]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
c7952afb | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015 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 | 22 | #include <fcntl.h> |
55bc98d6 | 23 | #include <arpa/inet.h> |
8b61709d | 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 | 35 | #include <linux/sockios.h> |
8b61709d BP |
36 | #include <sys/types.h> |
37 | #include <sys/ioctl.h> | |
38 | #include <sys/socket.h> | |
ac3e3aaa | 39 | #include <sys/utsname.h> |
55bc98d6 | 40 | #include <netpacket/packet.h> |
8b61709d BP |
41 | #include <net/if.h> |
42 | #include <net/if_arp.h> | |
55bc98d6 | 43 | #include <net/if_packet.h> |
8b61709d BP |
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" | |
e14deea0 | 52 | #include "dp-packet.h" |
93451a0a | 53 | #include "dpif-netlink.h" |
df1e5a3b | 54 | #include "dpif-netdev.h" |
8b61709d BP |
55 | #include "dynamic-string.h" |
56 | #include "fatal-signal.h" | |
93b13be8 BP |
57 | #include "hash.h" |
58 | #include "hmap.h" | |
8b61709d | 59 | #include "netdev-provider.h" |
7fbef77a | 60 | #include "netdev-vport.h" |
45c8d3a1 | 61 | #include "netlink-notifier.h" |
2fe27d5a | 62 | #include "netlink-socket.h" |
c060c4cf | 63 | #include "netlink.h" |
e9e28be3 | 64 | #include "ofpbuf.h" |
8b61709d | 65 | #include "openflow/openflow.h" |
19c8e9c1 | 66 | #include "ovs-atomic.h" |
8b61709d BP |
67 | #include "packets.h" |
68 | #include "poll-loop.h" | |
7e9dcc0f | 69 | #include "rtnetlink.h" |
8b61709d | 70 | #include "shash.h" |
c060c4cf | 71 | #include "socket-util.h" |
19993ef3 | 72 | #include "sset.h" |
1670c579 | 73 | #include "timer.h" |
c060c4cf | 74 | #include "unaligned.h" |
e6211adc | 75 | #include "openvswitch/vlog.h" |
5136ce49 | 76 | |
d98e6007 | 77 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea | 78 | |
d76f09ea BP |
79 | COVERAGE_DEFINE(netdev_set_policing); |
80 | COVERAGE_DEFINE(netdev_arp_lookup); | |
81 | COVERAGE_DEFINE(netdev_get_ifindex); | |
82 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
83 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
ab985a77 BP |
84 | COVERAGE_DEFINE(netdev_get_ethtool); |
85 | COVERAGE_DEFINE(netdev_set_ethtool); | |
4f925bd3 | 86 | |
8b61709d BP |
87 | \f |
88 | /* These were introduced in Linux 2.6.14, so they might be missing if we have | |
89 | * old headers. */ | |
90 | #ifndef ADVERTISED_Pause | |
91 | #define ADVERTISED_Pause (1 << 13) | |
92 | #endif | |
93 | #ifndef ADVERTISED_Asym_Pause | |
94 | #define ADVERTISED_Asym_Pause (1 << 14) | |
95 | #endif | |
96 | ||
e47bd51a JP |
97 | /* These were introduced in Linux 2.6.24, so they might be missing if we |
98 | * have old headers. */ | |
99 | #ifndef ETHTOOL_GFLAGS | |
100 | #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ | |
101 | #endif | |
102 | #ifndef ETHTOOL_SFLAGS | |
103 | #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ | |
104 | #endif | |
105 | ||
c1c9c9c4 BP |
106 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
107 | * headers. */ | |
108 | #ifndef TC_RTAB_SIZE | |
109 | #define TC_RTAB_SIZE 1024 | |
110 | #endif | |
111 | ||
b73c8518 SH |
112 | /* Linux 2.6.21 introduced struct tpacket_auxdata. |
113 | * Linux 2.6.27 added the tp_vlan_tci member. | |
114 | * Linux 3.0 defined TP_STATUS_VLAN_VALID. | |
115 | * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined | |
116 | * TP_STATUS_VLAN_TPID_VALID. | |
117 | * | |
118 | * With all this churn it's easiest to unconditionally define a replacement | |
119 | * structure that has everything we want. | |
120 | */ | |
55bc98d6 BP |
121 | #ifndef PACKET_AUXDATA |
122 | #define PACKET_AUXDATA 8 | |
123 | #endif | |
b73c8518 SH |
124 | #ifndef TP_STATUS_VLAN_VALID |
125 | #define TP_STATUS_VLAN_VALID (1 << 4) | |
126 | #endif | |
127 | #ifndef TP_STATUS_VLAN_TPID_VALID | |
128 | #define TP_STATUS_VLAN_TPID_VALID (1 << 6) | |
129 | #endif | |
130 | #undef tpacket_auxdata | |
131 | #define tpacket_auxdata rpl_tpacket_auxdata | |
132 | struct tpacket_auxdata { | |
133 | uint32_t tp_status; | |
134 | uint32_t tp_len; | |
135 | uint32_t tp_snaplen; | |
136 | uint16_t tp_mac; | |
137 | uint16_t tp_net; | |
138 | uint16_t tp_vlan_tci; | |
139 | uint16_t tp_vlan_tpid; | |
140 | }; | |
141 | ||
fa373af4 BP |
142 | /* Linux 2.6.35 introduced IFLA_STATS64 and rtnl_link_stats64. |
143 | * | |
144 | * Tests for rtnl_link_stats64 don't seem to consistently work, e.g. on | |
145 | * 2.6.32-431.29.2.el6.x86_64 (see report at | |
146 | * http://openvswitch.org/pipermail/dev/2014-October/047978.html). Maybe | |
147 | * if_link.h is not self-contained on those kernels. It is easiest to | |
148 | * unconditionally define a replacement. */ | |
149 | #ifndef IFLA_STATS64 | |
337c9b99 | 150 | #define IFLA_STATS64 23 |
fa373af4 BP |
151 | #endif |
152 | #define rtnl_link_stats64 rpl_rtnl_link_stats64 | |
337c9b99 BP |
153 | struct rtnl_link_stats64 { |
154 | uint64_t rx_packets; | |
155 | uint64_t tx_packets; | |
156 | uint64_t rx_bytes; | |
157 | uint64_t tx_bytes; | |
158 | uint64_t rx_errors; | |
159 | uint64_t tx_errors; | |
160 | uint64_t rx_dropped; | |
161 | uint64_t tx_dropped; | |
162 | uint64_t multicast; | |
163 | uint64_t collisions; | |
164 | ||
165 | uint64_t rx_length_errors; | |
166 | uint64_t rx_over_errors; | |
167 | uint64_t rx_crc_errors; | |
168 | uint64_t rx_frame_errors; | |
169 | uint64_t rx_fifo_errors; | |
170 | uint64_t rx_missed_errors; | |
171 | ||
172 | uint64_t tx_aborted_errors; | |
173 | uint64_t tx_carrier_errors; | |
174 | uint64_t tx_fifo_errors; | |
175 | uint64_t tx_heartbeat_errors; | |
176 | uint64_t tx_window_errors; | |
177 | ||
178 | uint64_t rx_compressed; | |
179 | uint64_t tx_compressed; | |
180 | }; | |
337c9b99 | 181 | |
8b61709d | 182 | enum { |
7fbef77a JG |
183 | VALID_IFINDEX = 1 << 0, |
184 | VALID_ETHERADDR = 1 << 1, | |
185 | VALID_IN4 = 1 << 2, | |
186 | VALID_IN6 = 1 << 3, | |
187 | VALID_MTU = 1 << 4, | |
3a183124 | 188 | VALID_POLICING = 1 << 5, |
4f925bd3 PS |
189 | VALID_VPORT_STAT_ERROR = 1 << 6, |
190 | VALID_DRVINFO = 1 << 7, | |
51f87458 | 191 | VALID_FEATURES = 1 << 8, |
8b61709d | 192 | }; |
c1c9c9c4 BP |
193 | \f |
194 | /* Traffic control. */ | |
195 | ||
196 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
197 | * network device. |
198 | * | |
199 | * Each TC implementation subclasses this with whatever additional data it | |
200 | * needs. */ | |
c1c9c9c4 BP |
201 | struct tc { |
202 | const struct tc_ops *ops; | |
93b13be8 BP |
203 | struct hmap queues; /* Contains "struct tc_queue"s. |
204 | * Read by generic TC layer. | |
205 | * Written only by TC implementation. */ | |
206 | }; | |
c1c9c9c4 | 207 | |
559eb230 BP |
208 | #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) } |
209 | ||
93b13be8 BP |
210 | /* One traffic control queue. |
211 | * | |
212 | * Each TC implementation subclasses this with whatever additional data it | |
213 | * needs. */ | |
214 | struct tc_queue { | |
215 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
216 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
6dc34a0d | 217 | long long int created; /* Time queue was created, in msecs. */ |
c1c9c9c4 BP |
218 | }; |
219 | ||
220 | /* A particular kind of traffic control. Each implementation generally maps to | |
221 | * one particular Linux qdisc class. | |
222 | * | |
223 | * The functions below return 0 if successful or a positive errno value on | |
224 | * failure, except where otherwise noted. All of them must be provided, except | |
225 | * where otherwise noted. */ | |
226 | struct tc_ops { | |
227 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
228 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
229 | * appropriate values. */ | |
230 | const char *linux_name; | |
231 | ||
232 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
233 | const char *ovs_name; | |
234 | ||
235 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
236 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
237 | unsigned int n_queues; | |
238 | ||
239 | /* Called to install this TC class on 'netdev'. The implementation should | |
240 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
241 | * and configure it according to 'details'. The implementation may assume | |
242 | * that the current qdisc is the default; that is, there is no need for it | |
243 | * to delete the current qdisc before installing itself. | |
244 | * | |
245 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
246 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
247 | * (which is built as ovs-vswitchd.conf.db(8)). | |
248 | * | |
249 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
250 | * initialized 'struct tc'. | |
251 | * | |
252 | * (This function is null for tc_ops_other, which cannot be installed. For | |
253 | * other TC classes it should always be nonnull.) */ | |
79f1cbe9 | 254 | int (*tc_install)(struct netdev *netdev, const struct smap *details); |
c1c9c9c4 BP |
255 | |
256 | /* Called when the netdev code determines (through a Netlink query) that | |
257 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
258 | * it ourselves and so don't know any of the details. | |
259 | * | |
260 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
261 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
262 | * implementation should parse the other attributes of 'nlmsg' as | |
263 | * necessary to determine its configuration. If necessary it should also | |
264 | * use Netlink queries to determine the configuration of queues on | |
265 | * 'netdev'. | |
266 | * | |
267 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
268 | * initialized 'struct tc'. */ | |
269 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
270 | ||
271 | /* Destroys the data structures allocated by the implementation as part of | |
272 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
273 | * tc_destroy(tc). | |
274 | * | |
275 | * The implementation should not need to perform any Netlink calls. If | |
276 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
277 | * (But it may not be desirable.) | |
278 | * | |
279 | * This function may be null if 'tc' is trivial. */ | |
280 | void (*tc_destroy)(struct tc *tc); | |
281 | ||
282 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
283 | * | |
284 | * The implementation should not need to perform any Netlink calls, because | |
285 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
286 | * cached the configuration. | |
287 | * | |
288 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
289 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
290 | * (which is built as ovs-vswitchd.conf.db(8)). | |
291 | * | |
292 | * This function may be null if 'tc' is not configurable. | |
293 | */ | |
79f1cbe9 | 294 | int (*qdisc_get)(const struct netdev *netdev, struct smap *details); |
c1c9c9c4 BP |
295 | |
296 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
297 | * required Netlink calls to complete the reconfiguration. | |
298 | * | |
299 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
300 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
301 | * (which is built as ovs-vswitchd.conf.db(8)). | |
302 | * | |
303 | * This function may be null if 'tc' is not configurable. | |
304 | */ | |
79f1cbe9 | 305 | int (*qdisc_set)(struct netdev *, const struct smap *details); |
c1c9c9c4 | 306 | |
93b13be8 BP |
307 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
308 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
309 | * |
310 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
311 | * in the "other_config" column in the "Queue" table in | |
312 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
313 | * | |
314 | * The implementation should not need to perform any Netlink calls, because | |
315 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
316 | * cached the queue configuration. | |
317 | * | |
318 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
319 | * 0). */ | |
93b13be8 | 320 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
79f1cbe9 | 321 | struct smap *details); |
c1c9c9c4 BP |
322 | |
323 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
324 | * 'details', perfoming any required Netlink calls to complete the | |
325 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
326 | * 'n_queues'. | |
327 | * | |
328 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
329 | * in the "other_config" column in the "Queue" table in | |
330 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
331 | * | |
332 | * This function may be null if 'tc' does not have queues or its queues are | |
333 | * not configurable. */ | |
334 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
79f1cbe9 | 335 | const struct smap *details); |
c1c9c9c4 | 336 | |
93b13be8 BP |
337 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
338 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
339 | * |
340 | * This function may be null if 'tc' does not have queues or its queues | |
341 | * cannot be deleted. */ | |
93b13be8 | 342 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 343 | |
93b13be8 BP |
344 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
345 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
346 | * |
347 | * On success, initializes '*stats'. | |
348 | * | |
349 | * This function may be null if 'tc' does not have queues or if it cannot | |
350 | * report queue statistics. */ | |
93b13be8 BP |
351 | int (*class_get_stats)(const struct netdev *netdev, |
352 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
353 | struct netdev_queue_stats *stats); |
354 | ||
355 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
356 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
357 | * | |
358 | * This function may be null if 'tc' does not have queues or if it cannot | |
359 | * report queue statistics. */ | |
360 | int (*class_dump_stats)(const struct netdev *netdev, | |
361 | const struct ofpbuf *nlmsg, | |
362 | netdev_dump_queue_stats_cb *cb, void *aux); | |
363 | }; | |
364 | ||
365 | static void | |
366 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
367 | { | |
368 | tc->ops = ops; | |
93b13be8 | 369 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
370 | } |
371 | ||
372 | static void | |
373 | tc_destroy(struct tc *tc) | |
374 | { | |
93b13be8 | 375 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
376 | } |
377 | ||
378 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 379 | static const struct tc_ops tc_ops_hfsc; |
677d9158 JV |
380 | static const struct tc_ops tc_ops_codel; |
381 | static const struct tc_ops tc_ops_fqcodel; | |
382 | static const struct tc_ops tc_ops_sfq; | |
c1c9c9c4 BP |
383 | static const struct tc_ops tc_ops_default; |
384 | static const struct tc_ops tc_ops_other; | |
385 | ||
559eb230 | 386 | static const struct tc_ops *const tcs[] = { |
c1c9c9c4 | 387 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ |
a339aa81 | 388 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
677d9158 JV |
389 | &tc_ops_codel, /* Controlled delay */ |
390 | &tc_ops_fqcodel, /* Fair queue controlled delay */ | |
391 | &tc_ops_sfq, /* Stochastic fair queueing */ | |
c1c9c9c4 BP |
392 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
393 | &tc_ops_other, /* Some other qdisc. */ | |
394 | NULL | |
395 | }; | |
149f577a | 396 | |
c1c9c9c4 BP |
397 | static unsigned int tc_make_handle(unsigned int major, unsigned int minor); |
398 | static unsigned int tc_get_major(unsigned int handle); | |
399 | static unsigned int tc_get_minor(unsigned int handle); | |
400 | ||
401 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); | |
402 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
403 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
404 | ||
405 | static struct tcmsg *tc_make_request(const struct netdev *, int type, | |
406 | unsigned int flags, struct ofpbuf *); | |
407 | static int tc_transact(struct ofpbuf *request, struct ofpbuf **replyp); | |
f8500004 | 408 | static int tc_add_del_ingress_qdisc(struct netdev *netdev, bool add); |
c7952afb BP |
409 | static int tc_add_policer(struct netdev *, |
410 | uint32_t kbits_rate, uint32_t kbits_burst); | |
c1c9c9c4 BP |
411 | |
412 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
413 | struct nlattr **options); | |
414 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
415 | struct nlattr **options, | |
416 | struct netdev_queue_stats *); | |
417 | static int tc_query_class(const struct netdev *, | |
418 | unsigned int handle, unsigned int parent, | |
419 | struct ofpbuf **replyp); | |
420 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
421 | ||
422 | static int tc_del_qdisc(struct netdev *netdev); | |
423 | static int tc_query_qdisc(const struct netdev *netdev); | |
424 | ||
425 | static int tc_calc_cell_log(unsigned int mtu); | |
426 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
427 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
428 | const struct tc_ratespec *rate); | |
429 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
430 | \f | |
b5d57fc8 BP |
431 | struct netdev_linux { |
432 | struct netdev up; | |
149f577a | 433 | |
86383816 BP |
434 | /* Protects all members below. */ |
435 | struct ovs_mutex mutex; | |
436 | ||
149f577a | 437 | unsigned int cache_valid; |
8b61709d | 438 | |
1670c579 EJ |
439 | bool miimon; /* Link status of last poll. */ |
440 | long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ | |
441 | struct timer miimon_timer; | |
442 | ||
8722022c BP |
443 | /* The following are figured out "on demand" only. They are only valid |
444 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d | 445 | int ifindex; |
74ff3298 | 446 | struct eth_addr etheraddr; |
f1acd62b | 447 | struct in_addr address, netmask; |
8b61709d BP |
448 | struct in6_addr in6; |
449 | int mtu; | |
059e5f4f | 450 | unsigned int ifi_flags; |
65c3058c | 451 | long long int carrier_resets; |
80a86fbe BP |
452 | uint32_t kbits_rate; /* Policing data. */ |
453 | uint32_t kbits_burst; | |
bba1e6f3 PS |
454 | int vport_stats_error; /* Cached error code from vport_get_stats(). |
455 | 0 or an errno value. */ | |
90a6637d | 456 | int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ |
44445cac | 457 | int ether_addr_error; /* Cached error code from set/get etheraddr. */ |
c9f71668 | 458 | int netdev_policing_error; /* Cached error code from set policing. */ |
51f87458 | 459 | int get_features_error; /* Cached error code from ETHTOOL_GSET. */ |
c7b1b0a5 | 460 | int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */ |
49af9a3d | 461 | int in4_error; /* Cached error code from reading in4 addr. */ |
7df6932e | 462 | int in6_error; /* Cached error code from reading in6 addr. */ |
51f87458 | 463 | |
a00ca915 EJ |
464 | enum netdev_features current; /* Cached from ETHTOOL_GSET. */ |
465 | enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ | |
466 | enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ | |
90a6637d | 467 | |
4f925bd3 | 468 | struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ |
c1c9c9c4 | 469 | struct tc *tc; |
149f577a | 470 | |
d0d08f8a BP |
471 | /* For devices of class netdev_tap_class only. */ |
472 | int tap_fd; | |
8b61709d BP |
473 | }; |
474 | ||
f7791740 PS |
475 | struct netdev_rxq_linux { |
476 | struct netdev_rxq up; | |
796223f5 | 477 | bool is_tap; |
5b7448ed | 478 | int fd; |
149f577a | 479 | }; |
8b61709d | 480 | |
8b61709d BP |
481 | /* This is set pretty low because we probably won't learn anything from the |
482 | * additional log messages. */ | |
483 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
484 | ||
19c8e9c1 JS |
485 | /* Polling miimon status for all ports causes performance degradation when |
486 | * handling a large number of ports. If there are no devices using miimon, then | |
812c272c JR |
487 | * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). |
488 | * | |
489 | * Readers do not depend on this variable synchronizing with the related | |
490 | * changes in the device miimon status, so we can use atomic_count. */ | |
491 | static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0); | |
19c8e9c1 | 492 | |
259e0b1a | 493 | static void netdev_linux_run(void); |
6f643e49 | 494 | |
0b0544d7 | 495 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 496 | int cmd, const char *cmd_name); |
f1acd62b BP |
497 | static int netdev_linux_get_ipv4(const struct netdev *, struct in_addr *, |
498 | int cmd, const char *cmd_name); | |
b5d57fc8 | 499 | static int get_flags(const struct netdev *, unsigned int *flags); |
4b609110 | 500 | static int set_flags(const char *, unsigned int flags); |
4f9f3f21 BP |
501 | static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
502 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
503 | OVS_REQUIRES(netdev->mutex); | |
8b61709d BP |
504 | static int do_get_ifindex(const char *netdev_name); |
505 | static int get_ifindex(const struct netdev *, int *ifindexp); | |
506 | static int do_set_addr(struct netdev *netdev, | |
507 | int ioctl_nr, const char *ioctl_name, | |
508 | struct in_addr addr); | |
74ff3298 JR |
509 | static int get_etheraddr(const char *netdev_name, struct eth_addr *ea); |
510 | static int set_etheraddr(const char *netdev_name, const struct eth_addr); | |
35eef899 | 511 | static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *); |
488d734d | 512 | static int af_packet_sock(void); |
19c8e9c1 | 513 | static bool netdev_linux_miimon_enabled(void); |
1670c579 EJ |
514 | static void netdev_linux_miimon_run(void); |
515 | static void netdev_linux_miimon_wait(void); | |
df1e5a3b | 516 | static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); |
8b61709d | 517 | |
15b3596a JG |
518 | static bool |
519 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
520 | { | |
259e0b1a | 521 | return netdev_class->run == netdev_linux_run; |
15b3596a JG |
522 | } |
523 | ||
796223f5 BP |
524 | static bool |
525 | is_tap_netdev(const struct netdev *netdev) | |
526 | { | |
b5d57fc8 | 527 | return netdev_get_class(netdev) == &netdev_tap_class; |
6c88d577 JP |
528 | } |
529 | ||
8b61709d BP |
530 | static struct netdev_linux * |
531 | netdev_linux_cast(const struct netdev *netdev) | |
532 | { | |
b5d57fc8 | 533 | ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); |
15b3596a | 534 | |
180c6d0b | 535 | return CONTAINER_OF(netdev, struct netdev_linux, up); |
8b61709d | 536 | } |
796223f5 | 537 | |
f7791740 PS |
538 | static struct netdev_rxq_linux * |
539 | netdev_rxq_linux_cast(const struct netdev_rxq *rx) | |
796223f5 | 540 | { |
9dc63482 | 541 | ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev))); |
f7791740 | 542 | return CONTAINER_OF(rx, struct netdev_rxq_linux, up); |
796223f5 | 543 | } |
ff4ed3c9 | 544 | \f |
cee87338 | 545 | static void netdev_linux_update(struct netdev_linux *netdev, |
7e9dcc0f | 546 | const struct rtnetlink_change *) |
86383816 | 547 | OVS_REQUIRES(netdev->mutex); |
cee87338 | 548 | static void netdev_linux_changed(struct netdev_linux *netdev, |
86383816 BP |
549 | unsigned int ifi_flags, unsigned int mask) |
550 | OVS_REQUIRES(netdev->mutex); | |
cee87338 | 551 | |
d6384a3a AW |
552 | /* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK, |
553 | * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL | |
cee87338 BP |
554 | * if no such socket could be created. */ |
555 | static struct nl_sock * | |
556 | netdev_linux_notify_sock(void) | |
557 | { | |
558 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
559 | static struct nl_sock *sock; | |
d6384a3a AW |
560 | unsigned int mcgroups[3] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR, |
561 | RTNLGRP_IPV6_IFADDR}; | |
cee87338 BP |
562 | |
563 | if (ovsthread_once_start(&once)) { | |
564 | int error; | |
565 | ||
566 | error = nl_sock_create(NETLINK_ROUTE, &sock); | |
567 | if (!error) { | |
d6384a3a AW |
568 | size_t i; |
569 | ||
570 | for (i = 0; i < ARRAY_SIZE(mcgroups); i++) { | |
571 | error = nl_sock_join_mcgroup(sock, mcgroups[i]); | |
572 | if (error) { | |
573 | nl_sock_destroy(sock); | |
574 | sock = NULL; | |
575 | break; | |
576 | } | |
cee87338 BP |
577 | } |
578 | } | |
579 | ovsthread_once_done(&once); | |
580 | } | |
581 | ||
582 | return sock; | |
583 | } | |
584 | ||
19c8e9c1 JS |
585 | static bool |
586 | netdev_linux_miimon_enabled(void) | |
587 | { | |
812c272c | 588 | return atomic_count_get(&miimon_cnt) > 0; |
19c8e9c1 JS |
589 | } |
590 | ||
8b61709d BP |
591 | static void |
592 | netdev_linux_run(void) | |
593 | { | |
cee87338 BP |
594 | struct nl_sock *sock; |
595 | int error; | |
596 | ||
19c8e9c1 JS |
597 | if (netdev_linux_miimon_enabled()) { |
598 | netdev_linux_miimon_run(); | |
599 | } | |
cee87338 BP |
600 | |
601 | sock = netdev_linux_notify_sock(); | |
602 | if (!sock) { | |
603 | return; | |
604 | } | |
605 | ||
606 | do { | |
607 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
608 | uint64_t buf_stub[4096 / 8]; | |
609 | struct ofpbuf buf; | |
610 | ||
611 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
612 | error = nl_sock_recv(sock, &buf, false); | |
613 | if (!error) { | |
7e9dcc0f | 614 | struct rtnetlink_change change; |
cee87338 | 615 | |
7e9dcc0f | 616 | if (rtnetlink_parse(&buf, &change)) { |
cee87338 BP |
617 | struct netdev *netdev_ = netdev_from_name(change.ifname); |
618 | if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { | |
619 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
86383816 BP |
620 | |
621 | ovs_mutex_lock(&netdev->mutex); | |
cee87338 | 622 | netdev_linux_update(netdev, &change); |
86383816 | 623 | ovs_mutex_unlock(&netdev->mutex); |
cee87338 | 624 | } |
38e0065b | 625 | netdev_close(netdev_); |
cee87338 BP |
626 | } |
627 | } else if (error == ENOBUFS) { | |
628 | struct shash device_shash; | |
629 | struct shash_node *node; | |
630 | ||
631 | nl_sock_drain(sock); | |
632 | ||
633 | shash_init(&device_shash); | |
634 | netdev_get_devices(&netdev_linux_class, &device_shash); | |
635 | SHASH_FOR_EACH (node, &device_shash) { | |
636 | struct netdev *netdev_ = node->data; | |
637 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
638 | unsigned int flags; | |
639 | ||
86383816 | 640 | ovs_mutex_lock(&netdev->mutex); |
cee87338 BP |
641 | get_flags(netdev_, &flags); |
642 | netdev_linux_changed(netdev, flags, 0); | |
86383816 BP |
643 | ovs_mutex_unlock(&netdev->mutex); |
644 | ||
cee87338 BP |
645 | netdev_close(netdev_); |
646 | } | |
647 | shash_destroy(&device_shash); | |
648 | } else if (error != EAGAIN) { | |
649 | VLOG_WARN_RL(&rl, "error reading or parsing netlink (%s)", | |
650 | ovs_strerror(error)); | |
651 | } | |
652 | ofpbuf_uninit(&buf); | |
653 | } while (!error); | |
8b61709d BP |
654 | } |
655 | ||
656 | static void | |
657 | netdev_linux_wait(void) | |
658 | { | |
cee87338 BP |
659 | struct nl_sock *sock; |
660 | ||
19c8e9c1 JS |
661 | if (netdev_linux_miimon_enabled()) { |
662 | netdev_linux_miimon_wait(); | |
663 | } | |
cee87338 BP |
664 | sock = netdev_linux_notify_sock(); |
665 | if (sock) { | |
666 | nl_sock_wait(sock, POLLIN); | |
667 | } | |
8b61709d BP |
668 | } |
669 | ||
ac4d3bcb | 670 | static void |
b5d57fc8 BP |
671 | netdev_linux_changed(struct netdev_linux *dev, |
672 | unsigned int ifi_flags, unsigned int mask) | |
86383816 | 673 | OVS_REQUIRES(dev->mutex) |
ac4d3bcb | 674 | { |
3e912ffc | 675 | netdev_change_seq_changed(&dev->up); |
8aa77183 BP |
676 | |
677 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
678 | dev->carrier_resets++; | |
679 | } | |
680 | dev->ifi_flags = ifi_flags; | |
681 | ||
4f925bd3 PS |
682 | dev->cache_valid &= mask; |
683 | } | |
684 | ||
685 | static void | |
b5d57fc8 | 686 | netdev_linux_update(struct netdev_linux *dev, |
7e9dcc0f | 687 | const struct rtnetlink_change *change) |
86383816 | 688 | OVS_REQUIRES(dev->mutex) |
4f925bd3 | 689 | { |
d6384a3a AW |
690 | if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)){ |
691 | if (change->nlmsg_type == RTM_NEWLINK) { | |
692 | /* Keep drv-info, in4, in6. */ | |
693 | netdev_linux_changed(dev, change->ifi_flags, | |
694 | VALID_DRVINFO | VALID_IN4 | VALID_IN6); | |
695 | ||
696 | /* Update netdev from rtnl-change msg. */ | |
697 | if (change->mtu) { | |
698 | dev->mtu = change->mtu; | |
699 | dev->cache_valid |= VALID_MTU; | |
700 | dev->netdev_mtu_error = 0; | |
701 | } | |
90a6637d | 702 | |
74ff3298 JR |
703 | if (!eth_addr_is_zero(change->mac)) { |
704 | dev->etheraddr = change->mac; | |
d6384a3a AW |
705 | dev->cache_valid |= VALID_ETHERADDR; |
706 | dev->ether_addr_error = 0; | |
707 | } | |
44445cac | 708 | |
d6384a3a AW |
709 | dev->ifindex = change->if_index; |
710 | dev->cache_valid |= VALID_IFINDEX; | |
711 | dev->get_ifindex_error = 0; | |
712 | } else { | |
713 | netdev_linux_changed(dev, change->ifi_flags, 0); | |
714 | } | |
715 | } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) { | |
716 | /* Invalidates in4, in6. */ | |
717 | netdev_linux_changed(dev, dev->ifi_flags, | |
718 | ~(VALID_IN4 | VALID_IN6)); | |
4f925bd3 | 719 | } else { |
d6384a3a | 720 | OVS_NOT_REACHED(); |
4f925bd3 | 721 | } |
ac4d3bcb EJ |
722 | } |
723 | ||
9dc63482 BP |
724 | static struct netdev * |
725 | netdev_linux_alloc(void) | |
726 | { | |
727 | struct netdev_linux *netdev = xzalloc(sizeof *netdev); | |
728 | return &netdev->up; | |
729 | } | |
730 | ||
cee87338 | 731 | static void |
9dc63482 BP |
732 | netdev_linux_common_construct(struct netdev_linux *netdev) |
733 | { | |
834d6caf | 734 | ovs_mutex_init(&netdev->mutex); |
9dc63482 BP |
735 | } |
736 | ||
1f6e0fbd BP |
737 | /* Creates system and internal devices. */ |
738 | static int | |
9dc63482 | 739 | netdev_linux_construct(struct netdev *netdev_) |
1f6e0fbd | 740 | { |
9dc63482 | 741 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1f6e0fbd BP |
742 | int error; |
743 | ||
cee87338 | 744 | netdev_linux_common_construct(netdev); |
1f6e0fbd | 745 | |
b5d57fc8 BP |
746 | error = get_flags(&netdev->up, &netdev->ifi_flags); |
747 | if (error == ENODEV) { | |
9dc63482 | 748 | if (netdev->up.netdev_class != &netdev_internal_class) { |
b5d57fc8 | 749 | /* The device does not exist, so don't allow it to be opened. */ |
b5d57fc8 BP |
750 | return ENODEV; |
751 | } else { | |
752 | /* "Internal" netdevs have to be created as netdev objects before | |
753 | * they exist in the kernel, because creating them in the kernel | |
754 | * happens by passing a netdev object to dpif_port_add(). | |
755 | * Therefore, ignore the error. */ | |
756 | } | |
757 | } | |
46415c90 | 758 | |
a740f0de JG |
759 | return 0; |
760 | } | |
761 | ||
5b7448ed JG |
762 | /* For most types of netdevs we open the device for each call of |
763 | * netdev_open(). However, this is not the case with tap devices, | |
764 | * since it is only possible to open the device once. In this | |
765 | * situation we share a single file descriptor, and consequently | |
766 | * buffers, across all readers. Therefore once data is read it will | |
767 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 768 | static int |
9dc63482 | 769 | netdev_linux_construct_tap(struct netdev *netdev_) |
a740f0de | 770 | { |
9dc63482 | 771 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a740f0de | 772 | static const char tap_dev[] = "/dev/net/tun"; |
9dc63482 | 773 | const char *name = netdev_->name; |
a740f0de JG |
774 | struct ifreq ifr; |
775 | int error; | |
776 | ||
cee87338 | 777 | netdev_linux_common_construct(netdev); |
1f6e0fbd | 778 | |
6c88d577 | 779 | /* Open tap device. */ |
d0d08f8a BP |
780 | netdev->tap_fd = open(tap_dev, O_RDWR); |
781 | if (netdev->tap_fd < 0) { | |
6c88d577 | 782 | error = errno; |
10a89ef0 | 783 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); |
cee87338 | 784 | return error; |
6c88d577 JP |
785 | } |
786 | ||
787 | /* Create tap device. */ | |
788 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; | |
71d7c22f | 789 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
d0d08f8a | 790 | if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 | 791 | VLOG_WARN("%s: creating tap device failed: %s", name, |
10a89ef0 | 792 | ovs_strerror(errno)); |
6c88d577 | 793 | error = errno; |
f61d8d29 | 794 | goto error_close; |
6c88d577 JP |
795 | } |
796 | ||
797 | /* Make non-blocking. */ | |
d0d08f8a | 798 | error = set_nonblocking(netdev->tap_fd); |
a740f0de | 799 | if (error) { |
f61d8d29 | 800 | goto error_close; |
a740f0de JG |
801 | } |
802 | ||
803 | return 0; | |
804 | ||
f61d8d29 | 805 | error_close: |
d0d08f8a | 806 | close(netdev->tap_fd); |
a740f0de JG |
807 | return error; |
808 | } | |
809 | ||
6c88d577 | 810 | static void |
9dc63482 | 811 | netdev_linux_destruct(struct netdev *netdev_) |
6c88d577 | 812 | { |
b5d57fc8 | 813 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
6c88d577 | 814 | |
b5d57fc8 BP |
815 | if (netdev->tc && netdev->tc->ops->tc_destroy) { |
816 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 BP |
817 | } |
818 | ||
d0d08f8a BP |
819 | if (netdev_get_class(netdev_) == &netdev_tap_class |
820 | && netdev->tap_fd >= 0) | |
821 | { | |
822 | close(netdev->tap_fd); | |
6c88d577 | 823 | } |
86383816 | 824 | |
19c8e9c1 | 825 | if (netdev->miimon_interval > 0) { |
812c272c | 826 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
827 | } |
828 | ||
86383816 | 829 | ovs_mutex_destroy(&netdev->mutex); |
6c88d577 JP |
830 | } |
831 | ||
9dc63482 BP |
832 | static void |
833 | netdev_linux_dealloc(struct netdev *netdev_) | |
834 | { | |
835 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
836 | free(netdev); | |
837 | } | |
838 | ||
f7791740 PS |
839 | static struct netdev_rxq * |
840 | netdev_linux_rxq_alloc(void) | |
9dc63482 | 841 | { |
f7791740 | 842 | struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); |
9dc63482 BP |
843 | return &rx->up; |
844 | } | |
845 | ||
7b6b0ef4 | 846 | static int |
f7791740 | 847 | netdev_linux_rxq_construct(struct netdev_rxq *rxq_) |
7b6b0ef4 | 848 | { |
f7791740 | 849 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 850 | struct netdev *netdev_ = rx->up.netdev; |
7b6b0ef4 | 851 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7b6b0ef4 | 852 | int error; |
7b6b0ef4 | 853 | |
86383816 | 854 | ovs_mutex_lock(&netdev->mutex); |
9dc63482 BP |
855 | rx->is_tap = is_tap_netdev(netdev_); |
856 | if (rx->is_tap) { | |
857 | rx->fd = netdev->tap_fd; | |
796223f5 BP |
858 | } else { |
859 | struct sockaddr_ll sll; | |
b73c8518 | 860 | int ifindex, val; |
32383c3b | 861 | /* Result of tcpdump -dd inbound */ |
259e0b1a | 862 | static const struct sock_filter filt[] = { |
32383c3b MM |
863 | { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ |
864 | { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ | |
865 | { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ | |
866 | { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ | |
867 | }; | |
259e0b1a BP |
868 | static const struct sock_fprog fprog = { |
869 | ARRAY_SIZE(filt), (struct sock_filter *) filt | |
870 | }; | |
7b6b0ef4 | 871 | |
796223f5 | 872 | /* Create file descriptor. */ |
9dc63482 BP |
873 | rx->fd = socket(PF_PACKET, SOCK_RAW, 0); |
874 | if (rx->fd < 0) { | |
796223f5 | 875 | error = errno; |
10a89ef0 | 876 | VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); |
796223f5 BP |
877 | goto error; |
878 | } | |
33d82a56 | 879 | |
b73c8518 SH |
880 | val = 1; |
881 | if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { | |
882 | error = errno; | |
883 | VLOG_ERR("%s: failed to mark socket for auxdata (%s)", | |
884 | netdev_get_name(netdev_), ovs_strerror(error)); | |
885 | goto error; | |
886 | } | |
887 | ||
796223f5 | 888 | /* Set non-blocking mode. */ |
9dc63482 | 889 | error = set_nonblocking(rx->fd); |
796223f5 BP |
890 | if (error) { |
891 | goto error; | |
892 | } | |
7b6b0ef4 | 893 | |
796223f5 | 894 | /* Get ethernet device index. */ |
180c6d0b | 895 | error = get_ifindex(&netdev->up, &ifindex); |
796223f5 BP |
896 | if (error) { |
897 | goto error; | |
898 | } | |
7b6b0ef4 | 899 | |
796223f5 BP |
900 | /* Bind to specific ethernet device. */ |
901 | memset(&sll, 0, sizeof sll); | |
902 | sll.sll_family = AF_PACKET; | |
903 | sll.sll_ifindex = ifindex; | |
b73c8518 | 904 | sll.sll_protocol = htons(ETH_P_ALL); |
9dc63482 | 905 | if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { |
796223f5 BP |
906 | error = errno; |
907 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
10a89ef0 | 908 | netdev_get_name(netdev_), ovs_strerror(error)); |
796223f5 BP |
909 | goto error; |
910 | } | |
32383c3b MM |
911 | |
912 | /* Filter for only inbound packets. */ | |
9dc63482 | 913 | error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, |
32383c3b MM |
914 | sizeof fprog); |
915 | if (error) { | |
916 | error = errno; | |
259e0b1a | 917 | VLOG_ERR("%s: failed to attach filter (%s)", |
10a89ef0 | 918 | netdev_get_name(netdev_), ovs_strerror(error)); |
32383c3b MM |
919 | goto error; |
920 | } | |
7b6b0ef4 | 921 | } |
86383816 | 922 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 | 923 | |
7b6b0ef4 BP |
924 | return 0; |
925 | ||
926 | error: | |
9dc63482 BP |
927 | if (rx->fd >= 0) { |
928 | close(rx->fd); | |
7b6b0ef4 | 929 | } |
86383816 | 930 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 BP |
931 | return error; |
932 | } | |
933 | ||
796223f5 | 934 | static void |
f7791740 | 935 | netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) |
8b61709d | 936 | { |
f7791740 | 937 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
8b61709d | 938 | |
796223f5 BP |
939 | if (!rx->is_tap) { |
940 | close(rx->fd); | |
8b61709d | 941 | } |
9dc63482 BP |
942 | } |
943 | ||
944 | static void | |
f7791740 | 945 | netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) |
9dc63482 | 946 | { |
f7791740 | 947 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 948 | |
796223f5 BP |
949 | free(rx); |
950 | } | |
8b61709d | 951 | |
b73c8518 SH |
952 | static ovs_be16 |
953 | auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux) | |
954 | { | |
955 | if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { | |
956 | return htons(aux->tp_vlan_tpid); | |
957 | } else { | |
958 | return htons(ETH_TYPE_VLAN); | |
959 | } | |
960 | } | |
961 | ||
962 | static bool | |
963 | auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) | |
964 | { | |
965 | return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; | |
966 | } | |
967 | ||
796223f5 | 968 | static int |
cf62fa4c | 969 | netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer) |
796223f5 | 970 | { |
b73c8518 | 971 | size_t size; |
796223f5 | 972 | ssize_t retval; |
b73c8518 SH |
973 | struct iovec iov; |
974 | struct cmsghdr *cmsg; | |
975 | union { | |
976 | struct cmsghdr cmsg; | |
977 | char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; | |
978 | } cmsg_buffer; | |
979 | struct msghdr msgh; | |
980 | ||
981 | /* Reserve headroom for a single VLAN tag */ | |
cf62fa4c PS |
982 | dp_packet_reserve(buffer, VLAN_HEADER_LEN); |
983 | size = dp_packet_tailroom(buffer); | |
b73c8518 | 984 | |
cf62fa4c | 985 | iov.iov_base = dp_packet_data(buffer); |
b73c8518 SH |
986 | iov.iov_len = size; |
987 | msgh.msg_name = NULL; | |
988 | msgh.msg_namelen = 0; | |
989 | msgh.msg_iov = &iov; | |
990 | msgh.msg_iovlen = 1; | |
991 | msgh.msg_control = &cmsg_buffer; | |
992 | msgh.msg_controllen = sizeof cmsg_buffer; | |
993 | msgh.msg_flags = 0; | |
8e8cddf7 | 994 | |
796223f5 | 995 | do { |
b73c8518 | 996 | retval = recvmsg(fd, &msgh, MSG_TRUNC); |
796223f5 BP |
997 | } while (retval < 0 && errno == EINTR); |
998 | ||
bfd3367b | 999 | if (retval < 0) { |
b73c8518 SH |
1000 | return errno; |
1001 | } else if (retval > size) { | |
1002 | return EMSGSIZE; | |
1003 | } | |
1004 | ||
cf62fa4c | 1005 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1006 | |
1007 | for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) { | |
1008 | const struct tpacket_auxdata *aux; | |
1009 | ||
1010 | if (cmsg->cmsg_level != SOL_PACKET | |
1011 | || cmsg->cmsg_type != PACKET_AUXDATA | |
1012 | || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) { | |
1013 | continue; | |
8b61709d | 1014 | } |
b73c8518 SH |
1015 | |
1016 | aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); | |
1017 | if (auxdata_has_vlan_tci(aux)) { | |
1018 | if (retval < ETH_HEADER_LEN) { | |
1019 | return EINVAL; | |
1020 | } | |
1021 | ||
1022 | eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux), | |
1023 | htons(aux->tp_vlan_tci)); | |
1024 | break; | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | return 0; | |
1029 | } | |
1030 | ||
1031 | static int | |
cf62fa4c | 1032 | netdev_linux_rxq_recv_tap(int fd, struct dp_packet *buffer) |
b73c8518 SH |
1033 | { |
1034 | ssize_t retval; | |
cf62fa4c | 1035 | size_t size = dp_packet_tailroom(buffer); |
b73c8518 SH |
1036 | |
1037 | do { | |
cf62fa4c | 1038 | retval = read(fd, dp_packet_data(buffer), size); |
b73c8518 SH |
1039 | } while (retval < 0 && errno == EINTR); |
1040 | ||
1041 | if (retval < 0) { | |
bfd3367b SH |
1042 | return errno; |
1043 | } else if (retval > size) { | |
1044 | return EMSGSIZE; | |
8b61709d | 1045 | } |
b73c8518 | 1046 | |
cf62fa4c | 1047 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1048 | return 0; |
1049 | } | |
1050 | ||
1051 | static int | |
e14deea0 | 1052 | netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 1053 | int *c) |
b73c8518 | 1054 | { |
f7791740 | 1055 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
df1e5a3b | 1056 | struct netdev *netdev = rx->up.netdev; |
cf62fa4c | 1057 | struct dp_packet *buffer; |
df1e5a3b PS |
1058 | ssize_t retval; |
1059 | int mtu; | |
1060 | ||
1061 | if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
1062 | mtu = ETH_PAYLOAD_MAX; | |
1063 | } | |
1064 | ||
cf62fa4c | 1065 | buffer = dp_packet_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, |
91088554 | 1066 | DP_NETDEV_HEADROOM); |
b73c8518 | 1067 | retval = (rx->is_tap |
f7791740 PS |
1068 | ? netdev_linux_rxq_recv_tap(rx->fd, buffer) |
1069 | : netdev_linux_rxq_recv_sock(rx->fd, buffer)); | |
df1e5a3b PS |
1070 | |
1071 | if (retval) { | |
1072 | if (retval != EAGAIN && retval != EMSGSIZE) { | |
1073 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
f7791740 | 1074 | ovs_strerror(errno), netdev_rxq_get_name(rxq_)); |
df1e5a3b | 1075 | } |
cf62fa4c | 1076 | dp_packet_delete(buffer); |
df1e5a3b PS |
1077 | } else { |
1078 | dp_packet_pad(buffer); | |
2bc1bbd2 | 1079 | dp_packet_set_rss_hash(buffer, 0); |
cf62fa4c | 1080 | packets[0] = buffer; |
df1e5a3b | 1081 | *c = 1; |
b73c8518 SH |
1082 | } |
1083 | ||
1084 | return retval; | |
8b61709d BP |
1085 | } |
1086 | ||
8b61709d | 1087 | static void |
f7791740 | 1088 | netdev_linux_rxq_wait(struct netdev_rxq *rxq_) |
8b61709d | 1089 | { |
f7791740 | 1090 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1091 | poll_fd_wait(rx->fd, POLLIN); |
8b61709d BP |
1092 | } |
1093 | ||
8b61709d | 1094 | static int |
f7791740 | 1095 | netdev_linux_rxq_drain(struct netdev_rxq *rxq_) |
8b61709d | 1096 | { |
f7791740 | 1097 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1098 | if (rx->is_tap) { |
8b61709d | 1099 | struct ifreq ifr; |
f7791740 | 1100 | int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, |
259e0b1a | 1101 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
8b61709d BP |
1102 | if (error) { |
1103 | return error; | |
1104 | } | |
796223f5 | 1105 | drain_fd(rx->fd, ifr.ifr_qlen); |
8b61709d BP |
1106 | return 0; |
1107 | } else { | |
796223f5 | 1108 | return drain_rcvbuf(rx->fd); |
8b61709d BP |
1109 | } |
1110 | } | |
1111 | ||
1112 | /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive | |
1113 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued | |
1114 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
1115 | * the packet is too big or too small to transmit on the device. | |
1116 | * | |
1117 | * The caller retains ownership of 'buffer' in all cases. | |
1118 | * | |
1119 | * The kernel maintains a packet transmission queue, so the caller is not | |
1120 | * expected to do additional queuing of packets. */ | |
1121 | static int | |
f00fa8cb | 1122 | netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED, |
e14deea0 | 1123 | struct dp_packet **pkts, int cnt, bool may_steal) |
8b61709d | 1124 | { |
f4fd623c DDP |
1125 | int i; |
1126 | int error = 0; | |
40d26f04 | 1127 | |
f4fd623c DDP |
1128 | /* 'i' is incremented only if there's no error */ |
1129 | for (i = 0; i < cnt;) { | |
cf62fa4c PS |
1130 | const void *data = dp_packet_data(pkts[i]); |
1131 | size_t size = dp_packet_size(pkts[i]); | |
f23347ea | 1132 | ssize_t retval; |
8b61709d | 1133 | |
796223f5 | 1134 | if (!is_tap_netdev(netdev_)) { |
f23347ea BP |
1135 | /* Use our AF_PACKET socket to send to this device. */ |
1136 | struct sockaddr_ll sll; | |
1137 | struct msghdr msg; | |
1138 | struct iovec iov; | |
1139 | int ifindex; | |
488d734d BP |
1140 | int sock; |
1141 | ||
1142 | sock = af_packet_sock(); | |
1143 | if (sock < 0) { | |
c4c7a3d7 | 1144 | return -sock; |
488d734d | 1145 | } |
f23347ea | 1146 | |
86383816 BP |
1147 | ifindex = netdev_get_ifindex(netdev_); |
1148 | if (ifindex < 0) { | |
1149 | return -ifindex; | |
f23347ea | 1150 | } |
8b61709d | 1151 | |
f23347ea BP |
1152 | /* We don't bother setting most fields in sockaddr_ll because the |
1153 | * kernel ignores them for SOCK_RAW. */ | |
1154 | memset(&sll, 0, sizeof sll); | |
1155 | sll.sll_family = AF_PACKET; | |
1156 | sll.sll_ifindex = ifindex; | |
76c308b5 | 1157 | |
ebc56baa | 1158 | iov.iov_base = CONST_CAST(void *, data); |
f23347ea | 1159 | iov.iov_len = size; |
76c308b5 | 1160 | |
f23347ea BP |
1161 | msg.msg_name = &sll; |
1162 | msg.msg_namelen = sizeof sll; | |
1163 | msg.msg_iov = &iov; | |
1164 | msg.msg_iovlen = 1; | |
1165 | msg.msg_control = NULL; | |
1166 | msg.msg_controllen = 0; | |
1167 | msg.msg_flags = 0; | |
1168 | ||
488d734d | 1169 | retval = sendmsg(sock, &msg, 0); |
f23347ea | 1170 | } else { |
796223f5 BP |
1171 | /* Use the tap fd to send to this device. This is essential for |
1172 | * tap devices, because packets sent to a tap device with an | |
1173 | * AF_PACKET socket will loop back to be *received* again on the | |
32383c3b MM |
1174 | * tap device. This doesn't occur on other interface types |
1175 | * because we attach a socket filter to the rx socket. */ | |
b5d57fc8 | 1176 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
796223f5 | 1177 | |
d0d08f8a | 1178 | retval = write(netdev->tap_fd, data, size); |
f23347ea | 1179 | } |
76c308b5 | 1180 | |
8b61709d BP |
1181 | if (retval < 0) { |
1182 | /* The Linux AF_PACKET implementation never blocks waiting for room | |
1183 | * for packets, instead returning ENOBUFS. Translate this into | |
1184 | * EAGAIN for the caller. */ | |
f4fd623c DDP |
1185 | error = errno == ENOBUFS ? EAGAIN : errno; |
1186 | if (error == EINTR) { | |
1187 | /* continue without incrementing 'i', i.e. retry this packet */ | |
8b61709d | 1188 | continue; |
8b61709d | 1189 | } |
f4fd623c | 1190 | break; |
8b61709d | 1191 | } else if (retval != size) { |
f4fd623c DDP |
1192 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" bytes" |
1193 | " of %"PRIuSIZE") on %s", retval, size, | |
1194 | netdev_get_name(netdev_)); | |
1195 | error = EMSGSIZE; | |
1196 | break; | |
1197 | } | |
1198 | ||
1199 | /* Process the next packet in the batch */ | |
1200 | i++; | |
1201 | } | |
1202 | ||
1203 | if (may_steal) { | |
1204 | for (i = 0; i < cnt; i++) { | |
e14deea0 | 1205 | dp_packet_delete(pkts[i]); |
8b61709d BP |
1206 | } |
1207 | } | |
f4fd623c DDP |
1208 | |
1209 | if (error && error != EAGAIN) { | |
1210 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", | |
1211 | netdev_get_name(netdev_), ovs_strerror(error)); | |
1212 | } | |
1213 | ||
1214 | return error; | |
1215 | ||
8b61709d BP |
1216 | } |
1217 | ||
1218 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
1219 | * when the packet transmission queue has sufficient room to transmit a packet | |
1220 | * with netdev_send(). | |
1221 | * | |
1222 | * The kernel maintains a packet transmission queue, so the client is not | |
1223 | * expected to do additional queuing of packets. Thus, this function is | |
1224 | * unlikely to ever be used. It is included for completeness. */ | |
1225 | static void | |
f00fa8cb | 1226 | netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED) |
8b61709d | 1227 | { |
796223f5 | 1228 | if (is_tap_netdev(netdev)) { |
8b61709d BP |
1229 | /* TAP device always accepts packets.*/ |
1230 | poll_immediate_wake(); | |
1231 | } | |
1232 | } | |
1233 | ||
1234 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
1235 | * otherwise a positive errno value. */ | |
1236 | static int | |
74ff3298 | 1237 | netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac) |
8b61709d | 1238 | { |
b5d57fc8 | 1239 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4f9f3f21 | 1240 | enum netdev_flags old_flags = 0; |
eb395f2e BP |
1241 | int error; |
1242 | ||
86383816 BP |
1243 | ovs_mutex_lock(&netdev->mutex); |
1244 | ||
b5d57fc8 | 1245 | if (netdev->cache_valid & VALID_ETHERADDR) { |
86383816 BP |
1246 | error = netdev->ether_addr_error; |
1247 | if (error || eth_addr_equals(netdev->etheraddr, mac)) { | |
1248 | goto exit; | |
44445cac | 1249 | } |
b5d57fc8 | 1250 | netdev->cache_valid &= ~VALID_ETHERADDR; |
44445cac PS |
1251 | } |
1252 | ||
7eb1bd81 | 1253 | /* Tap devices must be brought down before setting the address. */ |
796223f5 | 1254 | if (is_tap_netdev(netdev_)) { |
4f9f3f21 | 1255 | update_flags(netdev, NETDEV_UP, 0, &old_flags); |
7eb1bd81 | 1256 | } |
44445cac PS |
1257 | error = set_etheraddr(netdev_get_name(netdev_), mac); |
1258 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
1259 | netdev->ether_addr_error = error; |
1260 | netdev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 1261 | if (!error) { |
74ff3298 | 1262 | netdev->etheraddr = mac; |
eb395f2e | 1263 | } |
8b61709d | 1264 | } |
44445cac | 1265 | |
4f9f3f21 BP |
1266 | if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { |
1267 | update_flags(netdev, 0, NETDEV_UP, &old_flags); | |
1268 | } | |
7eb1bd81 | 1269 | |
86383816 BP |
1270 | exit: |
1271 | ovs_mutex_unlock(&netdev->mutex); | |
8b61709d BP |
1272 | return error; |
1273 | } | |
1274 | ||
44445cac | 1275 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d | 1276 | static int |
74ff3298 | 1277 | netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac) |
8b61709d | 1278 | { |
b5d57fc8 | 1279 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1280 | int error; |
44445cac | 1281 | |
86383816 | 1282 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1283 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { |
86383816 | 1284 | netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), |
74ff3298 | 1285 | &netdev->etheraddr); |
b5d57fc8 | 1286 | netdev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 1287 | } |
44445cac | 1288 | |
86383816 BP |
1289 | error = netdev->ether_addr_error; |
1290 | if (!error) { | |
74ff3298 | 1291 | *mac = netdev->etheraddr; |
44445cac | 1292 | } |
86383816 | 1293 | ovs_mutex_unlock(&netdev->mutex); |
44445cac | 1294 | |
86383816 | 1295 | return error; |
8b61709d BP |
1296 | } |
1297 | ||
8b61709d | 1298 | static int |
73371c09 | 1299 | netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) |
8b61709d | 1300 | { |
86383816 BP |
1301 | int error; |
1302 | ||
b5d57fc8 | 1303 | if (!(netdev->cache_valid & VALID_MTU)) { |
8b61709d | 1304 | struct ifreq ifr; |
90a6637d | 1305 | |
86383816 | 1306 | netdev->netdev_mtu_error = af_inet_ifreq_ioctl( |
73371c09 | 1307 | netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); |
b5d57fc8 BP |
1308 | netdev->mtu = ifr.ifr_mtu; |
1309 | netdev->cache_valid |= VALID_MTU; | |
8b61709d | 1310 | } |
90a6637d | 1311 | |
86383816 BP |
1312 | error = netdev->netdev_mtu_error; |
1313 | if (!error) { | |
b5d57fc8 | 1314 | *mtup = netdev->mtu; |
90a6637d | 1315 | } |
73371c09 BP |
1316 | |
1317 | return error; | |
1318 | } | |
1319 | ||
1320 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1321 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1322 | * bytes for Ethernet devices. */ | |
1323 | static int | |
1324 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1325 | { | |
1326 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1327 | int error; | |
1328 | ||
1329 | ovs_mutex_lock(&netdev->mutex); | |
1330 | error = netdev_linux_get_mtu__(netdev, mtup); | |
86383816 BP |
1331 | ovs_mutex_unlock(&netdev->mutex); |
1332 | ||
1333 | return error; | |
8b61709d BP |
1334 | } |
1335 | ||
9b020780 PS |
1336 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1337 | * networking ioctl interface. | |
1338 | */ | |
1339 | static int | |
1340 | netdev_linux_set_mtu(const struct netdev *netdev_, int mtu) | |
1341 | { | |
b5d57fc8 | 1342 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9b020780 PS |
1343 | struct ifreq ifr; |
1344 | int error; | |
1345 | ||
86383816 | 1346 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1347 | if (netdev->cache_valid & VALID_MTU) { |
86383816 BP |
1348 | error = netdev->netdev_mtu_error; |
1349 | if (error || netdev->mtu == mtu) { | |
1350 | goto exit; | |
90a6637d | 1351 | } |
b5d57fc8 | 1352 | netdev->cache_valid &= ~VALID_MTU; |
153e5481 | 1353 | } |
9b020780 | 1354 | ifr.ifr_mtu = mtu; |
259e0b1a BP |
1355 | error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, |
1356 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d | 1357 | if (!error || error == ENODEV) { |
b5d57fc8 BP |
1358 | netdev->netdev_mtu_error = error; |
1359 | netdev->mtu = ifr.ifr_mtu; | |
1360 | netdev->cache_valid |= VALID_MTU; | |
9b020780 | 1361 | } |
86383816 BP |
1362 | exit: |
1363 | ovs_mutex_unlock(&netdev->mutex); | |
90a6637d | 1364 | return error; |
9b020780 PS |
1365 | } |
1366 | ||
9ab3d9a3 BP |
1367 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1368 | * On failure, returns a negative errno value. */ | |
1369 | static int | |
86383816 | 1370 | netdev_linux_get_ifindex(const struct netdev *netdev_) |
9ab3d9a3 | 1371 | { |
86383816 | 1372 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9ab3d9a3 BP |
1373 | int ifindex, error; |
1374 | ||
86383816 BP |
1375 | ovs_mutex_lock(&netdev->mutex); |
1376 | error = get_ifindex(netdev_, &ifindex); | |
1377 | ovs_mutex_unlock(&netdev->mutex); | |
1378 | ||
9ab3d9a3 BP |
1379 | return error ? -error : ifindex; |
1380 | } | |
1381 | ||
8b61709d BP |
1382 | static int |
1383 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1384 | { | |
b5d57fc8 | 1385 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1386 | |
86383816 | 1387 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
1388 | if (netdev->miimon_interval > 0) { |
1389 | *carrier = netdev->miimon; | |
3a183124 | 1390 | } else { |
b5d57fc8 | 1391 | *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1392 | } |
86383816 | 1393 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d | 1394 | |
3a183124 | 1395 | return 0; |
8b61709d BP |
1396 | } |
1397 | ||
65c3058c | 1398 | static long long int |
86383816 | 1399 | netdev_linux_get_carrier_resets(const struct netdev *netdev_) |
65c3058c | 1400 | { |
86383816 BP |
1401 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1402 | long long int carrier_resets; | |
1403 | ||
1404 | ovs_mutex_lock(&netdev->mutex); | |
1405 | carrier_resets = netdev->carrier_resets; | |
1406 | ovs_mutex_unlock(&netdev->mutex); | |
1407 | ||
1408 | return carrier_resets; | |
65c3058c EJ |
1409 | } |
1410 | ||
63331829 | 1411 | static int |
1670c579 EJ |
1412 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1413 | struct mii_ioctl_data *data) | |
63331829 | 1414 | { |
63331829 | 1415 | struct ifreq ifr; |
782e6111 | 1416 | int error; |
63331829 | 1417 | |
63331829 | 1418 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1419 | memcpy(&ifr.ifr_data, data, sizeof *data); |
259e0b1a | 1420 | error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1421 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1422 | |
782e6111 EJ |
1423 | return error; |
1424 | } | |
1425 | ||
1426 | static int | |
1670c579 | 1427 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1428 | { |
782e6111 EJ |
1429 | struct mii_ioctl_data data; |
1430 | int error; | |
63331829 | 1431 | |
782e6111 EJ |
1432 | *miimon = false; |
1433 | ||
1434 | memset(&data, 0, sizeof data); | |
1670c579 | 1435 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
1436 | if (!error) { |
1437 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1438 | data.reg_num = MII_BMSR; | |
1670c579 | 1439 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 1440 | &data); |
63331829 EJ |
1441 | |
1442 | if (!error) { | |
782e6111 | 1443 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 EJ |
1444 | } else { |
1445 | VLOG_WARN_RL(&rl, "%s: failed to query MII", name); | |
1446 | } | |
1447 | } else { | |
1448 | struct ethtool_cmd ecmd; | |
63331829 EJ |
1449 | |
1450 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1451 | name); | |
1452 | ||
ab985a77 | 1453 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
1454 | memset(&ecmd, 0, sizeof ecmd); |
1455 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1456 | "ETHTOOL_GLINK"); | |
1457 | if (!error) { | |
782e6111 EJ |
1458 | struct ethtool_value eval; |
1459 | ||
1460 | memcpy(&eval, &ecmd, sizeof eval); | |
1461 | *miimon = !!eval.data; | |
63331829 EJ |
1462 | } else { |
1463 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1464 | } | |
1465 | } | |
1466 | ||
1467 | return error; | |
1468 | } | |
1469 | ||
1670c579 EJ |
1470 | static int |
1471 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
1472 | long long int interval) | |
1473 | { | |
b5d57fc8 | 1474 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1670c579 | 1475 | |
86383816 | 1476 | ovs_mutex_lock(&netdev->mutex); |
1670c579 | 1477 | interval = interval > 0 ? MAX(interval, 100) : 0; |
b5d57fc8 | 1478 | if (netdev->miimon_interval != interval) { |
19c8e9c1 | 1479 | if (interval && !netdev->miimon_interval) { |
812c272c | 1480 | atomic_count_inc(&miimon_cnt); |
19c8e9c1 | 1481 | } else if (!interval && netdev->miimon_interval) { |
812c272c | 1482 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
1483 | } |
1484 | ||
b5d57fc8 BP |
1485 | netdev->miimon_interval = interval; |
1486 | timer_set_expired(&netdev->miimon_timer); | |
1670c579 | 1487 | } |
86383816 | 1488 | ovs_mutex_unlock(&netdev->mutex); |
1670c579 EJ |
1489 | |
1490 | return 0; | |
1491 | } | |
1492 | ||
1493 | static void | |
1494 | netdev_linux_miimon_run(void) | |
1495 | { | |
1496 | struct shash device_shash; | |
1497 | struct shash_node *node; | |
1498 | ||
1499 | shash_init(&device_shash); | |
b5d57fc8 | 1500 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1501 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1502 | struct netdev *netdev = node->data; |
1503 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
1504 | bool miimon; |
1505 | ||
86383816 BP |
1506 | ovs_mutex_lock(&dev->mutex); |
1507 | if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { | |
1508 | netdev_linux_get_miimon(dev->up.name, &miimon); | |
1509 | if (miimon != dev->miimon) { | |
1510 | dev->miimon = miimon; | |
1511 | netdev_linux_changed(dev, dev->ifi_flags, 0); | |
1512 | } | |
1670c579 | 1513 | |
86383816 | 1514 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); |
1670c579 | 1515 | } |
86383816 | 1516 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1517 | netdev_close(netdev); |
1670c579 EJ |
1518 | } |
1519 | ||
1520 | shash_destroy(&device_shash); | |
1521 | } | |
1522 | ||
1523 | static void | |
1524 | netdev_linux_miimon_wait(void) | |
1525 | { | |
1526 | struct shash device_shash; | |
1527 | struct shash_node *node; | |
1528 | ||
1529 | shash_init(&device_shash); | |
b5d57fc8 | 1530 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1531 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1532 | struct netdev *netdev = node->data; |
1533 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 | 1534 | |
86383816 | 1535 | ovs_mutex_lock(&dev->mutex); |
1670c579 EJ |
1536 | if (dev->miimon_interval > 0) { |
1537 | timer_wait(&dev->miimon_timer); | |
1538 | } | |
86383816 | 1539 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1540 | netdev_close(netdev); |
1670c579 EJ |
1541 | } |
1542 | shash_destroy(&device_shash); | |
1543 | } | |
1544 | ||
92df599c JG |
1545 | static void |
1546 | swap_uint64(uint64_t *a, uint64_t *b) | |
1547 | { | |
1de0e8ae BP |
1548 | uint64_t tmp = *a; |
1549 | *a = *b; | |
1550 | *b = tmp; | |
92df599c JG |
1551 | } |
1552 | ||
c060c4cf EJ |
1553 | /* Copies 'src' into 'dst', performing format conversion in the process. |
1554 | * | |
1555 | * 'src' is allowed to be misaligned. */ | |
1556 | static void | |
1557 | netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, | |
1558 | const struct ovs_vport_stats *src) | |
1559 | { | |
6a54dedc BP |
1560 | dst->rx_packets = get_32aligned_u64(&src->rx_packets); |
1561 | dst->tx_packets = get_32aligned_u64(&src->tx_packets); | |
1562 | dst->rx_bytes = get_32aligned_u64(&src->rx_bytes); | |
1563 | dst->tx_bytes = get_32aligned_u64(&src->tx_bytes); | |
1564 | dst->rx_errors = get_32aligned_u64(&src->rx_errors); | |
1565 | dst->tx_errors = get_32aligned_u64(&src->tx_errors); | |
1566 | dst->rx_dropped = get_32aligned_u64(&src->rx_dropped); | |
1567 | dst->tx_dropped = get_32aligned_u64(&src->tx_dropped); | |
c060c4cf EJ |
1568 | dst->multicast = 0; |
1569 | dst->collisions = 0; | |
1570 | dst->rx_length_errors = 0; | |
1571 | dst->rx_over_errors = 0; | |
1572 | dst->rx_crc_errors = 0; | |
1573 | dst->rx_frame_errors = 0; | |
1574 | dst->rx_fifo_errors = 0; | |
1575 | dst->rx_missed_errors = 0; | |
1576 | dst->tx_aborted_errors = 0; | |
1577 | dst->tx_carrier_errors = 0; | |
1578 | dst->tx_fifo_errors = 0; | |
1579 | dst->tx_heartbeat_errors = 0; | |
1580 | dst->tx_window_errors = 0; | |
1581 | } | |
1582 | ||
1583 | static int | |
1584 | get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) | |
1585 | { | |
93451a0a | 1586 | struct dpif_netlink_vport reply; |
c060c4cf EJ |
1587 | struct ofpbuf *buf; |
1588 | int error; | |
1589 | ||
93451a0a | 1590 | error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf); |
c060c4cf EJ |
1591 | if (error) { |
1592 | return error; | |
1593 | } else if (!reply.stats) { | |
1594 | ofpbuf_delete(buf); | |
1595 | return EOPNOTSUPP; | |
1596 | } | |
1597 | ||
1598 | netdev_stats_from_ovs_vport_stats(stats, reply.stats); | |
1599 | ||
1600 | ofpbuf_delete(buf); | |
1601 | ||
1602 | return 0; | |
1603 | } | |
1604 | ||
f613a0d7 PS |
1605 | static void |
1606 | get_stats_via_vport(const struct netdev *netdev_, | |
1607 | struct netdev_stats *stats) | |
8b61709d | 1608 | { |
b5d57fc8 | 1609 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1610 | |
b5d57fc8 BP |
1611 | if (!netdev->vport_stats_error || |
1612 | !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 1613 | int error; |
7fbef77a | 1614 | |
c060c4cf | 1615 | error = get_stats_via_vport__(netdev_, stats); |
bb13fe5e | 1616 | if (error && error != ENOENT && error != ENODEV) { |
a57a8488 | 1617 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
10a89ef0 BP |
1618 | "(%s)", |
1619 | netdev_get_name(netdev_), ovs_strerror(error)); | |
f613a0d7 | 1620 | } |
b5d57fc8 BP |
1621 | netdev->vport_stats_error = error; |
1622 | netdev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 1623 | } |
f613a0d7 | 1624 | } |
8b61709d | 1625 | |
f613a0d7 PS |
1626 | /* Retrieves current device stats for 'netdev-linux'. */ |
1627 | static int | |
1628 | netdev_linux_get_stats(const struct netdev *netdev_, | |
1629 | struct netdev_stats *stats) | |
1630 | { | |
b5d57fc8 | 1631 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1632 | struct netdev_stats dev_stats; |
1633 | int error; | |
1634 | ||
86383816 | 1635 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1636 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1637 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1638 | if (error) { |
86383816 BP |
1639 | if (!netdev->vport_stats_error) { |
1640 | error = 0; | |
f613a0d7 | 1641 | } |
86383816 | 1642 | } else if (netdev->vport_stats_error) { |
04c881eb | 1643 | /* stats not available from OVS then use netdev stats. */ |
f613a0d7 PS |
1644 | *stats = dev_stats; |
1645 | } else { | |
04c881eb AZ |
1646 | /* Use kernel netdev's packet and byte counts since vport's counters |
1647 | * do not reflect packet counts on the wire when GSO, TSO or GRO are | |
1648 | * enabled. */ | |
1649 | stats->rx_packets = dev_stats.rx_packets; | |
1650 | stats->rx_bytes = dev_stats.rx_bytes; | |
1651 | stats->tx_packets = dev_stats.tx_packets; | |
1652 | stats->tx_bytes = dev_stats.tx_bytes; | |
1653 | ||
f613a0d7 PS |
1654 | stats->rx_errors += dev_stats.rx_errors; |
1655 | stats->tx_errors += dev_stats.tx_errors; | |
1656 | stats->rx_dropped += dev_stats.rx_dropped; | |
1657 | stats->tx_dropped += dev_stats.tx_dropped; | |
1658 | stats->multicast += dev_stats.multicast; | |
1659 | stats->collisions += dev_stats.collisions; | |
1660 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
1661 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
1662 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
1663 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
1664 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
1665 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
1666 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
1667 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
1668 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
1669 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
1670 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
1671 | } | |
86383816 BP |
1672 | ovs_mutex_unlock(&netdev->mutex); |
1673 | ||
1674 | return error; | |
f613a0d7 PS |
1675 | } |
1676 | ||
1677 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
1678 | * netdev-internal. */ | |
1679 | static int | |
15aee116 | 1680 | netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) |
f613a0d7 | 1681 | { |
b5d57fc8 | 1682 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1683 | struct netdev_stats dev_stats; |
1684 | int error; | |
1685 | ||
86383816 | 1686 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1687 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1688 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1689 | if (error) { |
86383816 BP |
1690 | if (!netdev->vport_stats_error) { |
1691 | error = 0; | |
8b61709d | 1692 | } |
86383816 BP |
1693 | } else if (netdev->vport_stats_error) { |
1694 | /* Transmit and receive stats will appear to be swapped relative to the | |
1695 | * other ports since we are the one sending the data, not a remote | |
1696 | * computer. For consistency, we swap them back here. This does not | |
1697 | * apply if we are getting stats from the vport layer because it always | |
1698 | * tracks stats from the perspective of the switch. */ | |
fe6b0e03 | 1699 | |
f613a0d7 | 1700 | *stats = dev_stats; |
92df599c JG |
1701 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1702 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1703 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1704 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1705 | stats->rx_length_errors = 0; |
1706 | stats->rx_over_errors = 0; | |
1707 | stats->rx_crc_errors = 0; | |
1708 | stats->rx_frame_errors = 0; | |
1709 | stats->rx_fifo_errors = 0; | |
1710 | stats->rx_missed_errors = 0; | |
1711 | stats->tx_aborted_errors = 0; | |
1712 | stats->tx_carrier_errors = 0; | |
1713 | stats->tx_fifo_errors = 0; | |
1714 | stats->tx_heartbeat_errors = 0; | |
1715 | stats->tx_window_errors = 0; | |
f613a0d7 | 1716 | } else { |
04c881eb AZ |
1717 | /* Use kernel netdev's packet and byte counts since vport counters |
1718 | * do not reflect packet counts on the wire when GSO, TSO or GRO | |
1719 | * are enabled. */ | |
1720 | stats->rx_packets = dev_stats.tx_packets; | |
1721 | stats->rx_bytes = dev_stats.tx_bytes; | |
1722 | stats->tx_packets = dev_stats.rx_packets; | |
1723 | stats->tx_bytes = dev_stats.rx_bytes; | |
1724 | ||
f613a0d7 PS |
1725 | stats->rx_dropped += dev_stats.tx_dropped; |
1726 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 1727 | |
f613a0d7 PS |
1728 | stats->rx_errors += dev_stats.tx_errors; |
1729 | stats->tx_errors += dev_stats.rx_errors; | |
1730 | ||
1731 | stats->multicast += dev_stats.multicast; | |
1732 | stats->collisions += dev_stats.collisions; | |
1733 | } | |
86383816 BP |
1734 | ovs_mutex_unlock(&netdev->mutex); |
1735 | ||
1736 | return error; | |
8b61709d BP |
1737 | } |
1738 | ||
bba1e6f3 PS |
1739 | static int |
1740 | netdev_internal_get_stats(const struct netdev *netdev_, | |
1741 | struct netdev_stats *stats) | |
1742 | { | |
b5d57fc8 | 1743 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1744 | int error; |
bba1e6f3 | 1745 | |
86383816 | 1746 | ovs_mutex_lock(&netdev->mutex); |
bba1e6f3 | 1747 | get_stats_via_vport(netdev_, stats); |
86383816 BP |
1748 | error = netdev->vport_stats_error; |
1749 | ovs_mutex_unlock(&netdev->mutex); | |
1750 | ||
1751 | return error; | |
bba1e6f3 PS |
1752 | } |
1753 | ||
51f87458 | 1754 | static void |
b5d57fc8 | 1755 | netdev_linux_read_features(struct netdev_linux *netdev) |
8b61709d BP |
1756 | { |
1757 | struct ethtool_cmd ecmd; | |
6c038611 | 1758 | uint32_t speed; |
8b61709d BP |
1759 | int error; |
1760 | ||
b5d57fc8 | 1761 | if (netdev->cache_valid & VALID_FEATURES) { |
51f87458 PS |
1762 | return; |
1763 | } | |
1764 | ||
ab985a77 | 1765 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1766 | memset(&ecmd, 0, sizeof ecmd); |
b5d57fc8 | 1767 | error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, |
8b61709d BP |
1768 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1769 | if (error) { | |
51f87458 | 1770 | goto out; |
8b61709d BP |
1771 | } |
1772 | ||
1773 | /* Supported features. */ | |
b5d57fc8 | 1774 | netdev->supported = 0; |
8b61709d | 1775 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
b5d57fc8 | 1776 | netdev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
1777 | } |
1778 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
b5d57fc8 | 1779 | netdev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
1780 | } |
1781 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
b5d57fc8 | 1782 | netdev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
1783 | } |
1784 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
b5d57fc8 | 1785 | netdev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
1786 | } |
1787 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
b5d57fc8 | 1788 | netdev->supported |= NETDEV_F_1GB_HD; |
8b61709d BP |
1789 | } |
1790 | if (ecmd.supported & SUPPORTED_1000baseT_Full) { | |
b5d57fc8 | 1791 | netdev->supported |= NETDEV_F_1GB_FD; |
8b61709d BP |
1792 | } |
1793 | if (ecmd.supported & SUPPORTED_10000baseT_Full) { | |
b5d57fc8 | 1794 | netdev->supported |= NETDEV_F_10GB_FD; |
8b61709d BP |
1795 | } |
1796 | if (ecmd.supported & SUPPORTED_TP) { | |
b5d57fc8 | 1797 | netdev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
1798 | } |
1799 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
b5d57fc8 | 1800 | netdev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
1801 | } |
1802 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
b5d57fc8 | 1803 | netdev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
1804 | } |
1805 | if (ecmd.supported & SUPPORTED_Pause) { | |
b5d57fc8 | 1806 | netdev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
1807 | } |
1808 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
b5d57fc8 | 1809 | netdev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1810 | } |
1811 | ||
1812 | /* Advertised features. */ | |
b5d57fc8 | 1813 | netdev->advertised = 0; |
8b61709d | 1814 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
b5d57fc8 | 1815 | netdev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
1816 | } |
1817 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
b5d57fc8 | 1818 | netdev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
1819 | } |
1820 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
b5d57fc8 | 1821 | netdev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
1822 | } |
1823 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
b5d57fc8 | 1824 | netdev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
1825 | } |
1826 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
b5d57fc8 | 1827 | netdev->advertised |= NETDEV_F_1GB_HD; |
8b61709d BP |
1828 | } |
1829 | if (ecmd.advertising & ADVERTISED_1000baseT_Full) { | |
b5d57fc8 | 1830 | netdev->advertised |= NETDEV_F_1GB_FD; |
8b61709d BP |
1831 | } |
1832 | if (ecmd.advertising & ADVERTISED_10000baseT_Full) { | |
b5d57fc8 | 1833 | netdev->advertised |= NETDEV_F_10GB_FD; |
8b61709d BP |
1834 | } |
1835 | if (ecmd.advertising & ADVERTISED_TP) { | |
b5d57fc8 | 1836 | netdev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
1837 | } |
1838 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
b5d57fc8 | 1839 | netdev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
1840 | } |
1841 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
b5d57fc8 | 1842 | netdev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
1843 | } |
1844 | if (ecmd.advertising & ADVERTISED_Pause) { | |
b5d57fc8 | 1845 | netdev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
1846 | } |
1847 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
b5d57fc8 | 1848 | netdev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
1849 | } |
1850 | ||
1851 | /* Current settings. */ | |
2a529ead | 1852 | speed = ecmd.speed; |
6c038611 | 1853 | if (speed == SPEED_10) { |
b5d57fc8 | 1854 | netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 1855 | } else if (speed == SPEED_100) { |
b5d57fc8 | 1856 | netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 1857 | } else if (speed == SPEED_1000) { |
b5d57fc8 | 1858 | netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 1859 | } else if (speed == SPEED_10000) { |
b5d57fc8 | 1860 | netdev->current = NETDEV_F_10GB_FD; |
6c038611 | 1861 | } else if (speed == 40000) { |
b5d57fc8 | 1862 | netdev->current = NETDEV_F_40GB_FD; |
6c038611 | 1863 | } else if (speed == 100000) { |
b5d57fc8 | 1864 | netdev->current = NETDEV_F_100GB_FD; |
6c038611 | 1865 | } else if (speed == 1000000) { |
b5d57fc8 | 1866 | netdev->current = NETDEV_F_1TB_FD; |
8b61709d | 1867 | } else { |
b5d57fc8 | 1868 | netdev->current = 0; |
8b61709d BP |
1869 | } |
1870 | ||
1871 | if (ecmd.port == PORT_TP) { | |
b5d57fc8 | 1872 | netdev->current |= NETDEV_F_COPPER; |
8b61709d | 1873 | } else if (ecmd.port == PORT_FIBRE) { |
b5d57fc8 | 1874 | netdev->current |= NETDEV_F_FIBER; |
8b61709d BP |
1875 | } |
1876 | ||
1877 | if (ecmd.autoneg) { | |
b5d57fc8 | 1878 | netdev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
1879 | } |
1880 | ||
51f87458 | 1881 | out: |
b5d57fc8 BP |
1882 | netdev->cache_valid |= VALID_FEATURES; |
1883 | netdev->get_features_error = error; | |
51f87458 PS |
1884 | } |
1885 | ||
887ed8b2 BP |
1886 | /* Stores the features supported by 'netdev' into of '*current', '*advertised', |
1887 | * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. | |
1888 | * Returns 0 if successful, otherwise a positive errno value. */ | |
51f87458 PS |
1889 | static int |
1890 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
1891 | enum netdev_features *current, |
1892 | enum netdev_features *advertised, | |
1893 | enum netdev_features *supported, | |
1894 | enum netdev_features *peer) | |
51f87458 | 1895 | { |
b5d57fc8 | 1896 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1897 | int error; |
51f87458 | 1898 | |
86383816 | 1899 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1900 | netdev_linux_read_features(netdev); |
b5d57fc8 BP |
1901 | if (!netdev->get_features_error) { |
1902 | *current = netdev->current; | |
1903 | *advertised = netdev->advertised; | |
1904 | *supported = netdev->supported; | |
887ed8b2 | 1905 | *peer = 0; /* XXX */ |
51f87458 | 1906 | } |
86383816 BP |
1907 | error = netdev->get_features_error; |
1908 | ovs_mutex_unlock(&netdev->mutex); | |
1909 | ||
1910 | return error; | |
8b61709d BP |
1911 | } |
1912 | ||
1913 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
1914 | static int | |
86383816 | 1915 | netdev_linux_set_advertisements(struct netdev *netdev_, |
6c038611 | 1916 | enum netdev_features advertise) |
8b61709d | 1917 | { |
86383816 | 1918 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
1919 | struct ethtool_cmd ecmd; |
1920 | int error; | |
1921 | ||
86383816 BP |
1922 | ovs_mutex_lock(&netdev->mutex); |
1923 | ||
ab985a77 | 1924 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1925 | memset(&ecmd, 0, sizeof ecmd); |
86383816 | 1926 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
8b61709d BP |
1927 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1928 | if (error) { | |
86383816 | 1929 | goto exit; |
8b61709d BP |
1930 | } |
1931 | ||
1932 | ecmd.advertising = 0; | |
6c038611 | 1933 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
1934 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
1935 | } | |
6c038611 | 1936 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
1937 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
1938 | } | |
6c038611 | 1939 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
1940 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
1941 | } | |
6c038611 | 1942 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
1943 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
1944 | } | |
6c038611 | 1945 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
1946 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
1947 | } | |
6c038611 | 1948 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
1949 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
1950 | } | |
6c038611 | 1951 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
1952 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
1953 | } | |
6c038611 | 1954 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
1955 | ecmd.advertising |= ADVERTISED_TP; |
1956 | } | |
6c038611 | 1957 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
1958 | ecmd.advertising |= ADVERTISED_FIBRE; |
1959 | } | |
6c038611 | 1960 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
1961 | ecmd.advertising |= ADVERTISED_Autoneg; |
1962 | } | |
6c038611 | 1963 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
1964 | ecmd.advertising |= ADVERTISED_Pause; |
1965 | } | |
6c038611 | 1966 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
1967 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
1968 | } | |
ab985a77 | 1969 | COVERAGE_INC(netdev_set_ethtool); |
86383816 BP |
1970 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
1971 | ETHTOOL_SSET, "ETHTOOL_SSET"); | |
1972 | ||
1973 | exit: | |
1974 | ovs_mutex_unlock(&netdev->mutex); | |
1975 | return error; | |
8b61709d BP |
1976 | } |
1977 | ||
f8500004 JP |
1978 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
1979 | * successful, otherwise a positive errno value. */ | |
8b61709d | 1980 | static int |
b5d57fc8 | 1981 | netdev_linux_set_policing(struct netdev *netdev_, |
8b61709d BP |
1982 | uint32_t kbits_rate, uint32_t kbits_burst) |
1983 | { | |
b5d57fc8 BP |
1984 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1985 | const char *netdev_name = netdev_get_name(netdev_); | |
f8500004 | 1986 | int error; |
8b61709d | 1987 | |
80a86fbe BP |
1988 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
1989 | : !kbits_burst ? 1000 /* Default to 1000 kbits if 0. */ | |
1990 | : kbits_burst); /* Stick with user-specified value. */ | |
1991 | ||
86383816 | 1992 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1993 | if (netdev->cache_valid & VALID_POLICING) { |
86383816 BP |
1994 | error = netdev->netdev_policing_error; |
1995 | if (error || (netdev->kbits_rate == kbits_rate && | |
1996 | netdev->kbits_burst == kbits_burst)) { | |
c9f71668 | 1997 | /* Assume that settings haven't changed since we last set them. */ |
86383816 | 1998 | goto out; |
c9f71668 | 1999 | } |
b5d57fc8 | 2000 | netdev->cache_valid &= ~VALID_POLICING; |
80a86fbe BP |
2001 | } |
2002 | ||
ac8c3412 | 2003 | COVERAGE_INC(netdev_set_policing); |
f8500004 | 2004 | /* Remove any existing ingress qdisc. */ |
b5d57fc8 | 2005 | error = tc_add_del_ingress_qdisc(netdev_, false); |
f8500004 JP |
2006 | if (error) { |
2007 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
10a89ef0 | 2008 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2009 | goto out; |
f8500004 JP |
2010 | } |
2011 | ||
8b61709d | 2012 | if (kbits_rate) { |
b5d57fc8 | 2013 | error = tc_add_del_ingress_qdisc(netdev_, true); |
f8500004 JP |
2014 | if (error) { |
2015 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
10a89ef0 | 2016 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2017 | goto out; |
8b61709d BP |
2018 | } |
2019 | ||
b5d57fc8 | 2020 | error = tc_add_policer(netdev_, kbits_rate, kbits_burst); |
f8500004 JP |
2021 | if (error){ |
2022 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
10a89ef0 | 2023 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2024 | goto out; |
8b61709d | 2025 | } |
8b61709d BP |
2026 | } |
2027 | ||
b5d57fc8 BP |
2028 | netdev->kbits_rate = kbits_rate; |
2029 | netdev->kbits_burst = kbits_burst; | |
f8500004 | 2030 | |
c9f71668 PS |
2031 | out: |
2032 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
2033 | netdev->netdev_policing_error = error; |
2034 | netdev->cache_valid |= VALID_POLICING; | |
c9f71668 | 2035 | } |
86383816 | 2036 | ovs_mutex_unlock(&netdev->mutex); |
c9f71668 | 2037 | return error; |
8b61709d BP |
2038 | } |
2039 | ||
c1c9c9c4 BP |
2040 | static int |
2041 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 2042 | struct sset *types) |
c1c9c9c4 | 2043 | { |
559eb230 | 2044 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2045 | |
2046 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2047 | const struct tc_ops *ops = *opsp; | |
2048 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 2049 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
2050 | } |
2051 | } | |
2052 | return 0; | |
2053 | } | |
2054 | ||
2055 | static const struct tc_ops * | |
2056 | tc_lookup_ovs_name(const char *name) | |
2057 | { | |
559eb230 | 2058 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2059 | |
2060 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2061 | const struct tc_ops *ops = *opsp; | |
2062 | if (!strcmp(name, ops->ovs_name)) { | |
2063 | return ops; | |
2064 | } | |
2065 | } | |
2066 | return NULL; | |
2067 | } | |
2068 | ||
2069 | static const struct tc_ops * | |
2070 | tc_lookup_linux_name(const char *name) | |
2071 | { | |
559eb230 | 2072 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2073 | |
2074 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2075 | const struct tc_ops *ops = *opsp; | |
2076 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
2077 | return ops; | |
2078 | } | |
2079 | } | |
2080 | return NULL; | |
2081 | } | |
2082 | ||
93b13be8 | 2083 | static struct tc_queue * |
b5d57fc8 | 2084 | tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, |
93b13be8 BP |
2085 | size_t hash) |
2086 | { | |
b5d57fc8 | 2087 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
93b13be8 BP |
2088 | struct tc_queue *queue; |
2089 | ||
b5d57fc8 | 2090 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { |
93b13be8 BP |
2091 | if (queue->queue_id == queue_id) { |
2092 | return queue; | |
2093 | } | |
2094 | } | |
2095 | return NULL; | |
2096 | } | |
2097 | ||
2098 | static struct tc_queue * | |
2099 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
2100 | { | |
2101 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
2102 | } | |
2103 | ||
c1c9c9c4 BP |
2104 | static int |
2105 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
2106 | const char *type, | |
2107 | struct netdev_qos_capabilities *caps) | |
2108 | { | |
2109 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
2110 | if (!ops) { | |
2111 | return EOPNOTSUPP; | |
2112 | } | |
2113 | caps->n_queues = ops->n_queues; | |
2114 | return 0; | |
2115 | } | |
2116 | ||
2117 | static int | |
b5d57fc8 | 2118 | netdev_linux_get_qos(const struct netdev *netdev_, |
79f1cbe9 | 2119 | const char **typep, struct smap *details) |
c1c9c9c4 | 2120 | { |
b5d57fc8 | 2121 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2122 | int error; |
2123 | ||
86383816 | 2124 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2125 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2126 | if (!error) { |
2127 | *typep = netdev->tc->ops->ovs_name; | |
2128 | error = (netdev->tc->ops->qdisc_get | |
2129 | ? netdev->tc->ops->qdisc_get(netdev_, details) | |
2130 | : 0); | |
c1c9c9c4 | 2131 | } |
86383816 | 2132 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2133 | |
86383816 | 2134 | return error; |
c1c9c9c4 BP |
2135 | } |
2136 | ||
2137 | static int | |
b5d57fc8 | 2138 | netdev_linux_set_qos(struct netdev *netdev_, |
79f1cbe9 | 2139 | const char *type, const struct smap *details) |
c1c9c9c4 | 2140 | { |
b5d57fc8 | 2141 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2142 | const struct tc_ops *new_ops; |
2143 | int error; | |
2144 | ||
2145 | new_ops = tc_lookup_ovs_name(type); | |
2146 | if (!new_ops || !new_ops->tc_install) { | |
2147 | return EOPNOTSUPP; | |
2148 | } | |
2149 | ||
86383816 | 2150 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2151 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 | 2152 | if (error) { |
86383816 | 2153 | goto exit; |
c1c9c9c4 BP |
2154 | } |
2155 | ||
b5d57fc8 | 2156 | if (new_ops == netdev->tc->ops) { |
86383816 | 2157 | error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; |
c1c9c9c4 BP |
2158 | } else { |
2159 | /* Delete existing qdisc. */ | |
b5d57fc8 | 2160 | error = tc_del_qdisc(netdev_); |
c1c9c9c4 | 2161 | if (error) { |
86383816 | 2162 | goto exit; |
c1c9c9c4 | 2163 | } |
b5d57fc8 | 2164 | ovs_assert(netdev->tc == NULL); |
c1c9c9c4 BP |
2165 | |
2166 | /* Install new qdisc. */ | |
b5d57fc8 BP |
2167 | error = new_ops->tc_install(netdev_, details); |
2168 | ovs_assert((error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 | 2169 | } |
86383816 BP |
2170 | |
2171 | exit: | |
2172 | ovs_mutex_unlock(&netdev->mutex); | |
2173 | return error; | |
c1c9c9c4 BP |
2174 | } |
2175 | ||
2176 | static int | |
b5d57fc8 | 2177 | netdev_linux_get_queue(const struct netdev *netdev_, |
79f1cbe9 | 2178 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 | 2179 | { |
b5d57fc8 | 2180 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2181 | int error; |
2182 | ||
86383816 | 2183 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2184 | error = tc_query_qdisc(netdev_); |
86383816 | 2185 | if (!error) { |
b5d57fc8 | 2186 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
86383816 | 2187 | error = (queue |
b5d57fc8 | 2188 | ? netdev->tc->ops->class_get(netdev_, queue, details) |
93b13be8 | 2189 | : ENOENT); |
c1c9c9c4 | 2190 | } |
86383816 BP |
2191 | ovs_mutex_unlock(&netdev->mutex); |
2192 | ||
2193 | return error; | |
c1c9c9c4 BP |
2194 | } |
2195 | ||
2196 | static int | |
b5d57fc8 | 2197 | netdev_linux_set_queue(struct netdev *netdev_, |
79f1cbe9 | 2198 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 | 2199 | { |
b5d57fc8 | 2200 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2201 | int error; |
2202 | ||
86383816 | 2203 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2204 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2205 | if (!error) { |
2206 | error = (queue_id < netdev->tc->ops->n_queues | |
2207 | && netdev->tc->ops->class_set | |
2208 | ? netdev->tc->ops->class_set(netdev_, queue_id, details) | |
2209 | : EINVAL); | |
c1c9c9c4 | 2210 | } |
86383816 | 2211 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2212 | |
86383816 | 2213 | return error; |
c1c9c9c4 BP |
2214 | } |
2215 | ||
2216 | static int | |
b5d57fc8 | 2217 | netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) |
c1c9c9c4 | 2218 | { |
b5d57fc8 | 2219 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2220 | int error; |
2221 | ||
86383816 | 2222 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2223 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2224 | if (!error) { |
2225 | if (netdev->tc->ops->class_delete) { | |
2226 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2227 | error = (queue | |
2228 | ? netdev->tc->ops->class_delete(netdev_, queue) | |
2229 | : ENOENT); | |
2230 | } else { | |
2231 | error = EINVAL; | |
2232 | } | |
c1c9c9c4 | 2233 | } |
86383816 BP |
2234 | ovs_mutex_unlock(&netdev->mutex); |
2235 | ||
2236 | return error; | |
c1c9c9c4 BP |
2237 | } |
2238 | ||
2239 | static int | |
b5d57fc8 | 2240 | netdev_linux_get_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2241 | unsigned int queue_id, |
2242 | struct netdev_queue_stats *stats) | |
2243 | { | |
b5d57fc8 | 2244 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2245 | int error; |
2246 | ||
86383816 | 2247 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2248 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2249 | if (!error) { |
2250 | if (netdev->tc->ops->class_get_stats) { | |
2251 | const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2252 | if (queue) { | |
2253 | stats->created = queue->created; | |
2254 | error = netdev->tc->ops->class_get_stats(netdev_, queue, | |
2255 | stats); | |
2256 | } else { | |
2257 | error = ENOENT; | |
2258 | } | |
2259 | } else { | |
2260 | error = EOPNOTSUPP; | |
6dc34a0d | 2261 | } |
c1c9c9c4 | 2262 | } |
86383816 BP |
2263 | ovs_mutex_unlock(&netdev->mutex); |
2264 | ||
2265 | return error; | |
c1c9c9c4 BP |
2266 | } |
2267 | ||
d57695d7 JS |
2268 | struct queue_dump_state { |
2269 | struct nl_dump dump; | |
2270 | struct ofpbuf buf; | |
2271 | }; | |
2272 | ||
23a98ffe | 2273 | static bool |
d57695d7 | 2274 | start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) |
c1c9c9c4 BP |
2275 | { |
2276 | struct ofpbuf request; | |
2277 | struct tcmsg *tcmsg; | |
2278 | ||
2279 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, 0, &request); | |
23a98ffe BP |
2280 | if (!tcmsg) { |
2281 | return false; | |
2282 | } | |
3c4de644 | 2283 | tcmsg->tcm_parent = 0; |
d57695d7 | 2284 | nl_dump_start(&state->dump, NETLINK_ROUTE, &request); |
c1c9c9c4 | 2285 | ofpbuf_uninit(&request); |
d57695d7 JS |
2286 | |
2287 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); | |
23a98ffe | 2288 | return true; |
c1c9c9c4 BP |
2289 | } |
2290 | ||
d57695d7 JS |
2291 | static int |
2292 | finish_queue_dump(struct queue_dump_state *state) | |
2293 | { | |
2294 | ofpbuf_uninit(&state->buf); | |
2295 | return nl_dump_done(&state->dump); | |
2296 | } | |
2297 | ||
89454bf4 BP |
2298 | struct netdev_linux_queue_state { |
2299 | unsigned int *queues; | |
2300 | size_t cur_queue; | |
2301 | size_t n_queues; | |
2302 | }; | |
2303 | ||
c1c9c9c4 | 2304 | static int |
89454bf4 | 2305 | netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) |
c1c9c9c4 | 2306 | { |
89454bf4 | 2307 | const struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2308 | int error; |
2309 | ||
86383816 | 2310 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2311 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2312 | if (!error) { |
2313 | if (netdev->tc->ops->class_get) { | |
89454bf4 BP |
2314 | struct netdev_linux_queue_state *state; |
2315 | struct tc_queue *queue; | |
2316 | size_t i; | |
2317 | ||
2318 | *statep = state = xmalloc(sizeof *state); | |
2319 | state->n_queues = hmap_count(&netdev->tc->queues); | |
2320 | state->cur_queue = 0; | |
2321 | state->queues = xmalloc(state->n_queues * sizeof *state->queues); | |
2322 | ||
2323 | i = 0; | |
2324 | HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { | |
2325 | state->queues[i++] = queue->queue_id; | |
86383816 | 2326 | } |
c1c9c9c4 | 2327 | } else { |
86383816 | 2328 | error = EOPNOTSUPP; |
c1c9c9c4 BP |
2329 | } |
2330 | } | |
86383816 | 2331 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2332 | |
86383816 | 2333 | return error; |
c1c9c9c4 BP |
2334 | } |
2335 | ||
89454bf4 BP |
2336 | static int |
2337 | netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, | |
2338 | unsigned int *queue_idp, struct smap *details) | |
2339 | { | |
2340 | const struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
2341 | struct netdev_linux_queue_state *state = state_; | |
2342 | int error = EOF; | |
2343 | ||
2344 | ovs_mutex_lock(&netdev->mutex); | |
2345 | while (state->cur_queue < state->n_queues) { | |
2346 | unsigned int queue_id = state->queues[state->cur_queue++]; | |
2347 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2348 | ||
2349 | if (queue) { | |
2350 | *queue_idp = queue_id; | |
2351 | error = netdev->tc->ops->class_get(netdev_, queue, details); | |
2352 | break; | |
2353 | } | |
2354 | } | |
2355 | ovs_mutex_unlock(&netdev->mutex); | |
2356 | ||
2357 | return error; | |
2358 | } | |
2359 | ||
2360 | static int | |
2361 | netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, | |
2362 | void *state_) | |
2363 | { | |
2364 | struct netdev_linux_queue_state *state = state_; | |
2365 | ||
2366 | free(state->queues); | |
2367 | free(state); | |
2368 | return 0; | |
2369 | } | |
2370 | ||
c1c9c9c4 | 2371 | static int |
b5d57fc8 | 2372 | netdev_linux_dump_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2373 | netdev_dump_queue_stats_cb *cb, void *aux) |
2374 | { | |
b5d57fc8 | 2375 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2376 | int error; |
2377 | ||
86383816 | 2378 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2379 | error = tc_query_qdisc(netdev_); |
86383816 | 2380 | if (!error) { |
d57695d7 | 2381 | struct queue_dump_state state; |
c1c9c9c4 | 2382 | |
86383816 BP |
2383 | if (!netdev->tc->ops->class_dump_stats) { |
2384 | error = EOPNOTSUPP; | |
d57695d7 | 2385 | } else if (!start_queue_dump(netdev_, &state)) { |
86383816 BP |
2386 | error = ENODEV; |
2387 | } else { | |
2388 | struct ofpbuf msg; | |
2389 | int retval; | |
2390 | ||
d57695d7 | 2391 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
86383816 BP |
2392 | retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, |
2393 | cb, aux); | |
2394 | if (retval) { | |
2395 | error = retval; | |
2396 | } | |
2397 | } | |
2398 | ||
d57695d7 | 2399 | retval = finish_queue_dump(&state); |
86383816 BP |
2400 | if (retval) { |
2401 | error = retval; | |
2402 | } | |
c1c9c9c4 BP |
2403 | } |
2404 | } | |
86383816 | 2405 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2406 | |
86383816 | 2407 | return error; |
c1c9c9c4 BP |
2408 | } |
2409 | ||
8b61709d | 2410 | static int |
f1acd62b BP |
2411 | netdev_linux_get_in4(const struct netdev *netdev_, |
2412 | struct in_addr *address, struct in_addr *netmask) | |
8b61709d | 2413 | { |
b5d57fc8 | 2414 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 2415 | int error; |
149f577a | 2416 | |
86383816 | 2417 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2418 | if (!(netdev->cache_valid & VALID_IN4)) { |
b5d57fc8 | 2419 | error = netdev_linux_get_ipv4(netdev_, &netdev->address, |
8b61709d | 2420 | SIOCGIFADDR, "SIOCGIFADDR"); |
86383816 BP |
2421 | if (!error) { |
2422 | error = netdev_linux_get_ipv4(netdev_, &netdev->netmask, | |
2423 | SIOCGIFNETMASK, "SIOCGIFNETMASK"); | |
8b61709d | 2424 | } |
49af9a3d AW |
2425 | netdev->in4_error = error; |
2426 | netdev->cache_valid |= VALID_IN4; | |
86383816 | 2427 | } else { |
49af9a3d | 2428 | error = netdev->in4_error; |
86383816 | 2429 | } |
8b61709d | 2430 | |
86383816 BP |
2431 | if (!error) { |
2432 | if (netdev->address.s_addr != INADDR_ANY) { | |
2433 | *address = netdev->address; | |
2434 | *netmask = netdev->netmask; | |
2435 | } else { | |
2436 | error = EADDRNOTAVAIL; | |
f1acd62b | 2437 | } |
8b61709d | 2438 | } |
86383816 BP |
2439 | ovs_mutex_unlock(&netdev->mutex); |
2440 | ||
2441 | return error; | |
8b61709d BP |
2442 | } |
2443 | ||
8b61709d | 2444 | static int |
f1acd62b BP |
2445 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
2446 | struct in_addr netmask) | |
8b61709d | 2447 | { |
b5d57fc8 | 2448 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2449 | int error; |
2450 | ||
86383816 | 2451 | ovs_mutex_lock(&netdev->mutex); |
f1acd62b | 2452 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 2453 | if (!error) { |
b5d57fc8 BP |
2454 | netdev->address = address; |
2455 | netdev->netmask = netmask; | |
f1acd62b | 2456 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 2457 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 2458 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
2459 | } |
2460 | } | |
49af9a3d AW |
2461 | |
2462 | if (!error) { | |
2463 | netdev->cache_valid |= VALID_IN4; | |
2464 | netdev->in4_error = 0; | |
2465 | } else { | |
2466 | netdev->cache_valid &= ~VALID_IN4; | |
2467 | } | |
86383816 BP |
2468 | ovs_mutex_unlock(&netdev->mutex); |
2469 | ||
8b61709d BP |
2470 | return error; |
2471 | } | |
2472 | ||
2473 | static bool | |
2474 | parse_if_inet6_line(const char *line, | |
2475 | struct in6_addr *in6, char ifname[16 + 1]) | |
2476 | { | |
2477 | uint8_t *s6 = in6->s6_addr; | |
2478 | #define X8 "%2"SCNx8 | |
c2c28dfd BP |
2479 | return ovs_scan(line, |
2480 | " "X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 X8 | |
2481 | "%*x %*x %*x %*x %16s\n", | |
2482 | &s6[0], &s6[1], &s6[2], &s6[3], | |
2483 | &s6[4], &s6[5], &s6[6], &s6[7], | |
2484 | &s6[8], &s6[9], &s6[10], &s6[11], | |
2485 | &s6[12], &s6[13], &s6[14], &s6[15], | |
2486 | ifname); | |
8b61709d BP |
2487 | } |
2488 | ||
7df6932e AW |
2489 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address. |
2490 | * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding | |
2491 | * error. */ | |
8b61709d BP |
2492 | static int |
2493 | netdev_linux_get_in6(const struct netdev *netdev_, struct in6_addr *in6) | |
2494 | { | |
b5d57fc8 | 2495 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7df6932e | 2496 | int error; |
86383816 BP |
2497 | |
2498 | ovs_mutex_lock(&netdev->mutex); | |
b5d57fc8 | 2499 | if (!(netdev->cache_valid & VALID_IN6)) { |
8b61709d BP |
2500 | FILE *file; |
2501 | char line[128]; | |
2502 | ||
b5d57fc8 | 2503 | netdev->in6 = in6addr_any; |
7df6932e | 2504 | netdev->in6_error = EADDRNOTAVAIL; |
8b61709d BP |
2505 | |
2506 | file = fopen("/proc/net/if_inet6", "r"); | |
2507 | if (file != NULL) { | |
2508 | const char *name = netdev_get_name(netdev_); | |
2509 | while (fgets(line, sizeof line, file)) { | |
2a022368 | 2510 | struct in6_addr in6_tmp; |
8b61709d | 2511 | char ifname[16 + 1]; |
2a022368 | 2512 | if (parse_if_inet6_line(line, &in6_tmp, ifname) |
8b61709d BP |
2513 | && !strcmp(name, ifname)) |
2514 | { | |
b5d57fc8 | 2515 | netdev->in6 = in6_tmp; |
7df6932e | 2516 | netdev->in6_error = 0; |
8b61709d BP |
2517 | break; |
2518 | } | |
2519 | } | |
2520 | fclose(file); | |
7df6932e AW |
2521 | } else { |
2522 | netdev->in6_error = EOPNOTSUPP; | |
8b61709d | 2523 | } |
b5d57fc8 | 2524 | netdev->cache_valid |= VALID_IN6; |
8b61709d | 2525 | } |
b5d57fc8 | 2526 | *in6 = netdev->in6; |
7df6932e | 2527 | error = netdev->in6_error; |
86383816 BP |
2528 | ovs_mutex_unlock(&netdev->mutex); |
2529 | ||
7df6932e | 2530 | return error; |
8b61709d BP |
2531 | } |
2532 | ||
2533 | static void | |
2534 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
2535 | { | |
2536 | struct sockaddr_in sin; | |
2537 | memset(&sin, 0, sizeof sin); | |
2538 | sin.sin_family = AF_INET; | |
2539 | sin.sin_addr = addr; | |
2540 | sin.sin_port = 0; | |
2541 | ||
2542 | memset(sa, 0, sizeof *sa); | |
2543 | memcpy(sa, &sin, sizeof sin); | |
2544 | } | |
2545 | ||
2546 | static int | |
2547 | do_set_addr(struct netdev *netdev, | |
2548 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
2549 | { | |
2550 | struct ifreq ifr; | |
149f577a | 2551 | |
259e0b1a BP |
2552 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
2553 | return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
2554 | ioctl_name); | |
8b61709d BP |
2555 | } |
2556 | ||
2557 | /* Adds 'router' as a default IP gateway. */ | |
2558 | static int | |
67a4917b | 2559 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
2560 | { |
2561 | struct in_addr any = { INADDR_ANY }; | |
2562 | struct rtentry rt; | |
2563 | int error; | |
2564 | ||
2565 | memset(&rt, 0, sizeof rt); | |
2566 | make_in4_sockaddr(&rt.rt_dst, any); | |
2567 | make_in4_sockaddr(&rt.rt_gateway, router); | |
2568 | make_in4_sockaddr(&rt.rt_genmask, any); | |
2569 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
259e0b1a | 2570 | error = af_inet_ioctl(SIOCADDRT, &rt); |
8b61709d | 2571 | if (error) { |
10a89ef0 | 2572 | VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); |
8b61709d BP |
2573 | } |
2574 | return error; | |
2575 | } | |
2576 | ||
f1acd62b BP |
2577 | static int |
2578 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
2579 | char **netdev_name) | |
2580 | { | |
2581 | static const char fn[] = "/proc/net/route"; | |
2582 | FILE *stream; | |
2583 | char line[256]; | |
2584 | int ln; | |
2585 | ||
2586 | *netdev_name = NULL; | |
2587 | stream = fopen(fn, "r"); | |
2588 | if (stream == NULL) { | |
10a89ef0 | 2589 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
f1acd62b BP |
2590 | return errno; |
2591 | } | |
2592 | ||
2593 | ln = 0; | |
2594 | while (fgets(line, sizeof line, stream)) { | |
2595 | if (++ln >= 2) { | |
2596 | char iface[17]; | |
dbba996b | 2597 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
2598 | int refcnt, metric, mtu; |
2599 | unsigned int flags, use, window, irtt; | |
2600 | ||
c2c28dfd BP |
2601 | if (!ovs_scan(line, |
2602 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
2603 | " %d %u %u\n", | |
2604 | iface, &dest, &gateway, &flags, &refcnt, | |
2605 | &use, &metric, &mask, &mtu, &window, &irtt)) { | |
d295e8e9 | 2606 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
2607 | fn, ln, line); |
2608 | continue; | |
2609 | } | |
2610 | if (!(flags & RTF_UP)) { | |
2611 | /* Skip routes that aren't up. */ | |
2612 | continue; | |
2613 | } | |
2614 | ||
2615 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 2616 | * network byte order, so we don't need need any endian |
f1acd62b BP |
2617 | * conversions here. */ |
2618 | if ((dest & mask) == (host->s_addr & mask)) { | |
2619 | if (!gateway) { | |
2620 | /* The host is directly reachable. */ | |
2621 | next_hop->s_addr = 0; | |
2622 | } else { | |
2623 | /* To reach the host, we must go through a gateway. */ | |
2624 | next_hop->s_addr = gateway; | |
2625 | } | |
2626 | *netdev_name = xstrdup(iface); | |
2627 | fclose(stream); | |
2628 | return 0; | |
2629 | } | |
2630 | } | |
2631 | } | |
2632 | ||
2633 | fclose(stream); | |
2634 | return ENXIO; | |
2635 | } | |
2636 | ||
e210037e | 2637 | static int |
b5d57fc8 | 2638 | netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) |
e210037e | 2639 | { |
b5d57fc8 | 2640 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
275707c3 EJ |
2641 | int error = 0; |
2642 | ||
86383816 | 2643 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
2644 | if (!(netdev->cache_valid & VALID_DRVINFO)) { |
2645 | struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; | |
275707c3 EJ |
2646 | |
2647 | COVERAGE_INC(netdev_get_ethtool); | |
b5d57fc8 BP |
2648 | memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); |
2649 | error = netdev_linux_do_ethtool(netdev->up.name, | |
275707c3 EJ |
2650 | cmd, |
2651 | ETHTOOL_GDRVINFO, | |
2652 | "ETHTOOL_GDRVINFO"); | |
2653 | if (!error) { | |
b5d57fc8 | 2654 | netdev->cache_valid |= VALID_DRVINFO; |
275707c3 EJ |
2655 | } |
2656 | } | |
e210037e | 2657 | |
e210037e | 2658 | if (!error) { |
b5d57fc8 BP |
2659 | smap_add(smap, "driver_name", netdev->drvinfo.driver); |
2660 | smap_add(smap, "driver_version", netdev->drvinfo.version); | |
2661 | smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); | |
e210037e | 2662 | } |
86383816 BP |
2663 | ovs_mutex_unlock(&netdev->mutex); |
2664 | ||
e210037e AE |
2665 | return error; |
2666 | } | |
2667 | ||
4f925bd3 | 2668 | static int |
275707c3 EJ |
2669 | netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, |
2670 | struct smap *smap) | |
4f925bd3 | 2671 | { |
79f1cbe9 | 2672 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
2673 | return 0; |
2674 | } | |
2675 | ||
8b61709d BP |
2676 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2677 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2678 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2679 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2680 | static int | |
2681 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
74ff3298 | 2682 | ovs_be32 ip, struct eth_addr *mac) |
8b61709d BP |
2683 | { |
2684 | struct arpreq r; | |
c100e025 | 2685 | struct sockaddr_in sin; |
8b61709d BP |
2686 | int retval; |
2687 | ||
2688 | memset(&r, 0, sizeof r); | |
f2cc621b | 2689 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2690 | sin.sin_family = AF_INET; |
2691 | sin.sin_addr.s_addr = ip; | |
2692 | sin.sin_port = 0; | |
2693 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2694 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2695 | r.arp_flags = 0; | |
71d7c22f | 2696 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d | 2697 | COVERAGE_INC(netdev_arp_lookup); |
259e0b1a | 2698 | retval = af_inet_ioctl(SIOCGARP, &r); |
8b61709d BP |
2699 | if (!retval) { |
2700 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2701 | } else if (retval != ENXIO) { | |
2702 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
10a89ef0 BP |
2703 | netdev_get_name(netdev), IP_ARGS(ip), |
2704 | ovs_strerror(retval)); | |
8b61709d BP |
2705 | } |
2706 | return retval; | |
2707 | } | |
2708 | ||
2709 | static int | |
2710 | nd_to_iff_flags(enum netdev_flags nd) | |
2711 | { | |
2712 | int iff = 0; | |
2713 | if (nd & NETDEV_UP) { | |
2714 | iff |= IFF_UP; | |
2715 | } | |
2716 | if (nd & NETDEV_PROMISC) { | |
2717 | iff |= IFF_PROMISC; | |
2718 | } | |
7ba19d41 AC |
2719 | if (nd & NETDEV_LOOPBACK) { |
2720 | iff |= IFF_LOOPBACK; | |
2721 | } | |
8b61709d BP |
2722 | return iff; |
2723 | } | |
2724 | ||
2725 | static int | |
2726 | iff_to_nd_flags(int iff) | |
2727 | { | |
2728 | enum netdev_flags nd = 0; | |
2729 | if (iff & IFF_UP) { | |
2730 | nd |= NETDEV_UP; | |
2731 | } | |
2732 | if (iff & IFF_PROMISC) { | |
2733 | nd |= NETDEV_PROMISC; | |
2734 | } | |
7ba19d41 AC |
2735 | if (iff & IFF_LOOPBACK) { |
2736 | nd |= NETDEV_LOOPBACK; | |
2737 | } | |
8b61709d BP |
2738 | return nd; |
2739 | } | |
2740 | ||
2741 | static int | |
4f9f3f21 BP |
2742 | update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
2743 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2744 | OVS_REQUIRES(netdev->mutex) | |
8b61709d BP |
2745 | { |
2746 | int old_flags, new_flags; | |
c37d4da4 EJ |
2747 | int error = 0; |
2748 | ||
b5d57fc8 | 2749 | old_flags = netdev->ifi_flags; |
c37d4da4 EJ |
2750 | *old_flagsp = iff_to_nd_flags(old_flags); |
2751 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2752 | if (new_flags != old_flags) { | |
4f9f3f21 BP |
2753 | error = set_flags(netdev_get_name(&netdev->up), new_flags); |
2754 | get_flags(&netdev->up, &netdev->ifi_flags); | |
8b61709d | 2755 | } |
4f9f3f21 BP |
2756 | |
2757 | return error; | |
2758 | } | |
2759 | ||
2760 | static int | |
2761 | netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, | |
2762 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2763 | { | |
2764 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
2765 | int error; | |
2766 | ||
2767 | ovs_mutex_lock(&netdev->mutex); | |
2768 | error = update_flags(netdev, off, on, old_flagsp); | |
86383816 BP |
2769 | ovs_mutex_unlock(&netdev->mutex); |
2770 | ||
8b61709d BP |
2771 | return error; |
2772 | } | |
2773 | ||
2f9dd77f | 2774 | #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, \ |
51f87458 | 2775 | GET_FEATURES, GET_STATUS) \ |
c3827f61 BP |
2776 | { \ |
2777 | NAME, \ | |
2778 | \ | |
259e0b1a | 2779 | NULL, \ |
c3827f61 BP |
2780 | netdev_linux_run, \ |
2781 | netdev_linux_wait, \ | |
2782 | \ | |
9dc63482 BP |
2783 | netdev_linux_alloc, \ |
2784 | CONSTRUCT, \ | |
2785 | netdev_linux_destruct, \ | |
2786 | netdev_linux_dealloc, \ | |
de5cdb90 | 2787 | NULL, /* get_config */ \ |
6d9e6eb4 | 2788 | NULL, /* set_config */ \ |
f431bf7d | 2789 | NULL, /* get_tunnel_config */ \ |
a36de779 PS |
2790 | NULL, /* build header */ \ |
2791 | NULL, /* push header */ \ | |
2792 | NULL, /* pop header */ \ | |
7dec44fe | 2793 | NULL, /* get_numa_id */ \ |
5496878c | 2794 | NULL, /* set_multiq */ \ |
c3827f61 | 2795 | \ |
c3827f61 BP |
2796 | netdev_linux_send, \ |
2797 | netdev_linux_send_wait, \ | |
2798 | \ | |
2799 | netdev_linux_set_etheraddr, \ | |
2800 | netdev_linux_get_etheraddr, \ | |
2801 | netdev_linux_get_mtu, \ | |
9b020780 | 2802 | netdev_linux_set_mtu, \ |
c3827f61 BP |
2803 | netdev_linux_get_ifindex, \ |
2804 | netdev_linux_get_carrier, \ | |
65c3058c | 2805 | netdev_linux_get_carrier_resets, \ |
1670c579 | 2806 | netdev_linux_set_miimon_interval, \ |
f613a0d7 | 2807 | GET_STATS, \ |
c3827f61 | 2808 | \ |
51f87458 | 2809 | GET_FEATURES, \ |
c3827f61 | 2810 | netdev_linux_set_advertisements, \ |
c3827f61 BP |
2811 | \ |
2812 | netdev_linux_set_policing, \ | |
2813 | netdev_linux_get_qos_types, \ | |
2814 | netdev_linux_get_qos_capabilities, \ | |
2815 | netdev_linux_get_qos, \ | |
2816 | netdev_linux_set_qos, \ | |
2817 | netdev_linux_get_queue, \ | |
2818 | netdev_linux_set_queue, \ | |
2819 | netdev_linux_delete_queue, \ | |
2820 | netdev_linux_get_queue_stats, \ | |
89454bf4 BP |
2821 | netdev_linux_queue_dump_start, \ |
2822 | netdev_linux_queue_dump_next, \ | |
2823 | netdev_linux_queue_dump_done, \ | |
c3827f61 BP |
2824 | netdev_linux_dump_queue_stats, \ |
2825 | \ | |
2826 | netdev_linux_get_in4, \ | |
2827 | netdev_linux_set_in4, \ | |
2828 | netdev_linux_get_in6, \ | |
2829 | netdev_linux_add_router, \ | |
2830 | netdev_linux_get_next_hop, \ | |
4f925bd3 | 2831 | GET_STATUS, \ |
c3827f61 BP |
2832 | netdev_linux_arp_lookup, \ |
2833 | \ | |
2834 | netdev_linux_update_flags, \ | |
2835 | \ | |
f7791740 PS |
2836 | netdev_linux_rxq_alloc, \ |
2837 | netdev_linux_rxq_construct, \ | |
2838 | netdev_linux_rxq_destruct, \ | |
2839 | netdev_linux_rxq_dealloc, \ | |
2840 | netdev_linux_rxq_recv, \ | |
2841 | netdev_linux_rxq_wait, \ | |
2842 | netdev_linux_rxq_drain, \ | |
c3827f61 BP |
2843 | } |
2844 | ||
2845 | const struct netdev_class netdev_linux_class = | |
2846 | NETDEV_LINUX_CLASS( | |
2847 | "system", | |
9dc63482 | 2848 | netdev_linux_construct, |
f613a0d7 | 2849 | netdev_linux_get_stats, |
51f87458 | 2850 | netdev_linux_get_features, |
275707c3 | 2851 | netdev_linux_get_status); |
c3827f61 BP |
2852 | |
2853 | const struct netdev_class netdev_tap_class = | |
2854 | NETDEV_LINUX_CLASS( | |
2855 | "tap", | |
9dc63482 | 2856 | netdev_linux_construct_tap, |
bba1e6f3 | 2857 | netdev_tap_get_stats, |
51f87458 | 2858 | netdev_linux_get_features, |
275707c3 | 2859 | netdev_linux_get_status); |
c3827f61 BP |
2860 | |
2861 | const struct netdev_class netdev_internal_class = | |
2862 | NETDEV_LINUX_CLASS( | |
2863 | "internal", | |
9dc63482 | 2864 | netdev_linux_construct, |
bba1e6f3 | 2865 | netdev_internal_get_stats, |
51f87458 | 2866 | NULL, /* get_features */ |
275707c3 | 2867 | netdev_internal_get_status); |
8b61709d | 2868 | \f |
677d9158 JV |
2869 | |
2870 | #define CODEL_N_QUEUES 0x0000 | |
2871 | ||
2f4298ce BP |
2872 | /* In sufficiently new kernel headers these are defined as enums in |
2873 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
2874 | * kernels. (This overrides any enum definition in the header file but that's | |
2875 | * harmless.) */ | |
2876 | #define TCA_CODEL_TARGET 1 | |
2877 | #define TCA_CODEL_LIMIT 2 | |
2878 | #define TCA_CODEL_INTERVAL 3 | |
2879 | ||
677d9158 JV |
2880 | struct codel { |
2881 | struct tc tc; | |
2882 | uint32_t target; | |
2883 | uint32_t limit; | |
2884 | uint32_t interval; | |
2885 | }; | |
2886 | ||
2887 | static struct codel * | |
2888 | codel_get__(const struct netdev *netdev_) | |
2889 | { | |
2890 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
2891 | return CONTAINER_OF(netdev->tc, struct codel, tc); | |
2892 | } | |
2893 | ||
2894 | static void | |
2895 | codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
2896 | uint32_t interval) | |
2897 | { | |
2898 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
2899 | struct codel *codel; | |
2900 | ||
2901 | codel = xmalloc(sizeof *codel); | |
2902 | tc_init(&codel->tc, &tc_ops_codel); | |
2903 | codel->target = target; | |
2904 | codel->limit = limit; | |
2905 | codel->interval = interval; | |
2906 | ||
2907 | netdev->tc = &codel->tc; | |
2908 | } | |
2909 | ||
2910 | static int | |
2911 | codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
2912 | uint32_t interval) | |
2913 | { | |
2914 | size_t opt_offset; | |
2915 | struct ofpbuf request; | |
2916 | struct tcmsg *tcmsg; | |
2917 | uint32_t otarget, olimit, ointerval; | |
2918 | int error; | |
2919 | ||
2920 | tc_del_qdisc(netdev); | |
2921 | ||
2922 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
2923 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
2924 | if (!tcmsg) { | |
2925 | return ENODEV; | |
2926 | } | |
2927 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
2928 | tcmsg->tcm_parent = TC_H_ROOT; | |
2929 | ||
2930 | otarget = target ? target : 5000; | |
2931 | olimit = limit ? limit : 10240; | |
2932 | ointerval = interval ? interval : 100000; | |
2933 | ||
2934 | nl_msg_put_string(&request, TCA_KIND, "codel"); | |
2935 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2936 | nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget); | |
2937 | nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit); | |
2938 | nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval); | |
2939 | nl_msg_end_nested(&request, opt_offset); | |
2940 | ||
2941 | error = tc_transact(&request, NULL); | |
2942 | if (error) { | |
2943 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
2944 | "target %u, limit %u, interval %u error %d(%s)", | |
2945 | netdev_get_name(netdev), | |
2946 | otarget, olimit, ointerval, | |
2947 | error, ovs_strerror(error)); | |
2948 | } | |
2949 | return error; | |
2950 | } | |
2951 | ||
2952 | static void | |
2953 | codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
2954 | const struct smap *details, struct codel *codel) | |
2955 | { | |
2956 | const char *target_s; | |
2957 | const char *limit_s; | |
2958 | const char *interval_s; | |
2959 | ||
2960 | target_s = smap_get(details, "target"); | |
2961 | limit_s = smap_get(details, "limit"); | |
2962 | interval_s = smap_get(details, "interval"); | |
2963 | ||
2964 | codel->target = target_s ? strtoull(target_s, NULL, 10) : 0; | |
2965 | codel->limit = limit_s ? strtoull(limit_s, NULL, 10) : 0; | |
2966 | codel->interval = interval_s ? strtoull(interval_s, NULL, 10) : 0; | |
2967 | ||
2968 | if (!codel->target) { | |
2969 | codel->target = 5000; | |
2970 | } | |
2971 | if (!codel->limit) { | |
2972 | codel->limit = 10240; | |
2973 | } | |
2974 | if (!codel->interval) { | |
2975 | codel->interval = 100000; | |
2976 | } | |
2977 | } | |
2978 | ||
2979 | static int | |
2980 | codel_tc_install(struct netdev *netdev, const struct smap *details) | |
2981 | { | |
2982 | int error; | |
2983 | struct codel codel; | |
2984 | ||
2985 | codel_parse_qdisc_details__(netdev, details, &codel); | |
2986 | error = codel_setup_qdisc__(netdev, codel.target, codel.limit, | |
2987 | codel.interval); | |
2988 | if (!error) { | |
2989 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
2990 | } | |
2991 | return error; | |
2992 | } | |
2993 | ||
2994 | static int | |
2995 | codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel) | |
2996 | { | |
2997 | static const struct nl_policy tca_codel_policy[] = { | |
2998 | [TCA_CODEL_TARGET] = { .type = NL_A_U32 }, | |
2999 | [TCA_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3000 | [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 } | |
3001 | }; | |
3002 | ||
3003 | struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)]; | |
3004 | ||
3005 | if (!nl_parse_nested(nl_options, tca_codel_policy, | |
3006 | attrs, ARRAY_SIZE(tca_codel_policy))) { | |
3007 | VLOG_WARN_RL(&rl, "failed to parse CoDel class options"); | |
3008 | return EPROTO; | |
3009 | } | |
3010 | ||
3011 | codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]); | |
3012 | codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]); | |
3013 | codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]); | |
3014 | return 0; | |
3015 | } | |
3016 | ||
3017 | static int | |
3018 | codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3019 | { | |
3020 | struct nlattr *nlattr; | |
3021 | const char * kind; | |
3022 | int error; | |
3023 | struct codel codel; | |
3024 | ||
3025 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3026 | if (error != 0) { | |
3027 | return error; | |
3028 | } | |
3029 | ||
3030 | error = codel_parse_tca_options__(nlattr, &codel); | |
3031 | if (error != 0) { | |
3032 | return error; | |
3033 | } | |
3034 | ||
3035 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3036 | return 0; | |
3037 | } | |
3038 | ||
3039 | ||
3040 | static void | |
3041 | codel_tc_destroy(struct tc *tc) | |
3042 | { | |
3043 | struct codel *codel = CONTAINER_OF(tc, struct codel, tc); | |
3044 | tc_destroy(tc); | |
3045 | free(codel); | |
3046 | } | |
3047 | ||
3048 | static int | |
3049 | codel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3050 | { | |
3051 | const struct codel *codel = codel_get__(netdev); | |
3052 | smap_add_format(details, "target", "%u", codel->target); | |
3053 | smap_add_format(details, "limit", "%u", codel->limit); | |
3054 | smap_add_format(details, "interval", "%u", codel->interval); | |
3055 | return 0; | |
3056 | } | |
3057 | ||
3058 | static int | |
3059 | codel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3060 | { | |
3061 | struct codel codel; | |
3062 | ||
3063 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3064 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3065 | codel_get__(netdev)->target = codel.target; | |
3066 | codel_get__(netdev)->limit = codel.limit; | |
3067 | codel_get__(netdev)->interval = codel.interval; | |
3068 | return 0; | |
3069 | } | |
3070 | ||
3071 | static const struct tc_ops tc_ops_codel = { | |
3072 | "codel", /* linux_name */ | |
3073 | "linux-codel", /* ovs_name */ | |
3074 | CODEL_N_QUEUES, /* n_queues */ | |
3075 | codel_tc_install, | |
3076 | codel_tc_load, | |
3077 | codel_tc_destroy, | |
3078 | codel_qdisc_get, | |
3079 | codel_qdisc_set, | |
3080 | NULL, | |
3081 | NULL, | |
3082 | NULL, | |
3083 | NULL, | |
3084 | NULL | |
3085 | }; | |
3086 | \f | |
3087 | /* FQ-CoDel traffic control class. */ | |
3088 | ||
3089 | #define FQCODEL_N_QUEUES 0x0000 | |
3090 | ||
2f4298ce BP |
3091 | /* In sufficiently new kernel headers these are defined as enums in |
3092 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3093 | * kernels. (This overrides any enum definition in the header file but that's | |
3094 | * harmless.) */ | |
3095 | #define TCA_FQ_CODEL_TARGET 1 | |
3096 | #define TCA_FQ_CODEL_LIMIT 2 | |
3097 | #define TCA_FQ_CODEL_INTERVAL 3 | |
3098 | #define TCA_FQ_CODEL_ECN 4 | |
3099 | #define TCA_FQ_CODEL_FLOWS 5 | |
3100 | #define TCA_FQ_CODEL_QUANTUM 6 | |
3101 | ||
677d9158 JV |
3102 | struct fqcodel { |
3103 | struct tc tc; | |
3104 | uint32_t target; | |
3105 | uint32_t limit; | |
3106 | uint32_t interval; | |
3107 | uint32_t flows; | |
3108 | uint32_t quantum; | |
3109 | }; | |
3110 | ||
3111 | static struct fqcodel * | |
3112 | fqcodel_get__(const struct netdev *netdev_) | |
3113 | { | |
3114 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3115 | return CONTAINER_OF(netdev->tc, struct fqcodel, tc); | |
3116 | } | |
3117 | ||
3118 | static void | |
3119 | fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3120 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3121 | { | |
3122 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3123 | struct fqcodel *fqcodel; | |
3124 | ||
3125 | fqcodel = xmalloc(sizeof *fqcodel); | |
3126 | tc_init(&fqcodel->tc, &tc_ops_fqcodel); | |
3127 | fqcodel->target = target; | |
3128 | fqcodel->limit = limit; | |
3129 | fqcodel->interval = interval; | |
3130 | fqcodel->flows = flows; | |
3131 | fqcodel->quantum = quantum; | |
3132 | ||
3133 | netdev->tc = &fqcodel->tc; | |
3134 | } | |
3135 | ||
3136 | static int | |
3137 | fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3138 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3139 | { | |
3140 | size_t opt_offset; | |
3141 | struct ofpbuf request; | |
3142 | struct tcmsg *tcmsg; | |
3143 | uint32_t otarget, olimit, ointerval, oflows, oquantum; | |
3144 | int error; | |
3145 | ||
3146 | tc_del_qdisc(netdev); | |
3147 | ||
3148 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3149 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3150 | if (!tcmsg) { | |
3151 | return ENODEV; | |
3152 | } | |
3153 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3154 | tcmsg->tcm_parent = TC_H_ROOT; | |
3155 | ||
3156 | otarget = target ? target : 5000; | |
3157 | olimit = limit ? limit : 10240; | |
3158 | ointerval = interval ? interval : 100000; | |
3159 | oflows = flows ? flows : 1024; | |
3160 | oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514 | |
3161 | not mtu */ | |
3162 | ||
3163 | nl_msg_put_string(&request, TCA_KIND, "fq_codel"); | |
3164 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3165 | nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget); | |
3166 | nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit); | |
3167 | nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval); | |
3168 | nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows); | |
3169 | nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum); | |
3170 | nl_msg_end_nested(&request, opt_offset); | |
3171 | ||
3172 | error = tc_transact(&request, NULL); | |
3173 | if (error) { | |
3174 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3175 | "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)", | |
3176 | netdev_get_name(netdev), | |
3177 | otarget, olimit, ointerval, oflows, oquantum, | |
3178 | error, ovs_strerror(error)); | |
3179 | } | |
3180 | return error; | |
3181 | } | |
3182 | ||
3183 | static void | |
3184 | fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3185 | const struct smap *details, struct fqcodel *fqcodel) | |
3186 | { | |
3187 | const char *target_s; | |
3188 | const char *limit_s; | |
3189 | const char *interval_s; | |
3190 | const char *flows_s; | |
3191 | const char *quantum_s; | |
3192 | ||
3193 | target_s = smap_get(details, "target"); | |
3194 | limit_s = smap_get(details, "limit"); | |
3195 | interval_s = smap_get(details, "interval"); | |
3196 | flows_s = smap_get(details, "flows"); | |
3197 | quantum_s = smap_get(details, "quantum"); | |
3198 | fqcodel->target = target_s ? strtoull(target_s, NULL, 10) : 0; | |
3199 | fqcodel->limit = limit_s ? strtoull(limit_s, NULL, 10) : 0; | |
3200 | fqcodel->interval = interval_s ? strtoull(interval_s, NULL, 10) : 0; | |
3201 | fqcodel->flows = flows_s ? strtoull(flows_s, NULL, 10) : 0; | |
3202 | fqcodel->quantum = quantum_s ? strtoull(quantum_s, NULL, 10) : 0; | |
3203 | if (!fqcodel->target) { | |
3204 | fqcodel->target = 5000; | |
3205 | } | |
3206 | if (!fqcodel->limit) { | |
3207 | fqcodel->limit = 10240; | |
3208 | } | |
3209 | if (!fqcodel->interval) { | |
3210 | fqcodel->interval = 1000000; | |
3211 | } | |
3212 | if (!fqcodel->flows) { | |
3213 | fqcodel->flows = 1024; | |
3214 | } | |
3215 | if (!fqcodel->quantum) { | |
3216 | fqcodel->quantum = 1514; | |
3217 | } | |
3218 | } | |
3219 | ||
3220 | static int | |
3221 | fqcodel_tc_install(struct netdev *netdev, const struct smap *details) | |
3222 | { | |
3223 | int error; | |
3224 | struct fqcodel fqcodel; | |
3225 | ||
3226 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3227 | error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit, | |
3228 | fqcodel.interval, fqcodel.flows, | |
3229 | fqcodel.quantum); | |
3230 | if (!error) { | |
3231 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, | |
3232 | fqcodel.interval, fqcodel.flows, fqcodel.quantum); | |
3233 | } | |
3234 | return error; | |
3235 | } | |
3236 | ||
3237 | static int | |
3238 | fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel) | |
3239 | { | |
3240 | static const struct nl_policy tca_fqcodel_policy[] = { | |
3241 | [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3242 | [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3243 | [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 }, | |
3244 | [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 }, | |
3245 | [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 } | |
3246 | }; | |
3247 | ||
3248 | struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)]; | |
3249 | ||
3250 | if (!nl_parse_nested(nl_options, tca_fqcodel_policy, | |
3251 | attrs, ARRAY_SIZE(tca_fqcodel_policy))) { | |
3252 | VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options"); | |
3253 | return EPROTO; | |
3254 | } | |
3255 | ||
3256 | fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]); | |
3257 | fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]); | |
3258 | fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]); | |
3259 | fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]); | |
3260 | fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]); | |
3261 | return 0; | |
3262 | } | |
3263 | ||
3264 | static int | |
3265 | fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3266 | { | |
3267 | struct nlattr *nlattr; | |
3268 | const char * kind; | |
3269 | int error; | |
3270 | struct fqcodel fqcodel; | |
3271 | ||
3272 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3273 | if (error != 0) { | |
3274 | return error; | |
3275 | } | |
3276 | ||
3277 | error = fqcodel_parse_tca_options__(nlattr, &fqcodel); | |
3278 | if (error != 0) { | |
3279 | return error; | |
3280 | } | |
3281 | ||
3282 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3283 | fqcodel.flows, fqcodel.quantum); | |
3284 | return 0; | |
3285 | } | |
3286 | ||
3287 | static void | |
3288 | fqcodel_tc_destroy(struct tc *tc) | |
3289 | { | |
3290 | struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc); | |
3291 | tc_destroy(tc); | |
3292 | free(fqcodel); | |
3293 | } | |
3294 | ||
3295 | static int | |
3296 | fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3297 | { | |
3298 | const struct fqcodel *fqcodel = fqcodel_get__(netdev); | |
3299 | smap_add_format(details, "target", "%u", fqcodel->target); | |
3300 | smap_add_format(details, "limit", "%u", fqcodel->limit); | |
3301 | smap_add_format(details, "interval", "%u", fqcodel->interval); | |
3302 | smap_add_format(details, "flows", "%u", fqcodel->flows); | |
3303 | smap_add_format(details, "quantum", "%u", fqcodel->quantum); | |
3304 | return 0; | |
3305 | } | |
3306 | ||
3307 | static int | |
3308 | fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3309 | { | |
3310 | struct fqcodel fqcodel; | |
3311 | ||
3312 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3313 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3314 | fqcodel.flows, fqcodel.quantum); | |
3315 | fqcodel_get__(netdev)->target = fqcodel.target; | |
3316 | fqcodel_get__(netdev)->limit = fqcodel.limit; | |
3317 | fqcodel_get__(netdev)->interval = fqcodel.interval; | |
3318 | fqcodel_get__(netdev)->flows = fqcodel.flows; | |
3319 | fqcodel_get__(netdev)->quantum = fqcodel.quantum; | |
3320 | return 0; | |
3321 | } | |
3322 | ||
3323 | static const struct tc_ops tc_ops_fqcodel = { | |
3324 | "fq_codel", /* linux_name */ | |
3325 | "linux-fq_codel", /* ovs_name */ | |
3326 | FQCODEL_N_QUEUES, /* n_queues */ | |
3327 | fqcodel_tc_install, | |
3328 | fqcodel_tc_load, | |
3329 | fqcodel_tc_destroy, | |
3330 | fqcodel_qdisc_get, | |
3331 | fqcodel_qdisc_set, | |
3332 | NULL, | |
3333 | NULL, | |
3334 | NULL, | |
3335 | NULL, | |
3336 | NULL | |
3337 | }; | |
3338 | \f | |
3339 | /* SFQ traffic control class. */ | |
3340 | ||
3341 | #define SFQ_N_QUEUES 0x0000 | |
3342 | ||
3343 | struct sfq { | |
3344 | struct tc tc; | |
3345 | uint32_t quantum; | |
3346 | uint32_t perturb; | |
3347 | }; | |
3348 | ||
3349 | static struct sfq * | |
3350 | sfq_get__(const struct netdev *netdev_) | |
3351 | { | |
3352 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3353 | return CONTAINER_OF(netdev->tc, struct sfq, tc); | |
3354 | } | |
3355 | ||
3356 | static void | |
3357 | sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb) | |
3358 | { | |
3359 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3360 | struct sfq *sfq; | |
3361 | ||
3362 | sfq = xmalloc(sizeof *sfq); | |
3363 | tc_init(&sfq->tc, &tc_ops_sfq); | |
3364 | sfq->perturb = perturb; | |
3365 | sfq->quantum = quantum; | |
3366 | ||
3367 | netdev->tc = &sfq->tc; | |
3368 | } | |
3369 | ||
3370 | static int | |
3371 | sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb) | |
3372 | { | |
3373 | struct tc_sfq_qopt opt; | |
3374 | struct ofpbuf request; | |
3375 | struct tcmsg *tcmsg; | |
3376 | int mtu; | |
3377 | int mtu_error, error; | |
3378 | mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); | |
3379 | ||
3380 | tc_del_qdisc(netdev); | |
3381 | ||
3382 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3383 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
3384 | if (!tcmsg) { | |
3385 | return ENODEV; | |
3386 | } | |
3387 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3388 | tcmsg->tcm_parent = TC_H_ROOT; | |
3389 | ||
3390 | memset(&opt, 0, sizeof opt); | |
3391 | if (!quantum) { | |
3392 | if (!mtu_error) { | |
3393 | opt.quantum = mtu; /* if we cannot find mtu, use default */ | |
3394 | } | |
3395 | } else { | |
3396 | opt.quantum = quantum; | |
3397 | } | |
3398 | ||
3399 | if (!perturb) { | |
3400 | opt.perturb_period = 10; | |
3401 | } else { | |
3402 | opt.perturb_period = perturb; | |
3403 | } | |
3404 | ||
3405 | nl_msg_put_string(&request, TCA_KIND, "sfq"); | |
3406 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3407 | ||
3408 | error = tc_transact(&request, NULL); | |
3409 | if (error) { | |
3410 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3411 | "quantum %u, perturb %u error %d(%s)", | |
3412 | netdev_get_name(netdev), | |
3413 | opt.quantum, opt.perturb_period, | |
3414 | error, ovs_strerror(error)); | |
3415 | } | |
3416 | return error; | |
3417 | } | |
3418 | ||
3419 | static void | |
3420 | sfq_parse_qdisc_details__(struct netdev *netdev, | |
3421 | const struct smap *details, struct sfq *sfq) | |
3422 | { | |
3423 | const char *perturb_s; | |
3424 | const char *quantum_s; | |
3425 | int mtu; | |
3426 | int mtu_error; | |
3427 | ||
3428 | perturb_s = smap_get(details, "perturb"); | |
3429 | quantum_s = smap_get(details, "quantum"); | |
3430 | sfq->perturb = perturb_s ? strtoull(perturb_s, NULL, 10) : 0; | |
3431 | sfq->quantum = quantum_s ? strtoull(quantum_s, NULL, 10) : 0; | |
3432 | if (!sfq->perturb) { | |
3433 | sfq->perturb = 10; | |
3434 | } | |
3435 | ||
3436 | if (!sfq->quantum) { | |
3437 | mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); | |
3438 | if (!mtu_error) { | |
3439 | sfq->quantum = mtu; | |
3440 | } else { | |
3441 | VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a " | |
3442 | "device without mtu"); | |
3443 | return; | |
3444 | } | |
3445 | } | |
3446 | } | |
3447 | ||
3448 | static int | |
3449 | sfq_tc_install(struct netdev *netdev, const struct smap *details) | |
3450 | { | |
3451 | int error; | |
3452 | struct sfq sfq; | |
3453 | ||
3454 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3455 | error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb); | |
3456 | if (!error) { | |
3457 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3458 | } | |
3459 | return error; | |
3460 | } | |
3461 | ||
3462 | static int | |
3463 | sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3464 | { | |
3465 | const struct tc_sfq_qopt *sfq; | |
3466 | struct nlattr *nlattr; | |
3467 | const char * kind; | |
3468 | int error; | |
3469 | ||
3470 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3471 | if (error == 0) { | |
3472 | sfq = nl_attr_get(nlattr); | |
3473 | sfq_install__(netdev, sfq->perturb_period, sfq->quantum); | |
3474 | return 0; | |
3475 | } | |
3476 | ||
3477 | return error; | |
3478 | } | |
3479 | ||
3480 | static void | |
3481 | sfq_tc_destroy(struct tc *tc) | |
3482 | { | |
3483 | struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc); | |
3484 | tc_destroy(tc); | |
3485 | free(sfq); | |
3486 | } | |
3487 | ||
3488 | static int | |
3489 | sfq_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3490 | { | |
3491 | const struct sfq *sfq = sfq_get__(netdev); | |
3492 | smap_add_format(details, "quantum", "%u", sfq->quantum); | |
3493 | smap_add_format(details, "perturb", "%u", sfq->perturb); | |
3494 | return 0; | |
3495 | } | |
3496 | ||
3497 | static int | |
3498 | sfq_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3499 | { | |
3500 | struct sfq sfq; | |
3501 | ||
3502 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3503 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3504 | sfq_get__(netdev)->quantum = sfq.quantum; | |
3505 | sfq_get__(netdev)->perturb = sfq.perturb; | |
3506 | return 0; | |
3507 | } | |
3508 | ||
3509 | static const struct tc_ops tc_ops_sfq = { | |
3510 | "sfq", /* linux_name */ | |
3511 | "linux-sfq", /* ovs_name */ | |
3512 | SFQ_N_QUEUES, /* n_queues */ | |
3513 | sfq_tc_install, | |
3514 | sfq_tc_load, | |
3515 | sfq_tc_destroy, | |
3516 | sfq_qdisc_get, | |
3517 | sfq_qdisc_set, | |
3518 | NULL, | |
3519 | NULL, | |
3520 | NULL, | |
3521 | NULL, | |
3522 | NULL | |
3523 | }; | |
3524 | \f | |
c1c9c9c4 | 3525 | /* HTB traffic control class. */ |
559843ed | 3526 | |
c1c9c9c4 | 3527 | #define HTB_N_QUEUES 0xf000 |
4f631ccd | 3528 | #define HTB_RATE2QUANTUM 10 |
8b61709d | 3529 | |
c1c9c9c4 BP |
3530 | struct htb { |
3531 | struct tc tc; | |
3532 | unsigned int max_rate; /* In bytes/s. */ | |
3533 | }; | |
8b61709d | 3534 | |
c1c9c9c4 | 3535 | struct htb_class { |
93b13be8 | 3536 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
3537 | unsigned int min_rate; /* In bytes/s. */ |
3538 | unsigned int max_rate; /* In bytes/s. */ | |
3539 | unsigned int burst; /* In bytes. */ | |
3540 | unsigned int priority; /* Lower values are higher priorities. */ | |
3541 | }; | |
8b61709d | 3542 | |
c1c9c9c4 | 3543 | static struct htb * |
b5d57fc8 | 3544 | htb_get__(const struct netdev *netdev_) |
c1c9c9c4 | 3545 | { |
b5d57fc8 BP |
3546 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
3547 | return CONTAINER_OF(netdev->tc, struct htb, tc); | |
c1c9c9c4 BP |
3548 | } |
3549 | ||
24045e35 | 3550 | static void |
b5d57fc8 | 3551 | htb_install__(struct netdev *netdev_, uint64_t max_rate) |
c1c9c9c4 | 3552 | { |
b5d57fc8 | 3553 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3554 | struct htb *htb; |
3555 | ||
3556 | htb = xmalloc(sizeof *htb); | |
3557 | tc_init(&htb->tc, &tc_ops_htb); | |
3558 | htb->max_rate = max_rate; | |
3559 | ||
b5d57fc8 | 3560 | netdev->tc = &htb->tc; |
c1c9c9c4 BP |
3561 | } |
3562 | ||
3563 | /* Create an HTB qdisc. | |
3564 | * | |
a339aa81 | 3565 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
3566 | static int |
3567 | htb_setup_qdisc__(struct netdev *netdev) | |
3568 | { | |
3569 | size_t opt_offset; | |
3570 | struct tc_htb_glob opt; | |
3571 | struct ofpbuf request; | |
3572 | struct tcmsg *tcmsg; | |
3573 | ||
3574 | tc_del_qdisc(netdev); | |
3575 | ||
3576 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
3577 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
3578 | if (!tcmsg) { |
3579 | return ENODEV; | |
3580 | } | |
c1c9c9c4 BP |
3581 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
3582 | tcmsg->tcm_parent = TC_H_ROOT; | |
3583 | ||
3584 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3585 | ||
3586 | memset(&opt, 0, sizeof opt); | |
4f631ccd | 3587 | opt.rate2quantum = HTB_RATE2QUANTUM; |
c1c9c9c4 | 3588 | opt.version = 3; |
4ecf12d5 | 3589 | opt.defcls = 1; |
c1c9c9c4 BP |
3590 | |
3591 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3592 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
3593 | nl_msg_end_nested(&request, opt_offset); | |
3594 | ||
3595 | return tc_transact(&request, NULL); | |
3596 | } | |
3597 | ||
3598 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
3599 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
3600 | static int | |
3601 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
3602 | unsigned int parent, struct htb_class *class) | |
3603 | { | |
3604 | size_t opt_offset; | |
3605 | struct tc_htb_opt opt; | |
3606 | struct ofpbuf request; | |
3607 | struct tcmsg *tcmsg; | |
3608 | int error; | |
3609 | int mtu; | |
3610 | ||
73371c09 | 3611 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 3612 | if (error) { |
f915f1a8 BP |
3613 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
3614 | netdev_get_name(netdev)); | |
9b020780 | 3615 | return error; |
f915f1a8 | 3616 | } |
c1c9c9c4 BP |
3617 | |
3618 | memset(&opt, 0, sizeof opt); | |
3619 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
3620 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
4f631ccd AW |
3621 | /* Makes sure the quantum is at least MTU. Setting quantum will |
3622 | * make htb ignore the r2q for this class. */ | |
3623 | if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) { | |
3624 | opt.quantum = mtu; | |
3625 | } | |
c1c9c9c4 BP |
3626 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); |
3627 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
3628 | opt.prio = class->priority; | |
3629 | ||
3630 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
23a98ffe BP |
3631 | if (!tcmsg) { |
3632 | return ENODEV; | |
3633 | } | |
c1c9c9c4 BP |
3634 | tcmsg->tcm_handle = handle; |
3635 | tcmsg->tcm_parent = parent; | |
3636 | ||
3637 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3638 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3639 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
3640 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
3641 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
3642 | nl_msg_end_nested(&request, opt_offset); | |
3643 | ||
3644 | error = tc_transact(&request, NULL); | |
3645 | if (error) { | |
3646 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3647 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
3648 | netdev_get_name(netdev), | |
3649 | tc_get_major(handle), tc_get_minor(handle), | |
3650 | tc_get_major(parent), tc_get_minor(parent), | |
3651 | class->min_rate, class->max_rate, | |
10a89ef0 | 3652 | class->burst, class->priority, ovs_strerror(error)); |
c1c9c9c4 BP |
3653 | } |
3654 | return error; | |
3655 | } | |
3656 | ||
3657 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
3658 | * description of them into 'details'. The description complies with the | |
3659 | * specification given in the vswitch database documentation for linux-htb | |
3660 | * queue details. */ | |
3661 | static int | |
3662 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
3663 | { | |
3664 | static const struct nl_policy tca_htb_policy[] = { | |
3665 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
3666 | .min_len = sizeof(struct tc_htb_opt) }, | |
3667 | }; | |
3668 | ||
3669 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
3670 | const struct tc_htb_opt *htb; | |
3671 | ||
3672 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
3673 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
3674 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
3675 | return EPROTO; | |
3676 | } | |
3677 | ||
3678 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
3679 | class->min_rate = htb->rate.rate; | |
3680 | class->max_rate = htb->ceil.rate; | |
3681 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
3682 | class->priority = htb->prio; | |
3683 | return 0; | |
3684 | } | |
3685 | ||
3686 | static int | |
3687 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
3688 | struct htb_class *options, | |
3689 | struct netdev_queue_stats *stats) | |
3690 | { | |
3691 | struct nlattr *nl_options; | |
3692 | unsigned int handle; | |
3693 | int error; | |
3694 | ||
3695 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
3696 | if (!error && queue_id) { | |
17ee3c1f BP |
3697 | unsigned int major = tc_get_major(handle); |
3698 | unsigned int minor = tc_get_minor(handle); | |
3699 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
3700 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
3701 | } else { |
3702 | error = EPROTO; | |
3703 | } | |
3704 | } | |
3705 | if (!error && options) { | |
3706 | error = htb_parse_tca_options__(nl_options, options); | |
3707 | } | |
3708 | return error; | |
3709 | } | |
3710 | ||
3711 | static void | |
73371c09 | 3712 | htb_parse_qdisc_details__(struct netdev *netdev_, |
79f1cbe9 | 3713 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 | 3714 | { |
73371c09 | 3715 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3716 | const char *max_rate_s; |
3717 | ||
79f1cbe9 | 3718 | max_rate_s = smap_get(details, "max-rate"); |
c1c9c9c4 BP |
3719 | hc->max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
3720 | if (!hc->max_rate) { | |
a00ca915 | 3721 | enum netdev_features current; |
c1c9c9c4 | 3722 | |
73371c09 BP |
3723 | netdev_linux_read_features(netdev); |
3724 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 3725 | hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
c1c9c9c4 BP |
3726 | } |
3727 | hc->min_rate = hc->max_rate; | |
3728 | hc->burst = 0; | |
3729 | hc->priority = 0; | |
3730 | } | |
3731 | ||
3732 | static int | |
3733 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 3734 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
3735 | { |
3736 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 EJ |
3737 | const char *min_rate_s = smap_get(details, "min-rate"); |
3738 | const char *max_rate_s = smap_get(details, "max-rate"); | |
3739 | const char *burst_s = smap_get(details, "burst"); | |
3740 | const char *priority_s = smap_get(details, "priority"); | |
9b020780 | 3741 | int mtu, error; |
c1c9c9c4 | 3742 | |
73371c09 | 3743 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 3744 | if (error) { |
f915f1a8 BP |
3745 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
3746 | netdev_get_name(netdev)); | |
9b020780 | 3747 | return error; |
f915f1a8 BP |
3748 | } |
3749 | ||
4f104611 EJ |
3750 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
3751 | * on uncongested links. */ | |
c45ab5e9 | 3752 | hc->min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
4f104611 | 3753 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
3754 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
3755 | ||
3756 | /* max-rate */ | |
3757 | hc->max_rate = (max_rate_s | |
3758 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
3759 | : htb->max_rate); | |
3760 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); | |
3761 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
3762 | ||
3763 | /* burst | |
3764 | * | |
3765 | * According to hints in the documentation that I've read, it is important | |
3766 | * that 'burst' be at least as big as the largest frame that might be | |
3767 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
3768 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
3769 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
3770 | * the MTU. We actually add 64, instead of 14, as a guard against | |
3771 | * additional headers get tacked on somewhere that we're not aware of. */ | |
c1c9c9c4 BP |
3772 | hc->burst = burst_s ? strtoull(burst_s, NULL, 10) / 8 : 0; |
3773 | hc->burst = MAX(hc->burst, mtu + 64); | |
3774 | ||
3775 | /* priority */ | |
3776 | hc->priority = priority_s ? strtoul(priority_s, NULL, 10) : 0; | |
3777 | ||
3778 | return 0; | |
3779 | } | |
3780 | ||
3781 | static int | |
3782 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
3783 | unsigned int parent, struct htb_class *options, | |
3784 | struct netdev_queue_stats *stats) | |
3785 | { | |
3786 | struct ofpbuf *reply; | |
3787 | int error; | |
3788 | ||
3789 | error = tc_query_class(netdev, handle, parent, &reply); | |
3790 | if (!error) { | |
3791 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
3792 | ofpbuf_delete(reply); | |
3793 | } | |
3794 | return error; | |
3795 | } | |
3796 | ||
3797 | static int | |
79f1cbe9 | 3798 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
3799 | { |
3800 | int error; | |
3801 | ||
3802 | error = htb_setup_qdisc__(netdev); | |
3803 | if (!error) { | |
3804 | struct htb_class hc; | |
3805 | ||
3806 | htb_parse_qdisc_details__(netdev, details, &hc); | |
3807 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3808 | tc_make_handle(1, 0), &hc); | |
3809 | if (!error) { | |
3810 | htb_install__(netdev, hc.max_rate); | |
3811 | } | |
3812 | } | |
3813 | return error; | |
3814 | } | |
3815 | ||
93b13be8 BP |
3816 | static struct htb_class * |
3817 | htb_class_cast__(const struct tc_queue *queue) | |
3818 | { | |
3819 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
3820 | } | |
3821 | ||
c1c9c9c4 BP |
3822 | static void |
3823 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
3824 | const struct htb_class *hc) | |
3825 | { | |
3826 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
3827 | size_t hash = hash_int(queue_id, 0); |
3828 | struct tc_queue *queue; | |
c1c9c9c4 BP |
3829 | struct htb_class *hcp; |
3830 | ||
93b13be8 BP |
3831 | queue = tc_find_queue__(netdev, queue_id, hash); |
3832 | if (queue) { | |
3833 | hcp = htb_class_cast__(queue); | |
3834 | } else { | |
c1c9c9c4 | 3835 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
3836 | queue = &hcp->tc_queue; |
3837 | queue->queue_id = queue_id; | |
6dc34a0d | 3838 | queue->created = time_msec(); |
93b13be8 | 3839 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); |
c1c9c9c4 | 3840 | } |
93b13be8 BP |
3841 | |
3842 | hcp->min_rate = hc->min_rate; | |
3843 | hcp->max_rate = hc->max_rate; | |
3844 | hcp->burst = hc->burst; | |
3845 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
3846 | } |
3847 | ||
3848 | static int | |
3849 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3850 | { | |
c1c9c9c4 | 3851 | struct ofpbuf msg; |
d57695d7 | 3852 | struct queue_dump_state state; |
c1c9c9c4 | 3853 | struct htb_class hc; |
c1c9c9c4 BP |
3854 | |
3855 | /* Get qdisc options. */ | |
3856 | hc.max_rate = 0; | |
3857 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 3858 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
3859 | |
3860 | /* Get queues. */ | |
d57695d7 | 3861 | if (!start_queue_dump(netdev, &state)) { |
23a98ffe BP |
3862 | return ENODEV; |
3863 | } | |
d57695d7 | 3864 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
c1c9c9c4 BP |
3865 | unsigned int queue_id; |
3866 | ||
3867 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
3868 | htb_update_queue__(netdev, queue_id, &hc); | |
3869 | } | |
3870 | } | |
d57695d7 | 3871 | finish_queue_dump(&state); |
c1c9c9c4 BP |
3872 | |
3873 | return 0; | |
3874 | } | |
3875 | ||
3876 | static void | |
3877 | htb_tc_destroy(struct tc *tc) | |
3878 | { | |
3879 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
93b13be8 | 3880 | struct htb_class *hc, *next; |
c1c9c9c4 | 3881 | |
4e8e4213 | 3882 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &htb->tc.queues) { |
93b13be8 | 3883 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 BP |
3884 | free(hc); |
3885 | } | |
3886 | tc_destroy(tc); | |
3887 | free(htb); | |
3888 | } | |
3889 | ||
3890 | static int | |
79f1cbe9 | 3891 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
3892 | { |
3893 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 3894 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
3895 | return 0; |
3896 | } | |
3897 | ||
3898 | static int | |
79f1cbe9 | 3899 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
3900 | { |
3901 | struct htb_class hc; | |
3902 | int error; | |
3903 | ||
3904 | htb_parse_qdisc_details__(netdev, details, &hc); | |
3905 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3906 | tc_make_handle(1, 0), &hc); | |
3907 | if (!error) { | |
3908 | htb_get__(netdev)->max_rate = hc.max_rate; | |
3909 | } | |
3910 | return error; | |
3911 | } | |
3912 | ||
3913 | static int | |
93b13be8 | 3914 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 3915 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 3916 | { |
93b13be8 | 3917 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 3918 | |
79f1cbe9 | 3919 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 3920 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 3921 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 3922 | } |
79f1cbe9 | 3923 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 3924 | if (hc->priority) { |
79f1cbe9 | 3925 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
3926 | } |
3927 | return 0; | |
3928 | } | |
3929 | ||
3930 | static int | |
3931 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 3932 | const struct smap *details) |
c1c9c9c4 BP |
3933 | { |
3934 | struct htb_class hc; | |
3935 | int error; | |
3936 | ||
3937 | error = htb_parse_class_details__(netdev, details, &hc); | |
3938 | if (error) { | |
3939 | return error; | |
3940 | } | |
3941 | ||
17ee3c1f | 3942 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
3943 | tc_make_handle(1, 0xfffe), &hc); |
3944 | if (error) { | |
3945 | return error; | |
3946 | } | |
3947 | ||
3948 | htb_update_queue__(netdev, queue_id, &hc); | |
3949 | return 0; | |
3950 | } | |
3951 | ||
3952 | static int | |
93b13be8 | 3953 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 3954 | { |
93b13be8 | 3955 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 3956 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
3957 | int error; |
3958 | ||
93b13be8 | 3959 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 3960 | if (!error) { |
93b13be8 | 3961 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 3962 | free(hc); |
c1c9c9c4 BP |
3963 | } |
3964 | return error; | |
3965 | } | |
3966 | ||
3967 | static int | |
93b13be8 | 3968 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
3969 | struct netdev_queue_stats *stats) |
3970 | { | |
93b13be8 | 3971 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
3972 | tc_make_handle(1, 0xfffe), NULL, stats); |
3973 | } | |
3974 | ||
3975 | static int | |
3976 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
3977 | const struct ofpbuf *nlmsg, | |
3978 | netdev_dump_queue_stats_cb *cb, void *aux) | |
3979 | { | |
3980 | struct netdev_queue_stats stats; | |
17ee3c1f | 3981 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
3982 | int error; |
3983 | ||
3984 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
3985 | if (error) { | |
3986 | return error; | |
3987 | } | |
3988 | ||
17ee3c1f BP |
3989 | major = tc_get_major(handle); |
3990 | minor = tc_get_minor(handle); | |
3991 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 3992 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
3993 | } |
3994 | return 0; | |
3995 | } | |
3996 | ||
3997 | static const struct tc_ops tc_ops_htb = { | |
3998 | "htb", /* linux_name */ | |
3999 | "linux-htb", /* ovs_name */ | |
4000 | HTB_N_QUEUES, /* n_queues */ | |
4001 | htb_tc_install, | |
4002 | htb_tc_load, | |
4003 | htb_tc_destroy, | |
4004 | htb_qdisc_get, | |
4005 | htb_qdisc_set, | |
4006 | htb_class_get, | |
4007 | htb_class_set, | |
4008 | htb_class_delete, | |
4009 | htb_class_get_stats, | |
4010 | htb_class_dump_stats | |
4011 | }; | |
4012 | \f | |
a339aa81 EJ |
4013 | /* "linux-hfsc" traffic control class. */ |
4014 | ||
4015 | #define HFSC_N_QUEUES 0xf000 | |
4016 | ||
4017 | struct hfsc { | |
4018 | struct tc tc; | |
4019 | uint32_t max_rate; | |
4020 | }; | |
4021 | ||
4022 | struct hfsc_class { | |
4023 | struct tc_queue tc_queue; | |
4024 | uint32_t min_rate; | |
4025 | uint32_t max_rate; | |
4026 | }; | |
4027 | ||
4028 | static struct hfsc * | |
b5d57fc8 | 4029 | hfsc_get__(const struct netdev *netdev_) |
a339aa81 | 4030 | { |
b5d57fc8 BP |
4031 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4032 | return CONTAINER_OF(netdev->tc, struct hfsc, tc); | |
a339aa81 EJ |
4033 | } |
4034 | ||
4035 | static struct hfsc_class * | |
4036 | hfsc_class_cast__(const struct tc_queue *queue) | |
4037 | { | |
4038 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
4039 | } | |
4040 | ||
24045e35 | 4041 | static void |
b5d57fc8 | 4042 | hfsc_install__(struct netdev *netdev_, uint32_t max_rate) |
a339aa81 | 4043 | { |
b5d57fc8 | 4044 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
4045 | struct hfsc *hfsc; |
4046 | ||
a339aa81 EJ |
4047 | hfsc = xmalloc(sizeof *hfsc); |
4048 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
4049 | hfsc->max_rate = max_rate; | |
b5d57fc8 | 4050 | netdev->tc = &hfsc->tc; |
a339aa81 EJ |
4051 | } |
4052 | ||
4053 | static void | |
4054 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4055 | const struct hfsc_class *hc) | |
4056 | { | |
4057 | size_t hash; | |
4058 | struct hfsc *hfsc; | |
4059 | struct hfsc_class *hcp; | |
4060 | struct tc_queue *queue; | |
4061 | ||
4062 | hfsc = hfsc_get__(netdev); | |
4063 | hash = hash_int(queue_id, 0); | |
4064 | ||
4065 | queue = tc_find_queue__(netdev, queue_id, hash); | |
4066 | if (queue) { | |
4067 | hcp = hfsc_class_cast__(queue); | |
4068 | } else { | |
4069 | hcp = xmalloc(sizeof *hcp); | |
4070 | queue = &hcp->tc_queue; | |
4071 | queue->queue_id = queue_id; | |
6dc34a0d | 4072 | queue->created = time_msec(); |
a339aa81 EJ |
4073 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); |
4074 | } | |
4075 | ||
4076 | hcp->min_rate = hc->min_rate; | |
4077 | hcp->max_rate = hc->max_rate; | |
4078 | } | |
4079 | ||
4080 | static int | |
4081 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
4082 | { | |
4083 | const struct tc_service_curve *rsc, *fsc, *usc; | |
4084 | static const struct nl_policy tca_hfsc_policy[] = { | |
4085 | [TCA_HFSC_RSC] = { | |
4086 | .type = NL_A_UNSPEC, | |
4087 | .optional = false, | |
4088 | .min_len = sizeof(struct tc_service_curve), | |
4089 | }, | |
4090 | [TCA_HFSC_FSC] = { | |
4091 | .type = NL_A_UNSPEC, | |
4092 | .optional = false, | |
4093 | .min_len = sizeof(struct tc_service_curve), | |
4094 | }, | |
4095 | [TCA_HFSC_USC] = { | |
4096 | .type = NL_A_UNSPEC, | |
4097 | .optional = false, | |
4098 | .min_len = sizeof(struct tc_service_curve), | |
4099 | }, | |
4100 | }; | |
4101 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
4102 | ||
4103 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
4104 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
4105 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
4106 | return EPROTO; | |
4107 | } | |
4108 | ||
4109 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
4110 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
4111 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
4112 | ||
4113 | if (rsc->m1 != 0 || rsc->d != 0 || | |
4114 | fsc->m1 != 0 || fsc->d != 0 || | |
4115 | usc->m1 != 0 || usc->d != 0) { | |
4116 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4117 | "Non-linear service curves are not supported."); | |
4118 | return EPROTO; | |
4119 | } | |
4120 | ||
4121 | if (rsc->m2 != fsc->m2) { | |
4122 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4123 | "Real-time service curves are not supported "); | |
4124 | return EPROTO; | |
4125 | } | |
4126 | ||
4127 | if (rsc->m2 > usc->m2) { | |
4128 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4129 | "Min-rate service curve is greater than " | |
4130 | "the max-rate service curve."); | |
4131 | return EPROTO; | |
4132 | } | |
4133 | ||
4134 | class->min_rate = fsc->m2; | |
4135 | class->max_rate = usc->m2; | |
4136 | return 0; | |
4137 | } | |
4138 | ||
4139 | static int | |
4140 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
4141 | struct hfsc_class *options, | |
4142 | struct netdev_queue_stats *stats) | |
4143 | { | |
4144 | int error; | |
4145 | unsigned int handle; | |
4146 | struct nlattr *nl_options; | |
4147 | ||
4148 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
4149 | if (error) { | |
4150 | return error; | |
4151 | } | |
4152 | ||
4153 | if (queue_id) { | |
4154 | unsigned int major, minor; | |
4155 | ||
4156 | major = tc_get_major(handle); | |
4157 | minor = tc_get_minor(handle); | |
4158 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4159 | *queue_id = minor - 1; | |
4160 | } else { | |
4161 | return EPROTO; | |
4162 | } | |
4163 | } | |
4164 | ||
4165 | if (options) { | |
4166 | error = hfsc_parse_tca_options__(nl_options, options); | |
4167 | } | |
4168 | ||
4169 | return error; | |
4170 | } | |
4171 | ||
4172 | static int | |
4173 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
4174 | unsigned int parent, struct hfsc_class *options, | |
4175 | struct netdev_queue_stats *stats) | |
4176 | { | |
4177 | int error; | |
4178 | struct ofpbuf *reply; | |
4179 | ||
4180 | error = tc_query_class(netdev, handle, parent, &reply); | |
4181 | if (error) { | |
4182 | return error; | |
4183 | } | |
4184 | ||
4185 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
4186 | ofpbuf_delete(reply); | |
4187 | return error; | |
4188 | } | |
4189 | ||
4190 | static void | |
73371c09 | 4191 | hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, |
a339aa81 EJ |
4192 | struct hfsc_class *class) |
4193 | { | |
73371c09 | 4194 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
4195 | uint32_t max_rate; |
4196 | const char *max_rate_s; | |
4197 | ||
79f1cbe9 | 4198 | max_rate_s = smap_get(details, "max-rate"); |
a339aa81 EJ |
4199 | max_rate = max_rate_s ? strtoull(max_rate_s, NULL, 10) / 8 : 0; |
4200 | ||
4201 | if (!max_rate) { | |
a00ca915 | 4202 | enum netdev_features current; |
a339aa81 | 4203 | |
73371c09 BP |
4204 | netdev_linux_read_features(netdev); |
4205 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 4206 | max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
a339aa81 EJ |
4207 | } |
4208 | ||
4209 | class->min_rate = max_rate; | |
4210 | class->max_rate = max_rate; | |
4211 | } | |
4212 | ||
4213 | static int | |
4214 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 4215 | const struct smap *details, |
a339aa81 EJ |
4216 | struct hfsc_class * class) |
4217 | { | |
4218 | const struct hfsc *hfsc; | |
4219 | uint32_t min_rate, max_rate; | |
4220 | const char *min_rate_s, *max_rate_s; | |
4221 | ||
4222 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 EJ |
4223 | min_rate_s = smap_get(details, "min-rate"); |
4224 | max_rate_s = smap_get(details, "max-rate"); | |
a339aa81 | 4225 | |
c45ab5e9 | 4226 | min_rate = min_rate_s ? strtoull(min_rate_s, NULL, 10) / 8 : 0; |
79398bad | 4227 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
4228 | min_rate = MIN(min_rate, hfsc->max_rate); |
4229 | ||
4230 | max_rate = (max_rate_s | |
4231 | ? strtoull(max_rate_s, NULL, 10) / 8 | |
4232 | : hfsc->max_rate); | |
4233 | max_rate = MAX(max_rate, min_rate); | |
4234 | max_rate = MIN(max_rate, hfsc->max_rate); | |
4235 | ||
4236 | class->min_rate = min_rate; | |
4237 | class->max_rate = max_rate; | |
4238 | ||
4239 | return 0; | |
4240 | } | |
4241 | ||
4242 | /* Create an HFSC qdisc. | |
4243 | * | |
4244 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
4245 | static int | |
4246 | hfsc_setup_qdisc__(struct netdev * netdev) | |
4247 | { | |
4248 | struct tcmsg *tcmsg; | |
4249 | struct ofpbuf request; | |
4250 | struct tc_hfsc_qopt opt; | |
4251 | ||
4252 | tc_del_qdisc(netdev); | |
4253 | ||
4254 | tcmsg = tc_make_request(netdev, RTM_NEWQDISC, | |
4255 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
4256 | ||
4257 | if (!tcmsg) { | |
4258 | return ENODEV; | |
4259 | } | |
4260 | ||
4261 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
4262 | tcmsg->tcm_parent = TC_H_ROOT; | |
4263 | ||
4264 | memset(&opt, 0, sizeof opt); | |
4265 | opt.defcls = 1; | |
4266 | ||
4267 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4268 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
4269 | ||
4270 | return tc_transact(&request, NULL); | |
4271 | } | |
4272 | ||
4273 | /* Create an HFSC class. | |
4274 | * | |
4275 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
4276 | * sc rate <min_rate> ul rate <max_rate>" */ | |
4277 | static int | |
4278 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
4279 | unsigned int parent, struct hfsc_class *class) | |
4280 | { | |
4281 | int error; | |
4282 | size_t opt_offset; | |
4283 | struct tcmsg *tcmsg; | |
4284 | struct ofpbuf request; | |
4285 | struct tc_service_curve min, max; | |
4286 | ||
4287 | tcmsg = tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, &request); | |
4288 | ||
4289 | if (!tcmsg) { | |
4290 | return ENODEV; | |
4291 | } | |
4292 | ||
4293 | tcmsg->tcm_handle = handle; | |
4294 | tcmsg->tcm_parent = parent; | |
4295 | ||
4296 | min.m1 = 0; | |
4297 | min.d = 0; | |
4298 | min.m2 = class->min_rate; | |
4299 | ||
4300 | max.m1 = 0; | |
4301 | max.d = 0; | |
4302 | max.m2 = class->max_rate; | |
4303 | ||
4304 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4305 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4306 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
4307 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
4308 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
4309 | nl_msg_end_nested(&request, opt_offset); | |
4310 | ||
4311 | error = tc_transact(&request, NULL); | |
4312 | if (error) { | |
4313 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
4314 | "min-rate %ubps, max-rate %ubps (%s)", | |
4315 | netdev_get_name(netdev), | |
4316 | tc_get_major(handle), tc_get_minor(handle), | |
4317 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 4318 | class->min_rate, class->max_rate, ovs_strerror(error)); |
a339aa81 EJ |
4319 | } |
4320 | ||
4321 | return error; | |
4322 | } | |
4323 | ||
4324 | static int | |
79f1cbe9 | 4325 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4326 | { |
4327 | int error; | |
4328 | struct hfsc_class class; | |
4329 | ||
4330 | error = hfsc_setup_qdisc__(netdev); | |
4331 | ||
4332 | if (error) { | |
4333 | return error; | |
4334 | } | |
4335 | ||
4336 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4337 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4338 | tc_make_handle(1, 0), &class); | |
4339 | ||
4340 | if (error) { | |
4341 | return error; | |
4342 | } | |
4343 | ||
4344 | hfsc_install__(netdev, class.max_rate); | |
4345 | return 0; | |
4346 | } | |
4347 | ||
4348 | static int | |
4349 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4350 | { | |
4351 | struct ofpbuf msg; | |
d57695d7 | 4352 | struct queue_dump_state state; |
a339aa81 EJ |
4353 | struct hfsc_class hc; |
4354 | ||
4355 | hc.max_rate = 0; | |
4356 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 4357 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 | 4358 | |
d57695d7 | 4359 | if (!start_queue_dump(netdev, &state)) { |
a339aa81 EJ |
4360 | return ENODEV; |
4361 | } | |
4362 | ||
d57695d7 | 4363 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
a339aa81 EJ |
4364 | unsigned int queue_id; |
4365 | ||
4366 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
4367 | hfsc_update_queue__(netdev, queue_id, &hc); | |
4368 | } | |
4369 | } | |
4370 | ||
d57695d7 | 4371 | finish_queue_dump(&state); |
a339aa81 EJ |
4372 | return 0; |
4373 | } | |
4374 | ||
4375 | static void | |
4376 | hfsc_tc_destroy(struct tc *tc) | |
4377 | { | |
4378 | struct hfsc *hfsc; | |
4379 | struct hfsc_class *hc, *next; | |
4380 | ||
4381 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
4382 | ||
4383 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
4384 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4385 | free(hc); | |
4386 | } | |
4387 | ||
4388 | tc_destroy(tc); | |
4389 | free(hfsc); | |
4390 | } | |
4391 | ||
4392 | static int | |
79f1cbe9 | 4393 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
4394 | { |
4395 | const struct hfsc *hfsc; | |
4396 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 4397 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
4398 | return 0; |
4399 | } | |
4400 | ||
4401 | static int | |
79f1cbe9 | 4402 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4403 | { |
4404 | int error; | |
4405 | struct hfsc_class class; | |
4406 | ||
4407 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4408 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4409 | tc_make_handle(1, 0), &class); | |
4410 | ||
4411 | if (!error) { | |
4412 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
4413 | } | |
4414 | ||
4415 | return error; | |
4416 | } | |
4417 | ||
4418 | static int | |
4419 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 4420 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
4421 | { |
4422 | const struct hfsc_class *hc; | |
4423 | ||
4424 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 4425 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 4426 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4427 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
4428 | } |
4429 | return 0; | |
4430 | } | |
4431 | ||
4432 | static int | |
4433 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4434 | const struct smap *details) |
a339aa81 EJ |
4435 | { |
4436 | int error; | |
4437 | struct hfsc_class class; | |
4438 | ||
4439 | error = hfsc_parse_class_details__(netdev, details, &class); | |
4440 | if (error) { | |
4441 | return error; | |
4442 | } | |
4443 | ||
4444 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
4445 | tc_make_handle(1, 0xfffe), &class); | |
4446 | if (error) { | |
4447 | return error; | |
4448 | } | |
4449 | ||
4450 | hfsc_update_queue__(netdev, queue_id, &class); | |
4451 | return 0; | |
4452 | } | |
4453 | ||
4454 | static int | |
4455 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
4456 | { | |
4457 | int error; | |
4458 | struct hfsc *hfsc; | |
4459 | struct hfsc_class *hc; | |
4460 | ||
4461 | hc = hfsc_class_cast__(queue); | |
4462 | hfsc = hfsc_get__(netdev); | |
4463 | ||
4464 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
4465 | if (!error) { | |
4466 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4467 | free(hc); | |
4468 | } | |
4469 | return error; | |
4470 | } | |
4471 | ||
4472 | static int | |
4473 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
4474 | struct netdev_queue_stats *stats) | |
4475 | { | |
4476 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
4477 | tc_make_handle(1, 0xfffe), NULL, stats); | |
4478 | } | |
4479 | ||
4480 | static int | |
4481 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4482 | const struct ofpbuf *nlmsg, | |
4483 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4484 | { | |
4485 | struct netdev_queue_stats stats; | |
4486 | unsigned int handle, major, minor; | |
4487 | int error; | |
4488 | ||
4489 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4490 | if (error) { | |
4491 | return error; | |
4492 | } | |
4493 | ||
4494 | major = tc_get_major(handle); | |
4495 | minor = tc_get_minor(handle); | |
4496 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4497 | (*cb)(minor - 1, &stats, aux); | |
4498 | } | |
4499 | return 0; | |
4500 | } | |
4501 | ||
4502 | static const struct tc_ops tc_ops_hfsc = { | |
4503 | "hfsc", /* linux_name */ | |
4504 | "linux-hfsc", /* ovs_name */ | |
4505 | HFSC_N_QUEUES, /* n_queues */ | |
4506 | hfsc_tc_install, /* tc_install */ | |
4507 | hfsc_tc_load, /* tc_load */ | |
4508 | hfsc_tc_destroy, /* tc_destroy */ | |
4509 | hfsc_qdisc_get, /* qdisc_get */ | |
4510 | hfsc_qdisc_set, /* qdisc_set */ | |
4511 | hfsc_class_get, /* class_get */ | |
4512 | hfsc_class_set, /* class_set */ | |
4513 | hfsc_class_delete, /* class_delete */ | |
4514 | hfsc_class_get_stats, /* class_get_stats */ | |
4515 | hfsc_class_dump_stats /* class_dump_stats */ | |
4516 | }; | |
4517 | \f | |
c1c9c9c4 BP |
4518 | /* "linux-default" traffic control class. |
4519 | * | |
4520 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
4521 | * the "" (empty string) QoS type in the OVS database. */ | |
4522 | ||
4523 | static void | |
b5d57fc8 | 4524 | default_install__(struct netdev *netdev_) |
c1c9c9c4 | 4525 | { |
b5d57fc8 | 4526 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4527 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); |
c1c9c9c4 | 4528 | |
559eb230 BP |
4529 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4530 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4531 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4532 | } |
4533 | ||
4534 | static int | |
4535 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 4536 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
4537 | { |
4538 | default_install__(netdev); | |
4539 | return 0; | |
4540 | } | |
4541 | ||
4542 | static int | |
4543 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4544 | { | |
4545 | default_install__(netdev); | |
4546 | return 0; | |
4547 | } | |
4548 | ||
4549 | static const struct tc_ops tc_ops_default = { | |
4550 | NULL, /* linux_name */ | |
4551 | "", /* ovs_name */ | |
4552 | 0, /* n_queues */ | |
4553 | default_tc_install, | |
4554 | default_tc_load, | |
4555 | NULL, /* tc_destroy */ | |
4556 | NULL, /* qdisc_get */ | |
4557 | NULL, /* qdisc_set */ | |
4558 | NULL, /* class_get */ | |
4559 | NULL, /* class_set */ | |
4560 | NULL, /* class_delete */ | |
4561 | NULL, /* class_get_stats */ | |
4562 | NULL /* class_dump_stats */ | |
4563 | }; | |
4564 | \f | |
4565 | /* "linux-other" traffic control class. | |
4566 | * | |
4567 | * */ | |
4568 | ||
4569 | static int | |
b5d57fc8 | 4570 | other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) |
c1c9c9c4 | 4571 | { |
b5d57fc8 | 4572 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4573 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); |
c1c9c9c4 | 4574 | |
559eb230 BP |
4575 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4576 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4577 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4578 | return 0; |
4579 | } | |
4580 | ||
4581 | static const struct tc_ops tc_ops_other = { | |
4582 | NULL, /* linux_name */ | |
4583 | "linux-other", /* ovs_name */ | |
4584 | 0, /* n_queues */ | |
4585 | NULL, /* tc_install */ | |
4586 | other_tc_load, | |
4587 | NULL, /* tc_destroy */ | |
4588 | NULL, /* qdisc_get */ | |
4589 | NULL, /* qdisc_set */ | |
4590 | NULL, /* class_get */ | |
4591 | NULL, /* class_set */ | |
4592 | NULL, /* class_delete */ | |
4593 | NULL, /* class_get_stats */ | |
4594 | NULL /* class_dump_stats */ | |
4595 | }; | |
4596 | \f | |
4597 | /* Traffic control. */ | |
4598 | ||
4599 | /* Number of kernel "tc" ticks per second. */ | |
4600 | static double ticks_per_s; | |
4601 | ||
4602 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
4603 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
4604 | * one jiffy's worth of data. | |
4605 | * | |
4606 | * There are two possibilities here: | |
4607 | * | |
4608 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
4609 | * approximate range of 100 to 1024. That means that we really need to | |
4610 | * make sure that the qdisc can buffer that much data. | |
4611 | * | |
4612 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
4613 | * has finely granular timers and there's no need to fudge additional room | |
4614 | * for buffers. (There's no extra effort needed to implement that: the | |
4615 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
4616 | * come out as 0 in the division. Small integer results in the case of | |
4617 | * really high dividends won't have any real effect anyhow.) | |
4618 | */ | |
4619 | static unsigned int buffer_hz; | |
4620 | ||
4621 | /* Returns tc handle 'major':'minor'. */ | |
4622 | static unsigned int | |
4623 | tc_make_handle(unsigned int major, unsigned int minor) | |
4624 | { | |
4625 | return TC_H_MAKE(major << 16, minor); | |
4626 | } | |
4627 | ||
4628 | /* Returns the major number from 'handle'. */ | |
4629 | static unsigned int | |
4630 | tc_get_major(unsigned int handle) | |
4631 | { | |
4632 | return TC_H_MAJ(handle) >> 16; | |
4633 | } | |
4634 | ||
4635 | /* Returns the minor number from 'handle'. */ | |
4636 | static unsigned int | |
4637 | tc_get_minor(unsigned int handle) | |
4638 | { | |
4639 | return TC_H_MIN(handle); | |
4640 | } | |
4641 | ||
4642 | static struct tcmsg * | |
4643 | tc_make_request(const struct netdev *netdev, int type, unsigned int flags, | |
4644 | struct ofpbuf *request) | |
4645 | { | |
4646 | struct tcmsg *tcmsg; | |
4647 | int ifindex; | |
4648 | int error; | |
4649 | ||
4650 | error = get_ifindex(netdev, &ifindex); | |
4651 | if (error) { | |
4652 | return NULL; | |
4653 | } | |
4654 | ||
4655 | ofpbuf_init(request, 512); | |
4656 | nl_msg_put_nlmsghdr(request, sizeof *tcmsg, type, NLM_F_REQUEST | flags); | |
4657 | tcmsg = ofpbuf_put_zeros(request, sizeof *tcmsg); | |
4658 | tcmsg->tcm_family = AF_UNSPEC; | |
4659 | tcmsg->tcm_ifindex = ifindex; | |
4660 | /* Caller should fill in tcmsg->tcm_handle. */ | |
4661 | /* Caller should fill in tcmsg->tcm_parent. */ | |
4662 | ||
4663 | return tcmsg; | |
4664 | } | |
4665 | ||
4666 | static int | |
4667 | tc_transact(struct ofpbuf *request, struct ofpbuf **replyp) | |
4668 | { | |
a88b4e04 | 4669 | int error = nl_transact(NETLINK_ROUTE, request, replyp); |
c1c9c9c4 BP |
4670 | ofpbuf_uninit(request); |
4671 | return error; | |
4672 | } | |
4673 | ||
f8500004 JP |
4674 | /* Adds or deletes a root ingress qdisc on 'netdev'. We use this for |
4675 | * policing configuration. | |
4676 | * | |
4677 | * This function is equivalent to running the following when 'add' is true: | |
4678 | * /sbin/tc qdisc add dev <devname> handle ffff: ingress | |
4679 | * | |
4680 | * This function is equivalent to running the following when 'add' is false: | |
4681 | * /sbin/tc qdisc del dev <devname> handle ffff: ingress | |
4682 | * | |
4683 | * The configuration and stats may be seen with the following command: | |
4684 | * /sbin/tc -s qdisc show dev <devname> | |
4685 | * | |
4686 | * Returns 0 if successful, otherwise a positive errno value. | |
4687 | */ | |
4688 | static int | |
4689 | tc_add_del_ingress_qdisc(struct netdev *netdev, bool add) | |
4690 | { | |
4691 | struct ofpbuf request; | |
4692 | struct tcmsg *tcmsg; | |
4693 | int error; | |
4694 | int type = add ? RTM_NEWQDISC : RTM_DELQDISC; | |
4695 | int flags = add ? NLM_F_EXCL | NLM_F_CREATE : 0; | |
4696 | ||
4697 | tcmsg = tc_make_request(netdev, type, flags, &request); | |
4698 | if (!tcmsg) { | |
4699 | return ENODEV; | |
4700 | } | |
4701 | tcmsg->tcm_handle = tc_make_handle(0xffff, 0); | |
4702 | tcmsg->tcm_parent = TC_H_INGRESS; | |
4703 | nl_msg_put_string(&request, TCA_KIND, "ingress"); | |
4704 | nl_msg_put_unspec(&request, TCA_OPTIONS, NULL, 0); | |
4705 | ||
4706 | error = tc_transact(&request, NULL); | |
4707 | if (error) { | |
4708 | /* If we're deleting the qdisc, don't worry about some of the | |
4709 | * error conditions. */ | |
4710 | if (!add && (error == ENOENT || error == EINVAL)) { | |
4711 | return 0; | |
4712 | } | |
4713 | return error; | |
4714 | } | |
4715 | ||
4716 | return 0; | |
4717 | } | |
4718 | ||
4719 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size | |
4720 | * of 'kbits_burst'. | |
4721 | * | |
4722 | * This function is equivalent to running: | |
4723 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
4724 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
4725 | * mtu 65535 drop | |
4726 | * | |
4727 | * The configuration and stats may be seen with the following command: | |
c7952afb | 4728 | * /sbin/tc -s filter show dev <devname> parent ffff: |
f8500004 JP |
4729 | * |
4730 | * Returns 0 if successful, otherwise a positive errno value. | |
4731 | */ | |
4732 | static int | |
c7952afb BP |
4733 | tc_add_policer(struct netdev *netdev, |
4734 | uint32_t kbits_rate, uint32_t kbits_burst) | |
f8500004 JP |
4735 | { |
4736 | struct tc_police tc_police; | |
4737 | struct ofpbuf request; | |
4738 | struct tcmsg *tcmsg; | |
4739 | size_t basic_offset; | |
4740 | size_t police_offset; | |
4741 | int error; | |
4742 | int mtu = 65535; | |
4743 | ||
4744 | memset(&tc_police, 0, sizeof tc_police); | |
4745 | tc_police.action = TC_POLICE_SHOT; | |
4746 | tc_police.mtu = mtu; | |
1aca400c | 4747 | tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu); |
c7952afb BP |
4748 | |
4749 | /* The following appears wrong in two ways: | |
4750 | * | |
4751 | * - tc_bytes_to_ticks() should take "bytes" as quantity for both of its | |
4752 | * arguments (or at least consistently "bytes" as both or "bits" as | |
4753 | * both), but this supplies bytes for the first argument and bits for the | |
4754 | * second. | |
4755 | * | |
4756 | * - In networking a kilobit is usually 1000 bits but this uses 1024 bits. | |
4757 | * | |
4758 | * However if you "fix" those problems then "tc filter show ..." shows | |
4759 | * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit == | |
4760 | * 1,000,000 bits, whereas this actually ends up doing the right thing from | |
4761 | * tc's point of view. Whatever. */ | |
4762 | tc_police.burst = tc_bytes_to_ticks( | |
4763 | tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024); | |
f8500004 JP |
4764 | |
4765 | tcmsg = tc_make_request(netdev, RTM_NEWTFILTER, | |
4766 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
4767 | if (!tcmsg) { | |
4768 | return ENODEV; | |
4769 | } | |
4770 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
4771 | tcmsg->tcm_info = tc_make_handle(49, | |
4772 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
4773 | ||
4774 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
4775 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4776 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
4777 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
4778 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
4779 | nl_msg_end_nested(&request, police_offset); | |
4780 | nl_msg_end_nested(&request, basic_offset); | |
4781 | ||
4782 | error = tc_transact(&request, NULL); | |
4783 | if (error) { | |
4784 | return error; | |
4785 | } | |
4786 | ||
4787 | return 0; | |
4788 | } | |
4789 | ||
c1c9c9c4 BP |
4790 | static void |
4791 | read_psched(void) | |
4792 | { | |
4793 | /* The values in psched are not individually very meaningful, but they are | |
4794 | * important. The tables below show some values seen in the wild. | |
4795 | * | |
4796 | * Some notes: | |
4797 | * | |
4798 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
4799 | * (Before that, there are hints that it was 1000000000.) | |
4800 | * | |
4801 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
4802 | * above. | |
4803 | * | |
4804 | * /proc/net/psched | |
4805 | * ----------------------------------- | |
4806 | * [1] 000c8000 000f4240 000f4240 00000064 | |
4807 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
4808 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
4809 | * [4] 000003e8 00000400 000f4240 00000064 | |
4810 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
4811 | * [6] 000003e8 00000040 000f4240 000000f9 | |
4812 | * | |
4813 | * a b c d ticks_per_s buffer_hz | |
4814 | * ------- --------- ---------- ------------- ----------- ------------- | |
4815 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
4816 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
4817 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
4818 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
4819 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
4820 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
4821 | * | |
4822 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
4823 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
4824 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
4825 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
4826 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
4827 | * [6] 2.6.34 from kernel.org on KVM | |
4828 | */ | |
23882115 | 4829 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
c1c9c9c4 BP |
4830 | static const char fn[] = "/proc/net/psched"; |
4831 | unsigned int a, b, c, d; | |
4832 | FILE *stream; | |
4833 | ||
23882115 BP |
4834 | if (!ovsthread_once_start(&once)) { |
4835 | return; | |
4836 | } | |
4837 | ||
c1c9c9c4 BP |
4838 | ticks_per_s = 1.0; |
4839 | buffer_hz = 100; | |
4840 | ||
4841 | stream = fopen(fn, "r"); | |
4842 | if (!stream) { | |
10a89ef0 | 4843 | VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); |
23882115 | 4844 | goto exit; |
c1c9c9c4 BP |
4845 | } |
4846 | ||
4847 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
4848 | VLOG_WARN("%s: read failed", fn); | |
4849 | fclose(stream); | |
23882115 | 4850 | goto exit; |
c1c9c9c4 BP |
4851 | } |
4852 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
4853 | fclose(stream); | |
4854 | ||
4855 | if (!a || !c) { | |
4856 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
23882115 | 4857 | goto exit; |
c1c9c9c4 BP |
4858 | } |
4859 | ||
4860 | ticks_per_s = (double) a * c / b; | |
4861 | if (c == 1000000) { | |
4862 | buffer_hz = d; | |
4863 | } else { | |
4864 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
4865 | fn, a, b, c, d); | |
4866 | } | |
4867 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
23882115 BP |
4868 | |
4869 | exit: | |
4870 | ovsthread_once_done(&once); | |
c1c9c9c4 BP |
4871 | } |
4872 | ||
4873 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
4874 | * rate of 'rate' bytes per second. */ | |
4875 | static unsigned int | |
4876 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
4877 | { | |
23882115 | 4878 | read_psched(); |
c1c9c9c4 BP |
4879 | return (rate * ticks) / ticks_per_s; |
4880 | } | |
4881 | ||
4882 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
4883 | * rate of 'rate' bytes per second. */ | |
4884 | static unsigned int | |
4885 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
4886 | { | |
23882115 | 4887 | read_psched(); |
015c93a4 | 4888 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
4889 | } |
4890 | ||
4891 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
4892 | * a transmission rate of 'rate' bytes per second. */ | |
4893 | static unsigned int | |
4894 | tc_buffer_per_jiffy(unsigned int rate) | |
4895 | { | |
23882115 | 4896 | read_psched(); |
c1c9c9c4 BP |
4897 | return rate / buffer_hz; |
4898 | } | |
4899 | ||
4900 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
4901 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
4902 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
4903 | * stores NULL into it if it is absent. | |
4904 | * | |
4905 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
4906 | * 'msg'. | |
4907 | * | |
4908 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4909 | static int | |
4910 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
4911 | struct nlattr **options) | |
4912 | { | |
4913 | static const struct nl_policy tca_policy[] = { | |
4914 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
4915 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
4916 | }; | |
4917 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
4918 | ||
4919 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
4920 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
4921 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
4922 | goto error; | |
4923 | } | |
4924 | ||
4925 | if (kind) { | |
4926 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
4927 | } | |
4928 | ||
4929 | if (options) { | |
4930 | *options = ta[TCA_OPTIONS]; | |
4931 | } | |
4932 | ||
4933 | return 0; | |
4934 | ||
4935 | error: | |
4936 | if (kind) { | |
4937 | *kind = NULL; | |
4938 | } | |
4939 | if (options) { | |
4940 | *options = NULL; | |
4941 | } | |
4942 | return EPROTO; | |
4943 | } | |
4944 | ||
4945 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
4946 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
4947 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
4948 | * arguments may be null. | |
4949 | * | |
4950 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4951 | static int | |
4952 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
4953 | struct nlattr **options, struct netdev_queue_stats *stats) | |
4954 | { | |
4955 | static const struct nl_policy tca_policy[] = { | |
4956 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
4957 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
4958 | }; | |
4959 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
4960 | ||
4961 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
4962 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
4963 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
4964 | goto error; | |
4965 | } | |
4966 | ||
4967 | if (handlep) { | |
4968 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
4969 | *handlep = tc->tcm_handle; | |
4970 | } | |
4971 | ||
4972 | if (options) { | |
4973 | *options = ta[TCA_OPTIONS]; | |
4974 | } | |
4975 | ||
4976 | if (stats) { | |
4977 | const struct gnet_stats_queue *gsq; | |
4978 | struct gnet_stats_basic gsb; | |
4979 | ||
4980 | static const struct nl_policy stats_policy[] = { | |
4981 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
4982 | .min_len = sizeof gsb }, | |
4983 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
4984 | .min_len = sizeof *gsq }, | |
4985 | }; | |
4986 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
4987 | ||
4988 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
4989 | sa, ARRAY_SIZE(sa))) { | |
4990 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
4991 | goto error; | |
4992 | } | |
4993 | ||
4994 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
4995 | * some arch/bitsize combinations. Newer versions of Linux have a | |
4996 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
4997 | * easiest thing to do is just to make a copy. */ | |
4998 | memset(&gsb, 0, sizeof gsb); | |
4999 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
5000 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
5001 | stats->tx_bytes = gsb.bytes; | |
5002 | stats->tx_packets = gsb.packets; | |
5003 | ||
5004 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
5005 | stats->tx_errors = gsq->drops; | |
5006 | } | |
5007 | ||
5008 | return 0; | |
5009 | ||
5010 | error: | |
5011 | if (options) { | |
5012 | *options = NULL; | |
5013 | } | |
5014 | if (stats) { | |
5015 | memset(stats, 0, sizeof *stats); | |
5016 | } | |
5017 | return EPROTO; | |
5018 | } | |
5019 | ||
5020 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
5021 | * on 'netdev'. */ | |
5022 | static int | |
5023 | tc_query_class(const struct netdev *netdev, | |
5024 | unsigned int handle, unsigned int parent, | |
5025 | struct ofpbuf **replyp) | |
5026 | { | |
5027 | struct ofpbuf request; | |
5028 | struct tcmsg *tcmsg; | |
5029 | int error; | |
5030 | ||
5031 | tcmsg = tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, &request); | |
23a98ffe BP |
5032 | if (!tcmsg) { |
5033 | return ENODEV; | |
5034 | } | |
c1c9c9c4 BP |
5035 | tcmsg->tcm_handle = handle; |
5036 | tcmsg->tcm_parent = parent; | |
5037 | ||
5038 | error = tc_transact(&request, replyp); | |
5039 | if (error) { | |
5040 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
5041 | netdev_get_name(netdev), | |
5042 | tc_get_major(handle), tc_get_minor(handle), | |
5043 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 5044 | ovs_strerror(error)); |
c1c9c9c4 BP |
5045 | } |
5046 | return error; | |
5047 | } | |
5048 | ||
5049 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
5050 | static int | |
5051 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
5052 | { | |
5053 | struct ofpbuf request; | |
5054 | struct tcmsg *tcmsg; | |
5055 | int error; | |
5056 | ||
5057 | tcmsg = tc_make_request(netdev, RTM_DELTCLASS, 0, &request); | |
23a98ffe BP |
5058 | if (!tcmsg) { |
5059 | return ENODEV; | |
5060 | } | |
c1c9c9c4 BP |
5061 | tcmsg->tcm_handle = handle; |
5062 | tcmsg->tcm_parent = 0; | |
5063 | ||
5064 | error = tc_transact(&request, NULL); | |
5065 | if (error) { | |
5066 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
5067 | netdev_get_name(netdev), | |
5068 | tc_get_major(handle), tc_get_minor(handle), | |
10a89ef0 | 5069 | ovs_strerror(error)); |
c1c9c9c4 BP |
5070 | } |
5071 | return error; | |
5072 | } | |
5073 | ||
5074 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
5075 | static int | |
b5d57fc8 | 5076 | tc_del_qdisc(struct netdev *netdev_) |
c1c9c9c4 | 5077 | { |
b5d57fc8 | 5078 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5079 | struct ofpbuf request; |
5080 | struct tcmsg *tcmsg; | |
5081 | int error; | |
5082 | ||
b5d57fc8 | 5083 | tcmsg = tc_make_request(netdev_, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
5084 | if (!tcmsg) { |
5085 | return ENODEV; | |
5086 | } | |
c1c9c9c4 BP |
5087 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
5088 | tcmsg->tcm_parent = TC_H_ROOT; | |
5089 | ||
5090 | error = tc_transact(&request, NULL); | |
5091 | if (error == EINVAL) { | |
5092 | /* EINVAL probably means that the default qdisc was in use, in which | |
5093 | * case we've accomplished our purpose. */ | |
5094 | error = 0; | |
5095 | } | |
b5d57fc8 BP |
5096 | if (!error && netdev->tc) { |
5097 | if (netdev->tc->ops->tc_destroy) { | |
5098 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 | 5099 | } |
b5d57fc8 | 5100 | netdev->tc = NULL; |
c1c9c9c4 BP |
5101 | } |
5102 | return error; | |
5103 | } | |
5104 | ||
ac3e3aaa BP |
5105 | static bool |
5106 | getqdisc_is_safe(void) | |
5107 | { | |
5108 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
5109 | static bool safe = false; | |
5110 | ||
5111 | if (ovsthread_once_start(&once)) { | |
5112 | struct utsname utsname; | |
5113 | int major, minor; | |
5114 | ||
5115 | if (uname(&utsname) == -1) { | |
5116 | VLOG_WARN("uname failed (%s)", ovs_strerror(errno)); | |
5117 | } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) { | |
5118 | VLOG_WARN("uname reported bad OS release (%s)", utsname.release); | |
5119 | } else if (major < 2 || (major == 2 && minor < 35)) { | |
5120 | VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s", | |
5121 | utsname.release); | |
5122 | } else { | |
5123 | safe = true; | |
5124 | } | |
5125 | ovsthread_once_done(&once); | |
5126 | } | |
5127 | return safe; | |
5128 | } | |
5129 | ||
c1c9c9c4 BP |
5130 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the |
5131 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
5132 | * positive errno value. */ | |
5133 | static int | |
b5d57fc8 | 5134 | tc_query_qdisc(const struct netdev *netdev_) |
c1c9c9c4 | 5135 | { |
b5d57fc8 | 5136 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5137 | struct ofpbuf request, *qdisc; |
5138 | const struct tc_ops *ops; | |
5139 | struct tcmsg *tcmsg; | |
5140 | int load_error; | |
5141 | int error; | |
5142 | ||
b5d57fc8 | 5143 | if (netdev->tc) { |
c1c9c9c4 BP |
5144 | return 0; |
5145 | } | |
5146 | ||
5147 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
5148 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
5149 | * 2.6.35 without that fix backported to it. | |
5150 | * | |
5151 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
5152 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
5153 | * few others. There are a few ways that I can see to do this, but most of | |
5154 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
5155 | * technique chosen here is to assume that any non-default qdisc that we | |
5156 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
5157 | * a class with handle 0:0. | |
5158 | * | |
ac3e3aaa BP |
5159 | * On Linux 2.6.35+ we use the straightforward method because it allows us |
5160 | * to handle non-builtin qdiscs without handle 1:0 (e.g. codel). However, | |
5161 | * in such a case we get no response at all from the kernel (!) if a | |
5162 | * builtin qdisc is in use (which is later caught by "!error && | |
5163 | * !qdisc->size"). */ | |
b5d57fc8 | 5164 | tcmsg = tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, &request); |
23a98ffe BP |
5165 | if (!tcmsg) { |
5166 | return ENODEV; | |
5167 | } | |
ac3e3aaa BP |
5168 | tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0); |
5169 | tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0; | |
c1c9c9c4 BP |
5170 | |
5171 | /* Figure out what tc class to instantiate. */ | |
5172 | error = tc_transact(&request, &qdisc); | |
ac3e3aaa | 5173 | if (!error && qdisc->size) { |
c1c9c9c4 BP |
5174 | const char *kind; |
5175 | ||
5176 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
5177 | if (error) { | |
5178 | ops = &tc_ops_other; | |
5179 | } else { | |
5180 | ops = tc_lookup_linux_name(kind); | |
5181 | if (!ops) { | |
5182 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
ac3e3aaa | 5183 | VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind); |
c1c9c9c4 BP |
5184 | |
5185 | ops = &tc_ops_other; | |
5186 | } | |
5187 | } | |
ac3e3aaa BP |
5188 | } else if ((!error && !qdisc->size) || error == ENOENT) { |
5189 | /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc | |
5190 | * set up by some other entity that doesn't have a handle 1:0. We will | |
5191 | * assume that it's the system default qdisc. */ | |
c1c9c9c4 BP |
5192 | ops = &tc_ops_default; |
5193 | error = 0; | |
5194 | } else { | |
5195 | /* Who knows? Maybe the device got deleted. */ | |
5196 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
10a89ef0 | 5197 | netdev_get_name(netdev_), ovs_strerror(error)); |
c1c9c9c4 BP |
5198 | ops = &tc_ops_other; |
5199 | } | |
5200 | ||
5201 | /* Instantiate it. */ | |
b5d57fc8 BP |
5202 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); |
5203 | ovs_assert((load_error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
5204 | ofpbuf_delete(qdisc); |
5205 | ||
5206 | return error ? error : load_error; | |
5207 | } | |
5208 | ||
5209 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
5210 | approximate the time to transmit packets of various lengths. For an MTU of | |
5211 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
5212 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
5213 | possible lengths; and so on. | |
5214 | ||
5215 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
5216 | need to be shifted right to fit within such a 256-entry table. */ | |
5217 | static int | |
5218 | tc_calc_cell_log(unsigned int mtu) | |
5219 | { | |
5220 | int cell_log; | |
5221 | ||
5222 | if (!mtu) { | |
5223 | mtu = ETH_PAYLOAD_MAX; | |
5224 | } | |
5225 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
5226 | ||
5227 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
5228 | mtu >>= 1; | |
5229 | } | |
5230 | ||
5231 | return cell_log; | |
5232 | } | |
5233 | ||
5234 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
5235 | * of 'mtu'. */ | |
5236 | static void | |
5237 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
5238 | { | |
5239 | memset(rate, 0, sizeof *rate); | |
5240 | rate->cell_log = tc_calc_cell_log(mtu); | |
5241 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
5242 | /* rate->cell_align = 0; */ /* distro headers. */ | |
5243 | rate->mpu = ETH_TOTAL_MIN; | |
5244 | rate->rate = Bps; | |
5245 | } | |
5246 | ||
5247 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
5248 | * attribute of the specified "type". | |
5249 | * | |
5250 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
5251 | static void | |
5252 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
5253 | { | |
5254 | uint32_t *rtab; | |
5255 | unsigned int i; | |
5256 | ||
5257 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
5258 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
5259 | unsigned packet_size = (i + 1) << rate->cell_log; | |
5260 | if (packet_size < rate->mpu) { | |
5261 | packet_size = rate->mpu; | |
5262 | } | |
5263 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
5264 | } | |
5265 | } | |
5266 | ||
5267 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
5268 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
5269 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 5270 | * 0 is fine.) */ |
c1c9c9c4 BP |
5271 | static int |
5272 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
5273 | { | |
5274 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
5275 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
5276 | } | |
d3980822 | 5277 | \f |
aaf2fb1a BP |
5278 | /* Linux-only functions declared in netdev-linux.h */ |
5279 | ||
5280 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If | |
5281 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
5282 | int | |
5283 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
5284 | const char *flag_name, bool enable) | |
5285 | { | |
5286 | const char *netdev_name = netdev_get_name(netdev); | |
5287 | struct ethtool_value evalue; | |
5288 | uint32_t new_flags; | |
5289 | int error; | |
5290 | ||
ab985a77 | 5291 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5292 | memset(&evalue, 0, sizeof evalue); |
5293 | error = netdev_linux_do_ethtool(netdev_name, | |
5294 | (struct ethtool_cmd *)&evalue, | |
5295 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5296 | if (error) { | |
5297 | return error; | |
5298 | } | |
5299 | ||
ab985a77 | 5300 | COVERAGE_INC(netdev_set_ethtool); |
aaf2fb1a BP |
5301 | evalue.data = new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
5302 | error = netdev_linux_do_ethtool(netdev_name, | |
5303 | (struct ethtool_cmd *)&evalue, | |
5304 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
5305 | if (error) { | |
5306 | return error; | |
5307 | } | |
5308 | ||
ab985a77 | 5309 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5310 | memset(&evalue, 0, sizeof evalue); |
5311 | error = netdev_linux_do_ethtool(netdev_name, | |
5312 | (struct ethtool_cmd *)&evalue, | |
5313 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5314 | if (error) { | |
5315 | return error; | |
5316 | } | |
5317 | ||
5318 | if (new_flags != evalue.data) { | |
5319 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
5320 | "device %s failed", enable ? "enable" : "disable", | |
5321 | flag_name, netdev_name); | |
5322 | return EOPNOTSUPP; | |
5323 | } | |
5324 | ||
5325 | return 0; | |
5326 | } | |
5327 | \f | |
5328 | /* Utility functions. */ | |
5329 | ||
d3980822 | 5330 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 5331 | static void |
d3980822 BP |
5332 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
5333 | const struct rtnl_link_stats *src) | |
5334 | { | |
f613a0d7 PS |
5335 | dst->rx_packets = src->rx_packets; |
5336 | dst->tx_packets = src->tx_packets; | |
5337 | dst->rx_bytes = src->rx_bytes; | |
5338 | dst->tx_bytes = src->tx_bytes; | |
5339 | dst->rx_errors = src->rx_errors; | |
5340 | dst->tx_errors = src->tx_errors; | |
5341 | dst->rx_dropped = src->rx_dropped; | |
5342 | dst->tx_dropped = src->tx_dropped; | |
5343 | dst->multicast = src->multicast; | |
5344 | dst->collisions = src->collisions; | |
5345 | dst->rx_length_errors = src->rx_length_errors; | |
5346 | dst->rx_over_errors = src->rx_over_errors; | |
5347 | dst->rx_crc_errors = src->rx_crc_errors; | |
5348 | dst->rx_frame_errors = src->rx_frame_errors; | |
5349 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5350 | dst->rx_missed_errors = src->rx_missed_errors; | |
5351 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5352 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5353 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5354 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5355 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
5356 | } |
5357 | ||
337c9b99 BP |
5358 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
5359 | static void | |
5360 | netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst, | |
5361 | const struct rtnl_link_stats64 *src) | |
5362 | { | |
5363 | dst->rx_packets = src->rx_packets; | |
5364 | dst->tx_packets = src->tx_packets; | |
5365 | dst->rx_bytes = src->rx_bytes; | |
5366 | dst->tx_bytes = src->tx_bytes; | |
5367 | dst->rx_errors = src->rx_errors; | |
5368 | dst->tx_errors = src->tx_errors; | |
5369 | dst->rx_dropped = src->rx_dropped; | |
5370 | dst->tx_dropped = src->tx_dropped; | |
5371 | dst->multicast = src->multicast; | |
5372 | dst->collisions = src->collisions; | |
5373 | dst->rx_length_errors = src->rx_length_errors; | |
5374 | dst->rx_over_errors = src->rx_over_errors; | |
5375 | dst->rx_crc_errors = src->rx_crc_errors; | |
5376 | dst->rx_frame_errors = src->rx_frame_errors; | |
5377 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5378 | dst->rx_missed_errors = src->rx_missed_errors; | |
5379 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5380 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5381 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5382 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5383 | dst->tx_window_errors = src->tx_window_errors; | |
5384 | } | |
5385 | ||
c1c9c9c4 | 5386 | static int |
35eef899 | 5387 | get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) |
c1c9c9c4 | 5388 | { |
c1c9c9c4 BP |
5389 | struct ofpbuf request; |
5390 | struct ofpbuf *reply; | |
c1c9c9c4 BP |
5391 | int error; |
5392 | ||
5393 | ofpbuf_init(&request, 0); | |
13a24df8 BP |
5394 | nl_msg_put_nlmsghdr(&request, |
5395 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
5396 | RTM_GETLINK, NLM_F_REQUEST); | |
5397 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
5398 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); | |
a88b4e04 | 5399 | error = nl_transact(NETLINK_ROUTE, &request, &reply); |
c1c9c9c4 BP |
5400 | ofpbuf_uninit(&request); |
5401 | if (error) { | |
5402 | return error; | |
5403 | } | |
5404 | ||
13a24df8 | 5405 | if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { |
337c9b99 BP |
5406 | const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64); |
5407 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) { | |
5408 | netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a)); | |
13a24df8 BP |
5409 | error = 0; |
5410 | } else { | |
337c9b99 BP |
5411 | const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS); |
5412 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { | |
5413 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); | |
5414 | error = 0; | |
5415 | } else { | |
5416 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
5417 | error = EPROTO; | |
5418 | } | |
13a24df8 BP |
5419 | } |
5420 | } else { | |
5421 | VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); | |
5422 | error = EPROTO; | |
c1c9c9c4 | 5423 | } |
8b61709d | 5424 | |
8b61709d | 5425 | |
576e26d7 | 5426 | ofpbuf_delete(reply); |
35eef899 | 5427 | return error; |
8b61709d | 5428 | } |
c1c9c9c4 | 5429 | |
3a183124 | 5430 | static int |
b5d57fc8 | 5431 | get_flags(const struct netdev *dev, unsigned int *flags) |
8b61709d BP |
5432 | { |
5433 | struct ifreq ifr; | |
5434 | int error; | |
5435 | ||
755be9ea | 5436 | *flags = 0; |
259e0b1a | 5437 | error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); |
755be9ea EJ |
5438 | if (!error) { |
5439 | *flags = ifr.ifr_flags; | |
5440 | } | |
8b61709d BP |
5441 | return error; |
5442 | } | |
5443 | ||
5444 | static int | |
4b609110 | 5445 | set_flags(const char *name, unsigned int flags) |
8b61709d BP |
5446 | { |
5447 | struct ifreq ifr; | |
5448 | ||
5449 | ifr.ifr_flags = flags; | |
259e0b1a | 5450 | return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); |
8b61709d BP |
5451 | } |
5452 | ||
5453 | static int | |
5454 | do_get_ifindex(const char *netdev_name) | |
5455 | { | |
5456 | struct ifreq ifr; | |
259e0b1a | 5457 | int error; |
8b61709d | 5458 | |
71d7c22f | 5459 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5460 | COVERAGE_INC(netdev_get_ifindex); |
259e0b1a BP |
5461 | |
5462 | error = af_inet_ioctl(SIOCGIFINDEX, &ifr); | |
5463 | if (error) { | |
8b61709d | 5464 | VLOG_WARN_RL(&rl, "ioctl(SIOCGIFINDEX) on %s device failed: %s", |
259e0b1a BP |
5465 | netdev_name, ovs_strerror(error)); |
5466 | return -error; | |
8b61709d BP |
5467 | } |
5468 | return ifr.ifr_ifindex; | |
5469 | } | |
5470 | ||
5471 | static int | |
5472 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
5473 | { | |
b5d57fc8 | 5474 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c7b1b0a5 | 5475 | |
b5d57fc8 | 5476 | if (!(netdev->cache_valid & VALID_IFINDEX)) { |
8b61709d | 5477 | int ifindex = do_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 5478 | |
8b61709d | 5479 | if (ifindex < 0) { |
b5d57fc8 BP |
5480 | netdev->get_ifindex_error = -ifindex; |
5481 | netdev->ifindex = 0; | |
c7b1b0a5 | 5482 | } else { |
b5d57fc8 BP |
5483 | netdev->get_ifindex_error = 0; |
5484 | netdev->ifindex = ifindex; | |
8b61709d | 5485 | } |
b5d57fc8 | 5486 | netdev->cache_valid |= VALID_IFINDEX; |
8b61709d | 5487 | } |
c7b1b0a5 | 5488 | |
b5d57fc8 BP |
5489 | *ifindexp = netdev->ifindex; |
5490 | return netdev->get_ifindex_error; | |
8b61709d BP |
5491 | } |
5492 | ||
5493 | static int | |
74ff3298 | 5494 | get_etheraddr(const char *netdev_name, struct eth_addr *ea) |
8b61709d BP |
5495 | { |
5496 | struct ifreq ifr; | |
5497 | int hwaddr_family; | |
259e0b1a | 5498 | int error; |
8b61709d BP |
5499 | |
5500 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5501 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5502 | COVERAGE_INC(netdev_get_hwaddr); |
259e0b1a BP |
5503 | error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); |
5504 | if (error) { | |
78857dfb BP |
5505 | /* ENODEV probably means that a vif disappeared asynchronously and |
5506 | * hasn't been removed from the database yet, so reduce the log level | |
5507 | * to INFO for that case. */ | |
259e0b1a | 5508 | VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, |
78857dfb | 5509 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", |
259e0b1a BP |
5510 | netdev_name, ovs_strerror(error)); |
5511 | return error; | |
8b61709d BP |
5512 | } |
5513 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
5514 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER) { | |
5515 | VLOG_WARN("%s device has unknown hardware address family %d", | |
5516 | netdev_name, hwaddr_family); | |
5517 | } | |
5518 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
5519 | return 0; | |
5520 | } | |
5521 | ||
5522 | static int | |
74ff3298 | 5523 | set_etheraddr(const char *netdev_name, const struct eth_addr mac) |
8b61709d BP |
5524 | { |
5525 | struct ifreq ifr; | |
259e0b1a | 5526 | int error; |
8b61709d BP |
5527 | |
5528 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5529 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 5530 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
74ff3298 | 5531 | memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN); |
8b61709d | 5532 | COVERAGE_INC(netdev_set_hwaddr); |
259e0b1a BP |
5533 | error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); |
5534 | if (error) { | |
8b61709d | 5535 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", |
259e0b1a | 5536 | netdev_name, ovs_strerror(error)); |
8b61709d | 5537 | } |
259e0b1a | 5538 | return error; |
8b61709d BP |
5539 | } |
5540 | ||
5541 | static int | |
0b0544d7 | 5542 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
5543 | int cmd, const char *cmd_name) |
5544 | { | |
5545 | struct ifreq ifr; | |
259e0b1a | 5546 | int error; |
8b61709d BP |
5547 | |
5548 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5549 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
5550 | ifr.ifr_data = (caddr_t) ecmd; |
5551 | ||
5552 | ecmd->cmd = cmd; | |
259e0b1a BP |
5553 | error = af_inet_ioctl(SIOCETHTOOL, &ifr); |
5554 | if (error) { | |
5555 | if (error != EOPNOTSUPP) { | |
8b61709d | 5556 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " |
259e0b1a | 5557 | "failed: %s", cmd_name, name, ovs_strerror(error)); |
8b61709d BP |
5558 | } else { |
5559 | /* The device doesn't support this operation. That's pretty | |
5560 | * common, so there's no point in logging anything. */ | |
5561 | } | |
8b61709d | 5562 | } |
259e0b1a | 5563 | return error; |
8b61709d | 5564 | } |
f1acd62b BP |
5565 | |
5566 | static int | |
5567 | netdev_linux_get_ipv4(const struct netdev *netdev, struct in_addr *ip, | |
5568 | int cmd, const char *cmd_name) | |
5569 | { | |
5570 | struct ifreq ifr; | |
5571 | int error; | |
5572 | ||
5573 | ifr.ifr_addr.sa_family = AF_INET; | |
259e0b1a | 5574 | error = af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, cmd, cmd_name); |
f1acd62b | 5575 | if (!error) { |
db5a1019 AW |
5576 | const struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, |
5577 | &ifr.ifr_addr); | |
f1acd62b BP |
5578 | *ip = sin->sin_addr; |
5579 | } | |
5580 | return error; | |
5581 | } | |
488d734d BP |
5582 | |
5583 | /* Returns an AF_PACKET raw socket or a negative errno value. */ | |
5584 | static int | |
5585 | af_packet_sock(void) | |
5586 | { | |
23882115 BP |
5587 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
5588 | static int sock; | |
488d734d | 5589 | |
23882115 | 5590 | if (ovsthread_once_start(&once)) { |
488d734d BP |
5591 | sock = socket(AF_PACKET, SOCK_RAW, 0); |
5592 | if (sock >= 0) { | |
8450059e BP |
5593 | int error = set_nonblocking(sock); |
5594 | if (error) { | |
5595 | close(sock); | |
5596 | sock = -error; | |
5597 | } | |
488d734d BP |
5598 | } else { |
5599 | sock = -errno; | |
10a89ef0 BP |
5600 | VLOG_ERR("failed to create packet socket: %s", |
5601 | ovs_strerror(errno)); | |
488d734d | 5602 | } |
23882115 | 5603 | ovsthread_once_done(&once); |
488d734d BP |
5604 | } |
5605 | ||
5606 | return sock; | |
5607 | } |