]>
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 */ | |
3d9c99ab JH |
512 | |
513 | /* LAG information. */ | |
514 | bool is_lag_master; /* True if the netdev is a LAG master. */ | |
8b61709d BP |
515 | }; |
516 | ||
f7791740 PS |
517 | struct netdev_rxq_linux { |
518 | struct netdev_rxq up; | |
796223f5 | 519 | bool is_tap; |
5b7448ed | 520 | int fd; |
149f577a | 521 | }; |
8b61709d | 522 | |
8b61709d BP |
523 | /* This is set pretty low because we probably won't learn anything from the |
524 | * additional log messages. */ | |
525 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
526 | ||
19c8e9c1 JS |
527 | /* Polling miimon status for all ports causes performance degradation when |
528 | * handling a large number of ports. If there are no devices using miimon, then | |
812c272c JR |
529 | * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). |
530 | * | |
531 | * Readers do not depend on this variable synchronizing with the related | |
532 | * changes in the device miimon status, so we can use atomic_count. */ | |
533 | static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0); | |
19c8e9c1 | 534 | |
1c33f0c3 | 535 | static void netdev_linux_run(const struct netdev_class *); |
6f643e49 | 536 | |
0b0544d7 | 537 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 538 | int cmd, const char *cmd_name); |
b5d57fc8 | 539 | static int get_flags(const struct netdev *, unsigned int *flags); |
4b609110 | 540 | static int set_flags(const char *, unsigned int flags); |
4f9f3f21 BP |
541 | static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
542 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
543 | OVS_REQUIRES(netdev->mutex); | |
8b61709d BP |
544 | static int get_ifindex(const struct netdev *, int *ifindexp); |
545 | static int do_set_addr(struct netdev *netdev, | |
546 | int ioctl_nr, const char *ioctl_name, | |
547 | struct in_addr addr); | |
74ff3298 JR |
548 | static int get_etheraddr(const char *netdev_name, struct eth_addr *ea); |
549 | static int set_etheraddr(const char *netdev_name, const struct eth_addr); | |
35eef899 | 550 | static int get_stats_via_netlink(const struct netdev *, struct netdev_stats *); |
488d734d | 551 | static int af_packet_sock(void); |
19c8e9c1 | 552 | static bool netdev_linux_miimon_enabled(void); |
1670c579 EJ |
553 | static void netdev_linux_miimon_run(void); |
554 | static void netdev_linux_miimon_wait(void); | |
df1e5a3b | 555 | static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); |
8b61709d | 556 | |
15b3596a JG |
557 | static bool |
558 | is_netdev_linux_class(const struct netdev_class *netdev_class) | |
559 | { | |
259e0b1a | 560 | return netdev_class->run == netdev_linux_run; |
15b3596a JG |
561 | } |
562 | ||
796223f5 BP |
563 | static bool |
564 | is_tap_netdev(const struct netdev *netdev) | |
565 | { | |
b5d57fc8 | 566 | return netdev_get_class(netdev) == &netdev_tap_class; |
6c88d577 JP |
567 | } |
568 | ||
8b61709d BP |
569 | static struct netdev_linux * |
570 | netdev_linux_cast(const struct netdev *netdev) | |
571 | { | |
b5d57fc8 | 572 | ovs_assert(is_netdev_linux_class(netdev_get_class(netdev))); |
15b3596a | 573 | |
180c6d0b | 574 | return CONTAINER_OF(netdev, struct netdev_linux, up); |
8b61709d | 575 | } |
796223f5 | 576 | |
f7791740 PS |
577 | static struct netdev_rxq_linux * |
578 | netdev_rxq_linux_cast(const struct netdev_rxq *rx) | |
796223f5 | 579 | { |
9dc63482 | 580 | ovs_assert(is_netdev_linux_class(netdev_get_class(rx->netdev))); |
f7791740 | 581 | return CONTAINER_OF(rx, struct netdev_rxq_linux, up); |
796223f5 | 582 | } |
ff4ed3c9 | 583 | \f |
bfda5239 FL |
584 | static int |
585 | netdev_linux_netnsid_update__(struct netdev_linux *netdev) | |
586 | { | |
587 | struct dpif_netlink_vport reply; | |
588 | struct ofpbuf *buf; | |
589 | int error; | |
590 | ||
591 | error = dpif_netlink_vport_get(netdev_get_name(&netdev->up), &reply, &buf); | |
592 | if (error) { | |
629e1476 FL |
593 | if (error == ENOENT) { |
594 | /* Assume it is local if there is no API (e.g. if the openvswitch | |
595 | * kernel module is not loaded). */ | |
596 | netnsid_set_local(&netdev->netnsid); | |
597 | } else { | |
598 | netnsid_unset(&netdev->netnsid); | |
599 | } | |
bfda5239 FL |
600 | return error; |
601 | } | |
602 | ||
603 | netnsid_set(&netdev->netnsid, reply.netnsid); | |
604 | ofpbuf_delete(buf); | |
605 | return 0; | |
606 | } | |
607 | ||
608 | static int | |
609 | netdev_linux_netnsid_update(struct netdev_linux *netdev) | |
610 | { | |
611 | if (netnsid_is_unset(netdev->netnsid)) { | |
3dbcbfe4 FL |
612 | if (netdev_get_class(&netdev->up) == &netdev_tap_class) { |
613 | netnsid_set_local(&netdev->netnsid); | |
614 | } else { | |
615 | return netdev_linux_netnsid_update__(netdev); | |
616 | } | |
bfda5239 FL |
617 | } |
618 | ||
619 | return 0; | |
620 | } | |
621 | ||
622 | static bool | |
623 | netdev_linux_netnsid_is_eq(struct netdev_linux *netdev, int nsid) | |
624 | { | |
625 | netdev_linux_netnsid_update(netdev); | |
626 | return netnsid_eq(netdev->netnsid, nsid); | |
627 | } | |
628 | ||
756819dd FL |
629 | static bool |
630 | netdev_linux_netnsid_is_remote(struct netdev_linux *netdev) | |
631 | { | |
632 | netdev_linux_netnsid_update(netdev); | |
633 | return netnsid_is_remote(netdev->netnsid); | |
634 | } | |
635 | ||
636 | static int netdev_linux_update_via_netlink(struct netdev_linux *); | |
bfda5239 | 637 | static void netdev_linux_update(struct netdev_linux *netdev, int, |
7e9dcc0f | 638 | const struct rtnetlink_change *) |
86383816 | 639 | OVS_REQUIRES(netdev->mutex); |
cee87338 | 640 | static void netdev_linux_changed(struct netdev_linux *netdev, |
86383816 BP |
641 | unsigned int ifi_flags, unsigned int mask) |
642 | OVS_REQUIRES(netdev->mutex); | |
cee87338 | 643 | |
d6384a3a AW |
644 | /* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK, |
645 | * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL | |
cee87338 BP |
646 | * if no such socket could be created. */ |
647 | static struct nl_sock * | |
648 | netdev_linux_notify_sock(void) | |
649 | { | |
650 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
651 | static struct nl_sock *sock; | |
989d7135 PS |
652 | unsigned int mcgroups[] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR, |
653 | RTNLGRP_IPV6_IFADDR, RTNLGRP_IPV6_IFINFO}; | |
cee87338 BP |
654 | |
655 | if (ovsthread_once_start(&once)) { | |
656 | int error; | |
657 | ||
658 | error = nl_sock_create(NETLINK_ROUTE, &sock); | |
659 | if (!error) { | |
d6384a3a AW |
660 | size_t i; |
661 | ||
662 | for (i = 0; i < ARRAY_SIZE(mcgroups); i++) { | |
663 | error = nl_sock_join_mcgroup(sock, mcgroups[i]); | |
664 | if (error) { | |
665 | nl_sock_destroy(sock); | |
666 | sock = NULL; | |
667 | break; | |
668 | } | |
cee87338 BP |
669 | } |
670 | } | |
cf114a7f | 671 | nl_sock_listen_all_nsid(sock, true); |
cee87338 BP |
672 | ovsthread_once_done(&once); |
673 | } | |
674 | ||
675 | return sock; | |
676 | } | |
677 | ||
19c8e9c1 JS |
678 | static bool |
679 | netdev_linux_miimon_enabled(void) | |
680 | { | |
812c272c | 681 | return atomic_count_get(&miimon_cnt) > 0; |
19c8e9c1 JS |
682 | } |
683 | ||
3d9c99ab JH |
684 | static bool |
685 | netdev_linux_kind_is_lag(const char *kind) | |
686 | { | |
687 | if (!strcmp(kind, "bond") || !strcmp(kind, "team")) { | |
688 | return true; | |
689 | } | |
690 | ||
691 | return false; | |
692 | } | |
693 | ||
8b61709d | 694 | static void |
1c33f0c3 | 695 | netdev_linux_run(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 696 | { |
cee87338 BP |
697 | struct nl_sock *sock; |
698 | int error; | |
699 | ||
19c8e9c1 JS |
700 | if (netdev_linux_miimon_enabled()) { |
701 | netdev_linux_miimon_run(); | |
702 | } | |
cee87338 BP |
703 | |
704 | sock = netdev_linux_notify_sock(); | |
705 | if (!sock) { | |
706 | return; | |
707 | } | |
708 | ||
709 | do { | |
cee87338 | 710 | uint64_t buf_stub[4096 / 8]; |
bfda5239 | 711 | int nsid; |
cee87338 BP |
712 | struct ofpbuf buf; |
713 | ||
714 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
bfda5239 | 715 | error = nl_sock_recv(sock, &buf, &nsid, false); |
cee87338 | 716 | if (!error) { |
7e9dcc0f | 717 | struct rtnetlink_change change; |
cee87338 | 718 | |
7e9dcc0f | 719 | if (rtnetlink_parse(&buf, &change)) { |
989d7135 PS |
720 | struct netdev *netdev_ = NULL; |
721 | char dev_name[IFNAMSIZ]; | |
722 | ||
723 | if (!change.ifname) { | |
724 | change.ifname = if_indextoname(change.if_index, dev_name); | |
725 | } | |
726 | ||
727 | if (change.ifname) { | |
728 | netdev_ = netdev_from_name(change.ifname); | |
729 | } | |
cee87338 BP |
730 | if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { |
731 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
86383816 BP |
732 | |
733 | ovs_mutex_lock(&netdev->mutex); | |
bfda5239 | 734 | netdev_linux_update(netdev, nsid, &change); |
86383816 | 735 | ovs_mutex_unlock(&netdev->mutex); |
cee87338 | 736 | } |
38e0065b | 737 | netdev_close(netdev_); |
cee87338 BP |
738 | } |
739 | } else if (error == ENOBUFS) { | |
740 | struct shash device_shash; | |
741 | struct shash_node *node; | |
742 | ||
743 | nl_sock_drain(sock); | |
744 | ||
745 | shash_init(&device_shash); | |
746 | netdev_get_devices(&netdev_linux_class, &device_shash); | |
747 | SHASH_FOR_EACH (node, &device_shash) { | |
748 | struct netdev *netdev_ = node->data; | |
749 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
750 | unsigned int flags; | |
751 | ||
86383816 | 752 | ovs_mutex_lock(&netdev->mutex); |
cee87338 BP |
753 | get_flags(netdev_, &flags); |
754 | netdev_linux_changed(netdev, flags, 0); | |
86383816 BP |
755 | ovs_mutex_unlock(&netdev->mutex); |
756 | ||
cee87338 BP |
757 | netdev_close(netdev_); |
758 | } | |
759 | shash_destroy(&device_shash); | |
760 | } else if (error != EAGAIN) { | |
7ed58d4a JP |
761 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5); |
762 | VLOG_WARN_RL(&rll, "error reading or parsing netlink (%s)", | |
cee87338 BP |
763 | ovs_strerror(error)); |
764 | } | |
765 | ofpbuf_uninit(&buf); | |
766 | } while (!error); | |
8b61709d BP |
767 | } |
768 | ||
769 | static void | |
1c33f0c3 | 770 | netdev_linux_wait(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 771 | { |
cee87338 BP |
772 | struct nl_sock *sock; |
773 | ||
19c8e9c1 JS |
774 | if (netdev_linux_miimon_enabled()) { |
775 | netdev_linux_miimon_wait(); | |
776 | } | |
cee87338 BP |
777 | sock = netdev_linux_notify_sock(); |
778 | if (sock) { | |
779 | nl_sock_wait(sock, POLLIN); | |
780 | } | |
8b61709d BP |
781 | } |
782 | ||
ac4d3bcb | 783 | static void |
b5d57fc8 BP |
784 | netdev_linux_changed(struct netdev_linux *dev, |
785 | unsigned int ifi_flags, unsigned int mask) | |
86383816 | 786 | OVS_REQUIRES(dev->mutex) |
ac4d3bcb | 787 | { |
3e912ffc | 788 | netdev_change_seq_changed(&dev->up); |
8aa77183 BP |
789 | |
790 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
791 | dev->carrier_resets++; | |
792 | } | |
793 | dev->ifi_flags = ifi_flags; | |
794 | ||
4f925bd3 | 795 | dev->cache_valid &= mask; |
6b6e1329 | 796 | if (!(mask & VALID_IN)) { |
a8704b50 PS |
797 | netdev_get_addrs_list_flush(); |
798 | } | |
4f925bd3 PS |
799 | } |
800 | ||
801 | static void | |
bfda5239 FL |
802 | netdev_linux_update__(struct netdev_linux *dev, |
803 | const struct rtnetlink_change *change) | |
86383816 | 804 | OVS_REQUIRES(dev->mutex) |
4f925bd3 | 805 | { |
bfda5239 | 806 | if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)) { |
d6384a3a | 807 | if (change->nlmsg_type == RTM_NEWLINK) { |
6b6e1329 | 808 | /* Keep drv-info, and ip addresses. */ |
d6384a3a | 809 | netdev_linux_changed(dev, change->ifi_flags, |
6b6e1329 | 810 | VALID_DRVINFO | VALID_IN); |
d6384a3a AW |
811 | |
812 | /* Update netdev from rtnl-change msg. */ | |
813 | if (change->mtu) { | |
814 | dev->mtu = change->mtu; | |
815 | dev->cache_valid |= VALID_MTU; | |
816 | dev->netdev_mtu_error = 0; | |
817 | } | |
90a6637d | 818 | |
74ff3298 JR |
819 | if (!eth_addr_is_zero(change->mac)) { |
820 | dev->etheraddr = change->mac; | |
d6384a3a AW |
821 | dev->cache_valid |= VALID_ETHERADDR; |
822 | dev->ether_addr_error = 0; | |
e8e1a409 TZ |
823 | |
824 | /* The mac addr has been changed, report it now. */ | |
825 | rtnetlink_report_link(); | |
d6384a3a | 826 | } |
44445cac | 827 | |
3d9c99ab JH |
828 | if (change->master && netdev_linux_kind_is_lag(change->master)) { |
829 | dev->is_lag_master = true; | |
830 | } | |
831 | ||
d6384a3a AW |
832 | dev->ifindex = change->if_index; |
833 | dev->cache_valid |= VALID_IFINDEX; | |
834 | dev->get_ifindex_error = 0; | |
22dcb534 | 835 | dev->present = true; |
d6384a3a | 836 | } else { |
bfda5239 | 837 | /* FIXME */ |
d6384a3a | 838 | netdev_linux_changed(dev, change->ifi_flags, 0); |
22dcb534 | 839 | dev->present = false; |
bfda5239 | 840 | netnsid_unset(&dev->netnsid); |
d6384a3a AW |
841 | } |
842 | } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) { | |
843 | /* Invalidates in4, in6. */ | |
6b6e1329 | 844 | netdev_linux_changed(dev, dev->ifi_flags, ~VALID_IN); |
4f925bd3 | 845 | } else { |
d6384a3a | 846 | OVS_NOT_REACHED(); |
4f925bd3 | 847 | } |
ac4d3bcb EJ |
848 | } |
849 | ||
bfda5239 FL |
850 | static void |
851 | netdev_linux_update(struct netdev_linux *dev, int nsid, | |
852 | const struct rtnetlink_change *change) | |
853 | OVS_REQUIRES(dev->mutex) | |
854 | { | |
855 | if (netdev_linux_netnsid_is_eq(dev, nsid)) { | |
856 | netdev_linux_update__(dev, change); | |
857 | } | |
858 | } | |
859 | ||
9dc63482 BP |
860 | static struct netdev * |
861 | netdev_linux_alloc(void) | |
862 | { | |
863 | struct netdev_linux *netdev = xzalloc(sizeof *netdev); | |
864 | return &netdev->up; | |
865 | } | |
866 | ||
48c6733c WT |
867 | static int |
868 | netdev_linux_common_construct(struct netdev *netdev_) | |
9dc63482 | 869 | { |
48c6733c WT |
870 | /* Prevent any attempt to create (or open) a network device named "default" |
871 | * or "all". These device names are effectively reserved on Linux because | |
872 | * /proc/sys/net/ipv4/conf/ always contains directories by these names. By | |
873 | * itself this wouldn't call for any special treatment, but in practice if | |
874 | * a program tries to create devices with these names, it causes the kernel | |
875 | * to fire a "new device" notification event even though creation failed, | |
876 | * and in turn that causes OVS to wake up and try to create them again, | |
877 | * which ends up as a 100% CPU loop. */ | |
878 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
879 | const char *name = netdev_->name; | |
880 | if (!strcmp(name, "default") || !strcmp(name, "all")) { | |
7ed58d4a JP |
881 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 1); |
882 | VLOG_WARN_RL(&rll, "%s: Linux forbids network device with this name", | |
48c6733c WT |
883 | name); |
884 | return EINVAL; | |
885 | } | |
886 | ||
bfda5239 FL |
887 | /* The device could be in the same network namespace or in another one. */ |
888 | netnsid_unset(&netdev->netnsid); | |
834d6caf | 889 | ovs_mutex_init(&netdev->mutex); |
48c6733c | 890 | return 0; |
9dc63482 BP |
891 | } |
892 | ||
1f6e0fbd BP |
893 | /* Creates system and internal devices. */ |
894 | static int | |
9dc63482 | 895 | netdev_linux_construct(struct netdev *netdev_) |
1f6e0fbd | 896 | { |
9dc63482 | 897 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
48c6733c WT |
898 | int error = netdev_linux_common_construct(netdev_); |
899 | if (error) { | |
900 | return error; | |
901 | } | |
1f6e0fbd | 902 | |
b5d57fc8 BP |
903 | error = get_flags(&netdev->up, &netdev->ifi_flags); |
904 | if (error == ENODEV) { | |
9dc63482 | 905 | if (netdev->up.netdev_class != &netdev_internal_class) { |
b5d57fc8 | 906 | /* The device does not exist, so don't allow it to be opened. */ |
b5d57fc8 BP |
907 | return ENODEV; |
908 | } else { | |
909 | /* "Internal" netdevs have to be created as netdev objects before | |
910 | * they exist in the kernel, because creating them in the kernel | |
911 | * happens by passing a netdev object to dpif_port_add(). | |
912 | * Therefore, ignore the error. */ | |
913 | } | |
914 | } | |
46415c90 | 915 | |
a740f0de JG |
916 | return 0; |
917 | } | |
918 | ||
5b7448ed JG |
919 | /* For most types of netdevs we open the device for each call of |
920 | * netdev_open(). However, this is not the case with tap devices, | |
921 | * since it is only possible to open the device once. In this | |
922 | * situation we share a single file descriptor, and consequently | |
923 | * buffers, across all readers. Therefore once data is read it will | |
924 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 925 | static int |
9dc63482 | 926 | netdev_linux_construct_tap(struct netdev *netdev_) |
a740f0de | 927 | { |
9dc63482 | 928 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a740f0de | 929 | static const char tap_dev[] = "/dev/net/tun"; |
9dc63482 | 930 | const char *name = netdev_->name; |
a740f0de | 931 | struct ifreq ifr; |
a740f0de | 932 | |
48c6733c WT |
933 | int error = netdev_linux_common_construct(netdev_); |
934 | if (error) { | |
935 | return error; | |
936 | } | |
1f6e0fbd | 937 | |
6c88d577 | 938 | /* Open tap device. */ |
d0d08f8a BP |
939 | netdev->tap_fd = open(tap_dev, O_RDWR); |
940 | if (netdev->tap_fd < 0) { | |
6c88d577 | 941 | error = errno; |
10a89ef0 | 942 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); |
cee87338 | 943 | return error; |
6c88d577 JP |
944 | } |
945 | ||
946 | /* Create tap device. */ | |
61b9d078 | 947 | get_flags(&netdev->up, &netdev->ifi_flags); |
6c88d577 | 948 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; |
71d7c22f | 949 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
d0d08f8a | 950 | if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 | 951 | VLOG_WARN("%s: creating tap device failed: %s", name, |
10a89ef0 | 952 | ovs_strerror(errno)); |
6c88d577 | 953 | error = errno; |
f61d8d29 | 954 | goto error_close; |
6c88d577 JP |
955 | } |
956 | ||
957 | /* Make non-blocking. */ | |
d0d08f8a | 958 | error = set_nonblocking(netdev->tap_fd); |
a740f0de | 959 | if (error) { |
f61d8d29 | 960 | goto error_close; |
a740f0de JG |
961 | } |
962 | ||
0f28164b FL |
963 | if (ioctl(netdev->tap_fd, TUNSETPERSIST, 1)) { |
964 | VLOG_WARN("%s: creating tap device failed (persist): %s", name, | |
965 | ovs_strerror(errno)); | |
966 | error = errno; | |
967 | goto error_close; | |
968 | } | |
969 | ||
19aac14a | 970 | netdev->present = true; |
a740f0de JG |
971 | return 0; |
972 | ||
f61d8d29 | 973 | error_close: |
d0d08f8a | 974 | close(netdev->tap_fd); |
a740f0de JG |
975 | return error; |
976 | } | |
977 | ||
6c88d577 | 978 | static void |
9dc63482 | 979 | netdev_linux_destruct(struct netdev *netdev_) |
6c88d577 | 980 | { |
b5d57fc8 | 981 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
6c88d577 | 982 | |
b5d57fc8 BP |
983 | if (netdev->tc && netdev->tc->ops->tc_destroy) { |
984 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 BP |
985 | } |
986 | ||
d0d08f8a BP |
987 | if (netdev_get_class(netdev_) == &netdev_tap_class |
988 | && netdev->tap_fd >= 0) | |
989 | { | |
0f28164b | 990 | ioctl(netdev->tap_fd, TUNSETPERSIST, 0); |
d0d08f8a | 991 | close(netdev->tap_fd); |
6c88d577 | 992 | } |
86383816 | 993 | |
19c8e9c1 | 994 | if (netdev->miimon_interval > 0) { |
812c272c | 995 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
996 | } |
997 | ||
86383816 | 998 | ovs_mutex_destroy(&netdev->mutex); |
6c88d577 JP |
999 | } |
1000 | ||
9dc63482 BP |
1001 | static void |
1002 | netdev_linux_dealloc(struct netdev *netdev_) | |
1003 | { | |
1004 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1005 | free(netdev); | |
1006 | } | |
1007 | ||
f7791740 PS |
1008 | static struct netdev_rxq * |
1009 | netdev_linux_rxq_alloc(void) | |
9dc63482 | 1010 | { |
f7791740 | 1011 | struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); |
9dc63482 BP |
1012 | return &rx->up; |
1013 | } | |
1014 | ||
7b6b0ef4 | 1015 | static int |
f7791740 | 1016 | netdev_linux_rxq_construct(struct netdev_rxq *rxq_) |
7b6b0ef4 | 1017 | { |
f7791740 | 1018 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 1019 | struct netdev *netdev_ = rx->up.netdev; |
7b6b0ef4 | 1020 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7b6b0ef4 | 1021 | int error; |
7b6b0ef4 | 1022 | |
86383816 | 1023 | ovs_mutex_lock(&netdev->mutex); |
9dc63482 BP |
1024 | rx->is_tap = is_tap_netdev(netdev_); |
1025 | if (rx->is_tap) { | |
1026 | rx->fd = netdev->tap_fd; | |
796223f5 BP |
1027 | } else { |
1028 | struct sockaddr_ll sll; | |
b73c8518 | 1029 | int ifindex, val; |
32383c3b | 1030 | /* Result of tcpdump -dd inbound */ |
259e0b1a | 1031 | static const struct sock_filter filt[] = { |
32383c3b MM |
1032 | { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ |
1033 | { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ | |
1034 | { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ | |
1035 | { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ | |
1036 | }; | |
259e0b1a BP |
1037 | static const struct sock_fprog fprog = { |
1038 | ARRAY_SIZE(filt), (struct sock_filter *) filt | |
1039 | }; | |
7b6b0ef4 | 1040 | |
796223f5 | 1041 | /* Create file descriptor. */ |
9dc63482 BP |
1042 | rx->fd = socket(PF_PACKET, SOCK_RAW, 0); |
1043 | if (rx->fd < 0) { | |
796223f5 | 1044 | error = errno; |
10a89ef0 | 1045 | VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); |
796223f5 BP |
1046 | goto error; |
1047 | } | |
33d82a56 | 1048 | |
b73c8518 SH |
1049 | val = 1; |
1050 | if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { | |
1051 | error = errno; | |
1052 | VLOG_ERR("%s: failed to mark socket for auxdata (%s)", | |
1053 | netdev_get_name(netdev_), ovs_strerror(error)); | |
1054 | goto error; | |
1055 | } | |
1056 | ||
796223f5 | 1057 | /* Set non-blocking mode. */ |
9dc63482 | 1058 | error = set_nonblocking(rx->fd); |
796223f5 BP |
1059 | if (error) { |
1060 | goto error; | |
1061 | } | |
7b6b0ef4 | 1062 | |
796223f5 | 1063 | /* Get ethernet device index. */ |
180c6d0b | 1064 | error = get_ifindex(&netdev->up, &ifindex); |
796223f5 BP |
1065 | if (error) { |
1066 | goto error; | |
1067 | } | |
7b6b0ef4 | 1068 | |
796223f5 BP |
1069 | /* Bind to specific ethernet device. */ |
1070 | memset(&sll, 0, sizeof sll); | |
1071 | sll.sll_family = AF_PACKET; | |
1072 | sll.sll_ifindex = ifindex; | |
b73c8518 | 1073 | sll.sll_protocol = htons(ETH_P_ALL); |
9dc63482 | 1074 | if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { |
796223f5 BP |
1075 | error = errno; |
1076 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
10a89ef0 | 1077 | netdev_get_name(netdev_), ovs_strerror(error)); |
796223f5 BP |
1078 | goto error; |
1079 | } | |
32383c3b MM |
1080 | |
1081 | /* Filter for only inbound packets. */ | |
9dc63482 | 1082 | error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, |
32383c3b MM |
1083 | sizeof fprog); |
1084 | if (error) { | |
1085 | error = errno; | |
259e0b1a | 1086 | VLOG_ERR("%s: failed to attach filter (%s)", |
10a89ef0 | 1087 | netdev_get_name(netdev_), ovs_strerror(error)); |
32383c3b MM |
1088 | goto error; |
1089 | } | |
7b6b0ef4 | 1090 | } |
86383816 | 1091 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 | 1092 | |
7b6b0ef4 BP |
1093 | return 0; |
1094 | ||
1095 | error: | |
9dc63482 BP |
1096 | if (rx->fd >= 0) { |
1097 | close(rx->fd); | |
7b6b0ef4 | 1098 | } |
86383816 | 1099 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 BP |
1100 | return error; |
1101 | } | |
1102 | ||
796223f5 | 1103 | static void |
f7791740 | 1104 | netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) |
8b61709d | 1105 | { |
f7791740 | 1106 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
8b61709d | 1107 | |
796223f5 BP |
1108 | if (!rx->is_tap) { |
1109 | close(rx->fd); | |
8b61709d | 1110 | } |
9dc63482 BP |
1111 | } |
1112 | ||
1113 | static void | |
f7791740 | 1114 | netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) |
9dc63482 | 1115 | { |
f7791740 | 1116 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 1117 | |
796223f5 BP |
1118 | free(rx); |
1119 | } | |
8b61709d | 1120 | |
b73c8518 | 1121 | static ovs_be16 |
1ebdc7eb | 1122 | auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux, bool double_tagged) |
b73c8518 SH |
1123 | { |
1124 | if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { | |
1125 | return htons(aux->tp_vlan_tpid); | |
1ebdc7eb EG |
1126 | } else if (double_tagged) { |
1127 | return htons(ETH_TYPE_VLAN_8021AD); | |
b73c8518 | 1128 | } else { |
1ebdc7eb | 1129 | return htons(ETH_TYPE_VLAN_8021Q); |
b73c8518 SH |
1130 | } |
1131 | } | |
1132 | ||
1133 | static bool | |
1134 | auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) | |
1135 | { | |
1136 | return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; | |
1137 | } | |
1138 | ||
796223f5 | 1139 | static int |
cf62fa4c | 1140 | netdev_linux_rxq_recv_sock(int fd, struct dp_packet *buffer) |
796223f5 | 1141 | { |
b73c8518 | 1142 | size_t size; |
796223f5 | 1143 | ssize_t retval; |
b73c8518 SH |
1144 | struct iovec iov; |
1145 | struct cmsghdr *cmsg; | |
1146 | union { | |
1147 | struct cmsghdr cmsg; | |
1148 | char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; | |
1149 | } cmsg_buffer; | |
1150 | struct msghdr msgh; | |
1151 | ||
1152 | /* Reserve headroom for a single VLAN tag */ | |
cf62fa4c PS |
1153 | dp_packet_reserve(buffer, VLAN_HEADER_LEN); |
1154 | size = dp_packet_tailroom(buffer); | |
b73c8518 | 1155 | |
cf62fa4c | 1156 | iov.iov_base = dp_packet_data(buffer); |
b73c8518 SH |
1157 | iov.iov_len = size; |
1158 | msgh.msg_name = NULL; | |
1159 | msgh.msg_namelen = 0; | |
1160 | msgh.msg_iov = &iov; | |
1161 | msgh.msg_iovlen = 1; | |
1162 | msgh.msg_control = &cmsg_buffer; | |
1163 | msgh.msg_controllen = sizeof cmsg_buffer; | |
1164 | msgh.msg_flags = 0; | |
8e8cddf7 | 1165 | |
796223f5 | 1166 | do { |
b73c8518 | 1167 | retval = recvmsg(fd, &msgh, MSG_TRUNC); |
796223f5 BP |
1168 | } while (retval < 0 && errno == EINTR); |
1169 | ||
bfd3367b | 1170 | if (retval < 0) { |
b73c8518 SH |
1171 | return errno; |
1172 | } else if (retval > size) { | |
1173 | return EMSGSIZE; | |
1174 | } | |
1175 | ||
cf62fa4c | 1176 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1177 | |
1178 | for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg; cmsg = CMSG_NXTHDR(&msgh, cmsg)) { | |
1179 | const struct tpacket_auxdata *aux; | |
1180 | ||
1181 | if (cmsg->cmsg_level != SOL_PACKET | |
1182 | || cmsg->cmsg_type != PACKET_AUXDATA | |
1183 | || cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata))) { | |
1184 | continue; | |
8b61709d | 1185 | } |
b73c8518 SH |
1186 | |
1187 | aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); | |
1188 | if (auxdata_has_vlan_tci(aux)) { | |
1ebdc7eb EG |
1189 | struct eth_header *eth; |
1190 | bool double_tagged; | |
1191 | ||
b73c8518 SH |
1192 | if (retval < ETH_HEADER_LEN) { |
1193 | return EINVAL; | |
1194 | } | |
1195 | ||
1ebdc7eb EG |
1196 | eth = dp_packet_data(buffer); |
1197 | double_tagged = eth->eth_type == htons(ETH_TYPE_VLAN_8021Q); | |
1198 | ||
1199 | eth_push_vlan(buffer, auxdata_to_vlan_tpid(aux, double_tagged), | |
b73c8518 SH |
1200 | htons(aux->tp_vlan_tci)); |
1201 | break; | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | return 0; | |
1206 | } | |
1207 | ||
1208 | static int | |
cf62fa4c | 1209 | netdev_linux_rxq_recv_tap(int fd, struct dp_packet *buffer) |
b73c8518 SH |
1210 | { |
1211 | ssize_t retval; | |
cf62fa4c | 1212 | size_t size = dp_packet_tailroom(buffer); |
b73c8518 SH |
1213 | |
1214 | do { | |
cf62fa4c | 1215 | retval = read(fd, dp_packet_data(buffer), size); |
b73c8518 SH |
1216 | } while (retval < 0 && errno == EINTR); |
1217 | ||
1218 | if (retval < 0) { | |
bfd3367b | 1219 | return errno; |
8b61709d | 1220 | } |
b73c8518 | 1221 | |
cf62fa4c | 1222 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); |
b73c8518 SH |
1223 | return 0; |
1224 | } | |
1225 | ||
1226 | static int | |
8492adc2 JS |
1227 | netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch, |
1228 | int *qfill) | |
b73c8518 | 1229 | { |
f7791740 | 1230 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
df1e5a3b | 1231 | struct netdev *netdev = rx->up.netdev; |
cf62fa4c | 1232 | struct dp_packet *buffer; |
df1e5a3b PS |
1233 | ssize_t retval; |
1234 | int mtu; | |
1235 | ||
1236 | if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
1237 | mtu = ETH_PAYLOAD_MAX; | |
1238 | } | |
1239 | ||
2482b0b0 | 1240 | /* Assume Ethernet port. No need to set packet_type. */ |
cf62fa4c | 1241 | buffer = dp_packet_new_with_headroom(VLAN_ETH_HEADER_LEN + mtu, |
91088554 | 1242 | DP_NETDEV_HEADROOM); |
b73c8518 | 1243 | retval = (rx->is_tap |
f7791740 PS |
1244 | ? netdev_linux_rxq_recv_tap(rx->fd, buffer) |
1245 | : netdev_linux_rxq_recv_sock(rx->fd, buffer)); | |
df1e5a3b PS |
1246 | |
1247 | if (retval) { | |
1248 | if (retval != EAGAIN && retval != EMSGSIZE) { | |
1249 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
7c6401ca | 1250 | netdev_rxq_get_name(rxq_), ovs_strerror(errno)); |
df1e5a3b | 1251 | } |
cf62fa4c | 1252 | dp_packet_delete(buffer); |
df1e5a3b | 1253 | } else { |
72c84bc2 | 1254 | dp_packet_batch_init_packet(batch, buffer); |
b73c8518 SH |
1255 | } |
1256 | ||
8492adc2 JS |
1257 | if (qfill) { |
1258 | *qfill = -ENOTSUP; | |
1259 | } | |
1260 | ||
b73c8518 | 1261 | return retval; |
8b61709d BP |
1262 | } |
1263 | ||
8b61709d | 1264 | static void |
f7791740 | 1265 | netdev_linux_rxq_wait(struct netdev_rxq *rxq_) |
8b61709d | 1266 | { |
f7791740 | 1267 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1268 | poll_fd_wait(rx->fd, POLLIN); |
8b61709d BP |
1269 | } |
1270 | ||
8b61709d | 1271 | static int |
f7791740 | 1272 | netdev_linux_rxq_drain(struct netdev_rxq *rxq_) |
8b61709d | 1273 | { |
f7791740 | 1274 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1275 | if (rx->is_tap) { |
8b61709d | 1276 | struct ifreq ifr; |
f7791740 | 1277 | int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, |
259e0b1a | 1278 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
8b61709d BP |
1279 | if (error) { |
1280 | return error; | |
1281 | } | |
796223f5 | 1282 | drain_fd(rx->fd, ifr.ifr_qlen); |
8b61709d BP |
1283 | return 0; |
1284 | } else { | |
796223f5 | 1285 | return drain_rcvbuf(rx->fd); |
8b61709d BP |
1286 | } |
1287 | } | |
1288 | ||
d19cf8bb ZG |
1289 | static int |
1290 | netdev_linux_sock_batch_send(int sock, int ifindex, | |
1291 | struct dp_packet_batch *batch) | |
1292 | { | |
e0a00cee | 1293 | const size_t size = dp_packet_batch_size(batch); |
d19cf8bb ZG |
1294 | /* We don't bother setting most fields in sockaddr_ll because the |
1295 | * kernel ignores them for SOCK_RAW. */ | |
1296 | struct sockaddr_ll sll = { .sll_family = AF_PACKET, | |
1297 | .sll_ifindex = ifindex }; | |
1298 | ||
e0a00cee BB |
1299 | struct mmsghdr *mmsg = xmalloc(sizeof(*mmsg) * size); |
1300 | struct iovec *iov = xmalloc(sizeof(*iov) * size); | |
d19cf8bb | 1301 | |
e0a00cee | 1302 | struct dp_packet *packet; |
e883448e | 1303 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
d19cf8bb | 1304 | iov[i].iov_base = dp_packet_data(packet); |
ad8b0b4f | 1305 | iov[i].iov_len = dp_packet_size(packet); |
d19cf8bb ZG |
1306 | mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll, |
1307 | .msg_namelen = sizeof sll, | |
1308 | .msg_iov = &iov[i], | |
1309 | .msg_iovlen = 1 }; | |
1310 | } | |
1311 | ||
1312 | int error = 0; | |
e0a00cee | 1313 | for (uint32_t ofs = 0; ofs < size; ) { |
d19cf8bb ZG |
1314 | ssize_t retval; |
1315 | do { | |
e0a00cee | 1316 | retval = sendmmsg(sock, mmsg + ofs, size - ofs, 0); |
d19cf8bb ZG |
1317 | error = retval < 0 ? errno : 0; |
1318 | } while (error == EINTR); | |
1319 | if (error) { | |
1320 | break; | |
1321 | } | |
1322 | ofs += retval; | |
1323 | } | |
1324 | ||
1325 | free(mmsg); | |
1326 | free(iov); | |
1327 | return error; | |
1328 | } | |
1329 | ||
1330 | /* Use the tap fd to send 'batch' to tap device 'netdev'. Using the tap fd is | |
1331 | * essential, because packets sent to a tap device with an AF_PACKET socket | |
1332 | * will loop back to be *received* again on the tap device. This doesn't occur | |
1333 | * on other interface types because we attach a socket filter to the rx | |
1334 | * socket. */ | |
1335 | static int | |
1336 | netdev_linux_tap_batch_send(struct netdev *netdev_, | |
1337 | struct dp_packet_batch *batch) | |
1338 | { | |
1339 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
13708b21 | 1340 | struct dp_packet *packet; |
22dcb534 FL |
1341 | |
1342 | /* The Linux tap driver returns EIO if the device is not up, | |
1343 | * so if the device is not up, don't waste time sending it. | |
1344 | * However, if the device is in another network namespace | |
1345 | * then OVS can't retrieve the state. In that case, send the | |
1346 | * packets anyway. */ | |
1347 | if (netdev->present && !(netdev->ifi_flags & IFF_UP)) { | |
1348 | netdev->tx_dropped += dp_packet_batch_size(batch); | |
1349 | return 0; | |
1350 | } | |
1351 | ||
e883448e | 1352 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
ad8b0b4f | 1353 | size_t size = dp_packet_size(packet); |
d19cf8bb ZG |
1354 | ssize_t retval; |
1355 | int error; | |
1356 | ||
1357 | do { | |
1358 | retval = write(netdev->tap_fd, dp_packet_data(packet), size); | |
1359 | error = retval < 0 ? errno : 0; | |
1360 | } while (error == EINTR); | |
1361 | ||
1362 | if (error) { | |
1363 | /* The Linux tap driver returns EIO if the device is not up. From | |
1364 | * the OVS side this is not an error, so we ignore it; otherwise, | |
1365 | * return the erro. */ | |
1366 | if (error != EIO) { | |
1367 | return error; | |
1368 | } | |
1369 | } else if (retval != size) { | |
1370 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" " | |
1371 | "bytes of %"PRIuSIZE") on %s", | |
1372 | retval, size, netdev_get_name(netdev_)); | |
1373 | return EMSGSIZE; | |
1374 | } | |
1375 | } | |
1376 | return 0; | |
1377 | } | |
1378 | ||
1379 | /* Sends 'batch' on 'netdev'. Returns 0 if successful, otherwise a positive | |
8b61709d BP |
1380 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued |
1381 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
1382 | * the packet is too big or too small to transmit on the device. | |
1383 | * | |
8b61709d BP |
1384 | * The kernel maintains a packet transmission queue, so the caller is not |
1385 | * expected to do additional queuing of packets. */ | |
1386 | static int | |
f00fa8cb | 1387 | netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED, |
b30896c9 | 1388 | struct dp_packet_batch *batch, |
324c8374 | 1389 | bool concurrent_txq OVS_UNUSED) |
8b61709d | 1390 | { |
f4fd623c | 1391 | int error = 0; |
0a62ae2c ZG |
1392 | int sock = 0; |
1393 | ||
0a62ae2c | 1394 | if (!is_tap_netdev(netdev_)) { |
e0e2410d FL |
1395 | if (netdev_linux_netnsid_is_remote(netdev_linux_cast(netdev_))) { |
1396 | error = EOPNOTSUPP; | |
1397 | goto free_batch; | |
1398 | } | |
1399 | ||
0a62ae2c ZG |
1400 | sock = af_packet_sock(); |
1401 | if (sock < 0) { | |
1402 | error = -sock; | |
1403 | goto free_batch; | |
1404 | } | |
1405 | ||
1406 | int ifindex = netdev_get_ifindex(netdev_); | |
1407 | if (ifindex < 0) { | |
1408 | error = -ifindex; | |
1409 | goto free_batch; | |
1410 | } | |
1411 | ||
d19cf8bb ZG |
1412 | error = netdev_linux_sock_batch_send(sock, ifindex, batch); |
1413 | } else { | |
1414 | error = netdev_linux_tap_batch_send(netdev_, batch); | |
0a62ae2c | 1415 | } |
d19cf8bb ZG |
1416 | if (error) { |
1417 | if (error == ENOBUFS) { | |
1418 | /* The Linux AF_PACKET implementation never blocks waiting | |
1419 | * for room for packets, instead returning ENOBUFS. | |
1420 | * Translate this into EAGAIN for the caller. */ | |
1421 | error = EAGAIN; | |
f23347ea | 1422 | } else { |
f4fd623c DDP |
1423 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", |
1424 | netdev_get_name(netdev_), ovs_strerror(error)); | |
d19cf8bb | 1425 | } |
f4fd623c DDP |
1426 | } |
1427 | ||
0a62ae2c | 1428 | free_batch: |
b30896c9 | 1429 | dp_packet_delete_batch(batch, true); |
f4fd623c | 1430 | return error; |
8b61709d BP |
1431 | } |
1432 | ||
1433 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
1434 | * when the packet transmission queue has sufficient room to transmit a packet | |
1435 | * with netdev_send(). | |
1436 | * | |
1437 | * The kernel maintains a packet transmission queue, so the client is not | |
1438 | * expected to do additional queuing of packets. Thus, this function is | |
1439 | * unlikely to ever be used. It is included for completeness. */ | |
1440 | static void | |
f00fa8cb | 1441 | netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED) |
8b61709d | 1442 | { |
796223f5 | 1443 | if (is_tap_netdev(netdev)) { |
8b61709d BP |
1444 | /* TAP device always accepts packets.*/ |
1445 | poll_immediate_wake(); | |
1446 | } | |
1447 | } | |
1448 | ||
1449 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
1450 | * otherwise a positive errno value. */ | |
1451 | static int | |
74ff3298 | 1452 | netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac) |
8b61709d | 1453 | { |
b5d57fc8 | 1454 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4f9f3f21 | 1455 | enum netdev_flags old_flags = 0; |
eb395f2e BP |
1456 | int error; |
1457 | ||
86383816 | 1458 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1459 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1460 | error = EOPNOTSUPP; | |
1461 | goto exit; | |
1462 | } | |
86383816 | 1463 | |
b5d57fc8 | 1464 | if (netdev->cache_valid & VALID_ETHERADDR) { |
86383816 BP |
1465 | error = netdev->ether_addr_error; |
1466 | if (error || eth_addr_equals(netdev->etheraddr, mac)) { | |
1467 | goto exit; | |
44445cac | 1468 | } |
b5d57fc8 | 1469 | netdev->cache_valid &= ~VALID_ETHERADDR; |
44445cac PS |
1470 | } |
1471 | ||
7eb1bd81 | 1472 | /* Tap devices must be brought down before setting the address. */ |
796223f5 | 1473 | if (is_tap_netdev(netdev_)) { |
4f9f3f21 | 1474 | update_flags(netdev, NETDEV_UP, 0, &old_flags); |
7eb1bd81 | 1475 | } |
44445cac PS |
1476 | error = set_etheraddr(netdev_get_name(netdev_), mac); |
1477 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
1478 | netdev->ether_addr_error = error; |
1479 | netdev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 1480 | if (!error) { |
74ff3298 | 1481 | netdev->etheraddr = mac; |
eb395f2e | 1482 | } |
8b61709d | 1483 | } |
44445cac | 1484 | |
4f9f3f21 BP |
1485 | if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { |
1486 | update_flags(netdev, 0, NETDEV_UP, &old_flags); | |
1487 | } | |
7eb1bd81 | 1488 | |
86383816 BP |
1489 | exit: |
1490 | ovs_mutex_unlock(&netdev->mutex); | |
8b61709d BP |
1491 | return error; |
1492 | } | |
1493 | ||
44445cac | 1494 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d | 1495 | static int |
74ff3298 | 1496 | netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac) |
8b61709d | 1497 | { |
b5d57fc8 | 1498 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1499 | int error; |
44445cac | 1500 | |
86383816 | 1501 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1502 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { |
756819dd FL |
1503 | netdev_linux_update_via_netlink(netdev); |
1504 | } | |
1505 | ||
1506 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { | |
1507 | /* Fall back to ioctl if netlink fails */ | |
86383816 | 1508 | netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), |
74ff3298 | 1509 | &netdev->etheraddr); |
b5d57fc8 | 1510 | netdev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 1511 | } |
44445cac | 1512 | |
86383816 BP |
1513 | error = netdev->ether_addr_error; |
1514 | if (!error) { | |
74ff3298 | 1515 | *mac = netdev->etheraddr; |
44445cac | 1516 | } |
86383816 | 1517 | ovs_mutex_unlock(&netdev->mutex); |
44445cac | 1518 | |
86383816 | 1519 | return error; |
8b61709d BP |
1520 | } |
1521 | ||
8b61709d | 1522 | static int |
73371c09 | 1523 | netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) |
8b61709d | 1524 | { |
86383816 BP |
1525 | int error; |
1526 | ||
b5d57fc8 | 1527 | if (!(netdev->cache_valid & VALID_MTU)) { |
756819dd FL |
1528 | netdev_linux_update_via_netlink(netdev); |
1529 | } | |
1530 | ||
1531 | if (!(netdev->cache_valid & VALID_MTU)) { | |
1532 | /* Fall back to ioctl if netlink fails */ | |
8b61709d | 1533 | struct ifreq ifr; |
90a6637d | 1534 | |
86383816 | 1535 | netdev->netdev_mtu_error = af_inet_ifreq_ioctl( |
73371c09 | 1536 | netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); |
b5d57fc8 BP |
1537 | netdev->mtu = ifr.ifr_mtu; |
1538 | netdev->cache_valid |= VALID_MTU; | |
8b61709d | 1539 | } |
90a6637d | 1540 | |
86383816 BP |
1541 | error = netdev->netdev_mtu_error; |
1542 | if (!error) { | |
b5d57fc8 | 1543 | *mtup = netdev->mtu; |
90a6637d | 1544 | } |
73371c09 BP |
1545 | |
1546 | return error; | |
1547 | } | |
1548 | ||
1549 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1550 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1551 | * bytes for Ethernet devices. */ | |
1552 | static int | |
1553 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1554 | { | |
1555 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1556 | int error; | |
1557 | ||
1558 | ovs_mutex_lock(&netdev->mutex); | |
1559 | error = netdev_linux_get_mtu__(netdev, mtup); | |
86383816 BP |
1560 | ovs_mutex_unlock(&netdev->mutex); |
1561 | ||
1562 | return error; | |
8b61709d BP |
1563 | } |
1564 | ||
9b020780 PS |
1565 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1566 | * networking ioctl interface. | |
1567 | */ | |
1568 | static int | |
4124cb12 | 1569 | netdev_linux_set_mtu(struct netdev *netdev_, int mtu) |
9b020780 | 1570 | { |
b5d57fc8 | 1571 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9b020780 PS |
1572 | struct ifreq ifr; |
1573 | int error; | |
1574 | ||
86383816 | 1575 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1576 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1577 | error = EOPNOTSUPP; | |
1578 | goto exit; | |
1579 | } | |
1580 | ||
b5d57fc8 | 1581 | if (netdev->cache_valid & VALID_MTU) { |
86383816 BP |
1582 | error = netdev->netdev_mtu_error; |
1583 | if (error || netdev->mtu == mtu) { | |
1584 | goto exit; | |
90a6637d | 1585 | } |
b5d57fc8 | 1586 | netdev->cache_valid &= ~VALID_MTU; |
153e5481 | 1587 | } |
9b020780 | 1588 | ifr.ifr_mtu = mtu; |
259e0b1a BP |
1589 | error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, |
1590 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d | 1591 | if (!error || error == ENODEV) { |
b5d57fc8 BP |
1592 | netdev->netdev_mtu_error = error; |
1593 | netdev->mtu = ifr.ifr_mtu; | |
1594 | netdev->cache_valid |= VALID_MTU; | |
9b020780 | 1595 | } |
86383816 BP |
1596 | exit: |
1597 | ovs_mutex_unlock(&netdev->mutex); | |
90a6637d | 1598 | return error; |
9b020780 PS |
1599 | } |
1600 | ||
9ab3d9a3 BP |
1601 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1602 | * On failure, returns a negative errno value. */ | |
1603 | static int | |
86383816 | 1604 | netdev_linux_get_ifindex(const struct netdev *netdev_) |
9ab3d9a3 | 1605 | { |
86383816 | 1606 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9ab3d9a3 BP |
1607 | int ifindex, error; |
1608 | ||
86383816 | 1609 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1610 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1611 | error = EOPNOTSUPP; | |
1612 | goto exit; | |
1613 | } | |
86383816 | 1614 | error = get_ifindex(netdev_, &ifindex); |
86383816 | 1615 | |
e0e2410d FL |
1616 | exit: |
1617 | ovs_mutex_unlock(&netdev->mutex); | |
9ab3d9a3 BP |
1618 | return error ? -error : ifindex; |
1619 | } | |
1620 | ||
8b61709d BP |
1621 | static int |
1622 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1623 | { | |
b5d57fc8 | 1624 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1625 | |
86383816 | 1626 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
1627 | if (netdev->miimon_interval > 0) { |
1628 | *carrier = netdev->miimon; | |
3a183124 | 1629 | } else { |
b5d57fc8 | 1630 | *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1631 | } |
86383816 | 1632 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d | 1633 | |
3a183124 | 1634 | return 0; |
8b61709d BP |
1635 | } |
1636 | ||
65c3058c | 1637 | static long long int |
86383816 | 1638 | netdev_linux_get_carrier_resets(const struct netdev *netdev_) |
65c3058c | 1639 | { |
86383816 BP |
1640 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1641 | long long int carrier_resets; | |
1642 | ||
1643 | ovs_mutex_lock(&netdev->mutex); | |
1644 | carrier_resets = netdev->carrier_resets; | |
1645 | ovs_mutex_unlock(&netdev->mutex); | |
1646 | ||
1647 | return carrier_resets; | |
65c3058c EJ |
1648 | } |
1649 | ||
63331829 | 1650 | static int |
1670c579 EJ |
1651 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1652 | struct mii_ioctl_data *data) | |
63331829 | 1653 | { |
63331829 | 1654 | struct ifreq ifr; |
782e6111 | 1655 | int error; |
63331829 | 1656 | |
63331829 | 1657 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1658 | memcpy(&ifr.ifr_data, data, sizeof *data); |
259e0b1a | 1659 | error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1660 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1661 | |
782e6111 EJ |
1662 | return error; |
1663 | } | |
1664 | ||
1665 | static int | |
1670c579 | 1666 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1667 | { |
782e6111 EJ |
1668 | struct mii_ioctl_data data; |
1669 | int error; | |
63331829 | 1670 | |
782e6111 EJ |
1671 | *miimon = false; |
1672 | ||
1673 | memset(&data, 0, sizeof data); | |
1670c579 | 1674 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
1675 | if (!error) { |
1676 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
1677 | data.reg_num = MII_BMSR; | |
1670c579 | 1678 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 1679 | &data); |
63331829 EJ |
1680 | |
1681 | if (!error) { | |
782e6111 | 1682 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 | 1683 | } |
9120cfc0 DH |
1684 | } |
1685 | if (error) { | |
63331829 | 1686 | struct ethtool_cmd ecmd; |
63331829 EJ |
1687 | |
1688 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
1689 | name); | |
1690 | ||
ab985a77 | 1691 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
1692 | memset(&ecmd, 0, sizeof ecmd); |
1693 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
1694 | "ETHTOOL_GLINK"); | |
1695 | if (!error) { | |
782e6111 EJ |
1696 | struct ethtool_value eval; |
1697 | ||
1698 | memcpy(&eval, &ecmd, sizeof eval); | |
1699 | *miimon = !!eval.data; | |
63331829 EJ |
1700 | } else { |
1701 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | return error; | |
1706 | } | |
1707 | ||
1670c579 EJ |
1708 | static int |
1709 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
1710 | long long int interval) | |
1711 | { | |
b5d57fc8 | 1712 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1670c579 | 1713 | |
86383816 | 1714 | ovs_mutex_lock(&netdev->mutex); |
1670c579 | 1715 | interval = interval > 0 ? MAX(interval, 100) : 0; |
b5d57fc8 | 1716 | if (netdev->miimon_interval != interval) { |
19c8e9c1 | 1717 | if (interval && !netdev->miimon_interval) { |
812c272c | 1718 | atomic_count_inc(&miimon_cnt); |
19c8e9c1 | 1719 | } else if (!interval && netdev->miimon_interval) { |
812c272c | 1720 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
1721 | } |
1722 | ||
b5d57fc8 BP |
1723 | netdev->miimon_interval = interval; |
1724 | timer_set_expired(&netdev->miimon_timer); | |
1670c579 | 1725 | } |
86383816 | 1726 | ovs_mutex_unlock(&netdev->mutex); |
1670c579 EJ |
1727 | |
1728 | return 0; | |
1729 | } | |
1730 | ||
1731 | static void | |
1732 | netdev_linux_miimon_run(void) | |
1733 | { | |
1734 | struct shash device_shash; | |
1735 | struct shash_node *node; | |
1736 | ||
1737 | shash_init(&device_shash); | |
b5d57fc8 | 1738 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1739 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1740 | struct netdev *netdev = node->data; |
1741 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
1742 | bool miimon; |
1743 | ||
86383816 BP |
1744 | ovs_mutex_lock(&dev->mutex); |
1745 | if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { | |
1746 | netdev_linux_get_miimon(dev->up.name, &miimon); | |
1747 | if (miimon != dev->miimon) { | |
1748 | dev->miimon = miimon; | |
1749 | netdev_linux_changed(dev, dev->ifi_flags, 0); | |
1750 | } | |
1670c579 | 1751 | |
86383816 | 1752 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); |
1670c579 | 1753 | } |
86383816 | 1754 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1755 | netdev_close(netdev); |
1670c579 EJ |
1756 | } |
1757 | ||
1758 | shash_destroy(&device_shash); | |
1759 | } | |
1760 | ||
1761 | static void | |
1762 | netdev_linux_miimon_wait(void) | |
1763 | { | |
1764 | struct shash device_shash; | |
1765 | struct shash_node *node; | |
1766 | ||
1767 | shash_init(&device_shash); | |
b5d57fc8 | 1768 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 1769 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
1770 | struct netdev *netdev = node->data; |
1771 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 | 1772 | |
86383816 | 1773 | ovs_mutex_lock(&dev->mutex); |
1670c579 EJ |
1774 | if (dev->miimon_interval > 0) { |
1775 | timer_wait(&dev->miimon_timer); | |
1776 | } | |
86383816 | 1777 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 1778 | netdev_close(netdev); |
1670c579 EJ |
1779 | } |
1780 | shash_destroy(&device_shash); | |
1781 | } | |
1782 | ||
92df599c JG |
1783 | static void |
1784 | swap_uint64(uint64_t *a, uint64_t *b) | |
1785 | { | |
1de0e8ae BP |
1786 | uint64_t tmp = *a; |
1787 | *a = *b; | |
1788 | *b = tmp; | |
92df599c JG |
1789 | } |
1790 | ||
c060c4cf EJ |
1791 | /* Copies 'src' into 'dst', performing format conversion in the process. |
1792 | * | |
1793 | * 'src' is allowed to be misaligned. */ | |
1794 | static void | |
1795 | netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, | |
1796 | const struct ovs_vport_stats *src) | |
1797 | { | |
6a54dedc BP |
1798 | dst->rx_packets = get_32aligned_u64(&src->rx_packets); |
1799 | dst->tx_packets = get_32aligned_u64(&src->tx_packets); | |
1800 | dst->rx_bytes = get_32aligned_u64(&src->rx_bytes); | |
1801 | dst->tx_bytes = get_32aligned_u64(&src->tx_bytes); | |
1802 | dst->rx_errors = get_32aligned_u64(&src->rx_errors); | |
1803 | dst->tx_errors = get_32aligned_u64(&src->tx_errors); | |
1804 | dst->rx_dropped = get_32aligned_u64(&src->rx_dropped); | |
1805 | dst->tx_dropped = get_32aligned_u64(&src->tx_dropped); | |
c060c4cf EJ |
1806 | dst->multicast = 0; |
1807 | dst->collisions = 0; | |
1808 | dst->rx_length_errors = 0; | |
1809 | dst->rx_over_errors = 0; | |
1810 | dst->rx_crc_errors = 0; | |
1811 | dst->rx_frame_errors = 0; | |
1812 | dst->rx_fifo_errors = 0; | |
1813 | dst->rx_missed_errors = 0; | |
1814 | dst->tx_aborted_errors = 0; | |
1815 | dst->tx_carrier_errors = 0; | |
1816 | dst->tx_fifo_errors = 0; | |
1817 | dst->tx_heartbeat_errors = 0; | |
1818 | dst->tx_window_errors = 0; | |
1819 | } | |
1820 | ||
1821 | static int | |
1822 | get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) | |
1823 | { | |
93451a0a | 1824 | struct dpif_netlink_vport reply; |
c060c4cf EJ |
1825 | struct ofpbuf *buf; |
1826 | int error; | |
1827 | ||
93451a0a | 1828 | error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf); |
c060c4cf EJ |
1829 | if (error) { |
1830 | return error; | |
1831 | } else if (!reply.stats) { | |
1832 | ofpbuf_delete(buf); | |
1833 | return EOPNOTSUPP; | |
1834 | } | |
1835 | ||
1836 | netdev_stats_from_ovs_vport_stats(stats, reply.stats); | |
1837 | ||
1838 | ofpbuf_delete(buf); | |
1839 | ||
1840 | return 0; | |
1841 | } | |
1842 | ||
f613a0d7 PS |
1843 | static void |
1844 | get_stats_via_vport(const struct netdev *netdev_, | |
1845 | struct netdev_stats *stats) | |
8b61709d | 1846 | { |
b5d57fc8 | 1847 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1848 | |
b5d57fc8 BP |
1849 | if (!netdev->vport_stats_error || |
1850 | !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 1851 | int error; |
7fbef77a | 1852 | |
c060c4cf | 1853 | error = get_stats_via_vport__(netdev_, stats); |
bb13fe5e | 1854 | if (error && error != ENOENT && error != ENODEV) { |
a57a8488 | 1855 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
10a89ef0 BP |
1856 | "(%s)", |
1857 | netdev_get_name(netdev_), ovs_strerror(error)); | |
f613a0d7 | 1858 | } |
b5d57fc8 BP |
1859 | netdev->vport_stats_error = error; |
1860 | netdev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 1861 | } |
f613a0d7 | 1862 | } |
8b61709d | 1863 | |
f613a0d7 PS |
1864 | /* Retrieves current device stats for 'netdev-linux'. */ |
1865 | static int | |
1866 | netdev_linux_get_stats(const struct netdev *netdev_, | |
1867 | struct netdev_stats *stats) | |
1868 | { | |
b5d57fc8 | 1869 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1870 | struct netdev_stats dev_stats; |
1871 | int error; | |
1872 | ||
86383816 | 1873 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1874 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1875 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1876 | if (error) { |
86383816 BP |
1877 | if (!netdev->vport_stats_error) { |
1878 | error = 0; | |
f613a0d7 | 1879 | } |
86383816 | 1880 | } else if (netdev->vport_stats_error) { |
04c881eb | 1881 | /* stats not available from OVS then use netdev stats. */ |
f613a0d7 PS |
1882 | *stats = dev_stats; |
1883 | } else { | |
04c881eb AZ |
1884 | /* Use kernel netdev's packet and byte counts since vport's counters |
1885 | * do not reflect packet counts on the wire when GSO, TSO or GRO are | |
1886 | * enabled. */ | |
1887 | stats->rx_packets = dev_stats.rx_packets; | |
1888 | stats->rx_bytes = dev_stats.rx_bytes; | |
1889 | stats->tx_packets = dev_stats.tx_packets; | |
1890 | stats->tx_bytes = dev_stats.tx_bytes; | |
1891 | ||
f613a0d7 PS |
1892 | stats->rx_errors += dev_stats.rx_errors; |
1893 | stats->tx_errors += dev_stats.tx_errors; | |
1894 | stats->rx_dropped += dev_stats.rx_dropped; | |
1895 | stats->tx_dropped += dev_stats.tx_dropped; | |
1896 | stats->multicast += dev_stats.multicast; | |
1897 | stats->collisions += dev_stats.collisions; | |
1898 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
1899 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
1900 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
1901 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
1902 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
1903 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
1904 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
1905 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
1906 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
1907 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
1908 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
1909 | } | |
86383816 BP |
1910 | ovs_mutex_unlock(&netdev->mutex); |
1911 | ||
1912 | return error; | |
f613a0d7 PS |
1913 | } |
1914 | ||
1915 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
1916 | * netdev-internal. */ | |
1917 | static int | |
15aee116 | 1918 | netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) |
f613a0d7 | 1919 | { |
b5d57fc8 | 1920 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
1921 | struct netdev_stats dev_stats; |
1922 | int error; | |
1923 | ||
86383816 | 1924 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 1925 | get_stats_via_vport(netdev_, stats); |
35eef899 | 1926 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 1927 | if (error) { |
86383816 BP |
1928 | if (!netdev->vport_stats_error) { |
1929 | error = 0; | |
8b61709d | 1930 | } |
86383816 BP |
1931 | } else if (netdev->vport_stats_error) { |
1932 | /* Transmit and receive stats will appear to be swapped relative to the | |
1933 | * other ports since we are the one sending the data, not a remote | |
1934 | * computer. For consistency, we swap them back here. This does not | |
1935 | * apply if we are getting stats from the vport layer because it always | |
1936 | * tracks stats from the perspective of the switch. */ | |
fe6b0e03 | 1937 | |
f613a0d7 | 1938 | *stats = dev_stats; |
92df599c JG |
1939 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
1940 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
1941 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
1942 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
1943 | stats->rx_length_errors = 0; |
1944 | stats->rx_over_errors = 0; | |
1945 | stats->rx_crc_errors = 0; | |
1946 | stats->rx_frame_errors = 0; | |
1947 | stats->rx_fifo_errors = 0; | |
1948 | stats->rx_missed_errors = 0; | |
1949 | stats->tx_aborted_errors = 0; | |
1950 | stats->tx_carrier_errors = 0; | |
1951 | stats->tx_fifo_errors = 0; | |
1952 | stats->tx_heartbeat_errors = 0; | |
1953 | stats->tx_window_errors = 0; | |
f613a0d7 | 1954 | } else { |
04c881eb AZ |
1955 | /* Use kernel netdev's packet and byte counts since vport counters |
1956 | * do not reflect packet counts on the wire when GSO, TSO or GRO | |
1957 | * are enabled. */ | |
1958 | stats->rx_packets = dev_stats.tx_packets; | |
1959 | stats->rx_bytes = dev_stats.tx_bytes; | |
1960 | stats->tx_packets = dev_stats.rx_packets; | |
1961 | stats->tx_bytes = dev_stats.rx_bytes; | |
1962 | ||
f613a0d7 PS |
1963 | stats->rx_dropped += dev_stats.tx_dropped; |
1964 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 1965 | |
f613a0d7 PS |
1966 | stats->rx_errors += dev_stats.tx_errors; |
1967 | stats->tx_errors += dev_stats.rx_errors; | |
1968 | ||
1969 | stats->multicast += dev_stats.multicast; | |
1970 | stats->collisions += dev_stats.collisions; | |
1971 | } | |
22dcb534 | 1972 | stats->tx_dropped += netdev->tx_dropped; |
86383816 BP |
1973 | ovs_mutex_unlock(&netdev->mutex); |
1974 | ||
1975 | return error; | |
8b61709d BP |
1976 | } |
1977 | ||
bba1e6f3 PS |
1978 | static int |
1979 | netdev_internal_get_stats(const struct netdev *netdev_, | |
1980 | struct netdev_stats *stats) | |
1981 | { | |
b5d57fc8 | 1982 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1983 | int error; |
bba1e6f3 | 1984 | |
86383816 | 1985 | ovs_mutex_lock(&netdev->mutex); |
bba1e6f3 | 1986 | get_stats_via_vport(netdev_, stats); |
86383816 BP |
1987 | error = netdev->vport_stats_error; |
1988 | ovs_mutex_unlock(&netdev->mutex); | |
1989 | ||
1990 | return error; | |
bba1e6f3 PS |
1991 | } |
1992 | ||
51f87458 | 1993 | static void |
b5d57fc8 | 1994 | netdev_linux_read_features(struct netdev_linux *netdev) |
8b61709d BP |
1995 | { |
1996 | struct ethtool_cmd ecmd; | |
6c038611 | 1997 | uint32_t speed; |
8b61709d BP |
1998 | int error; |
1999 | ||
b5d57fc8 | 2000 | if (netdev->cache_valid & VALID_FEATURES) { |
51f87458 PS |
2001 | return; |
2002 | } | |
2003 | ||
ab985a77 | 2004 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 2005 | memset(&ecmd, 0, sizeof ecmd); |
b5d57fc8 | 2006 | error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, |
8b61709d BP |
2007 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
2008 | if (error) { | |
51f87458 | 2009 | goto out; |
8b61709d BP |
2010 | } |
2011 | ||
2012 | /* Supported features. */ | |
b5d57fc8 | 2013 | netdev->supported = 0; |
8b61709d | 2014 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
b5d57fc8 | 2015 | netdev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
2016 | } |
2017 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
b5d57fc8 | 2018 | netdev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
2019 | } |
2020 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
b5d57fc8 | 2021 | netdev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
2022 | } |
2023 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
b5d57fc8 | 2024 | netdev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
2025 | } |
2026 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
b5d57fc8 | 2027 | netdev->supported |= NETDEV_F_1GB_HD; |
8b61709d | 2028 | } |
67bed84c SH |
2029 | if ((ecmd.supported & SUPPORTED_1000baseT_Full) || |
2030 | (ecmd.supported & SUPPORTED_1000baseKX_Full)) { | |
b5d57fc8 | 2031 | netdev->supported |= NETDEV_F_1GB_FD; |
8b61709d | 2032 | } |
67bed84c SH |
2033 | if ((ecmd.supported & SUPPORTED_10000baseT_Full) || |
2034 | (ecmd.supported & SUPPORTED_10000baseKX4_Full) || | |
2035 | (ecmd.supported & SUPPORTED_10000baseKR_Full) || | |
2036 | (ecmd.supported & SUPPORTED_10000baseR_FEC)) { | |
b5d57fc8 | 2037 | netdev->supported |= NETDEV_F_10GB_FD; |
8b61709d | 2038 | } |
67bed84c SH |
2039 | if ((ecmd.supported & SUPPORTED_40000baseKR4_Full) || |
2040 | (ecmd.supported & SUPPORTED_40000baseCR4_Full) || | |
2041 | (ecmd.supported & SUPPORTED_40000baseSR4_Full) || | |
2042 | (ecmd.supported & SUPPORTED_40000baseLR4_Full)) { | |
2043 | netdev->supported |= NETDEV_F_40GB_FD; | |
2044 | } | |
8b61709d | 2045 | if (ecmd.supported & SUPPORTED_TP) { |
b5d57fc8 | 2046 | netdev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
2047 | } |
2048 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
b5d57fc8 | 2049 | netdev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
2050 | } |
2051 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
b5d57fc8 | 2052 | netdev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
2053 | } |
2054 | if (ecmd.supported & SUPPORTED_Pause) { | |
b5d57fc8 | 2055 | netdev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
2056 | } |
2057 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
b5d57fc8 | 2058 | netdev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2059 | } |
2060 | ||
2061 | /* Advertised features. */ | |
b5d57fc8 | 2062 | netdev->advertised = 0; |
8b61709d | 2063 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
b5d57fc8 | 2064 | netdev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
2065 | } |
2066 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
b5d57fc8 | 2067 | netdev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
2068 | } |
2069 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
b5d57fc8 | 2070 | netdev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
2071 | } |
2072 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
b5d57fc8 | 2073 | netdev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
2074 | } |
2075 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
b5d57fc8 | 2076 | netdev->advertised |= NETDEV_F_1GB_HD; |
8b61709d | 2077 | } |
67bed84c SH |
2078 | if ((ecmd.advertising & ADVERTISED_1000baseT_Full) || |
2079 | (ecmd.advertising & ADVERTISED_1000baseKX_Full)) { | |
b5d57fc8 | 2080 | netdev->advertised |= NETDEV_F_1GB_FD; |
8b61709d | 2081 | } |
67bed84c SH |
2082 | if ((ecmd.advertising & ADVERTISED_10000baseT_Full) || |
2083 | (ecmd.advertising & ADVERTISED_10000baseKX4_Full) || | |
2084 | (ecmd.advertising & ADVERTISED_10000baseKR_Full) || | |
2085 | (ecmd.advertising & ADVERTISED_10000baseR_FEC)) { | |
b5d57fc8 | 2086 | netdev->advertised |= NETDEV_F_10GB_FD; |
8b61709d | 2087 | } |
67bed84c SH |
2088 | if ((ecmd.advertising & ADVERTISED_40000baseKR4_Full) || |
2089 | (ecmd.advertising & ADVERTISED_40000baseCR4_Full) || | |
2090 | (ecmd.advertising & ADVERTISED_40000baseSR4_Full) || | |
2091 | (ecmd.advertising & ADVERTISED_40000baseLR4_Full)) { | |
2092 | netdev->advertised |= NETDEV_F_40GB_FD; | |
2093 | } | |
8b61709d | 2094 | if (ecmd.advertising & ADVERTISED_TP) { |
b5d57fc8 | 2095 | netdev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
2096 | } |
2097 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
b5d57fc8 | 2098 | netdev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
2099 | } |
2100 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
b5d57fc8 | 2101 | netdev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
2102 | } |
2103 | if (ecmd.advertising & ADVERTISED_Pause) { | |
b5d57fc8 | 2104 | netdev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
2105 | } |
2106 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
b5d57fc8 | 2107 | netdev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2108 | } |
2109 | ||
2110 | /* Current settings. */ | |
0c615356 | 2111 | speed = ethtool_cmd_speed(&ecmd); |
6c038611 | 2112 | if (speed == SPEED_10) { |
b5d57fc8 | 2113 | netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 2114 | } else if (speed == SPEED_100) { |
b5d57fc8 | 2115 | netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 2116 | } else if (speed == SPEED_1000) { |
b5d57fc8 | 2117 | netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 2118 | } else if (speed == SPEED_10000) { |
b5d57fc8 | 2119 | netdev->current = NETDEV_F_10GB_FD; |
6c038611 | 2120 | } else if (speed == 40000) { |
b5d57fc8 | 2121 | netdev->current = NETDEV_F_40GB_FD; |
6c038611 | 2122 | } else if (speed == 100000) { |
b5d57fc8 | 2123 | netdev->current = NETDEV_F_100GB_FD; |
6c038611 | 2124 | } else if (speed == 1000000) { |
b5d57fc8 | 2125 | netdev->current = NETDEV_F_1TB_FD; |
8b61709d | 2126 | } else { |
b5d57fc8 | 2127 | netdev->current = 0; |
8b61709d BP |
2128 | } |
2129 | ||
2130 | if (ecmd.port == PORT_TP) { | |
b5d57fc8 | 2131 | netdev->current |= NETDEV_F_COPPER; |
8b61709d | 2132 | } else if (ecmd.port == PORT_FIBRE) { |
b5d57fc8 | 2133 | netdev->current |= NETDEV_F_FIBER; |
8b61709d BP |
2134 | } |
2135 | ||
2136 | if (ecmd.autoneg) { | |
b5d57fc8 | 2137 | netdev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
2138 | } |
2139 | ||
51f87458 | 2140 | out: |
b5d57fc8 BP |
2141 | netdev->cache_valid |= VALID_FEATURES; |
2142 | netdev->get_features_error = error; | |
51f87458 PS |
2143 | } |
2144 | ||
887ed8b2 BP |
2145 | /* Stores the features supported by 'netdev' into of '*current', '*advertised', |
2146 | * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. | |
2147 | * Returns 0 if successful, otherwise a positive errno value. */ | |
51f87458 PS |
2148 | static int |
2149 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
2150 | enum netdev_features *current, |
2151 | enum netdev_features *advertised, | |
2152 | enum netdev_features *supported, | |
2153 | enum netdev_features *peer) | |
51f87458 | 2154 | { |
b5d57fc8 | 2155 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 2156 | int error; |
51f87458 | 2157 | |
86383816 | 2158 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2159 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2160 | error = EOPNOTSUPP; | |
2161 | goto exit; | |
2162 | } | |
2163 | ||
b5d57fc8 | 2164 | netdev_linux_read_features(netdev); |
b5d57fc8 BP |
2165 | if (!netdev->get_features_error) { |
2166 | *current = netdev->current; | |
2167 | *advertised = netdev->advertised; | |
2168 | *supported = netdev->supported; | |
887ed8b2 | 2169 | *peer = 0; /* XXX */ |
51f87458 | 2170 | } |
86383816 | 2171 | error = netdev->get_features_error; |
86383816 | 2172 | |
e0e2410d FL |
2173 | exit: |
2174 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2175 | return error; |
8b61709d BP |
2176 | } |
2177 | ||
2178 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
2179 | static int | |
86383816 | 2180 | netdev_linux_set_advertisements(struct netdev *netdev_, |
6c038611 | 2181 | enum netdev_features advertise) |
8b61709d | 2182 | { |
86383816 | 2183 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2184 | struct ethtool_cmd ecmd; |
2185 | int error; | |
2186 | ||
86383816 BP |
2187 | ovs_mutex_lock(&netdev->mutex); |
2188 | ||
ab985a77 | 2189 | COVERAGE_INC(netdev_get_ethtool); |
e0e2410d FL |
2190 | |
2191 | if (netdev_linux_netnsid_is_remote(netdev)) { | |
2192 | error = EOPNOTSUPP; | |
2193 | goto exit; | |
2194 | } | |
2195 | ||
8b61709d | 2196 | memset(&ecmd, 0, sizeof ecmd); |
86383816 | 2197 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
8b61709d BP |
2198 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
2199 | if (error) { | |
86383816 | 2200 | goto exit; |
8b61709d BP |
2201 | } |
2202 | ||
2203 | ecmd.advertising = 0; | |
6c038611 | 2204 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
2205 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
2206 | } | |
6c038611 | 2207 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
2208 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
2209 | } | |
6c038611 | 2210 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
2211 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
2212 | } | |
6c038611 | 2213 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
2214 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
2215 | } | |
6c038611 | 2216 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
2217 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
2218 | } | |
6c038611 | 2219 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
2220 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
2221 | } | |
6c038611 | 2222 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
2223 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
2224 | } | |
6c038611 | 2225 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
2226 | ecmd.advertising |= ADVERTISED_TP; |
2227 | } | |
6c038611 | 2228 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
2229 | ecmd.advertising |= ADVERTISED_FIBRE; |
2230 | } | |
6c038611 | 2231 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
2232 | ecmd.advertising |= ADVERTISED_Autoneg; |
2233 | } | |
6c038611 | 2234 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
2235 | ecmd.advertising |= ADVERTISED_Pause; |
2236 | } | |
6c038611 | 2237 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
2238 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
2239 | } | |
ab985a77 | 2240 | COVERAGE_INC(netdev_set_ethtool); |
86383816 BP |
2241 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
2242 | ETHTOOL_SSET, "ETHTOOL_SSET"); | |
2243 | ||
2244 | exit: | |
2245 | ovs_mutex_unlock(&netdev->mutex); | |
2246 | return error; | |
8b61709d BP |
2247 | } |
2248 | ||
f8500004 JP |
2249 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
2250 | * successful, otherwise a positive errno value. */ | |
8b61709d | 2251 | static int |
b5d57fc8 | 2252 | netdev_linux_set_policing(struct netdev *netdev_, |
8b61709d BP |
2253 | uint32_t kbits_rate, uint32_t kbits_burst) |
2254 | { | |
b5d57fc8 BP |
2255 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
2256 | const char *netdev_name = netdev_get_name(netdev_); | |
7874bdff | 2257 | int ifindex; |
f8500004 | 2258 | int error; |
8b61709d | 2259 | |
d5ae4a60 PB |
2260 | if (netdev_is_flow_api_enabled()) { |
2261 | if (kbits_rate) { | |
2262 | VLOG_WARN_RL(&rl, "%s: policing with offload isn't supported", | |
2263 | netdev_name); | |
2264 | } | |
2265 | return EOPNOTSUPP; | |
2266 | } | |
2267 | ||
80a86fbe | 2268 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
79abacc8 | 2269 | : !kbits_burst ? 8000 /* Default to 8000 kbits if 0. */ |
80a86fbe BP |
2270 | : kbits_burst); /* Stick with user-specified value. */ |
2271 | ||
86383816 | 2272 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2273 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2274 | error = EOPNOTSUPP; | |
2275 | goto out; | |
2276 | } | |
2277 | ||
b5d57fc8 | 2278 | if (netdev->cache_valid & VALID_POLICING) { |
86383816 BP |
2279 | error = netdev->netdev_policing_error; |
2280 | if (error || (netdev->kbits_rate == kbits_rate && | |
2281 | netdev->kbits_burst == kbits_burst)) { | |
c9f71668 | 2282 | /* Assume that settings haven't changed since we last set them. */ |
86383816 | 2283 | goto out; |
c9f71668 | 2284 | } |
b5d57fc8 | 2285 | netdev->cache_valid &= ~VALID_POLICING; |
80a86fbe BP |
2286 | } |
2287 | ||
7874bdff RD |
2288 | error = get_ifindex(netdev_, &ifindex); |
2289 | if (error) { | |
2290 | goto out; | |
2291 | } | |
2292 | ||
ac8c3412 | 2293 | COVERAGE_INC(netdev_set_policing); |
f8500004 | 2294 | /* Remove any existing ingress qdisc. */ |
093c9458 | 2295 | error = tc_add_del_ingress_qdisc(ifindex, false, 0); |
f8500004 JP |
2296 | if (error) { |
2297 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
10a89ef0 | 2298 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2299 | goto out; |
f8500004 JP |
2300 | } |
2301 | ||
8b61709d | 2302 | if (kbits_rate) { |
093c9458 | 2303 | error = tc_add_del_ingress_qdisc(ifindex, true, 0); |
f8500004 JP |
2304 | if (error) { |
2305 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
10a89ef0 | 2306 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2307 | goto out; |
8b61709d BP |
2308 | } |
2309 | ||
b5d57fc8 | 2310 | error = tc_add_policer(netdev_, kbits_rate, kbits_burst); |
f8500004 JP |
2311 | if (error){ |
2312 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
10a89ef0 | 2313 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2314 | goto out; |
8b61709d | 2315 | } |
8b61709d BP |
2316 | } |
2317 | ||
b5d57fc8 BP |
2318 | netdev->kbits_rate = kbits_rate; |
2319 | netdev->kbits_burst = kbits_burst; | |
f8500004 | 2320 | |
c9f71668 PS |
2321 | out: |
2322 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
2323 | netdev->netdev_policing_error = error; |
2324 | netdev->cache_valid |= VALID_POLICING; | |
c9f71668 | 2325 | } |
86383816 | 2326 | ovs_mutex_unlock(&netdev->mutex); |
c9f71668 | 2327 | return error; |
8b61709d BP |
2328 | } |
2329 | ||
c1c9c9c4 BP |
2330 | static int |
2331 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 2332 | struct sset *types) |
c1c9c9c4 | 2333 | { |
559eb230 | 2334 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2335 | for (opsp = tcs; *opsp != NULL; opsp++) { |
2336 | const struct tc_ops *ops = *opsp; | |
2337 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 2338 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
2339 | } |
2340 | } | |
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | static const struct tc_ops * | |
2345 | tc_lookup_ovs_name(const char *name) | |
2346 | { | |
559eb230 | 2347 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2348 | |
2349 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2350 | const struct tc_ops *ops = *opsp; | |
2351 | if (!strcmp(name, ops->ovs_name)) { | |
2352 | return ops; | |
2353 | } | |
2354 | } | |
2355 | return NULL; | |
2356 | } | |
2357 | ||
2358 | static const struct tc_ops * | |
2359 | tc_lookup_linux_name(const char *name) | |
2360 | { | |
559eb230 | 2361 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2362 | |
2363 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2364 | const struct tc_ops *ops = *opsp; | |
2365 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
2366 | return ops; | |
2367 | } | |
2368 | } | |
2369 | return NULL; | |
2370 | } | |
2371 | ||
93b13be8 | 2372 | static struct tc_queue * |
b5d57fc8 | 2373 | tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, |
93b13be8 BP |
2374 | size_t hash) |
2375 | { | |
b5d57fc8 | 2376 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
93b13be8 BP |
2377 | struct tc_queue *queue; |
2378 | ||
b5d57fc8 | 2379 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { |
93b13be8 BP |
2380 | if (queue->queue_id == queue_id) { |
2381 | return queue; | |
2382 | } | |
2383 | } | |
2384 | return NULL; | |
2385 | } | |
2386 | ||
2387 | static struct tc_queue * | |
2388 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
2389 | { | |
2390 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
2391 | } | |
2392 | ||
c1c9c9c4 BP |
2393 | static int |
2394 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
2395 | const char *type, | |
2396 | struct netdev_qos_capabilities *caps) | |
2397 | { | |
2398 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
2399 | if (!ops) { | |
2400 | return EOPNOTSUPP; | |
2401 | } | |
2402 | caps->n_queues = ops->n_queues; | |
2403 | return 0; | |
2404 | } | |
2405 | ||
2406 | static int | |
b5d57fc8 | 2407 | netdev_linux_get_qos(const struct netdev *netdev_, |
79f1cbe9 | 2408 | const char **typep, struct smap *details) |
c1c9c9c4 | 2409 | { |
b5d57fc8 | 2410 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2411 | int error; |
2412 | ||
86383816 | 2413 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2414 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2415 | error = EOPNOTSUPP; | |
2416 | goto exit; | |
2417 | } | |
2418 | ||
b5d57fc8 | 2419 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2420 | if (!error) { |
2421 | *typep = netdev->tc->ops->ovs_name; | |
2422 | error = (netdev->tc->ops->qdisc_get | |
2423 | ? netdev->tc->ops->qdisc_get(netdev_, details) | |
2424 | : 0); | |
c1c9c9c4 BP |
2425 | } |
2426 | ||
e0e2410d FL |
2427 | exit: |
2428 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2429 | return error; |
c1c9c9c4 BP |
2430 | } |
2431 | ||
2432 | static int | |
b5d57fc8 | 2433 | netdev_linux_set_qos(struct netdev *netdev_, |
79f1cbe9 | 2434 | const char *type, const struct smap *details) |
c1c9c9c4 | 2435 | { |
b5d57fc8 | 2436 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2437 | const struct tc_ops *new_ops; |
2438 | int error; | |
2439 | ||
2440 | new_ops = tc_lookup_ovs_name(type); | |
2441 | if (!new_ops || !new_ops->tc_install) { | |
2442 | return EOPNOTSUPP; | |
2443 | } | |
2444 | ||
6cf888b8 BS |
2445 | if (new_ops == &tc_ops_noop) { |
2446 | return new_ops->tc_install(netdev_, details); | |
2447 | } | |
2448 | ||
86383816 | 2449 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2450 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2451 | error = EOPNOTSUPP; | |
2452 | goto exit; | |
2453 | } | |
2454 | ||
b5d57fc8 | 2455 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 | 2456 | if (error) { |
86383816 | 2457 | goto exit; |
c1c9c9c4 BP |
2458 | } |
2459 | ||
b5d57fc8 | 2460 | if (new_ops == netdev->tc->ops) { |
86383816 | 2461 | error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; |
c1c9c9c4 BP |
2462 | } else { |
2463 | /* Delete existing qdisc. */ | |
b5d57fc8 | 2464 | error = tc_del_qdisc(netdev_); |
c1c9c9c4 | 2465 | if (error) { |
86383816 | 2466 | goto exit; |
c1c9c9c4 | 2467 | } |
b5d57fc8 | 2468 | ovs_assert(netdev->tc == NULL); |
c1c9c9c4 BP |
2469 | |
2470 | /* Install new qdisc. */ | |
b5d57fc8 BP |
2471 | error = new_ops->tc_install(netdev_, details); |
2472 | ovs_assert((error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 | 2473 | } |
86383816 BP |
2474 | |
2475 | exit: | |
2476 | ovs_mutex_unlock(&netdev->mutex); | |
2477 | return error; | |
c1c9c9c4 BP |
2478 | } |
2479 | ||
2480 | static int | |
b5d57fc8 | 2481 | netdev_linux_get_queue(const struct netdev *netdev_, |
79f1cbe9 | 2482 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 | 2483 | { |
b5d57fc8 | 2484 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2485 | int error; |
2486 | ||
86383816 | 2487 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2488 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2489 | error = EOPNOTSUPP; | |
2490 | goto exit; | |
2491 | } | |
2492 | ||
b5d57fc8 | 2493 | error = tc_query_qdisc(netdev_); |
86383816 | 2494 | if (!error) { |
b5d57fc8 | 2495 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
86383816 | 2496 | error = (queue |
b5d57fc8 | 2497 | ? netdev->tc->ops->class_get(netdev_, queue, details) |
93b13be8 | 2498 | : ENOENT); |
c1c9c9c4 | 2499 | } |
86383816 | 2500 | |
e0e2410d FL |
2501 | exit: |
2502 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2503 | return error; |
c1c9c9c4 BP |
2504 | } |
2505 | ||
2506 | static int | |
b5d57fc8 | 2507 | netdev_linux_set_queue(struct netdev *netdev_, |
79f1cbe9 | 2508 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 | 2509 | { |
b5d57fc8 | 2510 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2511 | int error; |
2512 | ||
86383816 | 2513 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2514 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2515 | error = EOPNOTSUPP; | |
2516 | goto exit; | |
2517 | } | |
2518 | ||
b5d57fc8 | 2519 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2520 | if (!error) { |
2521 | error = (queue_id < netdev->tc->ops->n_queues | |
2522 | && netdev->tc->ops->class_set | |
2523 | ? netdev->tc->ops->class_set(netdev_, queue_id, details) | |
2524 | : EINVAL); | |
c1c9c9c4 BP |
2525 | } |
2526 | ||
e0e2410d FL |
2527 | exit: |
2528 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2529 | return error; |
c1c9c9c4 BP |
2530 | } |
2531 | ||
2532 | static int | |
b5d57fc8 | 2533 | netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) |
c1c9c9c4 | 2534 | { |
b5d57fc8 | 2535 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2536 | int error; |
2537 | ||
86383816 | 2538 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2539 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2540 | error = EOPNOTSUPP; | |
2541 | goto exit; | |
2542 | } | |
2543 | ||
b5d57fc8 | 2544 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2545 | if (!error) { |
2546 | if (netdev->tc->ops->class_delete) { | |
2547 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2548 | error = (queue | |
2549 | ? netdev->tc->ops->class_delete(netdev_, queue) | |
2550 | : ENOENT); | |
2551 | } else { | |
2552 | error = EINVAL; | |
2553 | } | |
c1c9c9c4 | 2554 | } |
86383816 | 2555 | |
e0e2410d FL |
2556 | exit: |
2557 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2558 | return error; |
c1c9c9c4 BP |
2559 | } |
2560 | ||
2561 | static int | |
b5d57fc8 | 2562 | netdev_linux_get_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2563 | unsigned int queue_id, |
2564 | struct netdev_queue_stats *stats) | |
2565 | { | |
b5d57fc8 | 2566 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2567 | int error; |
2568 | ||
86383816 | 2569 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2570 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2571 | error = EOPNOTSUPP; | |
2572 | goto exit; | |
2573 | } | |
2574 | ||
b5d57fc8 | 2575 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2576 | if (!error) { |
2577 | if (netdev->tc->ops->class_get_stats) { | |
2578 | const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2579 | if (queue) { | |
2580 | stats->created = queue->created; | |
2581 | error = netdev->tc->ops->class_get_stats(netdev_, queue, | |
2582 | stats); | |
2583 | } else { | |
2584 | error = ENOENT; | |
2585 | } | |
2586 | } else { | |
2587 | error = EOPNOTSUPP; | |
6dc34a0d | 2588 | } |
c1c9c9c4 | 2589 | } |
86383816 | 2590 | |
e0e2410d FL |
2591 | exit: |
2592 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2593 | return error; |
c1c9c9c4 BP |
2594 | } |
2595 | ||
d57695d7 JS |
2596 | struct queue_dump_state { |
2597 | struct nl_dump dump; | |
2598 | struct ofpbuf buf; | |
2599 | }; | |
2600 | ||
23a98ffe | 2601 | static bool |
d57695d7 | 2602 | start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) |
c1c9c9c4 BP |
2603 | { |
2604 | struct ofpbuf request; | |
2605 | struct tcmsg *tcmsg; | |
2606 | ||
7874bdff | 2607 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, 0, &request); |
23a98ffe BP |
2608 | if (!tcmsg) { |
2609 | return false; | |
2610 | } | |
3c4de644 | 2611 | tcmsg->tcm_parent = 0; |
d57695d7 | 2612 | nl_dump_start(&state->dump, NETLINK_ROUTE, &request); |
c1c9c9c4 | 2613 | ofpbuf_uninit(&request); |
d57695d7 JS |
2614 | |
2615 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); | |
23a98ffe | 2616 | return true; |
c1c9c9c4 BP |
2617 | } |
2618 | ||
d57695d7 JS |
2619 | static int |
2620 | finish_queue_dump(struct queue_dump_state *state) | |
2621 | { | |
2622 | ofpbuf_uninit(&state->buf); | |
2623 | return nl_dump_done(&state->dump); | |
2624 | } | |
2625 | ||
89454bf4 BP |
2626 | struct netdev_linux_queue_state { |
2627 | unsigned int *queues; | |
2628 | size_t cur_queue; | |
2629 | size_t n_queues; | |
2630 | }; | |
2631 | ||
c1c9c9c4 | 2632 | static int |
89454bf4 | 2633 | netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) |
c1c9c9c4 | 2634 | { |
e0e2410d | 2635 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2636 | int error; |
2637 | ||
86383816 | 2638 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2639 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2640 | error = EOPNOTSUPP; | |
2641 | goto exit; | |
2642 | } | |
2643 | ||
b5d57fc8 | 2644 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2645 | if (!error) { |
2646 | if (netdev->tc->ops->class_get) { | |
89454bf4 BP |
2647 | struct netdev_linux_queue_state *state; |
2648 | struct tc_queue *queue; | |
2649 | size_t i; | |
2650 | ||
2651 | *statep = state = xmalloc(sizeof *state); | |
2652 | state->n_queues = hmap_count(&netdev->tc->queues); | |
2653 | state->cur_queue = 0; | |
2654 | state->queues = xmalloc(state->n_queues * sizeof *state->queues); | |
2655 | ||
2656 | i = 0; | |
2657 | HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { | |
2658 | state->queues[i++] = queue->queue_id; | |
86383816 | 2659 | } |
c1c9c9c4 | 2660 | } else { |
86383816 | 2661 | error = EOPNOTSUPP; |
c1c9c9c4 BP |
2662 | } |
2663 | } | |
c1c9c9c4 | 2664 | |
e0e2410d FL |
2665 | exit: |
2666 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2667 | return error; |
c1c9c9c4 BP |
2668 | } |
2669 | ||
89454bf4 BP |
2670 | static int |
2671 | netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, | |
2672 | unsigned int *queue_idp, struct smap *details) | |
2673 | { | |
e0e2410d | 2674 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
89454bf4 BP |
2675 | struct netdev_linux_queue_state *state = state_; |
2676 | int error = EOF; | |
2677 | ||
2678 | ovs_mutex_lock(&netdev->mutex); | |
e0e2410d FL |
2679 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2680 | error = EOPNOTSUPP; | |
2681 | goto exit; | |
2682 | } | |
2683 | ||
89454bf4 BP |
2684 | while (state->cur_queue < state->n_queues) { |
2685 | unsigned int queue_id = state->queues[state->cur_queue++]; | |
2686 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2687 | ||
2688 | if (queue) { | |
2689 | *queue_idp = queue_id; | |
2690 | error = netdev->tc->ops->class_get(netdev_, queue, details); | |
2691 | break; | |
2692 | } | |
2693 | } | |
89454bf4 | 2694 | |
e0e2410d FL |
2695 | exit: |
2696 | ovs_mutex_unlock(&netdev->mutex); | |
89454bf4 BP |
2697 | return error; |
2698 | } | |
2699 | ||
2700 | static int | |
2701 | netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, | |
2702 | void *state_) | |
2703 | { | |
2704 | struct netdev_linux_queue_state *state = state_; | |
2705 | ||
2706 | free(state->queues); | |
2707 | free(state); | |
2708 | return 0; | |
2709 | } | |
2710 | ||
c1c9c9c4 | 2711 | static int |
b5d57fc8 | 2712 | netdev_linux_dump_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
2713 | netdev_dump_queue_stats_cb *cb, void *aux) |
2714 | { | |
b5d57fc8 | 2715 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2716 | int error; |
2717 | ||
86383816 | 2718 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2719 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2720 | error = EOPNOTSUPP; | |
2721 | goto exit; | |
2722 | } | |
2723 | ||
b5d57fc8 | 2724 | error = tc_query_qdisc(netdev_); |
86383816 | 2725 | if (!error) { |
d57695d7 | 2726 | struct queue_dump_state state; |
c1c9c9c4 | 2727 | |
86383816 BP |
2728 | if (!netdev->tc->ops->class_dump_stats) { |
2729 | error = EOPNOTSUPP; | |
d57695d7 | 2730 | } else if (!start_queue_dump(netdev_, &state)) { |
86383816 BP |
2731 | error = ENODEV; |
2732 | } else { | |
2733 | struct ofpbuf msg; | |
2734 | int retval; | |
2735 | ||
d57695d7 | 2736 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
86383816 BP |
2737 | retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, |
2738 | cb, aux); | |
2739 | if (retval) { | |
2740 | error = retval; | |
2741 | } | |
2742 | } | |
2743 | ||
d57695d7 | 2744 | retval = finish_queue_dump(&state); |
86383816 BP |
2745 | if (retval) { |
2746 | error = retval; | |
2747 | } | |
c1c9c9c4 BP |
2748 | } |
2749 | } | |
2750 | ||
e0e2410d FL |
2751 | exit: |
2752 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2753 | return error; |
c1c9c9c4 BP |
2754 | } |
2755 | ||
8b61709d | 2756 | static int |
f1acd62b BP |
2757 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
2758 | struct in_addr netmask) | |
8b61709d | 2759 | { |
b5d57fc8 | 2760 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2761 | int error; |
2762 | ||
86383816 | 2763 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2764 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2765 | error = EOPNOTSUPP; | |
2766 | goto exit; | |
2767 | } | |
2768 | ||
f1acd62b | 2769 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 2770 | if (!error) { |
f1acd62b | 2771 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 2772 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 2773 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
2774 | } |
2775 | } | |
49af9a3d | 2776 | |
e0e2410d | 2777 | exit: |
86383816 | 2778 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d BP |
2779 | return error; |
2780 | } | |
2781 | ||
7df6932e AW |
2782 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address. |
2783 | * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding | |
2784 | * error. */ | |
8b61709d | 2785 | static int |
a8704b50 PS |
2786 | netdev_linux_get_addr_list(const struct netdev *netdev_, |
2787 | struct in6_addr **addr, struct in6_addr **mask, int *n_cnt) | |
8b61709d | 2788 | { |
b5d57fc8 | 2789 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7df6932e | 2790 | int error; |
86383816 BP |
2791 | |
2792 | ovs_mutex_lock(&netdev->mutex); | |
e0e2410d FL |
2793 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2794 | error = EOPNOTSUPP; | |
2795 | goto exit; | |
2796 | } | |
2797 | ||
a8704b50 | 2798 | error = netdev_get_addrs(netdev_get_name(netdev_), addr, mask, n_cnt); |
86383816 | 2799 | |
e0e2410d FL |
2800 | exit: |
2801 | ovs_mutex_unlock(&netdev->mutex); | |
7df6932e | 2802 | return error; |
8b61709d BP |
2803 | } |
2804 | ||
2805 | static void | |
2806 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
2807 | { | |
2808 | struct sockaddr_in sin; | |
2809 | memset(&sin, 0, sizeof sin); | |
2810 | sin.sin_family = AF_INET; | |
2811 | sin.sin_addr = addr; | |
2812 | sin.sin_port = 0; | |
2813 | ||
2814 | memset(sa, 0, sizeof *sa); | |
2815 | memcpy(sa, &sin, sizeof sin); | |
2816 | } | |
2817 | ||
2818 | static int | |
2819 | do_set_addr(struct netdev *netdev, | |
2820 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
2821 | { | |
2822 | struct ifreq ifr; | |
149f577a | 2823 | |
259e0b1a BP |
2824 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
2825 | return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
2826 | ioctl_name); | |
8b61709d BP |
2827 | } |
2828 | ||
2829 | /* Adds 'router' as a default IP gateway. */ | |
2830 | static int | |
67a4917b | 2831 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
2832 | { |
2833 | struct in_addr any = { INADDR_ANY }; | |
2834 | struct rtentry rt; | |
2835 | int error; | |
2836 | ||
2837 | memset(&rt, 0, sizeof rt); | |
2838 | make_in4_sockaddr(&rt.rt_dst, any); | |
2839 | make_in4_sockaddr(&rt.rt_gateway, router); | |
2840 | make_in4_sockaddr(&rt.rt_genmask, any); | |
2841 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
259e0b1a | 2842 | error = af_inet_ioctl(SIOCADDRT, &rt); |
8b61709d | 2843 | if (error) { |
10a89ef0 | 2844 | VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); |
8b61709d BP |
2845 | } |
2846 | return error; | |
2847 | } | |
2848 | ||
f1acd62b BP |
2849 | static int |
2850 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
2851 | char **netdev_name) | |
2852 | { | |
2853 | static const char fn[] = "/proc/net/route"; | |
2854 | FILE *stream; | |
2855 | char line[256]; | |
2856 | int ln; | |
2857 | ||
2858 | *netdev_name = NULL; | |
2859 | stream = fopen(fn, "r"); | |
2860 | if (stream == NULL) { | |
10a89ef0 | 2861 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
f1acd62b BP |
2862 | return errno; |
2863 | } | |
2864 | ||
2865 | ln = 0; | |
2866 | while (fgets(line, sizeof line, stream)) { | |
2867 | if (++ln >= 2) { | |
2868 | char iface[17]; | |
dbba996b | 2869 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
2870 | int refcnt, metric, mtu; |
2871 | unsigned int flags, use, window, irtt; | |
2872 | ||
c2c28dfd BP |
2873 | if (!ovs_scan(line, |
2874 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
2875 | " %d %u %u\n", | |
2876 | iface, &dest, &gateway, &flags, &refcnt, | |
2877 | &use, &metric, &mask, &mtu, &window, &irtt)) { | |
d295e8e9 | 2878 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
2879 | fn, ln, line); |
2880 | continue; | |
2881 | } | |
2882 | if (!(flags & RTF_UP)) { | |
2883 | /* Skip routes that aren't up. */ | |
2884 | continue; | |
2885 | } | |
2886 | ||
2887 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 2888 | * network byte order, so we don't need need any endian |
f1acd62b BP |
2889 | * conversions here. */ |
2890 | if ((dest & mask) == (host->s_addr & mask)) { | |
2891 | if (!gateway) { | |
2892 | /* The host is directly reachable. */ | |
2893 | next_hop->s_addr = 0; | |
2894 | } else { | |
2895 | /* To reach the host, we must go through a gateway. */ | |
2896 | next_hop->s_addr = gateway; | |
2897 | } | |
2898 | *netdev_name = xstrdup(iface); | |
2899 | fclose(stream); | |
2900 | return 0; | |
2901 | } | |
2902 | } | |
2903 | } | |
2904 | ||
2905 | fclose(stream); | |
2906 | return ENXIO; | |
2907 | } | |
2908 | ||
e210037e | 2909 | static int |
b5d57fc8 | 2910 | netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) |
e210037e | 2911 | { |
b5d57fc8 | 2912 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
275707c3 EJ |
2913 | int error = 0; |
2914 | ||
86383816 | 2915 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
2916 | if (!(netdev->cache_valid & VALID_DRVINFO)) { |
2917 | struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; | |
275707c3 EJ |
2918 | |
2919 | COVERAGE_INC(netdev_get_ethtool); | |
b5d57fc8 BP |
2920 | memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); |
2921 | error = netdev_linux_do_ethtool(netdev->up.name, | |
275707c3 EJ |
2922 | cmd, |
2923 | ETHTOOL_GDRVINFO, | |
2924 | "ETHTOOL_GDRVINFO"); | |
2925 | if (!error) { | |
b5d57fc8 | 2926 | netdev->cache_valid |= VALID_DRVINFO; |
275707c3 EJ |
2927 | } |
2928 | } | |
e210037e | 2929 | |
e210037e | 2930 | if (!error) { |
b5d57fc8 BP |
2931 | smap_add(smap, "driver_name", netdev->drvinfo.driver); |
2932 | smap_add(smap, "driver_version", netdev->drvinfo.version); | |
2933 | smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); | |
e210037e | 2934 | } |
86383816 BP |
2935 | ovs_mutex_unlock(&netdev->mutex); |
2936 | ||
e210037e AE |
2937 | return error; |
2938 | } | |
2939 | ||
4f925bd3 | 2940 | static int |
275707c3 EJ |
2941 | netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, |
2942 | struct smap *smap) | |
4f925bd3 | 2943 | { |
79f1cbe9 | 2944 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
2945 | return 0; |
2946 | } | |
2947 | ||
8b61709d BP |
2948 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
2949 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
2950 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
2951 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
2952 | static int | |
2953 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
74ff3298 | 2954 | ovs_be32 ip, struct eth_addr *mac) |
8b61709d BP |
2955 | { |
2956 | struct arpreq r; | |
c100e025 | 2957 | struct sockaddr_in sin; |
8b61709d BP |
2958 | int retval; |
2959 | ||
2960 | memset(&r, 0, sizeof r); | |
f2cc621b | 2961 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
2962 | sin.sin_family = AF_INET; |
2963 | sin.sin_addr.s_addr = ip; | |
2964 | sin.sin_port = 0; | |
2965 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
2966 | r.arp_ha.sa_family = ARPHRD_ETHER; |
2967 | r.arp_flags = 0; | |
71d7c22f | 2968 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d | 2969 | COVERAGE_INC(netdev_arp_lookup); |
259e0b1a | 2970 | retval = af_inet_ioctl(SIOCGARP, &r); |
8b61709d BP |
2971 | if (!retval) { |
2972 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
2973 | } else if (retval != ENXIO) { | |
2974 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
10a89ef0 BP |
2975 | netdev_get_name(netdev), IP_ARGS(ip), |
2976 | ovs_strerror(retval)); | |
8b61709d BP |
2977 | } |
2978 | return retval; | |
2979 | } | |
2980 | ||
2981 | static int | |
2982 | nd_to_iff_flags(enum netdev_flags nd) | |
2983 | { | |
2984 | int iff = 0; | |
2985 | if (nd & NETDEV_UP) { | |
2986 | iff |= IFF_UP; | |
2987 | } | |
2988 | if (nd & NETDEV_PROMISC) { | |
2989 | iff |= IFF_PROMISC; | |
2990 | } | |
7ba19d41 AC |
2991 | if (nd & NETDEV_LOOPBACK) { |
2992 | iff |= IFF_LOOPBACK; | |
2993 | } | |
8b61709d BP |
2994 | return iff; |
2995 | } | |
2996 | ||
2997 | static int | |
2998 | iff_to_nd_flags(int iff) | |
2999 | { | |
3000 | enum netdev_flags nd = 0; | |
3001 | if (iff & IFF_UP) { | |
3002 | nd |= NETDEV_UP; | |
3003 | } | |
3004 | if (iff & IFF_PROMISC) { | |
3005 | nd |= NETDEV_PROMISC; | |
3006 | } | |
7ba19d41 AC |
3007 | if (iff & IFF_LOOPBACK) { |
3008 | nd |= NETDEV_LOOPBACK; | |
3009 | } | |
8b61709d BP |
3010 | return nd; |
3011 | } | |
3012 | ||
3013 | static int | |
4f9f3f21 BP |
3014 | update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
3015 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
3016 | OVS_REQUIRES(netdev->mutex) | |
8b61709d BP |
3017 | { |
3018 | int old_flags, new_flags; | |
c37d4da4 EJ |
3019 | int error = 0; |
3020 | ||
b5d57fc8 | 3021 | old_flags = netdev->ifi_flags; |
c37d4da4 EJ |
3022 | *old_flagsp = iff_to_nd_flags(old_flags); |
3023 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
3024 | if (new_flags != old_flags) { | |
4f9f3f21 BP |
3025 | error = set_flags(netdev_get_name(&netdev->up), new_flags); |
3026 | get_flags(&netdev->up, &netdev->ifi_flags); | |
8b61709d | 3027 | } |
4f9f3f21 BP |
3028 | |
3029 | return error; | |
3030 | } | |
3031 | ||
3032 | static int | |
3033 | netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, | |
3034 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
3035 | { | |
3036 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
756819dd | 3037 | int error = 0; |
4f9f3f21 BP |
3038 | |
3039 | ovs_mutex_lock(&netdev->mutex); | |
756819dd FL |
3040 | if (on || off) { |
3041 | /* Changing flags over netlink isn't support yet. */ | |
e0e2410d FL |
3042 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3043 | error = EOPNOTSUPP; | |
3044 | goto exit; | |
3045 | } | |
756819dd FL |
3046 | error = update_flags(netdev, off, on, old_flagsp); |
3047 | } else { | |
3048 | /* Try reading flags over netlink, or fall back to ioctl. */ | |
3049 | if (!netdev_linux_update_via_netlink(netdev)) { | |
3050 | *old_flagsp = iff_to_nd_flags(netdev->ifi_flags); | |
3051 | } else { | |
3052 | error = update_flags(netdev, off, on, old_flagsp); | |
3053 | } | |
3054 | } | |
e0e2410d FL |
3055 | |
3056 | exit: | |
86383816 | 3057 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d BP |
3058 | return error; |
3059 | } | |
3060 | ||
2f9dd77f | 3061 | #define NETDEV_LINUX_CLASS(NAME, CONSTRUCT, GET_STATS, \ |
18ebd48c PB |
3062 | GET_FEATURES, GET_STATUS, \ |
3063 | FLOW_OFFLOAD_API) \ | |
c3827f61 BP |
3064 | { \ |
3065 | NAME, \ | |
118c77b1 | 3066 | false, /* is_pmd */ \ |
c3827f61 | 3067 | \ |
259e0b1a | 3068 | NULL, \ |
c3827f61 BP |
3069 | netdev_linux_run, \ |
3070 | netdev_linux_wait, \ | |
3071 | \ | |
9dc63482 BP |
3072 | netdev_linux_alloc, \ |
3073 | CONSTRUCT, \ | |
3074 | netdev_linux_destruct, \ | |
3075 | netdev_linux_dealloc, \ | |
de5cdb90 | 3076 | NULL, /* get_config */ \ |
6d9e6eb4 | 3077 | NULL, /* set_config */ \ |
f431bf7d | 3078 | NULL, /* get_tunnel_config */ \ |
a36de779 PS |
3079 | NULL, /* build header */ \ |
3080 | NULL, /* push header */ \ | |
3081 | NULL, /* pop header */ \ | |
7dec44fe | 3082 | NULL, /* get_numa_id */ \ |
050c60bf | 3083 | NULL, /* set_tx_multiq */ \ |
c3827f61 | 3084 | \ |
c3827f61 BP |
3085 | netdev_linux_send, \ |
3086 | netdev_linux_send_wait, \ | |
3087 | \ | |
3088 | netdev_linux_set_etheraddr, \ | |
3089 | netdev_linux_get_etheraddr, \ | |
3090 | netdev_linux_get_mtu, \ | |
9b020780 | 3091 | netdev_linux_set_mtu, \ |
c3827f61 BP |
3092 | netdev_linux_get_ifindex, \ |
3093 | netdev_linux_get_carrier, \ | |
65c3058c | 3094 | netdev_linux_get_carrier_resets, \ |
1670c579 | 3095 | netdev_linux_set_miimon_interval, \ |
f613a0d7 | 3096 | GET_STATS, \ |
971f4b39 | 3097 | NULL, \ |
c3827f61 | 3098 | \ |
51f87458 | 3099 | GET_FEATURES, \ |
c3827f61 | 3100 | netdev_linux_set_advertisements, \ |
875ab130 | 3101 | NULL, /* get_pt_mode */ \ |
c3827f61 BP |
3102 | \ |
3103 | netdev_linux_set_policing, \ | |
3104 | netdev_linux_get_qos_types, \ | |
3105 | netdev_linux_get_qos_capabilities, \ | |
3106 | netdev_linux_get_qos, \ | |
3107 | netdev_linux_set_qos, \ | |
3108 | netdev_linux_get_queue, \ | |
3109 | netdev_linux_set_queue, \ | |
3110 | netdev_linux_delete_queue, \ | |
3111 | netdev_linux_get_queue_stats, \ | |
89454bf4 BP |
3112 | netdev_linux_queue_dump_start, \ |
3113 | netdev_linux_queue_dump_next, \ | |
3114 | netdev_linux_queue_dump_done, \ | |
c3827f61 BP |
3115 | netdev_linux_dump_queue_stats, \ |
3116 | \ | |
c3827f61 | 3117 | netdev_linux_set_in4, \ |
a8704b50 | 3118 | netdev_linux_get_addr_list, \ |
c3827f61 BP |
3119 | netdev_linux_add_router, \ |
3120 | netdev_linux_get_next_hop, \ | |
4f925bd3 | 3121 | GET_STATUS, \ |
c3827f61 BP |
3122 | netdev_linux_arp_lookup, \ |
3123 | \ | |
3124 | netdev_linux_update_flags, \ | |
790fb3b7 | 3125 | NULL, /* reconfigure */ \ |
c3827f61 | 3126 | \ |
f7791740 PS |
3127 | netdev_linux_rxq_alloc, \ |
3128 | netdev_linux_rxq_construct, \ | |
3129 | netdev_linux_rxq_destruct, \ | |
3130 | netdev_linux_rxq_dealloc, \ | |
3131 | netdev_linux_rxq_recv, \ | |
3132 | netdev_linux_rxq_wait, \ | |
3133 | netdev_linux_rxq_drain, \ | |
18ebd48c | 3134 | \ |
88dcf2aa JH |
3135 | FLOW_OFFLOAD_API, \ |
3136 | NULL /* get_block_id */ \ | |
c3827f61 BP |
3137 | } |
3138 | ||
3139 | const struct netdev_class netdev_linux_class = | |
3140 | NETDEV_LINUX_CLASS( | |
3141 | "system", | |
9dc63482 | 3142 | netdev_linux_construct, |
f613a0d7 | 3143 | netdev_linux_get_stats, |
51f87458 | 3144 | netdev_linux_get_features, |
18ebd48c PB |
3145 | netdev_linux_get_status, |
3146 | LINUX_FLOW_OFFLOAD_API); | |
c3827f61 BP |
3147 | |
3148 | const struct netdev_class netdev_tap_class = | |
3149 | NETDEV_LINUX_CLASS( | |
3150 | "tap", | |
9dc63482 | 3151 | netdev_linux_construct_tap, |
bba1e6f3 | 3152 | netdev_tap_get_stats, |
51f87458 | 3153 | netdev_linux_get_features, |
18ebd48c PB |
3154 | netdev_linux_get_status, |
3155 | NO_OFFLOAD_API); | |
c3827f61 BP |
3156 | |
3157 | const struct netdev_class netdev_internal_class = | |
3158 | NETDEV_LINUX_CLASS( | |
3159 | "internal", | |
9dc63482 | 3160 | netdev_linux_construct, |
bba1e6f3 | 3161 | netdev_internal_get_stats, |
51f87458 | 3162 | NULL, /* get_features */ |
18ebd48c PB |
3163 | netdev_internal_get_status, |
3164 | NO_OFFLOAD_API); | |
8b61709d | 3165 | \f |
677d9158 JV |
3166 | |
3167 | #define CODEL_N_QUEUES 0x0000 | |
3168 | ||
2f4298ce BP |
3169 | /* In sufficiently new kernel headers these are defined as enums in |
3170 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3171 | * kernels. (This overrides any enum definition in the header file but that's | |
3172 | * harmless.) */ | |
3173 | #define TCA_CODEL_TARGET 1 | |
3174 | #define TCA_CODEL_LIMIT 2 | |
3175 | #define TCA_CODEL_INTERVAL 3 | |
3176 | ||
677d9158 JV |
3177 | struct codel { |
3178 | struct tc tc; | |
3179 | uint32_t target; | |
3180 | uint32_t limit; | |
3181 | uint32_t interval; | |
3182 | }; | |
3183 | ||
3184 | static struct codel * | |
3185 | codel_get__(const struct netdev *netdev_) | |
3186 | { | |
3187 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3188 | return CONTAINER_OF(netdev->tc, struct codel, tc); | |
3189 | } | |
3190 | ||
3191 | static void | |
3192 | codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3193 | uint32_t interval) | |
3194 | { | |
3195 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3196 | struct codel *codel; | |
3197 | ||
3198 | codel = xmalloc(sizeof *codel); | |
3199 | tc_init(&codel->tc, &tc_ops_codel); | |
3200 | codel->target = target; | |
3201 | codel->limit = limit; | |
3202 | codel->interval = interval; | |
3203 | ||
3204 | netdev->tc = &codel->tc; | |
3205 | } | |
3206 | ||
3207 | static int | |
3208 | codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3209 | uint32_t interval) | |
3210 | { | |
3211 | size_t opt_offset; | |
3212 | struct ofpbuf request; | |
3213 | struct tcmsg *tcmsg; | |
3214 | uint32_t otarget, olimit, ointerval; | |
3215 | int error; | |
3216 | ||
3217 | tc_del_qdisc(netdev); | |
3218 | ||
7874bdff RD |
3219 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3220 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3221 | if (!tcmsg) { |
3222 | return ENODEV; | |
3223 | } | |
3224 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3225 | tcmsg->tcm_parent = TC_H_ROOT; | |
3226 | ||
3227 | otarget = target ? target : 5000; | |
3228 | olimit = limit ? limit : 10240; | |
3229 | ointerval = interval ? interval : 100000; | |
3230 | ||
3231 | nl_msg_put_string(&request, TCA_KIND, "codel"); | |
3232 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3233 | nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget); | |
3234 | nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit); | |
3235 | nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval); | |
3236 | nl_msg_end_nested(&request, opt_offset); | |
3237 | ||
3238 | error = tc_transact(&request, NULL); | |
3239 | if (error) { | |
3240 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3241 | "target %u, limit %u, interval %u error %d(%s)", | |
3242 | netdev_get_name(netdev), | |
3243 | otarget, olimit, ointerval, | |
3244 | error, ovs_strerror(error)); | |
3245 | } | |
3246 | return error; | |
3247 | } | |
3248 | ||
3249 | static void | |
3250 | codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3251 | const struct smap *details, struct codel *codel) | |
3252 | { | |
13c1637f BP |
3253 | codel->target = smap_get_ullong(details, "target", 0); |
3254 | codel->limit = smap_get_ullong(details, "limit", 0); | |
3255 | codel->interval = smap_get_ullong(details, "interval", 0); | |
677d9158 JV |
3256 | |
3257 | if (!codel->target) { | |
3258 | codel->target = 5000; | |
3259 | } | |
3260 | if (!codel->limit) { | |
3261 | codel->limit = 10240; | |
3262 | } | |
3263 | if (!codel->interval) { | |
3264 | codel->interval = 100000; | |
3265 | } | |
3266 | } | |
3267 | ||
3268 | static int | |
3269 | codel_tc_install(struct netdev *netdev, const struct smap *details) | |
3270 | { | |
3271 | int error; | |
3272 | struct codel codel; | |
3273 | ||
3274 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3275 | error = codel_setup_qdisc__(netdev, codel.target, codel.limit, | |
3276 | codel.interval); | |
3277 | if (!error) { | |
3278 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3279 | } | |
3280 | return error; | |
3281 | } | |
3282 | ||
3283 | static int | |
3284 | codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel) | |
3285 | { | |
3286 | static const struct nl_policy tca_codel_policy[] = { | |
3287 | [TCA_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3288 | [TCA_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3289 | [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 } | |
3290 | }; | |
3291 | ||
3292 | struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)]; | |
3293 | ||
3294 | if (!nl_parse_nested(nl_options, tca_codel_policy, | |
3295 | attrs, ARRAY_SIZE(tca_codel_policy))) { | |
3296 | VLOG_WARN_RL(&rl, "failed to parse CoDel class options"); | |
3297 | return EPROTO; | |
3298 | } | |
3299 | ||
3300 | codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]); | |
3301 | codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]); | |
3302 | codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]); | |
3303 | return 0; | |
3304 | } | |
3305 | ||
3306 | static int | |
3307 | codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3308 | { | |
3309 | struct nlattr *nlattr; | |
3310 | const char * kind; | |
3311 | int error; | |
3312 | struct codel codel; | |
3313 | ||
3314 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3315 | if (error != 0) { | |
3316 | return error; | |
3317 | } | |
3318 | ||
3319 | error = codel_parse_tca_options__(nlattr, &codel); | |
3320 | if (error != 0) { | |
3321 | return error; | |
3322 | } | |
3323 | ||
3324 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3325 | return 0; | |
3326 | } | |
3327 | ||
3328 | ||
3329 | static void | |
3330 | codel_tc_destroy(struct tc *tc) | |
3331 | { | |
3332 | struct codel *codel = CONTAINER_OF(tc, struct codel, tc); | |
3333 | tc_destroy(tc); | |
3334 | free(codel); | |
3335 | } | |
3336 | ||
3337 | static int | |
3338 | codel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3339 | { | |
3340 | const struct codel *codel = codel_get__(netdev); | |
3341 | smap_add_format(details, "target", "%u", codel->target); | |
3342 | smap_add_format(details, "limit", "%u", codel->limit); | |
3343 | smap_add_format(details, "interval", "%u", codel->interval); | |
3344 | return 0; | |
3345 | } | |
3346 | ||
3347 | static int | |
3348 | codel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3349 | { | |
3350 | struct codel codel; | |
3351 | ||
3352 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3353 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3354 | codel_get__(netdev)->target = codel.target; | |
3355 | codel_get__(netdev)->limit = codel.limit; | |
3356 | codel_get__(netdev)->interval = codel.interval; | |
3357 | return 0; | |
3358 | } | |
3359 | ||
3360 | static const struct tc_ops tc_ops_codel = { | |
3361 | "codel", /* linux_name */ | |
3362 | "linux-codel", /* ovs_name */ | |
3363 | CODEL_N_QUEUES, /* n_queues */ | |
3364 | codel_tc_install, | |
3365 | codel_tc_load, | |
3366 | codel_tc_destroy, | |
3367 | codel_qdisc_get, | |
3368 | codel_qdisc_set, | |
3369 | NULL, | |
3370 | NULL, | |
3371 | NULL, | |
3372 | NULL, | |
3373 | NULL | |
3374 | }; | |
3375 | \f | |
3376 | /* FQ-CoDel traffic control class. */ | |
3377 | ||
3378 | #define FQCODEL_N_QUEUES 0x0000 | |
3379 | ||
2f4298ce BP |
3380 | /* In sufficiently new kernel headers these are defined as enums in |
3381 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3382 | * kernels. (This overrides any enum definition in the header file but that's | |
3383 | * harmless.) */ | |
3384 | #define TCA_FQ_CODEL_TARGET 1 | |
3385 | #define TCA_FQ_CODEL_LIMIT 2 | |
3386 | #define TCA_FQ_CODEL_INTERVAL 3 | |
3387 | #define TCA_FQ_CODEL_ECN 4 | |
3388 | #define TCA_FQ_CODEL_FLOWS 5 | |
3389 | #define TCA_FQ_CODEL_QUANTUM 6 | |
3390 | ||
677d9158 JV |
3391 | struct fqcodel { |
3392 | struct tc tc; | |
3393 | uint32_t target; | |
3394 | uint32_t limit; | |
3395 | uint32_t interval; | |
3396 | uint32_t flows; | |
3397 | uint32_t quantum; | |
3398 | }; | |
3399 | ||
3400 | static struct fqcodel * | |
3401 | fqcodel_get__(const struct netdev *netdev_) | |
3402 | { | |
3403 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3404 | return CONTAINER_OF(netdev->tc, struct fqcodel, tc); | |
3405 | } | |
3406 | ||
3407 | static void | |
3408 | fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3409 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3410 | { | |
3411 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3412 | struct fqcodel *fqcodel; | |
3413 | ||
3414 | fqcodel = xmalloc(sizeof *fqcodel); | |
3415 | tc_init(&fqcodel->tc, &tc_ops_fqcodel); | |
3416 | fqcodel->target = target; | |
3417 | fqcodel->limit = limit; | |
3418 | fqcodel->interval = interval; | |
3419 | fqcodel->flows = flows; | |
3420 | fqcodel->quantum = quantum; | |
3421 | ||
3422 | netdev->tc = &fqcodel->tc; | |
3423 | } | |
3424 | ||
3425 | static int | |
3426 | fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3427 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3428 | { | |
3429 | size_t opt_offset; | |
3430 | struct ofpbuf request; | |
3431 | struct tcmsg *tcmsg; | |
3432 | uint32_t otarget, olimit, ointerval, oflows, oquantum; | |
3433 | int error; | |
3434 | ||
3435 | tc_del_qdisc(netdev); | |
3436 | ||
7874bdff RD |
3437 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3438 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3439 | if (!tcmsg) { |
3440 | return ENODEV; | |
3441 | } | |
3442 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3443 | tcmsg->tcm_parent = TC_H_ROOT; | |
3444 | ||
3445 | otarget = target ? target : 5000; | |
3446 | olimit = limit ? limit : 10240; | |
3447 | ointerval = interval ? interval : 100000; | |
3448 | oflows = flows ? flows : 1024; | |
3449 | oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514 | |
3450 | not mtu */ | |
3451 | ||
3452 | nl_msg_put_string(&request, TCA_KIND, "fq_codel"); | |
3453 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3454 | nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget); | |
3455 | nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit); | |
3456 | nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval); | |
3457 | nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows); | |
3458 | nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum); | |
3459 | nl_msg_end_nested(&request, opt_offset); | |
3460 | ||
3461 | error = tc_transact(&request, NULL); | |
3462 | if (error) { | |
3463 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3464 | "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)", | |
3465 | netdev_get_name(netdev), | |
3466 | otarget, olimit, ointerval, oflows, oquantum, | |
3467 | error, ovs_strerror(error)); | |
3468 | } | |
3469 | return error; | |
3470 | } | |
3471 | ||
3472 | static void | |
3473 | fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3474 | const struct smap *details, struct fqcodel *fqcodel) | |
3475 | { | |
13c1637f BP |
3476 | fqcodel->target = smap_get_ullong(details, "target", 0); |
3477 | fqcodel->limit = smap_get_ullong(details, "limit", 0); | |
3478 | fqcodel->interval = smap_get_ullong(details, "interval", 0); | |
3479 | fqcodel->flows = smap_get_ullong(details, "flows", 0); | |
3480 | fqcodel->quantum = smap_get_ullong(details, "quantum", 0); | |
3481 | ||
677d9158 JV |
3482 | if (!fqcodel->target) { |
3483 | fqcodel->target = 5000; | |
3484 | } | |
3485 | if (!fqcodel->limit) { | |
3486 | fqcodel->limit = 10240; | |
3487 | } | |
3488 | if (!fqcodel->interval) { | |
3489 | fqcodel->interval = 1000000; | |
3490 | } | |
3491 | if (!fqcodel->flows) { | |
3492 | fqcodel->flows = 1024; | |
3493 | } | |
3494 | if (!fqcodel->quantum) { | |
3495 | fqcodel->quantum = 1514; | |
3496 | } | |
3497 | } | |
3498 | ||
3499 | static int | |
3500 | fqcodel_tc_install(struct netdev *netdev, const struct smap *details) | |
3501 | { | |
3502 | int error; | |
3503 | struct fqcodel fqcodel; | |
3504 | ||
3505 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3506 | error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit, | |
3507 | fqcodel.interval, fqcodel.flows, | |
3508 | fqcodel.quantum); | |
3509 | if (!error) { | |
3510 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, | |
3511 | fqcodel.interval, fqcodel.flows, fqcodel.quantum); | |
3512 | } | |
3513 | return error; | |
3514 | } | |
3515 | ||
3516 | static int | |
3517 | fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel) | |
3518 | { | |
3519 | static const struct nl_policy tca_fqcodel_policy[] = { | |
3520 | [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3521 | [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3522 | [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 }, | |
3523 | [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 }, | |
3524 | [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 } | |
3525 | }; | |
3526 | ||
3527 | struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)]; | |
3528 | ||
3529 | if (!nl_parse_nested(nl_options, tca_fqcodel_policy, | |
3530 | attrs, ARRAY_SIZE(tca_fqcodel_policy))) { | |
3531 | VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options"); | |
3532 | return EPROTO; | |
3533 | } | |
3534 | ||
3535 | fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]); | |
3536 | fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]); | |
3537 | fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]); | |
3538 | fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]); | |
3539 | fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]); | |
3540 | return 0; | |
3541 | } | |
3542 | ||
3543 | static int | |
3544 | fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3545 | { | |
3546 | struct nlattr *nlattr; | |
3547 | const char * kind; | |
3548 | int error; | |
3549 | struct fqcodel fqcodel; | |
3550 | ||
3551 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3552 | if (error != 0) { | |
3553 | return error; | |
3554 | } | |
3555 | ||
3556 | error = fqcodel_parse_tca_options__(nlattr, &fqcodel); | |
3557 | if (error != 0) { | |
3558 | return error; | |
3559 | } | |
3560 | ||
3561 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3562 | fqcodel.flows, fqcodel.quantum); | |
3563 | return 0; | |
3564 | } | |
3565 | ||
3566 | static void | |
3567 | fqcodel_tc_destroy(struct tc *tc) | |
3568 | { | |
3569 | struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc); | |
3570 | tc_destroy(tc); | |
3571 | free(fqcodel); | |
3572 | } | |
3573 | ||
3574 | static int | |
3575 | fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3576 | { | |
3577 | const struct fqcodel *fqcodel = fqcodel_get__(netdev); | |
3578 | smap_add_format(details, "target", "%u", fqcodel->target); | |
3579 | smap_add_format(details, "limit", "%u", fqcodel->limit); | |
3580 | smap_add_format(details, "interval", "%u", fqcodel->interval); | |
3581 | smap_add_format(details, "flows", "%u", fqcodel->flows); | |
3582 | smap_add_format(details, "quantum", "%u", fqcodel->quantum); | |
3583 | return 0; | |
3584 | } | |
3585 | ||
3586 | static int | |
3587 | fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3588 | { | |
3589 | struct fqcodel fqcodel; | |
3590 | ||
3591 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3592 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
3593 | fqcodel.flows, fqcodel.quantum); | |
3594 | fqcodel_get__(netdev)->target = fqcodel.target; | |
3595 | fqcodel_get__(netdev)->limit = fqcodel.limit; | |
3596 | fqcodel_get__(netdev)->interval = fqcodel.interval; | |
3597 | fqcodel_get__(netdev)->flows = fqcodel.flows; | |
3598 | fqcodel_get__(netdev)->quantum = fqcodel.quantum; | |
3599 | return 0; | |
3600 | } | |
3601 | ||
3602 | static const struct tc_ops tc_ops_fqcodel = { | |
3603 | "fq_codel", /* linux_name */ | |
3604 | "linux-fq_codel", /* ovs_name */ | |
3605 | FQCODEL_N_QUEUES, /* n_queues */ | |
3606 | fqcodel_tc_install, | |
3607 | fqcodel_tc_load, | |
3608 | fqcodel_tc_destroy, | |
3609 | fqcodel_qdisc_get, | |
3610 | fqcodel_qdisc_set, | |
3611 | NULL, | |
3612 | NULL, | |
3613 | NULL, | |
3614 | NULL, | |
3615 | NULL | |
3616 | }; | |
3617 | \f | |
3618 | /* SFQ traffic control class. */ | |
3619 | ||
3620 | #define SFQ_N_QUEUES 0x0000 | |
3621 | ||
3622 | struct sfq { | |
3623 | struct tc tc; | |
3624 | uint32_t quantum; | |
3625 | uint32_t perturb; | |
3626 | }; | |
3627 | ||
3628 | static struct sfq * | |
3629 | sfq_get__(const struct netdev *netdev_) | |
3630 | { | |
3631 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3632 | return CONTAINER_OF(netdev->tc, struct sfq, tc); | |
3633 | } | |
3634 | ||
3635 | static void | |
3636 | sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb) | |
3637 | { | |
3638 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3639 | struct sfq *sfq; | |
3640 | ||
3641 | sfq = xmalloc(sizeof *sfq); | |
3642 | tc_init(&sfq->tc, &tc_ops_sfq); | |
3643 | sfq->perturb = perturb; | |
3644 | sfq->quantum = quantum; | |
3645 | ||
3646 | netdev->tc = &sfq->tc; | |
3647 | } | |
3648 | ||
3649 | static int | |
3650 | sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb) | |
3651 | { | |
3652 | struct tc_sfq_qopt opt; | |
3653 | struct ofpbuf request; | |
3654 | struct tcmsg *tcmsg; | |
3655 | int mtu; | |
3656 | int mtu_error, error; | |
3657 | mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); | |
3658 | ||
3659 | tc_del_qdisc(netdev); | |
3660 | ||
7874bdff RD |
3661 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3662 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3663 | if (!tcmsg) { |
3664 | return ENODEV; | |
3665 | } | |
3666 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3667 | tcmsg->tcm_parent = TC_H_ROOT; | |
3668 | ||
3669 | memset(&opt, 0, sizeof opt); | |
3670 | if (!quantum) { | |
3671 | if (!mtu_error) { | |
3672 | opt.quantum = mtu; /* if we cannot find mtu, use default */ | |
3673 | } | |
3674 | } else { | |
3675 | opt.quantum = quantum; | |
3676 | } | |
3677 | ||
3678 | if (!perturb) { | |
3679 | opt.perturb_period = 10; | |
3680 | } else { | |
3681 | opt.perturb_period = perturb; | |
3682 | } | |
3683 | ||
3684 | nl_msg_put_string(&request, TCA_KIND, "sfq"); | |
3685 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
3686 | ||
3687 | error = tc_transact(&request, NULL); | |
3688 | if (error) { | |
3689 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3690 | "quantum %u, perturb %u error %d(%s)", | |
3691 | netdev_get_name(netdev), | |
3692 | opt.quantum, opt.perturb_period, | |
3693 | error, ovs_strerror(error)); | |
3694 | } | |
3695 | return error; | |
3696 | } | |
3697 | ||
3698 | static void | |
3699 | sfq_parse_qdisc_details__(struct netdev *netdev, | |
3700 | const struct smap *details, struct sfq *sfq) | |
3701 | { | |
13c1637f BP |
3702 | sfq->perturb = smap_get_ullong(details, "perturb", 0); |
3703 | sfq->quantum = smap_get_ullong(details, "quantum", 0); | |
677d9158 | 3704 | |
677d9158 JV |
3705 | if (!sfq->perturb) { |
3706 | sfq->perturb = 10; | |
3707 | } | |
3708 | ||
3709 | if (!sfq->quantum) { | |
13c1637f BP |
3710 | int mtu; |
3711 | if (!netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
677d9158 JV |
3712 | sfq->quantum = mtu; |
3713 | } else { | |
3714 | VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a " | |
3715 | "device without mtu"); | |
677d9158 JV |
3716 | } |
3717 | } | |
3718 | } | |
3719 | ||
3720 | static int | |
3721 | sfq_tc_install(struct netdev *netdev, const struct smap *details) | |
3722 | { | |
3723 | int error; | |
3724 | struct sfq sfq; | |
3725 | ||
3726 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3727 | error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb); | |
3728 | if (!error) { | |
3729 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3730 | } | |
3731 | return error; | |
3732 | } | |
3733 | ||
3734 | static int | |
3735 | sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3736 | { | |
3737 | const struct tc_sfq_qopt *sfq; | |
3738 | struct nlattr *nlattr; | |
3739 | const char * kind; | |
3740 | int error; | |
3741 | ||
3742 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3743 | if (error == 0) { | |
3744 | sfq = nl_attr_get(nlattr); | |
3745 | sfq_install__(netdev, sfq->perturb_period, sfq->quantum); | |
3746 | return 0; | |
3747 | } | |
3748 | ||
3749 | return error; | |
3750 | } | |
3751 | ||
3752 | static void | |
3753 | sfq_tc_destroy(struct tc *tc) | |
3754 | { | |
3755 | struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc); | |
3756 | tc_destroy(tc); | |
3757 | free(sfq); | |
3758 | } | |
3759 | ||
3760 | static int | |
3761 | sfq_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3762 | { | |
3763 | const struct sfq *sfq = sfq_get__(netdev); | |
3764 | smap_add_format(details, "quantum", "%u", sfq->quantum); | |
3765 | smap_add_format(details, "perturb", "%u", sfq->perturb); | |
3766 | return 0; | |
3767 | } | |
3768 | ||
3769 | static int | |
3770 | sfq_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3771 | { | |
3772 | struct sfq sfq; | |
3773 | ||
3774 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
3775 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
3776 | sfq_get__(netdev)->quantum = sfq.quantum; | |
3777 | sfq_get__(netdev)->perturb = sfq.perturb; | |
3778 | return 0; | |
3779 | } | |
3780 | ||
3781 | static const struct tc_ops tc_ops_sfq = { | |
3782 | "sfq", /* linux_name */ | |
3783 | "linux-sfq", /* ovs_name */ | |
3784 | SFQ_N_QUEUES, /* n_queues */ | |
3785 | sfq_tc_install, | |
3786 | sfq_tc_load, | |
3787 | sfq_tc_destroy, | |
3788 | sfq_qdisc_get, | |
3789 | sfq_qdisc_set, | |
3790 | NULL, | |
3791 | NULL, | |
3792 | NULL, | |
3793 | NULL, | |
3794 | NULL | |
3795 | }; | |
3796 | \f | |
c1c9c9c4 | 3797 | /* HTB traffic control class. */ |
559843ed | 3798 | |
c1c9c9c4 | 3799 | #define HTB_N_QUEUES 0xf000 |
4f631ccd | 3800 | #define HTB_RATE2QUANTUM 10 |
8b61709d | 3801 | |
c1c9c9c4 BP |
3802 | struct htb { |
3803 | struct tc tc; | |
3804 | unsigned int max_rate; /* In bytes/s. */ | |
3805 | }; | |
8b61709d | 3806 | |
c1c9c9c4 | 3807 | struct htb_class { |
93b13be8 | 3808 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
3809 | unsigned int min_rate; /* In bytes/s. */ |
3810 | unsigned int max_rate; /* In bytes/s. */ | |
3811 | unsigned int burst; /* In bytes. */ | |
3812 | unsigned int priority; /* Lower values are higher priorities. */ | |
3813 | }; | |
8b61709d | 3814 | |
c1c9c9c4 | 3815 | static struct htb * |
b5d57fc8 | 3816 | htb_get__(const struct netdev *netdev_) |
c1c9c9c4 | 3817 | { |
b5d57fc8 BP |
3818 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
3819 | return CONTAINER_OF(netdev->tc, struct htb, tc); | |
c1c9c9c4 BP |
3820 | } |
3821 | ||
24045e35 | 3822 | static void |
b5d57fc8 | 3823 | htb_install__(struct netdev *netdev_, uint64_t max_rate) |
c1c9c9c4 | 3824 | { |
b5d57fc8 | 3825 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3826 | struct htb *htb; |
3827 | ||
3828 | htb = xmalloc(sizeof *htb); | |
3829 | tc_init(&htb->tc, &tc_ops_htb); | |
3830 | htb->max_rate = max_rate; | |
3831 | ||
b5d57fc8 | 3832 | netdev->tc = &htb->tc; |
c1c9c9c4 BP |
3833 | } |
3834 | ||
3835 | /* Create an HTB qdisc. | |
3836 | * | |
a339aa81 | 3837 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
3838 | static int |
3839 | htb_setup_qdisc__(struct netdev *netdev) | |
3840 | { | |
3841 | size_t opt_offset; | |
3842 | struct tc_htb_glob opt; | |
3843 | struct ofpbuf request; | |
3844 | struct tcmsg *tcmsg; | |
3845 | ||
3846 | tc_del_qdisc(netdev); | |
3847 | ||
7874bdff RD |
3848 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3849 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
3850 | if (!tcmsg) { |
3851 | return ENODEV; | |
3852 | } | |
c1c9c9c4 BP |
3853 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
3854 | tcmsg->tcm_parent = TC_H_ROOT; | |
3855 | ||
3856 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3857 | ||
3858 | memset(&opt, 0, sizeof opt); | |
4f631ccd | 3859 | opt.rate2quantum = HTB_RATE2QUANTUM; |
c1c9c9c4 | 3860 | opt.version = 3; |
4ecf12d5 | 3861 | opt.defcls = 1; |
c1c9c9c4 BP |
3862 | |
3863 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3864 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
3865 | nl_msg_end_nested(&request, opt_offset); | |
3866 | ||
3867 | return tc_transact(&request, NULL); | |
3868 | } | |
3869 | ||
3870 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
3871 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
3872 | static int | |
3873 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
3874 | unsigned int parent, struct htb_class *class) | |
3875 | { | |
3876 | size_t opt_offset; | |
3877 | struct tc_htb_opt opt; | |
3878 | struct ofpbuf request; | |
3879 | struct tcmsg *tcmsg; | |
3880 | int error; | |
3881 | int mtu; | |
3882 | ||
73371c09 | 3883 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 3884 | if (error) { |
f915f1a8 BP |
3885 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
3886 | netdev_get_name(netdev)); | |
9b020780 | 3887 | return error; |
f915f1a8 | 3888 | } |
c1c9c9c4 BP |
3889 | |
3890 | memset(&opt, 0, sizeof opt); | |
3891 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
3892 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
4f631ccd AW |
3893 | /* Makes sure the quantum is at least MTU. Setting quantum will |
3894 | * make htb ignore the r2q for this class. */ | |
3895 | if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) { | |
3896 | opt.quantum = mtu; | |
3897 | } | |
c1c9c9c4 BP |
3898 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); |
3899 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
3900 | opt.prio = class->priority; | |
3901 | ||
7874bdff RD |
3902 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
3903 | &request); | |
23a98ffe BP |
3904 | if (!tcmsg) { |
3905 | return ENODEV; | |
3906 | } | |
c1c9c9c4 BP |
3907 | tcmsg->tcm_handle = handle; |
3908 | tcmsg->tcm_parent = parent; | |
3909 | ||
3910 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
3911 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3912 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
3913 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
3914 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
3915 | nl_msg_end_nested(&request, opt_offset); | |
3916 | ||
3917 | error = tc_transact(&request, NULL); | |
3918 | if (error) { | |
3919 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
3920 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
3921 | netdev_get_name(netdev), | |
3922 | tc_get_major(handle), tc_get_minor(handle), | |
3923 | tc_get_major(parent), tc_get_minor(parent), | |
3924 | class->min_rate, class->max_rate, | |
10a89ef0 | 3925 | class->burst, class->priority, ovs_strerror(error)); |
c1c9c9c4 BP |
3926 | } |
3927 | return error; | |
3928 | } | |
3929 | ||
3930 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
3931 | * description of them into 'details'. The description complies with the | |
3932 | * specification given in the vswitch database documentation for linux-htb | |
3933 | * queue details. */ | |
3934 | static int | |
3935 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
3936 | { | |
3937 | static const struct nl_policy tca_htb_policy[] = { | |
3938 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
3939 | .min_len = sizeof(struct tc_htb_opt) }, | |
3940 | }; | |
3941 | ||
3942 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
3943 | const struct tc_htb_opt *htb; | |
3944 | ||
3945 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
3946 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
3947 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
3948 | return EPROTO; | |
3949 | } | |
3950 | ||
3951 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
3952 | class->min_rate = htb->rate.rate; | |
3953 | class->max_rate = htb->ceil.rate; | |
3954 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
3955 | class->priority = htb->prio; | |
3956 | return 0; | |
3957 | } | |
3958 | ||
3959 | static int | |
3960 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
3961 | struct htb_class *options, | |
3962 | struct netdev_queue_stats *stats) | |
3963 | { | |
3964 | struct nlattr *nl_options; | |
3965 | unsigned int handle; | |
3966 | int error; | |
3967 | ||
3968 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
3969 | if (!error && queue_id) { | |
17ee3c1f BP |
3970 | unsigned int major = tc_get_major(handle); |
3971 | unsigned int minor = tc_get_minor(handle); | |
3972 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
3973 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
3974 | } else { |
3975 | error = EPROTO; | |
3976 | } | |
3977 | } | |
3978 | if (!error && options) { | |
3979 | error = htb_parse_tca_options__(nl_options, options); | |
3980 | } | |
3981 | return error; | |
3982 | } | |
3983 | ||
3984 | static void | |
73371c09 | 3985 | htb_parse_qdisc_details__(struct netdev *netdev_, |
79f1cbe9 | 3986 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 | 3987 | { |
73371c09 | 3988 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 | 3989 | |
13c1637f | 3990 | hc->max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
c1c9c9c4 | 3991 | if (!hc->max_rate) { |
a00ca915 | 3992 | enum netdev_features current; |
c1c9c9c4 | 3993 | |
73371c09 BP |
3994 | netdev_linux_read_features(netdev); |
3995 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 3996 | hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
c1c9c9c4 BP |
3997 | } |
3998 | hc->min_rate = hc->max_rate; | |
3999 | hc->burst = 0; | |
4000 | hc->priority = 0; | |
4001 | } | |
4002 | ||
4003 | static int | |
4004 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 4005 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
4006 | { |
4007 | const struct htb *htb = htb_get__(netdev); | |
9b020780 | 4008 | int mtu, error; |
214117fd | 4009 | unsigned long long int max_rate_bit; |
c1c9c9c4 | 4010 | |
73371c09 | 4011 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 4012 | if (error) { |
f915f1a8 BP |
4013 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
4014 | netdev_get_name(netdev)); | |
9b020780 | 4015 | return error; |
f915f1a8 BP |
4016 | } |
4017 | ||
4f104611 EJ |
4018 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
4019 | * on uncongested links. */ | |
13c1637f | 4020 | hc->min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
4f104611 | 4021 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
4022 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
4023 | ||
4024 | /* max-rate */ | |
214117fd KF |
4025 | max_rate_bit = smap_get_ullong(details, "max-rate", 0); |
4026 | hc->max_rate = max_rate_bit ? max_rate_bit / 8 : htb->max_rate; | |
c1c9c9c4 BP |
4027 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); |
4028 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
4029 | ||
4030 | /* burst | |
4031 | * | |
4032 | * According to hints in the documentation that I've read, it is important | |
4033 | * that 'burst' be at least as big as the largest frame that might be | |
4034 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
4035 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
4036 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
4037 | * the MTU. We actually add 64, instead of 14, as a guard against | |
4038 | * additional headers get tacked on somewhere that we're not aware of. */ | |
13c1637f | 4039 | hc->burst = smap_get_ullong(details, "burst", 0) / 8; |
c1c9c9c4 BP |
4040 | hc->burst = MAX(hc->burst, mtu + 64); |
4041 | ||
4042 | /* priority */ | |
13c1637f | 4043 | hc->priority = smap_get_ullong(details, "priority", 0); |
c1c9c9c4 BP |
4044 | |
4045 | return 0; | |
4046 | } | |
4047 | ||
4048 | static int | |
4049 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
4050 | unsigned int parent, struct htb_class *options, | |
4051 | struct netdev_queue_stats *stats) | |
4052 | { | |
4053 | struct ofpbuf *reply; | |
4054 | int error; | |
4055 | ||
4056 | error = tc_query_class(netdev, handle, parent, &reply); | |
4057 | if (!error) { | |
4058 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
4059 | ofpbuf_delete(reply); | |
4060 | } | |
4061 | return error; | |
4062 | } | |
4063 | ||
4064 | static int | |
79f1cbe9 | 4065 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
4066 | { |
4067 | int error; | |
4068 | ||
4069 | error = htb_setup_qdisc__(netdev); | |
4070 | if (!error) { | |
4071 | struct htb_class hc; | |
4072 | ||
4073 | htb_parse_qdisc_details__(netdev, details, &hc); | |
4074 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4075 | tc_make_handle(1, 0), &hc); | |
4076 | if (!error) { | |
4077 | htb_install__(netdev, hc.max_rate); | |
4078 | } | |
4079 | } | |
4080 | return error; | |
4081 | } | |
4082 | ||
93b13be8 BP |
4083 | static struct htb_class * |
4084 | htb_class_cast__(const struct tc_queue *queue) | |
4085 | { | |
4086 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
4087 | } | |
4088 | ||
c1c9c9c4 BP |
4089 | static void |
4090 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4091 | const struct htb_class *hc) | |
4092 | { | |
4093 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
4094 | size_t hash = hash_int(queue_id, 0); |
4095 | struct tc_queue *queue; | |
c1c9c9c4 BP |
4096 | struct htb_class *hcp; |
4097 | ||
93b13be8 BP |
4098 | queue = tc_find_queue__(netdev, queue_id, hash); |
4099 | if (queue) { | |
4100 | hcp = htb_class_cast__(queue); | |
4101 | } else { | |
c1c9c9c4 | 4102 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
4103 | queue = &hcp->tc_queue; |
4104 | queue->queue_id = queue_id; | |
6dc34a0d | 4105 | queue->created = time_msec(); |
93b13be8 | 4106 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); |
c1c9c9c4 | 4107 | } |
93b13be8 BP |
4108 | |
4109 | hcp->min_rate = hc->min_rate; | |
4110 | hcp->max_rate = hc->max_rate; | |
4111 | hcp->burst = hc->burst; | |
4112 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
4113 | } |
4114 | ||
4115 | static int | |
4116 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4117 | { | |
c1c9c9c4 | 4118 | struct ofpbuf msg; |
d57695d7 | 4119 | struct queue_dump_state state; |
c1c9c9c4 | 4120 | struct htb_class hc; |
c1c9c9c4 BP |
4121 | |
4122 | /* Get qdisc options. */ | |
4123 | hc.max_rate = 0; | |
4124 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 4125 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
4126 | |
4127 | /* Get queues. */ | |
d57695d7 | 4128 | if (!start_queue_dump(netdev, &state)) { |
23a98ffe BP |
4129 | return ENODEV; |
4130 | } | |
d57695d7 | 4131 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
c1c9c9c4 BP |
4132 | unsigned int queue_id; |
4133 | ||
4134 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
4135 | htb_update_queue__(netdev, queue_id, &hc); | |
4136 | } | |
4137 | } | |
d57695d7 | 4138 | finish_queue_dump(&state); |
c1c9c9c4 BP |
4139 | |
4140 | return 0; | |
4141 | } | |
4142 | ||
4143 | static void | |
4144 | htb_tc_destroy(struct tc *tc) | |
4145 | { | |
4146 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
4ec3d7c7 | 4147 | struct htb_class *hc; |
c1c9c9c4 | 4148 | |
4ec3d7c7 | 4149 | HMAP_FOR_EACH_POP (hc, tc_queue.hmap_node, &htb->tc.queues) { |
c1c9c9c4 BP |
4150 | free(hc); |
4151 | } | |
4152 | tc_destroy(tc); | |
4153 | free(htb); | |
4154 | } | |
4155 | ||
4156 | static int | |
79f1cbe9 | 4157 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
4158 | { |
4159 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 4160 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
4161 | return 0; |
4162 | } | |
4163 | ||
4164 | static int | |
79f1cbe9 | 4165 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
4166 | { |
4167 | struct htb_class hc; | |
4168 | int error; | |
4169 | ||
4170 | htb_parse_qdisc_details__(netdev, details, &hc); | |
4171 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4172 | tc_make_handle(1, 0), &hc); | |
4173 | if (!error) { | |
4174 | htb_get__(netdev)->max_rate = hc.max_rate; | |
4175 | } | |
4176 | return error; | |
4177 | } | |
4178 | ||
4179 | static int | |
93b13be8 | 4180 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 4181 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 4182 | { |
93b13be8 | 4183 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4184 | |
79f1cbe9 | 4185 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 4186 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4187 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 4188 | } |
79f1cbe9 | 4189 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 4190 | if (hc->priority) { |
79f1cbe9 | 4191 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
4192 | } |
4193 | return 0; | |
4194 | } | |
4195 | ||
4196 | static int | |
4197 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4198 | const struct smap *details) |
c1c9c9c4 BP |
4199 | { |
4200 | struct htb_class hc; | |
4201 | int error; | |
4202 | ||
4203 | error = htb_parse_class_details__(netdev, details, &hc); | |
4204 | if (error) { | |
4205 | return error; | |
4206 | } | |
4207 | ||
17ee3c1f | 4208 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
4209 | tc_make_handle(1, 0xfffe), &hc); |
4210 | if (error) { | |
4211 | return error; | |
4212 | } | |
4213 | ||
4214 | htb_update_queue__(netdev, queue_id, &hc); | |
4215 | return 0; | |
4216 | } | |
4217 | ||
4218 | static int | |
93b13be8 | 4219 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 4220 | { |
93b13be8 | 4221 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4222 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
4223 | int error; |
4224 | ||
93b13be8 | 4225 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 4226 | if (!error) { |
93b13be8 | 4227 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 4228 | free(hc); |
c1c9c9c4 BP |
4229 | } |
4230 | return error; | |
4231 | } | |
4232 | ||
4233 | static int | |
93b13be8 | 4234 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
4235 | struct netdev_queue_stats *stats) |
4236 | { | |
93b13be8 | 4237 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
4238 | tc_make_handle(1, 0xfffe), NULL, stats); |
4239 | } | |
4240 | ||
4241 | static int | |
4242 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4243 | const struct ofpbuf *nlmsg, | |
4244 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4245 | { | |
4246 | struct netdev_queue_stats stats; | |
17ee3c1f | 4247 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
4248 | int error; |
4249 | ||
4250 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4251 | if (error) { | |
4252 | return error; | |
4253 | } | |
4254 | ||
17ee3c1f BP |
4255 | major = tc_get_major(handle); |
4256 | minor = tc_get_minor(handle); | |
4257 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 4258 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
4259 | } |
4260 | return 0; | |
4261 | } | |
4262 | ||
4263 | static const struct tc_ops tc_ops_htb = { | |
4264 | "htb", /* linux_name */ | |
4265 | "linux-htb", /* ovs_name */ | |
4266 | HTB_N_QUEUES, /* n_queues */ | |
4267 | htb_tc_install, | |
4268 | htb_tc_load, | |
4269 | htb_tc_destroy, | |
4270 | htb_qdisc_get, | |
4271 | htb_qdisc_set, | |
4272 | htb_class_get, | |
4273 | htb_class_set, | |
4274 | htb_class_delete, | |
4275 | htb_class_get_stats, | |
4276 | htb_class_dump_stats | |
4277 | }; | |
4278 | \f | |
a339aa81 EJ |
4279 | /* "linux-hfsc" traffic control class. */ |
4280 | ||
4281 | #define HFSC_N_QUEUES 0xf000 | |
4282 | ||
4283 | struct hfsc { | |
4284 | struct tc tc; | |
4285 | uint32_t max_rate; | |
4286 | }; | |
4287 | ||
4288 | struct hfsc_class { | |
4289 | struct tc_queue tc_queue; | |
4290 | uint32_t min_rate; | |
4291 | uint32_t max_rate; | |
4292 | }; | |
4293 | ||
4294 | static struct hfsc * | |
b5d57fc8 | 4295 | hfsc_get__(const struct netdev *netdev_) |
a339aa81 | 4296 | { |
b5d57fc8 BP |
4297 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4298 | return CONTAINER_OF(netdev->tc, struct hfsc, tc); | |
a339aa81 EJ |
4299 | } |
4300 | ||
4301 | static struct hfsc_class * | |
4302 | hfsc_class_cast__(const struct tc_queue *queue) | |
4303 | { | |
4304 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
4305 | } | |
4306 | ||
24045e35 | 4307 | static void |
b5d57fc8 | 4308 | hfsc_install__(struct netdev *netdev_, uint32_t max_rate) |
a339aa81 | 4309 | { |
b5d57fc8 | 4310 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
4311 | struct hfsc *hfsc; |
4312 | ||
a339aa81 EJ |
4313 | hfsc = xmalloc(sizeof *hfsc); |
4314 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
4315 | hfsc->max_rate = max_rate; | |
b5d57fc8 | 4316 | netdev->tc = &hfsc->tc; |
a339aa81 EJ |
4317 | } |
4318 | ||
4319 | static void | |
4320 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4321 | const struct hfsc_class *hc) | |
4322 | { | |
4323 | size_t hash; | |
4324 | struct hfsc *hfsc; | |
4325 | struct hfsc_class *hcp; | |
4326 | struct tc_queue *queue; | |
4327 | ||
4328 | hfsc = hfsc_get__(netdev); | |
4329 | hash = hash_int(queue_id, 0); | |
4330 | ||
4331 | queue = tc_find_queue__(netdev, queue_id, hash); | |
4332 | if (queue) { | |
4333 | hcp = hfsc_class_cast__(queue); | |
4334 | } else { | |
4335 | hcp = xmalloc(sizeof *hcp); | |
4336 | queue = &hcp->tc_queue; | |
4337 | queue->queue_id = queue_id; | |
6dc34a0d | 4338 | queue->created = time_msec(); |
a339aa81 EJ |
4339 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); |
4340 | } | |
4341 | ||
4342 | hcp->min_rate = hc->min_rate; | |
4343 | hcp->max_rate = hc->max_rate; | |
4344 | } | |
4345 | ||
4346 | static int | |
4347 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
4348 | { | |
4349 | const struct tc_service_curve *rsc, *fsc, *usc; | |
4350 | static const struct nl_policy tca_hfsc_policy[] = { | |
4351 | [TCA_HFSC_RSC] = { | |
4352 | .type = NL_A_UNSPEC, | |
4353 | .optional = false, | |
4354 | .min_len = sizeof(struct tc_service_curve), | |
4355 | }, | |
4356 | [TCA_HFSC_FSC] = { | |
4357 | .type = NL_A_UNSPEC, | |
4358 | .optional = false, | |
4359 | .min_len = sizeof(struct tc_service_curve), | |
4360 | }, | |
4361 | [TCA_HFSC_USC] = { | |
4362 | .type = NL_A_UNSPEC, | |
4363 | .optional = false, | |
4364 | .min_len = sizeof(struct tc_service_curve), | |
4365 | }, | |
4366 | }; | |
4367 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
4368 | ||
4369 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
4370 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
4371 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
4372 | return EPROTO; | |
4373 | } | |
4374 | ||
4375 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
4376 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
4377 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
4378 | ||
4379 | if (rsc->m1 != 0 || rsc->d != 0 || | |
4380 | fsc->m1 != 0 || fsc->d != 0 || | |
4381 | usc->m1 != 0 || usc->d != 0) { | |
4382 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4383 | "Non-linear service curves are not supported."); | |
4384 | return EPROTO; | |
4385 | } | |
4386 | ||
4387 | if (rsc->m2 != fsc->m2) { | |
4388 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4389 | "Real-time service curves are not supported "); | |
4390 | return EPROTO; | |
4391 | } | |
4392 | ||
4393 | if (rsc->m2 > usc->m2) { | |
4394 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
4395 | "Min-rate service curve is greater than " | |
4396 | "the max-rate service curve."); | |
4397 | return EPROTO; | |
4398 | } | |
4399 | ||
4400 | class->min_rate = fsc->m2; | |
4401 | class->max_rate = usc->m2; | |
4402 | return 0; | |
4403 | } | |
4404 | ||
4405 | static int | |
4406 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
4407 | struct hfsc_class *options, | |
4408 | struct netdev_queue_stats *stats) | |
4409 | { | |
4410 | int error; | |
4411 | unsigned int handle; | |
4412 | struct nlattr *nl_options; | |
4413 | ||
4414 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
4415 | if (error) { | |
4416 | return error; | |
4417 | } | |
4418 | ||
4419 | if (queue_id) { | |
4420 | unsigned int major, minor; | |
4421 | ||
4422 | major = tc_get_major(handle); | |
4423 | minor = tc_get_minor(handle); | |
4424 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4425 | *queue_id = minor - 1; | |
4426 | } else { | |
4427 | return EPROTO; | |
4428 | } | |
4429 | } | |
4430 | ||
4431 | if (options) { | |
4432 | error = hfsc_parse_tca_options__(nl_options, options); | |
4433 | } | |
4434 | ||
4435 | return error; | |
4436 | } | |
4437 | ||
4438 | static int | |
4439 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
4440 | unsigned int parent, struct hfsc_class *options, | |
4441 | struct netdev_queue_stats *stats) | |
4442 | { | |
4443 | int error; | |
4444 | struct ofpbuf *reply; | |
4445 | ||
4446 | error = tc_query_class(netdev, handle, parent, &reply); | |
4447 | if (error) { | |
4448 | return error; | |
4449 | } | |
4450 | ||
4451 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
4452 | ofpbuf_delete(reply); | |
4453 | return error; | |
4454 | } | |
4455 | ||
4456 | static void | |
73371c09 | 4457 | hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, |
a339aa81 EJ |
4458 | struct hfsc_class *class) |
4459 | { | |
73371c09 | 4460 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 | 4461 | |
13c1637f | 4462 | uint32_t max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
a339aa81 | 4463 | if (!max_rate) { |
a00ca915 | 4464 | enum netdev_features current; |
a339aa81 | 4465 | |
73371c09 BP |
4466 | netdev_linux_read_features(netdev); |
4467 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 4468 | max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
a339aa81 EJ |
4469 | } |
4470 | ||
4471 | class->min_rate = max_rate; | |
4472 | class->max_rate = max_rate; | |
4473 | } | |
4474 | ||
4475 | static int | |
4476 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 4477 | const struct smap *details, |
a339aa81 EJ |
4478 | struct hfsc_class * class) |
4479 | { | |
4480 | const struct hfsc *hfsc; | |
4481 | uint32_t min_rate, max_rate; | |
a339aa81 EJ |
4482 | |
4483 | hfsc = hfsc_get__(netdev); | |
a339aa81 | 4484 | |
13c1637f | 4485 | min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
79398bad | 4486 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
4487 | min_rate = MIN(min_rate, hfsc->max_rate); |
4488 | ||
13c1637f | 4489 | max_rate = smap_get_ullong(details, "max-rate", hfsc->max_rate * 8) / 8; |
a339aa81 EJ |
4490 | max_rate = MAX(max_rate, min_rate); |
4491 | max_rate = MIN(max_rate, hfsc->max_rate); | |
4492 | ||
4493 | class->min_rate = min_rate; | |
4494 | class->max_rate = max_rate; | |
4495 | ||
4496 | return 0; | |
4497 | } | |
4498 | ||
4499 | /* Create an HFSC qdisc. | |
4500 | * | |
4501 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
4502 | static int | |
4503 | hfsc_setup_qdisc__(struct netdev * netdev) | |
4504 | { | |
4505 | struct tcmsg *tcmsg; | |
4506 | struct ofpbuf request; | |
4507 | struct tc_hfsc_qopt opt; | |
4508 | ||
4509 | tc_del_qdisc(netdev); | |
4510 | ||
7874bdff RD |
4511 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
4512 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
a339aa81 EJ |
4513 | |
4514 | if (!tcmsg) { | |
4515 | return ENODEV; | |
4516 | } | |
4517 | ||
4518 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
4519 | tcmsg->tcm_parent = TC_H_ROOT; | |
4520 | ||
4521 | memset(&opt, 0, sizeof opt); | |
4522 | opt.defcls = 1; | |
4523 | ||
4524 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4525 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
4526 | ||
4527 | return tc_transact(&request, NULL); | |
4528 | } | |
4529 | ||
4530 | /* Create an HFSC class. | |
4531 | * | |
4532 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
4533 | * sc rate <min_rate> ul rate <max_rate>" */ | |
4534 | static int | |
4535 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
4536 | unsigned int parent, struct hfsc_class *class) | |
4537 | { | |
4538 | int error; | |
4539 | size_t opt_offset; | |
4540 | struct tcmsg *tcmsg; | |
4541 | struct ofpbuf request; | |
4542 | struct tc_service_curve min, max; | |
4543 | ||
7874bdff RD |
4544 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
4545 | &request); | |
a339aa81 EJ |
4546 | |
4547 | if (!tcmsg) { | |
4548 | return ENODEV; | |
4549 | } | |
4550 | ||
4551 | tcmsg->tcm_handle = handle; | |
4552 | tcmsg->tcm_parent = parent; | |
4553 | ||
4554 | min.m1 = 0; | |
4555 | min.d = 0; | |
4556 | min.m2 = class->min_rate; | |
4557 | ||
4558 | max.m1 = 0; | |
4559 | max.d = 0; | |
4560 | max.m2 = class->max_rate; | |
4561 | ||
4562 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
4563 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4564 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
4565 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
4566 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
4567 | nl_msg_end_nested(&request, opt_offset); | |
4568 | ||
4569 | error = tc_transact(&request, NULL); | |
4570 | if (error) { | |
4571 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
4572 | "min-rate %ubps, max-rate %ubps (%s)", | |
4573 | netdev_get_name(netdev), | |
4574 | tc_get_major(handle), tc_get_minor(handle), | |
4575 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 4576 | class->min_rate, class->max_rate, ovs_strerror(error)); |
a339aa81 EJ |
4577 | } |
4578 | ||
4579 | return error; | |
4580 | } | |
4581 | ||
4582 | static int | |
79f1cbe9 | 4583 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4584 | { |
4585 | int error; | |
4586 | struct hfsc_class class; | |
4587 | ||
4588 | error = hfsc_setup_qdisc__(netdev); | |
4589 | ||
4590 | if (error) { | |
4591 | return error; | |
4592 | } | |
4593 | ||
4594 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4595 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4596 | tc_make_handle(1, 0), &class); | |
4597 | ||
4598 | if (error) { | |
4599 | return error; | |
4600 | } | |
4601 | ||
4602 | hfsc_install__(netdev, class.max_rate); | |
4603 | return 0; | |
4604 | } | |
4605 | ||
4606 | static int | |
4607 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4608 | { | |
4609 | struct ofpbuf msg; | |
d57695d7 | 4610 | struct queue_dump_state state; |
a339aa81 EJ |
4611 | struct hfsc_class hc; |
4612 | ||
4613 | hc.max_rate = 0; | |
4614 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 4615 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 | 4616 | |
d57695d7 | 4617 | if (!start_queue_dump(netdev, &state)) { |
a339aa81 EJ |
4618 | return ENODEV; |
4619 | } | |
4620 | ||
d57695d7 | 4621 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
a339aa81 EJ |
4622 | unsigned int queue_id; |
4623 | ||
4624 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
4625 | hfsc_update_queue__(netdev, queue_id, &hc); | |
4626 | } | |
4627 | } | |
4628 | ||
d57695d7 | 4629 | finish_queue_dump(&state); |
a339aa81 EJ |
4630 | return 0; |
4631 | } | |
4632 | ||
4633 | static void | |
4634 | hfsc_tc_destroy(struct tc *tc) | |
4635 | { | |
4636 | struct hfsc *hfsc; | |
4637 | struct hfsc_class *hc, *next; | |
4638 | ||
4639 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
4640 | ||
4641 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
4642 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4643 | free(hc); | |
4644 | } | |
4645 | ||
4646 | tc_destroy(tc); | |
4647 | free(hfsc); | |
4648 | } | |
4649 | ||
4650 | static int | |
79f1cbe9 | 4651 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
4652 | { |
4653 | const struct hfsc *hfsc; | |
4654 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 4655 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
4656 | return 0; |
4657 | } | |
4658 | ||
4659 | static int | |
79f1cbe9 | 4660 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
4661 | { |
4662 | int error; | |
4663 | struct hfsc_class class; | |
4664 | ||
4665 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
4666 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4667 | tc_make_handle(1, 0), &class); | |
4668 | ||
4669 | if (!error) { | |
4670 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
4671 | } | |
4672 | ||
4673 | return error; | |
4674 | } | |
4675 | ||
4676 | static int | |
4677 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 4678 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
4679 | { |
4680 | const struct hfsc_class *hc; | |
4681 | ||
4682 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 4683 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 4684 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4685 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
4686 | } |
4687 | return 0; | |
4688 | } | |
4689 | ||
4690 | static int | |
4691 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4692 | const struct smap *details) |
a339aa81 EJ |
4693 | { |
4694 | int error; | |
4695 | struct hfsc_class class; | |
4696 | ||
4697 | error = hfsc_parse_class_details__(netdev, details, &class); | |
4698 | if (error) { | |
4699 | return error; | |
4700 | } | |
4701 | ||
4702 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
4703 | tc_make_handle(1, 0xfffe), &class); | |
4704 | if (error) { | |
4705 | return error; | |
4706 | } | |
4707 | ||
4708 | hfsc_update_queue__(netdev, queue_id, &class); | |
4709 | return 0; | |
4710 | } | |
4711 | ||
4712 | static int | |
4713 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
4714 | { | |
4715 | int error; | |
4716 | struct hfsc *hfsc; | |
4717 | struct hfsc_class *hc; | |
4718 | ||
4719 | hc = hfsc_class_cast__(queue); | |
4720 | hfsc = hfsc_get__(netdev); | |
4721 | ||
4722 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
4723 | if (!error) { | |
4724 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
4725 | free(hc); | |
4726 | } | |
4727 | return error; | |
4728 | } | |
4729 | ||
4730 | static int | |
4731 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
4732 | struct netdev_queue_stats *stats) | |
4733 | { | |
4734 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
4735 | tc_make_handle(1, 0xfffe), NULL, stats); | |
4736 | } | |
4737 | ||
4738 | static int | |
4739 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4740 | const struct ofpbuf *nlmsg, | |
4741 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4742 | { | |
4743 | struct netdev_queue_stats stats; | |
4744 | unsigned int handle, major, minor; | |
4745 | int error; | |
4746 | ||
4747 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4748 | if (error) { | |
4749 | return error; | |
4750 | } | |
4751 | ||
4752 | major = tc_get_major(handle); | |
4753 | minor = tc_get_minor(handle); | |
4754 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
4755 | (*cb)(minor - 1, &stats, aux); | |
4756 | } | |
4757 | return 0; | |
4758 | } | |
4759 | ||
4760 | static const struct tc_ops tc_ops_hfsc = { | |
4761 | "hfsc", /* linux_name */ | |
4762 | "linux-hfsc", /* ovs_name */ | |
4763 | HFSC_N_QUEUES, /* n_queues */ | |
4764 | hfsc_tc_install, /* tc_install */ | |
4765 | hfsc_tc_load, /* tc_load */ | |
4766 | hfsc_tc_destroy, /* tc_destroy */ | |
4767 | hfsc_qdisc_get, /* qdisc_get */ | |
4768 | hfsc_qdisc_set, /* qdisc_set */ | |
4769 | hfsc_class_get, /* class_get */ | |
4770 | hfsc_class_set, /* class_set */ | |
4771 | hfsc_class_delete, /* class_delete */ | |
4772 | hfsc_class_get_stats, /* class_get_stats */ | |
4773 | hfsc_class_dump_stats /* class_dump_stats */ | |
4774 | }; | |
4775 | \f | |
6cf888b8 BS |
4776 | /* "linux-noop" traffic control class. */ |
4777 | ||
4778 | static void | |
4779 | noop_install__(struct netdev *netdev_) | |
4780 | { | |
4781 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4782 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); | |
4783 | ||
4784 | netdev->tc = CONST_CAST(struct tc *, &tc); | |
4785 | } | |
4786 | ||
4787 | static int | |
4788 | noop_tc_install(struct netdev *netdev, | |
4789 | const struct smap *details OVS_UNUSED) | |
4790 | { | |
4791 | noop_install__(netdev); | |
4792 | return 0; | |
4793 | } | |
4794 | ||
4795 | static int | |
4796 | noop_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4797 | { | |
4798 | noop_install__(netdev); | |
4799 | return 0; | |
4800 | } | |
4801 | ||
4802 | static const struct tc_ops tc_ops_noop = { | |
4803 | NULL, /* linux_name */ | |
4804 | "linux-noop", /* ovs_name */ | |
4805 | 0, /* n_queues */ | |
4806 | noop_tc_install, | |
4807 | noop_tc_load, | |
4808 | NULL, /* tc_destroy */ | |
4809 | NULL, /* qdisc_get */ | |
4810 | NULL, /* qdisc_set */ | |
4811 | NULL, /* class_get */ | |
4812 | NULL, /* class_set */ | |
4813 | NULL, /* class_delete */ | |
4814 | NULL, /* class_get_stats */ | |
4815 | NULL /* class_dump_stats */ | |
4816 | }; | |
4817 | \f | |
c1c9c9c4 BP |
4818 | /* "linux-default" traffic control class. |
4819 | * | |
4820 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
4821 | * the "" (empty string) QoS type in the OVS database. */ | |
4822 | ||
4823 | static void | |
b5d57fc8 | 4824 | default_install__(struct netdev *netdev_) |
c1c9c9c4 | 4825 | { |
b5d57fc8 | 4826 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4827 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); |
c1c9c9c4 | 4828 | |
559eb230 BP |
4829 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4830 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4831 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4832 | } |
4833 | ||
4834 | static int | |
4835 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 4836 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
4837 | { |
4838 | default_install__(netdev); | |
4839 | return 0; | |
4840 | } | |
4841 | ||
4842 | static int | |
4843 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4844 | { | |
4845 | default_install__(netdev); | |
4846 | return 0; | |
4847 | } | |
4848 | ||
4849 | static const struct tc_ops tc_ops_default = { | |
4850 | NULL, /* linux_name */ | |
4851 | "", /* ovs_name */ | |
4852 | 0, /* n_queues */ | |
4853 | default_tc_install, | |
4854 | default_tc_load, | |
4855 | NULL, /* tc_destroy */ | |
4856 | NULL, /* qdisc_get */ | |
4857 | NULL, /* qdisc_set */ | |
4858 | NULL, /* class_get */ | |
4859 | NULL, /* class_set */ | |
4860 | NULL, /* class_delete */ | |
4861 | NULL, /* class_get_stats */ | |
4862 | NULL /* class_dump_stats */ | |
4863 | }; | |
4864 | \f | |
4865 | /* "linux-other" traffic control class. | |
4866 | * | |
4867 | * */ | |
4868 | ||
4869 | static int | |
b5d57fc8 | 4870 | other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) |
c1c9c9c4 | 4871 | { |
b5d57fc8 | 4872 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 4873 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); |
c1c9c9c4 | 4874 | |
559eb230 BP |
4875 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
4876 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 4877 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
4878 | return 0; |
4879 | } | |
4880 | ||
4881 | static const struct tc_ops tc_ops_other = { | |
4882 | NULL, /* linux_name */ | |
4883 | "linux-other", /* ovs_name */ | |
4884 | 0, /* n_queues */ | |
4885 | NULL, /* tc_install */ | |
4886 | other_tc_load, | |
4887 | NULL, /* tc_destroy */ | |
4888 | NULL, /* qdisc_get */ | |
4889 | NULL, /* qdisc_set */ | |
4890 | NULL, /* class_get */ | |
4891 | NULL, /* class_set */ | |
4892 | NULL, /* class_delete */ | |
4893 | NULL, /* class_get_stats */ | |
4894 | NULL /* class_dump_stats */ | |
4895 | }; | |
4896 | \f | |
4897 | /* Traffic control. */ | |
4898 | ||
4899 | /* Number of kernel "tc" ticks per second. */ | |
4900 | static double ticks_per_s; | |
4901 | ||
4902 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
4903 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
4904 | * one jiffy's worth of data. | |
4905 | * | |
4906 | * There are two possibilities here: | |
4907 | * | |
4908 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
4909 | * approximate range of 100 to 1024. That means that we really need to | |
4910 | * make sure that the qdisc can buffer that much data. | |
4911 | * | |
4912 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
4913 | * has finely granular timers and there's no need to fudge additional room | |
4914 | * for buffers. (There's no extra effort needed to implement that: the | |
4915 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
4916 | * come out as 0 in the division. Small integer results in the case of | |
4917 | * really high dividends won't have any real effect anyhow.) | |
4918 | */ | |
4919 | static unsigned int buffer_hz; | |
4920 | ||
7874bdff RD |
4921 | static struct tcmsg * |
4922 | netdev_linux_tc_make_request(const struct netdev *netdev, int type, | |
4923 | unsigned int flags, struct ofpbuf *request) | |
4924 | { | |
4925 | int ifindex; | |
4926 | int error; | |
4927 | ||
4928 | error = get_ifindex(netdev, &ifindex); | |
4929 | if (error) { | |
4930 | return NULL; | |
4931 | } | |
4932 | ||
4933 | return tc_make_request(ifindex, type, flags, request); | |
4934 | } | |
4935 | ||
f8500004 JP |
4936 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size |
4937 | * of 'kbits_burst'. | |
4938 | * | |
4939 | * This function is equivalent to running: | |
4940 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
4941 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
4942 | * mtu 65535 drop | |
4943 | * | |
4944 | * The configuration and stats may be seen with the following command: | |
c7952afb | 4945 | * /sbin/tc -s filter show dev <devname> parent ffff: |
f8500004 JP |
4946 | * |
4947 | * Returns 0 if successful, otherwise a positive errno value. | |
4948 | */ | |
4949 | static int | |
c7952afb BP |
4950 | tc_add_policer(struct netdev *netdev, |
4951 | uint32_t kbits_rate, uint32_t kbits_burst) | |
f8500004 JP |
4952 | { |
4953 | struct tc_police tc_police; | |
4954 | struct ofpbuf request; | |
4955 | struct tcmsg *tcmsg; | |
4956 | size_t basic_offset; | |
4957 | size_t police_offset; | |
4958 | int error; | |
4959 | int mtu = 65535; | |
4960 | ||
4961 | memset(&tc_police, 0, sizeof tc_police); | |
4962 | tc_police.action = TC_POLICE_SHOT; | |
4963 | tc_police.mtu = mtu; | |
1aca400c | 4964 | tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu); |
c7952afb | 4965 | |
79abacc8 MAA |
4966 | /* The following appears wrong in one way: In networking a kilobit is |
4967 | * usually 1000 bits but this uses 1024 bits. | |
c7952afb BP |
4968 | * |
4969 | * However if you "fix" those problems then "tc filter show ..." shows | |
4970 | * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit == | |
4971 | * 1,000,000 bits, whereas this actually ends up doing the right thing from | |
4972 | * tc's point of view. Whatever. */ | |
4973 | tc_police.burst = tc_bytes_to_ticks( | |
79abacc8 | 4974 | tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 8); |
f8500004 | 4975 | |
7874bdff RD |
4976 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTFILTER, |
4977 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
f8500004 JP |
4978 | if (!tcmsg) { |
4979 | return ENODEV; | |
4980 | } | |
4981 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
4982 | tcmsg->tcm_info = tc_make_handle(49, | |
4983 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
4984 | ||
4985 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
4986 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4987 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
4988 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
4989 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
4990 | nl_msg_end_nested(&request, police_offset); | |
4991 | nl_msg_end_nested(&request, basic_offset); | |
4992 | ||
4993 | error = tc_transact(&request, NULL); | |
4994 | if (error) { | |
4995 | return error; | |
4996 | } | |
4997 | ||
4998 | return 0; | |
4999 | } | |
5000 | ||
c1c9c9c4 BP |
5001 | static void |
5002 | read_psched(void) | |
5003 | { | |
5004 | /* The values in psched are not individually very meaningful, but they are | |
5005 | * important. The tables below show some values seen in the wild. | |
5006 | * | |
5007 | * Some notes: | |
5008 | * | |
5009 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
5010 | * (Before that, there are hints that it was 1000000000.) | |
5011 | * | |
5012 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
5013 | * above. | |
5014 | * | |
5015 | * /proc/net/psched | |
5016 | * ----------------------------------- | |
5017 | * [1] 000c8000 000f4240 000f4240 00000064 | |
5018 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
5019 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
5020 | * [4] 000003e8 00000400 000f4240 00000064 | |
5021 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
5022 | * [6] 000003e8 00000040 000f4240 000000f9 | |
5023 | * | |
5024 | * a b c d ticks_per_s buffer_hz | |
5025 | * ------- --------- ---------- ------------- ----------- ------------- | |
5026 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
5027 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
5028 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
5029 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
5030 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
5031 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
5032 | * | |
5033 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
5034 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
5035 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
5036 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
5037 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
5038 | * [6] 2.6.34 from kernel.org on KVM | |
5039 | */ | |
23882115 | 5040 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
c1c9c9c4 BP |
5041 | static const char fn[] = "/proc/net/psched"; |
5042 | unsigned int a, b, c, d; | |
5043 | FILE *stream; | |
5044 | ||
23882115 BP |
5045 | if (!ovsthread_once_start(&once)) { |
5046 | return; | |
5047 | } | |
5048 | ||
c1c9c9c4 BP |
5049 | ticks_per_s = 1.0; |
5050 | buffer_hz = 100; | |
5051 | ||
5052 | stream = fopen(fn, "r"); | |
5053 | if (!stream) { | |
10a89ef0 | 5054 | VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); |
23882115 | 5055 | goto exit; |
c1c9c9c4 BP |
5056 | } |
5057 | ||
5058 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
5059 | VLOG_WARN("%s: read failed", fn); | |
5060 | fclose(stream); | |
23882115 | 5061 | goto exit; |
c1c9c9c4 BP |
5062 | } |
5063 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
5064 | fclose(stream); | |
5065 | ||
5066 | if (!a || !c) { | |
5067 | VLOG_WARN("%s: invalid scheduler parameters", fn); | |
23882115 | 5068 | goto exit; |
c1c9c9c4 BP |
5069 | } |
5070 | ||
5071 | ticks_per_s = (double) a * c / b; | |
5072 | if (c == 1000000) { | |
5073 | buffer_hz = d; | |
5074 | } else { | |
5075 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
5076 | fn, a, b, c, d); | |
5077 | } | |
5078 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
23882115 BP |
5079 | |
5080 | exit: | |
5081 | ovsthread_once_done(&once); | |
c1c9c9c4 BP |
5082 | } |
5083 | ||
5084 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
5085 | * rate of 'rate' bytes per second. */ | |
5086 | static unsigned int | |
5087 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
5088 | { | |
23882115 | 5089 | read_psched(); |
c1c9c9c4 BP |
5090 | return (rate * ticks) / ticks_per_s; |
5091 | } | |
5092 | ||
5093 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
5094 | * rate of 'rate' bytes per second. */ | |
5095 | static unsigned int | |
5096 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
5097 | { | |
23882115 | 5098 | read_psched(); |
015c93a4 | 5099 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
5100 | } |
5101 | ||
5102 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
5103 | * a transmission rate of 'rate' bytes per second. */ | |
5104 | static unsigned int | |
5105 | tc_buffer_per_jiffy(unsigned int rate) | |
5106 | { | |
23882115 | 5107 | read_psched(); |
c1c9c9c4 BP |
5108 | return rate / buffer_hz; |
5109 | } | |
5110 | ||
5111 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, | |
5112 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
5113 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
5114 | * stores NULL into it if it is absent. | |
5115 | * | |
5116 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
5117 | * 'msg'. | |
5118 | * | |
5119 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5120 | static int | |
5121 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
5122 | struct nlattr **options) | |
5123 | { | |
5124 | static const struct nl_policy tca_policy[] = { | |
5125 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
5126 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
5127 | }; | |
5128 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
5129 | ||
5130 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
5131 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
5132 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
5133 | goto error; | |
5134 | } | |
5135 | ||
5136 | if (kind) { | |
5137 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
5138 | } | |
5139 | ||
5140 | if (options) { | |
5141 | *options = ta[TCA_OPTIONS]; | |
5142 | } | |
5143 | ||
5144 | return 0; | |
5145 | ||
5146 | error: | |
5147 | if (kind) { | |
5148 | *kind = NULL; | |
5149 | } | |
5150 | if (options) { | |
5151 | *options = NULL; | |
5152 | } | |
5153 | return EPROTO; | |
5154 | } | |
5155 | ||
5156 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
5157 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
5158 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
5159 | * arguments may be null. | |
5160 | * | |
5161 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5162 | static int | |
5163 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
5164 | struct nlattr **options, struct netdev_queue_stats *stats) | |
5165 | { | |
5166 | static const struct nl_policy tca_policy[] = { | |
5167 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
5168 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
5169 | }; | |
5170 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
5171 | ||
5172 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
5173 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
5174 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
5175 | goto error; | |
5176 | } | |
5177 | ||
5178 | if (handlep) { | |
5179 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
5180 | *handlep = tc->tcm_handle; | |
5181 | } | |
5182 | ||
5183 | if (options) { | |
5184 | *options = ta[TCA_OPTIONS]; | |
5185 | } | |
5186 | ||
5187 | if (stats) { | |
5188 | const struct gnet_stats_queue *gsq; | |
5189 | struct gnet_stats_basic gsb; | |
5190 | ||
5191 | static const struct nl_policy stats_policy[] = { | |
5192 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
5193 | .min_len = sizeof gsb }, | |
5194 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
5195 | .min_len = sizeof *gsq }, | |
5196 | }; | |
5197 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
5198 | ||
5199 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
5200 | sa, ARRAY_SIZE(sa))) { | |
5201 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
5202 | goto error; | |
5203 | } | |
5204 | ||
5205 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
5206 | * some arch/bitsize combinations. Newer versions of Linux have a | |
5207 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
5208 | * easiest thing to do is just to make a copy. */ | |
5209 | memset(&gsb, 0, sizeof gsb); | |
5210 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
5211 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
5212 | stats->tx_bytes = gsb.bytes; | |
5213 | stats->tx_packets = gsb.packets; | |
5214 | ||
5215 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
5216 | stats->tx_errors = gsq->drops; | |
5217 | } | |
5218 | ||
5219 | return 0; | |
5220 | ||
5221 | error: | |
5222 | if (options) { | |
5223 | *options = NULL; | |
5224 | } | |
5225 | if (stats) { | |
5226 | memset(stats, 0, sizeof *stats); | |
5227 | } | |
5228 | return EPROTO; | |
5229 | } | |
5230 | ||
5231 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
5232 | * on 'netdev'. */ | |
5233 | static int | |
5234 | tc_query_class(const struct netdev *netdev, | |
5235 | unsigned int handle, unsigned int parent, | |
5236 | struct ofpbuf **replyp) | |
5237 | { | |
5238 | struct ofpbuf request; | |
5239 | struct tcmsg *tcmsg; | |
5240 | int error; | |
5241 | ||
7874bdff RD |
5242 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, |
5243 | &request); | |
23a98ffe BP |
5244 | if (!tcmsg) { |
5245 | return ENODEV; | |
5246 | } | |
c1c9c9c4 BP |
5247 | tcmsg->tcm_handle = handle; |
5248 | tcmsg->tcm_parent = parent; | |
5249 | ||
5250 | error = tc_transact(&request, replyp); | |
5251 | if (error) { | |
5252 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
5253 | netdev_get_name(netdev), | |
5254 | tc_get_major(handle), tc_get_minor(handle), | |
5255 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 5256 | ovs_strerror(error)); |
c1c9c9c4 BP |
5257 | } |
5258 | return error; | |
5259 | } | |
5260 | ||
5261 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
5262 | static int | |
5263 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
5264 | { | |
5265 | struct ofpbuf request; | |
5266 | struct tcmsg *tcmsg; | |
5267 | int error; | |
5268 | ||
7874bdff | 5269 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_DELTCLASS, 0, &request); |
23a98ffe BP |
5270 | if (!tcmsg) { |
5271 | return ENODEV; | |
5272 | } | |
c1c9c9c4 BP |
5273 | tcmsg->tcm_handle = handle; |
5274 | tcmsg->tcm_parent = 0; | |
5275 | ||
5276 | error = tc_transact(&request, NULL); | |
5277 | if (error) { | |
5278 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
5279 | netdev_get_name(netdev), | |
5280 | tc_get_major(handle), tc_get_minor(handle), | |
10a89ef0 | 5281 | ovs_strerror(error)); |
c1c9c9c4 BP |
5282 | } |
5283 | return error; | |
5284 | } | |
5285 | ||
5286 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
5287 | static int | |
b5d57fc8 | 5288 | tc_del_qdisc(struct netdev *netdev_) |
c1c9c9c4 | 5289 | { |
b5d57fc8 | 5290 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5291 | struct ofpbuf request; |
5292 | struct tcmsg *tcmsg; | |
5293 | int error; | |
5294 | ||
7874bdff | 5295 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
5296 | if (!tcmsg) { |
5297 | return ENODEV; | |
5298 | } | |
c1c9c9c4 BP |
5299 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
5300 | tcmsg->tcm_parent = TC_H_ROOT; | |
5301 | ||
5302 | error = tc_transact(&request, NULL); | |
5303 | if (error == EINVAL) { | |
5304 | /* EINVAL probably means that the default qdisc was in use, in which | |
5305 | * case we've accomplished our purpose. */ | |
5306 | error = 0; | |
5307 | } | |
b5d57fc8 BP |
5308 | if (!error && netdev->tc) { |
5309 | if (netdev->tc->ops->tc_destroy) { | |
5310 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 | 5311 | } |
b5d57fc8 | 5312 | netdev->tc = NULL; |
c1c9c9c4 BP |
5313 | } |
5314 | return error; | |
5315 | } | |
5316 | ||
ac3e3aaa BP |
5317 | static bool |
5318 | getqdisc_is_safe(void) | |
5319 | { | |
5320 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
5321 | static bool safe = false; | |
5322 | ||
5323 | if (ovsthread_once_start(&once)) { | |
5324 | struct utsname utsname; | |
5325 | int major, minor; | |
5326 | ||
5327 | if (uname(&utsname) == -1) { | |
5328 | VLOG_WARN("uname failed (%s)", ovs_strerror(errno)); | |
5329 | } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) { | |
5330 | VLOG_WARN("uname reported bad OS release (%s)", utsname.release); | |
5331 | } else if (major < 2 || (major == 2 && minor < 35)) { | |
5332 | VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s", | |
5333 | utsname.release); | |
5334 | } else { | |
5335 | safe = true; | |
5336 | } | |
5337 | ovsthread_once_done(&once); | |
5338 | } | |
5339 | return safe; | |
5340 | } | |
5341 | ||
c1c9c9c4 BP |
5342 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the |
5343 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
5344 | * positive errno value. */ | |
5345 | static int | |
b5d57fc8 | 5346 | tc_query_qdisc(const struct netdev *netdev_) |
c1c9c9c4 | 5347 | { |
b5d57fc8 | 5348 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5349 | struct ofpbuf request, *qdisc; |
5350 | const struct tc_ops *ops; | |
5351 | struct tcmsg *tcmsg; | |
5352 | int load_error; | |
5353 | int error; | |
5354 | ||
b5d57fc8 | 5355 | if (netdev->tc) { |
c1c9c9c4 BP |
5356 | return 0; |
5357 | } | |
5358 | ||
5359 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
5360 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
5361 | * 2.6.35 without that fix backported to it. | |
5362 | * | |
5363 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
5364 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
5365 | * few others. There are a few ways that I can see to do this, but most of | |
5366 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
5367 | * technique chosen here is to assume that any non-default qdisc that we | |
5368 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
5369 | * a class with handle 0:0. | |
5370 | * | |
ac3e3aaa BP |
5371 | * On Linux 2.6.35+ we use the straightforward method because it allows us |
5372 | * to handle non-builtin qdiscs without handle 1:0 (e.g. codel). However, | |
5373 | * in such a case we get no response at all from the kernel (!) if a | |
5374 | * builtin qdisc is in use (which is later caught by "!error && | |
5375 | * !qdisc->size"). */ | |
7874bdff RD |
5376 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, |
5377 | &request); | |
23a98ffe BP |
5378 | if (!tcmsg) { |
5379 | return ENODEV; | |
5380 | } | |
ac3e3aaa BP |
5381 | tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0); |
5382 | tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0; | |
c1c9c9c4 BP |
5383 | |
5384 | /* Figure out what tc class to instantiate. */ | |
5385 | error = tc_transact(&request, &qdisc); | |
ac3e3aaa | 5386 | if (!error && qdisc->size) { |
c1c9c9c4 BP |
5387 | const char *kind; |
5388 | ||
5389 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
5390 | if (error) { | |
5391 | ops = &tc_ops_other; | |
5392 | } else { | |
5393 | ops = tc_lookup_linux_name(kind); | |
5394 | if (!ops) { | |
5395 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
ac3e3aaa | 5396 | VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind); |
c1c9c9c4 BP |
5397 | |
5398 | ops = &tc_ops_other; | |
5399 | } | |
5400 | } | |
ac3e3aaa BP |
5401 | } else if ((!error && !qdisc->size) || error == ENOENT) { |
5402 | /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc | |
5403 | * set up by some other entity that doesn't have a handle 1:0. We will | |
5404 | * assume that it's the system default qdisc. */ | |
c1c9c9c4 BP |
5405 | ops = &tc_ops_default; |
5406 | error = 0; | |
5407 | } else { | |
5408 | /* Who knows? Maybe the device got deleted. */ | |
5409 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
10a89ef0 | 5410 | netdev_get_name(netdev_), ovs_strerror(error)); |
c1c9c9c4 BP |
5411 | ops = &tc_ops_other; |
5412 | } | |
5413 | ||
5414 | /* Instantiate it. */ | |
b5d57fc8 BP |
5415 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); |
5416 | ovs_assert((load_error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
5417 | ofpbuf_delete(qdisc); |
5418 | ||
5419 | return error ? error : load_error; | |
5420 | } | |
5421 | ||
5422 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
5423 | approximate the time to transmit packets of various lengths. For an MTU of | |
5424 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
5425 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
5426 | possible lengths; and so on. | |
5427 | ||
5428 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
5429 | need to be shifted right to fit within such a 256-entry table. */ | |
5430 | static int | |
5431 | tc_calc_cell_log(unsigned int mtu) | |
5432 | { | |
5433 | int cell_log; | |
5434 | ||
5435 | if (!mtu) { | |
5436 | mtu = ETH_PAYLOAD_MAX; | |
5437 | } | |
5438 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
5439 | ||
5440 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
5441 | mtu >>= 1; | |
5442 | } | |
5443 | ||
5444 | return cell_log; | |
5445 | } | |
5446 | ||
5447 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
5448 | * of 'mtu'. */ | |
5449 | static void | |
5450 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
5451 | { | |
5452 | memset(rate, 0, sizeof *rate); | |
5453 | rate->cell_log = tc_calc_cell_log(mtu); | |
5454 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
5455 | /* rate->cell_align = 0; */ /* distro headers. */ | |
5456 | rate->mpu = ETH_TOTAL_MIN; | |
5457 | rate->rate = Bps; | |
5458 | } | |
5459 | ||
5460 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
5461 | * attribute of the specified "type". | |
5462 | * | |
5463 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
5464 | static void | |
5465 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) | |
5466 | { | |
5467 | uint32_t *rtab; | |
5468 | unsigned int i; | |
5469 | ||
5470 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
5471 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
5472 | unsigned packet_size = (i + 1) << rate->cell_log; | |
5473 | if (packet_size < rate->mpu) { | |
5474 | packet_size = rate->mpu; | |
5475 | } | |
5476 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
5477 | } | |
5478 | } | |
5479 | ||
5480 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
5481 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
5482 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 5483 | * 0 is fine.) */ |
c1c9c9c4 BP |
5484 | static int |
5485 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
5486 | { | |
5487 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
5488 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
5489 | } | |
d3980822 | 5490 | \f |
aaf2fb1a BP |
5491 | /* Linux-only functions declared in netdev-linux.h */ |
5492 | ||
5493 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If | |
5494 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
5495 | int | |
5496 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
5497 | const char *flag_name, bool enable) | |
5498 | { | |
5499 | const char *netdev_name = netdev_get_name(netdev); | |
5500 | struct ethtool_value evalue; | |
5501 | uint32_t new_flags; | |
5502 | int error; | |
5503 | ||
ab985a77 | 5504 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5505 | memset(&evalue, 0, sizeof evalue); |
5506 | error = netdev_linux_do_ethtool(netdev_name, | |
5507 | (struct ethtool_cmd *)&evalue, | |
5508 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5509 | if (error) { | |
5510 | return error; | |
5511 | } | |
5512 | ||
ab985a77 | 5513 | COVERAGE_INC(netdev_set_ethtool); |
ad2dade5 AS |
5514 | new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
5515 | if (new_flags == evalue.data) { | |
5516 | return 0; | |
5517 | } | |
5518 | evalue.data = new_flags; | |
aaf2fb1a BP |
5519 | error = netdev_linux_do_ethtool(netdev_name, |
5520 | (struct ethtool_cmd *)&evalue, | |
5521 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
5522 | if (error) { | |
5523 | return error; | |
5524 | } | |
5525 | ||
ab985a77 | 5526 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
5527 | memset(&evalue, 0, sizeof evalue); |
5528 | error = netdev_linux_do_ethtool(netdev_name, | |
5529 | (struct ethtool_cmd *)&evalue, | |
5530 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
5531 | if (error) { | |
5532 | return error; | |
5533 | } | |
5534 | ||
5535 | if (new_flags != evalue.data) { | |
5536 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
5537 | "device %s failed", enable ? "enable" : "disable", | |
5538 | flag_name, netdev_name); | |
5539 | return EOPNOTSUPP; | |
5540 | } | |
5541 | ||
5542 | return 0; | |
5543 | } | |
5544 | \f | |
5545 | /* Utility functions. */ | |
5546 | ||
d3980822 | 5547 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 5548 | static void |
d3980822 BP |
5549 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
5550 | const struct rtnl_link_stats *src) | |
5551 | { | |
f613a0d7 PS |
5552 | dst->rx_packets = src->rx_packets; |
5553 | dst->tx_packets = src->tx_packets; | |
5554 | dst->rx_bytes = src->rx_bytes; | |
5555 | dst->tx_bytes = src->tx_bytes; | |
5556 | dst->rx_errors = src->rx_errors; | |
5557 | dst->tx_errors = src->tx_errors; | |
5558 | dst->rx_dropped = src->rx_dropped; | |
5559 | dst->tx_dropped = src->tx_dropped; | |
5560 | dst->multicast = src->multicast; | |
5561 | dst->collisions = src->collisions; | |
5562 | dst->rx_length_errors = src->rx_length_errors; | |
5563 | dst->rx_over_errors = src->rx_over_errors; | |
5564 | dst->rx_crc_errors = src->rx_crc_errors; | |
5565 | dst->rx_frame_errors = src->rx_frame_errors; | |
5566 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5567 | dst->rx_missed_errors = src->rx_missed_errors; | |
5568 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5569 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5570 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5571 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5572 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
5573 | } |
5574 | ||
337c9b99 BP |
5575 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
5576 | static void | |
5577 | netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst, | |
5578 | const struct rtnl_link_stats64 *src) | |
5579 | { | |
5580 | dst->rx_packets = src->rx_packets; | |
5581 | dst->tx_packets = src->tx_packets; | |
5582 | dst->rx_bytes = src->rx_bytes; | |
5583 | dst->tx_bytes = src->tx_bytes; | |
5584 | dst->rx_errors = src->rx_errors; | |
5585 | dst->tx_errors = src->tx_errors; | |
5586 | dst->rx_dropped = src->rx_dropped; | |
5587 | dst->tx_dropped = src->tx_dropped; | |
5588 | dst->multicast = src->multicast; | |
5589 | dst->collisions = src->collisions; | |
5590 | dst->rx_length_errors = src->rx_length_errors; | |
5591 | dst->rx_over_errors = src->rx_over_errors; | |
5592 | dst->rx_crc_errors = src->rx_crc_errors; | |
5593 | dst->rx_frame_errors = src->rx_frame_errors; | |
5594 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
5595 | dst->rx_missed_errors = src->rx_missed_errors; | |
5596 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
5597 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
5598 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
5599 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
5600 | dst->tx_window_errors = src->tx_window_errors; | |
5601 | } | |
5602 | ||
c1c9c9c4 | 5603 | static int |
35eef899 | 5604 | get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) |
c1c9c9c4 | 5605 | { |
c1c9c9c4 BP |
5606 | struct ofpbuf request; |
5607 | struct ofpbuf *reply; | |
c1c9c9c4 BP |
5608 | int error; |
5609 | ||
d6e3feb5 | 5610 | /* Filtering all counters by default */ |
5611 | memset(stats, 0xFF, sizeof(struct netdev_stats)); | |
5612 | ||
c1c9c9c4 | 5613 | ofpbuf_init(&request, 0); |
13a24df8 BP |
5614 | nl_msg_put_nlmsghdr(&request, |
5615 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
5616 | RTM_GETLINK, NLM_F_REQUEST); | |
5617 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
5618 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); | |
a88b4e04 | 5619 | error = nl_transact(NETLINK_ROUTE, &request, &reply); |
c1c9c9c4 BP |
5620 | ofpbuf_uninit(&request); |
5621 | if (error) { | |
5622 | return error; | |
5623 | } | |
5624 | ||
13a24df8 | 5625 | if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { |
337c9b99 BP |
5626 | const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64); |
5627 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) { | |
5628 | netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a)); | |
13a24df8 BP |
5629 | error = 0; |
5630 | } else { | |
71f21279 | 5631 | a = nl_attr_find(reply, 0, IFLA_STATS); |
337c9b99 BP |
5632 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { |
5633 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); | |
5634 | error = 0; | |
5635 | } else { | |
5636 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
5637 | error = EPROTO; | |
5638 | } | |
13a24df8 BP |
5639 | } |
5640 | } else { | |
5641 | VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); | |
5642 | error = EPROTO; | |
c1c9c9c4 | 5643 | } |
8b61709d | 5644 | |
8b61709d | 5645 | |
576e26d7 | 5646 | ofpbuf_delete(reply); |
35eef899 | 5647 | return error; |
8b61709d | 5648 | } |
c1c9c9c4 | 5649 | |
3a183124 | 5650 | static int |
b5d57fc8 | 5651 | get_flags(const struct netdev *dev, unsigned int *flags) |
8b61709d BP |
5652 | { |
5653 | struct ifreq ifr; | |
5654 | int error; | |
5655 | ||
755be9ea | 5656 | *flags = 0; |
259e0b1a | 5657 | error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); |
755be9ea EJ |
5658 | if (!error) { |
5659 | *flags = ifr.ifr_flags; | |
5660 | } | |
8b61709d BP |
5661 | return error; |
5662 | } | |
5663 | ||
5664 | static int | |
4b609110 | 5665 | set_flags(const char *name, unsigned int flags) |
8b61709d BP |
5666 | { |
5667 | struct ifreq ifr; | |
5668 | ||
5669 | ifr.ifr_flags = flags; | |
259e0b1a | 5670 | return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); |
8b61709d BP |
5671 | } |
5672 | ||
01b25786 PB |
5673 | int |
5674 | linux_get_ifindex(const char *netdev_name) | |
8b61709d BP |
5675 | { |
5676 | struct ifreq ifr; | |
259e0b1a | 5677 | int error; |
8b61709d | 5678 | |
71d7c22f | 5679 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5680 | COVERAGE_INC(netdev_get_ifindex); |
259e0b1a BP |
5681 | |
5682 | error = af_inet_ioctl(SIOCGIFINDEX, &ifr); | |
5683 | if (error) { | |
580e1152 RD |
5684 | /* ENODEV probably means that a vif disappeared asynchronously and |
5685 | * hasn't been removed from the database yet, so reduce the log level | |
5686 | * to INFO for that case. */ | |
5687 | VLOG_RL(&rl, error == ENODEV ? VLL_INFO : VLL_ERR, | |
5688 | "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
5689 | netdev_name, ovs_strerror(error)); | |
259e0b1a | 5690 | return -error; |
8b61709d BP |
5691 | } |
5692 | return ifr.ifr_ifindex; | |
5693 | } | |
5694 | ||
5695 | static int | |
5696 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
5697 | { | |
b5d57fc8 | 5698 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c7b1b0a5 | 5699 | |
b5d57fc8 | 5700 | if (!(netdev->cache_valid & VALID_IFINDEX)) { |
756819dd FL |
5701 | netdev_linux_update_via_netlink(netdev); |
5702 | } | |
5703 | ||
5704 | if (!(netdev->cache_valid & VALID_IFINDEX)) { | |
5705 | /* Fall back to ioctl if netlink fails */ | |
01b25786 | 5706 | int ifindex = linux_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 5707 | |
8b61709d | 5708 | if (ifindex < 0) { |
b5d57fc8 BP |
5709 | netdev->get_ifindex_error = -ifindex; |
5710 | netdev->ifindex = 0; | |
c7b1b0a5 | 5711 | } else { |
b5d57fc8 BP |
5712 | netdev->get_ifindex_error = 0; |
5713 | netdev->ifindex = ifindex; | |
8b61709d | 5714 | } |
b5d57fc8 | 5715 | netdev->cache_valid |= VALID_IFINDEX; |
8b61709d | 5716 | } |
c7b1b0a5 | 5717 | |
b5d57fc8 BP |
5718 | *ifindexp = netdev->ifindex; |
5719 | return netdev->get_ifindex_error; | |
8b61709d BP |
5720 | } |
5721 | ||
5722 | static int | |
756819dd FL |
5723 | netdev_linux_update_via_netlink(struct netdev_linux *netdev) |
5724 | { | |
5725 | struct ofpbuf request; | |
5726 | struct ofpbuf *reply; | |
5727 | struct rtnetlink_change chg; | |
5728 | struct rtnetlink_change *change = &chg; | |
5729 | int error; | |
5730 | ||
5731 | ofpbuf_init(&request, 0); | |
5732 | nl_msg_put_nlmsghdr(&request, | |
5733 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
5734 | RTM_GETLINK, NLM_F_REQUEST); | |
5735 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
5736 | ||
5737 | /* The correct identifiers for a Linux device are netnsid and ifindex, | |
5738 | * but ifindex changes as the port is moved to another network namespace | |
5739 | * and the interface name statically stored in ovsdb. */ | |
5740 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(&netdev->up)); | |
5741 | if (netdev_linux_netnsid_is_remote(netdev)) { | |
5742 | nl_msg_push_u32(&request, IFLA_IF_NETNSID, netdev->netnsid); | |
5743 | } | |
5744 | error = nl_transact(NETLINK_ROUTE, &request, &reply); | |
5745 | ofpbuf_uninit(&request); | |
5746 | if (error) { | |
5747 | ofpbuf_delete(reply); | |
5748 | return error; | |
5749 | } | |
5750 | ||
5751 | if (rtnetlink_parse(reply, change) | |
5752 | && change->nlmsg_type == RTM_NEWLINK) { | |
5753 | bool changed = false; | |
5754 | error = 0; | |
5755 | ||
5756 | /* Update netdev from rtnl msg and increment its seq if needed. */ | |
5757 | if ((change->ifi_flags ^ netdev->ifi_flags) & IFF_RUNNING) { | |
5758 | netdev->carrier_resets++; | |
5759 | changed = true; | |
5760 | } | |
5761 | if (change->ifi_flags != netdev->ifi_flags) { | |
5762 | netdev->ifi_flags = change->ifi_flags; | |
5763 | changed = true; | |
5764 | } | |
5765 | if (change->mtu && change->mtu != netdev->mtu) { | |
5766 | netdev->mtu = change->mtu; | |
5767 | netdev->cache_valid |= VALID_MTU; | |
5768 | netdev->netdev_mtu_error = 0; | |
5769 | changed = true; | |
5770 | } | |
5771 | if (!eth_addr_is_zero(change->mac) | |
5772 | && !eth_addr_equals(change->mac, netdev->etheraddr)) { | |
5773 | netdev->etheraddr = change->mac; | |
5774 | netdev->cache_valid |= VALID_ETHERADDR; | |
5775 | netdev->ether_addr_error = 0; | |
5776 | changed = true; | |
5777 | } | |
5778 | if (change->if_index != netdev->ifindex) { | |
5779 | netdev->ifindex = change->if_index; | |
5780 | netdev->cache_valid |= VALID_IFINDEX; | |
5781 | netdev->get_ifindex_error = 0; | |
5782 | changed = true; | |
5783 | } | |
3d9c99ab JH |
5784 | if (change->master && netdev_linux_kind_is_lag(change->master)) { |
5785 | netdev->is_lag_master = true; | |
5786 | } | |
756819dd FL |
5787 | if (changed) { |
5788 | netdev_change_seq_changed(&netdev->up); | |
5789 | } | |
5790 | } else { | |
5791 | error = EINVAL; | |
5792 | } | |
5793 | ||
5794 | ofpbuf_delete(reply); | |
5795 | return error; | |
5796 | } | |
5797 | ||
5798 | static int | |
74ff3298 | 5799 | get_etheraddr(const char *netdev_name, struct eth_addr *ea) |
8b61709d BP |
5800 | { |
5801 | struct ifreq ifr; | |
5802 | int hwaddr_family; | |
259e0b1a | 5803 | int error; |
8b61709d BP |
5804 | |
5805 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5806 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 5807 | COVERAGE_INC(netdev_get_hwaddr); |
259e0b1a BP |
5808 | error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); |
5809 | if (error) { | |
78857dfb BP |
5810 | /* ENODEV probably means that a vif disappeared asynchronously and |
5811 | * hasn't been removed from the database yet, so reduce the log level | |
5812 | * to INFO for that case. */ | |
259e0b1a | 5813 | VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, |
78857dfb | 5814 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", |
259e0b1a BP |
5815 | netdev_name, ovs_strerror(error)); |
5816 | return error; | |
8b61709d BP |
5817 | } |
5818 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
2482b0b0 JS |
5819 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER && |
5820 | hwaddr_family != ARPHRD_NONE) { | |
c9697f35 | 5821 | VLOG_INFO("%s device has unknown hardware address family %d", |
8b61709d | 5822 | netdev_name, hwaddr_family); |
c9697f35 | 5823 | return EINVAL; |
8b61709d BP |
5824 | } |
5825 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
5826 | return 0; | |
5827 | } | |
5828 | ||
5829 | static int | |
74ff3298 | 5830 | set_etheraddr(const char *netdev_name, const struct eth_addr mac) |
8b61709d BP |
5831 | { |
5832 | struct ifreq ifr; | |
259e0b1a | 5833 | int error; |
8b61709d BP |
5834 | |
5835 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5836 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 5837 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
74ff3298 | 5838 | memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN); |
8b61709d | 5839 | COVERAGE_INC(netdev_set_hwaddr); |
259e0b1a BP |
5840 | error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); |
5841 | if (error) { | |
8b61709d | 5842 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", |
259e0b1a | 5843 | netdev_name, ovs_strerror(error)); |
8b61709d | 5844 | } |
259e0b1a | 5845 | return error; |
8b61709d BP |
5846 | } |
5847 | ||
5848 | static int | |
0b0544d7 | 5849 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
5850 | int cmd, const char *cmd_name) |
5851 | { | |
5852 | struct ifreq ifr; | |
259e0b1a | 5853 | int error; |
8b61709d BP |
5854 | |
5855 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 5856 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
5857 | ifr.ifr_data = (caddr_t) ecmd; |
5858 | ||
5859 | ecmd->cmd = cmd; | |
259e0b1a BP |
5860 | error = af_inet_ioctl(SIOCETHTOOL, &ifr); |
5861 | if (error) { | |
5862 | if (error != EOPNOTSUPP) { | |
8b61709d | 5863 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " |
259e0b1a | 5864 | "failed: %s", cmd_name, name, ovs_strerror(error)); |
8b61709d BP |
5865 | } else { |
5866 | /* The device doesn't support this operation. That's pretty | |
5867 | * common, so there's no point in logging anything. */ | |
5868 | } | |
8b61709d | 5869 | } |
259e0b1a | 5870 | return error; |
8b61709d | 5871 | } |
f1acd62b | 5872 | |
488d734d BP |
5873 | /* Returns an AF_PACKET raw socket or a negative errno value. */ |
5874 | static int | |
5875 | af_packet_sock(void) | |
5876 | { | |
23882115 BP |
5877 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
5878 | static int sock; | |
488d734d | 5879 | |
23882115 | 5880 | if (ovsthread_once_start(&once)) { |
488d734d BP |
5881 | sock = socket(AF_PACKET, SOCK_RAW, 0); |
5882 | if (sock >= 0) { | |
8450059e BP |
5883 | int error = set_nonblocking(sock); |
5884 | if (error) { | |
5885 | close(sock); | |
5886 | sock = -error; | |
5887 | } | |
488d734d BP |
5888 | } else { |
5889 | sock = -errno; | |
10a89ef0 BP |
5890 | VLOG_ERR("failed to create packet socket: %s", |
5891 | ovs_strerror(errno)); | |
488d734d | 5892 | } |
23882115 | 5893 | ovsthread_once_done(&once); |
488d734d BP |
5894 | } |
5895 | ||
5896 | return sock; | |
5897 | } |