]>
Commit | Line | Data |
---|---|---|
e9e28be3 | 1 | /* |
48c6733c | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017 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> |
b2befd5b BP |
23 | #include <sys/types.h> |
24 | #include <netinet/in.h> | |
55bc98d6 | 25 | #include <arpa/inet.h> |
8b61709d | 26 | #include <inttypes.h> |
32383c3b | 27 | #include <linux/filter.h> |
c1c9c9c4 | 28 | #include <linux/gen_stats.h> |
bb7d0e22 | 29 | #include <linux/if_ether.h> |
8b61709d BP |
30 | #include <linux/if_tun.h> |
31 | #include <linux/types.h> | |
32 | #include <linux/ethtool.h> | |
63331829 | 33 | #include <linux/mii.h> |
ef3767f5 | 34 | #include <linux/rtnetlink.h> |
8b61709d | 35 | #include <linux/sockios.h> |
8b61709d BP |
36 | #include <sys/ioctl.h> |
37 | #include <sys/socket.h> | |
ac3e3aaa | 38 | #include <sys/utsname.h> |
55bc98d6 | 39 | #include <netpacket/packet.h> |
8b61709d BP |
40 | #include <net/if.h> |
41 | #include <net/if_arp.h> | |
55bc98d6 | 42 | #include <net/if_packet.h> |
8b61709d | 43 | #include <net/route.h> |
e9e28be3 | 44 | #include <poll.h> |
8b61709d BP |
45 | #include <stdlib.h> |
46 | #include <string.h> | |
47 | #include <unistd.h> | |
e9e28be3 BP |
48 | |
49 | #include "coverage.h" | |
e14deea0 | 50 | #include "dp-packet.h" |
93451a0a | 51 | #include "dpif-netlink.h" |
df1e5a3b | 52 | #include "dpif-netdev.h" |
3e8a2ad1 | 53 | #include "openvswitch/dynamic-string.h" |
8b61709d | 54 | #include "fatal-signal.h" |
93b13be8 | 55 | #include "hash.h" |
ee89ea7b | 56 | #include "openvswitch/hmap.h" |
8b61709d | 57 | #include "netdev-provider.h" |
18ebd48c | 58 | #include "netdev-tc-offloads.h" |
7fbef77a | 59 | #include "netdev-vport.h" |
45c8d3a1 | 60 | #include "netlink-notifier.h" |
2fe27d5a | 61 | #include "netlink-socket.h" |
c060c4cf | 62 | #include "netlink.h" |
bfda5239 | 63 | #include "netnsid.h" |
64c96779 | 64 | #include "openvswitch/ofpbuf.h" |
8b61709d | 65 | #include "openflow/openflow.h" |
19c8e9c1 | 66 | #include "ovs-atomic.h" |
8b61709d | 67 | #include "packets.h" |
fd016ae3 | 68 | #include "openvswitch/poll-loop.h" |
7e9dcc0f | 69 | #include "rtnetlink.h" |
ee89ea7b | 70 | #include "openvswitch/shash.h" |
c060c4cf | 71 | #include "socket-util.h" |
19993ef3 | 72 | #include "sset.h" |
c1c5c723 | 73 | #include "tc.h" |
1670c579 | 74 | #include "timer.h" |
c060c4cf | 75 | #include "unaligned.h" |
e6211adc | 76 | #include "openvswitch/vlog.h" |
ee89ea7b | 77 | #include "util.h" |
5136ce49 | 78 | |
d98e6007 | 79 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea | 80 | |
d76f09ea BP |
81 | COVERAGE_DEFINE(netdev_set_policing); |
82 | COVERAGE_DEFINE(netdev_arp_lookup); | |
83 | COVERAGE_DEFINE(netdev_get_ifindex); | |
84 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
85 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
ab985a77 BP |
86 | COVERAGE_DEFINE(netdev_get_ethtool); |
87 | COVERAGE_DEFINE(netdev_set_ethtool); | |
4f925bd3 | 88 | |
8b61709d | 89 | \f |
756819dd FL |
90 | #ifndef IFLA_IF_NETNSID |
91 | #define IFLA_IF_NETNSID 0x45 | |
92 | #endif | |
8b61709d BP |
93 | /* These were introduced in Linux 2.6.14, so they might be missing if we have |
94 | * old headers. */ | |
95 | #ifndef ADVERTISED_Pause | |
96 | #define ADVERTISED_Pause (1 << 13) | |
97 | #endif | |
98 | #ifndef ADVERTISED_Asym_Pause | |
99 | #define ADVERTISED_Asym_Pause (1 << 14) | |
100 | #endif | |
101 | ||
e47bd51a JP |
102 | /* These were introduced in Linux 2.6.24, so they might be missing if we |
103 | * have old headers. */ | |
104 | #ifndef ETHTOOL_GFLAGS | |
105 | #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ | |
106 | #endif | |
107 | #ifndef ETHTOOL_SFLAGS | |
108 | #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ | |
109 | #endif | |
110 | ||
c1c9c9c4 BP |
111 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
112 | * headers. */ | |
113 | #ifndef TC_RTAB_SIZE | |
114 | #define TC_RTAB_SIZE 1024 | |
115 | #endif | |
116 | ||
b73c8518 SH |
117 | /* Linux 2.6.21 introduced struct tpacket_auxdata. |
118 | * Linux 2.6.27 added the tp_vlan_tci member. | |
119 | * Linux 3.0 defined TP_STATUS_VLAN_VALID. | |
120 | * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined | |
121 | * TP_STATUS_VLAN_TPID_VALID. | |
122 | * | |
123 | * With all this churn it's easiest to unconditionally define a replacement | |
124 | * structure that has everything we want. | |
125 | */ | |
55bc98d6 BP |
126 | #ifndef PACKET_AUXDATA |
127 | #define PACKET_AUXDATA 8 | |
128 | #endif | |
b73c8518 SH |
129 | #ifndef TP_STATUS_VLAN_VALID |
130 | #define TP_STATUS_VLAN_VALID (1 << 4) | |
131 | #endif | |
132 | #ifndef TP_STATUS_VLAN_TPID_VALID | |
133 | #define TP_STATUS_VLAN_TPID_VALID (1 << 6) | |
134 | #endif | |
135 | #undef tpacket_auxdata | |
136 | #define tpacket_auxdata rpl_tpacket_auxdata | |
137 | struct tpacket_auxdata { | |
138 | uint32_t tp_status; | |
139 | uint32_t tp_len; | |
140 | uint32_t tp_snaplen; | |
141 | uint16_t tp_mac; | |
142 | uint16_t tp_net; | |
143 | uint16_t tp_vlan_tci; | |
144 | uint16_t tp_vlan_tpid; | |
145 | }; | |
146 | ||
0c615356 SH |
147 | /* Linux 2.6.27 introduced ethtool_cmd_speed |
148 | * | |
149 | * To avoid revisiting problems reported with using configure to detect | |
150 | * compatibility (see report at | |
8a7903c6 | 151 | * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html) |
0c615356 SH |
152 | * unconditionally replace ethtool_cmd_speed. */ |
153 | #define ethtool_cmd_speed rpl_ethtool_cmd_speed | |
154 | static inline uint32_t rpl_ethtool_cmd_speed(const struct ethtool_cmd *ep) | |
155 | { | |
156 | return ep->speed | (ep->speed_hi << 16); | |
157 | } | |
158 | ||
67bed84c SH |
159 | /* Linux 2.6.30 introduced supported and advertised flags for |
160 | * 1G base KX, and 10G base KX4, KR and R. */ | |
161 | #ifndef SUPPORTED_1000baseKX_Full | |
162 | #define SUPPORTED_1000baseKX_Full (1 << 17) | |
163 | #define SUPPORTED_10000baseKX4_Full (1 << 18) | |
164 | #define SUPPORTED_10000baseKR_Full (1 << 19) | |
165 | #define SUPPORTED_10000baseR_FEC (1 << 20) | |
166 | #define ADVERTISED_1000baseKX_Full (1 << 17) | |
167 | #define ADVERTISED_10000baseKX4_Full (1 << 18) | |
168 | #define ADVERTISED_10000baseKR_Full (1 << 19) | |
169 | #define ADVERTISED_10000baseR_FEC (1 << 20) | |
170 | #endif | |
171 | ||
172 | /* Linux 3.5 introduced supported and advertised flags for | |
173 | * 40G base KR4, CR4, SR4 and LR4. */ | |
174 | #ifndef SUPPORTED_40000baseKR4_Full | |
175 | #define SUPPORTED_40000baseKR4_Full (1 << 23) | |
176 | #define SUPPORTED_40000baseCR4_Full (1 << 24) | |
177 | #define SUPPORTED_40000baseSR4_Full (1 << 25) | |
178 | #define SUPPORTED_40000baseLR4_Full (1 << 26) | |
179 | #define ADVERTISED_40000baseKR4_Full (1 << 23) | |
180 | #define ADVERTISED_40000baseCR4_Full (1 << 24) | |
181 | #define ADVERTISED_40000baseSR4_Full (1 << 25) | |
182 | #define ADVERTISED_40000baseLR4_Full (1 << 26) | |
183 | #endif | |
184 | ||
fa373af4 BP |
185 | /* Linux 2.6.35 introduced IFLA_STATS64 and rtnl_link_stats64. |
186 | * | |
187 | * Tests for rtnl_link_stats64 don't seem to consistently work, e.g. on | |
188 | * 2.6.32-431.29.2.el6.x86_64 (see report at | |
8a7903c6 JP |
189 | * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html). |
190 | * Maybe if_link.h is not self-contained on those kernels. It is easiest to | |
fa373af4 BP |
191 | * unconditionally define a replacement. */ |
192 | #ifndef IFLA_STATS64 | |
337c9b99 | 193 | #define IFLA_STATS64 23 |
fa373af4 BP |
194 | #endif |
195 | #define rtnl_link_stats64 rpl_rtnl_link_stats64 | |
337c9b99 BP |
196 | struct rtnl_link_stats64 { |
197 | uint64_t rx_packets; | |
198 | uint64_t tx_packets; | |
199 | uint64_t rx_bytes; | |
200 | uint64_t tx_bytes; | |
201 | uint64_t rx_errors; | |
202 | uint64_t tx_errors; | |
203 | uint64_t rx_dropped; | |
204 | uint64_t tx_dropped; | |
205 | uint64_t multicast; | |
206 | uint64_t collisions; | |
207 | ||
208 | uint64_t rx_length_errors; | |
209 | uint64_t rx_over_errors; | |
210 | uint64_t rx_crc_errors; | |
211 | uint64_t rx_frame_errors; | |
212 | uint64_t rx_fifo_errors; | |
213 | uint64_t rx_missed_errors; | |
214 | ||
215 | uint64_t tx_aborted_errors; | |
216 | uint64_t tx_carrier_errors; | |
217 | uint64_t tx_fifo_errors; | |
218 | uint64_t tx_heartbeat_errors; | |
219 | uint64_t tx_window_errors; | |
220 | ||
221 | uint64_t rx_compressed; | |
222 | uint64_t tx_compressed; | |
223 | }; | |
337c9b99 | 224 | |
8b61709d | 225 | enum { |
7fbef77a JG |
226 | VALID_IFINDEX = 1 << 0, |
227 | VALID_ETHERADDR = 1 << 1, | |
6b6e1329 PS |
228 | VALID_IN = 1 << 2, |
229 | VALID_MTU = 1 << 3, | |
230 | VALID_POLICING = 1 << 4, | |
231 | VALID_VPORT_STAT_ERROR = 1 << 5, | |
232 | VALID_DRVINFO = 1 << 6, | |
233 | VALID_FEATURES = 1 << 7, | |
8b61709d | 234 | }; |
c1c9c9c4 BP |
235 | \f |
236 | /* Traffic control. */ | |
237 | ||
238 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
239 | * network device. |
240 | * | |
241 | * Each TC implementation subclasses this with whatever additional data it | |
242 | * needs. */ | |
c1c9c9c4 BP |
243 | struct tc { |
244 | const struct tc_ops *ops; | |
93b13be8 BP |
245 | struct hmap queues; /* Contains "struct tc_queue"s. |
246 | * Read by generic TC layer. | |
247 | * Written only by TC implementation. */ | |
248 | }; | |
c1c9c9c4 | 249 | |
559eb230 BP |
250 | #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) } |
251 | ||
93b13be8 BP |
252 | /* One traffic control queue. |
253 | * | |
254 | * Each TC implementation subclasses this with whatever additional data it | |
255 | * needs. */ | |
256 | struct tc_queue { | |
257 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
258 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
6dc34a0d | 259 | long long int created; /* Time queue was created, in msecs. */ |
c1c9c9c4 BP |
260 | }; |
261 | ||
262 | /* A particular kind of traffic control. Each implementation generally maps to | |
263 | * one particular Linux qdisc class. | |
264 | * | |
265 | * The functions below return 0 if successful or a positive errno value on | |
266 | * failure, except where otherwise noted. All of them must be provided, except | |
267 | * where otherwise noted. */ | |
268 | struct tc_ops { | |
269 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
270 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
271 | * appropriate values. */ | |
272 | const char *linux_name; | |
273 | ||
274 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
275 | const char *ovs_name; | |
276 | ||
277 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
278 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
279 | unsigned int n_queues; | |
280 | ||
281 | /* Called to install this TC class on 'netdev'. The implementation should | |
282 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
283 | * and configure it according to 'details'. The implementation may assume | |
284 | * that the current qdisc is the default; that is, there is no need for it | |
285 | * to delete the current qdisc before installing itself. | |
286 | * | |
287 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
288 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
289 | * (which is built as ovs-vswitchd.conf.db(8)). | |
290 | * | |
291 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
292 | * initialized 'struct tc'. | |
293 | * | |
294 | * (This function is null for tc_ops_other, which cannot be installed. For | |
295 | * other TC classes it should always be nonnull.) */ | |
79f1cbe9 | 296 | int (*tc_install)(struct netdev *netdev, const struct smap *details); |
c1c9c9c4 BP |
297 | |
298 | /* Called when the netdev code determines (through a Netlink query) that | |
299 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
300 | * it ourselves and so don't know any of the details. | |
301 | * | |
302 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
303 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
304 | * implementation should parse the other attributes of 'nlmsg' as | |
305 | * necessary to determine its configuration. If necessary it should also | |
306 | * use Netlink queries to determine the configuration of queues on | |
307 | * 'netdev'. | |
308 | * | |
309 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
310 | * initialized 'struct tc'. */ | |
311 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
312 | ||
313 | /* Destroys the data structures allocated by the implementation as part of | |
314 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
315 | * tc_destroy(tc). | |
316 | * | |
317 | * The implementation should not need to perform any Netlink calls. If | |
318 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
319 | * (But it may not be desirable.) | |
320 | * | |
321 | * This function may be null if 'tc' is trivial. */ | |
322 | void (*tc_destroy)(struct tc *tc); | |
323 | ||
324 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
325 | * | |
326 | * The implementation should not need to perform any Netlink calls, because | |
327 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
328 | * cached the configuration. | |
329 | * | |
330 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
331 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
332 | * (which is built as ovs-vswitchd.conf.db(8)). | |
333 | * | |
334 | * This function may be null if 'tc' is not configurable. | |
335 | */ | |
79f1cbe9 | 336 | int (*qdisc_get)(const struct netdev *netdev, struct smap *details); |
c1c9c9c4 BP |
337 | |
338 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
339 | * required Netlink calls to complete the reconfiguration. | |
340 | * | |
341 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
342 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
343 | * (which is built as ovs-vswitchd.conf.db(8)). | |
344 | * | |
345 | * This function may be null if 'tc' is not configurable. | |
346 | */ | |
79f1cbe9 | 347 | int (*qdisc_set)(struct netdev *, const struct smap *details); |
c1c9c9c4 | 348 | |
93b13be8 BP |
349 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
350 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
351 | * |
352 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
353 | * in the "other_config" column in the "Queue" table in | |
354 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
355 | * | |
356 | * The implementation should not need to perform any Netlink calls, because | |
357 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
358 | * cached the queue configuration. | |
359 | * | |
360 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
361 | * 0). */ | |
93b13be8 | 362 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
79f1cbe9 | 363 | struct smap *details); |
c1c9c9c4 BP |
364 | |
365 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
366 | * 'details', perfoming any required Netlink calls to complete the | |
367 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
368 | * 'n_queues'. | |
369 | * | |
370 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
371 | * in the "other_config" column in the "Queue" table in | |
372 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
373 | * | |
374 | * This function may be null if 'tc' does not have queues or its queues are | |
375 | * not configurable. */ | |
376 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
79f1cbe9 | 377 | const struct smap *details); |
c1c9c9c4 | 378 | |
93b13be8 BP |
379 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
380 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
381 | * |
382 | * This function may be null if 'tc' does not have queues or its queues | |
383 | * cannot be deleted. */ | |
93b13be8 | 384 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 385 | |
93b13be8 BP |
386 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
387 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
388 | * |
389 | * On success, initializes '*stats'. | |
390 | * | |
391 | * This function may be null if 'tc' does not have queues or if it cannot | |
392 | * report queue statistics. */ | |
93b13be8 BP |
393 | int (*class_get_stats)(const struct netdev *netdev, |
394 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
395 | struct netdev_queue_stats *stats); |
396 | ||
397 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
398 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
399 | * | |
400 | * This function may be null if 'tc' does not have queues or if it cannot | |
401 | * report queue statistics. */ | |
402 | int (*class_dump_stats)(const struct netdev *netdev, | |
403 | const struct ofpbuf *nlmsg, | |
404 | netdev_dump_queue_stats_cb *cb, void *aux); | |
405 | }; | |
406 | ||
407 | static void | |
408 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
409 | { | |
410 | tc->ops = ops; | |
93b13be8 | 411 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
412 | } |
413 | ||
414 | static void | |
415 | tc_destroy(struct tc *tc) | |
416 | { | |
93b13be8 | 417 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
418 | } |
419 | ||
420 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 421 | static const struct tc_ops tc_ops_hfsc; |
677d9158 JV |
422 | static const struct tc_ops tc_ops_codel; |
423 | static const struct tc_ops tc_ops_fqcodel; | |
424 | static const struct tc_ops tc_ops_sfq; | |
c1c9c9c4 | 425 | static const struct tc_ops tc_ops_default; |
6cf888b8 | 426 | static const struct tc_ops tc_ops_noop; |
c1c9c9c4 BP |
427 | static const struct tc_ops tc_ops_other; |
428 | ||
559eb230 | 429 | static const struct tc_ops *const tcs[] = { |
c1c9c9c4 | 430 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ |
a339aa81 | 431 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
677d9158 JV |
432 | &tc_ops_codel, /* Controlled delay */ |
433 | &tc_ops_fqcodel, /* Fair queue controlled delay */ | |
434 | &tc_ops_sfq, /* Stochastic fair queueing */ | |
6cf888b8 | 435 | &tc_ops_noop, /* Non operating qos type. */ |
c1c9c9c4 BP |
436 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
437 | &tc_ops_other, /* Some other qdisc. */ | |
438 | NULL | |
439 | }; | |
149f577a | 440 | |
c1c9c9c4 BP |
441 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); |
442 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
443 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
444 | ||
7874bdff RD |
445 | static struct tcmsg *netdev_linux_tc_make_request(const struct netdev *, |
446 | int type, | |
447 | unsigned int flags, | |
448 | struct ofpbuf *); | |
c7952afb BP |
449 | static int tc_add_policer(struct netdev *, |
450 | uint32_t kbits_rate, uint32_t kbits_burst); | |
c1c9c9c4 BP |
451 | |
452 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
453 | struct nlattr **options); | |
454 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
455 | struct nlattr **options, | |
456 | struct netdev_queue_stats *); | |
457 | static int tc_query_class(const struct netdev *, | |
458 | unsigned int handle, unsigned int parent, | |
459 | struct ofpbuf **replyp); | |
460 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
461 | ||
462 | static int tc_del_qdisc(struct netdev *netdev); | |
463 | static int tc_query_qdisc(const struct netdev *netdev); | |
464 | ||
465 | static int tc_calc_cell_log(unsigned int mtu); | |
466 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
467 | static void tc_put_rtab(struct ofpbuf *, uint16_t type, | |
468 | const struct tc_ratespec *rate); | |
469 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); | |
470 | \f | |
b5d57fc8 BP |
471 | struct netdev_linux { |
472 | struct netdev up; | |
149f577a | 473 | |
86383816 BP |
474 | /* Protects all members below. */ |
475 | struct ovs_mutex mutex; | |
476 | ||
149f577a | 477 | unsigned int cache_valid; |
8b61709d | 478 | |
1670c579 EJ |
479 | bool miimon; /* Link status of last poll. */ |
480 | long long int miimon_interval; /* Miimon Poll rate. Disabled if <= 0. */ | |
481 | struct timer miimon_timer; | |
482 | ||
bfda5239 | 483 | int netnsid; /* Network namespace ID. */ |
8722022c BP |
484 | /* The following are figured out "on demand" only. They are only valid |
485 | * when the corresponding VALID_* bit in 'cache_valid' is set. */ | |
8b61709d | 486 | int ifindex; |
74ff3298 | 487 | struct eth_addr etheraddr; |
8b61709d | 488 | int mtu; |
059e5f4f | 489 | unsigned int ifi_flags; |
65c3058c | 490 | long long int carrier_resets; |
80a86fbe BP |
491 | uint32_t kbits_rate; /* Policing data. */ |
492 | uint32_t kbits_burst; | |
bba1e6f3 PS |
493 | int vport_stats_error; /* Cached error code from vport_get_stats(). |
494 | 0 or an errno value. */ | |
90a6637d | 495 | int netdev_mtu_error; /* Cached error code from SIOCGIFMTU or SIOCSIFMTU. */ |
44445cac | 496 | int ether_addr_error; /* Cached error code from set/get etheraddr. */ |
c9f71668 | 497 | int netdev_policing_error; /* Cached error code from set policing. */ |
51f87458 | 498 | int get_features_error; /* Cached error code from ETHTOOL_GSET. */ |
c7b1b0a5 | 499 | int get_ifindex_error; /* Cached error code from SIOCGIFINDEX. */ |
51f87458 | 500 | |
a00ca915 EJ |
501 | enum netdev_features current; /* Cached from ETHTOOL_GSET. */ |
502 | enum netdev_features advertised; /* Cached from ETHTOOL_GSET. */ | |
503 | enum netdev_features supported; /* Cached from ETHTOOL_GSET. */ | |
90a6637d | 504 | |
4f925bd3 | 505 | struct ethtool_drvinfo drvinfo; /* Cached from ETHTOOL_GDRVINFO. */ |
c1c9c9c4 | 506 | struct tc *tc; |
149f577a | 507 | |
d0d08f8a BP |
508 | /* For devices of class netdev_tap_class only. */ |
509 | int tap_fd; | |
22dcb534 FL |
510 | bool present; /* If the device is present in the namespace */ |
511 | uint64_t tx_dropped; /* tap device can drop if the iface is down */ | |
8b61709d BP |
512 | }; |
513 | ||
f7791740 PS |
514 | struct netdev_rxq_linux { |
515 | struct netdev_rxq up; | |
796223f5 | 516 | bool is_tap; |
5b7448ed | 517 | int fd; |
149f577a | 518 | }; |
8b61709d | 519 | |
8b61709d BP |
520 | /* This is set pretty low because we probably won't learn anything from the |
521 | * additional log messages. */ | |
522 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
523 | ||
19c8e9c1 JS |
524 | /* Polling miimon status for all ports causes performance degradation when |
525 | * handling a large number of ports. If there are no devices using miimon, then | |
812c272c JR |
526 | * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). |
527 | * | |
528 | * Readers do not depend on this variable synchronizing with the related | |
529 | * changes in the device miimon status, so we can use atomic_count. */ | |
530 | static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0); | |
19c8e9c1 | 531 | |
1c33f0c3 | 532 | static void netdev_linux_run(const struct netdev_class *); |
6f643e49 | 533 | |
0b0544d7 | 534 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 535 | int cmd, const char *cmd_name); |
b5d57fc8 | 536 | static int get_flags(const struct netdev *, unsigned int *flags); |
4b609110 | 537 | static int set_flags(const char *, unsigned int flags); |
4f9f3f21 BP |
538 | static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
539 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
540 | OVS_REQUIRES(netdev->mutex); | |
8b61709d BP |
541 | static int get_ifindex(const struct netdev *, int *ifindexp); |
542 | static int do_set_addr(struct netdev *netdev, | |
543 | int ioctl_nr, const char *ioctl_name, | |
544 | struct in_addr addr); | |
74ff3298 JR |
545 | static int get_etheraddr(const char *netdev_name, struct eth_addr *ea); |
546 | static int set_etheraddr(const char *netdev_name, const struct eth_addr); | |
35eef899 | 547 | static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *); |
488d734d | 548 | static int af_packet_sock(void); |
19c8e9c1 | 549 | static bool netdev_linux_miimon_enabled(void); |
1670c579 EJ |
550 | static void netdev_linux_miimon_run(void); |
551 | static void netdev_linux_miimon_wait(void); | |
df1e5a3b | 552 | static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); |
8b61709d | 553 | |
15b3596a JG |
554 | static bool |
555 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
556 | { | |
259e0b1a | 557 | return netdev_class->run == netdev_linux_run; |
15b3596a JG |
558 | } |
559 | ||
796223f5 BP |
560 | static bool |
561 | is_tap_netdev(const struct netdev *netdev) | |
562 | { | |
b5d57fc8 | 563 | return netdev_get_class(netdev) == &netdev_tap_class; |
6c88d577 JP |
564 | } |
565 | ||
8b61709d BP |
566 | static struct netdev_linux * |
567 | netdev_linux_cast(const struct netdev *netdev) | |
568 | { | |
b5d57fc8 | 569 | ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); |
15b3596a | 570 | |
180c6d0b | 571 | return CONTAINER_OF(netdev, struct netdev_linux, up); |
8b61709d | 572 | } |
796223f5 | 573 | |
f7791740 PS |
574 | static struct netdev_rxq_linux * |
575 | netdev_rxq_linux_cast(const struct netdev_rxq *rx) | |
796223f5 | 576 | { |
9dc63482 | 577 | ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev))); |
f7791740 | 578 | return CONTAINER_OF(rx, struct netdev_rxq_linux, up); |
796223f5 | 579 | } |
ff4ed3c9 | 580 | \f |
bfda5239 FL |
581 | static int |
582 | netdev_linux_netnsid_update__(struct netdev_linux *netdev) | |
583 | { | |
584 | struct dpif_netlink_vport reply; | |
585 | struct ofpbuf *buf; | |
586 | int error; | |
587 | ||
588 | error = dpif_netlink_vport_get(netdev_get_name(&netdev->up), &reply, &buf); | |
589 | if (error) { | |
590 | netnsid_unset(&netdev->netnsid); | |
591 | return error; | |
592 | } | |
593 | ||
594 | netnsid_set(&netdev->netnsid, reply.netnsid); | |
595 | ofpbuf_delete(buf); | |
596 | return 0; | |
597 | } | |
598 | ||
599 | static int | |
600 | netdev_linux_netnsid_update(struct netdev_linux *netdev) | |
601 | { | |
602 | if (netnsid_is_unset(netdev->netnsid)) { | |
603 | return netdev_linux_netnsid_update__(netdev); | |
604 | } | |
605 | ||
606 | return 0; | |
607 | } | |
608 | ||
609 | static bool | |
610 | netdev_linux_netnsid_is_eq(struct netdev_linux *netdev, int nsid) | |
611 | { | |
612 | netdev_linux_netnsid_update(netdev); | |
613 | return netnsid_eq(netdev->netnsid, nsid); | |
614 | } | |
615 | ||
756819dd FL |
616 | static bool |
617 | netdev_linux_netnsid_is_remote(struct netdev_linux *netdev) | |
618 | { | |
619 | netdev_linux_netnsid_update(netdev); | |
620 | return netnsid_is_remote(netdev->netnsid); | |
621 | } | |
622 | ||
623 | static int netdev_linux_update_via_netlink(struct netdev_linux *); | |
bfda5239 | 624 | static void netdev_linux_update(struct netdev_linux *netdev, int, |
7e9dcc0f | 625 | const struct rtnetlink_change *) |
86383816 | 626 | OVS_REQUIRES(netdev->mutex); |
cee87338 | 627 | static void netdev_linux_changed(struct netdev_linux *netdev, |
86383816 BP |
628 | unsigned int ifi_flags, unsigned int mask) |
629 | OVS_REQUIRES(netdev->mutex); | |
cee87338 | 630 | |
d6384a3a AW |
631 | /* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK, |
632 | * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL | |
cee87338 BP |
633 | * if no such socket could be created. */ |
634 | static struct nl_sock * | |
635 | netdev_linux_notify_sock(void) | |
636 | { | |
637 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
638 | static struct nl_sock *sock; | |
989d7135 PS |
639 | unsigned int mcgroups[] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR, |
640 | RTNLGRP_IPV6_IFADDR, RTNLGRP_IPV6_IFINFO}; | |
cee87338 BP |
641 | |
642 | if (ovsthread_once_start(&once)) { | |
643 | int error; | |
644 | ||
645 | error = nl_sock_create(NETLINK_ROUTE, &sock); | |
646 | if (!error) { | |
d6384a3a AW |
647 | size_t i; |
648 | ||
649 | for (i = 0; i < ARRAY_SIZE(mcgroups); i++) { | |
650 | error = nl_sock_join_mcgroup(sock, mcgroups[i]); | |
651 | if (error) { | |
652 | nl_sock_destroy(sock); | |
653 | sock = NULL; | |
654 | break; | |
655 | } | |
cee87338 BP |
656 | } |
657 | } | |
658 | ovsthread_once_done(&once); | |
659 | } | |
660 | ||
661 | return sock; | |
662 | } | |
663 | ||
19c8e9c1 JS |
664 | static bool |
665 | netdev_linux_miimon_enabled(void) | |
666 | { | |
812c272c | 667 | return atomic_count_get(&miimon_cnt) > 0; |
19c8e9c1 JS |
668 | } |
669 | ||
8b61709d | 670 | static void |
1c33f0c3 | 671 | netdev_linux_run(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 672 | { |
cee87338 BP |
673 | struct nl_sock *sock; |
674 | int error; | |
675 | ||
19c8e9c1 JS |
676 | if (netdev_linux_miimon_enabled()) { |
677 | netdev_linux_miimon_run(); | |
678 | } | |
cee87338 BP |
679 | |
680 | sock = netdev_linux_notify_sock(); | |
681 | if (!sock) { | |
682 | return; | |
683 | } | |
684 | ||
685 | do { | |
cee87338 | 686 | uint64_t buf_stub[4096 / 8]; |
bfda5239 | 687 | int nsid; |
cee87338 BP |
688 | struct ofpbuf buf; |
689 | ||
690 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
bfda5239 | 691 | error = nl_sock_recv(sock, &buf, &nsid, false); |
cee87338 | 692 | if (!error) { |
7e9dcc0f | 693 | struct rtnetlink_change change; |
cee87338 | 694 | |
7e9dcc0f | 695 | if (rtnetlink_parse(&buf, &change)) { |
989d7135 PS |
696 | struct netdev *netdev_ = NULL; |
697 | char dev_name[IFNAMSIZ]; | |
698 | ||
699 | if (!change.ifname) { | |
700 | change.ifname = if_indextoname(change.if_index, dev_name); | |
701 | } | |
702 | ||
703 | if (change.ifname) { | |
704 | netdev_ = netdev_from_name(change.ifname); | |
705 | } | |
cee87338 BP |
706 | if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { |
707 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
86383816 BP |
708 | |
709 | ovs_mutex_lock(&netdev->mutex); | |
bfda5239 | 710 | netdev_linux_update(netdev, nsid, &change); |
86383816 | 711 | ovs_mutex_unlock(&netdev->mutex); |
cee87338 | 712 | } |
38e0065b | 713 | netdev_close(netdev_); |
cee87338 BP |
714 | } |
715 | } else if (error == ENOBUFS) { | |
716 | struct shash device_shash; | |
717 | struct shash_node *node; | |
718 | ||
719 | nl_sock_drain(sock); | |
720 | ||
721 | shash_init(&device_shash); | |
722 | netdev_get_devices(&netdev_linux_class, &device_shash); | |
723 | SHASH_FOR_EACH (node, &device_shash) { | |
724 | struct netdev *netdev_ = node->data; | |
725 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
726 | unsigned int flags; | |
727 | ||
86383816 | 728 | ovs_mutex_lock(&netdev->mutex); |
cee87338 BP |
729 | get_flags(netdev_, &flags); |
730 | netdev_linux_changed(netdev, flags, 0); | |
86383816 BP |
731 | ovs_mutex_unlock(&netdev->mutex); |
732 | ||
cee87338 BP |
733 | netdev_close(netdev_); |
734 | } | |
735 | shash_destroy(&device_shash); | |
736 | } else if (error != EAGAIN) { | |
7ed58d4a JP |
737 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5); |
738 | VLOG_WARN_RL(&rll, "error reading or parsing netlink (%s)", | |
cee87338 BP |
739 | ovs_strerror(error)); |
740 | } | |
741 | ofpbuf_uninit(&buf); | |
742 | } while (!error); | |
8b61709d BP |
743 | } |
744 | ||
745 | static void | |
1c33f0c3 | 746 | netdev_linux_wait(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 747 | { |
cee87338 BP |
748 | struct nl_sock *sock; |
749 | ||
19c8e9c1 JS |
750 | if (netdev_linux_miimon_enabled()) { |
751 | netdev_linux_miimon_wait(); | |
752 | } | |
cee87338 BP |
753 | sock = netdev_linux_notify_sock(); |
754 | if (sock) { | |
755 | nl_sock_wait(sock, POLLIN); | |
756 | } | |
8b61709d BP |
757 | } |
758 | ||
ac4d3bcb | 759 | static void |
b5d57fc8 BP |
760 | netdev_linux_changed(struct netdev_linux *dev, |
761 | unsigned int ifi_flags, unsigned int mask) | |
86383816 | 762 | OVS_REQUIRES(dev->mutex) |
ac4d3bcb | 763 | { |
3e912ffc | 764 | netdev_change_seq_changed(&dev->up); |
8aa77183 BP |
765 | |
766 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
767 | dev->carrier_resets++; | |
768 | } | |
769 | dev->ifi_flags = ifi_flags; | |
770 | ||
4f925bd3 | 771 | dev->cache_valid &= mask; |
6b6e1329 | 772 | if (!(mask & VALID_IN)) { |
a8704b50 PS |
773 | netdev_get_addrs_list_flush(); |
774 | } | |
4f925bd3 PS |
775 | } |
776 | ||
777 | static void | |
bfda5239 FL |
778 | netdev_linux_update__(struct netdev_linux *dev, |
779 | const struct rtnetlink_change *change) | |
86383816 | 780 | OVS_REQUIRES(dev->mutex) |
4f925bd3 | 781 | { |
bfda5239 | 782 | if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)) { |
d6384a3a | 783 | if (change->nlmsg_type == RTM_NEWLINK) { |
6b6e1329 | 784 | /* Keep drv-info, and ip addresses. */ |
d6384a3a | 785 | netdev_linux_changed(dev, change->ifi_flags, |
6b6e1329 | 786 | VALID_DRVINFO | VALID_IN); |
d6384a3a AW |
787 | |
788 | /* Update netdev from rtnl-change msg. */ | |
789 | if (change->mtu) { | |
790 | dev->mtu = change->mtu; | |
791 | dev->cache_valid |= VALID_MTU; | |
792 | dev->netdev_mtu_error = 0; | |
793 | } | |
90a6637d | 794 | |
74ff3298 JR |
795 | if (!eth_addr_is_zero(change->mac)) { |
796 | dev->etheraddr = change->mac; | |
d6384a3a AW |
797 | dev->cache_valid |= VALID_ETHERADDR; |
798 | dev->ether_addr_error = 0; | |
e8e1a409 TZ |
799 | |
800 | /* The mac addr has been changed, report it now. */ | |
801 | rtnetlink_report_link(); | |
d6384a3a | 802 | } |
44445cac | 803 | |
d6384a3a AW |
804 | dev->ifindex = change->if_index; |
805 | dev->cache_valid |= VALID_IFINDEX; | |
806 | dev->get_ifindex_error = 0; | |
22dcb534 | 807 | dev->present = true; |
d6384a3a | 808 | } else { |
bfda5239 | 809 | /* FIXME */ |
d6384a3a | 810 | netdev_linux_changed(dev, change->ifi_flags, 0); |
22dcb534 | 811 | dev->present = false; |
bfda5239 | 812 | netnsid_unset(&dev->netnsid); |
d6384a3a AW |
813 | } |
814 | } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) { | |
815 | /* Invalidates in4, in6. */ | |
6b6e1329 | 816 | netdev_linux_changed(dev, dev->ifi_flags, ~VALID_IN); |
4f925bd3 | 817 | } else { |
d6384a3a | 818 | OVS_NOT_REACHED(); |
4f925bd3 | 819 | } |
ac4d3bcb EJ |
820 | } |
821 | ||
bfda5239 FL |
822 | static void |
823 | netdev_linux_update(struct netdev_linux *dev, int nsid, | |
824 | const struct rtnetlink_change *change) | |
825 | OVS_REQUIRES(dev->mutex) | |
826 | { | |
827 | if (netdev_linux_netnsid_is_eq(dev, nsid)) { | |
828 | netdev_linux_update__(dev, change); | |
829 | } | |
830 | } | |
831 | ||
9dc63482 BP |
832 | static struct netdev * |
833 | netdev_linux_alloc(void) | |
834 | { | |
835 | struct netdev_linux *netdev = xzalloc(sizeof *netdev); | |
836 | return &netdev->up; | |
837 | } | |
838 | ||
48c6733c WT |
839 | static int |
840 | netdev_linux_common_construct(struct netdev *netdev_) | |
9dc63482 | 841 | { |
48c6733c WT |
842 | /* Prevent any attempt to create (or open) a network device named "default" |
843 | * or "all". These device names are effectively reserved on Linux because | |
844 | * /proc/sys/net/ipv4/conf/ always contains directories by these names. By | |
845 | * itself this wouldn't call for any special treatment, but in practice if | |
846 | * a program tries to create devices with these names, it causes the kernel | |
847 | * to fire a "new device" notification event even though creation failed, | |
848 | * and in turn that causes OVS to wake up and try to create them again, | |
849 | * which ends up as a 100% CPU loop. */ | |
850 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
851 | const char *name = netdev_->name; | |
852 | if (!strcmp(name, "default") || !strcmp(name, "all")) { | |
7ed58d4a JP |
853 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 1); |
854 | VLOG_WARN_RL(&rll, "%s: Linux forbids network device with this name", | |
48c6733c WT |
855 | name); |
856 | return EINVAL; | |
857 | } | |
858 | ||
bfda5239 FL |
859 | /* The device could be in the same network namespace or in another one. */ |
860 | netnsid_unset(&netdev->netnsid); | |
834d6caf | 861 | ovs_mutex_init(&netdev->mutex); |
48c6733c | 862 | return 0; |
9dc63482 BP |
863 | } |
864 | ||
1f6e0fbd BP |
865 | /* Creates system and internal devices. */ |
866 | static int | |
9dc63482 | 867 | netdev_linux_construct(struct netdev *netdev_) |
1f6e0fbd | 868 | { |
9dc63482 | 869 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
48c6733c WT |
870 | int error = netdev_linux_common_construct(netdev_); |
871 | if (error) { | |
872 | return error; | |
873 | } | |
1f6e0fbd | 874 | |
b5d57fc8 BP |
875 | error = get_flags(&netdev->up, &netdev->ifi_flags); |
876 | if (error == ENODEV) { | |
9dc63482 | 877 | if (netdev->up.netdev_class != &netdev_internal_class) { |
b5d57fc8 | 878 | /* The device does not exist, so don't allow it to be opened. */ |
b5d57fc8 BP |
879 | return ENODEV; |
880 | } else { | |
881 | /* "Internal" netdevs have to be created as netdev objects before | |
882 | * they exist in the kernel, because creating them in the kernel | |
883 | * happens by passing a netdev object to dpif_port_add(). | |
884 | * Therefore, ignore the error. */ | |
885 | } | |
886 | } | |
46415c90 | 887 | |
a740f0de JG |
888 | return 0; |
889 | } | |
890 | ||
5b7448ed JG |
891 | /* For most types of netdevs we open the device for each call of |
892 | * netdev_open(). However, this is not the case with tap devices, | |
893 | * since it is only possible to open the device once. In this | |
894 | * situation we share a single file descriptor, and consequently | |
895 | * buffers, across all readers. Therefore once data is read it will | |
896 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 897 | static int |
9dc63482 | 898 | netdev_linux_construct_tap(struct netdev *netdev_) |
a740f0de | 899 | { |
9dc63482 | 900 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a740f0de | 901 | static const char tap_dev[] = "/dev/net/tun"; |
9dc63482 | 902 | const char *name = netdev_->name; |
a740f0de | 903 | struct ifreq ifr; |
a740f0de | 904 | |
48c6733c WT |
905 | int error = netdev_linux_common_construct(netdev_); |
906 | if (error) { | |
907 | return error; | |
908 | } | |
1f6e0fbd | 909 | |
6c88d577 | 910 | /* Open tap device. */ |
d0d08f8a BP |
911 | netdev->tap_fd = open(tap_dev, O_RDWR); |
912 | if (netdev->tap_fd < 0) { | |
6c88d577 | 913 | error = errno; |
10a89ef0 | 914 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); |
cee87338 | 915 | return error; |
6c88d577 JP |
916 | } |
917 | ||
918 | /* Create tap device. */ | |
61b9d078 | 919 | get_flags(&netdev->up, &netdev->ifi_flags); |
6c88d577 | 920 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; |
71d7c22f | 921 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
d0d08f8a | 922 | if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 | 923 | VLOG_WARN("%s: creating tap device failed: %s", name, |
10a89ef0 | 924 | ovs_strerror(errno)); |
6c88d577 | 925 | error = errno; |
f61d8d29 | 926 | goto error_close; |
6c88d577 JP |
927 | } |
928 | ||
929 | /* Make non-blocking. */ | |
d0d08f8a | 930 | error = set_nonblocking(netdev->tap_fd); |
a740f0de | 931 | if (error) { |
f61d8d29 | 932 | goto error_close; |
a740f0de JG |
933 | } |
934 | ||
0f28164b FL |
935 | if (ioctl(netdev->tap_fd, TUNSETPERSIST, 1)) { |
936 | VLOG_WARN("%s: creating tap device failed (persist): %s", name, | |
937 | ovs_strerror(errno)); | |
938 | error = errno; | |
939 | goto error_close; | |
940 | } | |
941 | ||
a740f0de JG |
942 | return 0; |
943 | ||
f61d8d29 | 944 | error_close: |
d0d08f8a | 945 | close(netdev->tap_fd); |
a740f0de JG |
946 | return error; |
947 | } | |
948 | ||
6c88d577 | 949 | static void |
9dc63482 | 950 | netdev_linux_destruct(struct netdev *netdev_) |
6c88d577 | 951 | { |
b5d57fc8 | 952 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
6c88d577 | 953 | |
b5d57fc8 BP |
954 | if (netdev->tc && netdev->tc->ops->tc_destroy) { |
955 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 BP |
956 | } |
957 | ||
d0d08f8a BP |
958 | if (netdev_get_class(netdev_) == &netdev_tap_class |
959 | && netdev->tap_fd >= 0) | |
960 | { | |
0f28164b | 961 | ioctl(netdev->tap_fd, TUNSETPERSIST, 0); |
d0d08f8a | 962 | close(netdev->tap_fd); |
6c88d577 | 963 | } |
86383816 | 964 | |
19c8e9c1 | 965 | if (netdev->miimon_interval > 0) { |
812c272c | 966 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
967 | } |
968 | ||
86383816 | 969 | ovs_mutex_destroy(&netdev->mutex); |
6c88d577 JP |
970 | } |
971 | ||
9dc63482 BP |
972 | static void |
973 | netdev_linux_dealloc(struct netdev *netdev_) | |
974 | { | |
975 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
976 | free(netdev); | |
977 | } | |
978 | ||
f7791740 PS |
979 | static struct netdev_rxq * |
980 | netdev_linux_rxq_alloc(void) | |
9dc63482 | 981 | { |
f7791740 | 982 | struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); |
9dc63482 BP |
983 | return &rx->up; |
984 | } | |
985 | ||
7b6b0ef4 | 986 | static int |
f7791740 | 987 | netdev_linux_rxq_construct(struct netdev_rxq *rxq_) |
7b6b0ef4 | 988 | { |
f7791740 | 989 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 990 | struct netdev *netdev_ = rx->up.netdev; |
7b6b0ef4 | 991 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7b6b0ef4 | 992 | int error; |
7b6b0ef4 | 993 | |
86383816 | 994 | ovs_mutex_lock(&netdev->mutex); |
9dc63482 BP |
995 | rx->is_tap = is_tap_netdev(netdev_); |
996 | if (rx->is_tap) { | |
997 | rx->fd = netdev->tap_fd; | |
796223f5 BP |
998 | } else { |
999 | struct sockaddr_ll sll; | |
b73c8518 | 1000 | int ifindex, val; |
32383c3b | 1001 | /* Result of tcpdump -dd inbound */ |
259e0b1a | 1002 | static const struct sock_filter filt[] = { |
32383c3b MM |
1003 | { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ |
1004 | { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ | |
1005 | { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ | |
1006 | { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ | |
1007 | }; | |
259e0b1a BP |
1008 | static const struct sock_fprog fprog = { |
1009 | ARRAY_SIZE(filt), (struct sock_filter *) filt | |
1010 | }; | |
7b6b0ef4 | 1011 | |
796223f5 | 1012 | /* Create file descriptor. */ |
9dc63482 BP |
1013 | rx->fd = socket(PF_PACKET, SOCK_RAW, 0); |
1014 | if (rx->fd < 0) { | |
796223f5 | 1015 | error = errno; |
10a89ef0 | 1016 | VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); |
796223f5 BP |
1017 | goto error; |
1018 | } | |
33d82a56 | 1019 | |
b73c8518 SH |
1020 | val = 1; |
1021 | if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { | |
1022 | error = errno; | |
1023 | VLOG_ERR("%s: failed to mark socket for auxdata (%s)", | |
1024 | netdev_get_name(netdev_), ovs_strerror(error)); | |
1025 | goto error; | |
1026 | } | |
1027 | ||
796223f5 | 1028 | /* Set non-blocking mode. */ |
9dc63482 | 1029 | error = set_nonblocking(rx->fd); |
796223f5 BP |
1030 | if (error) { |
1031 | goto error; | |
1032 | } | |
7b6b0ef4 | 1033 | |
796223f5 | 1034 | /* Get ethernet device index. */ |
180c6d0b | 1035 | error = get_ifindex(&netdev->up, &ifindex); |
796223f5 BP |
1036 | if (error) { |
1037 | goto error; | |
1038 | } | |
7b6b0ef4 | 1039 | |
796223f5 BP |
1040 | /* Bind to specific ethernet device. */ |
1041 | memset(&sll, 0, sizeof sll); | |
1042 | sll.sll_family = AF_PACKET; | |
1043 | sll.sll_ifindex = ifindex; | |
b73c8518 | 1044 | sll.sll_protocol = htons(ETH_P_ALL); |
9dc63482 | 1045 | if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { |
796223f5 BP |
1046 | error = errno; |
1047 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
10a89ef0 | 1048 | netdev_get_name(netdev_), ovs_strerror(error)); |
796223f5 BP |
1049 | goto error; |
1050 | } | |
32383c3b MM |
1051 | |
1052 | /* Filter for only inbound packets. */ | |
9dc63482 | 1053 | error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, |
32383c3b MM |
1054 | sizeof fprog); |
1055 | if (error) { | |
1056 | error = errno; | |
259e0b1a | 1057 | VLOG_ERR("%s: failed to attach filter (%s)", |
10a89ef0 | 1058 | netdev_get_name(netdev_), ovs_strerror(error)); |
32383c3b MM |
1059 | goto error; |
1060 | } | |
7b6b0ef4 | 1061 | } |
86383816 | 1062 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 | 1063 | |
7b6b0ef4 BP |
1064 | return 0; |
1065 | ||
1066 | error: | |
9dc63482 BP |
1067 | if (rx->fd >= 0) { |
1068 | close(rx->fd); | |
7b6b0ef4 | 1069 | } |
86383816 | 1070 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 BP |
1071 | return error; |
1072 | } | |
1073 | ||
796223f5 | 1074 | static void |
f7791740 | 1075 | netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) |
8b61709d | 1076 | { |
f7791740 | 1077 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
8b61709d | 1078 | |
796223f5 BP |
1079 | if (!rx->is_tap) { |
1080 | close(rx->fd); | |
8b61709d | 1081 | } |
9dc63482 BP |
1082 | } |
1083 | ||
1084 | static void | |
f7791740 | 1085 | netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) |
9dc63482 | 1086 | { |
f7791740 | 1087 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 1088 | |
796223f5 BP |
1089 | free(rx); |
1090 | } | |
8b61709d | 1091 | |
b73c8518 | 1092 | static ovs_be16 |
1ebdc7eb | 1093 | auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux, bool double_tagged) |
b73c8518 SH |
1094 | { |
1095 | if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { | |
1096 | return htons(aux->tp_vlan_tpid); | |
1ebdc7eb EG |
1097 | } else if (double_tagged) { |
1098 | return htons(ETH_TYPE_VLAN_8021AD); | |
b73c8518 | 1099 | } else { |
1ebdc7eb | 1100 | return htons(ETH_TYPE_VLAN_8021Q); |
b73c8518 SH |
1101 | } |
1102 | } | |
1103 | ||
1104 | static bool | |
1105 | auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) | |
1106 | { | |
1107 | return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; | |
1108 | } | |
1109 | ||
796223f5 | 1110 | static int |
cf62fa4c | 1111 | netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer) |
796223f5 | 1112 | { |
b73c8518 | 1113 | size_t size; |
796223f5 | 1114 | ssize_t retval; |
b73c8518 SH |
1115 | struct iovec iov; |
1116 | struct cmsghdr *cmsg; | |
1117 | union { | |
1118 | struct cmsghdr cmsg; | |
1119 | char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; | |
1120 | } cmsg_buffer; | |
1121 | struct msghdr msgh; | |
1122 | ||
1123 | /* Reserve headroom for a single VLAN tag */ | |
cf62fa4c PS |
1124 | dp_packet_reserve(buffer, VLAN_HEADER_LEN); |
1125 | size = dp_packet_tailroom(buffer); | |
b73c8518 | 1126 | |
cf62fa4c | 1127 | iov.iov_base = dp_packet_data(buffer); |
b73c8518 SH |
1128 | iov.iov_len = size; |
1129 | msgh.msg_name = NULL; | |
1130 | msgh.msg_namelen = 0; | |
1131 | msgh.msg_iov = &iov; | |
1132 | msgh.msg_iovlen = 1; | |
1133 | msgh.msg_control = &cmsg_buffer; | |
1134 | msgh.msg_controllen = sizeof cmsg_buffer; | |
1135 | msgh.msg_flags = 0; | |
8e8cddf7 | 1136 | |
796223f5 | 1137 | do { |
b73c8518 | 1138 | retval = recvmsg(fd, &msgh, MSG_TRUNC); |
796223f5 BP |
1139 | } while (retval < 0 && errno == EINTR); |
1140 | ||
bfd3367b | 1141 | if (retval < 0) { |
b73c8518 SH |
1142 | return errno; |
1143 | } else if (retval > size) { | |
1144 | return EMSGSIZE; | |
1145 | } | |
1146 | ||
cf62fa4c | 1147 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1148 | |
1149 | for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) { | |
1150 | const struct tpacket_auxdata *aux; | |
1151 | ||
1152 | if (cmsg->cmsg_level != SOL_PACKET | |
1153 | || cmsg->cmsg_type != PACKET_AUXDATA | |
1154 | || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) { | |
1155 | continue; | |
8b61709d | 1156 | } |
b73c8518 SH |
1157 | |
1158 | aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); | |
1159 | if (auxdata_has_vlan_tci(aux)) { | |
1ebdc7eb EG |
1160 | struct eth_header *eth; |
1161 | bool double_tagged; | |
1162 | ||
b73c8518 SH |
1163 | if (retval < ETH_HEADER_LEN) { |
1164 | return EINVAL; | |
1165 | } | |
1166 | ||
1ebdc7eb EG |
1167 | eth = dp_packet_data(buffer); |
1168 | double_tagged = eth->eth_type == htons(ETH_TYPE_VLAN_8021Q); | |
1169 | ||
1170 | eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux, double_tagged), | |
b73c8518 SH |
1171 | htons(aux->tp_vlan_tci)); |
1172 | break; | |
1173 | } | |
1174 | } | |
1175 | ||
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | static int | |
cf62fa4c | 1180 | netdev_linux_rxq_recv_tap(int fd, struct dp_packet *buffer) |
b73c8518 SH |
1181 | { |
1182 | ssize_t retval; | |
cf62fa4c | 1183 | size_t size = dp_packet_tailroom(buffer); |
b73c8518 SH |
1184 | |
1185 | do { | |
cf62fa4c | 1186 | retval = read(fd, dp_packet_data(buffer), size); |
b73c8518 SH |
1187 | } while (retval < 0 && errno == EINTR); |
1188 | ||
1189 | if (retval < 0) { | |
bfd3367b | 1190 | return errno; |
8b61709d | 1191 | } |
b73c8518 | 1192 | |
cf62fa4c | 1193 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1194 | return 0; |
1195 | } | |
1196 | ||
1197 | static int | |
64839cf4 | 1198 | netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch) |
b73c8518 | 1199 | { |
f7791740 | 1200 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
df1e5a3b | 1201 | struct netdev *netdev = rx->up.netdev; |
cf62fa4c | 1202 | struct dp_packet *buffer; |
df1e5a3b PS |
1203 | ssize_t retval; |
1204 | int mtu; | |
1205 | ||
1206 | if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
1207 | mtu = ETH_PAYLOAD_MAX; | |
1208 | } | |
1209 | ||
2482b0b0 | 1210 | /* Assume Ethernet port. No need to set packet_type. */ |
cf62fa4c | 1211 | buffer = dp_packet_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, |
91088554 | 1212 | DP_NETDEV_HEADROOM); |
b73c8518 | 1213 | retval = (rx->is_tap |
f7791740 PS |
1214 | ? netdev_linux_rxq_recv_tap(rx->fd, buffer) |
1215 | : netdev_linux_rxq_recv_sock(rx->fd, buffer)); | |
df1e5a3b PS |
1216 | |
1217 | if (retval) { | |
1218 | if (retval != EAGAIN && retval != EMSGSIZE) { | |
1219 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
7c6401ca | 1220 | netdev_rxq_get_name(rxq_), ovs_strerror(errno)); |
df1e5a3b | 1221 | } |
cf62fa4c | 1222 | dp_packet_delete(buffer); |
df1e5a3b | 1223 | } else { |
72c84bc2 | 1224 | dp_packet_batch_init_packet(batch, buffer); |
b73c8518 SH |
1225 | } |
1226 | ||
1227 | return retval; | |
8b61709d BP |
1228 | } |
1229 | ||
8b61709d | 1230 | static void |
f7791740 | 1231 | netdev_linux_rxq_wait(struct netdev_rxq *rxq_) |
8b61709d | 1232 | { |
f7791740 | 1233 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1234 | poll_fd_wait(rx->fd, POLLIN); |
8b61709d BP |
1235 | } |
1236 | ||
8b61709d | 1237 | static int |
f7791740 | 1238 | netdev_linux_rxq_drain(struct netdev_rxq *rxq_) |
8b61709d | 1239 | { |
f7791740 | 1240 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1241 | if (rx->is_tap) { |
8b61709d | 1242 | struct ifreq ifr; |
f7791740 | 1243 | int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, |
259e0b1a | 1244 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
8b61709d BP |
1245 | if (error) { |
1246 | return error; | |
1247 | } | |
796223f5 | 1248 | drain_fd(rx->fd, ifr.ifr_qlen); |
8b61709d BP |
1249 | return 0; |
1250 | } else { | |
796223f5 | 1251 | return drain_rcvbuf(rx->fd); |
8b61709d BP |
1252 | } |
1253 | } | |
1254 | ||
d19cf8bb ZG |
1255 | static int |
1256 | netdev_linux_sock_batch_send(int sock, int ifindex, | |
1257 | struct dp_packet_batch *batch) | |
1258 | { | |
e0a00cee | 1259 | const size_t size = dp_packet_batch_size(batch); |
d19cf8bb ZG |
1260 | /* We don't bother setting most fields in sockaddr_ll because the |
1261 | * kernel ignores them for SOCK_RAW. */ | |
1262 | struct sockaddr_ll sll = { .sll_family = AF_PACKET, | |
1263 | .sll_ifindex = ifindex }; | |
1264 | ||
e0a00cee BB |
1265 | struct mmsghdr *mmsg = xmalloc(sizeof(*mmsg) * size); |
1266 | struct iovec *iov = xmalloc(sizeof(*iov) * size); | |
d19cf8bb | 1267 | |
e0a00cee | 1268 | struct dp_packet *packet; |
e883448e | 1269 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
d19cf8bb | 1270 | iov[i].iov_base = dp_packet_data(packet); |
ad8b0b4f | 1271 | iov[i].iov_len = dp_packet_size(packet); |
d19cf8bb ZG |
1272 | mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll, |
1273 | .msg_namelen = sizeof sll, | |
1274 | .msg_iov = &iov[i], | |
1275 | .msg_iovlen = 1 }; | |
1276 | } | |
1277 | ||
1278 | int error = 0; | |
e0a00cee | 1279 | for (uint32_t ofs = 0; ofs < size; ) { |
d19cf8bb ZG |
1280 | ssize_t retval; |
1281 | do { | |
e0a00cee | 1282 | retval = sendmmsg(sock, mmsg + ofs, size - ofs, 0); |
d19cf8bb ZG |
1283 | error = retval < 0 ? errno : 0; |
1284 | } while (error == EINTR); | |
1285 | if (error) { | |
1286 | break; | |
1287 | } | |
1288 | ofs += retval; | |
1289 | } | |
1290 | ||
1291 | free(mmsg); | |
1292 | free(iov); | |
1293 | return error; | |
1294 | } | |
1295 | ||
1296 | /* Use the tap fd to send 'batch' to tap device 'netdev'. Using the tap fd is | |
1297 | * essential, because packets sent to a tap device with an AF_PACKET socket | |
1298 | * will loop back to be *received* again on the tap device. This doesn't occur | |
1299 | * on other interface types because we attach a socket filter to the rx | |
1300 | * socket. */ | |
1301 | static int | |
1302 | netdev_linux_tap_batch_send(struct netdev *netdev_, | |
1303 | struct dp_packet_batch *batch) | |
1304 | { | |
1305 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
13708b21 | 1306 | struct dp_packet *packet; |
22dcb534 FL |
1307 | |
1308 | /* The Linux tap driver returns EIO if the device is not up, | |
1309 | * so if the device is not up, don't waste time sending it. | |
1310 | * However, if the device is in another network namespace | |
1311 | * then OVS can't retrieve the state. In that case, send the | |
1312 | * packets anyway. */ | |
1313 | if (netdev->present && !(netdev->ifi_flags & IFF_UP)) { | |
1314 | netdev->tx_dropped += dp_packet_batch_size(batch); | |
1315 | return 0; | |
1316 | } | |
1317 | ||
e883448e | 1318 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
ad8b0b4f | 1319 | size_t size = dp_packet_size(packet); |
d19cf8bb ZG |
1320 | ssize_t retval; |
1321 | int error; | |
1322 | ||
1323 | do { | |
1324 | retval = write(netdev->tap_fd, dp_packet_data(packet), size); | |
1325 | error = retval < 0 ? errno : 0; | |
1326 | } while (error == EINTR); | |
1327 | ||
1328 | if (error) { | |
1329 | /* The Linux tap driver returns EIO if the device is not up. From | |
1330 | * the OVS side this is not an error, so we ignore it; otherwise, | |
1331 | * return the erro. */ | |
1332 | if (error != EIO) { | |
1333 | return error; | |
1334 | } | |
1335 | } else if (retval != size) { | |
1336 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" " | |
1337 | "bytes of %"PRIuSIZE") on %s", | |
1338 | retval, size, netdev_get_name(netdev_)); | |
1339 | return EMSGSIZE; | |
1340 | } | |
1341 | } | |
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | /* Sends 'batch' on 'netdev'. Returns 0 if successful, otherwise a positive | |
8b61709d BP |
1346 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued |
1347 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
1348 | * the packet is too big or too small to transmit on the device. | |
1349 | * | |
8b61709d BP |
1350 | * The kernel maintains a packet transmission queue, so the caller is not |
1351 | * expected to do additional queuing of packets. */ | |
1352 | static int | |
f00fa8cb | 1353 | netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED, |
b30896c9 | 1354 | struct dp_packet_batch *batch, |
324c8374 | 1355 | bool concurrent_txq OVS_UNUSED) |
8b61709d | 1356 | { |
f4fd623c | 1357 | int error = 0; |
0a62ae2c ZG |
1358 | int sock = 0; |
1359 | ||
0a62ae2c ZG |
1360 | if (!is_tap_netdev(netdev_)) { |
1361 | sock = af_packet_sock(); | |
1362 | if (sock < 0) { | |
1363 | error = -sock; | |
1364 | goto free_batch; | |
1365 | } | |
1366 | ||
1367 | int ifindex = netdev_get_ifindex(netdev_); | |
1368 | if (ifindex < 0) { | |
1369 | error = -ifindex; | |
1370 | goto free_batch; | |
1371 | } | |
1372 | ||
d19cf8bb ZG |
1373 | error = netdev_linux_sock_batch_send(sock, ifindex, batch); |
1374 | } else { | |
1375 | error = netdev_linux_tap_batch_send(netdev_, batch); | |
0a62ae2c | 1376 | } |
d19cf8bb ZG |
1377 | if (error) { |
1378 | if (error == ENOBUFS) { | |
1379 | /* The Linux AF_PACKET implementation never blocks waiting | |
1380 | * for room for packets, instead returning ENOBUFS. | |
1381 | * Translate this into EAGAIN for the caller. */ | |
1382 | error = EAGAIN; | |
f23347ea | 1383 | } else { |
f4fd623c DDP |
1384 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", |
1385 | netdev_get_name(netdev_), ovs_strerror(error)); | |
d19cf8bb | 1386 | } |
f4fd623c DDP |
1387 | } |
1388 | ||
0a62ae2c | 1389 | free_batch: |
b30896c9 | 1390 | dp_packet_delete_batch(batch, true); |
f4fd623c | 1391 | return error; |
8b61709d BP |
1392 | } |
1393 | ||
1394 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
1395 | * when the packet transmission queue has sufficient room to transmit a packet | |
1396 | * with netdev_send(). | |
1397 | * | |
1398 | * The kernel maintains a packet transmission queue, so the client is not | |
1399 | * expected to do additional queuing of packets. Thus, this function is | |
1400 | * unlikely to ever be used. It is included for completeness. */ | |
1401 | static void | |
f00fa8cb | 1402 | netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED) |
8b61709d | 1403 | { |
796223f5 | 1404 | if (is_tap_netdev(netdev)) { |
8b61709d BP |
1405 | /* TAP device always accepts packets.*/ |
1406 | poll_immediate_wake(); | |
1407 | } | |
1408 | } | |
1409 | ||
1410 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
1411 | * otherwise a positive errno value. */ | |
1412 | static int | |
74ff3298 | 1413 | netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac) |
8b61709d | 1414 | { |
b5d57fc8 | 1415 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4f9f3f21 | 1416 | enum netdev_flags old_flags = 0; |
eb395f2e BP |
1417 | int error; |
1418 | ||
86383816 BP |
1419 | ovs_mutex_lock(&netdev->mutex); |
1420 | ||
b5d57fc8 | 1421 | if (netdev->cache_valid & VALID_ETHERADDR) { |
86383816 BP |
1422 | error = netdev->ether_addr_error; |
1423 | if (error || eth_addr_equals(netdev->etheraddr, mac)) { | |
1424 | goto exit; | |
44445cac | 1425 | } |
b5d57fc8 | 1426 | netdev->cache_valid &= ~VALID_ETHERADDR; |
44445cac PS |
1427 | } |
1428 | ||
7eb1bd81 | 1429 | /* Tap devices must be brought down before setting the address. */ |
796223f5 | 1430 | if (is_tap_netdev(netdev_)) { |
4f9f3f21 | 1431 | update_flags(netdev, NETDEV_UP, 0, &old_flags); |
7eb1bd81 | 1432 | } |
44445cac PS |
1433 | error = set_etheraddr(netdev_get_name(netdev_), mac); |
1434 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
1435 | netdev->ether_addr_error = error; |
1436 | netdev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 1437 | if (!error) { |
74ff3298 | 1438 | netdev->etheraddr = mac; |
eb395f2e | 1439 | } |
8b61709d | 1440 | } |
44445cac | 1441 | |
4f9f3f21 BP |
1442 | if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { |
1443 | update_flags(netdev, 0, NETDEV_UP, &old_flags); | |
1444 | } | |
7eb1bd81 | 1445 | |
86383816 BP |
1446 | exit: |
1447 | ovs_mutex_unlock(&netdev->mutex); | |
8b61709d BP |
1448 | return error; |
1449 | } | |
1450 | ||
44445cac | 1451 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d | 1452 | static int |
74ff3298 | 1453 | netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac) |
8b61709d | 1454 | { |
b5d57fc8 | 1455 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1456 | int error; |
44445cac | 1457 | |
86383816 | 1458 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1459 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { |
756819dd FL |
1460 | netdev_linux_update_via_netlink(netdev); |
1461 | } | |
1462 | ||
1463 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { | |
1464 | /* Fall back to ioctl if netlink fails */ | |
86383816 | 1465 | netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), |
74ff3298 | 1466 | &netdev->etheraddr); |
b5d57fc8 | 1467 | netdev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 1468 | } |
44445cac | 1469 | |
86383816 BP |
1470 | error = netdev->ether_addr_error; |
1471 | if (!error) { | |
74ff3298 | 1472 | *mac = netdev->etheraddr; |
44445cac | 1473 | } |
86383816 | 1474 | ovs_mutex_unlock(&netdev->mutex); |
44445cac | 1475 | |
86383816 | 1476 | return error; |
8b61709d BP |
1477 | } |
1478 | ||
8b61709d | 1479 | static int |
73371c09 | 1480 | netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) |
8b61709d | 1481 | { |
86383816 BP |
1482 | int error; |
1483 | ||
b5d57fc8 | 1484 | if (!(netdev->cache_valid & VALID_MTU)) { |
756819dd FL |
1485 | netdev_linux_update_via_netlink(netdev); |
1486 | } | |
1487 | ||
1488 | if (!(netdev->cache_valid & VALID_MTU)) { | |
1489 | /* Fall back to ioctl if netlink fails */ | |
8b61709d | 1490 | struct ifreq ifr; |
90a6637d | 1491 | |
86383816 | 1492 | netdev->netdev_mtu_error = af_inet_ifreq_ioctl( |
73371c09 | 1493 | netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); |
b5d57fc8 BP |
1494 | netdev->mtu = ifr.ifr_mtu; |
1495 | netdev->cache_valid |= VALID_MTU; | |
8b61709d | 1496 | } |
90a6637d | 1497 | |
86383816 BP |
1498 | error = netdev->netdev_mtu_error; |
1499 | if (!error) { | |
b5d57fc8 | 1500 | *mtup = netdev->mtu; |
90a6637d | 1501 | } |
73371c09 BP |
1502 | |
1503 | return error; | |
1504 | } | |
1505 | ||
1506 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1507 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1508 | * bytes for Ethernet devices. */ | |
1509 | static int | |
1510 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1511 | { | |
1512 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1513 | int error; | |
1514 | ||
1515 | ovs_mutex_lock(&netdev->mutex); | |
1516 | error = netdev_linux_get_mtu__(netdev, mtup); | |
86383816 BP |
1517 | ovs_mutex_unlock(&netdev->mutex); |
1518 | ||
1519 | return error; | |
8b61709d BP |
1520 | } |
1521 | ||
9b020780 PS |
1522 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1523 | * networking ioctl interface. | |
1524 | */ | |
1525 | static int | |
4124cb12 | 1526 | netdev_linux_set_mtu(struct netdev *netdev_, int mtu) |
9b020780 | 1527 | { |
b5d57fc8 | 1528 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9b020780 PS |
1529 | struct ifreq ifr; |
1530 | int error; | |
1531 | ||
86383816 | 1532 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1533 | if (netdev->cache_valid & VALID_MTU) { |
86383816 BP |
1534 | error = netdev->netdev_mtu_error; |
1535 | if (error || netdev->mtu == mtu) { | |
1536 | goto exit; | |
90a6637d | 1537 | } |
b5d57fc8 | 1538 | netdev->cache_valid &= ~VALID_MTU; |
153e5481 | 1539 | } |
9b020780 | 1540 | ifr.ifr_mtu = mtu; |
259e0b1a BP |
1541 | error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, |
1542 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d | 1543 | if (!error || error == ENODEV) { |
b5d57fc8 BP |
1544 | netdev->netdev_mtu_error = error; |
1545 | netdev->mtu = ifr.ifr_mtu; | |
1546 | netdev->cache_valid |= VALID_MTU; | |
9b020780 | 1547 | } |
86383816 BP |
1548 | exit: |
1549 | ovs_mutex_unlock(&netdev->mutex); | |
90a6637d | 1550 | return error; |
9b020780 PS |
1551 | } |
1552 | ||
9ab3d9a3 BP |
1553 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1554 | * On failure, returns a negative errno value. */ | |
1555 | static int | |
86383816 | 1556 | netdev_linux_get_ifindex(const struct netdev *netdev_) |
9ab3d9a3 | 1557 | { |
86383816 | 1558 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9ab3d9a3 BP |
1559 | int ifindex, error; |
1560 | ||
86383816 BP |
1561 | ovs_mutex_lock(&netdev->mutex); |
1562 | error = get_ifindex(netdev_, &ifindex); | |
1563 | ovs_mutex_unlock(&netdev->mutex); | |
1564 | ||
9ab3d9a3 BP |
1565 | return error ? -error : ifindex; |
1566 | } | |
1567 | ||
8b61709d BP |
1568 | static int |
1569 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1570 | { | |
b5d57fc8 | 1571 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1572 | |
86383816 | 1573 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
1574 | if (netdev->miimon_interval > 0) { |
1575 | *carrier = netdev->miimon; | |
3a183124 | 1576 | } else { |
b5d57fc8 | 1577 | *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1578 | } |
86383816 | 1579 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d | 1580 | |
3a183124 | 1581 | return 0; |
8b61709d BP |
1582 | } |
1583 | ||
65c3058c | 1584 | static long long int |
86383816 | 1585 | netdev_linux_get_carrier_resets(const struct netdev *netdev_) |
65c3058c | 1586 | { |
86383816 BP |
1587 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1588 | long long int carrier_resets; | |
1589 | ||
1590 | ovs_mutex_lock(&netdev->mutex); | |
1591 | carrier_resets = netdev->carrier_resets; | |
1592 | ovs_mutex_unlock(&netdev->mutex); | |
1593 | ||
1594 | return carrier_resets; | |
65c3058c EJ |
1595 | } |
1596 | ||
63331829 | 1597 | static int |
1670c579 EJ |
1598 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1599 | struct mii_ioctl_data *data) | |
63331829 | 1600 | { |
63331829 | 1601 | struct ifreq ifr; |
782e6111 | 1602 | int error; |
63331829 | 1603 | |
63331829 | 1604 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1605 | memcpy(&ifr.ifr_data, data, sizeof *data); |
259e0b1a | 1606 | error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1607 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1608 | |
782e6111 EJ |
1609 | return error; |
1610 | } | |
1611 | ||
1612 | static int | |
1670c579 | 1613 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1614 | { |
782e6111 EJ |
1615 | struct mii_ioctl_data data; |
1616 | int error; | |
63331829 | 1617 | |
782e6111 EJ |
1618 | *miimon = false; |
1619 | ||
1620 | memset(&data, 0, sizeof data); | |
1670c579 | 1621 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
1622 | if (!error) { |
1623 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1624 | data.reg_num = MII_BMSR; | |
1670c579 | 1625 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 1626 | &data); |
63331829 EJ |
1627 | |
1628 | if (!error) { | |
782e6111 | 1629 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 | 1630 | } |
9120cfc0 DH |
1631 | } |
1632 | if (error) { | |
63331829 | 1633 | struct ethtool_cmd ecmd; |
63331829 EJ |
1634 | |
1635 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1636 | name); | |
1637 | ||
ab985a77 | 1638 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
1639 | memset(&ecmd, 0, sizeof ecmd); |
1640 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1641 | "ETHTOOL_GLINK"); | |
1642 | if (!error) { | |
782e6111 EJ |
1643 | struct ethtool_value eval; |
1644 | ||
1645 | memcpy(&eval, &ecmd, sizeof eval); | |
1646 | *miimon = !!eval.data; | |
63331829 EJ |
1647 | } else { |
1648 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1649 | } | |
1650 | } | |
1651 | ||
1652 | return error; | |
1653 | } | |
1654 | ||
1670c579 EJ |
1655 | static int |
1656 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
1657 | long long int interval) | |
1658 | { | |
b5d57fc8 | 1659 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1670c579 | 1660 | |
86383816 | 1661 | ovs_mutex_lock(&netdev->mutex); |
1670c579 | 1662 | interval = interval > 0 ? MAX(interval, 100) : 0; |
b5d57fc8 | 1663 | if (netdev->miimon_interval != interval) { |
19c8e9c1 | 1664 | if (interval && !netdev->miimon_interval) { |
812c272c | 1665 | atomic_count_inc(&miimon_cnt); |
19c8e9c1 | 1666 | } else if (!interval && netdev->miimon_interval) { |
812c272c | 1667 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
1668 | } |
1669 | ||
b5d57fc8 BP |
1670 | netdev->miimon_interval = interval; |
1671 | timer_set_expired(&netdev->miimon_timer); | |
1670c579 | 1672 | } |
86383816 | 1673 | ovs_mutex_unlock(&netdev->mutex); |
1670c579 EJ |
1674 | |
1675 | return 0; | |
1676 | } | |
1677 | ||
1678 | static void | |
1679 | netdev_linux_miimon_run(void) | |
1680 | { | |
1681 | struct shash device_shash; | |
1682 | struct shash_node *node; | |
1683 | ||
1684 | shash_init(&device_shash); | |
b5d57fc8 | 1685 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1686 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1687 | struct netdev *netdev = node->data; |
1688 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
1689 | bool miimon; |
1690 | ||
86383816 BP |
1691 | ovs_mutex_lock(&dev->mutex); |
1692 | if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { | |
1693 | netdev_linux_get_miimon(dev->up.name, &miimon); | |
1694 | if (miimon != dev->miimon) { | |
1695 | dev->miimon = miimon; | |
1696 | netdev_linux_changed(dev, dev->ifi_flags, 0); | |
1697 | } | |
1670c579 | 1698 | |
86383816 | 1699 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); |
1670c579 | 1700 | } |
86383816 | 1701 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1702 | netdev_close(netdev); |
1670c579 EJ |
1703 | } |
1704 | ||
1705 | shash_destroy(&device_shash); | |
1706 | } | |
1707 | ||
1708 | static void | |
1709 | netdev_linux_miimon_wait(void) | |
1710 | { | |
1711 | struct shash device_shash; | |
1712 | struct shash_node *node; | |
1713 | ||
1714 | shash_init(&device_shash); | |
b5d57fc8 | 1715 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1716 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1717 | struct netdev *netdev = node->data; |
1718 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 | 1719 | |
86383816 | 1720 | ovs_mutex_lock(&dev->mutex); |
1670c579 EJ |
1721 | if (dev->miimon_interval > 0) { |
1722 | timer_wait(&dev->miimon_timer); | |
1723 | } | |
86383816 | 1724 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1725 | netdev_close(netdev); |
1670c579 EJ |
1726 | } |
1727 | shash_destroy(&device_shash); | |
1728 | } | |
1729 | ||
92df599c JG |
1730 | static void |
1731 | swap_uint64(uint64_t *a, uint64_t *b) | |
1732 | { | |
1de0e8ae BP |
1733 | uint64_t tmp = *a; |
1734 | *a = *b; | |
1735 | *b = tmp; | |
92df599c JG |
1736 | } |
1737 | ||
c060c4cf EJ |
1738 | /* Copies 'src' into 'dst', performing format conversion in the process. |
1739 | * | |
1740 | * 'src' is allowed to be misaligned. */ | |
1741 | static void | |
1742 | netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, | |
1743 | const struct ovs_vport_stats *src) | |
1744 | { | |
6a54dedc BP |
1745 | dst->rx_packets = get_32aligned_u64(&src->rx_packets); |
1746 | dst->tx_packets = get_32aligned_u64(&src->tx_packets); | |
1747 | dst->rx_bytes = get_32aligned_u64(&src->rx_bytes); | |
1748 | dst->tx_bytes = get_32aligned_u64(&src->tx_bytes); | |
1749 | dst->rx_errors = get_32aligned_u64(&src->rx_errors); | |
1750 | dst->tx_errors = get_32aligned_u64(&src->tx_errors); | |
1751 | dst->rx_dropped = get_32aligned_u64(&src->rx_dropped); | |
1752 | dst->tx_dropped = get_32aligned_u64(&src->tx_dropped); | |
c060c4cf EJ |
1753 | dst->multicast = 0; |
1754 | dst->collisions = 0; | |
1755 | dst->rx_length_errors = 0; | |
1756 | dst->rx_over_errors = 0; | |
1757 | dst->rx_crc_errors = 0; | |
1758 | dst->rx_frame_errors = 0; | |
1759 | dst->rx_fifo_errors = 0; | |
1760 | dst->rx_missed_errors = 0; | |
1761 | dst->tx_aborted_errors = 0; | |
1762 | dst->tx_carrier_errors = 0; | |
1763 | dst->tx_fifo_errors = 0; | |
1764 | dst->tx_heartbeat_errors = 0; | |
1765 | dst->tx_window_errors = 0; | |
1766 | } | |
1767 | ||
1768 | static int | |
1769 | get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) | |
1770 | { | |
93451a0a | 1771 | struct dpif_netlink_vport reply; |
c060c4cf EJ |
1772 | struct ofpbuf *buf; |
1773 | int error; | |
1774 | ||
93451a0a | 1775 | error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf); |
c060c4cf EJ |
1776 | if (error) { |
1777 | return error; | |
1778 | } else if (!reply.stats) { | |
1779 | ofpbuf_delete(buf); | |
1780 | return EOPNOTSUPP; | |
1781 | } | |
1782 | ||
1783 | netdev_stats_from_ovs_vport_stats(stats, reply.stats); | |
1784 | ||
1785 | ofpbuf_delete(buf); | |
1786 | ||
1787 | return 0; | |
1788 | } | |
1789 | ||
f613a0d7 PS |
1790 | static void |
1791 | get_stats_via_vport(const struct netdev *netdev_, | |
1792 | struct netdev_stats *stats) | |
8b61709d | 1793 | { |
b5d57fc8 | 1794 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1795 | |
b5d57fc8 BP |
1796 | if (!netdev->vport_stats_error || |
1797 | !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 1798 | int error; |
7fbef77a | 1799 | |
c060c4cf | 1800 | error = get_stats_via_vport__(netdev_, stats); |
bb13fe5e | 1801 | if (error && error != ENOENT && error != ENODEV) { |
a57a8488 | 1802 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
10a89ef0 BP |
1803 | "(%s)", |
1804 | netdev_get_name(netdev_), ovs_strerror(error)); | |
f613a0d7 | 1805 | } |
b5d57fc8 BP |
1806 | netdev->vport_stats_error = error; |
1807 | netdev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 1808 | } |
f613a0d7 | 1809 | } |
8b61709d | 1810 | |
f613a0d7 PS |
1811 | /* Retrieves current device stats for 'netdev-linux'. */ |
1812 | static int | |
1813 | netdev_linux_get_stats(const struct netdev *netdev_, | |
1814 | struct netdev_stats *stats) | |
1815 | { | |
b5d57fc8 | 1816 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1817 | struct netdev_stats dev_stats; |
1818 | int error; | |
1819 | ||
86383816 | 1820 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1821 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1822 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1823 | if (error) { |
86383816 BP |
1824 | if (!netdev->vport_stats_error) { |
1825 | error = 0; | |
f613a0d7 | 1826 | } |
86383816 | 1827 | } else if (netdev->vport_stats_error) { |
04c881eb | 1828 | /* stats not available from OVS then use netdev stats. */ |
f613a0d7 PS |
1829 | *stats = dev_stats; |
1830 | } else { | |
04c881eb AZ |
1831 | /* Use kernel netdev's packet and byte counts since vport's counters |
1832 | * do not reflect packet counts on the wire when GSO, TSO or GRO are | |
1833 | * enabled. */ | |
1834 | stats->rx_packets = dev_stats.rx_packets; | |
1835 | stats->rx_bytes = dev_stats.rx_bytes; | |
1836 | stats->tx_packets = dev_stats.tx_packets; | |
1837 | stats->tx_bytes = dev_stats.tx_bytes; | |
1838 | ||
f613a0d7 PS |
1839 | stats->rx_errors += dev_stats.rx_errors; |
1840 | stats->tx_errors += dev_stats.tx_errors; | |
1841 | stats->rx_dropped += dev_stats.rx_dropped; | |
1842 | stats->tx_dropped += dev_stats.tx_dropped; | |
1843 | stats->multicast += dev_stats.multicast; | |
1844 | stats->collisions += dev_stats.collisions; | |
1845 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
1846 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
1847 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
1848 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
1849 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
1850 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
1851 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
1852 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
1853 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
1854 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
1855 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
1856 | } | |
86383816 BP |
1857 | ovs_mutex_unlock(&netdev->mutex); |
1858 | ||
1859 | return error; | |
f613a0d7 PS |
1860 | } |
1861 | ||
1862 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
1863 | * netdev-internal. */ | |
1864 | static int | |
15aee116 | 1865 | netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) |
f613a0d7 | 1866 | { |
b5d57fc8 | 1867 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1868 | struct netdev_stats dev_stats; |
1869 | int error; | |
1870 | ||
86383816 | 1871 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1872 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1873 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1874 | if (error) { |
86383816 BP |
1875 | if (!netdev->vport_stats_error) { |
1876 | error = 0; | |
8b61709d | 1877 | } |
86383816 BP |
1878 | } else if (netdev->vport_stats_error) { |
1879 | /* Transmit and receive stats will appear to be swapped relative to the | |
1880 | * other ports since we are the one sending the data, not a remote | |
1881 | * computer. For consistency, we swap them back here. This does not | |
1882 | * apply if we are getting stats from the vport layer because it always | |
1883 | * tracks stats from the perspective of the switch. */ | |
fe6b0e03 | 1884 | |
f613a0d7 | 1885 | *stats = dev_stats; |
92df599c JG |
1886 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1887 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1888 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1889 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1890 | stats->rx_length_errors = 0; |
1891 | stats->rx_over_errors = 0; | |
1892 | stats->rx_crc_errors = 0; | |
1893 | stats->rx_frame_errors = 0; | |
1894 | stats->rx_fifo_errors = 0; | |
1895 | stats->rx_missed_errors = 0; | |
1896 | stats->tx_aborted_errors = 0; | |
1897 | stats->tx_carrier_errors = 0; | |
1898 | stats->tx_fifo_errors = 0; | |
1899 | stats->tx_heartbeat_errors = 0; | |
1900 | stats->tx_window_errors = 0; | |
f613a0d7 | 1901 | } else { |
04c881eb AZ |
1902 | /* Use kernel netdev's packet and byte counts since vport counters |
1903 | * do not reflect packet counts on the wire when GSO, TSO or GRO | |
1904 | * are enabled. */ | |
1905 | stats->rx_packets = dev_stats.tx_packets; | |
1906 | stats->rx_bytes = dev_stats.tx_bytes; | |
1907 | stats->tx_packets = dev_stats.rx_packets; | |
1908 | stats->tx_bytes = dev_stats.rx_bytes; | |
1909 | ||
f613a0d7 PS |
1910 | stats->rx_dropped += dev_stats.tx_dropped; |
1911 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 1912 | |
f613a0d7 PS |
1913 | stats->rx_errors += dev_stats.tx_errors; |
1914 | stats->tx_errors += dev_stats.rx_errors; | |
1915 | ||
1916 | stats->multicast += dev_stats.multicast; | |
1917 | stats->collisions += dev_stats.collisions; | |
1918 | } | |
22dcb534 | 1919 | stats->tx_dropped += netdev->tx_dropped; |
86383816 BP |
1920 | ovs_mutex_unlock(&netdev->mutex); |
1921 | ||
1922 | return error; | |
8b61709d BP |
1923 | } |
1924 | ||
bba1e6f3 PS |
1925 | static int |
1926 | netdev_internal_get_stats(const struct netdev *netdev_, | |
1927 | struct netdev_stats *stats) | |
1928 | { | |
b5d57fc8 | 1929 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1930 | int error; |
bba1e6f3 | 1931 | |
86383816 | 1932 | ovs_mutex_lock(&netdev->mutex); |
bba1e6f3 | 1933 | get_stats_via_vport(netdev_, stats); |
86383816 BP |
1934 | error = netdev->vport_stats_error; |
1935 | ovs_mutex_unlock(&netdev->mutex); | |
1936 | ||
1937 | return error; | |
bba1e6f3 PS |
1938 | } |
1939 | ||
51f87458 | 1940 | static void |
b5d57fc8 | 1941 | netdev_linux_read_features(struct netdev_linux *netdev) |
8b61709d BP |
1942 | { |
1943 | struct ethtool_cmd ecmd; | |
6c038611 | 1944 | uint32_t speed; |
8b61709d BP |
1945 | int error; |
1946 | ||
b5d57fc8 | 1947 | if (netdev->cache_valid & VALID_FEATURES) { |
51f87458 PS |
1948 | return; |
1949 | } | |
1950 | ||
ab985a77 | 1951 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 1952 | memset(&ecmd, 0, sizeof ecmd); |
b5d57fc8 | 1953 | error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, |
8b61709d BP |
1954 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
1955 | if (error) { | |
51f87458 | 1956 | goto out; |
8b61709d BP |
1957 | } |
1958 | ||
1959 | /* Supported features. */ | |
b5d57fc8 | 1960 | netdev->supported = 0; |
8b61709d | 1961 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
b5d57fc8 | 1962 | netdev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
1963 | } |
1964 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
b5d57fc8 | 1965 | netdev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
1966 | } |
1967 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
b5d57fc8 | 1968 | netdev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
1969 | } |
1970 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
b5d57fc8 | 1971 | netdev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
1972 | } |
1973 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
b5d57fc8 | 1974 | netdev->supported |= NETDEV_F_1GB_HD; |
8b61709d | 1975 | } |
67bed84c SH |
1976 | if ((ecmd.supported & SUPPORTED_1000baseT_Full) || |
1977 | (ecmd.supported & SUPPORTED_1000baseKX_Full)) { | |
b5d57fc8 | 1978 | netdev->supported |= NETDEV_F_1GB_FD; |
8b61709d | 1979 | } |
67bed84c SH |
1980 | if ((ecmd.supported & SUPPORTED_10000baseT_Full) || |
1981 | (ecmd.supported & SUPPORTED_10000baseKX4_Full) || | |
1982 | (ecmd.supported & SUPPORTED_10000baseKR_Full) || | |
1983 | (ecmd.supported & SUPPORTED_10000baseR_FEC)) { | |
b5d57fc8 | 1984 | netdev->supported |= NETDEV_F_10GB_FD; |
8b61709d | 1985 | } |
67bed84c SH |
1986 | if ((ecmd.supported & SUPPORTED_40000baseKR4_Full) || |
1987 | (ecmd.supported & SUPPORTED_40000baseCR4_Full) || | |
1988 | (ecmd.supported & SUPPORTED_40000baseSR4_Full) || | |
1989 | (ecmd.supported & SUPPORTED_40000baseLR4_Full)) { | |
1990 | netdev->supported |= NETDEV_F_40GB_FD; | |
1991 | } | |
8b61709d | 1992 | if (ecmd.supported & SUPPORTED_TP) { |
b5d57fc8 | 1993 | netdev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
1994 | } |
1995 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
b5d57fc8 | 1996 | netdev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
1997 | } |
1998 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
b5d57fc8 | 1999 | netdev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
2000 | } |
2001 | if (ecmd.supported & SUPPORTED_Pause) { | |
b5d57fc8 | 2002 | netdev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
2003 | } |
2004 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
b5d57fc8 | 2005 | netdev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2006 | } |
2007 | ||
2008 | /* Advertised features. */ | |
b5d57fc8 | 2009 | netdev->advertised = 0; |
8b61709d | 2010 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
b5d57fc8 | 2011 | netdev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
2012 | } |
2013 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
b5d57fc8 | 2014 | netdev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
2015 | } |
2016 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
b5d57fc8 | 2017 | netdev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
2018 | } |
2019 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
b5d57fc8 | 2020 | netdev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
2021 | } |
2022 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
b5d57fc8 | 2023 | netdev->advertised |= NETDEV_F_1GB_HD; |
8b61709d | 2024 | } |
67bed84c SH |
2025 | if ((ecmd.advertising & ADVERTISED_1000baseT_Full) || |
2026 | (ecmd.advertising & ADVERTISED_1000baseKX_Full)) { | |
b5d57fc8 | 2027 | netdev->advertised |= NETDEV_F_1GB_FD; |
8b61709d | 2028 | } |
67bed84c SH |
2029 | if ((ecmd.advertising & ADVERTISED_10000baseT_Full) || |
2030 | (ecmd.advertising & ADVERTISED_10000baseKX4_Full) || | |
2031 | (ecmd.advertising & ADVERTISED_10000baseKR_Full) || | |
2032 | (ecmd.advertising & ADVERTISED_10000baseR_FEC)) { | |
b5d57fc8 | 2033 | netdev->advertised |= NETDEV_F_10GB_FD; |
8b61709d | 2034 | } |
67bed84c SH |
2035 | if ((ecmd.advertising & ADVERTISED_40000baseKR4_Full) || |
2036 | (ecmd.advertising & ADVERTISED_40000baseCR4_Full) || | |
2037 | (ecmd.advertising & ADVERTISED_40000baseSR4_Full) || | |
2038 | (ecmd.advertising & ADVERTISED_40000baseLR4_Full)) { | |
2039 | netdev->advertised |= NETDEV_F_40GB_FD; | |
2040 | } | |
8b61709d | 2041 | if (ecmd.advertising & ADVERTISED_TP) { |
b5d57fc8 | 2042 | netdev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
2043 | } |
2044 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
b5d57fc8 | 2045 | netdev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
2046 | } |
2047 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
b5d57fc8 | 2048 | netdev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
2049 | } |
2050 | if (ecmd.advertising & ADVERTISED_Pause) { | |
b5d57fc8 | 2051 | netdev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
2052 | } |
2053 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
b5d57fc8 | 2054 | netdev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2055 | } |
2056 | ||
2057 | /* Current settings. */ | |
0c615356 | 2058 | speed = ethtool_cmd_speed(&ecmd); |
6c038611 | 2059 | if (speed == SPEED_10) { |
b5d57fc8 | 2060 | netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 2061 | } else if (speed == SPEED_100) { |
b5d57fc8 | 2062 | netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 2063 | } else if (speed == SPEED_1000) { |
b5d57fc8 | 2064 | netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 2065 | } else if (speed == SPEED_10000) { |
b5d57fc8 | 2066 | netdev->current = NETDEV_F_10GB_FD; |
6c038611 | 2067 | } else if (speed == 40000) { |
b5d57fc8 | 2068 | netdev->current = NETDEV_F_40GB_FD; |
6c038611 | 2069 | } else if (speed == 100000) { |
b5d57fc8 | 2070 | netdev->current = NETDEV_F_100GB_FD; |
6c038611 | 2071 | } else if (speed == 1000000) { |
b5d57fc8 | 2072 | netdev->current = NETDEV_F_1TB_FD; |
8b61709d | 2073 | } else { |
b5d57fc8 | 2074 | netdev->current = 0; |
8b61709d BP |
2075 | } |
2076 | ||
2077 | if (ecmd.port == PORT_TP) { | |
b5d57fc8 | 2078 | netdev->current |= NETDEV_F_COPPER; |
8b61709d | 2079 | } else if (ecmd.port == PORT_FIBRE) { |
b5d57fc8 | 2080 | netdev->current |= NETDEV_F_FIBER; |
8b61709d BP |
2081 | } |
2082 | ||
2083 | if (ecmd.autoneg) { | |
b5d57fc8 | 2084 | netdev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
2085 | } |
2086 | ||
51f87458 | 2087 | out: |
b5d57fc8 BP |
2088 | netdev->cache_valid |= VALID_FEATURES; |
2089 | netdev->get_features_error = error; | |
51f87458 PS |
2090 | } |
2091 | ||
887ed8b2 BP |
2092 | /* Stores the features supported by 'netdev' into of '*current', '*advertised', |
2093 | * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. | |
2094 | * Returns 0 if successful, otherwise a positive errno value. */ | |
51f87458 PS |
2095 | static int |
2096 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
2097 | enum netdev_features *current, |
2098 | enum netdev_features *advertised, | |
2099 | enum netdev_features *supported, | |
2100 | enum netdev_features *peer) | |
51f87458 | 2101 | { |
b5d57fc8 | 2102 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 2103 | int error; |
51f87458 | 2104 | |
86383816 | 2105 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2106 | netdev_linux_read_features(netdev); |
b5d57fc8 BP |
2107 | if (!netdev->get_features_error) { |
2108 | *current = netdev->current; | |
2109 | *advertised = netdev->advertised; | |
2110 | *supported = netdev->supported; | |
887ed8b2 | 2111 | *peer = 0; /* XXX */ |
51f87458 | 2112 | } |
86383816 BP |
2113 | error = netdev->get_features_error; |
2114 | ovs_mutex_unlock(&netdev->mutex); | |
2115 | ||
2116 | return error; | |
8b61709d BP |
2117 | } |
2118 | ||
2119 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
2120 | static int | |
86383816 | 2121 | netdev_linux_set_advertisements(struct netdev *netdev_, |
6c038611 | 2122 | enum netdev_features advertise) |
8b61709d | 2123 | { |
86383816 | 2124 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2125 | struct ethtool_cmd ecmd; |
2126 | int error; | |
2127 | ||
86383816 BP |
2128 | ovs_mutex_lock(&netdev->mutex); |
2129 | ||
ab985a77 | 2130 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 2131 | memset(&ecmd, 0, sizeof ecmd); |
86383816 | 2132 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
8b61709d BP |
2133 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
2134 | if (error) { | |
86383816 | 2135 | goto exit; |
8b61709d BP |
2136 | } |
2137 | ||
2138 | ecmd.advertising = 0; | |
6c038611 | 2139 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
2140 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
2141 | } | |
6c038611 | 2142 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
2143 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
2144 | } | |
6c038611 | 2145 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
2146 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
2147 | } | |
6c038611 | 2148 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
2149 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
2150 | } | |
6c038611 | 2151 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
2152 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
2153 | } | |
6c038611 | 2154 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
2155 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
2156 | } | |
6c038611 | 2157 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
2158 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
2159 | } | |
6c038611 | 2160 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
2161 | ecmd.advertising |= ADVERTISED_TP; |
2162 | } | |
6c038611 | 2163 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
2164 | ecmd.advertising |= ADVERTISED_FIBRE; |
2165 | } | |
6c038611 | 2166 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
2167 | ecmd.advertising |= ADVERTISED_Autoneg; |
2168 | } | |
6c038611 | 2169 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
2170 | ecmd.advertising |= ADVERTISED_Pause; |
2171 | } | |
6c038611 | 2172 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
2173 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
2174 | } | |
ab985a77 | 2175 | COVERAGE_INC(netdev_set_ethtool); |
86383816 BP |
2176 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
2177 | ETHTOOL_SSET, "ETHTOOL_SSET"); | |
2178 | ||
2179 | exit: | |
2180 | ovs_mutex_unlock(&netdev->mutex); | |
2181 | return error; | |
8b61709d BP |
2182 | } |
2183 | ||
f8500004 JP |
2184 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
2185 | * successful, otherwise a positive errno value. */ | |
8b61709d | 2186 | static int |
b5d57fc8 | 2187 | netdev_linux_set_policing(struct netdev *netdev_, |
8b61709d BP |
2188 | uint32_t kbits_rate, uint32_t kbits_burst) |
2189 | { | |
b5d57fc8 BP |
2190 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
2191 | const char *netdev_name = netdev_get_name(netdev_); | |
7874bdff | 2192 | int ifindex; |
f8500004 | 2193 | int error; |
8b61709d | 2194 | |
d5ae4a60 PB |
2195 | if (netdev_is_flow_api_enabled()) { |
2196 | if (kbits_rate) { | |
2197 | VLOG_WARN_RL(&rl, "%s: policing with offload isn't supported", | |
2198 | netdev_name); | |
2199 | } | |
2200 | return EOPNOTSUPP; | |
2201 | } | |
2202 | ||
80a86fbe | 2203 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
79abacc8 | 2204 | : !kbits_burst ? 8000 /* Default to 8000 kbits if 0. */ |
80a86fbe BP |
2205 | : kbits_burst); /* Stick with user-specified value. */ |
2206 | ||
86383816 | 2207 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2208 | if (netdev->cache_valid & VALID_POLICING) { |
86383816 BP |
2209 | error = netdev->netdev_policing_error; |
2210 | if (error || (netdev->kbits_rate == kbits_rate && | |
2211 | netdev->kbits_burst == kbits_burst)) { | |
c9f71668 | 2212 | /* Assume that settings haven't changed since we last set them. */ |
86383816 | 2213 | goto out; |
c9f71668 | 2214 | } |
b5d57fc8 | 2215 | netdev->cache_valid &= ~VALID_POLICING; |
80a86fbe BP |
2216 | } |
2217 | ||
7874bdff RD |
2218 | error = get_ifindex(netdev_, &ifindex); |
2219 | if (error) { | |
2220 | goto out; | |
2221 | } | |
2222 | ||
ac8c3412 | 2223 | COVERAGE_INC(netdev_set_policing); |
f8500004 | 2224 | /* Remove any existing ingress qdisc. */ |
7874bdff | 2225 | error = tc_add_del_ingress_qdisc(ifindex, false); |
f8500004 JP |
2226 | if (error) { |
2227 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
10a89ef0 | 2228 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2229 | goto out; |
f8500004 JP |
2230 | } |
2231 | ||
8b61709d | 2232 | if (kbits_rate) { |
7874bdff | 2233 | error = tc_add_del_ingress_qdisc(ifindex, true); |
f8500004 JP |
2234 | if (error) { |
2235 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
10a89ef0 | 2236 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2237 | goto out; |
8b61709d BP |
2238 | } |
2239 | ||
b5d57fc8 | 2240 | error = tc_add_policer(netdev_, kbits_rate, kbits_burst); |
f8500004 JP |
2241 | if (error){ |
2242 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
10a89ef0 | 2243 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2244 | goto out; |
8b61709d | 2245 | } |
8b61709d BP |
2246 | } |
2247 | ||
b5d57fc8 BP |
2248 | netdev->kbits_rate = kbits_rate; |
2249 | netdev->kbits_burst = kbits_burst; | |
f8500004 | 2250 | |
c9f71668 PS |
2251 | out: |
2252 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
2253 | netdev->netdev_policing_error = error; |
2254 | netdev->cache_valid |= VALID_POLICING; | |
c9f71668 | 2255 | } |
86383816 | 2256 | ovs_mutex_unlock(&netdev->mutex); |
c9f71668 | 2257 | return error; |
8b61709d BP |
2258 | } |
2259 | ||
c1c9c9c4 BP |
2260 | static int |
2261 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 2262 | struct sset *types) |
c1c9c9c4 | 2263 | { |
559eb230 | 2264 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2265 | for (opsp = tcs; *opsp != NULL; opsp++) { |
2266 | const struct tc_ops *ops = *opsp; | |
2267 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 2268 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
2269 | } |
2270 | } | |
2271 | return 0; | |
2272 | } | |
2273 | ||
2274 | static const struct tc_ops * | |
2275 | tc_lookup_ovs_name(const char *name) | |
2276 | { | |
559eb230 | 2277 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2278 | |
2279 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2280 | const struct tc_ops *ops = *opsp; | |
2281 | if (!strcmp(name, ops->ovs_name)) { | |
2282 | return ops; | |
2283 | } | |
2284 | } | |
2285 | return NULL; | |
2286 | } | |
2287 | ||
2288 | static const struct tc_ops * | |
2289 | tc_lookup_linux_name(const char *name) | |
2290 | { | |
559eb230 | 2291 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2292 | |
2293 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2294 | const struct tc_ops *ops = *opsp; | |
2295 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
2296 | return ops; | |
2297 | } | |
2298 | } | |
2299 | return NULL; | |
2300 | } | |
2301 | ||
93b13be8 | 2302 | static struct tc_queue * |
b5d57fc8 | 2303 | tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, |
93b13be8 BP |
2304 | size_t hash) |
2305 | { | |
b5d57fc8 | 2306 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
93b13be8 BP |
2307 | struct tc_queue *queue; |
2308 | ||
b5d57fc8 | 2309 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { |
93b13be8 BP |
2310 | if (queue->queue_id == queue_id) { |
2311 | return queue; | |
2312 | } | |
2313 | } | |
2314 | return NULL; | |
2315 | } | |
2316 | ||
2317 | static struct tc_queue * | |
2318 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
2319 | { | |
2320 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
2321 | } | |
2322 | ||
c1c9c9c4 BP |
2323 | static int |
2324 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
2325 | const char *type, | |
2326 | struct netdev_qos_capabilities *caps) | |
2327 | { | |
2328 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
2329 | if (!ops) { | |
2330 | return EOPNOTSUPP; | |
2331 | } | |
2332 | caps->n_queues = ops->n_queues; | |
2333 | return 0; | |
2334 | } | |
2335 | ||
2336 | static int | |
b5d57fc8 | 2337 | netdev_linux_get_qos(const struct netdev *netdev_, |
79f1cbe9 | 2338 | const char **typep, struct smap *details) |
c1c9c9c4 | 2339 | { |
b5d57fc8 | 2340 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2341 | int error; |
2342 | ||
86383816 | 2343 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2344 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2345 | if (!error) { |
2346 | *typep = netdev->tc->ops->ovs_name; | |
2347 | error = (netdev->tc->ops->qdisc_get | |
2348 | ? netdev->tc->ops->qdisc_get(netdev_, details) | |
2349 | : 0); | |
c1c9c9c4 | 2350 | } |
86383816 | 2351 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2352 | |
86383816 | 2353 | return error; |
c1c9c9c4 BP |
2354 | } |
2355 | ||
2356 | static int | |
b5d57fc8 | 2357 | netdev_linux_set_qos(struct netdev *netdev_, |
79f1cbe9 | 2358 | const char *type, const struct smap *details) |
c1c9c9c4 | 2359 | { |
b5d57fc8 | 2360 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2361 | const struct tc_ops *new_ops; |
2362 | int error; | |
2363 | ||
2364 | new_ops = tc_lookup_ovs_name(type); | |
2365 | if (!new_ops || !new_ops->tc_install) { | |
2366 | return EOPNOTSUPP; | |
2367 | } | |
2368 | ||
6cf888b8 BS |
2369 | if (new_ops == &tc_ops_noop) { |
2370 | return new_ops->tc_install(netdev_, details); | |
2371 | } | |
2372 | ||
86383816 | 2373 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2374 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 | 2375 | if (error) { |
86383816 | 2376 | goto exit; |
c1c9c9c4 BP |
2377 | } |
2378 | ||
b5d57fc8 | 2379 | if (new_ops == netdev->tc->ops) { |
86383816 | 2380 | error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; |
c1c9c9c4 BP |
2381 | } else { |
2382 | /* Delete existing qdisc. */ | |
b5d57fc8 | 2383 | error = tc_del_qdisc(netdev_); |
c1c9c9c4 | 2384 | if (error) { |
86383816 | 2385 | goto exit; |
c1c9c9c4 | 2386 | } |
b5d57fc8 | 2387 | ovs_assert(netdev->tc == NULL); |
c1c9c9c4 BP |
2388 | |
2389 | /* Install new qdisc. */ | |
b5d57fc8 BP |
2390 | error = new_ops->tc_install(netdev_, details); |
2391 | ovs_assert((error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 | 2392 | } |
86383816 BP |
2393 | |
2394 | exit: | |
2395 | ovs_mutex_unlock(&netdev->mutex); | |
2396 | return error; | |
c1c9c9c4 BP |
2397 | } |
2398 | ||
2399 | static int | |
b5d57fc8 | 2400 | netdev_linux_get_queue(const struct netdev *netdev_, |
79f1cbe9 | 2401 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 | 2402 | { |
b5d57fc8 | 2403 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2404 | int error; |
2405 | ||
86383816 | 2406 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2407 | error = tc_query_qdisc(netdev_); |
86383816 | 2408 | if (!error) { |
b5d57fc8 | 2409 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
86383816 | 2410 | error = (queue |
b5d57fc8 | 2411 | ? netdev->tc->ops->class_get(netdev_, queue, details) |
93b13be8 | 2412 | : ENOENT); |
c1c9c9c4 | 2413 | } |
86383816 BP |
2414 | ovs_mutex_unlock(&netdev->mutex); |
2415 | ||
2416 | return error; | |
c1c9c9c4 BP |
2417 | } |
2418 | ||
2419 | static int | |
b5d57fc8 | 2420 | netdev_linux_set_queue(struct netdev *netdev_, |
79f1cbe9 | 2421 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 | 2422 | { |
b5d57fc8 | 2423 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2424 | int error; |
2425 | ||
86383816 | 2426 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2427 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2428 | if (!error) { |
2429 | error = (queue_id < netdev->tc->ops->n_queues | |
2430 | && netdev->tc->ops->class_set | |
2431 | ? netdev->tc->ops->class_set(netdev_, queue_id, details) | |
2432 | : EINVAL); | |
c1c9c9c4 | 2433 | } |
86383816 | 2434 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2435 | |
86383816 | 2436 | return error; |
c1c9c9c4 BP |
2437 | } |
2438 | ||
2439 | static int | |
b5d57fc8 | 2440 | netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) |
c1c9c9c4 | 2441 | { |
b5d57fc8 | 2442 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2443 | int error; |
2444 | ||
86383816 | 2445 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2446 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2447 | if (!error) { |
2448 | if (netdev->tc->ops->class_delete) { | |
2449 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2450 | error = (queue | |
2451 | ? netdev->tc->ops->class_delete(netdev_, queue) | |
2452 | : ENOENT); | |
2453 | } else { | |
2454 | error = EINVAL; | |
2455 | } | |
c1c9c9c4 | 2456 | } |
86383816 BP |
2457 | ovs_mutex_unlock(&netdev->mutex); |
2458 | ||
2459 | return error; | |
c1c9c9c4 BP |
2460 | } |
2461 | ||
2462 | static int | |
b5d57fc8 | 2463 | netdev_linux_get_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2464 | unsigned int queue_id, |
2465 | struct netdev_queue_stats *stats) | |
2466 | { | |
b5d57fc8 | 2467 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2468 | int error; |
2469 | ||
86383816 | 2470 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2471 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2472 | if (!error) { |
2473 | if (netdev->tc->ops->class_get_stats) { | |
2474 | const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2475 | if (queue) { | |
2476 | stats->created = queue->created; | |
2477 | error = netdev->tc->ops->class_get_stats(netdev_, queue, | |
2478 | stats); | |
2479 | } else { | |
2480 | error = ENOENT; | |
2481 | } | |
2482 | } else { | |
2483 | error = EOPNOTSUPP; | |
6dc34a0d | 2484 | } |
c1c9c9c4 | 2485 | } |
86383816 BP |
2486 | ovs_mutex_unlock(&netdev->mutex); |
2487 | ||
2488 | return error; | |
c1c9c9c4 BP |
2489 | } |
2490 | ||
d57695d7 JS |
2491 | struct queue_dump_state { |
2492 | struct nl_dump dump; | |
2493 | struct ofpbuf buf; | |
2494 | }; | |
2495 | ||
23a98ffe | 2496 | static bool |
d57695d7 | 2497 | start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) |
c1c9c9c4 BP |
2498 | { |
2499 | struct ofpbuf request; | |
2500 | struct tcmsg *tcmsg; | |
2501 | ||
7874bdff | 2502 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, 0, &request); |
23a98ffe BP |
2503 | if (!tcmsg) { |
2504 | return false; | |
2505 | } | |
3c4de644 | 2506 | tcmsg->tcm_parent = 0; |
d57695d7 | 2507 | nl_dump_start(&state->dump, NETLINK_ROUTE, &request); |
c1c9c9c4 | 2508 | ofpbuf_uninit(&request); |
d57695d7 JS |
2509 | |
2510 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); | |
23a98ffe | 2511 | return true; |
c1c9c9c4 BP |
2512 | } |
2513 | ||
d57695d7 JS |
2514 | static int |
2515 | finish_queue_dump(struct queue_dump_state *state) | |
2516 | { | |
2517 | ofpbuf_uninit(&state->buf); | |
2518 | return nl_dump_done(&state->dump); | |
2519 | } | |
2520 | ||
89454bf4 BP |
2521 | struct netdev_linux_queue_state { |
2522 | unsigned int *queues; | |
2523 | size_t cur_queue; | |
2524 | size_t n_queues; | |
2525 | }; | |
2526 | ||
c1c9c9c4 | 2527 | static int |
89454bf4 | 2528 | netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) |
c1c9c9c4 | 2529 | { |
89454bf4 | 2530 | const struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2531 | int error; |
2532 | ||
86383816 | 2533 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2534 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2535 | if (!error) { |
2536 | if (netdev->tc->ops->class_get) { | |
89454bf4 BP |
2537 | struct netdev_linux_queue_state *state; |
2538 | struct tc_queue *queue; | |
2539 | size_t i; | |
2540 | ||
2541 | *statep = state = xmalloc(sizeof *state); | |
2542 | state->n_queues = hmap_count(&netdev->tc->queues); | |
2543 | state->cur_queue = 0; | |
2544 | state->queues = xmalloc(state->n_queues * sizeof *state->queues); | |
2545 | ||
2546 | i = 0; | |
2547 | HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { | |
2548 | state->queues[i++] = queue->queue_id; | |
86383816 | 2549 | } |
c1c9c9c4 | 2550 | } else { |
86383816 | 2551 | error = EOPNOTSUPP; |
c1c9c9c4 BP |
2552 | } |
2553 | } | |
86383816 | 2554 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2555 | |
86383816 | 2556 | return error; |
c1c9c9c4 BP |
2557 | } |
2558 | ||
89454bf4 BP |
2559 | static int |
2560 | netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, | |
2561 | unsigned int *queue_idp, struct smap *details) | |
2562 | { | |
2563 | const struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
2564 | struct netdev_linux_queue_state *state = state_; | |
2565 | int error = EOF; | |
2566 | ||
2567 | ovs_mutex_lock(&netdev->mutex); | |
2568 | while (state->cur_queue < state->n_queues) { | |
2569 | unsigned int queue_id = state->queues[state->cur_queue++]; | |
2570 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2571 | ||
2572 | if (queue) { | |
2573 | *queue_idp = queue_id; | |
2574 | error = netdev->tc->ops->class_get(netdev_, queue, details); | |
2575 | break; | |
2576 | } | |
2577 | } | |
2578 | ovs_mutex_unlock(&netdev->mutex); | |
2579 | ||
2580 | return error; | |
2581 | } | |
2582 | ||
2583 | static int | |
2584 | netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, | |
2585 | void *state_) | |
2586 | { | |
2587 | struct netdev_linux_queue_state *state = state_; | |
2588 | ||
2589 | free(state->queues); | |
2590 | free(state); | |
2591 | return 0; | |
2592 | } | |
2593 | ||
c1c9c9c4 | 2594 | static int |
b5d57fc8 | 2595 | netdev_linux_dump_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2596 | netdev_dump_queue_stats_cb *cb, void *aux) |
2597 | { | |
b5d57fc8 | 2598 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2599 | int error; |
2600 | ||
86383816 | 2601 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 2602 | error = tc_query_qdisc(netdev_); |
86383816 | 2603 | if (!error) { |
d57695d7 | 2604 | struct queue_dump_state state; |
c1c9c9c4 | 2605 | |
86383816 BP |
2606 | if (!netdev->tc->ops->class_dump_stats) { |
2607 | error = EOPNOTSUPP; | |
d57695d7 | 2608 | } else if (!start_queue_dump(netdev_, &state)) { |
86383816 BP |
2609 | error = ENODEV; |
2610 | } else { | |
2611 | struct ofpbuf msg; | |
2612 | int retval; | |
2613 | ||
d57695d7 | 2614 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
86383816 BP |
2615 | retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, |
2616 | cb, aux); | |
2617 | if (retval) { | |
2618 | error = retval; | |
2619 | } | |
2620 | } | |
2621 | ||
d57695d7 | 2622 | retval = finish_queue_dump(&state); |
86383816 BP |
2623 | if (retval) { |
2624 | error = retval; | |
2625 | } | |
c1c9c9c4 BP |
2626 | } |
2627 | } | |
86383816 | 2628 | ovs_mutex_unlock(&netdev->mutex); |
c1c9c9c4 | 2629 | |
86383816 | 2630 | return error; |
c1c9c9c4 BP |
2631 | } |
2632 | ||
8b61709d | 2633 | static int |
f1acd62b BP |
2634 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
2635 | struct in_addr netmask) | |
8b61709d | 2636 | { |
b5d57fc8 | 2637 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2638 | int error; |
2639 | ||
86383816 | 2640 | ovs_mutex_lock(&netdev->mutex); |
f1acd62b | 2641 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 2642 | if (!error) { |
f1acd62b | 2643 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 2644 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 2645 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
2646 | } |
2647 | } | |
49af9a3d | 2648 | |
86383816 BP |
2649 | ovs_mutex_unlock(&netdev->mutex); |
2650 | ||
8b61709d BP |
2651 | return error; |
2652 | } | |
2653 | ||
7df6932e AW |
2654 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address. |
2655 | * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding | |
2656 | * error. */ | |
8b61709d | 2657 | static int |
a8704b50 PS |
2658 | netdev_linux_get_addr_list(const struct netdev *netdev_, |
2659 | struct in6_addr **addr, struct in6_addr **mask, int *n_cnt) | |
8b61709d | 2660 | { |
b5d57fc8 | 2661 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7df6932e | 2662 | int error; |
86383816 BP |
2663 | |
2664 | ovs_mutex_lock(&netdev->mutex); | |
a8704b50 | 2665 | error = netdev_get_addrs(netdev_get_name(netdev_), addr, mask, n_cnt); |
86383816 BP |
2666 | ovs_mutex_unlock(&netdev->mutex); |
2667 | ||
7df6932e | 2668 | return error; |
8b61709d BP |
2669 | } |
2670 | ||
2671 | static void | |
2672 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
2673 | { | |
2674 | struct sockaddr_in sin; | |
2675 | memset(&sin, 0, sizeof sin); | |
2676 | sin.sin_family = AF_INET; | |
2677 | sin.sin_addr = addr; | |
2678 | sin.sin_port = 0; | |
2679 | ||
2680 | memset(sa, 0, sizeof *sa); | |
2681 | memcpy(sa, &sin, sizeof sin); | |
2682 | } | |
2683 | ||
2684 | static int | |
2685 | do_set_addr(struct netdev *netdev, | |
2686 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
2687 | { | |
2688 | struct ifreq ifr; | |
149f577a | 2689 | |
259e0b1a BP |
2690 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
2691 | return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
2692 | ioctl_name); | |
8b61709d BP |
2693 | } |
2694 | ||
2695 | /* Adds 'router' as a default IP gateway. */ | |
2696 | static int | |
67a4917b | 2697 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
2698 | { |
2699 | struct in_addr any = { INADDR_ANY }; | |
2700 | struct rtentry rt; | |
2701 | int error; | |
2702 | ||
2703 | memset(&rt, 0, sizeof rt); | |
2704 | make_in4_sockaddr(&rt.rt_dst, any); | |
2705 | make_in4_sockaddr(&rt.rt_gateway, router); | |
2706 | make_in4_sockaddr(&rt.rt_genmask, any); | |
2707 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
259e0b1a | 2708 | error = af_inet_ioctl(SIOCADDRT, &rt); |
8b61709d | 2709 | if (error) { |
10a89ef0 | 2710 | VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); |
8b61709d BP |
2711 | } |
2712 | return error; | |
2713 | } | |
2714 | ||
f1acd62b BP |
2715 | static int |
2716 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
2717 | char **netdev_name) | |
2718 | { | |
2719 | static const char fn[] = "/proc/net/route"; | |
2720 | FILE *stream; | |
2721 | char line[256]; | |
2722 | int ln; | |
2723 | ||
2724 | *netdev_name = NULL; | |
2725 | stream = fopen(fn, "r"); | |
2726 | if (stream == NULL) { | |
10a89ef0 | 2727 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
f1acd62b BP |
2728 | return errno; |
2729 | } | |
2730 | ||
2731 | ln = 0; | |
2732 | while (fgets(line, sizeof line, stream)) { | |
2733 | if (++ln >= 2) { | |
2734 | char iface[17]; | |
dbba996b | 2735 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
2736 | int refcnt, metric, mtu; |
2737 | unsigned int flags, use, window, irtt; | |
2738 | ||
c2c28dfd BP |
2739 | if (!ovs_scan(line, |
2740 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
2741 | " %d %u %u\n", | |
2742 | iface, &dest, &gateway, &flags, &refcnt, | |
2743 | &use, &metric, &mask, &mtu, &window, &irtt)) { | |
d295e8e9 | 2744 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
2745 | fn, ln, line); |
2746 | continue; | |
2747 | } | |
2748 | if (!(flags & RTF_UP)) { | |
2749 | /* Skip routes that aren't up. */ | |
2750 | continue; | |
2751 | } | |
2752 | ||
2753 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 2754 | * network byte order, so we don't need need any endian |
f1acd62b BP |
2755 | * conversions here. */ |
2756 | if ((dest & mask) == (host->s_addr & mask)) { | |
2757 | if (!gateway) { | |
2758 | /* The host is directly reachable. */ | |
2759 | next_hop->s_addr = 0; | |
2760 | } else { | |
2761 | /* To reach the host, we must go through a gateway. */ | |
2762 | next_hop->s_addr = gateway; | |
2763 | } | |
2764 | *netdev_name = xstrdup(iface); | |
2765 | fclose(stream); | |
2766 | return 0; | |
2767 | } | |
2768 | } | |
2769 | } | |
2770 | ||
2771 | fclose(stream); | |
2772 | return ENXIO; | |
2773 | } | |
2774 | ||
e210037e | 2775 | static int |
b5d57fc8 | 2776 | netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) |
e210037e | 2777 | { |
b5d57fc8 | 2778 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
275707c3 EJ |
2779 | int error = 0; |
2780 | ||
86383816 | 2781 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
2782 | if (!(netdev->cache_valid & VALID_DRVINFO)) { |
2783 | struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; | |
275707c3 EJ |
2784 | |
2785 | COVERAGE_INC(netdev_get_ethtool); | |
b5d57fc8 BP |
2786 | memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); |
2787 | error = netdev_linux_do_ethtool(netdev->up.name, | |
275707c3 EJ |
2788 | cmd, |
2789 | ETHTOOL_GDRVINFO, | |
2790 | "ETHTOOL_GDRVINFO"); | |
2791 | if (!error) { | |
b5d57fc8 | 2792 | netdev->cache_valid |= VALID_DRVINFO; |
275707c3 EJ |
2793 | } |
2794 | } | |
e210037e | 2795 | |
e210037e | 2796 | if (!error) { |
b5d57fc8 BP |
2797 | smap_add(smap, "driver_name", netdev->drvinfo.driver); |
2798 | smap_add(smap, "driver_version", netdev->drvinfo.version); | |
2799 | smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); | |
e210037e | 2800 | } |
86383816 BP |
2801 | ovs_mutex_unlock(&netdev->mutex); |
2802 | ||
e210037e AE |
2803 | return error; |
2804 | } | |
2805 | ||
4f925bd3 | 2806 | static int |
275707c3 EJ |
2807 | netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, |
2808 | struct smap *smap) | |
4f925bd3 | 2809 | { |
79f1cbe9 | 2810 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
2811 | return 0; |
2812 | } | |
2813 | ||
8b61709d BP |
2814 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2815 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2816 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2817 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2818 | static int | |
2819 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
74ff3298 | 2820 | ovs_be32 ip, struct eth_addr *mac) |
8b61709d BP |
2821 | { |
2822 | struct arpreq r; | |
c100e025 | 2823 | struct sockaddr_in sin; |
8b61709d BP |
2824 | int retval; |
2825 | ||
2826 | memset(&r, 0, sizeof r); | |
f2cc621b | 2827 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2828 | sin.sin_family = AF_INET; |
2829 | sin.sin_addr.s_addr = ip; | |
2830 | sin.sin_port = 0; | |
2831 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2832 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2833 | r.arp_flags = 0; | |
71d7c22f | 2834 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d | 2835 | COVERAGE_INC(netdev_arp_lookup); |
259e0b1a | 2836 | retval = af_inet_ioctl(SIOCGARP, &r); |
8b61709d BP |
2837 | if (!retval) { |
2838 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2839 | } else if (retval != ENXIO) { | |
2840 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
10a89ef0 BP |
2841 | netdev_get_name(netdev), IP_ARGS(ip), |
2842 | ovs_strerror(retval)); | |
8b61709d BP |
2843 | } |
2844 | return retval; | |
2845 | } | |
2846 | ||
2847 | static int | |
2848 | nd_to_iff_flags(enum netdev_flags nd) | |
2849 | { | |
2850 | int iff = 0; | |
2851 | if (nd & NETDEV_UP) { | |
2852 | iff |= IFF_UP; | |
2853 | } | |
2854 | if (nd & NETDEV_PROMISC) { | |
2855 | iff |= IFF_PROMISC; | |
2856 | } | |
7ba19d41 AC |
2857 | if (nd & NETDEV_LOOPBACK) { |
2858 | iff |= IFF_LOOPBACK; | |
2859 | } | |
8b61709d BP |
2860 | return iff; |
2861 | } | |
2862 | ||
2863 | static int | |
2864 | iff_to_nd_flags(int iff) | |
2865 | { | |
2866 | enum netdev_flags nd = 0; | |
2867 | if (iff & IFF_UP) { | |
2868 | nd |= NETDEV_UP; | |
2869 | } | |
2870 | if (iff & IFF_PROMISC) { | |
2871 | nd |= NETDEV_PROMISC; | |
2872 | } | |
7ba19d41 AC |
2873 | if (iff & IFF_LOOPBACK) { |
2874 | nd |= NETDEV_LOOPBACK; | |
2875 | } | |
8b61709d BP |
2876 | return nd; |
2877 | } | |
2878 | ||
2879 | static int | |
4f9f3f21 BP |
2880 | update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
2881 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2882 | OVS_REQUIRES(netdev->mutex) | |
8b61709d BP |
2883 | { |
2884 | int old_flags, new_flags; | |
c37d4da4 EJ |
2885 | int error = 0; |
2886 | ||
b5d57fc8 | 2887 | old_flags = netdev->ifi_flags; |
c37d4da4 EJ |
2888 | *old_flagsp = iff_to_nd_flags(old_flags); |
2889 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
2890 | if (new_flags != old_flags) { | |
4f9f3f21 BP |
2891 | error = set_flags(netdev_get_name(&netdev->up), new_flags); |
2892 | get_flags(&netdev->up, &netdev->ifi_flags); | |
8b61709d | 2893 | } |
4f9f3f21 BP |
2894 | |
2895 | return error; | |
2896 | } | |
2897 | ||
2898 | static int | |
2899 | netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, | |
2900 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
2901 | { | |
2902 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
756819dd | 2903 | int error = 0; |
4f9f3f21 BP |
2904 | |
2905 | ovs_mutex_lock(&netdev->mutex); | |
756819dd FL |
2906 | if (on || off) { |
2907 | /* Changing flags over netlink isn't support yet. */ | |
2908 | error = update_flags(netdev, off, on, old_flagsp); | |
2909 | } else { | |
2910 | /* Try reading flags over netlink, or fall back to ioctl. */ | |
2911 | if (!netdev_linux_update_via_netlink(netdev)) { | |
2912 | *old_flagsp = iff_to_nd_flags(netdev->ifi_flags); | |
2913 | } else { | |
2914 | error = update_flags(netdev, off, on, old_flagsp); | |
2915 | } | |
2916 | } | |
86383816 | 2917 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d BP |
2918 | return error; |
2919 | } | |
2920 | ||
2f9dd77f | 2921 | #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, \ |
18ebd48c PB |
2922 | GET_FEATURES, GET_STATUS, \ |
2923 | FLOW_OFFLOAD_API) \ | |
c3827f61 BP |
2924 | { \ |
2925 | NAME, \ | |
118c77b1 | 2926 | false, /* is_pmd */ \ |
c3827f61 | 2927 | \ |
259e0b1a | 2928 | NULL, \ |
c3827f61 BP |
2929 | netdev_linux_run, \ |
2930 | netdev_linux_wait, \ | |
2931 | \ | |
9dc63482 BP |
2932 | netdev_linux_alloc, \ |
2933 | CONSTRUCT, \ | |
2934 | netdev_linux_destruct, \ | |
2935 | netdev_linux_dealloc, \ | |
de5cdb90 | 2936 | NULL, /* get_config */ \ |
6d9e6eb4 | 2937 | NULL, /* set_config */ \ |
f431bf7d | 2938 | NULL, /* get_tunnel_config */ \ |
a36de779 PS |
2939 | NULL, /* build header */ \ |
2940 | NULL, /* push header */ \ | |
2941 | NULL, /* pop header */ \ | |
7dec44fe | 2942 | NULL, /* get_numa_id */ \ |
050c60bf | 2943 | NULL, /* set_tx_multiq */ \ |
c3827f61 | 2944 | \ |
c3827f61 BP |
2945 | netdev_linux_send, \ |
2946 | netdev_linux_send_wait, \ | |
2947 | \ | |
2948 | netdev_linux_set_etheraddr, \ | |
2949 | netdev_linux_get_etheraddr, \ | |
2950 | netdev_linux_get_mtu, \ | |
9b020780 | 2951 | netdev_linux_set_mtu, \ |
c3827f61 BP |
2952 | netdev_linux_get_ifindex, \ |
2953 | netdev_linux_get_carrier, \ | |
65c3058c | 2954 | netdev_linux_get_carrier_resets, \ |
1670c579 | 2955 | netdev_linux_set_miimon_interval, \ |
f613a0d7 | 2956 | GET_STATS, \ |
971f4b39 | 2957 | NULL, \ |
c3827f61 | 2958 | \ |
51f87458 | 2959 | GET_FEATURES, \ |
c3827f61 | 2960 | netdev_linux_set_advertisements, \ |
875ab130 | 2961 | NULL, /* get_pt_mode */ \ |
c3827f61 BP |
2962 | \ |
2963 | netdev_linux_set_policing, \ | |
2964 | netdev_linux_get_qos_types, \ | |
2965 | netdev_linux_get_qos_capabilities, \ | |
2966 | netdev_linux_get_qos, \ | |
2967 | netdev_linux_set_qos, \ | |
2968 | netdev_linux_get_queue, \ | |
2969 | netdev_linux_set_queue, \ | |
2970 | netdev_linux_delete_queue, \ | |
2971 | netdev_linux_get_queue_stats, \ | |
89454bf4 BP |
2972 | netdev_linux_queue_dump_start, \ |
2973 | netdev_linux_queue_dump_next, \ | |
2974 | netdev_linux_queue_dump_done, \ | |
c3827f61 BP |
2975 | netdev_linux_dump_queue_stats, \ |
2976 | \ | |
c3827f61 | 2977 | netdev_linux_set_in4, \ |
a8704b50 | 2978 | netdev_linux_get_addr_list, \ |
c3827f61 BP |
2979 | netdev_linux_add_router, \ |
2980 | netdev_linux_get_next_hop, \ | |
4f925bd3 | 2981 | GET_STATUS, \ |
c3827f61 BP |
2982 | netdev_linux_arp_lookup, \ |
2983 | \ | |
2984 | netdev_linux_update_flags, \ | |
790fb3b7 | 2985 | NULL, /* reconfigure */ \ |
c3827f61 | 2986 | \ |
f7791740 PS |
2987 | netdev_linux_rxq_alloc, \ |
2988 | netdev_linux_rxq_construct, \ | |
2989 | netdev_linux_rxq_destruct, \ | |
2990 | netdev_linux_rxq_dealloc, \ | |
2991 | netdev_linux_rxq_recv, \ | |
2992 | netdev_linux_rxq_wait, \ | |
2993 | netdev_linux_rxq_drain, \ | |
18ebd48c PB |
2994 | \ |
2995 | FLOW_OFFLOAD_API \ | |
c3827f61 BP |
2996 | } |
2997 | ||
2998 | const struct netdev_class netdev_linux_class = | |
2999 | NETDEV_LINUX_CLASS( | |
3000 | "system", | |
9dc63482 | 3001 | netdev_linux_construct, |
f613a0d7 | 3002 | netdev_linux_get_stats, |
51f87458 | 3003 | netdev_linux_get_features, |
18ebd48c PB |
3004 | netdev_linux_get_status, |
3005 | LINUX_FLOW_OFFLOAD_API); | |
c3827f61 BP |
3006 | |
3007 | const struct netdev_class netdev_tap_class = | |
3008 | NETDEV_LINUX_CLASS( | |
3009 | "tap", | |
9dc63482 | 3010 | netdev_linux_construct_tap, |
bba1e6f3 | 3011 | netdev_tap_get_stats, |
51f87458 | 3012 | netdev_linux_get_features, |
18ebd48c PB |
3013 | netdev_linux_get_status, |
3014 | NO_OFFLOAD_API); | |
c3827f61 BP |
3015 | |
3016 | const struct netdev_class netdev_internal_class = | |
3017 | NETDEV_LINUX_CLASS( | |
3018 | "internal", | |
9dc63482 | 3019 | netdev_linux_construct, |
bba1e6f3 | 3020 | netdev_internal_get_stats, |
51f87458 | 3021 | NULL, /* get_features */ |
18ebd48c PB |
3022 | netdev_internal_get_status, |
3023 | NO_OFFLOAD_API); | |
8b61709d | 3024 | \f |
677d9158 JV |
3025 | |
3026 | #define CODEL_N_QUEUES 0x0000 | |
3027 | ||
2f4298ce BP |
3028 | /* In sufficiently new kernel headers these are defined as enums in |
3029 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3030 | * kernels. (This overrides any enum definition in the header file but that's | |
3031 | * harmless.) */ | |
3032 | #define TCA_CODEL_TARGET 1 | |
3033 | #define TCA_CODEL_LIMIT 2 | |
3034 | #define TCA_CODEL_INTERVAL 3 | |
3035 | ||
677d9158 JV |
3036 | struct codel { |
3037 | struct tc tc; | |
3038 | uint32_t target; | |
3039 | uint32_t limit; | |
3040 | uint32_t interval; | |
3041 | }; | |
3042 | ||
3043 | static struct codel * | |
3044 | codel_get__(const struct netdev *netdev_) | |
3045 | { | |
3046 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3047 | return CONTAINER_OF(netdev->tc, struct codel, tc); | |
3048 | } | |
3049 | ||
3050 | static void | |
3051 | codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3052 | uint32_t interval) | |
3053 | { | |
3054 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3055 | struct codel *codel; | |
3056 | ||
3057 | codel = xmalloc(sizeof *codel); | |
3058 | tc_init(&codel->tc, &tc_ops_codel); | |
3059 | codel->target = target; | |
3060 | codel->limit = limit; | |
3061 | codel->interval = interval; | |
3062 | ||
3063 | netdev->tc = &codel->tc; | |
3064 | } | |
3065 | ||
3066 | static int | |
3067 | codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3068 | uint32_t interval) | |
3069 | { | |
3070 | size_t opt_offset; | |
3071 | struct ofpbuf request; | |
3072 | struct tcmsg *tcmsg; | |
3073 | uint32_t otarget, olimit, ointerval; | |
3074 | int error; | |
3075 | ||
3076 | tc_del_qdisc(netdev); | |
3077 | ||
7874bdff RD |
3078 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3079 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3080 | if (!tcmsg) { |
3081 | return ENODEV; | |
3082 | } | |
3083 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3084 | tcmsg->tcm_parent = TC_H_ROOT; | |
3085 | ||
3086 | otarget = target ? target : 5000; | |
3087 | olimit = limit ? limit : 10240; | |
3088 | ointerval = interval ? interval : 100000; | |
3089 | ||
3090 | nl_msg_put_string(&request, TCA_KIND, "codel"); | |
3091 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3092 | nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget); | |
3093 | nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit); | |
3094 | nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval); | |
3095 | nl_msg_end_nested(&request, opt_offset); | |
3096 | ||
3097 | error = tc_transact(&request, NULL); | |
3098 | if (error) { | |
3099 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3100 | "target %u, limit %u, interval %u error %d(%s)", | |
3101 | netdev_get_name(netdev), | |
3102 | otarget, olimit, ointerval, | |
3103 | error, ovs_strerror(error)); | |
3104 | } | |
3105 | return error; | |
3106 | } | |
3107 | ||
3108 | static void | |
3109 | codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3110 | const struct smap *details, struct codel *codel) | |
3111 | { | |
13c1637f BP |
3112 | codel->target = smap_get_ullong(details, "target", 0); |
3113 | codel->limit = smap_get_ullong(details, "limit", 0); | |
3114 | codel->interval = smap_get_ullong(details, "interval", 0); | |
677d9158 JV |
3115 | |
3116 | if (!codel->target) { | |
3117 | codel->target = 5000; | |
3118 | } | |
3119 | if (!codel->limit) { | |
3120 | codel->limit = 10240; | |
3121 | } | |
3122 | if (!codel->interval) { | |
3123 | codel->interval = 100000; | |
3124 | } | |
3125 | } | |
3126 | ||
3127 | static int | |
3128 | codel_tc_install(struct netdev *netdev, const struct smap *details) | |
3129 | { | |
3130 | int error; | |
3131 | struct codel codel; | |
3132 | ||
3133 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3134 | error = codel_setup_qdisc__(netdev, codel.target, codel.limit, | |
3135 | codel.interval); | |
3136 | if (!error) { | |
3137 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3138 | } | |
3139 | return error; | |
3140 | } | |
3141 | ||
3142 | static int | |
3143 | codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel) | |
3144 | { | |
3145 | static const struct nl_policy tca_codel_policy[] = { | |
3146 | [TCA_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3147 | [TCA_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3148 | [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 } | |
3149 | }; | |
3150 | ||
3151 | struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)]; | |
3152 | ||
3153 | if (!nl_parse_nested(nl_options, tca_codel_policy, | |
3154 | attrs, ARRAY_SIZE(tca_codel_policy))) { | |
3155 | VLOG_WARN_RL(&rl, "failed to parse CoDel class options"); | |
3156 | return EPROTO; | |
3157 | } | |
3158 | ||
3159 | codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]); | |
3160 | codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]); | |
3161 | codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]); | |
3162 | return 0; | |
3163 | } | |
3164 | ||
3165 | static int | |
3166 | codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3167 | { | |
3168 | struct nlattr *nlattr; | |
3169 | const char * kind; | |
3170 | int error; | |
3171 | struct codel codel; | |
3172 | ||
3173 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3174 | if (error != 0) { | |
3175 | return error; | |
3176 | } | |
3177 | ||
3178 | error = codel_parse_tca_options__(nlattr, &codel); | |
3179 | if (error != 0) { | |
3180 | return error; | |
3181 | } | |
3182 | ||
3183 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3184 | return 0; | |
3185 | } | |
3186 | ||
3187 | ||
3188 | static void | |
3189 | codel_tc_destroy(struct tc *tc) | |
3190 | { | |
3191 | struct codel *codel = CONTAINER_OF(tc, struct codel, tc); | |
3192 | tc_destroy(tc); | |
3193 | free(codel); | |
3194 | } | |
3195 | ||
3196 | static int | |
3197 | codel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3198 | { | |
3199 | const struct codel *codel = codel_get__(netdev); | |
3200 | smap_add_format(details, "target", "%u", codel->target); | |
3201 | smap_add_format(details, "limit", "%u", codel->limit); | |
3202 | smap_add_format(details, "interval", "%u", codel->interval); | |
3203 | return 0; | |
3204 | } | |
3205 | ||
3206 | static int | |
3207 | codel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3208 | { | |
3209 | struct codel codel; | |
3210 | ||
3211 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3212 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3213 | codel_get__(netdev)->target = codel.target; | |
3214 | codel_get__(netdev)->limit = codel.limit; | |
3215 | codel_get__(netdev)->interval = codel.interval; | |
3216 | return 0; | |
3217 | } | |
3218 | ||
3219 | static const struct tc_ops tc_ops_codel = { | |
3220 | "codel", /* linux_name */ | |
3221 | "linux-codel", /* ovs_name */ | |
3222 | CODEL_N_QUEUES, /* n_queues */ | |
3223 | codel_tc_install, | |
3224 | codel_tc_load, | |
3225 | codel_tc_destroy, | |
3226 | codel_qdisc_get, | |
3227 | codel_qdisc_set, | |
3228 | NULL, | |
3229 | NULL, | |
3230 | NULL, | |
3231 | NULL, | |
3232 | NULL | |
3233 | }; | |
3234 | \f | |
3235 | /* FQ-CoDel traffic control class. */ | |
3236 | ||
3237 | #define FQCODEL_N_QUEUES 0x0000 | |
3238 | ||
2f4298ce BP |
3239 | /* In sufficiently new kernel headers these are defined as enums in |
3240 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3241 | * kernels. (This overrides any enum definition in the header file but that's | |
3242 | * harmless.) */ | |
3243 | #define TCA_FQ_CODEL_TARGET 1 | |
3244 | #define TCA_FQ_CODEL_LIMIT 2 | |
3245 | #define TCA_FQ_CODEL_INTERVAL 3 | |
3246 | #define TCA_FQ_CODEL_ECN 4 | |
3247 | #define TCA_FQ_CODEL_FLOWS 5 | |
3248 | #define TCA_FQ_CODEL_QUANTUM 6 | |
3249 | ||
677d9158 JV |
3250 | struct fqcodel { |
3251 | struct tc tc; | |
3252 | uint32_t target; | |
3253 | uint32_t limit; | |
3254 | uint32_t interval; | |
3255 | uint32_t flows; | |
3256 | uint32_t quantum; | |
3257 | }; | |
3258 | ||
3259 | static struct fqcodel * | |
3260 | fqcodel_get__(const struct netdev *netdev_) | |
3261 | { | |
3262 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3263 | return CONTAINER_OF(netdev->tc, struct fqcodel, tc); | |
3264 | } | |
3265 | ||
3266 | static void | |
3267 | fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3268 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3269 | { | |
3270 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3271 | struct fqcodel *fqcodel; | |
3272 | ||
3273 | fqcodel = xmalloc(sizeof *fqcodel); | |
3274 | tc_init(&fqcodel->tc, &tc_ops_fqcodel); | |
3275 | fqcodel->target = target; | |
3276 | fqcodel->limit = limit; | |
3277 | fqcodel->interval = interval; | |
3278 | fqcodel->flows = flows; | |
3279 | fqcodel->quantum = quantum; | |
3280 | ||
3281 | netdev->tc = &fqcodel->tc; | |
3282 | } | |
3283 | ||
3284 | static int | |
3285 | fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3286 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3287 | { | |
3288 | size_t opt_offset; | |
3289 | struct ofpbuf request; | |
3290 | struct tcmsg *tcmsg; | |
3291 | uint32_t otarget, olimit, ointerval, oflows, oquantum; | |
3292 | int error; | |
3293 | ||
3294 | tc_del_qdisc(netdev); | |
3295 | ||
7874bdff RD |
3296 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3297 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3298 | if (!tcmsg) { |
3299 | return ENODEV; | |
3300 | } | |
3301 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3302 | tcmsg->tcm_parent = TC_H_ROOT; | |
3303 | ||
3304 | otarget = target ? target : 5000; | |
3305 | olimit = limit ? limit : 10240; | |
3306 | ointerval = interval ? interval : 100000; | |
3307 | oflows = flows ? flows : 1024; | |
3308 | oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514 | |
3309 | not mtu */ | |
3310 | ||
3311 | nl_msg_put_string(&request, TCA_KIND, "fq_codel"); | |
3312 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3313 | nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget); | |
3314 | nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit); | |
3315 | nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval); | |
3316 | nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows); | |
3317 | nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum); | |
3318 | nl_msg_end_nested(&request, opt_offset); | |
3319 | ||
3320 | error = tc_transact(&request, NULL); | |
3321 | if (error) { | |
3322 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3323 | "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)", | |
3324 | netdev_get_name(netdev), | |
3325 | otarget, olimit, ointerval, oflows, oquantum, | |
3326 | error, ovs_strerror(error)); | |
3327 | } | |
3328 | return error; | |
3329 | } | |
3330 | ||
3331 | static void | |
3332 | fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3333 | const struct smap *details, struct fqcodel *fqcodel) | |
3334 | { | |
13c1637f BP |
3335 | fqcodel->target = smap_get_ullong(details, "target", 0); |
3336 | fqcodel->limit = smap_get_ullong(details, "limit", 0); | |
3337 | fqcodel->interval = smap_get_ullong(details, "interval", 0); | |
3338 | fqcodel->flows = smap_get_ullong(details, "flows", 0); | |
3339 | fqcodel->quantum = smap_get_ullong(details, "quantum", 0); | |
3340 | ||
677d9158 JV |
3341 | if (!fqcodel->target) { |
3342 | fqcodel->target = 5000; | |
3343 | } | |
3344 | if (!fqcodel->limit) { | |
3345 | fqcodel->limit = 10240; | |
3346 | } | |
3347 | if (!fqcodel->interval) { | |
3348 | fqcodel->interval = 1000000; | |
3349 | } | |
3350 | if (!fqcodel->flows) { | |
3351 | fqcodel->flows = 1024; | |
3352 | } | |
3353 | if (!fqcodel->quantum) { | |
3354 | fqcodel->quantum = 1514; | |
3355 | } | |
3356 | } | |
3357 | ||
3358 | static int | |
3359 | fqcodel_tc_install(struct netdev *netdev, const struct smap *details) | |
3360 | { | |
3361 | int error; | |
3362 | struct fqcodel fqcodel; | |
3363 | ||
3364 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3365 | error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit, | |
3366 | fqcodel.interval, fqcodel.flows, | |
3367 | fqcodel.quantum); | |
3368 | if (!error) { | |
3369 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, | |
3370 | fqcodel.interval, fqcodel.flows, fqcodel.quantum); | |
3371 | } | |
3372 | return error; | |
3373 | } | |
3374 | ||
3375 | static int | |
3376 | fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel) | |
3377 | { | |
3378 | static const struct nl_policy tca_fqcodel_policy[] = { | |
3379 | [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3380 | [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3381 | [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 }, | |
3382 | [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 }, | |
3383 | [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 } | |
3384 | }; | |
3385 | ||
3386 | struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)]; | |
3387 | ||
3388 | if (!nl_parse_nested(nl_options, tca_fqcodel_policy, | |
3389 | attrs, ARRAY_SIZE(tca_fqcodel_policy))) { | |
3390 | VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options"); | |
3391 | return EPROTO; | |
3392 | } | |
3393 | ||
3394 | fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]); | |
3395 | fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]); | |
3396 | fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]); | |
3397 | fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]); | |
3398 | fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]); | |
3399 | return 0; | |
3400 | } | |
3401 | ||
3402 | static int | |
3403 | fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3404 | { | |
3405 | struct nlattr *nlattr; | |
3406 | const char * kind; | |
3407 | int error; | |
3408 | struct fqcodel fqcodel; | |
3409 | ||
3410 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3411 | if (error != 0) { | |
3412 | return error; | |
3413 | } | |
3414 | ||
3415 | error = fqcodel_parse_tca_options__(nlattr, &fqcodel); | |
3416 | if (error != 0) { | |
3417 | return error; | |
3418 | } | |
3419 | ||
3420 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3421 | fqcodel.flows, fqcodel.quantum); | |
3422 | return 0; | |
3423 | } | |
3424 | ||
3425 | static void | |
3426 | fqcodel_tc_destroy(struct tc *tc) | |
3427 | { | |
3428 | struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc); | |
3429 | tc_destroy(tc); | |
3430 | free(fqcodel); | |
3431 | } | |
3432 | ||
3433 | static int | |
3434 | fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3435 | { | |
3436 | const struct fqcodel *fqcodel = fqcodel_get__(netdev); | |
3437 | smap_add_format(details, "target", "%u", fqcodel->target); | |
3438 | smap_add_format(details, "limit", "%u", fqcodel->limit); | |
3439 | smap_add_format(details, "interval", "%u", fqcodel->interval); | |
3440 | smap_add_format(details, "flows", "%u", fqcodel->flows); | |
3441 | smap_add_format(details, "quantum", "%u", fqcodel->quantum); | |
3442 | return 0; | |
3443 | } | |
3444 | ||
3445 | static int | |
3446 | fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3447 | { | |
3448 | struct fqcodel fqcodel; | |
3449 | ||
3450 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3451 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3452 | fqcodel.flows, fqcodel.quantum); | |
3453 | fqcodel_get__(netdev)->target = fqcodel.target; | |
3454 | fqcodel_get__(netdev)->limit = fqcodel.limit; | |
3455 | fqcodel_get__(netdev)->interval = fqcodel.interval; | |
3456 | fqcodel_get__(netdev)->flows = fqcodel.flows; | |
3457 | fqcodel_get__(netdev)->quantum = fqcodel.quantum; | |
3458 | return 0; | |
3459 | } | |
3460 | ||
3461 | static const struct tc_ops tc_ops_fqcodel = { | |
3462 | "fq_codel", /* linux_name */ | |
3463 | "linux-fq_codel", /* ovs_name */ | |
3464 | FQCODEL_N_QUEUES, /* n_queues */ | |
3465 | fqcodel_tc_install, | |
3466 | fqcodel_tc_load, | |
3467 | fqcodel_tc_destroy, | |
3468 | fqcodel_qdisc_get, | |
3469 | fqcodel_qdisc_set, | |
3470 | NULL, | |
3471 | NULL, | |
3472 | NULL, | |
3473 | NULL, | |
3474 | NULL | |
3475 | }; | |
3476 | \f | |
3477 | /* SFQ traffic control class. */ | |
3478 | ||
3479 | #define SFQ_N_QUEUES 0x0000 | |
3480 | ||
3481 | struct sfq { | |
3482 | struct tc tc; | |
3483 | uint32_t quantum; | |
3484 | uint32_t perturb; | |
3485 | }; | |
3486 | ||
3487 | static struct sfq * | |
3488 | sfq_get__(const struct netdev *netdev_) | |
3489 | { | |
3490 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3491 | return CONTAINER_OF(netdev->tc, struct sfq, tc); | |
3492 | } | |
3493 | ||
3494 | static void | |
3495 | sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb) | |
3496 | { | |
3497 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3498 | struct sfq *sfq; | |
3499 | ||
3500 | sfq = xmalloc(sizeof *sfq); | |
3501 | tc_init(&sfq->tc, &tc_ops_sfq); | |
3502 | sfq->perturb = perturb; | |
3503 | sfq->quantum = quantum; | |
3504 | ||
3505 | netdev->tc = &sfq->tc; | |
3506 | } | |
3507 | ||
3508 | static int | |
3509 | sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb) | |
3510 | { | |
3511 | struct tc_sfq_qopt opt; | |
3512 | struct ofpbuf request; | |
3513 | struct tcmsg *tcmsg; | |
3514 | int mtu; | |
3515 | int mtu_error, error; | |
3516 | mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); | |
3517 | ||
3518 | tc_del_qdisc(netdev); | |
3519 | ||
7874bdff RD |
3520 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3521 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3522 | if (!tcmsg) { |
3523 | return ENODEV; | |
3524 | } | |
3525 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3526 | tcmsg->tcm_parent = TC_H_ROOT; | |
3527 | ||
3528 | memset(&opt, 0, sizeof opt); | |
3529 | if (!quantum) { | |
3530 | if (!mtu_error) { | |
3531 | opt.quantum = mtu; /* if we cannot find mtu, use default */ | |
3532 | } | |
3533 | } else { | |
3534 | opt.quantum = quantum; | |
3535 | } | |
3536 | ||
3537 | if (!perturb) { | |
3538 | opt.perturb_period = 10; | |
3539 | } else { | |
3540 | opt.perturb_period = perturb; | |
3541 | } | |
3542 | ||
3543 | nl_msg_put_string(&request, TCA_KIND, "sfq"); | |
3544 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3545 | ||
3546 | error = tc_transact(&request, NULL); | |
3547 | if (error) { | |
3548 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3549 | "quantum %u, perturb %u error %d(%s)", | |
3550 | netdev_get_name(netdev), | |
3551 | opt.quantum, opt.perturb_period, | |
3552 | error, ovs_strerror(error)); | |
3553 | } | |
3554 | return error; | |
3555 | } | |
3556 | ||
3557 | static void | |
3558 | sfq_parse_qdisc_details__(struct netdev *netdev, | |
3559 | const struct smap *details, struct sfq *sfq) | |
3560 | { | |
13c1637f BP |
3561 | sfq->perturb = smap_get_ullong(details, "perturb", 0); |
3562 | sfq->quantum = smap_get_ullong(details, "quantum", 0); | |
677d9158 | 3563 | |
677d9158 JV |
3564 | if (!sfq->perturb) { |
3565 | sfq->perturb = 10; | |
3566 | } | |
3567 | ||
3568 | if (!sfq->quantum) { | |
13c1637f BP |
3569 | int mtu; |
3570 | if (!netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
677d9158 JV |
3571 | sfq->quantum = mtu; |
3572 | } else { | |
3573 | VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a " | |
3574 | "device without mtu"); | |
677d9158 JV |
3575 | } |
3576 | } | |
3577 | } | |
3578 | ||
3579 | static int | |
3580 | sfq_tc_install(struct netdev *netdev, const struct smap *details) | |
3581 | { | |
3582 | int error; | |
3583 | struct sfq sfq; | |
3584 | ||
3585 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3586 | error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb); | |
3587 | if (!error) { | |
3588 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3589 | } | |
3590 | return error; | |
3591 | } | |
3592 | ||
3593 | static int | |
3594 | sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3595 | { | |
3596 | const struct tc_sfq_qopt *sfq; | |
3597 | struct nlattr *nlattr; | |
3598 | const char * kind; | |
3599 | int error; | |
3600 | ||
3601 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3602 | if (error == 0) { | |
3603 | sfq = nl_attr_get(nlattr); | |
3604 | sfq_install__(netdev, sfq->perturb_period, sfq->quantum); | |
3605 | return 0; | |
3606 | } | |
3607 | ||
3608 | return error; | |
3609 | } | |
3610 | ||
3611 | static void | |
3612 | sfq_tc_destroy(struct tc *tc) | |
3613 | { | |
3614 | struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc); | |
3615 | tc_destroy(tc); | |
3616 | free(sfq); | |
3617 | } | |
3618 | ||
3619 | static int | |
3620 | sfq_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3621 | { | |
3622 | const struct sfq *sfq = sfq_get__(netdev); | |
3623 | smap_add_format(details, "quantum", "%u", sfq->quantum); | |
3624 | smap_add_format(details, "perturb", "%u", sfq->perturb); | |
3625 | return 0; | |
3626 | } | |
3627 | ||
3628 | static int | |
3629 | sfq_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3630 | { | |
3631 | struct sfq sfq; | |
3632 | ||
3633 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3634 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3635 | sfq_get__(netdev)->quantum = sfq.quantum; | |
3636 | sfq_get__(netdev)->perturb = sfq.perturb; | |
3637 | return 0; | |
3638 | } | |
3639 | ||
3640 | static const struct tc_ops tc_ops_sfq = { | |
3641 | "sfq", /* linux_name */ | |
3642 | "linux-sfq", /* ovs_name */ | |
3643 | SFQ_N_QUEUES, /* n_queues */ | |
3644 | sfq_tc_install, | |
3645 | sfq_tc_load, | |
3646 | sfq_tc_destroy, | |
3647 | sfq_qdisc_get, | |
3648 | sfq_qdisc_set, | |
3649 | NULL, | |
3650 | NULL, | |
3651 | NULL, | |
3652 | NULL, | |
3653 | NULL | |
3654 | }; | |
3655 | \f | |
c1c9c9c4 | 3656 | /* HTB traffic control class. */ |
559843ed | 3657 | |
c1c9c9c4 | 3658 | #define HTB_N_QUEUES 0xf000 |
4f631ccd | 3659 | #define HTB_RATE2QUANTUM 10 |
8b61709d | 3660 | |
c1c9c9c4 BP |
3661 | struct htb { |
3662 | struct tc tc; | |
3663 | unsigned int max_rate; /* In bytes/s. */ | |
3664 | }; | |
8b61709d | 3665 | |
c1c9c9c4 | 3666 | struct htb_class { |
93b13be8 | 3667 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
3668 | unsigned int min_rate; /* In bytes/s. */ |
3669 | unsigned int max_rate; /* In bytes/s. */ | |
3670 | unsigned int burst; /* In bytes. */ | |
3671 | unsigned int priority; /* Lower values are higher priorities. */ | |
3672 | }; | |
8b61709d | 3673 | |
c1c9c9c4 | 3674 | static struct htb * |
b5d57fc8 | 3675 | htb_get__(const struct netdev *netdev_) |
c1c9c9c4 | 3676 | { |
b5d57fc8 BP |
3677 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
3678 | return CONTAINER_OF(netdev->tc, struct htb, tc); | |
c1c9c9c4 BP |
3679 | } |
3680 | ||
24045e35 | 3681 | static void |
b5d57fc8 | 3682 | htb_install__(struct netdev *netdev_, uint64_t max_rate) |
c1c9c9c4 | 3683 | { |
b5d57fc8 | 3684 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3685 | struct htb *htb; |
3686 | ||
3687 | htb = xmalloc(sizeof *htb); | |
3688 | tc_init(&htb->tc, &tc_ops_htb); | |
3689 | htb->max_rate = max_rate; | |
3690 | ||
b5d57fc8 | 3691 | netdev->tc = &htb->tc; |
c1c9c9c4 BP |
3692 | } |
3693 | ||
3694 | /* Create an HTB qdisc. | |
3695 | * | |
a339aa81 | 3696 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
3697 | static int |
3698 | htb_setup_qdisc__(struct netdev *netdev) | |
3699 | { | |
3700 | size_t opt_offset; | |
3701 | struct tc_htb_glob opt; | |
3702 | struct ofpbuf request; | |
3703 | struct tcmsg *tcmsg; | |
3704 | ||
3705 | tc_del_qdisc(netdev); | |
3706 | ||
7874bdff RD |
3707 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3708 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
3709 | if (!tcmsg) { |
3710 | return ENODEV; | |
3711 | } | |
c1c9c9c4 BP |
3712 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
3713 | tcmsg->tcm_parent = TC_H_ROOT; | |
3714 | ||
3715 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3716 | ||
3717 | memset(&opt, 0, sizeof opt); | |
4f631ccd | 3718 | opt.rate2quantum = HTB_RATE2QUANTUM; |
c1c9c9c4 | 3719 | opt.version = 3; |
4ecf12d5 | 3720 | opt.defcls = 1; |
c1c9c9c4 BP |
3721 | |
3722 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3723 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
3724 | nl_msg_end_nested(&request, opt_offset); | |
3725 | ||
3726 | return tc_transact(&request, NULL); | |
3727 | } | |
3728 | ||
3729 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
3730 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
3731 | static int | |
3732 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
3733 | unsigned int parent, struct htb_class *class) | |
3734 | { | |
3735 | size_t opt_offset; | |
3736 | struct tc_htb_opt opt; | |
3737 | struct ofpbuf request; | |
3738 | struct tcmsg *tcmsg; | |
3739 | int error; | |
3740 | int mtu; | |
3741 | ||
73371c09 | 3742 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 3743 | if (error) { |
f915f1a8 BP |
3744 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
3745 | netdev_get_name(netdev)); | |
9b020780 | 3746 | return error; |
f915f1a8 | 3747 | } |
c1c9c9c4 BP |
3748 | |
3749 | memset(&opt, 0, sizeof opt); | |
3750 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
3751 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
4f631ccd AW |
3752 | /* Makes sure the quantum is at least MTU. Setting quantum will |
3753 | * make htb ignore the r2q for this class. */ | |
3754 | if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) { | |
3755 | opt.quantum = mtu; | |
3756 | } | |
c1c9c9c4 BP |
3757 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); |
3758 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
3759 | opt.prio = class->priority; | |
3760 | ||
7874bdff RD |
3761 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
3762 | &request); | |
23a98ffe BP |
3763 | if (!tcmsg) { |
3764 | return ENODEV; | |
3765 | } | |
c1c9c9c4 BP |
3766 | tcmsg->tcm_handle = handle; |
3767 | tcmsg->tcm_parent = parent; | |
3768 | ||
3769 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3770 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3771 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
3772 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
3773 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
3774 | nl_msg_end_nested(&request, opt_offset); | |
3775 | ||
3776 | error = tc_transact(&request, NULL); | |
3777 | if (error) { | |
3778 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3779 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
3780 | netdev_get_name(netdev), | |
3781 | tc_get_major(handle), tc_get_minor(handle), | |
3782 | tc_get_major(parent), tc_get_minor(parent), | |
3783 | class->min_rate, class->max_rate, | |
10a89ef0 | 3784 | class->burst, class->priority, ovs_strerror(error)); |
c1c9c9c4 BP |
3785 | } |
3786 | return error; | |
3787 | } | |
3788 | ||
3789 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
3790 | * description of them into 'details'. The description complies with the | |
3791 | * specification given in the vswitch database documentation for linux-htb | |
3792 | * queue details. */ | |
3793 | static int | |
3794 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
3795 | { | |
3796 | static const struct nl_policy tca_htb_policy[] = { | |
3797 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
3798 | .min_len = sizeof(struct tc_htb_opt) }, | |
3799 | }; | |
3800 | ||
3801 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
3802 | const struct tc_htb_opt *htb; | |
3803 | ||
3804 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
3805 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
3806 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
3807 | return EPROTO; | |
3808 | } | |
3809 | ||
3810 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
3811 | class->min_rate = htb->rate.rate; | |
3812 | class->max_rate = htb->ceil.rate; | |
3813 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
3814 | class->priority = htb->prio; | |
3815 | return 0; | |
3816 | } | |
3817 | ||
3818 | static int | |
3819 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
3820 | struct htb_class *options, | |
3821 | struct netdev_queue_stats *stats) | |
3822 | { | |
3823 | struct nlattr *nl_options; | |
3824 | unsigned int handle; | |
3825 | int error; | |
3826 | ||
3827 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
3828 | if (!error && queue_id) { | |
17ee3c1f BP |
3829 | unsigned int major = tc_get_major(handle); |
3830 | unsigned int minor = tc_get_minor(handle); | |
3831 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
3832 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
3833 | } else { |
3834 | error = EPROTO; | |
3835 | } | |
3836 | } | |
3837 | if (!error && options) { | |
3838 | error = htb_parse_tca_options__(nl_options, options); | |
3839 | } | |
3840 | return error; | |
3841 | } | |
3842 | ||
3843 | static void | |
73371c09 | 3844 | htb_parse_qdisc_details__(struct netdev *netdev_, |
79f1cbe9 | 3845 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 | 3846 | { |
73371c09 | 3847 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 | 3848 | |
13c1637f | 3849 | hc->max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
c1c9c9c4 | 3850 | if (!hc->max_rate) { |
a00ca915 | 3851 | enum netdev_features current; |
c1c9c9c4 | 3852 | |
73371c09 BP |
3853 | netdev_linux_read_features(netdev); |
3854 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 3855 | hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
c1c9c9c4 BP |
3856 | } |
3857 | hc->min_rate = hc->max_rate; | |
3858 | hc->burst = 0; | |
3859 | hc->priority = 0; | |
3860 | } | |
3861 | ||
3862 | static int | |
3863 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 3864 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
3865 | { |
3866 | const struct htb *htb = htb_get__(netdev); | |
9b020780 | 3867 | int mtu, error; |
214117fd | 3868 | unsigned long long int max_rate_bit; |
c1c9c9c4 | 3869 | |
73371c09 | 3870 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 3871 | if (error) { |
f915f1a8 BP |
3872 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
3873 | netdev_get_name(netdev)); | |
9b020780 | 3874 | return error; |
f915f1a8 BP |
3875 | } |
3876 | ||
4f104611 EJ |
3877 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
3878 | * on uncongested links. */ | |
13c1637f | 3879 | hc->min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
4f104611 | 3880 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
3881 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
3882 | ||
3883 | /* max-rate */ | |
214117fd KF |
3884 | max_rate_bit = smap_get_ullong(details, "max-rate", 0); |
3885 | hc->max_rate = max_rate_bit ? max_rate_bit / 8 : htb->max_rate; | |
c1c9c9c4 BP |
3886 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); |
3887 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
3888 | ||
3889 | /* burst | |
3890 | * | |
3891 | * According to hints in the documentation that I've read, it is important | |
3892 | * that 'burst' be at least as big as the largest frame that might be | |
3893 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
3894 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
3895 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
3896 | * the MTU. We actually add 64, instead of 14, as a guard against | |
3897 | * additional headers get tacked on somewhere that we're not aware of. */ | |
13c1637f | 3898 | hc->burst = smap_get_ullong(details, "burst", 0) / 8; |
c1c9c9c4 BP |
3899 | hc->burst = MAX(hc->burst, mtu + 64); |
3900 | ||
3901 | /* priority */ | |
13c1637f | 3902 | hc->priority = smap_get_ullong(details, "priority", 0); |
c1c9c9c4 BP |
3903 | |
3904 | return 0; | |
3905 | } | |
3906 | ||
3907 | static int | |
3908 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
3909 | unsigned int parent, struct htb_class *options, | |
3910 | struct netdev_queue_stats *stats) | |
3911 | { | |
3912 | struct ofpbuf *reply; | |
3913 | int error; | |
3914 | ||
3915 | error = tc_query_class(netdev, handle, parent, &reply); | |
3916 | if (!error) { | |
3917 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
3918 | ofpbuf_delete(reply); | |
3919 | } | |
3920 | return error; | |
3921 | } | |
3922 | ||
3923 | static int | |
79f1cbe9 | 3924 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
3925 | { |
3926 | int error; | |
3927 | ||
3928 | error = htb_setup_qdisc__(netdev); | |
3929 | if (!error) { | |
3930 | struct htb_class hc; | |
3931 | ||
3932 | htb_parse_qdisc_details__(netdev, details, &hc); | |
3933 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
3934 | tc_make_handle(1, 0), &hc); | |
3935 | if (!error) { | |
3936 | htb_install__(netdev, hc.max_rate); | |
3937 | } | |
3938 | } | |
3939 | return error; | |
3940 | } | |
3941 | ||
93b13be8 BP |
3942 | static struct htb_class * |
3943 | htb_class_cast__(const struct tc_queue *queue) | |
3944 | { | |
3945 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
3946 | } | |
3947 | ||
c1c9c9c4 BP |
3948 | static void |
3949 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
3950 | const struct htb_class *hc) | |
3951 | { | |
3952 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
3953 | size_t hash = hash_int(queue_id, 0); |
3954 | struct tc_queue *queue; | |
c1c9c9c4 BP |
3955 | struct htb_class *hcp; |
3956 | ||
93b13be8 BP |
3957 | queue = tc_find_queue__(netdev, queue_id, hash); |
3958 | if (queue) { | |
3959 | hcp = htb_class_cast__(queue); | |
3960 | } else { | |
c1c9c9c4 | 3961 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
3962 | queue = &hcp->tc_queue; |
3963 | queue->queue_id = queue_id; | |
6dc34a0d | 3964 | queue->created = time_msec(); |
93b13be8 | 3965 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); |
c1c9c9c4 | 3966 | } |
93b13be8 BP |
3967 | |
3968 | hcp->min_rate = hc->min_rate; | |
3969 | hcp->max_rate = hc->max_rate; | |
3970 | hcp->burst = hc->burst; | |
3971 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
3972 | } |
3973 | ||
3974 | static int | |
3975 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
3976 | { | |
c1c9c9c4 | 3977 | struct ofpbuf msg; |
d57695d7 | 3978 | struct queue_dump_state state; |
c1c9c9c4 | 3979 | struct htb_class hc; |
c1c9c9c4 BP |
3980 | |
3981 | /* Get qdisc options. */ | |
3982 | hc.max_rate = 0; | |
3983 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 3984 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
3985 | |
3986 | /* Get queues. */ | |
d57695d7 | 3987 | if (!start_queue_dump(netdev, &state)) { |
23a98ffe BP |
3988 | return ENODEV; |
3989 | } | |
d57695d7 | 3990 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
c1c9c9c4 BP |
3991 | unsigned int queue_id; |
3992 | ||
3993 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
3994 | htb_update_queue__(netdev, queue_id, &hc); | |
3995 | } | |
3996 | } | |
d57695d7 | 3997 | finish_queue_dump(&state); |
c1c9c9c4 BP |
3998 | |
3999 | return 0; | |
4000 | } | |
4001 | ||
4002 | static void | |
4003 | htb_tc_destroy(struct tc *tc) | |
4004 | { | |
4005 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
4ec3d7c7 | 4006 | struct htb_class *hc; |
c1c9c9c4 | 4007 | |
4ec3d7c7 | 4008 | HMAP_FOR_EACH_POP (hc, tc_queue.hmap_node, &htb->tc.queues) { |
c1c9c9c4 BP |
4009 | free(hc); |
4010 | } | |
4011 | tc_destroy(tc); | |
4012 | free(htb); | |
4013 | } | |
4014 | ||
4015 | static int | |
79f1cbe9 | 4016 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
4017 | { |
4018 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 4019 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
4020 | return 0; |
4021 | } | |
4022 | ||
4023 | static int | |
79f1cbe9 | 4024 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
4025 | { |
4026 | struct htb_class hc; | |
4027 | int error; | |
4028 | ||
4029 | htb_parse_qdisc_details__(netdev, details, &hc); | |
4030 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4031 | tc_make_handle(1, 0), &hc); | |
4032 | if (!error) { | |
4033 | htb_get__(netdev)->max_rate = hc.max_rate; | |
4034 | } | |
4035 | return error; | |
4036 | } | |
4037 | ||
4038 | static int | |
93b13be8 | 4039 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 4040 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 4041 | { |
93b13be8 | 4042 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4043 | |
79f1cbe9 | 4044 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 4045 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4046 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 4047 | } |
79f1cbe9 | 4048 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 4049 | if (hc->priority) { |
79f1cbe9 | 4050 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
4051 | } |
4052 | return 0; | |
4053 | } | |
4054 | ||
4055 | static int | |
4056 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4057 | const struct smap *details) |
c1c9c9c4 BP |
4058 | { |
4059 | struct htb_class hc; | |
4060 | int error; | |
4061 | ||
4062 | error = htb_parse_class_details__(netdev, details, &hc); | |
4063 | if (error) { | |
4064 | return error; | |
4065 | } | |
4066 | ||
17ee3c1f | 4067 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
4068 | tc_make_handle(1, 0xfffe), &hc); |
4069 | if (error) { | |
4070 | return error; | |
4071 | } | |
4072 | ||
4073 | htb_update_queue__(netdev, queue_id, &hc); | |
4074 | return 0; | |
4075 | } | |
4076 | ||
4077 | static int | |
93b13be8 | 4078 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 4079 | { |
93b13be8 | 4080 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4081 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
4082 | int error; |
4083 | ||
93b13be8 | 4084 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 4085 | if (!error) { |
93b13be8 | 4086 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 4087 | free(hc); |
c1c9c9c4 BP |
4088 | } |
4089 | return error; | |
4090 | } | |
4091 | ||
4092 | static int | |
93b13be8 | 4093 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
4094 | struct netdev_queue_stats *stats) |
4095 | { | |
93b13be8 | 4096 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
4097 | tc_make_handle(1, 0xfffe), NULL, stats); |
4098 | } | |
4099 | ||
4100 | static int | |
4101 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4102 | const struct ofpbuf *nlmsg, | |
4103 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4104 | { | |
4105 | struct netdev_queue_stats stats; | |
17ee3c1f | 4106 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
4107 | int error; |
4108 | ||
4109 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4110 | if (error) { | |
4111 | return error; | |
4112 | } | |
4113 | ||
17ee3c1f BP |
4114 | major = tc_get_major(handle); |
4115 | minor = tc_get_minor(handle); | |
4116 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 4117 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
4118 | } |
4119 | return 0; | |
4120 | } | |
4121 | ||
4122 | static const struct tc_ops tc_ops_htb = { | |
4123 | "htb", /* linux_name */ | |
4124 | "linux-htb", /* ovs_name */ | |
4125 | HTB_N_QUEUES, /* n_queues */ | |
4126 | htb_tc_install, | |
4127 | htb_tc_load, | |
4128 | htb_tc_destroy, | |
4129 | htb_qdisc_get, | |
4130 | htb_qdisc_set, | |
4131 | htb_class_get, | |
4132 | htb_class_set, | |
4133 | htb_class_delete, | |
4134 | htb_class_get_stats, | |
4135 | htb_class_dump_stats | |
4136 | }; | |
4137 | \f | |
a339aa81 EJ |
4138 | /* "linux-hfsc" traffic control class. */ |
4139 | ||
4140 | #define HFSC_N_QUEUES 0xf000 | |
4141 | ||
4142 | struct hfsc { | |
4143 | struct tc tc; | |
4144 | uint32_t max_rate; | |
4145 | }; | |
4146 | ||
4147 | struct hfsc_class { | |
4148 | struct tc_queue tc_queue; | |
4149 | uint32_t min_rate; | |
4150 | uint32_t max_rate; | |
4151 | }; | |
4152 | ||
4153 | static struct hfsc * | |
b5d57fc8 | 4154 | hfsc_get__(const struct netdev *netdev_) |
a339aa81 | 4155 | { |
b5d57fc8 BP |
4156 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4157 | return CONTAINER_OF(netdev->tc, struct hfsc, tc); | |
a339aa81 EJ |
4158 | } |
4159 | ||
4160 | static struct hfsc_class * | |
4161 | hfsc_class_cast__(const struct tc_queue *queue) | |
4162 | { | |
4163 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
4164 | } | |
4165 | ||
24045e35 | 4166 | static void |
b5d57fc8 | 4167 | hfsc_install__(struct netdev *netdev_, uint32_t max_rate) |
a339aa81 | 4168 | { |
b5d57fc8 | 4169 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
4170 | struct hfsc *hfsc; |
4171 | ||
a339aa81 EJ |
4172 | hfsc = xmalloc(sizeof *hfsc); |
4173 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
4174 | hfsc->max_rate = max_rate; | |
b5d57fc8 | 4175 | netdev->tc = &hfsc->tc; |
a339aa81 EJ |
4176 | } |
4177 | ||
4178 | static void | |
4179 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4180 | const struct hfsc_class *hc) | |
4181 | { | |
4182 | size_t hash; | |
4183 | struct hfsc *hfsc; | |
4184 | struct hfsc_class *hcp; | |
4185 | struct tc_queue *queue; | |
4186 | ||
4187 | hfsc = hfsc_get__(netdev); | |
4188 | hash = hash_int(queue_id, 0); | |
4189 | ||
4190 | queue = tc_find_queue__(netdev, queue_id, hash); | |
4191 | if (queue) { | |
4192 | hcp = hfsc_class_cast__(queue); | |
4193 | } else { | |
4194 | hcp = xmalloc(sizeof *hcp); | |
4195 | queue = &hcp->tc_queue; | |
4196 | queue->queue_id = queue_id; | |
6dc34a0d | 4197 | queue->created = time_msec(); |
a339aa81 EJ |
4198 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); |
4199 | } | |
4200 | ||
4201 | hcp->min_rate = hc->min_rate; | |
4202 | hcp->max_rate = hc->max_rate; | |
4203 | } | |
4204 | ||
4205 | static int | |
4206 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
4207 | { | |
4208 | const struct tc_service_curve *rsc, *fsc, *usc; | |
4209 | static const struct nl_policy tca_hfsc_policy[] = { | |
4210 | [TCA_HFSC_RSC] = { | |
4211 | .type = NL_A_UNSPEC, | |
4212 | .optional = false, | |
4213 | .min_len = sizeof(struct tc_service_curve), | |
4214 | }, | |
4215 | [TCA_HFSC_FSC] = { | |
4216 | .type = NL_A_UNSPEC, | |
4217 | .optional = false, | |
4218 | .min_len = sizeof(struct tc_service_curve), | |
4219 | }, | |
4220 | [TCA_HFSC_USC] = { | |
4221 | .type = NL_A_UNSPEC, | |
4222 | .optional = false, | |
4223 | .min_len = sizeof(struct tc_service_curve), | |
4224 | }, | |
4225 | }; | |
4226 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
4227 | ||
4228 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
4229 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
4230 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
4231 | return EPROTO; | |
4232 | } | |
4233 | ||
4234 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
4235 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
4236 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
4237 | ||
4238 | if (rsc->m1 != 0 || rsc->d != 0 || | |
4239 | fsc->m1 != 0 || fsc->d != 0 || | |
4240 | usc->m1 != 0 || usc->d != 0) { | |
4241 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4242 | "Non-linear service curves are not supported."); | |
4243 | return EPROTO; | |
4244 | } | |
4245 | ||
4246 | if (rsc->m2 != fsc->m2) { | |
4247 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4248 | "Real-time service curves are not supported "); | |
4249 | return EPROTO; | |
4250 | } | |
4251 | ||
4252 | if (rsc->m2 > usc->m2) { | |
4253 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4254 | "Min-rate service curve is greater than " | |
4255 | "the max-rate service curve."); | |
4256 | return EPROTO; | |
4257 | } | |
4258 | ||
4259 | class->min_rate = fsc->m2; | |
4260 | class->max_rate = usc->m2; | |
4261 | return 0; | |
4262 | } | |
4263 | ||
4264 | static int | |
4265 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
4266 | struct hfsc_class *options, | |
4267 | struct netdev_queue_stats *stats) | |
4268 | { | |
4269 | int error; | |
4270 | unsigned int handle; | |
4271 | struct nlattr *nl_options; | |
4272 | ||
4273 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
4274 | if (error) { | |
4275 | return error; | |
4276 | } | |
4277 | ||
4278 | if (queue_id) { | |
4279 | unsigned int major, minor; | |
4280 | ||
4281 | major = tc_get_major(handle); | |
4282 | minor = tc_get_minor(handle); | |
4283 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4284 | *queue_id = minor - 1; | |
4285 | } else { | |
4286 | return EPROTO; | |
4287 | } | |
4288 | } | |
4289 | ||
4290 | if (options) { | |
4291 | error = hfsc_parse_tca_options__(nl_options, options); | |
4292 | } | |
4293 | ||
4294 | return error; | |
4295 | } | |
4296 | ||
4297 | static int | |
4298 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
4299 | unsigned int parent, struct hfsc_class *options, | |
4300 | struct netdev_queue_stats *stats) | |
4301 | { | |
4302 | int error; | |
4303 | struct ofpbuf *reply; | |
4304 | ||
4305 | error = tc_query_class(netdev, handle, parent, &reply); | |
4306 | if (error) { | |
4307 | return error; | |
4308 | } | |
4309 | ||
4310 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
4311 | ofpbuf_delete(reply); | |
4312 | return error; | |
4313 | } | |
4314 | ||
4315 | static void | |
73371c09 | 4316 | hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, |
a339aa81 EJ |
4317 | struct hfsc_class *class) |
4318 | { | |
73371c09 | 4319 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 | 4320 | |
13c1637f | 4321 | uint32_t max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
a339aa81 | 4322 | if (!max_rate) { |
a00ca915 | 4323 | enum netdev_features current; |
a339aa81 | 4324 | |
73371c09 BP |
4325 | netdev_linux_read_features(netdev); |
4326 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 4327 | max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
a339aa81 EJ |
4328 | } |
4329 | ||
4330 | class->min_rate = max_rate; | |
4331 | class->max_rate = max_rate; | |
4332 | } | |
4333 | ||
4334 | static int | |
4335 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 4336 | const struct smap *details, |
a339aa81 EJ |
4337 | struct hfsc_class * class) |
4338 | { | |
4339 | const struct hfsc *hfsc; | |
4340 | uint32_t min_rate, max_rate; | |
a339aa81 EJ |
4341 | |
4342 | hfsc = hfsc_get__(netdev); | |
a339aa81 | 4343 | |
13c1637f | 4344 | min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
79398bad | 4345 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
4346 | min_rate = MIN(min_rate, hfsc->max_rate); |
4347 | ||
13c1637f | 4348 | max_rate = smap_get_ullong(details, "max-rate", hfsc->max_rate * 8) / 8; |
a339aa81 EJ |
4349 | max_rate = MAX(max_rate, min_rate); |
4350 | max_rate = MIN(max_rate, hfsc->max_rate); | |
4351 | ||
4352 | class->min_rate = min_rate; | |
4353 | class->max_rate = max_rate; | |
4354 | ||
4355 | return 0; | |
4356 | } | |
4357 | ||
4358 | /* Create an HFSC qdisc. | |
4359 | * | |
4360 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
4361 | static int | |
4362 | hfsc_setup_qdisc__(struct netdev * netdev) | |
4363 | { | |
4364 | struct tcmsg *tcmsg; | |
4365 | struct ofpbuf request; | |
4366 | struct tc_hfsc_qopt opt; | |
4367 | ||
4368 | tc_del_qdisc(netdev); | |
4369 | ||
7874bdff RD |
4370 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
4371 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
a339aa81 EJ |
4372 | |
4373 | if (!tcmsg) { | |
4374 | return ENODEV; | |
4375 | } | |
4376 | ||
4377 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
4378 | tcmsg->tcm_parent = TC_H_ROOT; | |
4379 | ||
4380 | memset(&opt, 0, sizeof opt); | |
4381 | opt.defcls = 1; | |
4382 | ||
4383 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4384 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
4385 | ||
4386 | return tc_transact(&request, NULL); | |
4387 | } | |
4388 | ||
4389 | /* Create an HFSC class. | |
4390 | * | |
4391 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
4392 | * sc rate <min_rate> ul rate <max_rate>" */ | |
4393 | static int | |
4394 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
4395 | unsigned int parent, struct hfsc_class *class) | |
4396 | { | |
4397 | int error; | |
4398 | size_t opt_offset; | |
4399 | struct tcmsg *tcmsg; | |
4400 | struct ofpbuf request; | |
4401 | struct tc_service_curve min, max; | |
4402 | ||
7874bdff RD |
4403 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
4404 | &request); | |
a339aa81 EJ |
4405 | |
4406 | if (!tcmsg) { | |
4407 | return ENODEV; | |
4408 | } | |
4409 | ||
4410 | tcmsg->tcm_handle = handle; | |
4411 | tcmsg->tcm_parent = parent; | |
4412 | ||
4413 | min.m1 = 0; | |
4414 | min.d = 0; | |
4415 | min.m2 = class->min_rate; | |
4416 | ||
4417 | max.m1 = 0; | |
4418 | max.d = 0; | |
4419 | max.m2 = class->max_rate; | |
4420 | ||
4421 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4422 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4423 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
4424 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
4425 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
4426 | nl_msg_end_nested(&request, opt_offset); | |
4427 | ||
4428 | error = tc_transact(&request, NULL); | |
4429 | if (error) { | |
4430 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
4431 | "min-rate %ubps, max-rate %ubps (%s)", | |
4432 | netdev_get_name(netdev), | |
4433 | tc_get_major(handle), tc_get_minor(handle), | |
4434 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 4435 | class->min_rate, class->max_rate, ovs_strerror(error)); |
a339aa81 EJ |
4436 | } |
4437 | ||
4438 | return error; | |
4439 | } | |
4440 | ||
4441 | static int | |
79f1cbe9 | 4442 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4443 | { |
4444 | int error; | |
4445 | struct hfsc_class class; | |
4446 | ||
4447 | error = hfsc_setup_qdisc__(netdev); | |
4448 | ||
4449 | if (error) { | |
4450 | return error; | |
4451 | } | |
4452 | ||
4453 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4454 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4455 | tc_make_handle(1, 0), &class); | |
4456 | ||
4457 | if (error) { | |
4458 | return error; | |
4459 | } | |
4460 | ||
4461 | hfsc_install__(netdev, class.max_rate); | |
4462 | return 0; | |
4463 | } | |
4464 | ||
4465 | static int | |
4466 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4467 | { | |
4468 | struct ofpbuf msg; | |
d57695d7 | 4469 | struct queue_dump_state state; |
a339aa81 EJ |
4470 | struct hfsc_class hc; |
4471 | ||
4472 | hc.max_rate = 0; | |
4473 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 4474 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 | 4475 | |
d57695d7 | 4476 | if (!start_queue_dump(netdev, &state)) { |
a339aa81 EJ |
4477 | return ENODEV; |
4478 | } | |
4479 | ||
d57695d7 | 4480 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
a339aa81 EJ |
4481 | unsigned int queue_id; |
4482 | ||
4483 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
4484 | hfsc_update_queue__(netdev, queue_id, &hc); | |
4485 | } | |
4486 | } | |
4487 | ||
d57695d7 | 4488 | finish_queue_dump(&state); |
a339aa81 EJ |
4489 | return 0; |
4490 | } | |
4491 | ||
4492 | static void | |
4493 | hfsc_tc_destroy(struct tc *tc) | |
4494 | { | |
4495 | struct hfsc *hfsc; | |
4496 | struct hfsc_class *hc, *next; | |
4497 | ||
4498 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
4499 | ||
4500 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
4501 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4502 | free(hc); | |
4503 | } | |
4504 | ||
4505 | tc_destroy(tc); | |
4506 | free(hfsc); | |
4507 | } | |
4508 | ||
4509 | static int | |
79f1cbe9 | 4510 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
4511 | { |
4512 | const struct hfsc *hfsc; | |
4513 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 4514 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
4515 | return 0; |
4516 | } | |
4517 | ||
4518 | static int | |
79f1cbe9 | 4519 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4520 | { |
4521 | int error; | |
4522 | struct hfsc_class class; | |
4523 | ||
4524 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4525 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4526 | tc_make_handle(1, 0), &class); | |
4527 | ||
4528 | if (!error) { | |
4529 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
4530 | } | |
4531 | ||
4532 | return error; | |
4533 | } | |
4534 | ||
4535 | static int | |
4536 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 4537 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
4538 | { |
4539 | const struct hfsc_class *hc; | |
4540 | ||
4541 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 4542 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 4543 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4544 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
4545 | } |
4546 | return 0; | |
4547 | } | |
4548 | ||
4549 | static int | |
4550 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4551 | const struct smap *details) |
a339aa81 EJ |
4552 | { |
4553 | int error; | |
4554 | struct hfsc_class class; | |
4555 | ||
4556 | error = hfsc_parse_class_details__(netdev, details, &class); | |
4557 | if (error) { | |
4558 | return error; | |
4559 | } | |
4560 | ||
4561 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
4562 | tc_make_handle(1, 0xfffe), &class); | |
4563 | if (error) { | |
4564 | return error; | |
4565 | } | |
4566 | ||
4567 | hfsc_update_queue__(netdev, queue_id, &class); | |
4568 | return 0; | |
4569 | } | |
4570 | ||
4571 | static int | |
4572 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
4573 | { | |
4574 | int error; | |
4575 | struct hfsc *hfsc; | |
4576 | struct hfsc_class *hc; | |
4577 | ||
4578 | hc = hfsc_class_cast__(queue); | |
4579 | hfsc = hfsc_get__(netdev); | |
4580 | ||
4581 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
4582 | if (!error) { | |
4583 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4584 | free(hc); | |
4585 | } | |
4586 | return error; | |
4587 | } | |
4588 | ||
4589 | static int | |
4590 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
4591 | struct netdev_queue_stats *stats) | |
4592 | { | |
4593 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
4594 | tc_make_handle(1, 0xfffe), NULL, stats); | |
4595 | } | |
4596 | ||
4597 | static int | |
4598 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4599 | const struct ofpbuf *nlmsg, | |
4600 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4601 | { | |
4602 | struct netdev_queue_stats stats; | |
4603 | unsigned int handle, major, minor; | |
4604 | int error; | |
4605 | ||
4606 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4607 | if (error) { | |
4608 | return error; | |
4609 | } | |
4610 | ||
4611 | major = tc_get_major(handle); | |
4612 | minor = tc_get_minor(handle); | |
4613 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4614 | (*cb)(minor - 1, &stats, aux); | |
4615 | } | |
4616 | return 0; | |
4617 | } | |
4618 | ||
4619 | static const struct tc_ops tc_ops_hfsc = { | |
4620 | "hfsc", /* linux_name */ | |
4621 | "linux-hfsc", /* ovs_name */ | |
4622 | HFSC_N_QUEUES, /* n_queues */ | |
4623 | hfsc_tc_install, /* tc_install */ | |
4624 | hfsc_tc_load, /* tc_load */ | |
4625 | hfsc_tc_destroy, /* tc_destroy */ | |
4626 | hfsc_qdisc_get, /* qdisc_get */ | |
4627 | hfsc_qdisc_set, /* qdisc_set */ | |
4628 | hfsc_class_get, /* class_get */ | |
4629 | hfsc_class_set, /* class_set */ | |
4630 | hfsc_class_delete, /* class_delete */ | |
4631 | hfsc_class_get_stats, /* class_get_stats */ | |
4632 | hfsc_class_dump_stats /* class_dump_stats */ | |
4633 | }; | |
4634 | \f | |
6cf888b8 BS |
4635 | /* "linux-noop" traffic control class. */ |
4636 | ||
4637 | static void | |
4638 | noop_install__(struct netdev *netdev_) | |
4639 | { | |
4640 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4641 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); | |
4642 | ||
4643 | netdev->tc = CONST_CAST(struct tc *, &tc); | |
4644 | } | |
4645 | ||
4646 | static int | |
4647 | noop_tc_install(struct netdev *netdev, | |
4648 | const struct smap *details OVS_UNUSED) | |
4649 | { | |
4650 | noop_install__(netdev); | |
4651 | return 0; | |
4652 | } | |
4653 | ||
4654 | static int | |
4655 | noop_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4656 | { | |
4657 | noop_install__(netdev); | |
4658 | return 0; | |
4659 | } | |
4660 | ||
4661 | static const struct tc_ops tc_ops_noop = { | |
4662 | NULL, /* linux_name */ | |
4663 | "linux-noop", /* ovs_name */ | |
4664 | 0, /* n_queues */ | |
4665 | noop_tc_install, | |
4666 | noop_tc_load, | |
4667 | NULL, /* tc_destroy */ | |
4668 | NULL, /* qdisc_get */ | |
4669 | NULL, /* qdisc_set */ | |
4670 | NULL, /* class_get */ | |
4671 | NULL, /* class_set */ | |
4672 | NULL, /* class_delete */ | |
4673 | NULL, /* class_get_stats */ | |
4674 | NULL /* class_dump_stats */ | |
4675 | }; | |
4676 | \f | |
c1c9c9c4 BP |
4677 | /* "linux-default" traffic control class. |
4678 | * | |
4679 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
4680 | * the "" (empty string) QoS type in the OVS database. */ | |
4681 | ||
4682 | static void | |
b5d57fc8 | 4683 | default_install__(struct netdev *netdev_) |
c1c9c9c4 | 4684 | { |
b5d57fc8 | 4685 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4686 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); |
c1c9c9c4 | 4687 | |
559eb230 BP |
4688 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4689 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4690 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4691 | } |
4692 | ||
4693 | static int | |
4694 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 4695 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
4696 | { |
4697 | default_install__(netdev); | |
4698 | return 0; | |
4699 | } | |
4700 | ||
4701 | static int | |
4702 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4703 | { | |
4704 | default_install__(netdev); | |
4705 | return 0; | |
4706 | } | |
4707 | ||
4708 | static const struct tc_ops tc_ops_default = { | |
4709 | NULL, /* linux_name */ | |
4710 | "", /* ovs_name */ | |
4711 | 0, /* n_queues */ | |
4712 | default_tc_install, | |
4713 | default_tc_load, | |
4714 | NULL, /* tc_destroy */ | |
4715 | NULL, /* qdisc_get */ | |
4716 | NULL, /* qdisc_set */ | |
4717 | NULL, /* class_get */ | |
4718 | NULL, /* class_set */ | |
4719 | NULL, /* class_delete */ | |
4720 | NULL, /* class_get_stats */ | |
4721 | NULL /* class_dump_stats */ | |
4722 | }; | |
4723 | \f | |
4724 | /* "linux-other" traffic control class. | |
4725 | * | |
4726 | * */ | |
4727 | ||
4728 | static int | |
b5d57fc8 | 4729 | other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) |
c1c9c9c4 | 4730 | { |
b5d57fc8 | 4731 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4732 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); |
c1c9c9c4 | 4733 | |
559eb230 BP |
4734 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4735 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4736 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4737 | return 0; |
4738 | } | |
4739 | ||
4740 | static const struct tc_ops tc_ops_other = { | |
4741 | NULL, /* linux_name */ | |
4742 | "linux-other", /* ovs_name */ | |
4743 | 0, /* n_queues */ | |
4744 | NULL, /* tc_install */ | |
4745 | other_tc_load, | |
4746 | NULL, /* tc_destroy */ | |
4747 | NULL, /* qdisc_get */ | |
4748 | NULL, /* qdisc_set */ | |
4749 | NULL, /* class_get */ | |
4750 | NULL, /* class_set */ | |
4751 | NULL, /* class_delete */ | |
4752 | NULL, /* class_get_stats */ | |
4753 | NULL /* class_dump_stats */ | |
4754 | }; | |
4755 | \f | |
4756 | /* Traffic control. */ | |
4757 | ||
4758 | /* Number of kernel "tc" ticks per second. */ | |
4759 | static double ticks_per_s; | |
4760 | ||
4761 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
4762 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
4763 | * one jiffy's worth of data. | |
4764 | * | |
4765 | * There are two possibilities here: | |
4766 | * | |
4767 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
4768 | * approximate range of 100 to 1024. That means that we really need to | |
4769 | * make sure that the qdisc can buffer that much data. | |
4770 | * | |
4771 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
4772 | * has finely granular timers and there's no need to fudge additional room | |
4773 | * for buffers. (There's no extra effort needed to implement that: the | |
4774 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
4775 | * come out as 0 in the division. Small integer results in the case of | |
4776 | * really high dividends won't have any real effect anyhow.) | |
4777 | */ | |
4778 | static unsigned int buffer_hz; | |
4779 | ||
7874bdff RD |
4780 | static struct tcmsg * |
4781 | netdev_linux_tc_make_request(const struct netdev *netdev, int type, | |
4782 | unsigned int flags, struct ofpbuf *request) | |
4783 | { | |
4784 | int ifindex; | |
4785 | int error; | |
4786 | ||
4787 | error = get_ifindex(netdev, &ifindex); | |
4788 | if (error) { | |
4789 | return NULL; | |
4790 | } | |
4791 | ||
4792 | return tc_make_request(ifindex, type, flags, request); | |
4793 | } | |
4794 | ||
f8500004 JP |
4795 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size |
4796 | * of 'kbits_burst'. | |
4797 | * | |
4798 | * This function is equivalent to running: | |
4799 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
4800 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
4801 | * mtu 65535 drop | |
4802 | * | |
4803 | * The configuration and stats may be seen with the following command: | |
c7952afb | 4804 | * /sbin/tc -s filter show dev <devname> parent ffff: |
f8500004 JP |
4805 | * |
4806 | * Returns 0 if successful, otherwise a positive errno value. | |
4807 | */ | |
4808 | static int | |
c7952afb BP |
4809 | tc_add_policer(struct netdev *netdev, |
4810 | uint32_t kbits_rate, uint32_t kbits_burst) | |
f8500004 JP |
4811 | { |
4812 | struct tc_police tc_police; | |
4813 | struct ofpbuf request; | |
4814 | struct tcmsg *tcmsg; | |
4815 | size_t basic_offset; | |
4816 | size_t police_offset; | |
4817 | int error; | |
4818 | int mtu = 65535; | |
4819 | ||
4820 | memset(&tc_police, 0, sizeof tc_police); | |
4821 | tc_police.action = TC_POLICE_SHOT; | |
4822 | tc_police.mtu = mtu; | |
1aca400c | 4823 | tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu); |
c7952afb | 4824 | |
79abacc8 MAA |
4825 | /* The following appears wrong in one way: In networking a kilobit is |
4826 | * usually 1000 bits but this uses 1024 bits. | |
c7952afb BP |
4827 | * |
4828 | * However if you "fix" those problems then "tc filter show ..." shows | |
4829 | * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit == | |
4830 | * 1,000,000 bits, whereas this actually ends up doing the right thing from | |
4831 | * tc's point of view. Whatever. */ | |
4832 | tc_police.burst = tc_bytes_to_ticks( | |
79abacc8 | 4833 | tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 8); |
f8500004 | 4834 | |
7874bdff RD |
4835 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTFILTER, |
4836 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
f8500004 JP |
4837 | if (!tcmsg) { |
4838 | return ENODEV; | |
4839 | } | |
4840 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
4841 | tcmsg->tcm_info = tc_make_handle(49, | |
4842 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
4843 | ||
4844 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
4845 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4846 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
4847 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
4848 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
4849 | nl_msg_end_nested(&request, police_offset); | |
4850 | nl_msg_end_nested(&request, basic_offset); | |
4851 | ||
4852 | error = tc_transact(&request, NULL); | |
4853 | if (error) { | |
4854 | return error; | |
4855 | } | |
4856 | ||
4857 | return 0; | |
4858 | } | |
4859 | ||
c1c9c9c4 BP |
4860 | static void |
4861 | read_psched(void) | |
4862 | { | |
4863 | /* The values in psched are not individually very meaningful, but they are | |
4864 | * important. The tables below show some values seen in the wild. | |
4865 | * | |
4866 | * Some notes: | |
4867 | * | |
4868 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
4869 | * (Before that, there are hints that it was 1000000000.) | |
4870 | * | |
4871 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
4872 | * above. | |
4873 | * | |
4874 | * /proc/net/psched | |
4875 | * ----------------------------------- | |
4876 | * [1] 000c8000 000f4240 000f4240 00000064 | |
4877 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
4878 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
4879 | * [4] 000003e8 00000400 000f4240 00000064 | |
4880 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
4881 | * [6] 000003e8 00000040 000f4240 000000f9 | |
4882 | * | |
4883 | * a b c d ticks_per_s buffer_hz | |
4884 | * ------- --------- ---------- ------------- ----------- ------------- | |
4885 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
4886 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
4887 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
4888 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
4889 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
4890 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
4891 | * | |
4892 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
4893 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
4894 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
4895 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
4896 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
4897 | * [6] 2.6.34 from kernel.org on KVM | |
4898 | */ | |
23882115 | 4899 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
c1c9c9c4 BP |
4900 | static const char fn[] = "/proc/net/psched"; |
4901 | unsigned int a, b, c, d; | |
4902 | FILE *stream; | |
4903 | ||
23882115 BP |
4904 | if (!ovsthread_once_start(&once)) { |
4905 | return; | |
4906 | } | |
4907 | ||
c1c9c9c4 BP |
4908 | ticks_per_s = 1.0; |
4909 | buffer_hz = 100; | |
4910 | ||
4911 | stream = fopen(fn, "r"); | |
4912 | if (!stream) { | |
10a89ef0 | 4913 | VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); |
23882115 | 4914 | goto exit; |
c1c9c9c4 BP |
4915 | } |
4916 | ||
4917 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
4918 | VLOG_WARN("%s: read failed", fn); | |
4919 | fclose(stream); | |
23882115 | 4920 | goto exit; |
c1c9c9c4 BP |
4921 | } |
4922 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
4923 | fclose(stream); | |
4924 | ||
4925 | if (!a || !c) { | |
4926 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
23882115 | 4927 | goto exit; |
c1c9c9c4 BP |
4928 | } |
4929 | ||
4930 | ticks_per_s = (double) a * c / b; | |
4931 | if (c == 1000000) { | |
4932 | buffer_hz = d; | |
4933 | } else { | |
4934 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
4935 | fn, a, b, c, d); | |
4936 | } | |
4937 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
23882115 BP |
4938 | |
4939 | exit: | |
4940 | ovsthread_once_done(&once); | |
c1c9c9c4 BP |
4941 | } |
4942 | ||
4943 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
4944 | * rate of 'rate' bytes per second. */ | |
4945 | static unsigned int | |
4946 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
4947 | { | |
23882115 | 4948 | read_psched(); |
c1c9c9c4 BP |
4949 | return (rate * ticks) / ticks_per_s; |
4950 | } | |
4951 | ||
4952 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
4953 | * rate of 'rate' bytes per second. */ | |
4954 | static unsigned int | |
4955 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
4956 | { | |
23882115 | 4957 | read_psched(); |
015c93a4 | 4958 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
4959 | } |
4960 | ||
4961 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
4962 | * a transmission rate of 'rate' bytes per second. */ | |
4963 | static unsigned int | |
4964 | tc_buffer_per_jiffy(unsigned int rate) | |
4965 | { | |
23882115 | 4966 | read_psched(); |
c1c9c9c4 BP |
4967 | return rate / buffer_hz; |
4968 | } | |
4969 | ||
4970 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
4971 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
4972 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
4973 | * stores NULL into it if it is absent. | |
4974 | * | |
4975 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
4976 | * 'msg'. | |
4977 | * | |
4978 | * Returns 0 if successful, otherwise a positive errno value. */ | |
4979 | static int | |
4980 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
4981 | struct nlattr **options) | |
4982 | { | |
4983 | static const struct nl_policy tca_policy[] = { | |
4984 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
4985 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
4986 | }; | |
4987 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
4988 | ||
4989 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
4990 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
4991 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
4992 | goto error; | |
4993 | } | |
4994 | ||
4995 | if (kind) { | |
4996 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
4997 | } | |
4998 | ||
4999 | if (options) { | |
5000 | *options = ta[TCA_OPTIONS]; | |
5001 | } | |
5002 | ||
5003 | return 0; | |
5004 | ||
5005 | error: | |
5006 | if (kind) { | |
5007 | *kind = NULL; | |
5008 | } | |
5009 | if (options) { | |
5010 | *options = NULL; | |
5011 | } | |
5012 | return EPROTO; | |
5013 | } | |
5014 | ||
5015 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
5016 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
5017 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
5018 | * arguments may be null. | |
5019 | * | |
5020 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5021 | static int | |
5022 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
5023 | struct nlattr **options, struct netdev_queue_stats *stats) | |
5024 | { | |
5025 | static const struct nl_policy tca_policy[] = { | |
5026 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
5027 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
5028 | }; | |
5029 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
5030 | ||
5031 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
5032 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
5033 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
5034 | goto error; | |
5035 | } | |
5036 | ||
5037 | if (handlep) { | |
5038 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
5039 | *handlep = tc->tcm_handle; | |
5040 | } | |
5041 | ||
5042 | if (options) { | |
5043 | *options = ta[TCA_OPTIONS]; | |
5044 | } | |
5045 | ||
5046 | if (stats) { | |
5047 | const struct gnet_stats_queue *gsq; | |
5048 | struct gnet_stats_basic gsb; | |
5049 | ||
5050 | static const struct nl_policy stats_policy[] = { | |
5051 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
5052 | .min_len = sizeof gsb }, | |
5053 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
5054 | .min_len = sizeof *gsq }, | |
5055 | }; | |
5056 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
5057 | ||
5058 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
5059 | sa, ARRAY_SIZE(sa))) { | |
5060 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
5061 | goto error; | |
5062 | } | |
5063 | ||
5064 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
5065 | * some arch/bitsize combinations. Newer versions of Linux have a | |
5066 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
5067 | * easiest thing to do is just to make a copy. */ | |
5068 | memset(&gsb, 0, sizeof gsb); | |
5069 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
5070 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
5071 | stats->tx_bytes = gsb.bytes; | |
5072 | stats->tx_packets = gsb.packets; | |
5073 | ||
5074 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
5075 | stats->tx_errors = gsq->drops; | |
5076 | } | |
5077 | ||
5078 | return 0; | |
5079 | ||
5080 | error: | |
5081 | if (options) { | |
5082 | *options = NULL; | |
5083 | } | |
5084 | if (stats) { | |
5085 | memset(stats, 0, sizeof *stats); | |
5086 | } | |
5087 | return EPROTO; | |
5088 | } | |
5089 | ||
5090 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
5091 | * on 'netdev'. */ | |
5092 | static int | |
5093 | tc_query_class(const struct netdev *netdev, | |
5094 | unsigned int handle, unsigned int parent, | |
5095 | struct ofpbuf **replyp) | |
5096 | { | |
5097 | struct ofpbuf request; | |
5098 | struct tcmsg *tcmsg; | |
5099 | int error; | |
5100 | ||
7874bdff RD |
5101 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, |
5102 | &request); | |
23a98ffe BP |
5103 | if (!tcmsg) { |
5104 | return ENODEV; | |
5105 | } | |
c1c9c9c4 BP |
5106 | tcmsg->tcm_handle = handle; |
5107 | tcmsg->tcm_parent = parent; | |
5108 | ||
5109 | error = tc_transact(&request, replyp); | |
5110 | if (error) { | |
5111 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
5112 | netdev_get_name(netdev), | |
5113 | tc_get_major(handle), tc_get_minor(handle), | |
5114 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 5115 | ovs_strerror(error)); |
c1c9c9c4 BP |
5116 | } |
5117 | return error; | |
5118 | } | |
5119 | ||
5120 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
5121 | static int | |
5122 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
5123 | { | |
5124 | struct ofpbuf request; | |
5125 | struct tcmsg *tcmsg; | |
5126 | int error; | |
5127 | ||
7874bdff | 5128 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_DELTCLASS, 0, &request); |
23a98ffe BP |
5129 | if (!tcmsg) { |
5130 | return ENODEV; | |
5131 | } | |
c1c9c9c4 BP |
5132 | tcmsg->tcm_handle = handle; |
5133 | tcmsg->tcm_parent = 0; | |
5134 | ||
5135 | error = tc_transact(&request, NULL); | |
5136 | if (error) { | |
5137 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
5138 | netdev_get_name(netdev), | |
5139 | tc_get_major(handle), tc_get_minor(handle), | |
10a89ef0 | 5140 | ovs_strerror(error)); |
c1c9c9c4 BP |
5141 | } |
5142 | return error; | |
5143 | } | |
5144 | ||
5145 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
5146 | static int | |
b5d57fc8 | 5147 | tc_del_qdisc(struct netdev *netdev_) |
c1c9c9c4 | 5148 | { |
b5d57fc8 | 5149 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5150 | struct ofpbuf request; |
5151 | struct tcmsg *tcmsg; | |
5152 | int error; | |
5153 | ||
7874bdff | 5154 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
5155 | if (!tcmsg) { |
5156 | return ENODEV; | |
5157 | } | |
c1c9c9c4 BP |
5158 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
5159 | tcmsg->tcm_parent = TC_H_ROOT; | |
5160 | ||
5161 | error = tc_transact(&request, NULL); | |
5162 | if (error == EINVAL) { | |
5163 | /* EINVAL probably means that the default qdisc was in use, in which | |
5164 | * case we've accomplished our purpose. */ | |
5165 | error = 0; | |
5166 | } | |
b5d57fc8 BP |
5167 | if (!error && netdev->tc) { |
5168 | if (netdev->tc->ops->tc_destroy) { | |
5169 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 | 5170 | } |
b5d57fc8 | 5171 | netdev->tc = NULL; |
c1c9c9c4 BP |
5172 | } |
5173 | return error; | |
5174 | } | |
5175 | ||
ac3e3aaa BP |
5176 | static bool |
5177 | getqdisc_is_safe(void) | |
5178 | { | |
5179 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
5180 | static bool safe = false; | |
5181 | ||
5182 | if (ovsthread_once_start(&once)) { | |
5183 | struct utsname utsname; | |
5184 | int major, minor; | |
5185 | ||
5186 | if (uname(&utsname) == -1) { | |
5187 | VLOG_WARN("uname failed (%s)", ovs_strerror(errno)); | |
5188 | } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) { | |
5189 | VLOG_WARN("uname reported bad OS release (%s)", utsname.release); | |
5190 | } else if (major < 2 || (major == 2 && minor < 35)) { | |
5191 | VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s", | |
5192 | utsname.release); | |
5193 | } else { | |
5194 | safe = true; | |
5195 | } | |
5196 | ovsthread_once_done(&once); | |
5197 | } | |
5198 | return safe; | |
5199 | } | |
5200 | ||
c1c9c9c4 BP |
5201 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the |
5202 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
5203 | * positive errno value. */ | |
5204 | static int | |
b5d57fc8 | 5205 | tc_query_qdisc(const struct netdev *netdev_) |
c1c9c9c4 | 5206 | { |
b5d57fc8 | 5207 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5208 | struct ofpbuf request, *qdisc; |
5209 | const struct tc_ops *ops; | |
5210 | struct tcmsg *tcmsg; | |
5211 | int load_error; | |
5212 | int error; | |
5213 | ||
b5d57fc8 | 5214 | if (netdev->tc) { |
c1c9c9c4 BP |
5215 | return 0; |
5216 | } | |
5217 | ||
5218 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
5219 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
5220 | * 2.6.35 without that fix backported to it. | |
5221 | * | |
5222 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
5223 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
5224 | * few others. There are a few ways that I can see to do this, but most of | |
5225 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
5226 | * technique chosen here is to assume that any non-default qdisc that we | |
5227 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
5228 | * a class with handle 0:0. | |
5229 | * | |
ac3e3aaa BP |
5230 | * On Linux 2.6.35+ we use the straightforward method because it allows us |
5231 | * to handle non-builtin qdiscs without handle 1:0 (e.g. codel). However, | |
5232 | * in such a case we get no response at all from the kernel (!) if a | |
5233 | * builtin qdisc is in use (which is later caught by "!error && | |
5234 | * !qdisc->size"). */ | |
7874bdff RD |
5235 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, |
5236 | &request); | |
23a98ffe BP |
5237 | if (!tcmsg) { |
5238 | return ENODEV; | |
5239 | } | |
ac3e3aaa BP |
5240 | tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0); |
5241 | tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0; | |
c1c9c9c4 BP |
5242 | |
5243 | /* Figure out what tc class to instantiate. */ | |
5244 | error = tc_transact(&request, &qdisc); | |
ac3e3aaa | 5245 | if (!error && qdisc->size) { |
c1c9c9c4 BP |
5246 | const char *kind; |
5247 | ||
5248 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
5249 | if (error) { | |
5250 | ops = &tc_ops_other; | |
5251 | } else { | |
5252 | ops = tc_lookup_linux_name(kind); | |
5253 | if (!ops) { | |
5254 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
ac3e3aaa | 5255 | VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind); |
c1c9c9c4 BP |
5256 | |
5257 | ops = &tc_ops_other; | |
5258 | } | |
5259 | } | |
ac3e3aaa BP |
5260 | } else if ((!error && !qdisc->size) || error == ENOENT) { |
5261 | /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc | |
5262 | * set up by some other entity that doesn't have a handle 1:0. We will | |
5263 | * assume that it's the system default qdisc. */ | |
c1c9c9c4 BP |
5264 | ops = &tc_ops_default; |
5265 | error = 0; | |
5266 | } else { | |
5267 | /* Who knows? Maybe the device got deleted. */ | |
5268 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
10a89ef0 | 5269 | netdev_get_name(netdev_), ovs_strerror(error)); |
c1c9c9c4 BP |
5270 | ops = &tc_ops_other; |
5271 | } | |
5272 | ||
5273 | /* Instantiate it. */ | |
b5d57fc8 BP |
5274 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); |
5275 | ovs_assert((load_error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
5276 | ofpbuf_delete(qdisc); |
5277 | ||
5278 | return error ? error : load_error; | |
5279 | } | |
5280 | ||
5281 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
5282 | approximate the time to transmit packets of various lengths. For an MTU of | |
5283 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
5284 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
5285 | possible lengths; and so on. | |
5286 | ||
5287 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
5288 | need to be shifted right to fit within such a 256-entry table. */ | |
5289 | static int | |
5290 | tc_calc_cell_log(unsigned int mtu) | |
5291 | { | |
5292 | int cell_log; | |
5293 | ||
5294 | if (!mtu) { | |
5295 | mtu = ETH_PAYLOAD_MAX; | |
5296 | } | |
5297 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
5298 | ||
5299 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
5300 | mtu >>= 1; | |
5301 | } | |
5302 | ||
5303 | return cell_log; | |
5304 | } | |
5305 | ||
5306 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
5307 | * of 'mtu'. */ | |
5308 | static void | |
5309 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
5310 | { | |
5311 | memset(rate, 0, sizeof *rate); | |
5312 | rate->cell_log = tc_calc_cell_log(mtu); | |
5313 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
5314 | /* rate->cell_align = 0; */ /* distro headers. */ | |
5315 | rate->mpu = ETH_TOTAL_MIN; | |
5316 | rate->rate = Bps; | |
5317 | } | |
5318 | ||
5319 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
5320 | * attribute of the specified "type". | |
5321 | * | |
5322 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
5323 | static void | |
5324 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
5325 | { | |
5326 | uint32_t *rtab; | |
5327 | unsigned int i; | |
5328 | ||
5329 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
5330 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
5331 | unsigned packet_size = (i + 1) << rate->cell_log; | |
5332 | if (packet_size < rate->mpu) { | |
5333 | packet_size = rate->mpu; | |
5334 | } | |
5335 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
5336 | } | |
5337 | } | |
5338 | ||
5339 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
5340 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
5341 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 5342 | * 0 is fine.) */ |
c1c9c9c4 BP |
5343 | static int |
5344 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
5345 | { | |
5346 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
5347 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
5348 | } | |
d3980822 | 5349 | \f |
aaf2fb1a BP |
5350 | /* Linux-only functions declared in netdev-linux.h */ |
5351 | ||
5352 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If | |
5353 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
5354 | int | |
5355 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
5356 | const char *flag_name, bool enable) | |
5357 | { | |
5358 | const char *netdev_name = netdev_get_name(netdev); | |
5359 | struct ethtool_value evalue; | |
5360 | uint32_t new_flags; | |
5361 | int error; | |
5362 | ||
ab985a77 | 5363 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5364 | memset(&evalue, 0, sizeof evalue); |
5365 | error = netdev_linux_do_ethtool(netdev_name, | |
5366 | (struct ethtool_cmd *)&evalue, | |
5367 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5368 | if (error) { | |
5369 | return error; | |
5370 | } | |
5371 | ||
ab985a77 | 5372 | COVERAGE_INC(netdev_set_ethtool); |
ad2dade5 AS |
5373 | new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
5374 | if (new_flags == evalue.data) { | |
5375 | return 0; | |
5376 | } | |
5377 | evalue.data = new_flags; | |
aaf2fb1a BP |
5378 | error = netdev_linux_do_ethtool(netdev_name, |
5379 | (struct ethtool_cmd *)&evalue, | |
5380 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
5381 | if (error) { | |
5382 | return error; | |
5383 | } | |
5384 | ||
ab985a77 | 5385 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5386 | memset(&evalue, 0, sizeof evalue); |
5387 | error = netdev_linux_do_ethtool(netdev_name, | |
5388 | (struct ethtool_cmd *)&evalue, | |
5389 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5390 | if (error) { | |
5391 | return error; | |
5392 | } | |
5393 | ||
5394 | if (new_flags != evalue.data) { | |
5395 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
5396 | "device %s failed", enable ? "enable" : "disable", | |
5397 | flag_name, netdev_name); | |
5398 | return EOPNOTSUPP; | |
5399 | } | |
5400 | ||
5401 | return 0; | |
5402 | } | |
5403 | \f | |
5404 | /* Utility functions. */ | |
5405 | ||
d3980822 | 5406 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 5407 | static void |
d3980822 BP |
5408 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
5409 | const struct rtnl_link_stats *src) | |
5410 | { | |
f613a0d7 PS |
5411 | dst->rx_packets = src->rx_packets; |
5412 | dst->tx_packets = src->tx_packets; | |
5413 | dst->rx_bytes = src->rx_bytes; | |
5414 | dst->tx_bytes = src->tx_bytes; | |
5415 | dst->rx_errors = src->rx_errors; | |
5416 | dst->tx_errors = src->tx_errors; | |
5417 | dst->rx_dropped = src->rx_dropped; | |
5418 | dst->tx_dropped = src->tx_dropped; | |
5419 | dst->multicast = src->multicast; | |
5420 | dst->collisions = src->collisions; | |
5421 | dst->rx_length_errors = src->rx_length_errors; | |
5422 | dst->rx_over_errors = src->rx_over_errors; | |
5423 | dst->rx_crc_errors = src->rx_crc_errors; | |
5424 | dst->rx_frame_errors = src->rx_frame_errors; | |
5425 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5426 | dst->rx_missed_errors = src->rx_missed_errors; | |
5427 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5428 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5429 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5430 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5431 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
5432 | } |
5433 | ||
337c9b99 BP |
5434 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
5435 | static void | |
5436 | netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst, | |
5437 | const struct rtnl_link_stats64 *src) | |
5438 | { | |
5439 | dst->rx_packets = src->rx_packets; | |
5440 | dst->tx_packets = src->tx_packets; | |
5441 | dst->rx_bytes = src->rx_bytes; | |
5442 | dst->tx_bytes = src->tx_bytes; | |
5443 | dst->rx_errors = src->rx_errors; | |
5444 | dst->tx_errors = src->tx_errors; | |
5445 | dst->rx_dropped = src->rx_dropped; | |
5446 | dst->tx_dropped = src->tx_dropped; | |
5447 | dst->multicast = src->multicast; | |
5448 | dst->collisions = src->collisions; | |
5449 | dst->rx_length_errors = src->rx_length_errors; | |
5450 | dst->rx_over_errors = src->rx_over_errors; | |
5451 | dst->rx_crc_errors = src->rx_crc_errors; | |
5452 | dst->rx_frame_errors = src->rx_frame_errors; | |
5453 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5454 | dst->rx_missed_errors = src->rx_missed_errors; | |
5455 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5456 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5457 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5458 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5459 | dst->tx_window_errors = src->tx_window_errors; | |
5460 | } | |
5461 | ||
c1c9c9c4 | 5462 | static int |
35eef899 | 5463 | get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) |
c1c9c9c4 | 5464 | { |
c1c9c9c4 BP |
5465 | struct ofpbuf request; |
5466 | struct ofpbuf *reply; | |
c1c9c9c4 BP |
5467 | int error; |
5468 | ||
d6e3feb5 | 5469 | /* Filtering all counters by default */ |
5470 | memset(stats, 0xFF, sizeof(struct netdev_stats)); | |
5471 | ||
c1c9c9c4 | 5472 | ofpbuf_init(&request, 0); |
13a24df8 BP |
5473 | nl_msg_put_nlmsghdr(&request, |
5474 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
5475 | RTM_GETLINK, NLM_F_REQUEST); | |
5476 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
5477 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); | |
a88b4e04 | 5478 | error = nl_transact(NETLINK_ROUTE, &request, &reply); |
c1c9c9c4 BP |
5479 | ofpbuf_uninit(&request); |
5480 | if (error) { | |
5481 | return error; | |
5482 | } | |
5483 | ||
13a24df8 | 5484 | if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { |
337c9b99 BP |
5485 | const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64); |
5486 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) { | |
5487 | netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a)); | |
13a24df8 BP |
5488 | error = 0; |
5489 | } else { | |
71f21279 | 5490 | a = nl_attr_find(reply, 0, IFLA_STATS); |
337c9b99 BP |
5491 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { |
5492 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); | |
5493 | error = 0; | |
5494 | } else { | |
5495 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
5496 | error = EPROTO; | |
5497 | } | |
13a24df8 BP |
5498 | } |
5499 | } else { | |
5500 | VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); | |
5501 | error = EPROTO; | |
c1c9c9c4 | 5502 | } |
8b61709d | 5503 | |
8b61709d | 5504 | |
576e26d7 | 5505 | ofpbuf_delete(reply); |
35eef899 | 5506 | return error; |
8b61709d | 5507 | } |
c1c9c9c4 | 5508 | |
3a183124 | 5509 | static int |
b5d57fc8 | 5510 | get_flags(const struct netdev *dev, unsigned int *flags) |
8b61709d BP |
5511 | { |
5512 | struct ifreq ifr; | |
5513 | int error; | |
5514 | ||
755be9ea | 5515 | *flags = 0; |
259e0b1a | 5516 | error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); |
755be9ea EJ |
5517 | if (!error) { |
5518 | *flags = ifr.ifr_flags; | |
5519 | } | |
8b61709d BP |
5520 | return error; |
5521 | } | |
5522 | ||
5523 | static int | |
4b609110 | 5524 | set_flags(const char *name, unsigned int flags) |
8b61709d BP |
5525 | { |
5526 | struct ifreq ifr; | |
5527 | ||
5528 | ifr.ifr_flags = flags; | |
259e0b1a | 5529 | return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); |
8b61709d BP |
5530 | } |
5531 | ||
01b25786 PB |
5532 | int |
5533 | linux_get_ifindex(const char *netdev_name) | |
8b61709d BP |
5534 | { |
5535 | struct ifreq ifr; | |
259e0b1a | 5536 | int error; |
8b61709d | 5537 | |
71d7c22f | 5538 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5539 | COVERAGE_INC(netdev_get_ifindex); |
259e0b1a BP |
5540 | |
5541 | error = af_inet_ioctl(SIOCGIFINDEX, &ifr); | |
5542 | if (error) { | |
580e1152 RD |
5543 | /* ENODEV probably means that a vif disappeared asynchronously and |
5544 | * hasn't been removed from the database yet, so reduce the log level | |
5545 | * to INFO for that case. */ | |
5546 | VLOG_RL(&rl, error == ENODEV ? VLL_INFO : VLL_ERR, | |
5547 | "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
5548 | netdev_name, ovs_strerror(error)); | |
259e0b1a | 5549 | return -error; |
8b61709d BP |
5550 | } |
5551 | return ifr.ifr_ifindex; | |
5552 | } | |
5553 | ||
5554 | static int | |
5555 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
5556 | { | |
b5d57fc8 | 5557 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c7b1b0a5 | 5558 | |
b5d57fc8 | 5559 | if (!(netdev->cache_valid & VALID_IFINDEX)) { |
756819dd FL |
5560 | netdev_linux_update_via_netlink(netdev); |
5561 | } | |
5562 | ||
5563 | if (!(netdev->cache_valid & VALID_IFINDEX)) { | |
5564 | /* Fall back to ioctl if netlink fails */ | |
01b25786 | 5565 | int ifindex = linux_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 5566 | |
8b61709d | 5567 | if (ifindex < 0) { |
b5d57fc8 BP |
5568 | netdev->get_ifindex_error = -ifindex; |
5569 | netdev->ifindex = 0; | |
c7b1b0a5 | 5570 | } else { |
b5d57fc8 BP |
5571 | netdev->get_ifindex_error = 0; |
5572 | netdev->ifindex = ifindex; | |
8b61709d | 5573 | } |
b5d57fc8 | 5574 | netdev->cache_valid |= VALID_IFINDEX; |
8b61709d | 5575 | } |
c7b1b0a5 | 5576 | |
b5d57fc8 BP |
5577 | *ifindexp = netdev->ifindex; |
5578 | return netdev->get_ifindex_error; | |
8b61709d BP |
5579 | } |
5580 | ||
5581 | static int | |
756819dd FL |
5582 | netdev_linux_update_via_netlink(struct netdev_linux *netdev) |
5583 | { | |
5584 | struct ofpbuf request; | |
5585 | struct ofpbuf *reply; | |
5586 | struct rtnetlink_change chg; | |
5587 | struct rtnetlink_change *change = &chg; | |
5588 | int error; | |
5589 | ||
5590 | ofpbuf_init(&request, 0); | |
5591 | nl_msg_put_nlmsghdr(&request, | |
5592 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
5593 | RTM_GETLINK, NLM_F_REQUEST); | |
5594 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
5595 | ||
5596 | /* The correct identifiers for a Linux device are netnsid and ifindex, | |
5597 | * but ifindex changes as the port is moved to another network namespace | |
5598 | * and the interface name statically stored in ovsdb. */ | |
5599 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(&netdev->up)); | |
5600 | if (netdev_linux_netnsid_is_remote(netdev)) { | |
5601 | nl_msg_push_u32(&request, IFLA_IF_NETNSID, netdev->netnsid); | |
5602 | } | |
5603 | error = nl_transact(NETLINK_ROUTE, &request, &reply); | |
5604 | ofpbuf_uninit(&request); | |
5605 | if (error) { | |
5606 | ofpbuf_delete(reply); | |
5607 | return error; | |
5608 | } | |
5609 | ||
5610 | if (rtnetlink_parse(reply, change) | |
5611 | && change->nlmsg_type == RTM_NEWLINK) { | |
5612 | bool changed = false; | |
5613 | error = 0; | |
5614 | ||
5615 | /* Update netdev from rtnl msg and increment its seq if needed. */ | |
5616 | if ((change->ifi_flags ^ netdev->ifi_flags) & IFF_RUNNING) { | |
5617 | netdev->carrier_resets++; | |
5618 | changed = true; | |
5619 | } | |
5620 | if (change->ifi_flags != netdev->ifi_flags) { | |
5621 | netdev->ifi_flags = change->ifi_flags; | |
5622 | changed = true; | |
5623 | } | |
5624 | if (change->mtu && change->mtu != netdev->mtu) { | |
5625 | netdev->mtu = change->mtu; | |
5626 | netdev->cache_valid |= VALID_MTU; | |
5627 | netdev->netdev_mtu_error = 0; | |
5628 | changed = true; | |
5629 | } | |
5630 | if (!eth_addr_is_zero(change->mac) | |
5631 | && !eth_addr_equals(change->mac, netdev->etheraddr)) { | |
5632 | netdev->etheraddr = change->mac; | |
5633 | netdev->cache_valid |= VALID_ETHERADDR; | |
5634 | netdev->ether_addr_error = 0; | |
5635 | changed = true; | |
5636 | } | |
5637 | if (change->if_index != netdev->ifindex) { | |
5638 | netdev->ifindex = change->if_index; | |
5639 | netdev->cache_valid |= VALID_IFINDEX; | |
5640 | netdev->get_ifindex_error = 0; | |
5641 | changed = true; | |
5642 | } | |
5643 | if (changed) { | |
5644 | netdev_change_seq_changed(&netdev->up); | |
5645 | } | |
5646 | } else { | |
5647 | error = EINVAL; | |
5648 | } | |
5649 | ||
5650 | ofpbuf_delete(reply); | |
5651 | return error; | |
5652 | } | |
5653 | ||
5654 | static int | |
74ff3298 | 5655 | get_etheraddr(const char *netdev_name, struct eth_addr *ea) |
8b61709d BP |
5656 | { |
5657 | struct ifreq ifr; | |
5658 | int hwaddr_family; | |
259e0b1a | 5659 | int error; |
8b61709d BP |
5660 | |
5661 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5662 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5663 | COVERAGE_INC(netdev_get_hwaddr); |
259e0b1a BP |
5664 | error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); |
5665 | if (error) { | |
78857dfb BP |
5666 | /* ENODEV probably means that a vif disappeared asynchronously and |
5667 | * hasn't been removed from the database yet, so reduce the log level | |
5668 | * to INFO for that case. */ | |
259e0b1a | 5669 | VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, |
78857dfb | 5670 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", |
259e0b1a BP |
5671 | netdev_name, ovs_strerror(error)); |
5672 | return error; | |
8b61709d BP |
5673 | } |
5674 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
2482b0b0 JS |
5675 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER && |
5676 | hwaddr_family != ARPHRD_NONE) { | |
c9697f35 | 5677 | VLOG_INFO("%s device has unknown hardware address family %d", |
8b61709d | 5678 | netdev_name, hwaddr_family); |
c9697f35 | 5679 | return EINVAL; |
8b61709d BP |
5680 | } |
5681 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
5682 | return 0; | |
5683 | } | |
5684 | ||
5685 | static int | |
74ff3298 | 5686 | set_etheraddr(const char *netdev_name, const struct eth_addr mac) |
8b61709d BP |
5687 | { |
5688 | struct ifreq ifr; | |
259e0b1a | 5689 | int error; |
8b61709d BP |
5690 | |
5691 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5692 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 5693 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
74ff3298 | 5694 | memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN); |
8b61709d | 5695 | COVERAGE_INC(netdev_set_hwaddr); |
259e0b1a BP |
5696 | error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); |
5697 | if (error) { | |
8b61709d | 5698 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", |
259e0b1a | 5699 | netdev_name, ovs_strerror(error)); |
8b61709d | 5700 | } |
259e0b1a | 5701 | return error; |
8b61709d BP |
5702 | } |
5703 | ||
5704 | static int | |
0b0544d7 | 5705 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
5706 | int cmd, const char *cmd_name) |
5707 | { | |
5708 | struct ifreq ifr; | |
259e0b1a | 5709 | int error; |
8b61709d BP |
5710 | |
5711 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5712 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
5713 | ifr.ifr_data = (caddr_t) ecmd; |
5714 | ||
5715 | ecmd->cmd = cmd; | |
259e0b1a BP |
5716 | error = af_inet_ioctl(SIOCETHTOOL, &ifr); |
5717 | if (error) { | |
5718 | if (error != EOPNOTSUPP) { | |
8b61709d | 5719 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " |
259e0b1a | 5720 | "failed: %s", cmd_name, name, ovs_strerror(error)); |
8b61709d BP |
5721 | } else { |
5722 | /* The device doesn't support this operation. That's pretty | |
5723 | * common, so there's no point in logging anything. */ | |
5724 | } | |
8b61709d | 5725 | } |
259e0b1a | 5726 | return error; |
8b61709d | 5727 | } |
f1acd62b | 5728 | |
488d734d BP |
5729 | /* Returns an AF_PACKET raw socket or a negative errno value. */ |
5730 | static int | |
5731 | af_packet_sock(void) | |
5732 | { | |
23882115 BP |
5733 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
5734 | static int sock; | |
488d734d | 5735 | |
23882115 | 5736 | if (ovsthread_once_start(&once)) { |
488d734d BP |
5737 | sock = socket(AF_PACKET, SOCK_RAW, 0); |
5738 | if (sock >= 0) { | |
8450059e BP |
5739 | int error = set_nonblocking(sock); |
5740 | if (error) { | |
5741 | close(sock); | |
5742 | sock = -error; | |
5743 | } | |
488d734d BP |
5744 | } else { |
5745 | sock = -errno; | |
10a89ef0 BP |
5746 | VLOG_ERR("failed to create packet socket: %s", |
5747 | ovs_strerror(errno)); | |
488d734d | 5748 | } |
23882115 | 5749 | ovsthread_once_done(&once); |
488d734d BP |
5750 | } |
5751 | ||
5752 | return sock; | |
5753 | } |