
32383c3b | 25 | #include <linux/filter.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" |
45c8d3a1 | 59 | #include "netlink-notifier.h" |
2fe27d5a | 60 | #include "netlink-socket.h" |
c060c4cf | 61 | #include "netlink.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 | 67 | #include "shash.h" |
c060c4cf | 68 | #include "socket-util.h" |
19993ef3 | 69 | #include "sset.h" |
1670c579 | 70 | #include "timer.h" |
c060c4cf | 71 | #include "unaligned.h" |
e9e28be3 | 72 | #include "vlog.h" |
5136ce49 | 73 | |
d98e6007 | 74 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea | 75 | |
d76f09ea BP |
76 | COVERAGE_DEFINE(netdev_set_policing); |
77 | COVERAGE_DEFINE(netdev_arp_lookup); | |
78 | COVERAGE_DEFINE(netdev_get_ifindex); | |
79 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
80 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
ab985a77 BP |
81 | COVERAGE_DEFINE(netdev_get_ethtool); |
82 | COVERAGE_DEFINE(netdev_set_ethtool); | |
4f925bd3 | 83 | |
8b61709d BP |
84 | \f |
85 | /* These were introduced in Linux 2.6.14, so they might be missing if we have | |
86 | * old headers. */ | |
87 | #ifndef ADVERTISED_Pause | |
88 | #define ADVERTISED_Pause (1 << 13) | |
89 | #endif | |
90 | #ifndef ADVERTISED_Asym_Pause | |
91 | #define ADVERTISED_Asym_Pause (1 << 14) | |
92 | #endif | |
93 | ||
e47bd51a JP |
94 | /* These were introduced in Linux 2.6.24, so they might be missing if we |
95 | * have old headers. */ | |
96 | #ifndef ETHTOOL_GFLAGS | |
97 | #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ | |
98 | #endif | |
99 | #ifndef ETHTOOL_SFLAGS | |
100 | #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ | |
101 | #endif | |
102 | ||
c1c9c9c4 BP |
103 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
104 | * headers. */ | |
105 | #ifndef TC_RTAB_SIZE | |
106 | #define TC_RTAB_SIZE 1024 | |
107 | #endif | |
108 | ||
2ee6545f | 109 | static struct nln_notifier *netdev_linux_cache_notifier = NULL; |
46415c90 | 110 | static int cache_notifier_refcount; |
8b61709d BP |
111 | |
112 | enum { | |
7fbef77a JG |
113 | VALID_IFINDEX = 1 << 0, |
114 | VALID_ETHERADDR = 1 << 1, | |
115 | VALID_IN4 = 1 << 2, | |
116 | VALID_IN6 = 1 << 3, | |
117 | VALID_MTU = 1 << 4, | |
3a183124 | 118 | VALID_POLICING = 1 << 5, |
4f925bd3 PS |
119 | VALID_VPORT_STAT_ERROR = 1 << 6, |
120 | VALID_DRVINFO = 1 << 7, | |
51f87458 | 121 | VALID_FEATURES = 1 << 8, |
8b61709d BP |
122 | }; |
123 | ||
149f577a JG |
124 | struct tap_state { |
125 | int fd; | |
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 | |
559eb230 BP |
142 | #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) } |
143 | ||
93b13be8 BP |
144 | /* One traffic control queue. |
145 | * | |
146 | * Each TC implementation subclasses this with whatever additional data it | |
147 | * needs. */ | |
148 | struct tc_queue { | |
149 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
150 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
6dc34a0d | 151 | long long int created; /* Time queue was created, in msecs. */ |
c1c9c9c4 BP |
152 | }; |
153 | ||
154 | /* A particular kind of traffic control. Each implementation generally maps to | |
155 | * one particular Linux qdisc class. | |
156 | * | |
157 | * The functions below return 0 if successful or a positive errno value on | |
158 | * failure, except where otherwise noted. All of them must be provided, except | |
159 | * where otherwise noted. */ | |
160 | struct tc_ops { | |
161 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
162 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
163 | * appropriate values. */ | |
164 | const char *linux_name; | |
165 | ||
166 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
167 | const char *ovs_name; | |
168 | ||
169 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
170 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
171 | unsigned int n_queues; | |
172 | ||
173 | /* Called to install this TC class on 'netdev'. The implementation should | |
174 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
175 | * and configure it according to 'details'. The implementation may assume | |
176 | * that the current qdisc is the default; that is, there is no need for it | |
177 | * to delete the current qdisc before installing itself. | |
178 | * | |
179 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
180 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
181 | * (which is built as ovs-vswitchd.conf.db(8)). | |
182 | * | |
183 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
184 | * initialized 'struct tc'. | |
185 | * | |
186 | * (This function is null for tc_ops_other, which cannot be installed. For | |
187 | * other TC classes it should always be nonnull.) */ | |
79f1cbe9 | 188 | int (*tc_install)(struct netdev *netdev, const struct smap *details); |
c1c9c9c4 BP |
189 | |
190 | /* Called when the netdev code determines (through a Netlink query) that | |
191 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
192 | * it ourselves and so don't know any of the details. | |
193 | * | |
194 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
195 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
196 | * implementation should parse the other attributes of 'nlmsg' as | |
197 | * necessary to determine its configuration. If necessary it should also | |
198 | * use Netlink queries to determine the configuration of queues on | |
199 | * 'netdev'. | |
200 | * | |
201 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
202 | * initialized 'struct tc'. */ | |
203 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
204 | ||
205 | /* Destroys the data structures allocated by the implementation as part of | |
206 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
207 | * tc_destroy(tc). | |
208 | * | |
209 | * The implementation should not need to perform any Netlink calls. If | |
210 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
211 | * (But it may not be desirable.) | |
212 | * | |
213 | * This function may be null if 'tc' is trivial. */ | |
214 | void (*tc_destroy)(struct tc *tc); | |
215 | ||
216 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
217 | * | |
218 | * The implementation should not need to perform any Netlink calls, because | |
219 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
220 | * cached the configuration. | |
221 | * | |
222 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
223 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
224 | * (which is built as ovs-vswitchd.conf.db(8)). | |
225 | * | |
226 | * This function may be null if 'tc' is not configurable. | |
227 | */ | |
79f1cbe9 | 228 | int (*qdisc_get)(const struct netdev *netdev, struct smap *details); |
c1c9c9c4 BP |
229 | |
230 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
231 | * required Netlink calls to complete the reconfiguration. | |
232 | * | |
233 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
234 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
235 | * (which is built as ovs-vswitchd.conf.db(8)). | |
236 | * | |
237 | * This function may be null if 'tc' is not configurable. | |
238 | */ | |
79f1cbe9 | 239 | int (*qdisc_set)(struct netdev *, const struct smap *details); |
c1c9c9c4 | 240 | |
93b13be8 BP |
241 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
242 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
243 | * |
244 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
245 | * in the "other_config" column in the "Queue" table in | |
246 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
247 | * | |
248 | * The implementation should not need to perform any Netlink calls, because | |
249 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
250 | * cached the queue configuration. | |
251 | * | |
252 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
253 | * 0). */ | |
93b13be8 | 254 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
79f1cbe9 | 255 | struct smap *details); |
c1c9c9c4 BP |
256 | |
257 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
258 | * 'details', perfoming any required Netlink calls to complete the | |
259 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
260 | * 'n_queues'. | |
261 | * | |
262 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
263 | * in the "other_config" column in the "Queue" table in | |
264 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
265 | * | |
266 | * This function may be null if 'tc' does not have queues or its queues are | |
267 | * not configurable. */ | |
268 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
79f1cbe9 | 269 | const struct smap *details); |
c1c9c9c4 | 270 | |
93b13be8 BP |
271 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
272 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
273 | * |
274 | * This function may be null if 'tc' does not have queues or its queues | |
275 | * cannot be deleted. */ | |
93b13be8 | 276 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 277 | |
93b13be8 BP |
278 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
279 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
280 | * |
281 | * On success, initializes '*stats'. | |
282 | * | |
283 | * This function may be null if 'tc' does not have queues or if it cannot | |
284 | * report queue statistics. */ | |
93b13be8 BP |
285 | int (*class_get_stats)(const struct netdev *netdev, |
286 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
287 | struct netdev_queue_stats *stats); |
288 | ||
289 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
290 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
291 | * | |
292 | * This function may be null if 'tc' does not have queues or if it cannot | |
293 | * report queue statistics. */ | |
294 | int (*class_dump_stats)(const struct netdev *netdev, | |
295 | const struct ofpbuf *nlmsg, | |
296 | netdev_dump_queue_stats_cb *cb, void *aux); | |
297 | }; | |
298 | ||
299 | static void | |
300 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
301 | { | |
302 | tc->ops = ops; | |
93b13be8 | 303 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
304 | } |
305 | ||
306 | static void | |
307 | tc_destroy(struct tc *tc) | |
308 | { | |
93b13be8 | 309 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
310 | } |
311 | ||
312 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 313 | static const struct tc_ops tc_ops_hfsc; |
c1c9c9c4 BP |
314 | static const struct tc_ops tc_ops_default; |
315 | static const struct tc_ops tc_ops_other; | |
316 | ||
559eb230 | 317 | static const struct tc_ops *const tcs[] = { |
c1c9c9c4 | 318 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ |
a339aa81 | 319 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
c1c9c9c4 BP |
320 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
321 | &tc_ops_other, /* Some other qdisc. */ | |
322 | NULL | |
323 | }; | |
149f577a | 324 | |
c1c9c9c4 BP |
325 | static unsigned int tc_make_handle(unsigned int major, unsigned int minor); |
326 | static unsigned int tc_get_major(unsigned int handle); | |
327 | static unsigned int tc_get_minor(unsigned int handle); | |
328 | ||
329 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); | |
330 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
331 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
332 | ||
333 | static struct tcmsg *tc_make_request(const struct netdev *, int type, | |
334 | unsigned int flags, struct ofpbuf *); | |
335 | static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); | |
f8500004 JP |
336 | static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add); |
337 | static int tc_add_policer(struct netdev *netdev, int kbits_rate, | |
338 | int kbits_burst); | |
c1c9c9c4 BP |
339 | |
340 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
341 | struct nlattr **options); | |
342 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
343 | struct nlattr **options, | |
344 | struct netdev_queue_stats *); | |
345 | static int tc_query_class(const struct netdev *, | |
346 | unsigned int handle, unsigned int parent, | |
347 | struct ofpbuf **replyp); | |
348 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
349 | ||
350 | static int tc_del_qdisc(struct netdev *netdev); | |
351 | static int tc_query_qdisc(const struct netdev *netdev); | |
352 | ||
353 | static int tc_calc_cell_log(unsigned int mtu); | |
354 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
355 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
356 | const struct tc_ratespec *rate); | |
357 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
358 | \f | |
b5d57fc8 BP |
359 | struct netdev_linux { |
360 | struct netdev up; | |
149f577a | 361 | |
8b61709d | 362 | struct shash_node *shash_node; |
149f577a | 363 | unsigned int cache_valid; |
ac4d3bcb | 364 | unsigned int change_seq; |
8b61709d | 365 | |
1670c579 EJ |
366 | bool miimon; /* Link status of last poll. */ |
367 | long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ | |
368 | struct timer miimon_timer; | |
369 | ||
8722022c BP |
370 | /* The following are figured out "on demand" only. They are only valid |
371 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d BP |
372 | int ifindex; |
373 | uint8_t etheraddr[ETH_ADDR_LEN]; | |
f1acd62b | 374 | struct in_addr address, netmask; |
8b61709d BP |
375 | struct in6_addr in6; |
376 | int mtu; | |
059e5f4f | 377 | unsigned int ifi_flags; |
65c3058c | 378 | long long int carrier_resets; |
80a86fbe BP |
379 | uint32_t kbits_rate; /* Policing data. */ |
380 | uint32_t kbits_burst; | |
bba1e6f3 PS |
381 | int vport_stats_error; /* Cached error code from vport_get_stats(). |
382 | 0 or an errno value. */ | |
90a6637d | 383 | int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ |
44445cac | 384 | int ether_addr_error; /* Cached error code from set/get etheraddr. */ |
c9f71668 | 385 | int netdev_policing_error; /* Cached error code from set policing. */ |
51f87458 | 386 | int get_features_error; /* Cached error code from ETHTOOL_GSET. */ |
c7b1b0a5 | 387 | int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */ |
51f87458 | 388 | |
a00ca915 EJ |
389 | enum netdev_features current; /* Cached from ETHTOOL_GSET. */ |
390 | enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ | |
391 | enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ | |
392 | enum netdev_features peer; /* Cached from ETHTOOL_GSET. */ | |
90a6637d | 393 | |
4f925bd3 | 394 | struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ |
c1c9c9c4 | 395 | struct tc *tc; |
149f577a JG |
396 | |
397 | union { | |
398 | struct tap_state tap; | |
399 | } state; | |
8b61709d BP |
400 | }; |
401 | ||
796223f5 BP |
402 | struct netdev_rx_linux { |
403 | struct netdev_rx up; | |
404 | bool is_tap; | |
5b7448ed | 405 | int fd; |
149f577a | 406 | }; |
8b61709d | 407 | |
796223f5 BP |
408 | static const struct netdev_rx_class netdev_rx_linux_class; |
409 | ||
76c308b5 BP |
410 | /* Sockets used for ioctl operations. */ |
411 | static int af_inet_sock = -1; /* AF_INET, SOCK_DGRAM. */ | |
8b61709d | 412 | |
8b61709d BP |
413 | /* This is set pretty low because we probably won't learn anything from the |
414 | * additional log messages. */ | |
415 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
416 | ||
15b3596a | 417 | static int netdev_linux_init(void); |
6f643e49 | 418 | |
0b0544d7 | 419 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 420 | int cmd, const char *cmd_name); |
149f577a JG |
421 | static int netdev_linux_do_ioctl(const char *name, struct ifreq *, int cmd, |
422 | const char *cmd_name); | |
f1acd62b BP |
423 | static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, |
424 | int cmd, const char *cmd_name); | |
b5d57fc8 | 425 | static int get_flags(const struct netdev *, unsigned int *flags); |
4b609110 | 426 | static int set_flags(const char *, unsigned int flags); |
8b61709d BP |
427 | static int do_get_ifindex(const char *netdev_name); |
428 | static int get_ifindex(const struct netdev *, int *ifindexp); | |
429 | static int do_set_addr(struct netdev *netdev, | |
430 | int ioctl_nr, const char *ioctl_name, | |
431 | struct in_addr addr); | |
432 | static int get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]); | |
44445cac | 433 | static int set_etheraddr(const char *netdev_name, const uint8_t[ETH_ADDR_LEN]); |
8b61709d BP |
434 | static int get_stats_via_netlink(int ifindex, struct netdev_stats *stats); |
435 | static int get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats); | |
488d734d | 436 | static int af_packet_sock(void); |
1670c579 EJ |
437 | static void netdev_linux_miimon_run(void); |
438 | static void netdev_linux_miimon_wait(void); | |
8b61709d | 439 | |
15b3596a JG |
440 | static bool |
441 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
442 | { | |
443 | return netdev_class->init == netdev_linux_init; | |
444 | } | |
445 | ||
796223f5 BP |
446 | static bool |
447 | is_tap_netdev(const struct netdev *netdev) | |
448 | { | |
b5d57fc8 | 449 | return netdev_get_class(netdev) == &netdev_tap_class; |
6c88d577 JP |
450 | } |
451 | ||
8b61709d BP |
452 | static struct netdev_linux * |
453 | netdev_linux_cast(const struct netdev *netdev) | |
454 | { | |
b5d57fc8 | 455 | ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); |
15b3596a | 456 | |
180c6d0b | 457 | return CONTAINER_OF(netdev, struct netdev_linux, up); |
8b61709d | 458 | } |
796223f5 BP |
459 | |
460 | static struct netdev_rx_linux * | |
461 | netdev_rx_linux_cast(const struct netdev_rx *rx) | |
462 | { | |
463 | netdev_rx_assert_class(rx, &netdev_rx_linux_class); | |
464 | return CONTAINER_OF(rx, struct netdev_rx_linux, up); | |
465 | } | |
ff4ed3c9 | 466 | \f |
8b61709d BP |
467 | static int |
468 | netdev_linux_init(void) | |
469 | { | |
470 | static int status = -1; | |
471 | if (status < 0) { | |
ff4ed3c9 | 472 | /* Create AF_INET socket. */ |
8b61709d BP |
473 | af_inet_sock = socket(AF_INET, SOCK_DGRAM, 0); |
474 | status = af_inet_sock >= 0 ? 0 : errno; | |
475 | if (status) { | |
10a89ef0 | 476 | VLOG_ERR("failed to create inet socket: %s", ovs_strerror(status)); |
8b61709d BP |
477 | } |
478 | } | |
479 | return status; | |
480 | } | |
481 | ||
482 | static void | |
483 | netdev_linux_run(void) | |
484 | { | |
18a23781 | 485 | rtnetlink_link_run(); |
1670c579 | 486 | netdev_linux_miimon_run(); |
8b61709d BP |
487 | } |
488 | ||
489 | static void | |
490 | netdev_linux_wait(void) | |
491 | { | |
18a23781 | 492 | rtnetlink_link_wait(); |
1670c579 | 493 | netdev_linux_miimon_wait(); |
8b61709d BP |
494 | } |
495 | ||
ac4d3bcb | 496 | static void |
b5d57fc8 BP |
497 | netdev_linux_changed(struct netdev_linux *dev, |
498 | unsigned int ifi_flags, unsigned int mask) | |
ac4d3bcb EJ |
499 | { |
500 | dev->change_seq++; | |
501 | if (!dev->change_seq) { | |
502 | dev->change_seq++; | |
503 | } | |
8aa77183 BP |
504 | |
505 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
506 | dev->carrier_resets++; | |
507 | } | |
508 | dev->ifi_flags = ifi_flags; | |
509 | ||
4f925bd3 PS |
510 | dev->cache_valid &= mask; |
511 | } | |
512 | ||
513 | static void | |
b5d57fc8 BP |
514 | netdev_linux_update(struct netdev_linux *dev, |
515 | const struct rtnetlink_link_change *change) | |
4f925bd3 PS |
516 | { |
517 | if (change->nlmsg_type == RTM_NEWLINK) { | |
518 | /* Keep drv-info */ | |
b5d57fc8 | 519 | netdev_linux_changed(dev, change->ifi_flags, VALID_DRVINFO); |
90a6637d | 520 | |
c7b1b0a5 | 521 | /* Update netdev from rtnl-change msg. */ |
90a6637d PS |
522 | if (change->mtu) { |
523 | dev->mtu = change->mtu; | |
524 | dev->cache_valid |= VALID_MTU; | |
525 | dev->netdev_mtu_error = 0; | |
526 | } | |
527 | ||
44445cac PS |
528 | if (!eth_addr_is_zero(change->addr)) { |
529 | memcpy(dev->etheraddr, change->addr, ETH_ADDR_LEN); | |
530 | dev->cache_valid |= VALID_ETHERADDR; | |
531 | dev->ether_addr_error = 0; | |
532 | } | |
533 | ||
c7b1b0a5 PS |
534 | dev->ifindex = change->ifi_index; |
535 | dev->cache_valid |= VALID_IFINDEX; | |
536 | dev->get_ifindex_error = 0; | |
537 | ||
4f925bd3 | 538 | } else { |
b5d57fc8 | 539 | netdev_linux_changed(dev, change->ifi_flags, 0); |
4f925bd3 | 540 | } |
ac4d3bcb EJ |
541 | } |
542 | ||
8b61709d | 543 | static void |
21d6e22e | 544 | netdev_linux_cache_cb(const struct rtnetlink_link_change *change, |
67a4917b | 545 | void *aux OVS_UNUSED) |
8b61709d | 546 | { |
b5d57fc8 | 547 | struct netdev_linux *dev; |
8b61709d | 548 | if (change) { |
b5d57fc8 BP |
549 | struct netdev *base_dev = netdev_from_name(change->ifname); |
550 | if (base_dev && is_netdev_linux_class(netdev_get_class(base_dev))) { | |
551 | netdev_linux_update(netdev_linux_cast(base_dev), change); | |
8b61709d BP |
552 | } |
553 | } else { | |
46415c90 | 554 | struct shash device_shash; |
8b61709d | 555 | struct shash_node *node; |
46415c90 JG |
556 | |
557 | shash_init(&device_shash); | |
b5d57fc8 | 558 | netdev_get_devices(&netdev_linux_class, &device_shash); |
46415c90 | 559 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa | 560 | struct netdev *netdev = node->data; |
059e5f4f | 561 | unsigned int flags; |
3a183124 | 562 | |
96172faa | 563 | dev = netdev_linux_cast(netdev); |
3a183124 | 564 | |
180c6d0b | 565 | get_flags(&dev->up, &flags); |
b5d57fc8 | 566 | netdev_linux_changed(dev, flags, 0); |
8b61709d | 567 | } |
46415c90 | 568 | shash_destroy(&device_shash); |
8b61709d BP |
569 | } |
570 | } | |
571 | ||
572 | static int | |
1f6e0fbd | 573 | cache_notifier_ref(void) |
6c88d577 | 574 | { |
46415c90 | 575 | if (!cache_notifier_refcount) { |
cb22974d | 576 | ovs_assert(!netdev_linux_cache_notifier); |
2ee6545f EJ |
577 | |
578 | netdev_linux_cache_notifier = | |
579 | rtnetlink_link_notifier_create(netdev_linux_cache_cb, NULL); | |
580 | ||
581 | if (!netdev_linux_cache_notifier) { | |
582 | return EINVAL; | |
149f577a JG |
583 | } |
584 | } | |
46415c90 | 585 | cache_notifier_refcount++; |
6c88d577 | 586 | |
1f6e0fbd BP |
587 | return 0; |
588 | } | |
589 | ||
590 | static void | |
591 | cache_notifier_unref(void) | |
592 | { | |
cb22974d | 593 | ovs_assert(cache_notifier_refcount > 0); |
1f6e0fbd | 594 | if (!--cache_notifier_refcount) { |
cb22974d | 595 | ovs_assert(netdev_linux_cache_notifier); |
1f6e0fbd BP |
596 | rtnetlink_link_notifier_destroy(netdev_linux_cache_notifier); |
597 | netdev_linux_cache_notifier = NULL; | |
598 | } | |
599 | } | |
600 | ||
601 | /* Creates system and internal devices. */ | |
602 | static int | |
603 | netdev_linux_create(const struct netdev_class *class, const char *name, | |
b5d57fc8 | 604 | struct netdev **netdevp) |
1f6e0fbd | 605 | { |
b5d57fc8 | 606 | struct netdev_linux *netdev; |
1f6e0fbd BP |
607 | int error; |
608 | ||
609 | error = cache_notifier_ref(); | |
610 | if (error) { | |
611 | return error; | |
612 | } | |
613 | ||
b5d57fc8 BP |
614 | netdev = xzalloc(sizeof *netdev); |
615 | netdev->change_seq = 1; | |
616 | netdev_init(&netdev->up, name, class); | |
617 | error = get_flags(&netdev->up, &netdev->ifi_flags); | |
618 | if (error == ENODEV) { | |
619 | if (class != &netdev_internal_class) { | |
620 | /* The device does not exist, so don't allow it to be opened. */ | |
621 | netdev_uninit(&netdev->up, false); | |
622 | cache_notifier_unref(); | |
623 | free(netdev); | |
624 | return ENODEV; | |
625 | } else { | |
626 | /* "Internal" netdevs have to be created as netdev objects before | |
627 | * they exist in the kernel, because creating them in the kernel | |
628 | * happens by passing a netdev object to dpif_port_add(). | |
629 | * Therefore, ignore the error. */ | |
630 | } | |
631 | } | |
46415c90 | 632 | |
b5d57fc8 | 633 | *netdevp = &netdev->up; |
a740f0de JG |
634 | return 0; |
635 | } | |
636 | ||
5b7448ed JG |
637 | /* For most types of netdevs we open the device for each call of |
638 | * netdev_open(). However, this is not the case with tap devices, | |
639 | * since it is only possible to open the device once. In this | |
640 | * situation we share a single file descriptor, and consequently | |
641 | * buffers, across all readers. Therefore once data is read it will | |
642 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 643 | static int |
b8dcf5e9 | 644 | netdev_linux_create_tap(const struct netdev_class *class OVS_UNUSED, |
b5d57fc8 | 645 | const char *name, struct netdev **netdevp) |
a740f0de | 646 | { |
b5d57fc8 | 647 | struct netdev_linux *netdev; |
a740f0de JG |
648 | struct tap_state *state; |
649 | static const char tap_dev[] = "/dev/net/tun"; | |
650 | struct ifreq ifr; | |
651 | int error; | |
652 | ||
b5d57fc8 | 653 | netdev = xzalloc(sizeof *netdev); |
2e5ae318 | 654 | netdev->change_seq = 1; |
b5d57fc8 | 655 | state = &netdev->state.tap; |
a740f0de | 656 | |
1f6e0fbd BP |
657 | error = cache_notifier_ref(); |
658 | if (error) { | |
659 | goto error; | |
660 | } | |
661 | ||
6c88d577 | 662 | /* Open tap device. */ |
149f577a JG |
663 | state->fd = open(tap_dev, O_RDWR); |
664 | if (state->fd < 0) { | |
6c88d577 | 665 | error = errno; |
10a89ef0 | 666 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); |
1f6e0fbd | 667 | goto error_unref_notifier; |
6c88d577 JP |
668 | } |
669 | ||
670 | /* Create tap device. */ | |
671 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
71d7c22f | 672 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
149f577a | 673 | if (ioctl(state->fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 | 674 | VLOG_WARN("%s: creating tap device failed: %s", name, |
10a89ef0 | 675 | ovs_strerror(errno)); |
6c88d577 | 676 | error = errno; |
f61d8d29 | 677 | goto error_close; |
6c88d577 JP |
678 | } |
679 | ||
680 | /* Make non-blocking. */ | |
149f577a | 681 | error = set_nonblocking(state->fd); |
a740f0de | 682 | if (error) { |
f61d8d29 | 683 | goto error_close; |
a740f0de JG |
684 | } |
685 | ||
b5d57fc8 BP |
686 | netdev_init(&netdev->up, name, &netdev_tap_class); |
687 | *netdevp = &netdev->up; | |
a740f0de JG |
688 | return 0; |
689 | ||
f61d8d29 BP |
690 | error_close: |
691 | close(state->fd); | |
1f6e0fbd BP |
692 | error_unref_notifier: |
693 | cache_notifier_unref(); | |
a740f0de | 694 | error: |
b5d57fc8 | 695 | free(netdev); |
a740f0de JG |
696 | return error; |
697 | } | |
698 | ||
a740f0de | 699 | static void |
b5d57fc8 | 700 | destroy_tap(struct netdev_linux *netdev) |
a740f0de | 701 | { |
b5d57fc8 | 702 | struct tap_state *state = &netdev->state.tap; |
149f577a JG |
703 | |
704 | if (state->fd >= 0) { | |
705 | close(state->fd); | |
a740f0de JG |
706 | } |
707 | } | |
708 | ||
b5d57fc8 | 709 | /* Destroys the netdev device 'netdev_'. */ |
6c88d577 | 710 | static void |
b5d57fc8 | 711 | netdev_linux_destroy(struct netdev *netdev_) |
6c88d577 | 712 | { |
b5d57fc8 | 713 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
6c88d577 | 714 | |
b5d57fc8 BP |
715 | if (netdev->tc && netdev->tc->ops->tc_destroy) { |
716 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 BP |
717 | } |
718 | ||
b5d57fc8 BP |
719 | if (netdev_get_class(netdev_) == &netdev_tap_class) { |
720 | destroy_tap(netdev); | |
6c88d577 | 721 | } |
b5d57fc8 | 722 | free(netdev); |
1f6e0fbd BP |
723 | |
724 | cache_notifier_unref(); | |
6c88d577 JP |
725 | } |
726 | ||
7b6b0ef4 | 727 | static int |
796223f5 | 728 | netdev_linux_rx_open(struct netdev *netdev_, struct netdev_rx **rxp) |
7b6b0ef4 BP |
729 | { |
730 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
796223f5 BP |
731 | bool is_tap = is_tap_netdev(netdev_); |
732 | struct netdev_rx_linux *rx; | |
7b6b0ef4 BP |
733 | int error; |
734 | int fd; | |
735 | ||
796223f5 | 736 | if (is_tap) { |
b5d57fc8 | 737 | fd = netdev->state.tap.fd; |
796223f5 BP |
738 | } else { |
739 | struct sockaddr_ll sll; | |
740 | int ifindex; | |
32383c3b MM |
741 | /* Result of tcpdump -dd inbound */ |
742 | static struct sock_filter filt[] = { | |
743 | { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ | |
744 | { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ | |
745 | { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ | |
746 | { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ | |
747 | }; | |
748 | static struct sock_fprog fprog = { ARRAY_SIZE(filt), filt }; | |
7b6b0ef4 | 749 | |
796223f5 BP |
750 | /* Create file descriptor. */ |
751 | fd = socket(PF_PACKET, SOCK_RAW, 0); | |
752 | if (fd < 0) { | |
753 | error = errno; | |
10a89ef0 | 754 | VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); |
796223f5 BP |
755 | goto error; |
756 | } | |
33d82a56 | 757 | |
796223f5 BP |
758 | /* Set non-blocking mode. */ |
759 | error = set_nonblocking(fd); | |
760 | if (error) { | |
761 | goto error; | |
762 | } | |
7b6b0ef4 | 763 | |
796223f5 | 764 | /* Get ethernet device index. */ |
180c6d0b | 765 | error = get_ifindex(&netdev->up, &ifindex); |
796223f5 BP |
766 | if (error) { |
767 | goto error; | |
768 | } | |
7b6b0ef4 | 769 | |
796223f5 BP |
770 | /* Bind to specific ethernet device. */ |
771 | memset(&sll, 0, sizeof sll); | |
772 | sll.sll_family = AF_PACKET; | |
773 | sll.sll_ifindex = ifindex; | |
774 | sll.sll_protocol = (OVS_FORCE unsigned short int) htons(ETH_P_ALL); | |
775 | if (bind(fd, (struct sockaddr *) &sll, sizeof sll) < 0) { | |
776 | error = errno; | |
777 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
10a89ef0 | 778 | netdev_get_name(netdev_), ovs_strerror(error)); |
796223f5 BP |
779 | goto error; |
780 | } | |
32383c3b MM |
781 | |
782 | /* Filter for only inbound packets. */ | |
783 | error = setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, | |
784 | sizeof fprog); | |
785 | if (error) { | |
786 | error = errno; | |
787 | VLOG_ERR("%s: failed attach filter (%s)", | |
10a89ef0 | 788 | netdev_get_name(netdev_), ovs_strerror(error)); |
32383c3b MM |
789 | goto error; |
790 | } | |
7b6b0ef4 BP |
791 | } |
792 | ||
796223f5 | 793 | rx = xmalloc(sizeof *rx); |
b5d57fc8 | 794 | netdev_rx_init(&rx->up, netdev_, &netdev_rx_linux_class); |
796223f5 BP |
795 | rx->is_tap = is_tap; |
796 | rx->fd = fd; | |
7b6b0ef4 | 797 | |
796223f5 | 798 | *rxp = &rx->up; |
7b6b0ef4 BP |
799 | return 0; |
800 | ||
801 | error: | |
802 | if (fd >= 0) { | |
803 | close(fd); | |
804 | } | |
805 | return error; | |
806 | } | |
807 | ||
796223f5 BP |
808 | static void |
809 | netdev_rx_linux_destroy(struct netdev_rx *rx_) | |
8b61709d | 810 | { |
796223f5 | 811 | struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); |
8b61709d | 812 | |
796223f5 BP |
813 | if (!rx->is_tap) { |
814 | close(rx->fd); | |
8b61709d | 815 | } |
796223f5 BP |
816 | free(rx); |
817 | } | |
8b61709d | 818 | |
796223f5 BP |
819 | static int |
820 | netdev_rx_linux_recv(struct netdev_rx *rx_, void *data, size_t size) | |
821 | { | |
822 | struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); | |
823 | ssize_t retval; | |
8e8cddf7 | 824 | |
796223f5 BP |
825 | do { |
826 | retval = (rx->is_tap | |
827 | ? read(rx->fd, data, size) | |
828 | : recv(rx->fd, data, size, MSG_TRUNC)); | |
829 | } while (retval < 0 && errno == EINTR); | |
830 | ||
bb5c1468 Z |
831 | if (retval >= 0) { |
832 | return retval > size ? -EMSGSIZE : retval; | |
796223f5 BP |
833 | } else { |
834 | if (errno != EAGAIN) { | |
835 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
10a89ef0 | 836 | ovs_strerror(errno), netdev_rx_get_name(rx_)); |
8b61709d | 837 | } |
796223f5 | 838 | return -errno; |
8b61709d BP |
839 | } |
840 | } | |
841 | ||
8b61709d | 842 | static void |
796223f5 | 843 | netdev_rx_linux_wait(struct netdev_rx *rx_) |
8b61709d | 844 | { |
796223f5 BP |
845 | struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); |
846 | poll_fd_wait(rx->fd, POLLIN); | |
8b61709d BP |
847 | } |
848 | ||
8b61709d | 849 | static int |
796223f5 | 850 | netdev_rx_linux_drain(struct netdev_rx *rx_) |
8b61709d | 851 | { |
796223f5 BP |
852 | struct netdev_rx_linux *rx = netdev_rx_linux_cast(rx_); |
853 | if (rx->is_tap) { | |
8b61709d | 854 | struct ifreq ifr; |
796223f5 | 855 | int error = netdev_linux_do_ioctl(netdev_rx_get_name(rx_), &ifr, |
8b61709d BP |
856 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
857 | if (error) { | |
858 | return error; | |
859 | } | |
796223f5 | 860 | drain_fd(rx->fd, ifr.ifr_qlen); |
8b61709d BP |
861 | return 0; |
862 | } else { | |
796223f5 | 863 | return drain_rcvbuf(rx->fd); |
8b61709d BP |
864 | } |
865 | } | |
866 | ||
867 | /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive | |
868 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued | |
869 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
870 | * the packet is too big or too small to transmit on the device. | |
871 | * | |
872 | * The caller retains ownership of 'buffer' in all cases. | |
873 | * | |
874 | * The kernel maintains a packet transmission queue, so the caller is not | |
875 | * expected to do additional queuing of packets. */ | |
876 | static int | |
877 | netdev_linux_send(struct netdev *netdev_, const void *data, size_t size) | |
878 | { | |
f23347ea BP |
879 | for (;;) { |
880 | ssize_t retval; | |
8b61709d | 881 | |
796223f5 | 882 | if (!is_tap_netdev(netdev_)) { |
f23347ea BP |
883 | /* Use our AF_PACKET socket to send to this device. */ |
884 | struct sockaddr_ll sll; | |
885 | struct msghdr msg; | |
886 | struct iovec iov; | |
887 | int ifindex; | |
888 | int error; | |
488d734d BP |
889 | int sock; |
890 | ||
891 | sock = af_packet_sock(); | |
892 | if (sock < 0) { | |
c4c7a3d7 | 893 | return -sock; |
488d734d | 894 | } |
f23347ea BP |
895 | |
896 | error = get_ifindex(netdev_, &ifindex); | |
897 | if (error) { | |
898 | return error; | |
899 | } | |
8b61709d | 900 | |
f23347ea BP |
901 | /* We don't bother setting most fields in sockaddr_ll because the |
902 | * kernel ignores them for SOCK_RAW. */ | |
903 | memset(&sll, 0, sizeof sll); | |
904 | sll.sll_family = AF_PACKET; | |
905 | sll.sll_ifindex = ifindex; | |
76c308b5 | 906 | |
ebc56baa | 907 | iov.iov_base = CONST_CAST(void *, data); |
f23347ea | 908 | iov.iov_len = size; |
76c308b5 | 909 | |
f23347ea BP |
910 | msg.msg_name = &sll; |
911 | msg.msg_namelen = sizeof sll; | |
912 | msg.msg_iov = &iov; | |
913 | msg.msg_iovlen = 1; | |
914 | msg.msg_control = NULL; | |
915 | msg.msg_controllen = 0; | |
916 | msg.msg_flags = 0; | |
917 | ||
488d734d | 918 | retval = sendmsg(sock, &msg, 0); |
f23347ea | 919 | } else { |
796223f5 BP |
920 | /* Use the tap fd to send to this device. This is essential for |
921 | * tap devices, because packets sent to a tap device with an | |
922 | * AF_PACKET socket will loop back to be *received* again on the | |
32383c3b MM |
923 | * tap device. This doesn't occur on other interface types |
924 | * because we attach a socket filter to the rx socket. */ | |
b5d57fc8 | 925 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
796223f5 | 926 | |
b5d57fc8 | 927 | retval = write(netdev->state.tap.fd, data, size); |
f23347ea | 928 | } |
76c308b5 | 929 | |
8b61709d BP |
930 | if (retval < 0) { |
931 | /* The Linux AF_PACKET implementation never blocks waiting for room | |
932 | * for packets, instead returning ENOBUFS. Translate this into | |
933 | * EAGAIN for the caller. */ | |
934 | if (errno == ENOBUFS) { | |
935 | return EAGAIN; | |
936 | } else if (errno == EINTR) { | |
937 | continue; | |
938 | } else if (errno != EAGAIN) { | |
939 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", | |
10a89ef0 | 940 | netdev_get_name(netdev_), ovs_strerror(errno)); |
8b61709d BP |
941 | } |
942 | return errno; | |
943 | } else if (retval != size) { | |
944 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%zd bytes of " | |
945 | "%zu) on %s", retval, size, netdev_get_name(netdev_)); | |
946 | return EMSGSIZE; | |
947 | } else { | |
948 | return 0; | |
949 | } | |
950 | } | |
951 | } | |
952 | ||
953 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
954 | * when the packet transmission queue has sufficient room to transmit a packet | |
955 | * with netdev_send(). | |
956 | * | |
957 | * The kernel maintains a packet transmission queue, so the client is not | |
958 | * expected to do additional queuing of packets. Thus, this function is | |
959 | * unlikely to ever be used. It is included for completeness. */ | |
960 | static void | |
796223f5 | 961 | netdev_linux_send_wait(struct netdev *netdev) |
8b61709d | 962 | { |
796223f5 | 963 | if (is_tap_netdev(netdev)) { |
8b61709d BP |
964 | /* TAP device always accepts packets.*/ |
965 | poll_immediate_wake(); | |
966 | } | |
967 | } | |
968 | ||
969 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
970 | * otherwise a positive errno value. */ | |
971 | static int | |
972 | netdev_linux_set_etheraddr(struct netdev *netdev_, | |
973 | const uint8_t mac[ETH_ADDR_LEN]) | |
974 | { | |
b5d57fc8 | 975 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4b609110 | 976 | struct netdev_saved_flags *sf = NULL; |
eb395f2e BP |
977 | int error; |
978 | ||
b5d57fc8 BP |
979 | if (netdev->cache_valid & VALID_ETHERADDR) { |
980 | if (netdev->ether_addr_error) { | |
981 | return netdev->ether_addr_error; | |
44445cac | 982 | } |
b5d57fc8 | 983 | if (eth_addr_equals(netdev->etheraddr, mac)) { |
44445cac PS |
984 | return 0; |
985 | } | |
b5d57fc8 | 986 | netdev->cache_valid &= ~VALID_ETHERADDR; |
44445cac PS |
987 | } |
988 | ||
7eb1bd81 | 989 | /* Tap devices must be brought down before setting the address. */ |
796223f5 | 990 | if (is_tap_netdev(netdev_)) { |
bbd5b6f4 | 991 | netdev_turn_flags_off(netdev_, NETDEV_UP, &sf); |
7eb1bd81 | 992 | } |
44445cac PS |
993 | error = set_etheraddr(netdev_get_name(netdev_), mac); |
994 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
995 | netdev->ether_addr_error = error; |
996 | netdev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 997 | if (!error) { |
b5d57fc8 | 998 | memcpy(netdev->etheraddr, mac, ETH_ADDR_LEN); |
eb395f2e | 999 | } |
8b61709d | 1000 | } |
44445cac | 1001 | |
4b609110 | 1002 | netdev_restore_flags(sf); |
7eb1bd81 | 1003 | |
8b61709d BP |
1004 | return error; |
1005 | } | |
1006 | ||
44445cac | 1007 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d BP |
1008 | static int |
1009 | netdev_linux_get_etheraddr(const struct netdev *netdev_, | |
1010 | uint8_t mac[ETH_ADDR_LEN]) | |
1011 | { | |
b5d57fc8 | 1012 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
44445cac | 1013 | |
b5d57fc8 | 1014 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { |
8b61709d | 1015 | int error = get_etheraddr(netdev_get_name(netdev_), |
b5d57fc8 | 1016 | netdev->etheraddr); |
44445cac | 1017 | |
b5d57fc8 BP |
1018 | netdev->ether_addr_error = error; |
1019 | netdev->cache_valid |= VALID_ETHERADDR; | |
8b61709d | 1020 | } |
44445cac | 1021 | |
b5d57fc8 BP |
1022 | if (!netdev->ether_addr_error) { |
1023 | memcpy(mac, netdev->etheraddr, ETH_ADDR_LEN); | |
44445cac PS |
1024 | } |
1025 | ||
b5d57fc8 | 1026 | return netdev->ether_addr_error; |
8b61709d BP |
1027 | } |
1028 | ||
1029 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1030 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1031 | * bytes for Ethernet devices. */ | |
1032 | static int | |
1033 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1034 | { | |
b5d57fc8 BP |
1035 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1036 | if (!(netdev->cache_valid & VALID_MTU)) { | |
8b61709d BP |
1037 | struct ifreq ifr; |
1038 | int error; | |
1039 | ||
149f577a JG |
1040 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, |
1041 | SIOCGIFMTU, "SIOCGIFMTU"); | |
90a6637d | 1042 | |
b5d57fc8 BP |
1043 | netdev->netdev_mtu_error = error; |
1044 | netdev->mtu = ifr.ifr_mtu; | |
1045 | netdev->cache_valid |= VALID_MTU; | |
8b61709d | 1046 | } |
90a6637d | 1047 | |
b5d57fc8 BP |
1048 | if (!netdev->netdev_mtu_error) { |
1049 | *mtup = netdev->mtu; | |
90a6637d | 1050 | } |
b5d57fc8 | 1051 | return netdev->netdev_mtu_error; |
8b61709d BP |
1052 | } |
1053 | ||
9b020780 PS |
1054 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1055 | * networking ioctl interface. | |
1056 | */ | |
1057 | static int | |
1058 | netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) | |
1059 | { | |
b5d57fc8 | 1060 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9b020780 PS |
1061 | struct ifreq ifr; |
1062 | int error; | |
1063 | ||
b5d57fc8 BP |
1064 | if (netdev->cache_valid & VALID_MTU) { |
1065 | if (netdev->netdev_mtu_error) { | |
1066 | return netdev->netdev_mtu_error; | |
90a6637d | 1067 | } |
b5d57fc8 | 1068 | if (netdev->mtu == mtu) { |
90a6637d PS |
1069 | return 0; |
1070 | } | |
b5d57fc8 | 1071 | netdev->cache_valid &= ~VALID_MTU; |
153e5481 | 1072 | } |
9b020780 PS |
1073 | ifr.ifr_mtu = mtu; |
1074 | error = netdev_linux_do_ioctl(netdev_get_name(netdev_), &ifr, | |
1075 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d | 1076 | if (!error || error == ENODEV) { |
b5d57fc8 BP |
1077 | netdev->netdev_mtu_error = error; |
1078 | netdev->mtu = ifr.ifr_mtu; | |
1079 | netdev->cache_valid |= VALID_MTU; | |
9b020780 | 1080 | } |
90a6637d | 1081 | return error; |
9b020780 PS |
1082 | } |
1083 | ||
9ab3d9a3 BP |
1084 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1085 | * On failure, returns a negative errno value. */ | |
1086 | static int | |
1087 | netdev_linux_get_ifindex(const struct netdev *netdev) | |
1088 | { | |
1089 | int ifindex, error; | |
1090 | ||
1091 | error = get_ifindex(netdev, &ifindex); | |
1092 | return error ? -error : ifindex; | |
1093 | } | |
1094 | ||
8b61709d BP |
1095 | static int |
1096 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1097 | { | |
b5d57fc8 | 1098 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1099 | |
b5d57fc8 BP |
1100 | if (netdev->miimon_interval > 0) { |
1101 | *carrier = netdev->miimon; | |
3a183124 | 1102 | } else { |
b5d57fc8 | 1103 | *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1104 | } |
8b61709d | 1105 | |
3a183124 | 1106 | return 0; |
8b61709d BP |
1107 | } |
1108 | ||
65c3058c EJ |
1109 | static long long int |
1110 | netdev_linux_get_carrier_resets(const struct netdev *netdev) | |
1111 | { | |
b5d57fc8 | 1112 | return netdev_linux_cast(netdev)->carrier_resets; |
65c3058c EJ |
1113 | } |
1114 | ||
63331829 | 1115 | static int |
1670c579 EJ |
1116 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1117 | struct mii_ioctl_data *data) | |
63331829 | 1118 | { |
63331829 | 1119 | struct ifreq ifr; |
782e6111 | 1120 | int error; |
63331829 | 1121 | |
63331829 | 1122 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1123 | memcpy(&ifr.ifr_data, data, sizeof *data); |
1670c579 | 1124 | error = netdev_linux_do_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1125 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1126 | |
782e6111 EJ |
1127 | return error; |
1128 | } | |
1129 | ||
1130 | static int | |
1670c579 | 1131 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1132 | { |
782e6111 EJ |
1133 | struct mii_ioctl_data data; |
1134 | int error; | |
63331829 | 1135 | |
782e6111 EJ |
1136 | *miimon = false; |
1137 | ||
1138 | memset(&data, 0, sizeof data); | |
1670c579 | 1139 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
1140 | if (!error) { |
1141 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1142 | data.reg_num = MII_BMSR; | |
1670c579 | 1143 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 1144 | &data); |
63331829 EJ |
1145 | |
1146 | if (!error) { | |
782e6111 | 1147 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 EJ |
1148 | } else { |
1149 | VLOG_WARN_RL(&rl, "%s: failed to query MII", name); | |
1150 | } | |
1151 | } else { | |
1152 | struct ethtool_cmd ecmd; | |
63331829 EJ |
1153 | |
1154 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1155 | name); | |
1156 | ||
ab985a77 | 1157 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
1158 | memset(&ecmd, 0, sizeof ecmd); |
1159 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1160 | "ETHTOOL_GLINK"); | |
1161 | if (!error) { | |
782e6111 EJ |
1162 | struct ethtool_value eval; |
1163 | ||
1164 | memcpy(&eval, &ecmd, sizeof eval); | |
1165 | *miimon = !!eval.data; | |
63331829 EJ |
1166 | } else { |
1167 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1168 | } | |
1169 | } | |
1170 | ||
1171 | return error; | |
1172 | } | |
1173 | ||
1670c579 EJ |
1174 | static int |
1175 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
1176 | long long int interval) | |
1177 | { | |
b5d57fc8 | 1178 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1670c579 EJ |
1179 | |
1180 | interval = interval > 0 ? MAX(interval, 100) : 0; | |
b5d57fc8 BP |
1181 | if (netdev->miimon_interval != interval) { |
1182 | netdev->miimon_interval = interval; | |
1183 | timer_set_expired(&netdev->miimon_timer); | |
1670c579 EJ |
1184 | } |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
1189 | static void | |
1190 | netdev_linux_miimon_run(void) | |
1191 | { | |
1192 | struct shash device_shash; | |
1193 | struct shash_node *node; | |
1194 | ||
1195 | shash_init(&device_shash); | |
b5d57fc8 | 1196 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1197 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1198 | struct netdev *netdev = node->data; |
1199 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
1200 | bool miimon; |
1201 | ||
1202 | if (dev->miimon_interval <= 0 || !timer_expired(&dev->miimon_timer)) { | |
1203 | continue; | |
1204 | } | |
1205 | ||
180c6d0b | 1206 | netdev_linux_get_miimon(dev->up.name, &miimon); |
1670c579 | 1207 | if (miimon != dev->miimon) { |
1670c579 | 1208 | dev->miimon = miimon; |
b5d57fc8 | 1209 | netdev_linux_changed(dev, dev->ifi_flags, 0); |
1670c579 EJ |
1210 | } |
1211 | ||
1212 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); | |
1213 | } | |
1214 | ||
1215 | shash_destroy(&device_shash); | |
1216 | } | |
1217 | ||
1218 | static void | |
1219 | netdev_linux_miimon_wait(void) | |
1220 | { | |
1221 | struct shash device_shash; | |
1222 | struct shash_node *node; | |
1223 | ||
1224 | shash_init(&device_shash); | |
b5d57fc8 | 1225 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1226 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1227 | struct netdev *netdev = node->data; |
1228 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
1229 | |
1230 | if (dev->miimon_interval > 0) { | |
1231 | timer_wait(&dev->miimon_timer); | |
1232 | } | |
1233 | } | |
1234 | shash_destroy(&device_shash); | |
1235 | } | |
1236 | ||
8b61709d BP |
1237 | /* Check whether we can we use RTM_GETLINK to get network device statistics. |
1238 | * In pre-2.6.19 kernels, this was only available if wireless extensions were | |
1239 | * enabled. */ | |
1240 | static bool | |
1241 | check_for_working_netlink_stats(void) | |
1242 | { | |
1243 | /* Decide on the netdev_get_stats() implementation to use. Netlink is | |
1244 | * preferable, so if that works, we'll use it. */ | |
1245 | int ifindex = do_get_ifindex("lo"); | |
1246 | if (ifindex < 0) { | |
1247 | VLOG_WARN("failed to get ifindex for lo, " | |
1248 | "obtaining netdev stats from proc"); | |
1249 | return false; | |
1250 | } else { | |
1251 | struct netdev_stats stats; | |
1252 | int error = get_stats_via_netlink(ifindex, &stats); | |
1253 | if (!error) { | |
1254 | VLOG_DBG("obtaining netdev stats via rtnetlink"); | |
1255 | return true; | |
1256 | } else { | |
1257 | VLOG_INFO("RTM_GETLINK failed (%s), obtaining netdev stats " | |
1258 | "via proc (you are probably running a pre-2.6.19 " | |
10a89ef0 | 1259 | "kernel)", ovs_strerror(error)); |
8b61709d BP |
1260 | return false; |
1261 | } | |
1262 | } | |
1263 | } | |
1264 | ||
92df599c JG |
1265 | static void |
1266 | swap_uint64(uint64_t *a, uint64_t *b) | |
1267 | { | |
1de0e8ae BP |
1268 | uint64_t tmp = *a; |
1269 | *a = *b; | |
1270 | *b = tmp; | |
92df599c JG |
1271 | } |
1272 | ||
c060c4cf EJ |
1273 | /* Copies 'src' into 'dst', performing format conversion in the process. |
1274 | * | |
1275 | * 'src' is allowed to be misaligned. */ | |
1276 | static void | |
1277 | netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, | |
1278 | const struct ovs_vport_stats *src) | |
1279 | { | |
1280 | dst->rx_packets = get_unaligned_u64(&src->rx_packets); | |
1281 | dst->tx_packets = get_unaligned_u64(&src->tx_packets); | |
1282 | dst->rx_bytes = get_unaligned_u64(&src->rx_bytes); | |
1283 | dst->tx_bytes = get_unaligned_u64(&src->tx_bytes); | |
1284 | dst->rx_errors = get_unaligned_u64(&src->rx_errors); | |
1285 | dst->tx_errors = get_unaligned_u64(&src->tx_errors); | |
1286 | dst->rx_dropped = get_unaligned_u64(&src->rx_dropped); | |
1287 | dst->tx_dropped = get_unaligned_u64(&src->tx_dropped); | |
1288 | dst->multicast = 0; | |
1289 | dst->collisions = 0; | |
1290 | dst->rx_length_errors = 0; | |
1291 | dst->rx_over_errors = 0; | |
1292 | dst->rx_crc_errors = 0; | |
1293 | dst->rx_frame_errors = 0; | |
1294 | dst->rx_fifo_errors = 0; | |
1295 | dst->rx_missed_errors = 0; | |
1296 | dst->tx_aborted_errors = 0; | |
1297 | dst->tx_carrier_errors = 0; | |
1298 | dst->tx_fifo_errors = 0; | |
1299 | dst->tx_heartbeat_errors = 0; | |
1300 | dst->tx_window_errors = 0; | |
1301 | } | |
1302 | ||
1303 | static int | |
1304 | get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) | |
1305 | { | |
1306 | struct dpif_linux_vport reply; | |
1307 | struct ofpbuf *buf; | |
1308 | int error; | |
1309 | ||
1310 | error = dpif_linux_vport_get(netdev_get_name(netdev), &reply, &buf); | |
1311 | if (error) { | |
1312 | return error; | |
1313 | } else if (!reply.stats) { | |
1314 | ofpbuf_delete(buf); | |
1315 | return EOPNOTSUPP; | |
1316 | } | |
1317 | ||
1318 | netdev_stats_from_ovs_vport_stats(stats, reply.stats); | |
1319 | ||
1320 | ofpbuf_delete(buf); | |
1321 | ||
1322 | return 0; | |
1323 | } | |
1324 | ||
f613a0d7 PS |
1325 | static void |
1326 | get_stats_via_vport(const struct netdev *netdev_, | |
1327 | struct netdev_stats *stats) | |
8b61709d | 1328 | { |
b5d57fc8 | 1329 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1330 | |
b5d57fc8 BP |
1331 | if (!netdev->vport_stats_error || |
1332 | !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 1333 | int error; |
7fbef77a | 1334 | |
c060c4cf | 1335 | error = get_stats_via_vport__(netdev_, stats); |
bcb1f5a1 | 1336 | if (error && error != ENOENT) { |
a57a8488 | 1337 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
10a89ef0 BP |
1338 | "(%s)", |
1339 | netdev_get_name(netdev_), ovs_strerror(error)); | |
f613a0d7 | 1340 | } |
b5d57fc8 BP |
1341 | netdev->vport_stats_error = error; |
1342 | netdev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 1343 | } |
f613a0d7 | 1344 | } |
8b61709d | 1345 | |
f613a0d7 PS |
1346 | static int |
1347 | netdev_linux_sys_get_stats(const struct netdev *netdev_, | |
1348 | struct netdev_stats *stats) | |
1349 | { | |
23882115 BP |
1350 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
1351 | static int use_netlink_stats; | |
f613a0d7 PS |
1352 | int error; |
1353 | ||
23882115 | 1354 | if (ovsthread_once_start(&once)) { |
f613a0d7 | 1355 | use_netlink_stats = check_for_working_netlink_stats(); |
23882115 | 1356 | ovsthread_once_done(&once); |
f613a0d7 PS |
1357 | } |
1358 | ||
1359 | if (use_netlink_stats) { | |
1360 | int ifindex; | |
1361 | ||
1362 | error = get_ifindex(netdev_, &ifindex); | |
1363 | if (!error) { | |
1364 | error = get_stats_via_netlink(ifindex, stats); | |
7fbef77a | 1365 | } |
f613a0d7 PS |
1366 | } else { |
1367 | error = get_stats_via_proc(netdev_get_name(netdev_), stats); | |
1368 | } | |
7fbef77a | 1369 | |
f613a0d7 PS |
1370 | if (error) { |
1371 | VLOG_WARN_RL(&rl, "%s: linux-sys get stats failed %d", | |
1372 | netdev_get_name(netdev_), error); | |
1373 | } | |
1374 | return error; | |
1375 | ||
1376 | } | |
1377 | ||
1378 | /* Retrieves current device stats for 'netdev-linux'. */ | |
1379 | static int | |
1380 | netdev_linux_get_stats(const struct netdev *netdev_, | |
1381 | struct netdev_stats *stats) | |
1382 | { | |
b5d57fc8 | 1383 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1384 | struct netdev_stats dev_stats; |
1385 | int error; | |
1386 | ||
1387 | get_stats_via_vport(netdev_, stats); | |
1388 | ||
1389 | error = netdev_linux_sys_get_stats(netdev_, &dev_stats); | |
1390 | ||
1391 | if (error) { | |
b5d57fc8 | 1392 | if (netdev->vport_stats_error) { |
f613a0d7 | 1393 | return error; |
7fbef77a | 1394 | } else { |
f613a0d7 PS |
1395 | return 0; |
1396 | } | |
1397 | } | |
1398 | ||
b5d57fc8 | 1399 | if (netdev->vport_stats_error) { |
f613a0d7 PS |
1400 | /* stats not available from OVS then use ioctl stats. */ |
1401 | *stats = dev_stats; | |
1402 | } else { | |
1403 | stats->rx_errors += dev_stats.rx_errors; | |
1404 | stats->tx_errors += dev_stats.tx_errors; | |
1405 | stats->rx_dropped += dev_stats.rx_dropped; | |
1406 | stats->tx_dropped += dev_stats.tx_dropped; | |
1407 | stats->multicast += dev_stats.multicast; | |
1408 | stats->collisions += dev_stats.collisions; | |
1409 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
1410 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
1411 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
1412 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
1413 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
1414 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
1415 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
1416 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
1417 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
1418 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
1419 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
1420 | } | |
1421 | return 0; | |
1422 | } | |
1423 | ||
1424 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
1425 | * netdev-internal. */ | |
1426 | static int | |
bba1e6f3 | 1427 | netdev_tap_get_stats(const struct netdev *netdev_, |
f613a0d7 PS |
1428 | struct netdev_stats *stats) |
1429 | { | |
b5d57fc8 | 1430 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1431 | struct netdev_stats dev_stats; |
1432 | int error; | |
1433 | ||
1434 | get_stats_via_vport(netdev_, stats); | |
1435 | ||
1436 | error = netdev_linux_sys_get_stats(netdev_, &dev_stats); | |
1437 | if (error) { | |
b5d57fc8 | 1438 | if (netdev->vport_stats_error) { |
f613a0d7 PS |
1439 | return error; |
1440 | } else { | |
1441 | return 0; | |
8b61709d | 1442 | } |
8b61709d | 1443 | } |
fe6b0e03 JG |
1444 | |
1445 | /* If this port is an internal port then the transmit and receive stats | |
1446 | * will appear to be swapped relative to the other ports since we are the | |
1447 | * one sending the data, not a remote computer. For consistency, we swap | |
7fbef77a JG |
1448 | * them back here. This does not apply if we are getting stats from the |
1449 | * vport layer because it always tracks stats from the perspective of the | |
1450 | * switch. */ | |
b5d57fc8 | 1451 | if (netdev->vport_stats_error) { |
f613a0d7 | 1452 | *stats = dev_stats; |
92df599c JG |
1453 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1454 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1455 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1456 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1457 | stats->rx_length_errors = 0; |
1458 | stats->rx_over_errors = 0; | |
1459 | stats->rx_crc_errors = 0; | |
1460 | stats->rx_frame_errors = 0; | |
1461 | stats->rx_fifo_errors = 0; | |
1462 | stats->rx_missed_errors = 0; | |
1463 | stats->tx_aborted_errors = 0; | |
1464 | stats->tx_carrier_errors = 0; | |
1465 | stats->tx_fifo_errors = 0; | |
1466 | stats->tx_heartbeat_errors = 0; | |
1467 | stats->tx_window_errors = 0; | |
f613a0d7 PS |
1468 | } else { |
1469 | stats->rx_dropped += dev_stats.tx_dropped; | |
1470 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 1471 | |
f613a0d7 PS |
1472 | stats->rx_errors += dev_stats.tx_errors; |
1473 | stats->tx_errors += dev_stats.rx_errors; | |
1474 | ||
1475 | stats->multicast += dev_stats.multicast; | |
1476 | stats->collisions += dev_stats.collisions; | |
1477 | } | |
1478 | return 0; | |
8b61709d BP |
1479 | } |
1480 | ||
bba1e6f3 PS |
1481 | static int |
1482 | netdev_internal_get_stats(const struct netdev *netdev_, | |
1483 | struct netdev_stats *stats) | |
1484 | { | |
b5d57fc8 | 1485 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
bba1e6f3 PS |
1486 | |
1487 | get_stats_via_vport(netdev_, stats); | |
b5d57fc8 | 1488 | return netdev->vport_stats_error; |
bba1e6f3 PS |
1489 | } |
1490 | ||
2f31a822 EJ |
1491 | static int |
1492 | netdev_internal_set_stats(struct netdev *netdev, | |
1493 | const struct netdev_stats *stats) | |
1494 | { | |
1495 | struct ovs_vport_stats vport_stats; | |
1496 | struct dpif_linux_vport vport; | |
1497 | int err; | |
1498 | ||
1499 | vport_stats.rx_packets = stats->rx_packets; | |
1500 | vport_stats.tx_packets = stats->tx_packets; | |
1501 | vport_stats.rx_bytes = stats->rx_bytes; | |
1502 | vport_stats.tx_bytes = stats->tx_bytes; | |
1503 | vport_stats.rx_errors = stats->rx_errors; | |
1504 | vport_stats.tx_errors = stats->tx_errors; | |
1505 | vport_stats.rx_dropped = stats->rx_dropped; | |
1506 | vport_stats.tx_dropped = stats->tx_dropped; | |
1507 | ||
1508 | dpif_linux_vport_init(&vport); | |
1509 | vport.cmd = OVS_VPORT_CMD_SET; | |
1510 | vport.name = netdev_get_name(netdev); | |
1511 | vport.stats = &vport_stats; | |
1512 | ||
1513 | err = dpif_linux_vport_transact(&vport, NULL, NULL); | |
1514 | ||
1515 | /* If the vport layer doesn't know about the device, that doesn't mean it | |
1516 | * doesn't exist (after all were able to open it when netdev_open() was | |
1517 | * called), it just means that it isn't attached and we'll be getting | |
1518 | * stats a different way. */ | |
1519 | if (err == ENODEV) { | |
1520 | err = EOPNOTSUPP; | |
1521 | } | |
1522 | ||
1523 | return err; | |
1524 | } | |
1525 | ||
51f87458 | 1526 | static void |
b5d57fc8 | 1527 | netdev_linux_read_features(struct netdev_linux *netdev) |
8b61709d BP |
1528 | { |
1529 | struct ethtool_cmd ecmd; | |
6c038611 | 1530 | uint32_t speed; |
8b61709d BP |
1531 | int error; |
1532 | ||
b5d57fc8 | 1533 | if (netdev->cache_valid & VALID_FEATURES) { |
51f87458 PS |
1534 | return; |
1535 | } | |
1536 | ||
ab985a77 | 1537 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1538 | memset(&ecmd, 0, sizeof ecmd); |
b5d57fc8 | 1539 | error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, |
8b61709d BP |
1540 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1541 | if (error) { | |
51f87458 | 1542 | goto out; |
8b61709d BP |
1543 | } |
1544 | ||
1545 | /* Supported features. */ | |
b5d57fc8 | 1546 | netdev->supported = 0; |
8b61709d | 1547 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
b5d57fc8 | 1548 | netdev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
1549 | } |
1550 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
b5d57fc8 | 1551 | netdev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
1552 | } |
1553 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
b5d57fc8 | 1554 | netdev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
1555 | } |
1556 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
b5d57fc8 | 1557 | netdev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
1558 | } |
1559 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
b5d57fc8 | 1560 | netdev->supported |= NETDEV_F_1GB_HD; |
8b61709d BP |
1561 | } |
1562 | if (ecmd.supported & SUPPORTED_1000baseT_Full) { | |
b5d57fc8 | 1563 | netdev->supported |= NETDEV_F_1GB_FD; |
8b61709d BP |
1564 | } |
1565 | if (ecmd.supported & SUPPORTED_10000baseT_Full) { | |
b5d57fc8 | 1566 | netdev->supported |= NETDEV_F_10GB_FD; |
8b61709d BP |
1567 | } |
1568 | if (ecmd.supported & SUPPORTED_TP) { | |
b5d57fc8 | 1569 | netdev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
1570 | } |
1571 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
b5d57fc8 | 1572 | netdev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
1573 | } |
1574 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
b5d57fc8 | 1575 | netdev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
1576 | } |
1577 | if (ecmd.supported & SUPPORTED_Pause) { | |
b5d57fc8 | 1578 | netdev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
1579 | } |
1580 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
b5d57fc8 | 1581 | netdev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1582 | } |
1583 | ||
1584 | /* Advertised features. */ | |
b5d57fc8 | 1585 | netdev->advertised = 0; |
8b61709d | 1586 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
b5d57fc8 | 1587 | netdev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
1588 | } |
1589 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
b5d57fc8 | 1590 | netdev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
1591 | } |
1592 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
b5d57fc8 | 1593 | netdev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
1594 | } |
1595 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
b5d57fc8 | 1596 | netdev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
1597 | } |
1598 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
b5d57fc8 | 1599 | netdev->advertised |= NETDEV_F_1GB_HD; |
8b61709d BP |
1600 | } |
1601 | if (ecmd.advertising & ADVERTISED_1000baseT_Full) { | |
b5d57fc8 | 1602 | netdev->advertised |= NETDEV_F_1GB_FD; |
8b61709d BP |
1603 | } |
1604 | if (ecmd.advertising & ADVERTISED_10000baseT_Full) { | |
b5d57fc8 | 1605 | netdev->advertised |= NETDEV_F_10GB_FD; |
8b61709d BP |
1606 | } |
1607 | if (ecmd.advertising & ADVERTISED_TP) { | |
b5d57fc8 | 1608 | netdev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
1609 | } |
1610 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
b5d57fc8 | 1611 | netdev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
1612 | } |
1613 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
b5d57fc8 | 1614 | netdev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
1615 | } |
1616 | if (ecmd.advertising & ADVERTISED_Pause) { | |
b5d57fc8 | 1617 | netdev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
1618 | } |
1619 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
b5d57fc8 | 1620 | netdev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1621 | } |
1622 | ||
1623 | /* Current settings. */ | |
2a529ead | 1624 | speed = ecmd.speed; |
6c038611 | 1625 | if (speed == SPEED_10) { |
b5d57fc8 | 1626 | netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 1627 | } else if (speed == SPEED_100) { |
b5d57fc8 | 1628 | netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 1629 | } else if (speed == SPEED_1000) { |
b5d57fc8 | 1630 | netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 1631 | } else if (speed == SPEED_10000) { |
b5d57fc8 | 1632 | netdev->current = NETDEV_F_10GB_FD; |
6c038611 | 1633 | } else if (speed == 40000) { |
b5d57fc8 | 1634 | netdev->current = NETDEV_F_40GB_FD; |
6c038611 | 1635 | } else if (speed == 100000) { |
b5d57fc8 | 1636 | netdev->current = NETDEV_F_100GB_FD; |
6c038611 | 1637 | } else if (speed == 1000000) { |
b5d57fc8 | 1638 | netdev->current = NETDEV_F_1TB_FD; |
8b61709d | 1639 | } else { |
b5d57fc8 | 1640 | netdev->current = 0; |
8b61709d BP |
1641 | } |
1642 | ||
1643 | if (ecmd.port == PORT_TP) { | |
b5d57fc8 | 1644 | netdev->current |= NETDEV_F_COPPER; |
8b61709d | 1645 | } else if (ecmd.port == PORT_FIBRE) { |
b5d57fc8 | 1646 | netdev->current |= NETDEV_F_FIBER; |
8b61709d BP |
1647 | } |
1648 | ||
1649 | if (ecmd.autoneg) { | |
b5d57fc8 | 1650 | netdev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
1651 | } |
1652 | ||
1653 | /* Peer advertisements. */ | |
b5d57fc8 | 1654 | netdev->peer = 0; /* XXX */ |
8b61709d | 1655 | |
51f87458 | 1656 | out: |
b5d57fc8 BP |
1657 | netdev->cache_valid |= VALID_FEATURES; |
1658 | netdev->get_features_error = error; | |
51f87458 PS |
1659 | } |
1660 | ||
1661 | /* Stores the features supported by 'netdev' into each of '*current', | |
1662 | * '*advertised', '*supported', and '*peer' that are non-null. Each value is a | |
1663 | * bitmap of NETDEV_* bits. Returns 0 if successful, otherwise a positive | |
1664 | * errno value. */ | |
1665 | static int | |
1666 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
1667 | enum netdev_features *current, |
1668 | enum netdev_features *advertised, | |
1669 | enum netdev_features *supported, | |
1670 | enum netdev_features *peer) | |
51f87458 | 1671 | { |
b5d57fc8 | 1672 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
51f87458 | 1673 | |
b5d57fc8 | 1674 | netdev_linux_read_features(netdev); |
51f87458 | 1675 | |
b5d57fc8 BP |
1676 | if (!netdev->get_features_error) { |
1677 | *current = netdev->current; | |
1678 | *advertised = netdev->advertised; | |
1679 | *supported = netdev->supported; | |
1680 | *peer = netdev->peer; | |
51f87458 | 1681 | } |
b5d57fc8 | 1682 | return netdev->get_features_error; |
8b61709d BP |
1683 | } |
1684 | ||
1685 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
1686 | static int | |
6c038611 BP |
1687 | netdev_linux_set_advertisements(struct netdev *netdev, |
1688 | enum netdev_features advertise) | |
8b61709d BP |
1689 | { |
1690 | struct ethtool_cmd ecmd; | |
1691 | int error; | |
1692 | ||
ab985a77 | 1693 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1694 | memset(&ecmd, 0, sizeof ecmd); |
0b0544d7 | 1695 | error = netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1696 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1697 | if (error) { | |
1698 | return error; | |
1699 | } | |
1700 | ||
1701 | ecmd.advertising = 0; | |
6c038611 | 1702 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
1703 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
1704 | } | |
6c038611 | 1705 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
1706 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
1707 | } | |
6c038611 | 1708 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
1709 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
1710 | } | |
6c038611 | 1711 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
1712 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
1713 | } | |
6c038611 | 1714 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
1715 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
1716 | } | |
6c038611 | 1717 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
1718 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
1719 | } | |
6c038611 | 1720 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
1721 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
1722 | } | |
6c038611 | 1723 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
1724 | ecmd.advertising |= ADVERTISED_TP; |
1725 | } | |
6c038611 | 1726 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
1727 | ecmd.advertising |= ADVERTISED_FIBRE; |
1728 | } | |
6c038611 | 1729 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
1730 | ecmd.advertising |= ADVERTISED_Autoneg; |
1731 | } | |
6c038611 | 1732 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
1733 | ecmd.advertising |= ADVERTISED_Pause; |
1734 | } | |
6c038611 | 1735 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
1736 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
1737 | } | |
ab985a77 | 1738 | COVERAGE_INC(netdev_set_ethtool); |
0b0544d7 | 1739 | return netdev_linux_do_ethtool(netdev_get_name(netdev), &ecmd, |
8b61709d BP |
1740 | ETHTOOL_SSET, "ETHTOOL_SSET"); |
1741 | } | |
1742 | ||
f8500004 JP |
1743 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
1744 | * successful, otherwise a positive errno value. */ | |
8b61709d | 1745 | static int |
b5d57fc8 | 1746 | netdev_linux_set_policing(struct netdev *netdev_, |
8b61709d BP |
1747 | uint32_t kbits_rate, uint32_t kbits_burst) |
1748 | { | |
b5d57fc8 BP |
1749 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1750 | const char *netdev_name = netdev_get_name(netdev_); | |
f8500004 | 1751 | int error; |
8b61709d | 1752 | |
8e460221 | 1753 | |
80a86fbe BP |
1754 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
1755 | : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ | |
1756 | : kbits_burst); /* Stick with user-specified value. */ | |
1757 | ||
b5d57fc8 BP |
1758 | if (netdev->cache_valid & VALID_POLICING) { |
1759 | if (netdev->netdev_policing_error) { | |
1760 | return netdev->netdev_policing_error; | |
c9f71668 PS |
1761 | } |
1762 | ||
b5d57fc8 BP |
1763 | if (netdev->kbits_rate == kbits_rate && |
1764 | netdev->kbits_burst == kbits_burst) { | |
c9f71668 PS |
1765 | /* Assume that settings haven't changed since we last set them. */ |
1766 | return 0; | |
1767 | } | |
b5d57fc8 | 1768 | netdev->cache_valid &= ~VALID_POLICING; |
80a86fbe BP |
1769 | } |
1770 | ||
ac8c3412 | 1771 | COVERAGE_INC(netdev_set_policing); |
f8500004 | 1772 | /* Remove any existing ingress qdisc. */ |
b5d57fc8 | 1773 | error = tc_add_del_ingress_qdisc(netdev_, false); |
f8500004 JP |
1774 | if (error) { |
1775 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
10a89ef0 | 1776 | netdev_name, ovs_strerror(error)); |
c9f71668 | 1777 | goto out; |
f8500004 JP |
1778 | } |
1779 | ||
8b61709d | 1780 | if (kbits_rate) { |
b5d57fc8 | 1781 | error = tc_add_del_ingress_qdisc(netdev_, true); |
f8500004 JP |
1782 | if (error) { |
1783 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
10a89ef0 | 1784 | netdev_name, ovs_strerror(error)); |
c9f71668 | 1785 | goto out; |
8b61709d BP |
1786 | } |
1787 | ||
b5d57fc8 | 1788 | error = tc_add_policer(netdev_, kbits_rate, kbits_burst); |
f8500004 JP |
1789 | if (error){ |
1790 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
10a89ef0 | 1791 | netdev_name, ovs_strerror(error)); |
c9f71668 | 1792 | goto out; |
8b61709d | 1793 | } |
8b61709d BP |
1794 | } |
1795 | ||
b5d57fc8 BP |
1796 | netdev->kbits_rate = kbits_rate; |
1797 | netdev->kbits_burst = kbits_burst; | |
f8500004 | 1798 | |
c9f71668 PS |
1799 | out: |
1800 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
1801 | netdev->netdev_policing_error = error; |
1802 | netdev->cache_valid |= VALID_POLICING; | |
c9f71668 PS |
1803 | } |
1804 | return error; | |
8b61709d BP |
1805 | } |
1806 | ||
c1c9c9c4 BP |
1807 | static int |
1808 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 1809 | struct sset *types) |
c1c9c9c4 | 1810 | { |
559eb230 | 1811 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
1812 | |
1813 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1814 | const struct tc_ops *ops = *opsp; | |
1815 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 1816 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
1817 | } |
1818 | } | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | static const struct tc_ops * | |
1823 | tc_lookup_ovs_name(const char *name) | |
1824 | { | |
559eb230 | 1825 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
1826 | |
1827 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1828 | const struct tc_ops *ops = *opsp; | |
1829 | if (!strcmp(name, ops->ovs_name)) { | |
1830 | return ops; | |
1831 | } | |
1832 | } | |
1833 | return NULL; | |
1834 | } | |
1835 | ||
1836 | static const struct tc_ops * | |
1837 | tc_lookup_linux_name(const char *name) | |
1838 | { | |
559eb230 | 1839 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
1840 | |
1841 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
1842 | const struct tc_ops *ops = *opsp; | |
1843 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
1844 | return ops; | |
1845 | } | |
1846 | } | |
1847 | return NULL; | |
1848 | } | |
1849 | ||
93b13be8 | 1850 | static struct tc_queue * |
b5d57fc8 | 1851 | tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, |
93b13be8 BP |
1852 | size_t hash) |
1853 | { | |
b5d57fc8 | 1854 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
93b13be8 BP |
1855 | struct tc_queue *queue; |
1856 | ||
b5d57fc8 | 1857 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { |
93b13be8 BP |
1858 | if (queue->queue_id == queue_id) { |
1859 | return queue; | |
1860 | } | |
1861 | } | |
1862 | return NULL; | |
1863 | } | |
1864 | ||
1865 | static struct tc_queue * | |
1866 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
1867 | { | |
1868 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
1869 | } | |
1870 | ||
c1c9c9c4 BP |
1871 | static int |
1872 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
1873 | const char *type, | |
1874 | struct netdev_qos_capabilities *caps) | |
1875 | { | |
1876 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
1877 | if (!ops) { | |
1878 | return EOPNOTSUPP; | |
1879 | } | |
1880 | caps->n_queues = ops->n_queues; | |
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | static int | |
b5d57fc8 | 1885 | netdev_linux_get_qos(const struct netdev *netdev_, |
79f1cbe9 | 1886 | const char **typep, struct smap *details) |
c1c9c9c4 | 1887 | { |
b5d57fc8 | 1888 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1889 | int error; |
1890 | ||
b5d57fc8 | 1891 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
1892 | if (error) { |
1893 | return error; | |
1894 | } | |
1895 | ||
b5d57fc8 BP |
1896 | *typep = netdev->tc->ops->ovs_name; |
1897 | return (netdev->tc->ops->qdisc_get | |
1898 | ? netdev->tc->ops->qdisc_get(netdev_, details) | |
c1c9c9c4 BP |
1899 | : 0); |
1900 | } | |
1901 | ||
1902 | static int | |
b5d57fc8 | 1903 | netdev_linux_set_qos(struct netdev *netdev_, |
79f1cbe9 | 1904 | const char *type, const struct smap *details) |
c1c9c9c4 | 1905 | { |
b5d57fc8 | 1906 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1907 | const struct tc_ops *new_ops; |
1908 | int error; | |
1909 | ||
1910 | new_ops = tc_lookup_ovs_name(type); | |
1911 | if (!new_ops || !new_ops->tc_install) { | |
1912 | return EOPNOTSUPP; | |
1913 | } | |
1914 | ||
b5d57fc8 | 1915 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
1916 | if (error) { |
1917 | return error; | |
1918 | } | |
1919 | ||
b5d57fc8 BP |
1920 | if (new_ops == netdev->tc->ops) { |
1921 | return new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; | |
c1c9c9c4 BP |
1922 | } else { |
1923 | /* Delete existing qdisc. */ | |
b5d57fc8 | 1924 | error = tc_del_qdisc(netdev_); |
c1c9c9c4 BP |
1925 | if (error) { |
1926 | return error; | |
1927 | } | |
b5d57fc8 | 1928 | ovs_assert(netdev->tc == NULL); |
c1c9c9c4 BP |
1929 | |
1930 | /* Install new qdisc. */ | |
b5d57fc8 BP |
1931 | error = new_ops->tc_install(netdev_, details); |
1932 | ovs_assert((error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
1933 | |
1934 | return error; | |
1935 | } | |
1936 | } | |
1937 | ||
1938 | static int | |
b5d57fc8 | 1939 | netdev_linux_get_queue(const struct netdev *netdev_, |
79f1cbe9 | 1940 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 | 1941 | { |
b5d57fc8 | 1942 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1943 | int error; |
1944 | ||
b5d57fc8 | 1945 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
1946 | if (error) { |
1947 | return error; | |
93b13be8 | 1948 | } else { |
b5d57fc8 | 1949 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
93b13be8 | 1950 | return (queue |
b5d57fc8 | 1951 | ? netdev->tc->ops->class_get(netdev_, queue, details) |
93b13be8 | 1952 | : ENOENT); |
c1c9c9c4 | 1953 | } |
c1c9c9c4 BP |
1954 | } |
1955 | ||
1956 | static int | |
b5d57fc8 | 1957 | netdev_linux_set_queue(struct netdev *netdev_, |
79f1cbe9 | 1958 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 | 1959 | { |
b5d57fc8 | 1960 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1961 | int error; |
1962 | ||
b5d57fc8 | 1963 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
1964 | if (error) { |
1965 | return error; | |
b5d57fc8 BP |
1966 | } else if (queue_id >= netdev->tc->ops->n_queues |
1967 | || !netdev->tc->ops->class_set) { | |
c1c9c9c4 BP |
1968 | return EINVAL; |
1969 | } | |
1970 | ||
b5d57fc8 | 1971 | return netdev->tc->ops->class_set(netdev_, queue_id, details); |
c1c9c9c4 BP |
1972 | } |
1973 | ||
1974 | static int | |
b5d57fc8 | 1975 | netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) |
c1c9c9c4 | 1976 | { |
b5d57fc8 | 1977 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1978 | int error; |
1979 | ||
b5d57fc8 | 1980 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
1981 | if (error) { |
1982 | return error; | |
b5d57fc8 | 1983 | } else if (!netdev->tc->ops->class_delete) { |
c1c9c9c4 | 1984 | return EINVAL; |
93b13be8 | 1985 | } else { |
b5d57fc8 | 1986 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
93b13be8 | 1987 | return (queue |
b5d57fc8 | 1988 | ? netdev->tc->ops->class_delete(netdev_, queue) |
93b13be8 | 1989 | : ENOENT); |
c1c9c9c4 | 1990 | } |
c1c9c9c4 BP |
1991 | } |
1992 | ||
1993 | static int | |
b5d57fc8 | 1994 | netdev_linux_get_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
1995 | unsigned int queue_id, |
1996 | struct netdev_queue_stats *stats) | |
1997 | { | |
b5d57fc8 | 1998 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
1999 | int error; |
2000 | ||
b5d57fc8 | 2001 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
2002 | if (error) { |
2003 | return error; | |
b5d57fc8 | 2004 | } else if (!netdev->tc->ops->class_get_stats) { |
c1c9c9c4 | 2005 | return EOPNOTSUPP; |
93b13be8 | 2006 | } else { |
b5d57fc8 | 2007 | const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
6dc34a0d BP |
2008 | if (!queue) { |
2009 | return ENOENT; | |
2010 | } | |
2011 | stats->created = queue->created; | |
2012 | return netdev->tc->ops->class_get_stats(netdev_, queue, stats); | |
c1c9c9c4 | 2013 | } |
c1c9c9c4 BP |
2014 | } |
2015 | ||
23a98ffe | 2016 | static bool |
c1c9c9c4 BP |
2017 | start_queue_dump(const struct netdev *netdev, struct nl_dump *dump) |
2018 | { | |
2019 | struct ofpbuf request; | |
2020 | struct tcmsg *tcmsg; | |
2021 | ||
2022 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); | |
23a98ffe BP |
2023 | if (!tcmsg) { |
2024 | return false; | |
2025 | } | |
3c4de644 | 2026 | tcmsg->tcm_parent = 0; |
a88b4e04 | 2027 | nl_dump_start(dump, NETLINK_ROUTE, &request); |
c1c9c9c4 | 2028 | ofpbuf_uninit(&request); |
23a98ffe | 2029 | return true; |
c1c9c9c4 BP |
2030 | } |
2031 | ||
2032 | static int | |
b5d57fc8 | 2033 | netdev_linux_dump_queues(const struct netdev *netdev_, |
c1c9c9c4 BP |
2034 | netdev_dump_queues_cb *cb, void *aux) |
2035 | { | |
b5d57fc8 | 2036 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f486e840 | 2037 | struct tc_queue *queue, *next_queue; |
79f1cbe9 | 2038 | struct smap details; |
c1c9c9c4 | 2039 | int last_error; |
c1c9c9c4 BP |
2040 | int error; |
2041 | ||
b5d57fc8 | 2042 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
2043 | if (error) { |
2044 | return error; | |
b5d57fc8 | 2045 | } else if (!netdev->tc->ops->class_get) { |
c1c9c9c4 BP |
2046 | return EOPNOTSUPP; |
2047 | } | |
2048 | ||
2049 | last_error = 0; | |
79f1cbe9 | 2050 | smap_init(&details); |
f486e840 | 2051 | HMAP_FOR_EACH_SAFE (queue, next_queue, hmap_node, |
b5d57fc8 | 2052 | &netdev->tc->queues) { |
79f1cbe9 | 2053 | smap_clear(&details); |
c1c9c9c4 | 2054 | |
b5d57fc8 | 2055 | error = netdev->tc->ops->class_get(netdev_, queue, &details); |
c1c9c9c4 | 2056 | if (!error) { |
93b13be8 | 2057 | (*cb)(queue->queue_id, &details, aux); |
c1c9c9c4 BP |
2058 | } else { |
2059 | last_error = error; | |
2060 | } | |
2061 | } | |
79f1cbe9 | 2062 | smap_destroy(&details); |
c1c9c9c4 BP |
2063 | |
2064 | return last_error; | |
2065 | } | |
2066 | ||
2067 | static int | |
b5d57fc8 | 2068 | netdev_linux_dump_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2069 | netdev_dump_queue_stats_cb *cb, void *aux) |
2070 | { | |
b5d57fc8 | 2071 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2072 | struct nl_dump dump; |
2073 | struct ofpbuf msg; | |
2074 | int last_error; | |
2075 | int error; | |
2076 | ||
b5d57fc8 | 2077 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 BP |
2078 | if (error) { |
2079 | return error; | |
b5d57fc8 | 2080 | } else if (!netdev->tc->ops->class_dump_stats) { |
c1c9c9c4 BP |
2081 | return EOPNOTSUPP; |
2082 | } | |
2083 | ||
2084 | last_error = 0; | |
b5d57fc8 | 2085 | if (!start_queue_dump(netdev_, &dump)) { |
23a98ffe BP |
2086 | return ENODEV; |
2087 | } | |
c1c9c9c4 | 2088 | while (nl_dump_next(&dump, &msg)) { |
b5d57fc8 | 2089 | error = netdev->tc->ops->class_dump_stats(netdev_, &msg, cb, aux); |
c1c9c9c4 BP |
2090 | if (error) { |
2091 | last_error = error; | |
2092 | } | |
2093 | } | |
2094 | ||
2095 | error = nl_dump_done(&dump); | |
2096 | return error ? error : last_error; | |
2097 | } | |
2098 | ||
8b61709d | 2099 | static int |
f1acd62b BP |
2100 | netdev_linux_get_in4(const struct netdev *netdev_, |
2101 | struct in_addr *address, struct in_addr *netmask) | |
8b61709d | 2102 | { |
b5d57fc8 | 2103 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
149f577a | 2104 | |
b5d57fc8 | 2105 | if (!(netdev->cache_valid & VALID_IN4)) { |
8b61709d BP |
2106 | int error; |
2107 | ||
b5d57fc8 | 2108 | error = netdev_linux_get_ipv4(netdev_, &netdev->address, |
8b61709d BP |
2109 | SIOCGIFADDR, "SIOCGIFADDR"); |
2110 | if (error) { | |
2111 | return error; | |
2112 | } | |
2113 | ||
b5d57fc8 | 2114 | error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, |
f1acd62b BP |
2115 | SIOCGIFNETMASK, "SIOCGIFNETMASK"); |
2116 | if (error) { | |
2117 | return error; | |
2118 | } | |
2119 | ||
b5d57fc8 | 2120 | netdev->cache_valid |= VALID_IN4; |
8b61709d | 2121 | } |
b5d57fc8 BP |
2122 | *address = netdev->address; |
2123 | *netmask = netdev->netmask; | |
f1acd62b | 2124 | return address->s_addr == INADDR_ANY ? EADDRNOTAVAIL : 0; |
8b61709d BP |
2125 | } |
2126 | ||
8b61709d | 2127 | static int |
f1acd62b BP |
2128 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
2129 | struct in_addr netmask) | |
8b61709d | 2130 | { |
b5d57fc8 | 2131 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2132 | int error; |
2133 | ||
f1acd62b | 2134 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 2135 | if (!error) { |
b5d57fc8 BP |
2136 | netdev->cache_valid |= VALID_IN4; |
2137 | netdev->address = address; | |
2138 | netdev->netmask = netmask; | |
f1acd62b | 2139 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 2140 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 2141 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
2142 | } |
2143 | } | |
2144 | return error; | |
2145 | } | |
2146 | ||
2147 | static bool | |
2148 | parse_if_inet6_line(const char *line, | |
2149 | struct in6_addr *in6, char ifname[16 + 1]) | |
2150 | { | |
2151 | uint8_t *s6 = in6->s6_addr; | |
2152 | #define X8 "%2"SCNx8 | |
2153 | return sscanf(line, | |
2154 | " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 | |
2155 | "%*x %*x %*x %*x %16s\n", | |
2156 | &s6[0], &s6[1], &s6[2], &s6[3], | |
2157 | &s6[4], &s6[5], &s6[6], &s6[7], | |
2158 | &s6[8], &s6[9], &s6[10], &s6[11], | |
2159 | &s6[12], &s6[13], &s6[14], &s6[15], | |
2160 | ifname) == 17; | |
2161 | } | |
2162 | ||
2163 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address (if | |
2164 | * 'in6' is non-null) and returns true. Otherwise, returns false. */ | |
2165 | static int | |
2166 | netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) | |
2167 | { | |
b5d57fc8 BP |
2168 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
2169 | if (!(netdev->cache_valid & VALID_IN6)) { | |
8b61709d BP |
2170 | FILE *file; |
2171 | char line[128]; | |
2172 | ||
b5d57fc8 | 2173 | netdev->in6 = in6addr_any; |
8b61709d BP |
2174 | |
2175 | file = fopen("/proc/net/if_inet6", "r"); | |
2176 | if (file != NULL) { | |
2177 | const char *name = netdev_get_name(netdev_); | |
2178 | while (fgets(line, sizeof line, file)) { | |
2a022368 | 2179 | struct in6_addr in6_tmp; |
8b61709d | 2180 | char ifname[16 + 1]; |
2a022368 | 2181 | if (parse_if_inet6_line(line, &in6_tmp, ifname) |
8b61709d BP |
2182 | && !strcmp(name, ifname)) |
2183 | { | |
b5d57fc8 | 2184 | netdev->in6 = in6_tmp; |
8b61709d BP |
2185 | break; |
2186 | } | |
2187 | } | |
2188 | fclose(file); | |
2189 | } | |
b5d57fc8 | 2190 | netdev->cache_valid |= VALID_IN6; |
8b61709d | 2191 | } |
b5d57fc8 | 2192 | *in6 = netdev->in6; |
8b61709d BP |
2193 | return 0; |
2194 | } | |
2195 | ||
2196 | static void | |
2197 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
2198 | { | |
2199 | struct sockaddr_in sin; | |
2200 | memset(&sin, 0, sizeof sin); | |
2201 | sin.sin_family = AF_INET; | |
2202 | sin.sin_addr = addr; | |
2203 | sin.sin_port = 0; | |
2204 | ||
2205 | memset(sa, 0, sizeof *sa); | |
2206 | memcpy(sa, &sin, sizeof sin); | |
2207 | } | |
2208 | ||
2209 | static int | |
2210 | do_set_addr(struct netdev *netdev, | |
2211 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
2212 | { | |
2213 | struct ifreq ifr; | |
71d7c22f | 2214 | ovs_strzcpy(ifr.ifr_name, netdev_get_name(netdev), sizeof ifr.ifr_name); |
8b61709d | 2215 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
149f577a JG |
2216 | |
2217 | return netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
2218 | ioctl_name); | |
8b61709d BP |
2219 | } |
2220 | ||
2221 | /* Adds 'router' as a default IP gateway. */ | |
2222 | static int | |
67a4917b | 2223 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
2224 | { |
2225 | struct in_addr any = { INADDR_ANY }; | |
2226 | struct rtentry rt; | |
2227 | int error; | |
2228 | ||
2229 | memset(&rt, 0, sizeof rt); | |
2230 | make_in4_sockaddr(&rt.rt_dst, any); | |
2231 | make_in4_sockaddr(&rt.rt_gateway, router); | |
2232 | make_in4_sockaddr(&rt.rt_genmask, any); | |
2233 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
8b61709d BP |
2234 | error = ioctl(af_inet_sock, SIOCADDRT, &rt) < 0 ? errno : 0; |
2235 | if (error) { | |
10a89ef0 | 2236 | VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); |
8b61709d BP |
2237 | } |
2238 | return error; | |
2239 | } | |
2240 | ||
f1acd62b BP |
2241 | static int |
2242 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
2243 | char **netdev_name) | |
2244 | { | |
2245 | static const char fn[] = "/proc/net/route"; | |
2246 | FILE *stream; | |
2247 | char line[256]; | |
2248 | int ln; | |
2249 | ||
2250 | *netdev_name = NULL; | |
2251 | stream = fopen(fn, "r"); | |
2252 | if (stream == NULL) { | |
10a89ef0 | 2253 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
f1acd62b BP |
2254 | return errno; |
2255 | } | |
2256 | ||
2257 | ln = 0; | |
2258 | while (fgets(line, sizeof line, stream)) { | |
2259 | if (++ln >= 2) { | |
2260 | char iface[17]; | |
dbba996b | 2261 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
2262 | int refcnt, metric, mtu; |
2263 | unsigned int flags, use, window, irtt; | |
2264 | ||
2265 | if (sscanf(line, | |
2266 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
2267 | " %d %u %u\n", | |
2268 | iface, &dest, &gateway, &flags, &refcnt, | |
2269 | &use, &metric, &mask, &mtu, &window, &irtt) != 11) { | |
2270 | ||
d295e8e9 | 2271 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
2272 | fn, ln, line); |
2273 | continue; | |
2274 | } | |
2275 | if (!(flags & RTF_UP)) { | |
2276 | /* Skip routes that aren't up. */ | |
2277 | continue; | |
2278 | } | |
2279 | ||
2280 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 2281 | * network byte order, so we don't need need any endian |
f1acd62b BP |
2282 | * conversions here. */ |
2283 | if ((dest & mask) == (host->s_addr & mask)) { | |
2284 | if (!gateway) { | |
2285 | /* The host is directly reachable. */ | |
2286 | next_hop->s_addr = 0; | |
2287 | } else { | |
2288 | /* To reach the host, we must go through a gateway. */ | |
2289 | next_hop->s_addr = gateway; | |
2290 | } | |
2291 | *netdev_name = xstrdup(iface); | |
2292 | fclose(stream); | |
2293 | return 0; | |
2294 | } | |
2295 | } | |
2296 | } | |
2297 | ||
2298 | fclose(stream); | |
2299 | return ENXIO; | |
2300 | } | |
2301 | ||
e210037e | 2302 | static int |
b5d57fc8 | 2303 | netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) |
e210037e | 2304 | { |
b5d57fc8 | 2305 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
275707c3 EJ |
2306 | int error = 0; |
2307 | ||
b5d57fc8 BP |
2308 | if (!(netdev->cache_valid & VALID_DRVINFO)) { |
2309 | struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; | |
275707c3 EJ |
2310 | |
2311 | COVERAGE_INC(netdev_get_ethtool); | |
b5d57fc8 BP |
2312 | memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); |
2313 | error = netdev_linux_do_ethtool(netdev->up.name, | |
275707c3 EJ |
2314 | cmd, |
2315 | ETHTOOL_GDRVINFO, | |
2316 | "ETHTOOL_GDRVINFO"); | |
2317 | if (!error) { | |
b5d57fc8 | 2318 | netdev->cache_valid |= VALID_DRVINFO; |
275707c3 EJ |
2319 | } |
2320 | } | |
e210037e | 2321 | |
e210037e | 2322 | if (!error) { |
b5d57fc8 BP |
2323 | smap_add(smap, "driver_name", netdev->drvinfo.driver); |
2324 | smap_add(smap, "driver_version", netdev->drvinfo.version); | |
2325 | smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); | |
e210037e | 2326 | } |
e210037e AE |
2327 | return error; |
2328 | } | |
2329 | ||
4f925bd3 | 2330 | static int |
275707c3 EJ |
2331 | netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, |
2332 | struct smap *smap) | |
4f925bd3 | 2333 | { |
79f1cbe9 | 2334 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
2335 | return 0; |
2336 | } | |
2337 | ||
8b61709d BP |
2338 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2339 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2340 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2341 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2342 | static int | |
2343 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
dbba996b | 2344 | ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) |
8b61709d BP |
2345 | { |
2346 | struct arpreq r; | |
c100e025 | 2347 | struct sockaddr_in sin; |
8b61709d BP |
2348 | int retval; |
2349 | ||
2350 | memset(&r, 0, sizeof r); | |
f2cc621b | 2351 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2352 | sin.sin_family = AF_INET; |
2353 | sin.sin_addr.s_addr = ip; | |
2354 | sin.sin_port = 0; | |
2355 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2356 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2357 | r.arp_flags = 0; | |
71d7c22f | 2358 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d BP |
2359 | COVERAGE_INC(netdev_arp_lookup); |
2360 | retval = ioctl(af_inet_sock, SIOCGARP, &r) < 0 ? errno : 0; | |
2361 | if (!retval) { | |
2362 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2363 | } else if (retval != ENXIO) { | |
2364 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
10a89ef0 BP |
2365 | netdev_get_name(netdev), IP_ARGS(ip), |
2366 | ovs_strerror(retval)); | |
8b61709d BP |
2367 | } |
2368 | return retval; | |
2369 | } | |
2370 | ||
2371 | static int | |
2372 | nd_to_iff_flags(enum netdev_flags nd) | |
2373 | { | |
2374 | int iff = 0; | |
2375 | if (nd & NETDEV_UP) { | |
2376 | iff |= IFF_UP; | |
2377 | } | |
2378 | if (nd & NETDEV_PROMISC) { | |
2379 | iff |= IFF_PROMISC; | |
2380 | } | |
2381 | return iff; | |
2382 | } | |
2383 | ||
2384 | static int | |
2385 | iff_to_nd_flags(int iff) | |
2386 | { | |
2387 | enum netdev_flags nd = 0; | |
2388 | if (iff & IFF_UP) { | |
2389 | nd |= NETDEV_UP; | |
2390 | } | |
2391 | if (iff & IFF_PROMISC) { | |
2392 | nd |= NETDEV_PROMISC; | |
2393 | } | |
2394 | return nd; | |
2395 | } | |
2396 | ||
2397 | static int | |
b5d57fc8 | 2398 | netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, |
8b61709d BP |
2399 | enum netdev_flags on, enum netdev_flags *old_flagsp) |
2400 | { | |
b5d57fc8 | 2401 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 2402 | int old_flags, new_flags; |
c37d4da4 EJ |
2403 | int error = 0; |
2404 | ||
b5d57fc8 | 2405 | old_flags = netdev->ifi_flags; |
c37d4da4 EJ |
2406 | *old_flagsp = iff_to_nd_flags(old_flags); |
2407 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2408 | if (new_flags != old_flags) { | |
b5d57fc8 BP |
2409 | error = set_flags(netdev_get_name(netdev_), new_flags); |
2410 | get_flags(netdev_, &netdev->ifi_flags); | |
8b61709d BP |
2411 | } |
2412 | return error; | |
2413 | } | |
2414 | ||
ac4d3bcb EJ |
2415 | static unsigned int |
2416 | netdev_linux_change_seq(const struct netdev *netdev) | |
2417 | { | |
b5d57fc8 | 2418 | return netdev_linux_cast(netdev)->change_seq; |
ac4d3bcb EJ |
2419 | } |
2420 | ||
4f925bd3 | 2421 | #define NETDEV_LINUX_CLASS(NAME, CREATE, GET_STATS, SET_STATS, \ |
51f87458 | 2422 | GET_FEATURES, GET_STATUS) \ |
c3827f61 BP |
2423 | { \ |
2424 | NAME, \ | |
2425 | \ | |
2426 | netdev_linux_init, \ | |
2427 | netdev_linux_run, \ | |
2428 | netdev_linux_wait, \ | |
2429 | \ | |
2430 | CREATE, \ | |
2431 | netdev_linux_destroy, \ | |
de5cdb90 | 2432 | NULL, /* get_config */ \ |
6d9e6eb4 | 2433 | NULL, /* set_config */ \ |
f431bf7d | 2434 | NULL, /* get_tunnel_config */ \ |
c3827f61 | 2435 | \ |
796223f5 | 2436 | netdev_linux_rx_open, \ |
c3827f61 BP |
2437 | \ |
2438 | netdev_linux_send, \ | |
2439 | netdev_linux_send_wait, \ | |
2440 | \ | |
2441 | netdev_linux_set_etheraddr, \ | |
2442 | netdev_linux_get_etheraddr, \ | |
2443 | netdev_linux_get_mtu, \ | |
9b020780 | 2444 | netdev_linux_set_mtu, \ |
c3827f61 BP |
2445 | netdev_linux_get_ifindex, \ |
2446 | netdev_linux_get_carrier, \ | |
65c3058c | 2447 | netdev_linux_get_carrier_resets, \ |
1670c579 | 2448 | netdev_linux_set_miimon_interval, \ |
f613a0d7 | 2449 | GET_STATS, \ |
c3827f61 BP |
2450 | SET_STATS, \ |
2451 | \ | |
51f87458 | 2452 | GET_FEATURES, \ |
c3827f61 | 2453 | netdev_linux_set_advertisements, \ |
c3827f61 BP |
2454 | \ |
2455 | netdev_linux_set_policing, \ | |
2456 | netdev_linux_get_qos_types, \ | |
2457 | netdev_linux_get_qos_capabilities, \ | |
2458 | netdev_linux_get_qos, \ | |
2459 | netdev_linux_set_qos, \ | |
2460 | netdev_linux_get_queue, \ | |
2461 | netdev_linux_set_queue, \ | |
2462 | netdev_linux_delete_queue, \ | |
2463 | netdev_linux_get_queue_stats, \ | |
2464 | netdev_linux_dump_queues, \ | |
2465 | netdev_linux_dump_queue_stats, \ | |
2466 | \ | |
2467 | netdev_linux_get_in4, \ | |
2468 | netdev_linux_set_in4, \ | |
2469 | netdev_linux_get_in6, \ | |
2470 | netdev_linux_add_router, \ | |
2471 | netdev_linux_get_next_hop, \ | |
4f925bd3 | 2472 | GET_STATUS, \ |
c3827f61 BP |
2473 | netdev_linux_arp_lookup, \ |
2474 | \ | |
2475 | netdev_linux_update_flags, \ | |
2476 | \ | |
ac4d3bcb | 2477 | netdev_linux_change_seq \ |
c3827f61 BP |
2478 | } |
2479 | ||
2480 | const struct netdev_class netdev_linux_class = | |
2481 | NETDEV_LINUX_CLASS( | |
2482 | "system", | |
2483 | netdev_linux_create, | |
f613a0d7 | 2484 | netdev_linux_get_stats, |
4f925bd3 | 2485 | NULL, /* set_stats */ |
51f87458 | 2486 | netdev_linux_get_features, |
275707c3 | 2487 | netdev_linux_get_status); |
c3827f61 BP |
2488 | |
2489 | const struct netdev_class netdev_tap_class = | |
2490 | NETDEV_LINUX_CLASS( | |
2491 | "tap", | |
2492 | netdev_linux_create_tap, | |
bba1e6f3 | 2493 | netdev_tap_get_stats, |
4f925bd3 | 2494 | NULL, /* set_stats */ |
51f87458 | 2495 | netdev_linux_get_features, |
275707c3 | 2496 | netdev_linux_get_status); |
c3827f61 BP |
2497 | |
2498 | const struct netdev_class netdev_internal_class = | |
2499 | NETDEV_LINUX_CLASS( | |
2500 | "internal", | |
2501 | netdev_linux_create, | |
bba1e6f3 | 2502 | netdev_internal_get_stats, |
2f31a822 | 2503 | netdev_internal_set_stats, |
51f87458 | 2504 | NULL, /* get_features */ |
275707c3 | 2505 | netdev_internal_get_status); |
796223f5 BP |
2506 | |
2507 | static const struct netdev_rx_class netdev_rx_linux_class = { | |
2508 | netdev_rx_linux_destroy, | |
2509 | netdev_rx_linux_recv, | |
2510 | netdev_rx_linux_wait, | |
2511 | netdev_rx_linux_drain, | |
2512 | }; | |
8b61709d | 2513 | \f |
c1c9c9c4 | 2514 | /* HTB traffic control class. */ |
559843ed | 2515 | |
c1c9c9c4 | 2516 | #define HTB_N_QUEUES 0xf000 |
8b61709d | 2517 | |
c1c9c9c4 BP |
2518 | struct htb { |
2519 | struct tc tc; | |
2520 | unsigned int max_rate; /* In bytes/s. */ | |
2521 | }; | |
8b61709d | 2522 | |
c1c9c9c4 | 2523 | struct htb_class { |
93b13be8 | 2524 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
2525 | unsigned int min_rate; /* In bytes/s. */ |
2526 | unsigned int max_rate; /* In bytes/s. */ | |
2527 | unsigned int burst; /* In bytes. */ | |
2528 | unsigned int priority; /* Lower values are higher priorities. */ | |
2529 | }; | |
8b61709d | 2530 | |
c1c9c9c4 | 2531 | static struct htb * |
b5d57fc8 | 2532 | htb_get__(const struct netdev *netdev_) |
c1c9c9c4 | 2533 | { |
b5d57fc8 BP |
2534 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
2535 | return CONTAINER_OF(netdev->tc, struct htb, tc); | |
c1c9c9c4 BP |
2536 | } |
2537 | ||
24045e35 | 2538 | static void |
b5d57fc8 | 2539 | htb_install__(struct netdev *netdev_, uint64_t max_rate) |
c1c9c9c4 | 2540 | { |
b5d57fc8 | 2541 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2542 | struct htb *htb; |
2543 | ||
2544 | htb = xmalloc(sizeof *htb); | |
2545 | tc_init(&htb->tc, &tc_ops_htb); | |
2546 | htb->max_rate = max_rate; | |
2547 | ||
b5d57fc8 | 2548 | netdev->tc = &htb->tc; |
c1c9c9c4 BP |
2549 | } |
2550 | ||
2551 | /* Create an HTB qdisc. | |
2552 | * | |
a339aa81 | 2553 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
2554 | static int |
2555 | htb_setup_qdisc__(struct netdev *netdev) | |
2556 | { | |
2557 | size_t opt_offset; | |
2558 | struct tc_htb_glob opt; | |
2559 | struct ofpbuf request; | |
2560 | struct tcmsg *tcmsg; | |
2561 | ||
2562 | tc_del_qdisc(netdev); | |
2563 | ||
2564 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
2565 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
2566 | if (!tcmsg) { |
2567 | return ENODEV; | |
2568 | } | |
c1c9c9c4 BP |
2569 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
2570 | tcmsg->tcm_parent = TC_H_ROOT; | |
2571 | ||
2572 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2573 | ||
2574 | memset(&opt, 0, sizeof opt); | |
2575 | opt.rate2quantum = 10; | |
2576 | opt.version = 3; | |
4ecf12d5 | 2577 | opt.defcls = 1; |
c1c9c9c4 BP |
2578 | |
2579 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2580 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
2581 | nl_msg_end_nested(&request, opt_offset); | |
2582 | ||
2583 | return tc_transact(&request, NULL); | |
2584 | } | |
2585 | ||
2586 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
2587 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
2588 | static int | |
2589 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
2590 | unsigned int parent, struct htb_class *class) | |
2591 | { | |
2592 | size_t opt_offset; | |
2593 | struct tc_htb_opt opt; | |
2594 | struct ofpbuf request; | |
2595 | struct tcmsg *tcmsg; | |
2596 | int error; | |
2597 | int mtu; | |
2598 | ||
9b020780 PS |
2599 | error = netdev_get_mtu(netdev, &mtu); |
2600 | if (error) { | |
f915f1a8 BP |
2601 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
2602 | netdev_get_name(netdev)); | |
9b020780 | 2603 | return error; |
f915f1a8 | 2604 | } |
c1c9c9c4 BP |
2605 | |
2606 | memset(&opt, 0, sizeof opt); | |
2607 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
2608 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
2609 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); | |
2610 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
2611 | opt.prio = class->priority; | |
2612 | ||
2613 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
23a98ffe BP |
2614 | if (!tcmsg) { |
2615 | return ENODEV; | |
2616 | } | |
c1c9c9c4 BP |
2617 | tcmsg->tcm_handle = handle; |
2618 | tcmsg->tcm_parent = parent; | |
2619 | ||
2620 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
2621 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2622 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
2623 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
2624 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
2625 | nl_msg_end_nested(&request, opt_offset); | |
2626 | ||
2627 | error = tc_transact(&request, NULL); | |
2628 | if (error) { | |
2629 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
2630 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
2631 | netdev_get_name(netdev), | |
2632 | tc_get_major(handle), tc_get_minor(handle), | |
2633 | tc_get_major(parent), tc_get_minor(parent), | |
2634 | class->min_rate, class->max_rate, | |
10a89ef0 | 2635 | class->burst, class->priority, ovs_strerror(error)); |
c1c9c9c4 BP |
2636 | } |
2637 | return error; | |
2638 | } | |
2639 | ||
2640 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
2641 | * description of them into 'details'. The description complies with the | |
2642 | * specification given in the vswitch database documentation for linux-htb | |
2643 | * queue details. */ | |
2644 | static int | |
2645 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
2646 | { | |
2647 | static const struct nl_policy tca_htb_policy[] = { | |
2648 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
2649 | .min_len = sizeof(struct tc_htb_opt) }, | |
2650 | }; | |
2651 | ||
2652 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
2653 | const struct tc_htb_opt *htb; | |
2654 | ||
2655 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
2656 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
2657 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
2658 | return EPROTO; | |
2659 | } | |
2660 | ||
2661 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
2662 | class->min_rate = htb->rate.rate; | |
2663 | class->max_rate = htb->ceil.rate; | |
2664 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
2665 | class->priority = htb->prio; | |
2666 | return 0; | |
2667 | } | |
2668 | ||
2669 | static int | |
2670 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
2671 | struct htb_class *options, | |
2672 | struct netdev_queue_stats *stats) | |
2673 | { | |
2674 | struct nlattr *nl_options; | |
2675 | unsigned int handle; | |
2676 | int error; | |
2677 | ||
2678 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
2679 | if (!error && queue_id) { | |
17ee3c1f BP |
2680 | unsigned int major = tc_get_major(handle); |
2681 | unsigned int minor = tc_get_minor(handle); | |
2682 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
2683 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
2684 | } else { |
2685 | error = EPROTO; | |
2686 | } | |
2687 | } | |
2688 | if (!error && options) { | |
2689 | error = htb_parse_tca_options__(nl_options, options); | |
2690 | } | |
2691 | return error; | |
2692 | } | |
2693 | ||
2694 | static void | |
2695 | htb_parse_qdisc_details__(struct netdev *netdev, | |
79f1cbe9 | 2696 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
2697 | { |
2698 | const char *max_rate_s; | |
2699 | ||
79f1cbe9 | 2700 | max_rate_s = smap_get(details, "max-rate"); |
c1c9c9c4 BP |
2701 | hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
2702 | if (!hc->max_rate) { | |
a00ca915 | 2703 | enum netdev_features current; |
c1c9c9c4 BP |
2704 | |
2705 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
d02a5f8e | 2706 | hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
c1c9c9c4 BP |
2707 | } |
2708 | hc->min_rate = hc->max_rate; | |
2709 | hc->burst = 0; | |
2710 | hc->priority = 0; | |
2711 | } | |
2712 | ||
2713 | static int | |
2714 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 2715 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
2716 | { |
2717 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 EJ |
2718 | const char *min_rate_s = smap_get(details, "min-rate"); |
2719 | const char *max_rate_s = smap_get(details, "max-rate"); | |
2720 | const char *burst_s = smap_get(details, "burst"); | |
2721 | const char *priority_s = smap_get(details, "priority"); | |
9b020780 | 2722 | int mtu, error; |
c1c9c9c4 | 2723 | |
9b020780 PS |
2724 | error = netdev_get_mtu(netdev, &mtu); |
2725 | if (error) { | |
f915f1a8 BP |
2726 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
2727 | netdev_get_name(netdev)); | |
9b020780 | 2728 | return error; |
f915f1a8 BP |
2729 | } |
2730 | ||
4f104611 EJ |
2731 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
2732 | * on uncongested links. */ | |
c45ab5e9 | 2733 | hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
4f104611 | 2734 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
2735 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
2736 | ||
2737 | /* max-rate */ | |
2738 | hc->max_rate = (max_rate_s | |
2739 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
2740 | : htb->max_rate); | |
2741 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); | |
2742 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
2743 | ||
2744 | /* burst | |
2745 | * | |
2746 | * According to hints in the documentation that I've read, it is important | |
2747 | * that 'burst' be at least as big as the largest frame that might be | |
2748 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
2749 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
2750 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
2751 | * the MTU. We actually add 64, instead of 14, as a guard against | |
2752 | * additional headers get tacked on somewhere that we're not aware of. */ | |
c1c9c9c4 BP |
2753 | hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; |
2754 | hc->burst = MAX(hc->burst, mtu + 64); | |
2755 | ||
2756 | /* priority */ | |
2757 | hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; | |
2758 | ||
2759 | return 0; | |
2760 | } | |
2761 | ||
2762 | static int | |
2763 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
2764 | unsigned int parent, struct htb_class *options, | |
2765 | struct netdev_queue_stats *stats) | |
2766 | { | |
2767 | struct ofpbuf *reply; | |
2768 | int error; | |
2769 | ||
2770 | error = tc_query_class(netdev, handle, parent, &reply); | |
2771 | if (!error) { | |
2772 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
2773 | ofpbuf_delete(reply); | |
2774 | } | |
2775 | return error; | |
2776 | } | |
2777 | ||
2778 | static int | |
79f1cbe9 | 2779 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
2780 | { |
2781 | int error; | |
2782 | ||
2783 | error = htb_setup_qdisc__(netdev); | |
2784 | if (!error) { | |
2785 | struct htb_class hc; | |
2786 | ||
2787 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2788 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2789 | tc_make_handle(1, 0), &hc); | |
2790 | if (!error) { | |
2791 | htb_install__(netdev, hc.max_rate); | |
2792 | } | |
2793 | } | |
2794 | return error; | |
2795 | } | |
2796 | ||
93b13be8 BP |
2797 | static struct htb_class * |
2798 | htb_class_cast__(const struct tc_queue *queue) | |
2799 | { | |
2800 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
2801 | } | |
2802 | ||
c1c9c9c4 BP |
2803 | static void |
2804 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
2805 | const struct htb_class *hc) | |
2806 | { | |
2807 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
2808 | size_t hash = hash_int(queue_id, 0); |
2809 | struct tc_queue *queue; | |
c1c9c9c4 BP |
2810 | struct htb_class *hcp; |
2811 | ||
93b13be8 BP |
2812 | queue = tc_find_queue__(netdev, queue_id, hash); |
2813 | if (queue) { | |
2814 | hcp = htb_class_cast__(queue); | |
2815 | } else { | |
c1c9c9c4 | 2816 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
2817 | queue = &hcp->tc_queue; |
2818 | queue->queue_id = queue_id; | |
6dc34a0d | 2819 | queue->created = time_msec(); |
93b13be8 | 2820 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); |
c1c9c9c4 | 2821 | } |
93b13be8 BP |
2822 | |
2823 | hcp->min_rate = hc->min_rate; | |
2824 | hcp->max_rate = hc->max_rate; | |
2825 | hcp->burst = hc->burst; | |
2826 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
2827 | } |
2828 | ||
2829 | static int | |
2830 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
2831 | { | |
c1c9c9c4 BP |
2832 | struct ofpbuf msg; |
2833 | struct nl_dump dump; | |
2834 | struct htb_class hc; | |
c1c9c9c4 BP |
2835 | |
2836 | /* Get qdisc options. */ | |
2837 | hc.max_rate = 0; | |
2838 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 2839 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
2840 | |
2841 | /* Get queues. */ | |
23a98ffe BP |
2842 | if (!start_queue_dump(netdev, &dump)) { |
2843 | return ENODEV; | |
2844 | } | |
c1c9c9c4 BP |
2845 | while (nl_dump_next(&dump, &msg)) { |
2846 | unsigned int queue_id; | |
2847 | ||
2848 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
2849 | htb_update_queue__(netdev, queue_id, &hc); | |
2850 | } | |
2851 | } | |
2852 | nl_dump_done(&dump); | |
2853 | ||
2854 | return 0; | |
2855 | } | |
2856 | ||
2857 | static void | |
2858 | htb_tc_destroy(struct tc *tc) | |
2859 | { | |
2860 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
93b13be8 | 2861 | struct htb_class *hc, *next; |
c1c9c9c4 | 2862 | |
4e8e4213 | 2863 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { |
93b13be8 | 2864 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 BP |
2865 | free(hc); |
2866 | } | |
2867 | tc_destroy(tc); | |
2868 | free(htb); | |
2869 | } | |
2870 | ||
2871 | static int | |
79f1cbe9 | 2872 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
2873 | { |
2874 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 2875 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
2876 | return 0; |
2877 | } | |
2878 | ||
2879 | static int | |
79f1cbe9 | 2880 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
2881 | { |
2882 | struct htb_class hc; | |
2883 | int error; | |
2884 | ||
2885 | htb_parse_qdisc_details__(netdev, details, &hc); | |
2886 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
2887 | tc_make_handle(1, 0), &hc); | |
2888 | if (!error) { | |
2889 | htb_get__(netdev)->max_rate = hc.max_rate; | |
2890 | } | |
2891 | return error; | |
2892 | } | |
2893 | ||
2894 | static int | |
93b13be8 | 2895 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 2896 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 2897 | { |
93b13be8 | 2898 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2899 | |
79f1cbe9 | 2900 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 2901 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 2902 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 2903 | } |
79f1cbe9 | 2904 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 2905 | if (hc->priority) { |
79f1cbe9 | 2906 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
2907 | } |
2908 | return 0; | |
2909 | } | |
2910 | ||
2911 | static int | |
2912 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 2913 | const struct smap *details) |
c1c9c9c4 BP |
2914 | { |
2915 | struct htb_class hc; | |
2916 | int error; | |
2917 | ||
2918 | error = htb_parse_class_details__(netdev, details, &hc); | |
2919 | if (error) { | |
2920 | return error; | |
2921 | } | |
2922 | ||
17ee3c1f | 2923 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
2924 | tc_make_handle(1, 0xfffe), &hc); |
2925 | if (error) { | |
2926 | return error; | |
2927 | } | |
2928 | ||
2929 | htb_update_queue__(netdev, queue_id, &hc); | |
2930 | return 0; | |
2931 | } | |
2932 | ||
2933 | static int | |
93b13be8 | 2934 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 2935 | { |
93b13be8 | 2936 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 2937 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
2938 | int error; |
2939 | ||
93b13be8 | 2940 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 2941 | if (!error) { |
93b13be8 | 2942 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 2943 | free(hc); |
c1c9c9c4 BP |
2944 | } |
2945 | return error; | |
2946 | } | |
2947 | ||
2948 | static int | |
93b13be8 | 2949 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
2950 | struct netdev_queue_stats *stats) |
2951 | { | |
93b13be8 | 2952 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
2953 | tc_make_handle(1, 0xfffe), NULL, stats); |
2954 | } | |
2955 | ||
2956 | static int | |
2957 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
2958 | const struct ofpbuf *nlmsg, | |
2959 | netdev_dump_queue_stats_cb *cb, void *aux) | |
2960 | { | |
2961 | struct netdev_queue_stats stats; | |
17ee3c1f | 2962 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
2963 | int error; |
2964 | ||
2965 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
2966 | if (error) { | |
2967 | return error; | |
2968 | } | |
2969 | ||
17ee3c1f BP |
2970 | major = tc_get_major(handle); |
2971 | minor = tc_get_minor(handle); | |
2972 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 2973 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
2974 | } |
2975 | return 0; | |
2976 | } | |
2977 | ||
2978 | static const struct tc_ops tc_ops_htb = { | |
2979 | "htb", /* linux_name */ | |
2980 | "linux-htb", /* ovs_name */ | |
2981 | HTB_N_QUEUES, /* n_queues */ | |
2982 | htb_tc_install, | |
2983 | htb_tc_load, | |
2984 | htb_tc_destroy, | |
2985 | htb_qdisc_get, | |
2986 | htb_qdisc_set, | |
2987 | htb_class_get, | |
2988 | htb_class_set, | |
2989 | htb_class_delete, | |
2990 | htb_class_get_stats, | |
2991 | htb_class_dump_stats | |
2992 | }; | |
2993 | \f | |
a339aa81 EJ |
2994 | /* "linux-hfsc" traffic control class. */ |
2995 | ||
2996 | #define HFSC_N_QUEUES 0xf000 | |
2997 | ||
2998 | struct hfsc { | |
2999 | struct tc tc; | |
3000 | uint32_t max_rate; | |
3001 | }; | |
3002 | ||
3003 | struct hfsc_class { | |
3004 | struct tc_queue tc_queue; | |
3005 | uint32_t min_rate; | |
3006 | uint32_t max_rate; | |
3007 | }; | |
3008 | ||
3009 | static struct hfsc * | |
b5d57fc8 | 3010 | hfsc_get__(const struct netdev *netdev_) |
a339aa81 | 3011 | { |
b5d57fc8 BP |
3012 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
3013 | return CONTAINER_OF(netdev->tc, struct hfsc, tc); | |
a339aa81 EJ |
3014 | } |
3015 | ||
3016 | static struct hfsc_class * | |
3017 | hfsc_class_cast__(const struct tc_queue *queue) | |
3018 | { | |
3019 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
3020 | } | |
3021 | ||
24045e35 | 3022 | static void |
b5d57fc8 | 3023 | hfsc_install__(struct netdev *netdev_, uint32_t max_rate) |
a339aa81 | 3024 | { |
b5d57fc8 | 3025 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
3026 | struct hfsc *hfsc; |
3027 | ||
a339aa81 EJ |
3028 | hfsc = xmalloc(sizeof *hfsc); |
3029 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
3030 | hfsc->max_rate = max_rate; | |
b5d57fc8 | 3031 | netdev->tc = &hfsc->tc; |
a339aa81 EJ |
3032 | } |
3033 | ||
3034 | static void | |
3035 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
3036 | const struct hfsc_class *hc) | |
3037 | { | |
3038 | size_t hash; | |
3039 | struct hfsc *hfsc; | |
3040 | struct hfsc_class *hcp; | |
3041 | struct tc_queue *queue; | |
3042 | ||
3043 | hfsc = hfsc_get__(netdev); | |
3044 | hash = hash_int(queue_id, 0); | |
3045 | ||
3046 | queue = tc_find_queue__(netdev, queue_id, hash); | |
3047 | if (queue) { | |
3048 | hcp = hfsc_class_cast__(queue); | |
3049 | } else { | |
3050 | hcp = xmalloc(sizeof *hcp); | |
3051 | queue = &hcp->tc_queue; | |
3052 | queue->queue_id = queue_id; | |
6dc34a0d | 3053 | queue->created = time_msec(); |
a339aa81 EJ |
3054 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); |
3055 | } | |
3056 | ||
3057 | hcp->min_rate = hc->min_rate; | |
3058 | hcp->max_rate = hc->max_rate; | |
3059 | } | |
3060 | ||
3061 | static int | |
3062 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
3063 | { | |
3064 | const struct tc_service_curve *rsc, *fsc, *usc; | |
3065 | static const struct nl_policy tca_hfsc_policy[] = { | |
3066 | [TCA_HFSC_RSC] = { | |
3067 | .type = NL_A_UNSPEC, | |
3068 | .optional = false, | |
3069 | .min_len = sizeof(struct tc_service_curve), | |
3070 | }, | |
3071 | [TCA_HFSC_FSC] = { | |
3072 | .type = NL_A_UNSPEC, | |
3073 | .optional = false, | |
3074 | .min_len = sizeof(struct tc_service_curve), | |
3075 | }, | |
3076 | [TCA_HFSC_USC] = { | |
3077 | .type = NL_A_UNSPEC, | |
3078 | .optional = false, | |
3079 | .min_len = sizeof(struct tc_service_curve), | |
3080 | }, | |
3081 | }; | |
3082 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
3083 | ||
3084 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
3085 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
3086 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
3087 | return EPROTO; | |
3088 | } | |
3089 | ||
3090 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
3091 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
3092 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
3093 | ||
3094 | if (rsc->m1 != 0 || rsc->d != 0 || | |
3095 | fsc->m1 != 0 || fsc->d != 0 || | |
3096 | usc->m1 != 0 || usc->d != 0) { | |
3097 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3098 | "Non-linear service curves are not supported."); | |
3099 | return EPROTO; | |
3100 | } | |
3101 | ||
3102 | if (rsc->m2 != fsc->m2) { | |
3103 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3104 | "Real-time service curves are not supported "); | |
3105 | return EPROTO; | |
3106 | } | |
3107 | ||
3108 | if (rsc->m2 > usc->m2) { | |
3109 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
3110 | "Min-rate service curve is greater than " | |
3111 | "the max-rate service curve."); | |
3112 | return EPROTO; | |
3113 | } | |
3114 | ||
3115 | class->min_rate = fsc->m2; | |
3116 | class->max_rate = usc->m2; | |
3117 | return 0; | |
3118 | } | |
3119 | ||
3120 | static int | |
3121 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
3122 | struct hfsc_class *options, | |
3123 | struct netdev_queue_stats *stats) | |
3124 | { | |
3125 | int error; | |
3126 | unsigned int handle; | |
3127 | struct nlattr *nl_options; | |
3128 | ||
3129 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
3130 | if (error) { | |
3131 | return error; | |
3132 | } | |
3133 | ||
3134 | if (queue_id) { | |
3135 | unsigned int major, minor; | |
3136 | ||
3137 | major = tc_get_major(handle); | |
3138 | minor = tc_get_minor(handle); | |
3139 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
3140 | *queue_id = minor - 1; | |
3141 | } else { | |
3142 | return EPROTO; | |
3143 | } | |
3144 | } | |
3145 | ||
3146 | if (options) { | |
3147 | error = hfsc_parse_tca_options__(nl_options, options); | |
3148 | } | |
3149 | ||
3150 | return error; | |
3151 | } | |
3152 | ||
3153 | static int | |
3154 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
3155 | unsigned int parent, struct hfsc_class *options, | |
3156 | struct netdev_queue_stats *stats) | |
3157 | { | |
3158 | int error; | |
3159 | struct ofpbuf *reply; | |
3160 | ||
3161 | error = tc_query_class(netdev, handle, parent, &reply); | |
3162 | if (error) { | |
3163 | return error; | |
3164 | } | |
3165 | ||
3166 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
3167 | ofpbuf_delete(reply); | |
3168 | return error; | |
3169 | } | |
3170 | ||
3171 | static void | |
79f1cbe9 | 3172 | hfsc_parse_qdisc_details__(struct netdev *netdev, const struct smap *details, |
a339aa81 EJ |
3173 | struct hfsc_class *class) |
3174 | { | |
3175 | uint32_t max_rate; | |
3176 | const char *max_rate_s; | |
3177 | ||
79f1cbe9 | 3178 | max_rate_s = smap_get(details, "max-rate"); |
a339aa81 EJ |
3179 | max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
3180 | ||
3181 | if (!max_rate) { | |
a00ca915 | 3182 | enum netdev_features current; |
a339aa81 EJ |
3183 | |
3184 | netdev_get_features(netdev, ¤t, NULL, NULL, NULL); | |
d02a5f8e | 3185 | max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
a339aa81 EJ |
3186 | } |
3187 | ||
3188 | class->min_rate = max_rate; | |
3189 | class->max_rate = max_rate; | |
3190 | } | |
3191 | ||
3192 | static int | |
3193 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 3194 | const struct smap *details, |
a339aa81 EJ |
3195 | struct hfsc_class * class) |
3196 | { | |
3197 | const struct hfsc *hfsc; | |
3198 | uint32_t min_rate, max_rate; | |
3199 | const char *min_rate_s, *max_rate_s; | |
3200 | ||
3201 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 EJ |
3202 | min_rate_s = smap_get(details, "min-rate"); |
3203 | max_rate_s = smap_get(details, "max-rate"); | |
a339aa81 | 3204 | |
c45ab5e9 | 3205 | min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
79398bad | 3206 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
3207 | min_rate = MIN(min_rate, hfsc->max_rate); |
3208 | ||
3209 | max_rate = (max_rate_s | |
3210 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
3211 | : hfsc->max_rate); | |
3212 | max_rate = MAX(max_rate, min_rate); | |
3213 | max_rate = MIN(max_rate, hfsc->max_rate); | |
3214 | ||
3215 | class->min_rate = min_rate; | |
3216 | class->max_rate = max_rate; | |
3217 | ||
3218 | return 0; | |
3219 | } | |
3220 | ||
3221 | /* Create an HFSC qdisc. | |
3222 | * | |
3223 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
3224 | static int | |
3225 | hfsc_setup_qdisc__(struct netdev * netdev) | |
3226 | { | |
3227 | struct tcmsg *tcmsg; | |
3228 | struct ofpbuf request; | |
3229 | struct tc_hfsc_qopt opt; | |
3230 | ||
3231 | tc_del_qdisc(netdev); | |
3232 | ||
3233 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3234 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3235 | ||
3236 | if (!tcmsg) { | |
3237 | return ENODEV; | |
3238 | } | |
3239 | ||
3240 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3241 | tcmsg->tcm_parent = TC_H_ROOT; | |
3242 | ||
3243 | memset(&opt, 0, sizeof opt); | |
3244 | opt.defcls = 1; | |
3245 | ||
3246 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3247 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3248 | ||
3249 | return tc_transact(&request, NULL); | |
3250 | } | |
3251 | ||
3252 | /* Create an HFSC class. | |
3253 | * | |
3254 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
3255 | * sc rate <min_rate> ul rate <max_rate>" */ | |
3256 | static int | |
3257 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
3258 | unsigned int parent, struct hfsc_class *class) | |
3259 | { | |
3260 | int error; | |
3261 | size_t opt_offset; | |
3262 | struct tcmsg *tcmsg; | |
3263 | struct ofpbuf request; | |
3264 | struct tc_service_curve min, max; | |
3265 | ||
3266 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
3267 | ||
3268 | if (!tcmsg) { | |
3269 | return ENODEV; | |
3270 | } | |
3271 | ||
3272 | tcmsg->tcm_handle = handle; | |
3273 | tcmsg->tcm_parent = parent; | |
3274 | ||
3275 | min.m1 = 0; | |
3276 | min.d = 0; | |
3277 | min.m2 = class->min_rate; | |
3278 | ||
3279 | max.m1 = 0; | |
3280 | max.d = 0; | |
3281 | max.m2 = class->max_rate; | |
3282 | ||
3283 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
3284 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3285 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
3286 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
3287 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
3288 | nl_msg_end_nested(&request, opt_offset); | |
3289 | ||
3290 | error = tc_transact(&request, NULL); | |
3291 | if (error) { | |
3292 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3293 | "min-rate %ubps, max-rate %ubps (%s)", | |
3294 | netdev_get_name(netdev), | |
3295 | tc_get_major(handle), tc_get_minor(handle), | |
3296 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 3297 | class->min_rate, class->max_rate, ovs_strerror(error)); |
a339aa81 EJ |
3298 | } |
3299 | ||
3300 | return error; | |
3301 | } | |
3302 | ||
3303 | static int | |
79f1cbe9 | 3304 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
3305 | { |
3306 | int error; | |
3307 | struct hfsc_class class; | |
3308 | ||
3309 | error = hfsc_setup_qdisc__(netdev); | |
3310 | ||
3311 | if (error) { | |
3312 | return error; | |
3313 | } | |
3314 | ||
3315 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3316 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3317 | tc_make_handle(1, 0), &class); | |
3318 | ||
3319 | if (error) { | |
3320 | return error; | |
3321 | } | |
3322 | ||
3323 | hfsc_install__(netdev, class.max_rate); | |
3324 | return 0; | |
3325 | } | |
3326 | ||
3327 | static int | |
3328 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3329 | { | |
3330 | struct ofpbuf msg; | |
a339aa81 EJ |
3331 | struct nl_dump dump; |
3332 | struct hfsc_class hc; | |
3333 | ||
3334 | hc.max_rate = 0; | |
3335 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 3336 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 EJ |
3337 | |
3338 | if (!start_queue_dump(netdev, &dump)) { | |
3339 | return ENODEV; | |
3340 | } | |
3341 | ||
3342 | while (nl_dump_next(&dump, &msg)) { | |
3343 | unsigned int queue_id; | |
3344 | ||
3345 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
3346 | hfsc_update_queue__(netdev, queue_id, &hc); | |
3347 | } | |
3348 | } | |
3349 | ||
3350 | nl_dump_done(&dump); | |
3351 | return 0; | |
3352 | } | |
3353 | ||
3354 | static void | |
3355 | hfsc_tc_destroy(struct tc *tc) | |
3356 | { | |
3357 | struct hfsc *hfsc; | |
3358 | struct hfsc_class *hc, *next; | |
3359 | ||
3360 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
3361 | ||
3362 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
3363 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3364 | free(hc); | |
3365 | } | |
3366 | ||
3367 | tc_destroy(tc); | |
3368 | free(hfsc); | |
3369 | } | |
3370 | ||
3371 | static int | |
79f1cbe9 | 3372 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
3373 | { |
3374 | const struct hfsc *hfsc; | |
3375 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 3376 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
3377 | return 0; |
3378 | } | |
3379 | ||
3380 | static int | |
79f1cbe9 | 3381 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
3382 | { |
3383 | int error; | |
3384 | struct hfsc_class class; | |
3385 | ||
3386 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
3387 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3388 | tc_make_handle(1, 0), &class); | |
3389 | ||
3390 | if (!error) { | |
3391 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
3392 | } | |
3393 | ||
3394 | return error; | |
3395 | } | |
3396 | ||
3397 | static int | |
3398 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 3399 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
3400 | { |
3401 | const struct hfsc_class *hc; | |
3402 | ||
3403 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 3404 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 3405 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 3406 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
3407 | } |
3408 | return 0; | |
3409 | } | |
3410 | ||
3411 | static int | |
3412 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 3413 | const struct smap *details) |
a339aa81 EJ |
3414 | { |
3415 | int error; | |
3416 | struct hfsc_class class; | |
3417 | ||
3418 | error = hfsc_parse_class_details__(netdev, details, &class); | |
3419 | if (error) { | |
3420 | return error; | |
3421 | } | |
3422 | ||
3423 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
3424 | tc_make_handle(1, 0xfffe), &class); | |
3425 | if (error) { | |
3426 | return error; | |
3427 | } | |
3428 | ||
3429 | hfsc_update_queue__(netdev, queue_id, &class); | |
3430 | return 0; | |
3431 | } | |
3432 | ||
3433 | static int | |
3434 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
3435 | { | |
3436 | int error; | |
3437 | struct hfsc *hfsc; | |
3438 | struct hfsc_class *hc; | |
3439 | ||
3440 | hc = hfsc_class_cast__(queue); | |
3441 | hfsc = hfsc_get__(netdev); | |
3442 | ||
3443 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
3444 | if (!error) { | |
3445 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
3446 | free(hc); | |
3447 | } | |
3448 | return error; | |
3449 | } | |
3450 | ||
3451 | static int | |
3452 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
3453 | struct netdev_queue_stats *stats) | |
3454 | { | |
3455 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
3456 | tc_make_handle(1, 0xfffe), NULL, stats); | |
3457 | } | |
3458 | ||
3459 | static int | |
3460 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
3461 | const struct ofpbuf *nlmsg, | |
3462 | netdev_dump_queue_stats_cb *cb, void *aux) | |
3463 | { | |
3464 | struct netdev_queue_stats stats; | |
3465 | unsigned int handle, major, minor; | |
3466 | int error; | |
3467 | ||
3468 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
3469 | if (error) { | |
3470 | return error; | |
3471 | } | |
3472 | ||
3473 | major = tc_get_major(handle); | |
3474 | minor = tc_get_minor(handle); | |
3475 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
3476 | (*cb)(minor - 1, &stats, aux); | |
3477 | } | |
3478 | return 0; | |
3479 | } | |
3480 | ||
3481 | static const struct tc_ops tc_ops_hfsc = { | |
3482 | "hfsc", /* linux_name */ | |
3483 | "linux-hfsc", /* ovs_name */ | |
3484 | HFSC_N_QUEUES, /* n_queues */ | |
3485 | hfsc_tc_install, /* tc_install */ | |
3486 | hfsc_tc_load, /* tc_load */ | |
3487 | hfsc_tc_destroy, /* tc_destroy */ | |
3488 | hfsc_qdisc_get, /* qdisc_get */ | |
3489 | hfsc_qdisc_set, /* qdisc_set */ | |
3490 | hfsc_class_get, /* class_get */ | |
3491 | hfsc_class_set, /* class_set */ | |
3492 | hfsc_class_delete, /* class_delete */ | |
3493 | hfsc_class_get_stats, /* class_get_stats */ | |
3494 | hfsc_class_dump_stats /* class_dump_stats */ | |
3495 | }; | |
3496 | \f | |
c1c9c9c4 BP |
3497 | /* "linux-default" traffic control class. |
3498 | * | |
3499 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
3500 | * the "" (empty string) QoS type in the OVS database. */ | |
3501 | ||
3502 | static void | |
b5d57fc8 | 3503 | default_install__(struct netdev *netdev_) |
c1c9c9c4 | 3504 | { |
b5d57fc8 | 3505 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 3506 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); |
c1c9c9c4 | 3507 | |
559eb230 BP |
3508 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
3509 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 3510 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
3511 | } |
3512 | ||
3513 | static int | |
3514 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 3515 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
3516 | { |
3517 | default_install__(netdev); | |
3518 | return 0; | |
3519 | } | |
3520 | ||
3521 | static int | |
3522 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3523 | { | |
3524 | default_install__(netdev); | |
3525 | return 0; | |
3526 | } | |
3527 | ||
3528 | static const struct tc_ops tc_ops_default = { | |
3529 | NULL, /* linux_name */ | |
3530 | "", /* ovs_name */ | |
3531 | 0, /* n_queues */ | |
3532 | default_tc_install, | |
3533 | default_tc_load, | |
3534 | NULL, /* tc_destroy */ | |
3535 | NULL, /* qdisc_get */ | |
3536 | NULL, /* qdisc_set */ | |
3537 | NULL, /* class_get */ | |
3538 | NULL, /* class_set */ | |
3539 | NULL, /* class_delete */ | |
3540 | NULL, /* class_get_stats */ | |
3541 | NULL /* class_dump_stats */ | |
3542 | }; | |
3543 | \f | |
3544 | /* "linux-other" traffic control class. | |
3545 | * | |
3546 | * */ | |
3547 | ||
3548 | static int | |
b5d57fc8 | 3549 | other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) |
c1c9c9c4 | 3550 | { |
b5d57fc8 | 3551 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 3552 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); |
c1c9c9c4 | 3553 | |
559eb230 BP |
3554 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
3555 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 3556 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
3557 | return 0; |
3558 | } | |
3559 | ||
3560 | static const struct tc_ops tc_ops_other = { | |
3561 | NULL, /* linux_name */ | |
3562 | "linux-other", /* ovs_name */ | |
3563 | 0, /* n_queues */ | |
3564 | NULL, /* tc_install */ | |
3565 | other_tc_load, | |
3566 | NULL, /* tc_destroy */ | |
3567 | NULL, /* qdisc_get */ | |
3568 | NULL, /* qdisc_set */ | |
3569 | NULL, /* class_get */ | |
3570 | NULL, /* class_set */ | |
3571 | NULL, /* class_delete */ | |
3572 | NULL, /* class_get_stats */ | |
3573 | NULL /* class_dump_stats */ | |
3574 | }; | |
3575 | \f | |
3576 | /* Traffic control. */ | |
3577 | ||
3578 | /* Number of kernel "tc" ticks per second. */ | |
3579 | static double ticks_per_s; | |
3580 | ||
3581 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
3582 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
3583 | * one jiffy's worth of data. | |
3584 | * | |
3585 | * There are two possibilities here: | |
3586 | * | |
3587 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
3588 | * approximate range of 100 to 1024. That means that we really need to | |
3589 | * make sure that the qdisc can buffer that much data. | |
3590 | * | |
3591 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
3592 | * has finely granular timers and there's no need to fudge additional room | |
3593 | * for buffers. (There's no extra effort needed to implement that: the | |
3594 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
3595 | * come out as 0 in the division. Small integer results in the case of | |
3596 | * really high dividends won't have any real effect anyhow.) | |
3597 | */ | |
3598 | static unsigned int buffer_hz; | |
3599 | ||
3600 | /* Returns tc handle 'major':'minor'. */ | |
3601 | static unsigned int | |
3602 | tc_make_handle(unsigned int major, unsigned int minor) | |
3603 | { | |
3604 | return TC_H_MAKE(major << 16, minor); | |
3605 | } | |
3606 | ||
3607 | /* Returns the major number from 'handle'. */ | |
3608 | static unsigned int | |
3609 | tc_get_major(unsigned int handle) | |
3610 | { | |
3611 | return TC_H_MAJ(handle) >> 16; | |
3612 | } | |
3613 | ||
3614 | /* Returns the minor number from 'handle'. */ | |
3615 | static unsigned int | |
3616 | tc_get_minor(unsigned int handle) | |
3617 | { | |
3618 | return TC_H_MIN(handle); | |
3619 | } | |
3620 | ||
3621 | static struct tcmsg * | |
3622 | tc_make_request(const struct netdev *netdev, int type, unsigned int flags, | |
3623 | struct ofpbuf *request) | |
3624 | { | |
3625 | struct tcmsg *tcmsg; | |
3626 | int ifindex; | |
3627 | int error; | |
3628 | ||
3629 | error = get_ifindex(netdev, &ifindex); | |
3630 | if (error) { | |
3631 | return NULL; | |
3632 | } | |
3633 | ||
3634 | ofpbuf_init(request, 512); | |
3635 | nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); | |
3636 | tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); | |
3637 | tcmsg->tcm_family = AF_UNSPEC; | |
3638 | tcmsg->tcm_ifindex = ifindex; | |
3639 | /* Caller should fill in tcmsg->tcm_handle. */ | |
3640 | /* Caller should fill in tcmsg->tcm_parent. */ | |
3641 | ||
3642 | return tcmsg; | |
3643 | } | |
3644 | ||
3645 | static int | |
3646 | tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) | |
3647 | { | |
a88b4e04 | 3648 | int error = nl_transact(NETLINK_ROUTE, request, replyp); |
c1c9c9c4 BP |
3649 | ofpbuf_uninit(request); |
3650 | return error; | |
3651 | } | |
3652 | ||
f8500004 JP |
3653 | /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for |
3654 | * policing configuration. | |
3655 | * | |
3656 | * This function is equivalent to running the following when 'add' is true: | |
3657 | * /sbin/tc qdisc add dev <devname> handle ffff: ingress | |
3658 | * | |
3659 | * This function is equivalent to running the following when 'add' is false: | |
3660 | * /sbin/tc qdisc del dev <devname> handle ffff: ingress | |
3661 | * | |
3662 | * The configuration and stats may be seen with the following command: | |
3663 | * /sbin/tc -s qdisc show dev <devname> | |
3664 | * | |
3665 | * Returns 0 if successful, otherwise a positive errno value. | |
3666 | */ | |
3667 | static int | |
3668 | tc_add_del_ingress_qdisc(struct netdev *netdev, bool add) | |
3669 | { | |
3670 | struct ofpbuf request; | |
3671 | struct tcmsg *tcmsg; | |
3672 | int error; | |
3673 | int type = add ? RTM_NEWQDISC : RTM_DELQDISC; | |
3674 | int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0; | |
3675 | ||
3676 | tcmsg = tc_make_request(netdev, type, flags, &request); | |
3677 | if (!tcmsg) { | |
3678 | return ENODEV; | |
3679 | } | |
3680 | tcmsg->tcm_handle = tc_make_handle(0xffff, 0); | |
3681 | tcmsg->tcm_parent = TC_H_INGRESS; | |
3682 | nl_msg_put_string(&request, TCA_KIND, "ingress"); | |
3683 | nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); | |
3684 | ||
3685 | error = tc_transact(&request, NULL); | |
3686 | if (error) { | |
3687 | /* If we're deleting the qdisc, don't worry about some of the | |
3688 | * error conditions. */ | |
3689 | if (!add && (error == ENOENT || error == EINVAL)) { | |
3690 | return 0; | |
3691 | } | |
3692 | return error; | |
3693 | } | |
3694 | ||
3695 | return 0; | |
3696 | } | |
3697 | ||
3698 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size | |
3699 | * of 'kbits_burst'. | |
3700 | * | |
3701 | * This function is equivalent to running: | |
3702 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
3703 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
3704 | * mtu 65535 drop | |
3705 | * | |
3706 | * The configuration and stats may be seen with the following command: | |
3707 | * /sbin/tc -s filter show <devname> eth0 parent ffff: | |
3708 | * | |
3709 | * Returns 0 if successful, otherwise a positive errno value. | |
3710 | */ | |
3711 | static int | |
3712 | tc_add_policer(struct netdev *netdev, int kbits_rate, int kbits_burst) | |
3713 | { | |
3714 | struct tc_police tc_police; | |
3715 | struct ofpbuf request; | |
3716 | struct tcmsg *tcmsg; | |
3717 | size_t basic_offset; | |
3718 | size_t police_offset; | |
3719 | int error; | |
3720 | int mtu = 65535; | |
3721 | ||
3722 | memset(&tc_police, 0, sizeof tc_police); | |
3723 | tc_police.action = TC_POLICE_SHOT; | |
3724 | tc_police.mtu = mtu; | |
e5c08015 | 3725 | tc_fill_rate(&tc_police.rate, (kbits_rate * 1000)/8, mtu); |
f8500004 JP |
3726 | tc_police.burst = tc_bytes_to_ticks(tc_police.rate.rate, |
3727 | kbits_burst * 1024); | |
3728 | ||
3729 | tcmsg = tc_make_request(netdev, RTM_NEWTFILTER, | |
3730 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3731 | if (!tcmsg) { | |
3732 | return ENODEV; | |
3733 | } | |
3734 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
3735 | tcmsg->tcm_info = tc_make_handle(49, | |
3736 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
3737 | ||
3738 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
3739 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3740 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
3741 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
3742 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
3743 | nl_msg_end_nested(&request, police_offset); | |
3744 | nl_msg_end_nested(&request, basic_offset); | |
3745 | ||
3746 | error = tc_transact(&request, NULL); | |
3747 | if (error) { | |
3748 | return error; | |
3749 | } | |
3750 | ||
3751 | return 0; | |
3752 | } | |
3753 | ||
c1c9c9c4 BP |
3754 | static void |
3755 | read_psched(void) | |
3756 | { | |
3757 | /* The values in psched are not individually very meaningful, but they are | |
3758 | * important. The tables below show some values seen in the wild. | |
3759 | * | |
3760 | * Some notes: | |
3761 | * | |
3762 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
3763 | * (Before that, there are hints that it was 1000000000.) | |
3764 | * | |
3765 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
3766 | * above. | |
3767 | * | |
3768 | * /proc/net/psched | |
3769 | * ----------------------------------- | |
3770 | * [1] 000c8000 000f4240 000f4240 00000064 | |
3771 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
3772 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
3773 | * [4] 000003e8 00000400 000f4240 00000064 | |
3774 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
3775 | * [6] 000003e8 00000040 000f4240 000000f9 | |
3776 | * | |
3777 | * a b c d ticks_per_s buffer_hz | |
3778 | * ------- --------- ---------- ------------- ----------- ------------- | |
3779 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
3780 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3781 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
3782 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
3783 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
3784 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
3785 | * | |
3786 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
3787 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
3788 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
3789 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
3790 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
3791 | * [6] 2.6.34 from kernel.org on KVM | |
3792 | */ | |
23882115 | 3793 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
c1c9c9c4 BP |
3794 | static const char fn[] = "/proc/net/psched"; |
3795 | unsigned int a, b, c, d; | |
3796 | FILE *stream; | |
3797 | ||
23882115 BP |
3798 | if (!ovsthread_once_start(&once)) { |
3799 | return; | |
3800 | } | |
3801 | ||
c1c9c9c4 BP |
3802 | ticks_per_s = 1.0; |
3803 | buffer_hz = 100; | |
3804 | ||
3805 | stream = fopen(fn, "r"); | |
3806 | if (!stream) { | |
10a89ef0 | 3807 | VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); |
23882115 | 3808 | goto exit; |
c1c9c9c4 BP |
3809 | } |
3810 | ||
3811 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
3812 | VLOG_WARN("%s: read failed", fn); | |
3813 | fclose(stream); | |
23882115 | 3814 | goto exit; |
c1c9c9c4 BP |
3815 | } |
3816 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
3817 | fclose(stream); | |
3818 | ||
3819 | if (!a || !c) { | |
3820 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
23882115 | 3821 | goto exit; |
c1c9c9c4 BP |
3822 | } |
3823 | ||
3824 | ticks_per_s = (double) a * c / b; | |
3825 | if (c == 1000000) { | |
3826 | buffer_hz = d; | |
3827 | } else { | |
3828 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
3829 | fn, a, b, c, d); | |
3830 | } | |
3831 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
23882115 BP |
3832 | |
3833 | exit: | |
3834 | ovsthread_once_done(&once); | |
c1c9c9c4 BP |
3835 | } |
3836 | ||
3837 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
3838 | * rate of 'rate' bytes per second. */ | |
3839 | static unsigned int | |
3840 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
3841 | { | |
23882115 | 3842 | read_psched(); |
c1c9c9c4 BP |
3843 | return (rate * ticks) / ticks_per_s; |
3844 | } | |
3845 | ||
3846 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
3847 | * rate of 'rate' bytes per second. */ | |
3848 | static unsigned int | |
3849 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
3850 | { | |
23882115 | 3851 | read_psched(); |
015c93a4 | 3852 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
3853 | } |
3854 | ||
3855 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
3856 | * a transmission rate of 'rate' bytes per second. */ | |
3857 | static unsigned int | |
3858 | tc_buffer_per_jiffy(unsigned int rate) | |
3859 | { | |
23882115 | 3860 | read_psched(); |
c1c9c9c4 BP |
3861 | return rate / buffer_hz; |
3862 | } | |
3863 | ||
3864 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
3865 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
3866 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
3867 | * stores NULL into it if it is absent. | |
3868 | * | |
3869 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
3870 | * 'msg'. | |
3871 | * | |
3872 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3873 | static int | |
3874 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
3875 | struct nlattr **options) | |
3876 | { | |
3877 | static const struct nl_policy tca_policy[] = { | |
3878 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
3879 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
3880 | }; | |
3881 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3882 | ||
3883 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3884 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3885 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
3886 | goto error; | |
3887 | } | |
3888 | ||
3889 | if (kind) { | |
3890 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
3891 | } | |
3892 | ||
3893 | if (options) { | |
3894 | *options = ta[TCA_OPTIONS]; | |
3895 | } | |
3896 | ||
3897 | return 0; | |
3898 | ||
3899 | error: | |
3900 | if (kind) { | |
3901 | *kind = NULL; | |
3902 | } | |
3903 | if (options) { | |
3904 | *options = NULL; | |
3905 | } | |
3906 | return EPROTO; | |
3907 | } | |
3908 | ||
3909 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
3910 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
3911 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
3912 | * arguments may be null. | |
3913 | * | |
3914 | * Returns 0 if successful, otherwise a positive errno value. */ | |
3915 | static int | |
3916 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
3917 | struct nlattr **options, struct netdev_queue_stats *stats) | |
3918 | { | |
3919 | static const struct nl_policy tca_policy[] = { | |
3920 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
3921 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
3922 | }; | |
3923 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
3924 | ||
3925 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
3926 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
3927 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
3928 | goto error; | |
3929 | } | |
3930 | ||
3931 | if (handlep) { | |
3932 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
3933 | *handlep = tc->tcm_handle; | |
3934 | } | |
3935 | ||
3936 | if (options) { | |
3937 | *options = ta[TCA_OPTIONS]; | |
3938 | } | |
3939 | ||
3940 | if (stats) { | |
3941 | const struct gnet_stats_queue *gsq; | |
3942 | struct gnet_stats_basic gsb; | |
3943 | ||
3944 | static const struct nl_policy stats_policy[] = { | |
3945 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
3946 | .min_len = sizeof gsb }, | |
3947 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
3948 | .min_len = sizeof *gsq }, | |
3949 | }; | |
3950 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
3951 | ||
3952 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
3953 | sa, ARRAY_SIZE(sa))) { | |
3954 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
3955 | goto error; | |
3956 | } | |
3957 | ||
3958 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
3959 | * some arch/bitsize combinations. Newer versions of Linux have a | |
3960 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
3961 | * easiest thing to do is just to make a copy. */ | |
3962 | memset(&gsb, 0, sizeof gsb); | |
3963 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
3964 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
3965 | stats->tx_bytes = gsb.bytes; | |
3966 | stats->tx_packets = gsb.packets; | |
3967 | ||
3968 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
3969 | stats->tx_errors = gsq->drops; | |
3970 | } | |
3971 | ||
3972 | return 0; | |
3973 | ||
3974 | error: | |
3975 | if (options) { | |
3976 | *options = NULL; | |
3977 | } | |
3978 | if (stats) { | |
3979 | memset(stats, 0, sizeof *stats); | |
3980 | } | |
3981 | return EPROTO; | |
3982 | } | |
3983 | ||
3984 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
3985 | * on 'netdev'. */ | |
3986 | static int | |
3987 | tc_query_class(const struct netdev *netdev, | |
3988 | unsigned int handle, unsigned int parent, | |
3989 | struct ofpbuf **replyp) | |
3990 | { | |
3991 | struct ofpbuf request; | |
3992 | struct tcmsg *tcmsg; | |
3993 | int error; | |
3994 | ||
3995 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); | |
23a98ffe BP |
3996 | if (!tcmsg) { |
3997 | return ENODEV; | |
3998 | } | |
c1c9c9c4 BP |
3999 | tcmsg->tcm_handle = handle; |
4000 | tcmsg->tcm_parent = parent; | |
4001 | ||
4002 | error = tc_transact(&request, replyp); | |
4003 | if (error) { | |
4004 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
4005 | netdev_get_name(netdev), | |
4006 | tc_get_major(handle), tc_get_minor(handle), | |
4007 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 4008 | ovs_strerror(error)); |
c1c9c9c4 BP |
4009 | } |
4010 | return error; | |
4011 | } | |
4012 | ||
4013 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
4014 | static int | |
4015 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
4016 | { | |
4017 | struct ofpbuf request; | |
4018 | struct tcmsg *tcmsg; | |
4019 | int error; | |
4020 | ||
4021 | tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); | |
23a98ffe BP |
4022 | if (!tcmsg) { |
4023 | return ENODEV; | |
4024 | } | |
c1c9c9c4 BP |
4025 | tcmsg->tcm_handle = handle; |
4026 | tcmsg->tcm_parent = 0; | |
4027 | ||
4028 | error = tc_transact(&request, NULL); | |
4029 | if (error) { | |
4030 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
4031 | netdev_get_name(netdev), | |
4032 | tc_get_major(handle), tc_get_minor(handle), | |
10a89ef0 | 4033 | ovs_strerror(error)); |
c1c9c9c4 BP |
4034 | } |
4035 | return error; | |
4036 | } | |
4037 | ||
4038 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
4039 | static int | |
b5d57fc8 | 4040 | tc_del_qdisc(struct netdev *netdev_) |
c1c9c9c4 | 4041 | { |
b5d57fc8 | 4042 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
4043 | struct ofpbuf request; |
4044 | struct tcmsg *tcmsg; | |
4045 | int error; | |
4046 | ||
b5d57fc8 | 4047 | tcmsg = tc_make_request(netdev_, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
4048 | if (!tcmsg) { |
4049 | return ENODEV; | |
4050 | } | |
c1c9c9c4 BP |
4051 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
4052 | tcmsg->tcm_parent = TC_H_ROOT; | |
4053 | ||
4054 | error = tc_transact(&request, NULL); | |
4055 | if (error == EINVAL) { | |
4056 | /* EINVAL probably means that the default qdisc was in use, in which | |
4057 | * case we've accomplished our purpose. */ | |
4058 | error = 0; | |
4059 | } | |
b5d57fc8 BP |
4060 | if (!error && netdev->tc) { |
4061 | if (netdev->tc->ops->tc_destroy) { | |
4062 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 | 4063 | } |
b5d57fc8 | 4064 | netdev->tc = NULL; |
c1c9c9c4 BP |
4065 | } |
4066 | return error; | |
4067 | } | |
4068 | ||
4069 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the | |
4070 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
4071 | * positive errno value. */ | |
4072 | static int | |
b5d57fc8 | 4073 | tc_query_qdisc(const struct netdev *netdev_) |
c1c9c9c4 | 4074 | { |
b5d57fc8 | 4075 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
4076 | struct ofpbuf request, *qdisc; |
4077 | const struct tc_ops *ops; | |
4078 | struct tcmsg *tcmsg; | |
4079 | int load_error; | |
4080 | int error; | |
4081 | ||
b5d57fc8 | 4082 | if (netdev->tc) { |
c1c9c9c4 BP |
4083 | return 0; |
4084 | } | |
4085 | ||
4086 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
4087 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
4088 | * 2.6.35 without that fix backported to it. | |
4089 | * | |
4090 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
4091 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
4092 | * few others. There are a few ways that I can see to do this, but most of | |
4093 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
4094 | * technique chosen here is to assume that any non-default qdisc that we | |
4095 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
4096 | * a class with handle 0:0. | |
4097 | * | |
4098 | * We could check for Linux 2.6.35+ and use a more straightforward method | |
4099 | * there. */ | |
b5d57fc8 | 4100 | tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request); |
23a98ffe BP |
4101 | if (!tcmsg) { |
4102 | return ENODEV; | |
4103 | } | |
c1c9c9c4 BP |
4104 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
4105 | tcmsg->tcm_parent = 0; | |
4106 | ||
4107 | /* Figure out what tc class to instantiate. */ | |
4108 | error = tc_transact(&request, &qdisc); | |
4109 | if (!error) { | |
4110 | const char *kind; | |
4111 | ||
4112 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
4113 | if (error) { | |
4114 | ops = &tc_ops_other; | |
4115 | } else { | |
4116 | ops = tc_lookup_linux_name(kind); | |
4117 | if (!ops) { | |
4118 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
4119 | VLOG_INFO_RL(&rl2, "unknown qdisc \"%s\"", kind); | |
4120 | ||
4121 | ops = &tc_ops_other; | |
4122 | } | |
4123 | } | |
4124 | } else if (error == ENOENT) { | |
4125 | /* Either it's a built-in qdisc, or it's a qdisc set up by some | |
4126 | * other entity that doesn't have a handle 1:0. We will assume | |
4127 | * that it's the system default qdisc. */ | |
4128 | ops = &tc_ops_default; | |
4129 | error = 0; | |
4130 | } else { | |
4131 | /* Who knows? Maybe the device got deleted. */ | |
4132 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
10a89ef0 | 4133 | netdev_get_name(netdev_), ovs_strerror(error)); |
c1c9c9c4 BP |
4134 | ops = &tc_ops_other; |
4135 | } | |
4136 | ||
4137 | /* Instantiate it. */ | |
b5d57fc8 BP |
4138 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); |
4139 | ovs_assert((load_error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
4140 | ofpbuf_delete(qdisc); |
4141 | ||
4142 | return error ? error : load_error; | |
4143 | } | |
4144 | ||
4145 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
4146 | approximate the time to transmit packets of various lengths. For an MTU of | |
4147 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
4148 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
4149 | possible lengths; and so on. | |
4150 | ||
4151 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
4152 | need to be shifted right to fit within such a 256-entry table. */ | |
4153 | static int | |
4154 | tc_calc_cell_log(unsigned int mtu) | |
4155 | { | |
4156 | int cell_log; | |
4157 | ||
4158 | if (!mtu) { | |
4159 | mtu = ETH_PAYLOAD_MAX; | |
4160 | } | |
4161 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
4162 | ||
4163 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
4164 | mtu >>= 1; | |
4165 | } | |
4166 | ||
4167 | return cell_log; | |
4168 | } | |
4169 | ||
4170 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
4171 | * of 'mtu'. */ | |
4172 | static void | |
4173 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
4174 | { | |
4175 | memset(rate, 0, sizeof *rate); | |
4176 | rate->cell_log = tc_calc_cell_log(mtu); | |
4177 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
4178 | /* rate->cell_align = 0; */ /* distro headers. */ | |
4179 | rate->mpu = ETH_TOTAL_MIN; | |
4180 | rate->rate = Bps; | |
4181 | } | |
4182 | ||
4183 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
4184 | * attribute of the specified "type". | |
4185 | * | |
4186 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
4187 | static void | |
4188 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
4189 | { | |
4190 | uint32_t *rtab; | |
4191 | unsigned int i; | |
4192 | ||
4193 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
4194 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
4195 | unsigned packet_size = (i + 1) << rate->cell_log; | |
4196 | if (packet_size < rate->mpu) { | |
4197 | packet_size = rate->mpu; | |
4198 | } | |
4199 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
4200 | } | |
4201 | } | |
4202 | ||
4203 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
4204 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
4205 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 4206 | * 0 is fine.) */ |
c1c9c9c4 BP |
4207 | static int |
4208 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
4209 | { | |
4210 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
4211 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
4212 | } | |
d3980822 | 4213 | \f |
aaf2fb1a BP |
4214 | /* Linux-only functions declared in netdev-linux.h */ |
4215 | ||
025e874a BP |
4216 | /* Returns a fd for an AF_INET socket or a negative errno value. */ |
4217 | int | |
4218 | netdev_linux_get_af_inet_sock(void) | |
4219 | { | |
4220 | int error = netdev_linux_init(); | |
4221 | return error ? -error : af_inet_sock; | |
4222 | } | |
4223 | ||
aaf2fb1a BP |
4224 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If |
4225 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
4226 | int | |
4227 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
4228 | const char *flag_name, bool enable) | |
4229 | { | |
4230 | const char *netdev_name = netdev_get_name(netdev); | |
4231 | struct ethtool_value evalue; | |
4232 | uint32_t new_flags; | |
4233 | int error; | |
4234 | ||
ab985a77 | 4235 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
4236 | memset(&evalue, 0, sizeof evalue); |
4237 | error = netdev_linux_do_ethtool(netdev_name, | |
4238 | (struct ethtool_cmd *)&evalue, | |
4239 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
4240 | if (error) { | |
4241 | return error; | |
4242 | } | |
4243 | ||
ab985a77 | 4244 | COVERAGE_INC(netdev_set_ethtool); |
aaf2fb1a BP |
4245 | evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
4246 | error = netdev_linux_do_ethtool(netdev_name, | |
4247 | (struct ethtool_cmd *)&evalue, | |
4248 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
4249 | if (error) { | |
4250 | return error; | |
4251 | } | |
4252 | ||
ab985a77 | 4253 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
4254 | memset(&evalue, 0, sizeof evalue); |
4255 | error = netdev_linux_do_ethtool(netdev_name, | |
4256 | (struct ethtool_cmd *)&evalue, | |
4257 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
4258 | if (error) { | |
4259 | return error; | |
4260 | } | |
4261 | ||
4262 | if (new_flags != evalue.data) { | |
4263 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
4264 | "device %s failed", enable ? "enable" : "disable", | |
4265 | flag_name, netdev_name); | |
4266 | return EOPNOTSUPP; | |
4267 | } | |
4268 | ||
4269 | return 0; | |
4270 | } | |
4271 | \f | |
4272 | /* Utility functions. */ | |
4273 | ||
d3980822 | 4274 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 4275 | static void |
d3980822 BP |
4276 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
4277 | const struct rtnl_link_stats *src) | |
4278 | { | |
f613a0d7 PS |
4279 | dst->rx_packets = src->rx_packets; |
4280 | dst->tx_packets = src->tx_packets; | |
4281 | dst->rx_bytes = src->rx_bytes; | |
4282 | dst->tx_bytes = src->tx_bytes; | |
4283 | dst->rx_errors = src->rx_errors; | |
4284 | dst->tx_errors = src->tx_errors; | |
4285 | dst->rx_dropped = src->rx_dropped; | |
4286 | dst->tx_dropped = src->tx_dropped; | |
4287 | dst->multicast = src->multicast; | |
4288 | dst->collisions = src->collisions; | |
4289 | dst->rx_length_errors = src->rx_length_errors; | |
4290 | dst->rx_over_errors = src->rx_over_errors; | |
4291 | dst->rx_crc_errors = src->rx_crc_errors; | |
4292 | dst->rx_frame_errors = src->rx_frame_errors; | |
4293 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
4294 | dst->rx_missed_errors = src->rx_missed_errors; | |
4295 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
4296 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
4297 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
4298 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
4299 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
4300 | } |
4301 | ||
c1c9c9c4 BP |
4302 | static int |
4303 | get_stats_via_netlink(int ifindex, struct netdev_stats *stats) | |
4304 | { | |
4305 | /* Policy for RTNLGRP_LINK messages. | |
4306 | * | |
4307 | * There are *many* more fields in these messages, but currently we only | |
4308 | * care about these fields. */ | |
4309 | static const struct nl_policy rtnlgrp_link_policy[] = { | |
4310 | [IFLA_IFNAME] = { .type = NL_A_STRING, .optional = false }, | |
4311 | [IFLA_STATS] = { .type = NL_A_UNSPEC, .optional = true, | |
4312 | .min_len = sizeof(struct rtnl_link_stats) }, | |
4313 | }; | |
4314 | ||
4315 | struct ofpbuf request; | |
4316 | struct ofpbuf *reply; | |
4317 | struct ifinfomsg *ifi; | |
c1c9c9c4 BP |
4318 | struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; |
4319 | int error; | |
4320 | ||
4321 | ofpbuf_init(&request, 0); | |
4322 | nl_msg_put_nlmsghdr(&request, sizeof *ifi, RTM_GETLINK, NLM_F_REQUEST); | |
4323 | ifi = ofpbuf_put_zeros(&request, sizeof *ifi); | |
4324 | ifi->ifi_family = PF_UNSPEC; | |
4325 | ifi->ifi_index = ifindex; | |
a88b4e04 | 4326 | error = nl_transact(NETLINK_ROUTE, &request, &reply); |
c1c9c9c4 BP |
4327 | ofpbuf_uninit(&request); |
4328 | if (error) { | |
4329 | return error; | |
4330 | } | |
4331 | ||
4332 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg), | |
4333 | rtnlgrp_link_policy, | |
4334 | attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { | |
4335 | ofpbuf_delete(reply); | |
4336 | return EPROTO; | |
4337 | } | |
4338 | ||
4339 | if (!attrs[IFLA_STATS]) { | |
4340 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
4341 | ofpbuf_delete(reply); | |
4342 | return EPROTO; | |
4343 | } | |
8b61709d | 4344 | |
d3980822 | 4345 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(attrs[IFLA_STATS])); |
8b61709d | 4346 | |
576e26d7 BP |
4347 | ofpbuf_delete(reply); |
4348 | ||
8b61709d BP |
4349 | return 0; |
4350 | } | |
4351 | ||
4352 | static int | |
4353 | get_stats_via_proc(const char *netdev_name, struct netdev_stats *stats) | |
4354 | { | |
4355 | static const char fn[] = "/proc/net/dev"; | |
4356 | char line[1024]; | |
4357 | FILE *stream; | |
4358 | int ln; | |
4359 | ||
4360 | stream = fopen(fn, "r"); | |
4361 | if (!stream) { | |
10a89ef0 | 4362 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
8b61709d BP |
4363 | return errno; |
4364 | } | |
4365 | ||
4366 | ln = 0; | |
4367 | while (fgets(line, sizeof line, stream)) { | |
4368 | if (++ln >= 3) { | |
4369 | char devname[16]; | |
4370 | #define X64 "%"SCNu64 | |
4371 | if (sscanf(line, | |
4372 | " %15[^:]:" | |
4373 | X64 X64 X64 X64 X64 X64 X64 "%*u" | |
4374 | X64 X64 X64 X64 X64 X64 X64 "%*u", | |
4375 | devname, | |
4376 | &stats->rx_bytes, | |
4377 | &stats->rx_packets, | |
4378 | &stats->rx_errors, | |
4379 | &stats->rx_dropped, | |
4380 | &stats->rx_fifo_errors, | |
4381 | &stats->rx_frame_errors, | |
4382 | &stats->multicast, | |
4383 | &stats->tx_bytes, | |
4384 | &stats->tx_packets, | |
4385 | &stats->tx_errors, | |
4386 | &stats->tx_dropped, | |
4387 | &stats->tx_fifo_errors, | |
4388 | &stats->collisions, | |
4389 | &stats->tx_carrier_errors) != 15) { | |
4390 | VLOG_WARN_RL(&rl, "%s:%d: parse error", fn, ln); | |
4391 | } else if (!strcmp(devname, netdev_name)) { | |
4392 | stats->rx_length_errors = UINT64_MAX; | |
4393 | stats->rx_over_errors = UINT64_MAX; | |
4394 | stats->rx_crc_errors = UINT64_MAX; | |
4395 | stats->rx_missed_errors = UINT64_MAX; | |
4396 | stats->tx_aborted_errors = UINT64_MAX; | |
4397 | stats->tx_heartbeat_errors = UINT64_MAX; | |
4398 | stats->tx_window_errors = UINT64_MAX; | |
4399 | fclose(stream); | |
4400 | return 0; | |
4401 | } | |
4402 | } | |
4403 | } | |
4404 | VLOG_WARN_RL(&rl, "%s: no stats for %s", fn, netdev_name); | |
4405 | fclose(stream); | |
4406 | return ENODEV; | |
4407 | } | |
c1c9c9c4 | 4408 | |
3a183124 | 4409 | static int |
b5d57fc8 | 4410 | get_flags(const struct netdev *dev, unsigned int *flags) |
8b61709d BP |
4411 | { |
4412 | struct ifreq ifr; | |
4413 | int error; | |
4414 | ||
755be9ea EJ |
4415 | *flags = 0; |
4416 | error = netdev_linux_do_ioctl(dev->name, &ifr, SIOCGIFFLAGS, | |
149f577a | 4417 | "SIOCGIFFLAGS"); |
755be9ea EJ |
4418 | if (!error) { |
4419 | *flags = ifr.ifr_flags; | |
4420 | } | |
8b61709d BP |
4421 | return error; |
4422 | } | |
4423 | ||
4424 | static int | |
4b609110 | 4425 | set_flags(const char *name, unsigned int flags) |
8b61709d BP |
4426 | { |
4427 | struct ifreq ifr; | |
4428 | ||
4429 | ifr.ifr_flags = flags; | |
4b609110 | 4430 | return netdev_linux_do_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); |
8b61709d BP |
4431 | } |
4432 | ||
4433 | static int | |
4434 | do_get_ifindex(const char *netdev_name) | |
4435 | { | |
4436 | struct ifreq ifr; | |
4437 | ||
71d7c22f | 4438 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4439 | COVERAGE_INC(netdev_get_ifindex); |
4440 | if (ioctl(af_inet_sock, SIOCGIFINDEX, &ifr) < 0) { | |
4441 | VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
10a89ef0 | 4442 | netdev_name, ovs_strerror(errno)); |
8b61709d BP |
4443 | return -errno; |
4444 | } | |
4445 | return ifr.ifr_ifindex; | |
4446 | } | |
4447 | ||
4448 | static int | |
4449 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
4450 | { | |
b5d57fc8 | 4451 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c7b1b0a5 | 4452 | |
b5d57fc8 | 4453 | if (!(netdev->cache_valid & VALID_IFINDEX)) { |
8b61709d | 4454 | int ifindex = do_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 4455 | |
8b61709d | 4456 | if (ifindex < 0) { |
b5d57fc8 BP |
4457 | netdev->get_ifindex_error = -ifindex; |
4458 | netdev->ifindex = 0; | |
c7b1b0a5 | 4459 | } else { |
b5d57fc8 BP |
4460 | netdev->get_ifindex_error = 0; |
4461 | netdev->ifindex = ifindex; | |
8b61709d | 4462 | } |
b5d57fc8 | 4463 | netdev->cache_valid |= VALID_IFINDEX; |
8b61709d | 4464 | } |
c7b1b0a5 | 4465 | |
b5d57fc8 BP |
4466 | *ifindexp = netdev->ifindex; |
4467 | return netdev->get_ifindex_error; | |
8b61709d BP |
4468 | } |
4469 | ||
4470 | static int | |
4471 | get_etheraddr(const char *netdev_name, uint8_t ea[ETH_ADDR_LEN]) | |
4472 | { | |
4473 | struct ifreq ifr; | |
4474 | int hwaddr_family; | |
4475 | ||
4476 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4477 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d BP |
4478 | COVERAGE_INC(netdev_get_hwaddr); |
4479 | if (ioctl(af_inet_sock, SIOCGIFHWADDR, &ifr) < 0) { | |
78857dfb BP |
4480 | /* ENODEV probably means that a vif disappeared asynchronously and |
4481 | * hasn't been removed from the database yet, so reduce the log level | |
4482 | * to INFO for that case. */ | |
4483 | VLOG(errno == ENODEV ? VLL_INFO : VLL_ERR, | |
4484 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", | |
10a89ef0 | 4485 | netdev_name, ovs_strerror(errno)); |
8b61709d BP |
4486 | return errno; |
4487 | } | |
4488 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
4489 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { | |
4490 | VLOG_WARN("%s device has unknown hardware address family %d", | |
4491 | netdev_name, hwaddr_family); | |
4492 | } | |
4493 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
4494 | return 0; | |
4495 | } | |
4496 | ||
4497 | static int | |
44445cac | 4498 | set_etheraddr(const char *netdev_name, |
8b61709d BP |
4499 | const uint8_t mac[ETH_ADDR_LEN]) |
4500 | { | |
4501 | struct ifreq ifr; | |
4502 | ||
4503 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4504 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 4505 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
8b61709d BP |
4506 | memcpy(ifr.ifr_hwaddr.sa_data, mac, ETH_ADDR_LEN); |
4507 | COVERAGE_INC(netdev_set_hwaddr); | |
4508 | if (ioctl(af_inet_sock, SIOCSIFHWADDR, &ifr) < 0) { | |
4509 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", | |
10a89ef0 | 4510 | netdev_name, ovs_strerror(errno)); |
8b61709d BP |
4511 | return errno; |
4512 | } | |
4513 | return 0; | |
4514 | } | |
4515 | ||
4516 | static int | |
0b0544d7 | 4517 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
4518 | int cmd, const char *cmd_name) |
4519 | { | |
4520 | struct ifreq ifr; | |
4521 | ||
4522 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 4523 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
4524 | ifr.ifr_data = (caddr_t) ecmd; |
4525 | ||
4526 | ecmd->cmd = cmd; | |
8b61709d BP |
4527 | if (ioctl(af_inet_sock, SIOCETHTOOL, &ifr) == 0) { |
4528 | return 0; | |
4529 | } else { | |
4530 | if (errno != EOPNOTSUPP) { | |
4531 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " | |
10a89ef0 | 4532 | "failed: %s", cmd_name, name, ovs_strerror(errno)); |
8b61709d BP |
4533 | } else { |
4534 | /* The device doesn't support this operation. That's pretty | |
4535 | * common, so there's no point in logging anything. */ | |
4536 | } | |
4537 | return errno; | |
4538 | } | |
4539 | } | |
4540 | ||
4541 | static int | |
149f577a JG |
4542 | netdev_linux_do_ioctl(const char *name, struct ifreq *ifr, int cmd, |
4543 | const char *cmd_name) | |
8b61709d | 4544 | { |
71d7c22f | 4545 | ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name); |
8b61709d | 4546 | if (ioctl(af_inet_sock, cmd, ifr) == -1) { |
149f577a | 4547 | VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name, |
10a89ef0 | 4548 | ovs_strerror(errno)); |
8b61709d BP |
4549 | return errno; |
4550 | } | |
4551 | return 0; | |
4552 | } | |
f1acd62b BP |
4553 | |
4554 | static int | |
4555 | netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, | |
4556 | int cmd, const char *cmd_name) | |
4557 | { | |
4558 | struct ifreq ifr; | |
4559 | int error; | |
4560 | ||
4561 | ifr.ifr_addr.sa_family = AF_INET; | |
149f577a | 4562 | error = netdev_linux_do_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); |
f1acd62b | 4563 | if (!error) { |
db5a1019 AW |
4564 | const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, |
4565 | &ifr.ifr_addr); | |
f1acd62b BP |
4566 | *ip = sin->sin_addr; |
4567 | } | |
4568 | return error; | |
4569 | } | |
488d734d BP |
4570 | |
4571 | /* Returns an AF_PACKET raw socket or a negative errno value. */ | |
4572 | static int | |
4573 | af_packet_sock(void) | |
4574 | { | |
23882115 BP |
4575 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
4576 | static int sock; | |
488d734d | 4577 | |
23882115 | 4578 | if (ovsthread_once_start(&once)) { |
488d734d BP |
4579 | sock = socket(AF_PACKET, SOCK_RAW, 0); |
4580 | if (sock >= 0) { | |
8450059e BP |
4581 | int error = set_nonblocking(sock); |
4582 | if (error) { | |
4583 | close(sock); | |
4584 | sock = -error; | |
4585 | } | |
488d734d BP |
4586 | } else { |
4587 | sock = -errno; | |
10a89ef0 BP |
4588 | VLOG_ERR("failed to create packet socket: %s", |
4589 | ovs_strerror(errno)); | |
488d734d | 4590 | } |
23882115 | 4591 | ovsthread_once_done(&once); |
488d734d BP |
4592 | } |
4593 | ||
4594 | return sock; | |
4595 | } |