]>
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" | |
0de1b425 | 20 | #include "netdev-linux-private.h" |
d3980822 | 21 | |
e9e28be3 | 22 | #include <errno.h> |
8b61709d | 23 | #include <fcntl.h> |
b2befd5b BP |
24 | #include <sys/types.h> |
25 | #include <netinet/in.h> | |
55bc98d6 | 26 | #include <arpa/inet.h> |
8b61709d | 27 | #include <inttypes.h> |
2f564bb1 | 28 | #include <math.h> |
32383c3b | 29 | #include <linux/filter.h> |
c1c9c9c4 | 30 | #include <linux/gen_stats.h> |
bb7d0e22 | 31 | #include <linux/if_ether.h> |
29cf9c1b | 32 | #include <linux/if_packet.h> |
8b61709d BP |
33 | #include <linux/if_tun.h> |
34 | #include <linux/types.h> | |
35 | #include <linux/ethtool.h> | |
63331829 | 36 | #include <linux/mii.h> |
ef3767f5 | 37 | #include <linux/rtnetlink.h> |
8b61709d | 38 | #include <linux/sockios.h> |
29cf9c1b | 39 | #include <linux/virtio_net.h> |
8b61709d BP |
40 | #include <sys/ioctl.h> |
41 | #include <sys/socket.h> | |
29cf9c1b | 42 | #include <sys/uio.h> |
ac3e3aaa | 43 | #include <sys/utsname.h> |
8b61709d BP |
44 | #include <net/if.h> |
45 | #include <net/if_arp.h> | |
8b61709d | 46 | #include <net/route.h> |
e9e28be3 | 47 | #include <poll.h> |
8b61709d BP |
48 | #include <stdlib.h> |
49 | #include <string.h> | |
50 | #include <unistd.h> | |
e9e28be3 BP |
51 | |
52 | #include "coverage.h" | |
e14deea0 | 53 | #include "dp-packet.h" |
93451a0a | 54 | #include "dpif-netlink.h" |
df1e5a3b | 55 | #include "dpif-netdev.h" |
3e8a2ad1 | 56 | #include "openvswitch/dynamic-string.h" |
8b61709d | 57 | #include "fatal-signal.h" |
93b13be8 | 58 | #include "hash.h" |
ee89ea7b | 59 | #include "openvswitch/hmap.h" |
0de1b425 | 60 | #include "netdev-afxdp.h" |
8b61709d | 61 | #include "netdev-provider.h" |
7fbef77a | 62 | #include "netdev-vport.h" |
45c8d3a1 | 63 | #include "netlink-notifier.h" |
2fe27d5a | 64 | #include "netlink-socket.h" |
c060c4cf | 65 | #include "netlink.h" |
bfda5239 | 66 | #include "netnsid.h" |
64c96779 | 67 | #include "openvswitch/ofpbuf.h" |
8b61709d | 68 | #include "openflow/openflow.h" |
19c8e9c1 | 69 | #include "ovs-atomic.h" |
105cf8df | 70 | #include "ovs-numa.h" |
8b61709d | 71 | #include "packets.h" |
fd016ae3 | 72 | #include "openvswitch/poll-loop.h" |
7e9dcc0f | 73 | #include "rtnetlink.h" |
ee89ea7b | 74 | #include "openvswitch/shash.h" |
c060c4cf | 75 | #include "socket-util.h" |
19993ef3 | 76 | #include "sset.h" |
c1c5c723 | 77 | #include "tc.h" |
1670c579 | 78 | #include "timer.h" |
c060c4cf | 79 | #include "unaligned.h" |
e6211adc | 80 | #include "openvswitch/vlog.h" |
29cf9c1b | 81 | #include "userspace-tso.h" |
ee89ea7b | 82 | #include "util.h" |
5136ce49 | 83 | |
d98e6007 | 84 | VLOG_DEFINE_THIS_MODULE(netdev_linux); |
d76f09ea | 85 | |
d76f09ea BP |
86 | COVERAGE_DEFINE(netdev_set_policing); |
87 | COVERAGE_DEFINE(netdev_arp_lookup); | |
88 | COVERAGE_DEFINE(netdev_get_ifindex); | |
89 | COVERAGE_DEFINE(netdev_get_hwaddr); | |
90 | COVERAGE_DEFINE(netdev_set_hwaddr); | |
ab985a77 BP |
91 | COVERAGE_DEFINE(netdev_get_ethtool); |
92 | COVERAGE_DEFINE(netdev_set_ethtool); | |
4f925bd3 | 93 | |
8b61709d | 94 | \f |
756819dd FL |
95 | #ifndef IFLA_IF_NETNSID |
96 | #define IFLA_IF_NETNSID 0x45 | |
97 | #endif | |
8b61709d BP |
98 | /* These were introduced in Linux 2.6.14, so they might be missing if we have |
99 | * old headers. */ | |
100 | #ifndef ADVERTISED_Pause | |
101 | #define ADVERTISED_Pause (1 << 13) | |
102 | #endif | |
103 | #ifndef ADVERTISED_Asym_Pause | |
104 | #define ADVERTISED_Asym_Pause (1 << 14) | |
105 | #endif | |
106 | ||
e47bd51a JP |
107 | /* These were introduced in Linux 2.6.24, so they might be missing if we |
108 | * have old headers. */ | |
109 | #ifndef ETHTOOL_GFLAGS | |
110 | #define ETHTOOL_GFLAGS 0x00000025 /* Get flags bitmap(ethtool_value) */ | |
111 | #endif | |
112 | #ifndef ETHTOOL_SFLAGS | |
113 | #define ETHTOOL_SFLAGS 0x00000026 /* Set flags bitmap(ethtool_value) */ | |
114 | #endif | |
115 | ||
c1c9c9c4 BP |
116 | /* This was introduced in Linux 2.6.25, so it might be missing if we have old |
117 | * headers. */ | |
118 | #ifndef TC_RTAB_SIZE | |
119 | #define TC_RTAB_SIZE 1024 | |
120 | #endif | |
121 | ||
e7f6ba22 PJV |
122 | #ifndef TCM_IFINDEX_MAGIC_BLOCK |
123 | #define TCM_IFINDEX_MAGIC_BLOCK (0xFFFFFFFFU) | |
124 | #endif | |
125 | ||
b73c8518 SH |
126 | /* Linux 2.6.21 introduced struct tpacket_auxdata. |
127 | * Linux 2.6.27 added the tp_vlan_tci member. | |
128 | * Linux 3.0 defined TP_STATUS_VLAN_VALID. | |
129 | * Linux 3.13 repurposed a padding member for tp_vlan_tpid and defined | |
130 | * TP_STATUS_VLAN_TPID_VALID. | |
131 | * | |
132 | * With all this churn it's easiest to unconditionally define a replacement | |
133 | * structure that has everything we want. | |
134 | */ | |
55bc98d6 BP |
135 | #ifndef PACKET_AUXDATA |
136 | #define PACKET_AUXDATA 8 | |
137 | #endif | |
b73c8518 SH |
138 | #ifndef TP_STATUS_VLAN_VALID |
139 | #define TP_STATUS_VLAN_VALID (1 << 4) | |
140 | #endif | |
141 | #ifndef TP_STATUS_VLAN_TPID_VALID | |
142 | #define TP_STATUS_VLAN_TPID_VALID (1 << 6) | |
143 | #endif | |
144 | #undef tpacket_auxdata | |
145 | #define tpacket_auxdata rpl_tpacket_auxdata | |
146 | struct tpacket_auxdata { | |
147 | uint32_t tp_status; | |
148 | uint32_t tp_len; | |
149 | uint32_t tp_snaplen; | |
150 | uint16_t tp_mac; | |
151 | uint16_t tp_net; | |
152 | uint16_t tp_vlan_tci; | |
153 | uint16_t tp_vlan_tpid; | |
154 | }; | |
155 | ||
0c615356 SH |
156 | /* Linux 2.6.27 introduced ethtool_cmd_speed |
157 | * | |
158 | * To avoid revisiting problems reported with using configure to detect | |
159 | * compatibility (see report at | |
8a7903c6 | 160 | * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html) |
0c615356 SH |
161 | * unconditionally replace ethtool_cmd_speed. */ |
162 | #define ethtool_cmd_speed rpl_ethtool_cmd_speed | |
163 | static inline uint32_t rpl_ethtool_cmd_speed(const struct ethtool_cmd *ep) | |
164 | { | |
165 | return ep->speed | (ep->speed_hi << 16); | |
166 | } | |
167 | ||
67bed84c SH |
168 | /* Linux 2.6.30 introduced supported and advertised flags for |
169 | * 1G base KX, and 10G base KX4, KR and R. */ | |
170 | #ifndef SUPPORTED_1000baseKX_Full | |
171 | #define SUPPORTED_1000baseKX_Full (1 << 17) | |
172 | #define SUPPORTED_10000baseKX4_Full (1 << 18) | |
173 | #define SUPPORTED_10000baseKR_Full (1 << 19) | |
174 | #define SUPPORTED_10000baseR_FEC (1 << 20) | |
175 | #define ADVERTISED_1000baseKX_Full (1 << 17) | |
176 | #define ADVERTISED_10000baseKX4_Full (1 << 18) | |
177 | #define ADVERTISED_10000baseKR_Full (1 << 19) | |
178 | #define ADVERTISED_10000baseR_FEC (1 << 20) | |
179 | #endif | |
180 | ||
181 | /* Linux 3.5 introduced supported and advertised flags for | |
182 | * 40G base KR4, CR4, SR4 and LR4. */ | |
183 | #ifndef SUPPORTED_40000baseKR4_Full | |
184 | #define SUPPORTED_40000baseKR4_Full (1 << 23) | |
185 | #define SUPPORTED_40000baseCR4_Full (1 << 24) | |
186 | #define SUPPORTED_40000baseSR4_Full (1 << 25) | |
187 | #define SUPPORTED_40000baseLR4_Full (1 << 26) | |
188 | #define ADVERTISED_40000baseKR4_Full (1 << 23) | |
189 | #define ADVERTISED_40000baseCR4_Full (1 << 24) | |
190 | #define ADVERTISED_40000baseSR4_Full (1 << 25) | |
191 | #define ADVERTISED_40000baseLR4_Full (1 << 26) | |
192 | #endif | |
193 | ||
fa373af4 BP |
194 | /* Linux 2.6.35 introduced IFLA_STATS64 and rtnl_link_stats64. |
195 | * | |
196 | * Tests for rtnl_link_stats64 don't seem to consistently work, e.g. on | |
197 | * 2.6.32-431.29.2.el6.x86_64 (see report at | |
8a7903c6 JP |
198 | * https://mail.openvswitch.org/pipermail/ovs-dev/2014-October/291521.html). |
199 | * Maybe if_link.h is not self-contained on those kernels. It is easiest to | |
fa373af4 BP |
200 | * unconditionally define a replacement. */ |
201 | #ifndef IFLA_STATS64 | |
337c9b99 | 202 | #define IFLA_STATS64 23 |
fa373af4 BP |
203 | #endif |
204 | #define rtnl_link_stats64 rpl_rtnl_link_stats64 | |
337c9b99 BP |
205 | struct rtnl_link_stats64 { |
206 | uint64_t rx_packets; | |
207 | uint64_t tx_packets; | |
208 | uint64_t rx_bytes; | |
209 | uint64_t tx_bytes; | |
210 | uint64_t rx_errors; | |
211 | uint64_t tx_errors; | |
212 | uint64_t rx_dropped; | |
213 | uint64_t tx_dropped; | |
214 | uint64_t multicast; | |
215 | uint64_t collisions; | |
216 | ||
217 | uint64_t rx_length_errors; | |
218 | uint64_t rx_over_errors; | |
219 | uint64_t rx_crc_errors; | |
220 | uint64_t rx_frame_errors; | |
221 | uint64_t rx_fifo_errors; | |
222 | uint64_t rx_missed_errors; | |
223 | ||
224 | uint64_t tx_aborted_errors; | |
225 | uint64_t tx_carrier_errors; | |
226 | uint64_t tx_fifo_errors; | |
227 | uint64_t tx_heartbeat_errors; | |
228 | uint64_t tx_window_errors; | |
229 | ||
230 | uint64_t rx_compressed; | |
231 | uint64_t tx_compressed; | |
232 | }; | |
337c9b99 | 233 | |
8b61709d | 234 | enum { |
7fbef77a JG |
235 | VALID_IFINDEX = 1 << 0, |
236 | VALID_ETHERADDR = 1 << 1, | |
6b6e1329 PS |
237 | VALID_IN = 1 << 2, |
238 | VALID_MTU = 1 << 3, | |
239 | VALID_POLICING = 1 << 4, | |
240 | VALID_VPORT_STAT_ERROR = 1 << 5, | |
241 | VALID_DRVINFO = 1 << 6, | |
242 | VALID_FEATURES = 1 << 7, | |
105cf8df | 243 | VALID_NUMA_ID = 1 << 8, |
8b61709d | 244 | }; |
29cf9c1b FL |
245 | |
246 | /* Use one for the packet buffer and another for the aux buffer to receive | |
247 | * TSO packets. */ | |
248 | #define IOV_STD_SIZE 1 | |
249 | #define IOV_TSO_SIZE 2 | |
250 | ||
251 | enum { | |
252 | IOV_PACKET = 0, | |
253 | IOV_AUXBUF = 1, | |
254 | }; | |
c1c9c9c4 | 255 | \f |
d22f8927 JH |
256 | struct linux_lag_slave { |
257 | uint32_t block_id; | |
258 | struct shash_node *node; | |
259 | }; | |
260 | ||
261 | /* Protects 'lag_shash' and the mutable members of struct linux_lag_slave. */ | |
262 | static struct ovs_mutex lag_mutex = OVS_MUTEX_INITIALIZER; | |
263 | ||
264 | /* All slaves whose LAG masters are network devices in OvS. */ | |
265 | static struct shash lag_shash OVS_GUARDED_BY(lag_mutex) | |
266 | = SHASH_INITIALIZER(&lag_shash); | |
267 | ||
c1c9c9c4 BP |
268 | /* Traffic control. */ |
269 | ||
270 | /* An instance of a traffic control class. Always associated with a particular | |
93b13be8 BP |
271 | * network device. |
272 | * | |
273 | * Each TC implementation subclasses this with whatever additional data it | |
274 | * needs. */ | |
c1c9c9c4 BP |
275 | struct tc { |
276 | const struct tc_ops *ops; | |
93b13be8 BP |
277 | struct hmap queues; /* Contains "struct tc_queue"s. |
278 | * Read by generic TC layer. | |
279 | * Written only by TC implementation. */ | |
280 | }; | |
c1c9c9c4 | 281 | |
559eb230 BP |
282 | #define TC_INITIALIZER(TC, OPS) { OPS, HMAP_INITIALIZER(&(TC)->queues) } |
283 | ||
93b13be8 BP |
284 | /* One traffic control queue. |
285 | * | |
286 | * Each TC implementation subclasses this with whatever additional data it | |
287 | * needs. */ | |
288 | struct tc_queue { | |
289 | struct hmap_node hmap_node; /* In struct tc's "queues" hmap. */ | |
290 | unsigned int queue_id; /* OpenFlow queue ID. */ | |
6dc34a0d | 291 | long long int created; /* Time queue was created, in msecs. */ |
c1c9c9c4 BP |
292 | }; |
293 | ||
294 | /* A particular kind of traffic control. Each implementation generally maps to | |
295 | * one particular Linux qdisc class. | |
296 | * | |
297 | * The functions below return 0 if successful or a positive errno value on | |
298 | * failure, except where otherwise noted. All of them must be provided, except | |
299 | * where otherwise noted. */ | |
300 | struct tc_ops { | |
301 | /* Name used by kernel in the TCA_KIND attribute of tcmsg, e.g. "htb". | |
302 | * This is null for tc_ops_default and tc_ops_other, for which there are no | |
303 | * appropriate values. */ | |
304 | const char *linux_name; | |
305 | ||
306 | /* Name used in OVS database, e.g. "linux-htb". Must be nonnull. */ | |
307 | const char *ovs_name; | |
308 | ||
309 | /* Number of supported OpenFlow queues, 0 for qdiscs that have no | |
310 | * queues. The queues are numbered 0 through n_queues - 1. */ | |
311 | unsigned int n_queues; | |
312 | ||
313 | /* Called to install this TC class on 'netdev'. The implementation should | |
314 | * make the Netlink calls required to set up 'netdev' with the right qdisc | |
315 | * and configure it according to 'details'. The implementation may assume | |
316 | * that the current qdisc is the default; that is, there is no need for it | |
317 | * to delete the current qdisc before installing itself. | |
318 | * | |
319 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
320 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
321 | * (which is built as ovs-vswitchd.conf.db(8)). | |
322 | * | |
323 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
324 | * initialized 'struct tc'. | |
325 | * | |
326 | * (This function is null for tc_ops_other, which cannot be installed. For | |
327 | * other TC classes it should always be nonnull.) */ | |
79f1cbe9 | 328 | int (*tc_install)(struct netdev *netdev, const struct smap *details); |
c1c9c9c4 BP |
329 | |
330 | /* Called when the netdev code determines (through a Netlink query) that | |
331 | * this TC class's qdisc is installed on 'netdev', but we didn't install | |
332 | * it ourselves and so don't know any of the details. | |
333 | * | |
334 | * 'nlmsg' is the kernel reply to a RTM_GETQDISC Netlink message for | |
335 | * 'netdev'. The TCA_KIND attribute of 'nlmsg' is 'linux_name'. The | |
336 | * implementation should parse the other attributes of 'nlmsg' as | |
337 | * necessary to determine its configuration. If necessary it should also | |
338 | * use Netlink queries to determine the configuration of queues on | |
339 | * 'netdev'. | |
340 | * | |
341 | * This function must return 0 if and only if it sets 'netdev->tc' to an | |
342 | * initialized 'struct tc'. */ | |
343 | int (*tc_load)(struct netdev *netdev, struct ofpbuf *nlmsg); | |
344 | ||
345 | /* Destroys the data structures allocated by the implementation as part of | |
346 | * 'tc'. (This includes destroying 'tc->queues' by calling | |
347 | * tc_destroy(tc). | |
348 | * | |
349 | * The implementation should not need to perform any Netlink calls. If | |
350 | * desirable, the caller is responsible for deconfiguring the kernel qdisc. | |
351 | * (But it may not be desirable.) | |
352 | * | |
353 | * This function may be null if 'tc' is trivial. */ | |
354 | void (*tc_destroy)(struct tc *tc); | |
355 | ||
356 | /* Retrieves details of 'netdev->tc' configuration into 'details'. | |
357 | * | |
358 | * The implementation should not need to perform any Netlink calls, because | |
359 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
360 | * cached the configuration. | |
361 | * | |
362 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
363 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
364 | * (which is built as ovs-vswitchd.conf.db(8)). | |
365 | * | |
366 | * This function may be null if 'tc' is not configurable. | |
367 | */ | |
79f1cbe9 | 368 | int (*qdisc_get)(const struct netdev *netdev, struct smap *details); |
c1c9c9c4 BP |
369 | |
370 | /* Reconfigures 'netdev->tc' according to 'details', performing any | |
371 | * required Netlink calls to complete the reconfiguration. | |
372 | * | |
373 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
374 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
375 | * (which is built as ovs-vswitchd.conf.db(8)). | |
376 | * | |
377 | * This function may be null if 'tc' is not configurable. | |
378 | */ | |
79f1cbe9 | 379 | int (*qdisc_set)(struct netdev *, const struct smap *details); |
c1c9c9c4 | 380 | |
93b13be8 BP |
381 | /* Retrieves details of 'queue' on 'netdev->tc' into 'details'. 'queue' is |
382 | * one of the 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
383 | * |
384 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
385 | * in the "other_config" column in the "Queue" table in | |
386 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
387 | * | |
388 | * The implementation should not need to perform any Netlink calls, because | |
389 | * the 'tc_install' or 'tc_load' that instantiated 'netdev->tc' should have | |
390 | * cached the queue configuration. | |
391 | * | |
392 | * This function may be null if 'tc' does not have queues ('n_queues' is | |
393 | * 0). */ | |
93b13be8 | 394 | int (*class_get)(const struct netdev *netdev, const struct tc_queue *queue, |
79f1cbe9 | 395 | struct smap *details); |
c1c9c9c4 BP |
396 | |
397 | /* Configures or reconfigures 'queue_id' on 'netdev->tc' according to | |
398 | * 'details', perfoming any required Netlink calls to complete the | |
399 | * reconfiguration. The caller ensures that 'queue_id' is less than | |
400 | * 'n_queues'. | |
401 | * | |
402 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
403 | * in the "other_config" column in the "Queue" table in | |
404 | * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). | |
405 | * | |
406 | * This function may be null if 'tc' does not have queues or its queues are | |
407 | * not configurable. */ | |
408 | int (*class_set)(struct netdev *, unsigned int queue_id, | |
79f1cbe9 | 409 | const struct smap *details); |
c1c9c9c4 | 410 | |
93b13be8 BP |
411 | /* Deletes 'queue' from 'netdev->tc'. 'queue' is one of the 'struct |
412 | * tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
413 | * |
414 | * This function may be null if 'tc' does not have queues or its queues | |
415 | * cannot be deleted. */ | |
93b13be8 | 416 | int (*class_delete)(struct netdev *, struct tc_queue *queue); |
c1c9c9c4 | 417 | |
93b13be8 BP |
418 | /* Obtains stats for 'queue' from 'netdev->tc'. 'queue' is one of the |
419 | * 'struct tc_queue's within 'netdev->tc->queues'. | |
c1c9c9c4 BP |
420 | * |
421 | * On success, initializes '*stats'. | |
422 | * | |
423 | * This function may be null if 'tc' does not have queues or if it cannot | |
424 | * report queue statistics. */ | |
93b13be8 BP |
425 | int (*class_get_stats)(const struct netdev *netdev, |
426 | const struct tc_queue *queue, | |
c1c9c9c4 BP |
427 | struct netdev_queue_stats *stats); |
428 | ||
429 | /* Extracts queue stats from 'nlmsg', which is a response to a | |
430 | * RTM_GETTCLASS message, and passes them to 'cb' along with 'aux'. | |
431 | * | |
432 | * This function may be null if 'tc' does not have queues or if it cannot | |
433 | * report queue statistics. */ | |
434 | int (*class_dump_stats)(const struct netdev *netdev, | |
435 | const struct ofpbuf *nlmsg, | |
436 | netdev_dump_queue_stats_cb *cb, void *aux); | |
437 | }; | |
438 | ||
439 | static void | |
440 | tc_init(struct tc *tc, const struct tc_ops *ops) | |
441 | { | |
442 | tc->ops = ops; | |
93b13be8 | 443 | hmap_init(&tc->queues); |
c1c9c9c4 BP |
444 | } |
445 | ||
446 | static void | |
447 | tc_destroy(struct tc *tc) | |
448 | { | |
93b13be8 | 449 | hmap_destroy(&tc->queues); |
c1c9c9c4 BP |
450 | } |
451 | ||
452 | static const struct tc_ops tc_ops_htb; | |
a339aa81 | 453 | static const struct tc_ops tc_ops_hfsc; |
677d9158 JV |
454 | static const struct tc_ops tc_ops_codel; |
455 | static const struct tc_ops tc_ops_fqcodel; | |
456 | static const struct tc_ops tc_ops_sfq; | |
2f564bb1 | 457 | static const struct tc_ops tc_ops_netem; |
c1c9c9c4 | 458 | static const struct tc_ops tc_ops_default; |
6cf888b8 | 459 | static const struct tc_ops tc_ops_noop; |
c1c9c9c4 BP |
460 | static const struct tc_ops tc_ops_other; |
461 | ||
559eb230 | 462 | static const struct tc_ops *const tcs[] = { |
c1c9c9c4 | 463 | &tc_ops_htb, /* Hierarchy token bucket (see tc-htb(8)). */ |
a339aa81 | 464 | &tc_ops_hfsc, /* Hierarchical fair service curve. */ |
677d9158 JV |
465 | &tc_ops_codel, /* Controlled delay */ |
466 | &tc_ops_fqcodel, /* Fair queue controlled delay */ | |
467 | &tc_ops_sfq, /* Stochastic fair queueing */ | |
2f564bb1 | 468 | &tc_ops_netem, /* Network Emulator */ |
6cf888b8 | 469 | &tc_ops_noop, /* Non operating qos type. */ |
c1c9c9c4 BP |
470 | &tc_ops_default, /* Default qdisc (see tc-pfifo_fast(8)). */ |
471 | &tc_ops_other, /* Some other qdisc. */ | |
472 | NULL | |
473 | }; | |
149f577a | 474 | |
c1c9c9c4 BP |
475 | static unsigned int tc_ticks_to_bytes(unsigned int rate, unsigned int ticks); |
476 | static unsigned int tc_bytes_to_ticks(unsigned int rate, unsigned int size); | |
477 | static unsigned int tc_buffer_per_jiffy(unsigned int rate); | |
2f564bb1 | 478 | static uint32_t tc_time_to_ticks(uint32_t time); |
c1c9c9c4 | 479 | |
7874bdff RD |
480 | static struct tcmsg *netdev_linux_tc_make_request(const struct netdev *, |
481 | int type, | |
482 | unsigned int flags, | |
483 | struct ofpbuf *); | |
c7952afb BP |
484 | static int tc_add_policer(struct netdev *, |
485 | uint32_t kbits_rate, uint32_t kbits_burst); | |
c1c9c9c4 BP |
486 | |
487 | static int tc_parse_qdisc(const struct ofpbuf *, const char **kind, | |
488 | struct nlattr **options); | |
489 | static int tc_parse_class(const struct ofpbuf *, unsigned int *queue_id, | |
490 | struct nlattr **options, | |
491 | struct netdev_queue_stats *); | |
492 | static int tc_query_class(const struct netdev *, | |
493 | unsigned int handle, unsigned int parent, | |
494 | struct ofpbuf **replyp); | |
495 | static int tc_delete_class(const struct netdev *, unsigned int handle); | |
496 | ||
497 | static int tc_del_qdisc(struct netdev *netdev); | |
498 | static int tc_query_qdisc(const struct netdev *netdev); | |
499 | ||
e7f6ba22 PJV |
500 | void |
501 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate); | |
c1c9c9c4 BP |
502 | static int tc_calc_cell_log(unsigned int mtu); |
503 | static void tc_fill_rate(struct tc_ratespec *rate, uint64_t bps, int mtu); | |
c1c9c9c4 BP |
504 | static int tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes); |
505 | \f | |
8b61709d | 506 | |
8b61709d BP |
507 | /* This is set pretty low because we probably won't learn anything from the |
508 | * additional log messages. */ | |
509 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
510 | ||
19c8e9c1 JS |
511 | /* Polling miimon status for all ports causes performance degradation when |
512 | * handling a large number of ports. If there are no devices using miimon, then | |
812c272c JR |
513 | * we skip netdev_linux_miimon_run() and netdev_linux_miimon_wait(). |
514 | * | |
515 | * Readers do not depend on this variable synchronizing with the related | |
516 | * changes in the device miimon status, so we can use atomic_count. */ | |
517 | static atomic_count miimon_cnt = ATOMIC_COUNT_INIT(0); | |
19c8e9c1 | 518 | |
29cf9c1b FL |
519 | static int netdev_linux_parse_vnet_hdr(struct dp_packet *b); |
520 | static void netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu); | |
0b0544d7 | 521 | static int netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *, |
8b61709d | 522 | int cmd, const char *cmd_name); |
b5d57fc8 | 523 | static int get_flags(const struct netdev *, unsigned int *flags); |
4b609110 | 524 | static int set_flags(const char *, unsigned int flags); |
4f9f3f21 BP |
525 | static int update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
526 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
527 | OVS_REQUIRES(netdev->mutex); | |
8b61709d BP |
528 | static int get_ifindex(const struct netdev *, int *ifindexp); |
529 | static int do_set_addr(struct netdev *netdev, | |
530 | int ioctl_nr, const char *ioctl_name, | |
531 | struct in_addr addr); | |
74ff3298 JR |
532 | static int get_etheraddr(const char *netdev_name, struct eth_addr *ea); |
533 | static int set_etheraddr(const char *netdev_name, const struct eth_addr); | |
488d734d | 534 | static int af_packet_sock(void); |
19c8e9c1 | 535 | static bool netdev_linux_miimon_enabled(void); |
1670c579 EJ |
536 | static void netdev_linux_miimon_run(void); |
537 | static void netdev_linux_miimon_wait(void); | |
df1e5a3b | 538 | static int netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup); |
8b61709d | 539 | |
796223f5 BP |
540 | static bool |
541 | is_tap_netdev(const struct netdev *netdev) | |
542 | { | |
b5d57fc8 | 543 | return netdev_get_class(netdev) == &netdev_tap_class; |
6c88d577 | 544 | } |
ff4ed3c9 | 545 | \f |
bfda5239 FL |
546 | static int |
547 | netdev_linux_netnsid_update__(struct netdev_linux *netdev) | |
548 | { | |
549 | struct dpif_netlink_vport reply; | |
550 | struct ofpbuf *buf; | |
551 | int error; | |
552 | ||
553 | error = dpif_netlink_vport_get(netdev_get_name(&netdev->up), &reply, &buf); | |
554 | if (error) { | |
629e1476 FL |
555 | if (error == ENOENT) { |
556 | /* Assume it is local if there is no API (e.g. if the openvswitch | |
557 | * kernel module is not loaded). */ | |
558 | netnsid_set_local(&netdev->netnsid); | |
559 | } else { | |
560 | netnsid_unset(&netdev->netnsid); | |
561 | } | |
bfda5239 FL |
562 | return error; |
563 | } | |
564 | ||
565 | netnsid_set(&netdev->netnsid, reply.netnsid); | |
566 | ofpbuf_delete(buf); | |
567 | return 0; | |
568 | } | |
569 | ||
570 | static int | |
571 | netdev_linux_netnsid_update(struct netdev_linux *netdev) | |
572 | { | |
573 | if (netnsid_is_unset(netdev->netnsid)) { | |
3dbcbfe4 FL |
574 | if (netdev_get_class(&netdev->up) == &netdev_tap_class) { |
575 | netnsid_set_local(&netdev->netnsid); | |
576 | } else { | |
577 | return netdev_linux_netnsid_update__(netdev); | |
578 | } | |
bfda5239 FL |
579 | } |
580 | ||
581 | return 0; | |
582 | } | |
583 | ||
584 | static bool | |
585 | netdev_linux_netnsid_is_eq(struct netdev_linux *netdev, int nsid) | |
586 | { | |
587 | netdev_linux_netnsid_update(netdev); | |
588 | return netnsid_eq(netdev->netnsid, nsid); | |
589 | } | |
590 | ||
756819dd FL |
591 | static bool |
592 | netdev_linux_netnsid_is_remote(struct netdev_linux *netdev) | |
593 | { | |
594 | netdev_linux_netnsid_update(netdev); | |
595 | return netnsid_is_remote(netdev->netnsid); | |
596 | } | |
597 | ||
598 | static int netdev_linux_update_via_netlink(struct netdev_linux *); | |
bfda5239 | 599 | static void netdev_linux_update(struct netdev_linux *netdev, int, |
7e9dcc0f | 600 | const struct rtnetlink_change *) |
86383816 | 601 | OVS_REQUIRES(netdev->mutex); |
cee87338 | 602 | static void netdev_linux_changed(struct netdev_linux *netdev, |
86383816 BP |
603 | unsigned int ifi_flags, unsigned int mask) |
604 | OVS_REQUIRES(netdev->mutex); | |
cee87338 | 605 | |
d6384a3a AW |
606 | /* Returns a NETLINK_ROUTE socket listening for RTNLGRP_LINK, |
607 | * RTNLGRP_IPV4_IFADDR and RTNLGRP_IPV6_IFADDR changes, or NULL | |
cee87338 BP |
608 | * if no such socket could be created. */ |
609 | static struct nl_sock * | |
610 | netdev_linux_notify_sock(void) | |
611 | { | |
612 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
613 | static struct nl_sock *sock; | |
989d7135 PS |
614 | unsigned int mcgroups[] = {RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR, |
615 | RTNLGRP_IPV6_IFADDR, RTNLGRP_IPV6_IFINFO}; | |
cee87338 BP |
616 | |
617 | if (ovsthread_once_start(&once)) { | |
618 | int error; | |
619 | ||
620 | error = nl_sock_create(NETLINK_ROUTE, &sock); | |
621 | if (!error) { | |
d6384a3a AW |
622 | size_t i; |
623 | ||
624 | for (i = 0; i < ARRAY_SIZE(mcgroups); i++) { | |
625 | error = nl_sock_join_mcgroup(sock, mcgroups[i]); | |
626 | if (error) { | |
627 | nl_sock_destroy(sock); | |
628 | sock = NULL; | |
629 | break; | |
630 | } | |
cee87338 BP |
631 | } |
632 | } | |
cf114a7f | 633 | nl_sock_listen_all_nsid(sock, true); |
cee87338 BP |
634 | ovsthread_once_done(&once); |
635 | } | |
636 | ||
637 | return sock; | |
638 | } | |
639 | ||
19c8e9c1 JS |
640 | static bool |
641 | netdev_linux_miimon_enabled(void) | |
642 | { | |
812c272c | 643 | return atomic_count_get(&miimon_cnt) > 0; |
19c8e9c1 JS |
644 | } |
645 | ||
3d9c99ab JH |
646 | static bool |
647 | netdev_linux_kind_is_lag(const char *kind) | |
648 | { | |
649 | if (!strcmp(kind, "bond") || !strcmp(kind, "team")) { | |
650 | return true; | |
651 | } | |
652 | ||
653 | return false; | |
654 | } | |
655 | ||
d22f8927 JH |
656 | static void |
657 | netdev_linux_update_lag(struct rtnetlink_change *change) | |
658 | OVS_REQUIRES(lag_mutex) | |
659 | { | |
660 | struct linux_lag_slave *lag; | |
661 | ||
662 | if (!rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)) { | |
663 | return; | |
664 | } | |
665 | ||
666 | if (change->slave && netdev_linux_kind_is_lag(change->slave)) { | |
667 | lag = shash_find_data(&lag_shash, change->ifname); | |
668 | ||
669 | if (!lag) { | |
670 | struct netdev *master_netdev; | |
671 | char master_name[IFNAMSIZ]; | |
672 | uint32_t block_id; | |
673 | int error = 0; | |
674 | ||
675 | if_indextoname(change->master_ifindex, master_name); | |
676 | master_netdev = netdev_from_name(master_name); | |
e3b5d7c5 TL |
677 | if (!master_netdev) { |
678 | return; | |
679 | } | |
d22f8927 JH |
680 | |
681 | if (is_netdev_linux_class(master_netdev->netdev_class)) { | |
682 | block_id = netdev_get_block_id(master_netdev); | |
683 | if (!block_id) { | |
e3b5d7c5 TL |
684 | netdev_close(master_netdev); |
685 | return; | |
d22f8927 JH |
686 | } |
687 | ||
688 | lag = xmalloc(sizeof *lag); | |
689 | lag->block_id = block_id; | |
690 | lag->node = shash_add(&lag_shash, change->ifname, lag); | |
691 | ||
cae64353 | 692 | /* delete ingress block in case it exists */ |
95255018 | 693 | tc_add_del_qdisc(change->if_index, false, 0, TC_INGRESS); |
d22f8927 | 694 | /* LAG master is linux netdev so add slave to same block. */ |
95255018 JH |
695 | error = tc_add_del_qdisc(change->if_index, true, block_id, |
696 | TC_INGRESS); | |
d22f8927 | 697 | if (error) { |
cae64353 RD |
698 | VLOG_WARN("failed to bind LAG slave %s to master's block", |
699 | change->ifname); | |
d22f8927 JH |
700 | shash_delete(&lag_shash, lag->node); |
701 | free(lag); | |
702 | } | |
703 | } | |
e3b5d7c5 TL |
704 | |
705 | netdev_close(master_netdev); | |
d22f8927 JH |
706 | } |
707 | } else if (change->master_ifindex == 0) { | |
708 | /* Check if this was a lag slave that has been freed. */ | |
709 | lag = shash_find_data(&lag_shash, change->ifname); | |
710 | ||
711 | if (lag) { | |
95255018 JH |
712 | tc_add_del_qdisc(change->if_index, false, lag->block_id, |
713 | TC_INGRESS); | |
d22f8927 JH |
714 | shash_delete(&lag_shash, lag->node); |
715 | free(lag); | |
716 | } | |
717 | } | |
718 | } | |
719 | ||
0de1b425 | 720 | void |
1c33f0c3 | 721 | netdev_linux_run(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 722 | { |
cee87338 BP |
723 | struct nl_sock *sock; |
724 | int error; | |
725 | ||
19c8e9c1 JS |
726 | if (netdev_linux_miimon_enabled()) { |
727 | netdev_linux_miimon_run(); | |
728 | } | |
cee87338 BP |
729 | |
730 | sock = netdev_linux_notify_sock(); | |
731 | if (!sock) { | |
732 | return; | |
733 | } | |
734 | ||
735 | do { | |
cee87338 | 736 | uint64_t buf_stub[4096 / 8]; |
bfda5239 | 737 | int nsid; |
cee87338 BP |
738 | struct ofpbuf buf; |
739 | ||
740 | ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub); | |
bfda5239 | 741 | error = nl_sock_recv(sock, &buf, &nsid, false); |
cee87338 | 742 | if (!error) { |
7e9dcc0f | 743 | struct rtnetlink_change change; |
cee87338 | 744 | |
7e9dcc0f | 745 | if (rtnetlink_parse(&buf, &change)) { |
989d7135 PS |
746 | struct netdev *netdev_ = NULL; |
747 | char dev_name[IFNAMSIZ]; | |
748 | ||
749 | if (!change.ifname) { | |
750 | change.ifname = if_indextoname(change.if_index, dev_name); | |
751 | } | |
752 | ||
753 | if (change.ifname) { | |
754 | netdev_ = netdev_from_name(change.ifname); | |
755 | } | |
cee87338 BP |
756 | if (netdev_ && is_netdev_linux_class(netdev_->netdev_class)) { |
757 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
86383816 BP |
758 | |
759 | ovs_mutex_lock(&netdev->mutex); | |
bfda5239 | 760 | netdev_linux_update(netdev, nsid, &change); |
86383816 | 761 | ovs_mutex_unlock(&netdev->mutex); |
cee87338 | 762 | } |
d22f8927 JH |
763 | else if (!netdev_ && change.ifname) { |
764 | /* Netdev is not present in OvS but its master could be. */ | |
765 | ovs_mutex_lock(&lag_mutex); | |
766 | netdev_linux_update_lag(&change); | |
767 | ovs_mutex_unlock(&lag_mutex); | |
768 | } | |
38e0065b | 769 | netdev_close(netdev_); |
cee87338 BP |
770 | } |
771 | } else if (error == ENOBUFS) { | |
772 | struct shash device_shash; | |
773 | struct shash_node *node; | |
774 | ||
775 | nl_sock_drain(sock); | |
776 | ||
777 | shash_init(&device_shash); | |
778 | netdev_get_devices(&netdev_linux_class, &device_shash); | |
779 | SHASH_FOR_EACH (node, &device_shash) { | |
780 | struct netdev *netdev_ = node->data; | |
781 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
782 | unsigned int flags; | |
783 | ||
86383816 | 784 | ovs_mutex_lock(&netdev->mutex); |
cee87338 BP |
785 | get_flags(netdev_, &flags); |
786 | netdev_linux_changed(netdev, flags, 0); | |
86383816 BP |
787 | ovs_mutex_unlock(&netdev->mutex); |
788 | ||
cee87338 BP |
789 | netdev_close(netdev_); |
790 | } | |
791 | shash_destroy(&device_shash); | |
792 | } else if (error != EAGAIN) { | |
7ed58d4a JP |
793 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 5); |
794 | VLOG_WARN_RL(&rll, "error reading or parsing netlink (%s)", | |
cee87338 BP |
795 | ovs_strerror(error)); |
796 | } | |
797 | ofpbuf_uninit(&buf); | |
798 | } while (!error); | |
8b61709d BP |
799 | } |
800 | ||
801 | static void | |
1c33f0c3 | 802 | netdev_linux_wait(const struct netdev_class *netdev_class OVS_UNUSED) |
8b61709d | 803 | { |
cee87338 BP |
804 | struct nl_sock *sock; |
805 | ||
19c8e9c1 JS |
806 | if (netdev_linux_miimon_enabled()) { |
807 | netdev_linux_miimon_wait(); | |
808 | } | |
cee87338 BP |
809 | sock = netdev_linux_notify_sock(); |
810 | if (sock) { | |
811 | nl_sock_wait(sock, POLLIN); | |
812 | } | |
8b61709d BP |
813 | } |
814 | ||
ac4d3bcb | 815 | static void |
b5d57fc8 BP |
816 | netdev_linux_changed(struct netdev_linux *dev, |
817 | unsigned int ifi_flags, unsigned int mask) | |
86383816 | 818 | OVS_REQUIRES(dev->mutex) |
ac4d3bcb | 819 | { |
3e912ffc | 820 | netdev_change_seq_changed(&dev->up); |
8aa77183 BP |
821 | |
822 | if ((dev->ifi_flags ^ ifi_flags) & IFF_RUNNING) { | |
823 | dev->carrier_resets++; | |
824 | } | |
825 | dev->ifi_flags = ifi_flags; | |
826 | ||
4f925bd3 | 827 | dev->cache_valid &= mask; |
6b6e1329 | 828 | if (!(mask & VALID_IN)) { |
a8704b50 PS |
829 | netdev_get_addrs_list_flush(); |
830 | } | |
4f925bd3 PS |
831 | } |
832 | ||
833 | static void | |
bfda5239 FL |
834 | netdev_linux_update__(struct netdev_linux *dev, |
835 | const struct rtnetlink_change *change) | |
86383816 | 836 | OVS_REQUIRES(dev->mutex) |
4f925bd3 | 837 | { |
bfda5239 | 838 | if (rtnetlink_type_is_rtnlgrp_link(change->nlmsg_type)) { |
d6384a3a | 839 | if (change->nlmsg_type == RTM_NEWLINK) { |
105cf8df | 840 | /* Keep drv-info, ip addresses, and NUMA id. */ |
d6384a3a | 841 | netdev_linux_changed(dev, change->ifi_flags, |
105cf8df | 842 | VALID_DRVINFO | VALID_IN | VALID_NUMA_ID); |
d6384a3a AW |
843 | |
844 | /* Update netdev from rtnl-change msg. */ | |
845 | if (change->mtu) { | |
846 | dev->mtu = change->mtu; | |
847 | dev->cache_valid |= VALID_MTU; | |
848 | dev->netdev_mtu_error = 0; | |
849 | } | |
90a6637d | 850 | |
74ff3298 JR |
851 | if (!eth_addr_is_zero(change->mac)) { |
852 | dev->etheraddr = change->mac; | |
d6384a3a AW |
853 | dev->cache_valid |= VALID_ETHERADDR; |
854 | dev->ether_addr_error = 0; | |
e8e1a409 TZ |
855 | |
856 | /* The mac addr has been changed, report it now. */ | |
857 | rtnetlink_report_link(); | |
d6384a3a | 858 | } |
44445cac | 859 | |
3d9c99ab JH |
860 | if (change->master && netdev_linux_kind_is_lag(change->master)) { |
861 | dev->is_lag_master = true; | |
862 | } | |
863 | ||
d6384a3a AW |
864 | dev->ifindex = change->if_index; |
865 | dev->cache_valid |= VALID_IFINDEX; | |
866 | dev->get_ifindex_error = 0; | |
22dcb534 | 867 | dev->present = true; |
d6384a3a | 868 | } else { |
bfda5239 | 869 | /* FIXME */ |
d6384a3a | 870 | netdev_linux_changed(dev, change->ifi_flags, 0); |
22dcb534 | 871 | dev->present = false; |
bfda5239 | 872 | netnsid_unset(&dev->netnsid); |
d6384a3a AW |
873 | } |
874 | } else if (rtnetlink_type_is_rtnlgrp_addr(change->nlmsg_type)) { | |
875 | /* Invalidates in4, in6. */ | |
6b6e1329 | 876 | netdev_linux_changed(dev, dev->ifi_flags, ~VALID_IN); |
4f925bd3 | 877 | } else { |
d6384a3a | 878 | OVS_NOT_REACHED(); |
4f925bd3 | 879 | } |
ac4d3bcb EJ |
880 | } |
881 | ||
bfda5239 FL |
882 | static void |
883 | netdev_linux_update(struct netdev_linux *dev, int nsid, | |
884 | const struct rtnetlink_change *change) | |
885 | OVS_REQUIRES(dev->mutex) | |
886 | { | |
887 | if (netdev_linux_netnsid_is_eq(dev, nsid)) { | |
888 | netdev_linux_update__(dev, change); | |
889 | } | |
890 | } | |
891 | ||
9dc63482 BP |
892 | static struct netdev * |
893 | netdev_linux_alloc(void) | |
894 | { | |
895 | struct netdev_linux *netdev = xzalloc(sizeof *netdev); | |
896 | return &netdev->up; | |
897 | } | |
898 | ||
48c6733c WT |
899 | static int |
900 | netdev_linux_common_construct(struct netdev *netdev_) | |
9dc63482 | 901 | { |
48c6733c WT |
902 | /* Prevent any attempt to create (or open) a network device named "default" |
903 | * or "all". These device names are effectively reserved on Linux because | |
904 | * /proc/sys/net/ipv4/conf/ always contains directories by these names. By | |
905 | * itself this wouldn't call for any special treatment, but in practice if | |
906 | * a program tries to create devices with these names, it causes the kernel | |
907 | * to fire a "new device" notification event even though creation failed, | |
908 | * and in turn that causes OVS to wake up and try to create them again, | |
909 | * which ends up as a 100% CPU loop. */ | |
910 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
911 | const char *name = netdev_->name; | |
912 | if (!strcmp(name, "default") || !strcmp(name, "all")) { | |
7ed58d4a JP |
913 | static struct vlog_rate_limit rll = VLOG_RATE_LIMIT_INIT(1, 1); |
914 | VLOG_WARN_RL(&rll, "%s: Linux forbids network device with this name", | |
48c6733c WT |
915 | name); |
916 | return EINVAL; | |
917 | } | |
918 | ||
bfda5239 FL |
919 | /* The device could be in the same network namespace or in another one. */ |
920 | netnsid_unset(&netdev->netnsid); | |
834d6caf | 921 | ovs_mutex_init(&netdev->mutex); |
29cf9c1b FL |
922 | |
923 | if (userspace_tso_enabled()) { | |
924 | netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_TSO; | |
925 | netdev_->ol_flags |= NETDEV_TX_OFFLOAD_TCP_CKSUM; | |
8c5163fe | 926 | netdev_->ol_flags |= NETDEV_TX_OFFLOAD_UDP_CKSUM; |
35b5586b | 927 | netdev_->ol_flags |= NETDEV_TX_OFFLOAD_SCTP_CKSUM; |
29cf9c1b FL |
928 | netdev_->ol_flags |= NETDEV_TX_OFFLOAD_IPV4_CKSUM; |
929 | } | |
930 | ||
48c6733c | 931 | return 0; |
9dc63482 BP |
932 | } |
933 | ||
1f6e0fbd | 934 | /* Creates system and internal devices. */ |
f627cf1d | 935 | int |
9dc63482 | 936 | netdev_linux_construct(struct netdev *netdev_) |
1f6e0fbd | 937 | { |
9dc63482 | 938 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
48c6733c WT |
939 | int error = netdev_linux_common_construct(netdev_); |
940 | if (error) { | |
941 | return error; | |
942 | } | |
1f6e0fbd | 943 | |
b5d57fc8 BP |
944 | error = get_flags(&netdev->up, &netdev->ifi_flags); |
945 | if (error == ENODEV) { | |
9dc63482 | 946 | if (netdev->up.netdev_class != &netdev_internal_class) { |
b5d57fc8 | 947 | /* The device does not exist, so don't allow it to be opened. */ |
b5d57fc8 BP |
948 | return ENODEV; |
949 | } else { | |
950 | /* "Internal" netdevs have to be created as netdev objects before | |
951 | * they exist in the kernel, because creating them in the kernel | |
952 | * happens by passing a netdev object to dpif_port_add(). | |
953 | * Therefore, ignore the error. */ | |
954 | } | |
955 | } | |
46415c90 | 956 | |
a740f0de JG |
957 | return 0; |
958 | } | |
959 | ||
5b7448ed JG |
960 | /* For most types of netdevs we open the device for each call of |
961 | * netdev_open(). However, this is not the case with tap devices, | |
962 | * since it is only possible to open the device once. In this | |
963 | * situation we share a single file descriptor, and consequently | |
964 | * buffers, across all readers. Therefore once data is read it will | |
965 | * be unavailable to other reads for tap devices. */ | |
a740f0de | 966 | static int |
9dc63482 | 967 | netdev_linux_construct_tap(struct netdev *netdev_) |
a740f0de | 968 | { |
9dc63482 | 969 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a740f0de | 970 | static const char tap_dev[] = "/dev/net/tun"; |
9dc63482 | 971 | const char *name = netdev_->name; |
a740f0de | 972 | struct ifreq ifr; |
a740f0de | 973 | |
48c6733c WT |
974 | int error = netdev_linux_common_construct(netdev_); |
975 | if (error) { | |
976 | return error; | |
977 | } | |
1f6e0fbd | 978 | |
6c88d577 | 979 | /* Open tap device. */ |
d0d08f8a BP |
980 | netdev->tap_fd = open(tap_dev, O_RDWR); |
981 | if (netdev->tap_fd < 0) { | |
6c88d577 | 982 | error = errno; |
10a89ef0 | 983 | VLOG_WARN("opening \"%s\" failed: %s", tap_dev, ovs_strerror(error)); |
cee87338 | 984 | return error; |
6c88d577 JP |
985 | } |
986 | ||
987 | /* Create tap device. */ | |
61b9d078 | 988 | get_flags(&netdev->up, &netdev->ifi_flags); |
6c88d577 | 989 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI; |
29cf9c1b FL |
990 | if (userspace_tso_enabled()) { |
991 | ifr.ifr_flags |= IFF_VNET_HDR; | |
992 | } | |
993 | ||
71d7c22f | 994 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
d0d08f8a | 995 | if (ioctl(netdev->tap_fd, TUNSETIFF, &ifr) == -1) { |
6c88d577 | 996 | VLOG_WARN("%s: creating tap device failed: %s", name, |
10a89ef0 | 997 | ovs_strerror(errno)); |
6c88d577 | 998 | error = errno; |
f61d8d29 | 999 | goto error_close; |
6c88d577 JP |
1000 | } |
1001 | ||
1002 | /* Make non-blocking. */ | |
d0d08f8a | 1003 | error = set_nonblocking(netdev->tap_fd); |
a740f0de | 1004 | if (error) { |
f61d8d29 | 1005 | goto error_close; |
a740f0de JG |
1006 | } |
1007 | ||
0f28164b FL |
1008 | if (ioctl(netdev->tap_fd, TUNSETPERSIST, 1)) { |
1009 | VLOG_WARN("%s: creating tap device failed (persist): %s", name, | |
1010 | ovs_strerror(errno)); | |
1011 | error = errno; | |
1012 | goto error_close; | |
1013 | } | |
1014 | ||
6211ad57 FL |
1015 | if (userspace_tso_enabled()) { |
1016 | /* Old kernels don't support TUNSETOFFLOAD. If TUNSETOFFLOAD is | |
1017 | * available, it will return EINVAL when a flag is unknown. | |
1018 | * Therefore, try enabling offload with no flags to check | |
1019 | * if TUNSETOFFLOAD support is available or not. */ | |
1020 | if (ioctl(netdev->tap_fd, TUNSETOFFLOAD, 0) == 0 || errno != EINVAL) { | |
1021 | unsigned long oflags = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6; | |
1022 | ||
1023 | if (ioctl(netdev->tap_fd, TUNSETOFFLOAD, oflags) == -1) { | |
1024 | VLOG_WARN("%s: enabling tap offloading failed: %s", name, | |
1025 | ovs_strerror(errno)); | |
1026 | error = errno; | |
1027 | goto error_close; | |
1028 | } | |
1029 | } | |
1030 | } | |
1031 | ||
19aac14a | 1032 | netdev->present = true; |
a740f0de JG |
1033 | return 0; |
1034 | ||
f61d8d29 | 1035 | error_close: |
d0d08f8a | 1036 | close(netdev->tap_fd); |
a740f0de JG |
1037 | return error; |
1038 | } | |
1039 | ||
6c88d577 | 1040 | static void |
9dc63482 | 1041 | netdev_linux_destruct(struct netdev *netdev_) |
6c88d577 | 1042 | { |
b5d57fc8 | 1043 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
6c88d577 | 1044 | |
b5d57fc8 BP |
1045 | if (netdev->tc && netdev->tc->ops->tc_destroy) { |
1046 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 BP |
1047 | } |
1048 | ||
d0d08f8a BP |
1049 | if (netdev_get_class(netdev_) == &netdev_tap_class |
1050 | && netdev->tap_fd >= 0) | |
1051 | { | |
0f28164b | 1052 | ioctl(netdev->tap_fd, TUNSETPERSIST, 0); |
d0d08f8a | 1053 | close(netdev->tap_fd); |
6c88d577 | 1054 | } |
86383816 | 1055 | |
19c8e9c1 | 1056 | if (netdev->miimon_interval > 0) { |
812c272c | 1057 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
1058 | } |
1059 | ||
86383816 | 1060 | ovs_mutex_destroy(&netdev->mutex); |
6c88d577 JP |
1061 | } |
1062 | ||
9dc63482 BP |
1063 | static void |
1064 | netdev_linux_dealloc(struct netdev *netdev_) | |
1065 | { | |
1066 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1067 | free(netdev); | |
1068 | } | |
1069 | ||
f7791740 PS |
1070 | static struct netdev_rxq * |
1071 | netdev_linux_rxq_alloc(void) | |
9dc63482 | 1072 | { |
f7791740 | 1073 | struct netdev_rxq_linux *rx = xzalloc(sizeof *rx); |
9dc63482 BP |
1074 | return &rx->up; |
1075 | } | |
1076 | ||
7b6b0ef4 | 1077 | static int |
f7791740 | 1078 | netdev_linux_rxq_construct(struct netdev_rxq *rxq_) |
7b6b0ef4 | 1079 | { |
f7791740 | 1080 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 1081 | struct netdev *netdev_ = rx->up.netdev; |
7b6b0ef4 | 1082 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7b6b0ef4 | 1083 | int error; |
7b6b0ef4 | 1084 | |
86383816 | 1085 | ovs_mutex_lock(&netdev->mutex); |
9dc63482 BP |
1086 | rx->is_tap = is_tap_netdev(netdev_); |
1087 | if (rx->is_tap) { | |
1088 | rx->fd = netdev->tap_fd; | |
796223f5 BP |
1089 | } else { |
1090 | struct sockaddr_ll sll; | |
b73c8518 | 1091 | int ifindex, val; |
32383c3b | 1092 | /* Result of tcpdump -dd inbound */ |
259e0b1a | 1093 | static const struct sock_filter filt[] = { |
32383c3b MM |
1094 | { 0x28, 0, 0, 0xfffff004 }, /* ldh [0] */ |
1095 | { 0x15, 0, 1, 0x00000004 }, /* jeq #4 jt 2 jf 3 */ | |
1096 | { 0x6, 0, 0, 0x00000000 }, /* ret #0 */ | |
1097 | { 0x6, 0, 0, 0x0000ffff } /* ret #65535 */ | |
1098 | }; | |
259e0b1a BP |
1099 | static const struct sock_fprog fprog = { |
1100 | ARRAY_SIZE(filt), (struct sock_filter *) filt | |
1101 | }; | |
7b6b0ef4 | 1102 | |
796223f5 | 1103 | /* Create file descriptor. */ |
9dc63482 BP |
1104 | rx->fd = socket(PF_PACKET, SOCK_RAW, 0); |
1105 | if (rx->fd < 0) { | |
796223f5 | 1106 | error = errno; |
10a89ef0 | 1107 | VLOG_ERR("failed to create raw socket (%s)", ovs_strerror(error)); |
796223f5 BP |
1108 | goto error; |
1109 | } | |
33d82a56 | 1110 | |
b73c8518 SH |
1111 | val = 1; |
1112 | if (setsockopt(rx->fd, SOL_PACKET, PACKET_AUXDATA, &val, sizeof val)) { | |
1113 | error = errno; | |
1114 | VLOG_ERR("%s: failed to mark socket for auxdata (%s)", | |
1115 | netdev_get_name(netdev_), ovs_strerror(error)); | |
1116 | goto error; | |
1117 | } | |
1118 | ||
29cf9c1b FL |
1119 | if (userspace_tso_enabled() |
1120 | && setsockopt(rx->fd, SOL_PACKET, PACKET_VNET_HDR, &val, | |
1121 | sizeof val)) { | |
1122 | error = errno; | |
1123 | VLOG_ERR("%s: failed to enable vnet hdr in txq raw socket: %s", | |
1124 | netdev_get_name(netdev_), ovs_strerror(errno)); | |
1125 | goto error; | |
1126 | } | |
1127 | ||
796223f5 | 1128 | /* Set non-blocking mode. */ |
9dc63482 | 1129 | error = set_nonblocking(rx->fd); |
796223f5 BP |
1130 | if (error) { |
1131 | goto error; | |
1132 | } | |
7b6b0ef4 | 1133 | |
796223f5 | 1134 | /* Get ethernet device index. */ |
180c6d0b | 1135 | error = get_ifindex(&netdev->up, &ifindex); |
796223f5 BP |
1136 | if (error) { |
1137 | goto error; | |
1138 | } | |
7b6b0ef4 | 1139 | |
796223f5 BP |
1140 | /* Bind to specific ethernet device. */ |
1141 | memset(&sll, 0, sizeof sll); | |
1142 | sll.sll_family = AF_PACKET; | |
1143 | sll.sll_ifindex = ifindex; | |
b73c8518 | 1144 | sll.sll_protocol = htons(ETH_P_ALL); |
9dc63482 | 1145 | if (bind(rx->fd, (struct sockaddr *) &sll, sizeof sll) < 0) { |
796223f5 BP |
1146 | error = errno; |
1147 | VLOG_ERR("%s: failed to bind raw socket (%s)", | |
10a89ef0 | 1148 | netdev_get_name(netdev_), ovs_strerror(error)); |
796223f5 BP |
1149 | goto error; |
1150 | } | |
32383c3b MM |
1151 | |
1152 | /* Filter for only inbound packets. */ | |
9dc63482 | 1153 | error = setsockopt(rx->fd, SOL_SOCKET, SO_ATTACH_FILTER, &fprog, |
32383c3b MM |
1154 | sizeof fprog); |
1155 | if (error) { | |
1156 | error = errno; | |
259e0b1a | 1157 | VLOG_ERR("%s: failed to attach filter (%s)", |
10a89ef0 | 1158 | netdev_get_name(netdev_), ovs_strerror(error)); |
32383c3b MM |
1159 | goto error; |
1160 | } | |
7b6b0ef4 | 1161 | } |
86383816 | 1162 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 | 1163 | |
7b6b0ef4 BP |
1164 | return 0; |
1165 | ||
1166 | error: | |
9dc63482 BP |
1167 | if (rx->fd >= 0) { |
1168 | close(rx->fd); | |
7b6b0ef4 | 1169 | } |
86383816 | 1170 | ovs_mutex_unlock(&netdev->mutex); |
7b6b0ef4 BP |
1171 | return error; |
1172 | } | |
1173 | ||
796223f5 | 1174 | static void |
f7791740 | 1175 | netdev_linux_rxq_destruct(struct netdev_rxq *rxq_) |
8b61709d | 1176 | { |
f7791740 | 1177 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
29cf9c1b | 1178 | int i; |
8b61709d | 1179 | |
796223f5 BP |
1180 | if (!rx->is_tap) { |
1181 | close(rx->fd); | |
8b61709d | 1182 | } |
29cf9c1b FL |
1183 | |
1184 | for (i = 0; i < NETDEV_MAX_BURST; i++) { | |
73858f9d | 1185 | dp_packet_delete(rx->aux_bufs[i]); |
29cf9c1b | 1186 | } |
9dc63482 BP |
1187 | } |
1188 | ||
1189 | static void | |
f7791740 | 1190 | netdev_linux_rxq_dealloc(struct netdev_rxq *rxq_) |
9dc63482 | 1191 | { |
f7791740 | 1192 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
9dc63482 | 1193 | |
796223f5 BP |
1194 | free(rx); |
1195 | } | |
8b61709d | 1196 | |
b73c8518 | 1197 | static ovs_be16 |
1ebdc7eb | 1198 | auxdata_to_vlan_tpid(const struct tpacket_auxdata *aux, bool double_tagged) |
b73c8518 SH |
1199 | { |
1200 | if (aux->tp_status & TP_STATUS_VLAN_TPID_VALID) { | |
1201 | return htons(aux->tp_vlan_tpid); | |
1ebdc7eb EG |
1202 | } else if (double_tagged) { |
1203 | return htons(ETH_TYPE_VLAN_8021AD); | |
b73c8518 | 1204 | } else { |
1ebdc7eb | 1205 | return htons(ETH_TYPE_VLAN_8021Q); |
b73c8518 SH |
1206 | } |
1207 | } | |
1208 | ||
1209 | static bool | |
1210 | auxdata_has_vlan_tci(const struct tpacket_auxdata *aux) | |
1211 | { | |
1212 | return aux->tp_vlan_tci || aux->tp_status & TP_STATUS_VLAN_VALID; | |
1213 | } | |
1214 | ||
2109841b YY |
1215 | /* |
1216 | * Receive packets from raw socket in batch process for better performance, | |
1217 | * it can receive NETDEV_MAX_BURST packets at most once, the received | |
1218 | * packets are added into *batch. The return value is 0 or errno. | |
1219 | * | |
1220 | * It also used recvmmsg to reduce multiple syscalls overhead; | |
1221 | */ | |
796223f5 | 1222 | static int |
29cf9c1b | 1223 | netdev_linux_batch_rxq_recv_sock(struct netdev_rxq_linux *rx, int mtu, |
2109841b | 1224 | struct dp_packet_batch *batch) |
796223f5 | 1225 | { |
29cf9c1b FL |
1226 | int iovlen; |
1227 | size_t std_len; | |
796223f5 | 1228 | ssize_t retval; |
29cf9c1b FL |
1229 | int virtio_net_hdr_size; |
1230 | struct iovec iovs[NETDEV_MAX_BURST][IOV_TSO_SIZE]; | |
b73c8518 SH |
1231 | struct cmsghdr *cmsg; |
1232 | union { | |
1233 | struct cmsghdr cmsg; | |
1234 | char buffer[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; | |
2109841b YY |
1235 | } cmsg_buffers[NETDEV_MAX_BURST]; |
1236 | struct mmsghdr mmsgs[NETDEV_MAX_BURST]; | |
1237 | struct dp_packet *buffers[NETDEV_MAX_BURST]; | |
1238 | int i; | |
1239 | ||
29cf9c1b FL |
1240 | if (userspace_tso_enabled()) { |
1241 | /* Use the buffer from the allocated packet below to receive MTU | |
1242 | * sized packets and an aux_buf for extra TSO data. */ | |
1243 | iovlen = IOV_TSO_SIZE; | |
1244 | virtio_net_hdr_size = sizeof(struct virtio_net_hdr); | |
1245 | } else { | |
1246 | /* Use only the buffer from the allocated packet. */ | |
1247 | iovlen = IOV_STD_SIZE; | |
1248 | virtio_net_hdr_size = 0; | |
1249 | } | |
1250 | ||
73858f9d FL |
1251 | /* The length here needs to be accounted in the same way when the |
1252 | * aux_buf is allocated so that it can be prepended to TSO buffer. */ | |
1253 | std_len = virtio_net_hdr_size + VLAN_ETH_HEADER_LEN + mtu; | |
2109841b | 1254 | for (i = 0; i < NETDEV_MAX_BURST; i++) { |
29cf9c1b FL |
1255 | buffers[i] = dp_packet_new_with_headroom(std_len, DP_NETDEV_HEADROOM); |
1256 | iovs[i][IOV_PACKET].iov_base = dp_packet_data(buffers[i]); | |
1257 | iovs[i][IOV_PACKET].iov_len = std_len; | |
73858f9d FL |
1258 | if (iovlen == IOV_TSO_SIZE) { |
1259 | iovs[i][IOV_AUXBUF].iov_base = dp_packet_data(rx->aux_bufs[i]); | |
1260 | iovs[i][IOV_AUXBUF].iov_len = dp_packet_tailroom(rx->aux_bufs[i]); | |
1261 | } | |
1262 | ||
2109841b YY |
1263 | mmsgs[i].msg_hdr.msg_name = NULL; |
1264 | mmsgs[i].msg_hdr.msg_namelen = 0; | |
29cf9c1b FL |
1265 | mmsgs[i].msg_hdr.msg_iov = iovs[i]; |
1266 | mmsgs[i].msg_hdr.msg_iovlen = iovlen; | |
2109841b YY |
1267 | mmsgs[i].msg_hdr.msg_control = &cmsg_buffers[i]; |
1268 | mmsgs[i].msg_hdr.msg_controllen = sizeof cmsg_buffers[i]; | |
1269 | mmsgs[i].msg_hdr.msg_flags = 0; | |
1270 | } | |
8e8cddf7 | 1271 | |
796223f5 | 1272 | do { |
29cf9c1b | 1273 | retval = recvmmsg(rx->fd, mmsgs, NETDEV_MAX_BURST, MSG_TRUNC, NULL); |
796223f5 BP |
1274 | } while (retval < 0 && errno == EINTR); |
1275 | ||
bfd3367b | 1276 | if (retval < 0) { |
29cf9c1b FL |
1277 | retval = errno; |
1278 | for (i = 0; i < NETDEV_MAX_BURST; i++) { | |
1279 | dp_packet_delete(buffers[i]); | |
1280 | } | |
1281 | ||
1282 | return retval; | |
b73c8518 SH |
1283 | } |
1284 | ||
2109841b | 1285 | for (i = 0; i < retval; i++) { |
73858f9d FL |
1286 | struct dp_packet *pkt; |
1287 | ||
2109841b | 1288 | if (mmsgs[i].msg_len < ETH_HEADER_LEN) { |
29cf9c1b FL |
1289 | struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up); |
1290 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1291 | ||
1292 | dp_packet_delete(buffers[i]); | |
1293 | netdev->rx_dropped += 1; | |
1294 | VLOG_WARN_RL(&rl, "%s: Dropped packet: less than ether hdr size", | |
1295 | netdev_get_name(netdev_)); | |
1296 | continue; | |
1297 | } | |
1298 | ||
1299 | if (mmsgs[i].msg_len > std_len) { | |
73858f9d FL |
1300 | /* Build a single linear TSO packet by prepending the data from |
1301 | * std_len buffer to the aux_buf. */ | |
1302 | pkt = rx->aux_bufs[i]; | |
1303 | dp_packet_set_size(pkt, mmsgs[i].msg_len - std_len); | |
1304 | dp_packet_push(pkt, dp_packet_data(buffers[i]), std_len); | |
1305 | /* The headroom should be the same in buffers[i], pkt and | |
1306 | * DP_NETDEV_HEADROOM. */ | |
1307 | dp_packet_resize(pkt, DP_NETDEV_HEADROOM, 0); | |
1308 | dp_packet_delete(buffers[i]); | |
1309 | rx->aux_bufs[i] = NULL; | |
1310 | } else { | |
1311 | dp_packet_set_size(buffers[i], mmsgs[i].msg_len); | |
1312 | pkt = buffers[i]; | |
1313 | } | |
b73c8518 | 1314 | |
73858f9d | 1315 | if (virtio_net_hdr_size && netdev_linux_parse_vnet_hdr(pkt)) { |
29cf9c1b FL |
1316 | struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up); |
1317 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1318 | ||
1319 | /* Unexpected error situation: the virtio header is not present | |
1320 | * or corrupted. Drop the packet but continue in case next ones | |
1321 | * are correct. */ | |
73858f9d | 1322 | dp_packet_delete(pkt); |
29cf9c1b FL |
1323 | netdev->rx_dropped += 1; |
1324 | VLOG_WARN_RL(&rl, "%s: Dropped packet: Invalid virtio net header", | |
1325 | netdev_get_name(netdev_)); | |
1326 | continue; | |
1327 | } | |
1ebdc7eb | 1328 | |
2109841b YY |
1329 | for (cmsg = CMSG_FIRSTHDR(&mmsgs[i].msg_hdr); cmsg; |
1330 | cmsg = CMSG_NXTHDR(&mmsgs[i].msg_hdr, cmsg)) { | |
1331 | const struct tpacket_auxdata *aux; | |
1332 | ||
1333 | if (cmsg->cmsg_level != SOL_PACKET | |
1334 | || cmsg->cmsg_type != PACKET_AUXDATA | |
1335 | || cmsg->cmsg_len < | |
1336 | CMSG_LEN(sizeof(struct tpacket_auxdata))) { | |
1337 | continue; | |
b73c8518 SH |
1338 | } |
1339 | ||
2109841b YY |
1340 | aux = ALIGNED_CAST(struct tpacket_auxdata *, CMSG_DATA(cmsg)); |
1341 | if (auxdata_has_vlan_tci(aux)) { | |
1342 | struct eth_header *eth; | |
1343 | bool double_tagged; | |
1ebdc7eb | 1344 | |
73858f9d | 1345 | eth = dp_packet_data(pkt); |
2109841b YY |
1346 | double_tagged = eth->eth_type == htons(ETH_TYPE_VLAN_8021Q); |
1347 | ||
73858f9d | 1348 | eth_push_vlan(pkt, |
2109841b YY |
1349 | auxdata_to_vlan_tpid(aux, double_tagged), |
1350 | htons(aux->tp_vlan_tci)); | |
1351 | break; | |
1352 | } | |
b73c8518 | 1353 | } |
73858f9d | 1354 | dp_packet_batch_add(batch, pkt); |
2109841b YY |
1355 | } |
1356 | ||
29cf9c1b | 1357 | /* Delete unused buffers. */ |
2109841b YY |
1358 | for (; i < NETDEV_MAX_BURST; i++) { |
1359 | dp_packet_delete(buffers[i]); | |
1360 | } | |
1361 | ||
b73c8518 SH |
1362 | return 0; |
1363 | } | |
1364 | ||
2109841b YY |
1365 | /* |
1366 | * Receive packets from tap by batch process for better performance, | |
1367 | * it can receive NETDEV_MAX_BURST packets at most once, the received | |
1368 | * packets are added into *batch. The return value is 0 or errno. | |
1369 | */ | |
b73c8518 | 1370 | static int |
29cf9c1b FL |
1371 | netdev_linux_batch_rxq_recv_tap(struct netdev_rxq_linux *rx, int mtu, |
1372 | struct dp_packet_batch *batch) | |
b73c8518 | 1373 | { |
29cf9c1b | 1374 | int virtio_net_hdr_size; |
b73c8518 | 1375 | ssize_t retval; |
29cf9c1b FL |
1376 | size_t std_len; |
1377 | int iovlen; | |
2109841b YY |
1378 | int i; |
1379 | ||
29cf9c1b FL |
1380 | if (userspace_tso_enabled()) { |
1381 | /* Use the buffer from the allocated packet below to receive MTU | |
1382 | * sized packets and an aux_buf for extra TSO data. */ | |
1383 | iovlen = IOV_TSO_SIZE; | |
1384 | virtio_net_hdr_size = sizeof(struct virtio_net_hdr); | |
1385 | } else { | |
1386 | /* Use only the buffer from the allocated packet. */ | |
1387 | iovlen = IOV_STD_SIZE; | |
1388 | virtio_net_hdr_size = 0; | |
1389 | } | |
1390 | ||
73858f9d FL |
1391 | /* The length here needs to be accounted in the same way when the |
1392 | * aux_buf is allocated so that it can be prepended to TSO buffer. */ | |
1393 | std_len = virtio_net_hdr_size + VLAN_ETH_HEADER_LEN + mtu; | |
2109841b | 1394 | for (i = 0; i < NETDEV_MAX_BURST; i++) { |
73858f9d FL |
1395 | struct dp_packet *buffer; |
1396 | struct dp_packet *pkt; | |
29cf9c1b FL |
1397 | struct iovec iov[IOV_TSO_SIZE]; |
1398 | ||
2109841b | 1399 | /* Assume Ethernet port. No need to set packet_type. */ |
29cf9c1b FL |
1400 | buffer = dp_packet_new_with_headroom(std_len, DP_NETDEV_HEADROOM); |
1401 | iov[IOV_PACKET].iov_base = dp_packet_data(buffer); | |
1402 | iov[IOV_PACKET].iov_len = std_len; | |
73858f9d FL |
1403 | if (iovlen == IOV_TSO_SIZE) { |
1404 | iov[IOV_AUXBUF].iov_base = dp_packet_data(rx->aux_bufs[i]); | |
1405 | iov[IOV_AUXBUF].iov_len = dp_packet_tailroom(rx->aux_bufs[i]); | |
1406 | } | |
29cf9c1b | 1407 | |
2109841b | 1408 | do { |
29cf9c1b | 1409 | retval = readv(rx->fd, iov, iovlen); |
2109841b YY |
1410 | } while (retval < 0 && errno == EINTR); |
1411 | ||
1412 | if (retval < 0) { | |
1413 | dp_packet_delete(buffer); | |
1414 | break; | |
1415 | } | |
b73c8518 | 1416 | |
29cf9c1b | 1417 | if (retval > std_len) { |
73858f9d FL |
1418 | /* Build a single linear TSO packet by prepending the data from |
1419 | * std_len buffer to the aux_buf. */ | |
1420 | pkt = rx->aux_bufs[i]; | |
1421 | dp_packet_set_size(pkt, retval - std_len); | |
1422 | dp_packet_push(pkt, dp_packet_data(buffer), std_len); | |
1423 | /* The headroom should be the same in buffers[i], pkt and | |
1424 | * DP_NETDEV_HEADROOM. */ | |
1425 | dp_packet_resize(pkt, DP_NETDEV_HEADROOM, 0); | |
1426 | dp_packet_delete(buffer); | |
1427 | rx->aux_bufs[i] = NULL; | |
29cf9c1b FL |
1428 | } else { |
1429 | dp_packet_set_size(buffer, dp_packet_size(buffer) + retval); | |
73858f9d | 1430 | pkt = buffer; |
29cf9c1b FL |
1431 | } |
1432 | ||
73858f9d | 1433 | if (virtio_net_hdr_size && netdev_linux_parse_vnet_hdr(pkt)) { |
29cf9c1b FL |
1434 | struct netdev *netdev_ = netdev_rxq_get_netdev(&rx->up); |
1435 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1436 | ||
1437 | /* Unexpected error situation: the virtio header is not present | |
1438 | * or corrupted. Drop the packet but continue in case next ones | |
1439 | * are correct. */ | |
73858f9d | 1440 | dp_packet_delete(pkt); |
29cf9c1b FL |
1441 | netdev->rx_dropped += 1; |
1442 | VLOG_WARN_RL(&rl, "%s: Dropped packet: Invalid virtio net header", | |
1443 | netdev_get_name(netdev_)); | |
1444 | continue; | |
1445 | } | |
1446 | ||
73858f9d | 1447 | dp_packet_batch_add(batch, pkt); |
2109841b | 1448 | } |
b73c8518 | 1449 | |
2109841b | 1450 | if ((i == 0) && (retval < 0)) { |
bfd3367b | 1451 | return errno; |
8b61709d | 1452 | } |
b73c8518 | 1453 | |
b73c8518 SH |
1454 | return 0; |
1455 | } | |
1456 | ||
1457 | static int | |
8492adc2 JS |
1458 | netdev_linux_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet_batch *batch, |
1459 | int *qfill) | |
b73c8518 | 1460 | { |
f7791740 | 1461 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
df1e5a3b | 1462 | struct netdev *netdev = rx->up.netdev; |
df1e5a3b PS |
1463 | ssize_t retval; |
1464 | int mtu; | |
1465 | ||
1466 | if (netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
1467 | mtu = ETH_PAYLOAD_MAX; | |
1468 | } | |
1469 | ||
73858f9d FL |
1470 | if (userspace_tso_enabled()) { |
1471 | /* Allocate TSO packets. The packet has enough headroom to store | |
1472 | * a full non-TSO packet. When a TSO packet is received, the data | |
1473 | * from non-TSO buffer (std_len) is prepended to the TSO packet | |
1474 | * (aux_buf). */ | |
1475 | size_t std_len = sizeof(struct virtio_net_hdr) + VLAN_ETH_HEADER_LEN | |
1476 | + DP_NETDEV_HEADROOM + mtu; | |
1477 | size_t data_len = LINUX_RXQ_TSO_MAX_LEN - std_len; | |
1478 | for (int i = 0; i < NETDEV_MAX_BURST; i++) { | |
1479 | if (rx->aux_bufs[i]) { | |
1480 | continue; | |
1481 | } | |
1482 | ||
1483 | rx->aux_bufs[i] = dp_packet_new_with_headroom(data_len, std_len); | |
1484 | } | |
1485 | } | |
1486 | ||
2109841b | 1487 | dp_packet_batch_init(batch); |
b73c8518 | 1488 | retval = (rx->is_tap |
29cf9c1b FL |
1489 | ? netdev_linux_batch_rxq_recv_tap(rx, mtu, batch) |
1490 | : netdev_linux_batch_rxq_recv_sock(rx, mtu, batch)); | |
df1e5a3b PS |
1491 | |
1492 | if (retval) { | |
1493 | if (retval != EAGAIN && retval != EMSGSIZE) { | |
1494 | VLOG_WARN_RL(&rl, "error receiving Ethernet packet on %s: %s", | |
7c6401ca | 1495 | netdev_rxq_get_name(rxq_), ovs_strerror(errno)); |
df1e5a3b | 1496 | } |
b73c8518 SH |
1497 | } |
1498 | ||
8492adc2 JS |
1499 | if (qfill) { |
1500 | *qfill = -ENOTSUP; | |
1501 | } | |
1502 | ||
b73c8518 | 1503 | return retval; |
8b61709d BP |
1504 | } |
1505 | ||
8b61709d | 1506 | static void |
f7791740 | 1507 | netdev_linux_rxq_wait(struct netdev_rxq *rxq_) |
8b61709d | 1508 | { |
f7791740 | 1509 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1510 | poll_fd_wait(rx->fd, POLLIN); |
8b61709d BP |
1511 | } |
1512 | ||
8b61709d | 1513 | static int |
f7791740 | 1514 | netdev_linux_rxq_drain(struct netdev_rxq *rxq_) |
8b61709d | 1515 | { |
f7791740 | 1516 | struct netdev_rxq_linux *rx = netdev_rxq_linux_cast(rxq_); |
796223f5 | 1517 | if (rx->is_tap) { |
8b61709d | 1518 | struct ifreq ifr; |
f7791740 | 1519 | int error = af_inet_ifreq_ioctl(netdev_rxq_get_name(rxq_), &ifr, |
259e0b1a | 1520 | SIOCGIFTXQLEN, "SIOCGIFTXQLEN"); |
8b61709d BP |
1521 | if (error) { |
1522 | return error; | |
1523 | } | |
796223f5 | 1524 | drain_fd(rx->fd, ifr.ifr_qlen); |
8b61709d BP |
1525 | return 0; |
1526 | } else { | |
796223f5 | 1527 | return drain_rcvbuf(rx->fd); |
8b61709d BP |
1528 | } |
1529 | } | |
1530 | ||
d19cf8bb | 1531 | static int |
29cf9c1b | 1532 | netdev_linux_sock_batch_send(int sock, int ifindex, bool tso, int mtu, |
d19cf8bb ZG |
1533 | struct dp_packet_batch *batch) |
1534 | { | |
e0a00cee | 1535 | const size_t size = dp_packet_batch_size(batch); |
d19cf8bb ZG |
1536 | /* We don't bother setting most fields in sockaddr_ll because the |
1537 | * kernel ignores them for SOCK_RAW. */ | |
1538 | struct sockaddr_ll sll = { .sll_family = AF_PACKET, | |
1539 | .sll_ifindex = ifindex }; | |
1540 | ||
e0a00cee BB |
1541 | struct mmsghdr *mmsg = xmalloc(sizeof(*mmsg) * size); |
1542 | struct iovec *iov = xmalloc(sizeof(*iov) * size); | |
d19cf8bb | 1543 | |
e0a00cee | 1544 | struct dp_packet *packet; |
e883448e | 1545 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
29cf9c1b FL |
1546 | if (tso) { |
1547 | netdev_linux_prepend_vnet_hdr(packet, mtu); | |
1548 | } | |
1549 | ||
d19cf8bb | 1550 | iov[i].iov_base = dp_packet_data(packet); |
ad8b0b4f | 1551 | iov[i].iov_len = dp_packet_size(packet); |
d19cf8bb ZG |
1552 | mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll, |
1553 | .msg_namelen = sizeof sll, | |
1554 | .msg_iov = &iov[i], | |
1555 | .msg_iovlen = 1 }; | |
1556 | } | |
1557 | ||
1558 | int error = 0; | |
e0a00cee | 1559 | for (uint32_t ofs = 0; ofs < size; ) { |
d19cf8bb ZG |
1560 | ssize_t retval; |
1561 | do { | |
e0a00cee | 1562 | retval = sendmmsg(sock, mmsg + ofs, size - ofs, 0); |
d19cf8bb ZG |
1563 | error = retval < 0 ? errno : 0; |
1564 | } while (error == EINTR); | |
1565 | if (error) { | |
1566 | break; | |
1567 | } | |
1568 | ofs += retval; | |
1569 | } | |
1570 | ||
1571 | free(mmsg); | |
1572 | free(iov); | |
1573 | return error; | |
1574 | } | |
1575 | ||
1576 | /* Use the tap fd to send 'batch' to tap device 'netdev'. Using the tap fd is | |
1577 | * essential, because packets sent to a tap device with an AF_PACKET socket | |
1578 | * will loop back to be *received* again on the tap device. This doesn't occur | |
1579 | * on other interface types because we attach a socket filter to the rx | |
1580 | * socket. */ | |
1581 | static int | |
29cf9c1b | 1582 | netdev_linux_tap_batch_send(struct netdev *netdev_, bool tso, int mtu, |
d19cf8bb ZG |
1583 | struct dp_packet_batch *batch) |
1584 | { | |
1585 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
13708b21 | 1586 | struct dp_packet *packet; |
22dcb534 FL |
1587 | |
1588 | /* The Linux tap driver returns EIO if the device is not up, | |
1589 | * so if the device is not up, don't waste time sending it. | |
1590 | * However, if the device is in another network namespace | |
1591 | * then OVS can't retrieve the state. In that case, send the | |
1592 | * packets anyway. */ | |
1593 | if (netdev->present && !(netdev->ifi_flags & IFF_UP)) { | |
1594 | netdev->tx_dropped += dp_packet_batch_size(batch); | |
1595 | return 0; | |
1596 | } | |
1597 | ||
e883448e | 1598 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
29cf9c1b | 1599 | size_t size; |
d19cf8bb ZG |
1600 | ssize_t retval; |
1601 | int error; | |
1602 | ||
29cf9c1b FL |
1603 | if (tso) { |
1604 | netdev_linux_prepend_vnet_hdr(packet, mtu); | |
1605 | } | |
1606 | ||
1607 | size = dp_packet_size(packet); | |
d19cf8bb ZG |
1608 | do { |
1609 | retval = write(netdev->tap_fd, dp_packet_data(packet), size); | |
1610 | error = retval < 0 ? errno : 0; | |
1611 | } while (error == EINTR); | |
1612 | ||
1613 | if (error) { | |
1614 | /* The Linux tap driver returns EIO if the device is not up. From | |
1615 | * the OVS side this is not an error, so we ignore it; otherwise, | |
1616 | * return the erro. */ | |
1617 | if (error != EIO) { | |
1618 | return error; | |
1619 | } | |
1620 | } else if (retval != size) { | |
1621 | VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" " | |
1622 | "bytes of %"PRIuSIZE") on %s", | |
1623 | retval, size, netdev_get_name(netdev_)); | |
1624 | return EMSGSIZE; | |
1625 | } | |
1626 | } | |
1627 | return 0; | |
1628 | } | |
1629 | ||
105cf8df WT |
1630 | static int |
1631 | netdev_linux_get_numa_id__(struct netdev_linux *netdev) | |
1632 | OVS_REQUIRES(netdev->mutex) | |
1633 | { | |
1634 | char *numa_node_path; | |
1635 | const char *name; | |
1636 | int node_id; | |
1637 | FILE *stream; | |
1638 | ||
1639 | if (netdev->cache_valid & VALID_NUMA_ID) { | |
1640 | return netdev->numa_id; | |
1641 | } | |
1642 | ||
1643 | netdev->numa_id = 0; | |
1644 | netdev->cache_valid |= VALID_NUMA_ID; | |
1645 | ||
1646 | if (ovs_numa_get_n_numas() < 2) { | |
1647 | /* No need to check on system with a single NUMA node. */ | |
1648 | return 0; | |
1649 | } | |
1650 | ||
1651 | name = netdev_get_name(&netdev->up); | |
1652 | if (strpbrk(name, "/\\")) { | |
1653 | VLOG_ERR_RL(&rl, "\"%s\" is not a valid name for a port. " | |
1654 | "A valid name must not include '/' or '\\'." | |
1655 | "Using numa_id 0", name); | |
1656 | return 0; | |
1657 | } | |
1658 | ||
1659 | numa_node_path = xasprintf("/sys/class/net/%s/device/numa_node", name); | |
1660 | ||
1661 | stream = fopen(numa_node_path, "r"); | |
1662 | if (!stream) { | |
1663 | /* Virtual device does not have this info. */ | |
1664 | VLOG_INFO_RL(&rl, "%s: Can't open '%s': %s, using numa_id 0", | |
1665 | name, numa_node_path, ovs_strerror(errno)); | |
1666 | free(numa_node_path); | |
1667 | return 0; | |
1668 | } | |
1669 | ||
1670 | if (fscanf(stream, "%d", &node_id) != 1 | |
1671 | || !ovs_numa_numa_id_is_valid(node_id)) { | |
1672 | VLOG_WARN_RL(&rl, "%s: Can't detect NUMA node, using numa_id 0", name); | |
1673 | node_id = 0; | |
1674 | } | |
1675 | ||
1676 | netdev->numa_id = node_id; | |
1677 | fclose(stream); | |
1678 | free(numa_node_path); | |
1679 | return node_id; | |
1680 | } | |
1681 | ||
1682 | static int OVS_UNUSED | |
1683 | netdev_linux_get_numa_id(const struct netdev *netdev_) | |
1684 | { | |
1685 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1686 | int numa_id; | |
1687 | ||
1688 | ovs_mutex_lock(&netdev->mutex); | |
1689 | numa_id = netdev_linux_get_numa_id__(netdev); | |
1690 | ovs_mutex_unlock(&netdev->mutex); | |
1691 | ||
1692 | return numa_id; | |
1693 | } | |
1694 | ||
d19cf8bb | 1695 | /* Sends 'batch' on 'netdev'. Returns 0 if successful, otherwise a positive |
8b61709d BP |
1696 | * errno value. Returns EAGAIN without blocking if the packet cannot be queued |
1697 | * immediately. Returns EMSGSIZE if a partial packet was transmitted or if | |
1698 | * the packet is too big or too small to transmit on the device. | |
1699 | * | |
8b61709d BP |
1700 | * The kernel maintains a packet transmission queue, so the caller is not |
1701 | * expected to do additional queuing of packets. */ | |
1702 | static int | |
f00fa8cb | 1703 | netdev_linux_send(struct netdev *netdev_, int qid OVS_UNUSED, |
b30896c9 | 1704 | struct dp_packet_batch *batch, |
324c8374 | 1705 | bool concurrent_txq OVS_UNUSED) |
8b61709d | 1706 | { |
29cf9c1b FL |
1707 | bool tso = userspace_tso_enabled(); |
1708 | int mtu = ETH_PAYLOAD_MAX; | |
f4fd623c | 1709 | int error = 0; |
0a62ae2c ZG |
1710 | int sock = 0; |
1711 | ||
29cf9c1b FL |
1712 | if (tso) { |
1713 | netdev_linux_get_mtu__(netdev_linux_cast(netdev_), &mtu); | |
1714 | } | |
1715 | ||
0a62ae2c | 1716 | if (!is_tap_netdev(netdev_)) { |
e0e2410d FL |
1717 | if (netdev_linux_netnsid_is_remote(netdev_linux_cast(netdev_))) { |
1718 | error = EOPNOTSUPP; | |
1719 | goto free_batch; | |
1720 | } | |
1721 | ||
0a62ae2c ZG |
1722 | sock = af_packet_sock(); |
1723 | if (sock < 0) { | |
1724 | error = -sock; | |
1725 | goto free_batch; | |
1726 | } | |
1727 | ||
1728 | int ifindex = netdev_get_ifindex(netdev_); | |
1729 | if (ifindex < 0) { | |
1730 | error = -ifindex; | |
1731 | goto free_batch; | |
1732 | } | |
1733 | ||
29cf9c1b | 1734 | error = netdev_linux_sock_batch_send(sock, ifindex, tso, mtu, batch); |
d19cf8bb | 1735 | } else { |
29cf9c1b | 1736 | error = netdev_linux_tap_batch_send(netdev_, tso, mtu, batch); |
0a62ae2c | 1737 | } |
d19cf8bb ZG |
1738 | if (error) { |
1739 | if (error == ENOBUFS) { | |
1740 | /* The Linux AF_PACKET implementation never blocks waiting | |
1741 | * for room for packets, instead returning ENOBUFS. | |
1742 | * Translate this into EAGAIN for the caller. */ | |
1743 | error = EAGAIN; | |
f23347ea | 1744 | } else { |
f4fd623c DDP |
1745 | VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s", |
1746 | netdev_get_name(netdev_), ovs_strerror(error)); | |
d19cf8bb | 1747 | } |
f4fd623c DDP |
1748 | } |
1749 | ||
0a62ae2c | 1750 | free_batch: |
b30896c9 | 1751 | dp_packet_delete_batch(batch, true); |
f4fd623c | 1752 | return error; |
8b61709d BP |
1753 | } |
1754 | ||
1755 | /* Registers with the poll loop to wake up from the next call to poll_block() | |
1756 | * when the packet transmission queue has sufficient room to transmit a packet | |
1757 | * with netdev_send(). | |
1758 | * | |
1759 | * The kernel maintains a packet transmission queue, so the client is not | |
1760 | * expected to do additional queuing of packets. Thus, this function is | |
1761 | * unlikely to ever be used. It is included for completeness. */ | |
1762 | static void | |
f00fa8cb | 1763 | netdev_linux_send_wait(struct netdev *netdev, int qid OVS_UNUSED) |
8b61709d | 1764 | { |
796223f5 | 1765 | if (is_tap_netdev(netdev)) { |
8b61709d BP |
1766 | /* TAP device always accepts packets.*/ |
1767 | poll_immediate_wake(); | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, | |
1772 | * otherwise a positive errno value. */ | |
1773 | static int | |
74ff3298 | 1774 | netdev_linux_set_etheraddr(struct netdev *netdev_, const struct eth_addr mac) |
8b61709d | 1775 | { |
b5d57fc8 | 1776 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4f9f3f21 | 1777 | enum netdev_flags old_flags = 0; |
eb395f2e BP |
1778 | int error; |
1779 | ||
86383816 | 1780 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1781 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1782 | error = EOPNOTSUPP; | |
1783 | goto exit; | |
1784 | } | |
86383816 | 1785 | |
b5d57fc8 | 1786 | if (netdev->cache_valid & VALID_ETHERADDR) { |
86383816 BP |
1787 | error = netdev->ether_addr_error; |
1788 | if (error || eth_addr_equals(netdev->etheraddr, mac)) { | |
1789 | goto exit; | |
44445cac | 1790 | } |
b5d57fc8 | 1791 | netdev->cache_valid &= ~VALID_ETHERADDR; |
44445cac PS |
1792 | } |
1793 | ||
7eb1bd81 | 1794 | /* Tap devices must be brought down before setting the address. */ |
796223f5 | 1795 | if (is_tap_netdev(netdev_)) { |
4f9f3f21 | 1796 | update_flags(netdev, NETDEV_UP, 0, &old_flags); |
7eb1bd81 | 1797 | } |
44445cac PS |
1798 | error = set_etheraddr(netdev_get_name(netdev_), mac); |
1799 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
1800 | netdev->ether_addr_error = error; |
1801 | netdev->cache_valid |= VALID_ETHERADDR; | |
eb395f2e | 1802 | if (!error) { |
74ff3298 | 1803 | netdev->etheraddr = mac; |
eb395f2e | 1804 | } |
8b61709d | 1805 | } |
44445cac | 1806 | |
4f9f3f21 BP |
1807 | if (is_tap_netdev(netdev_) && old_flags & NETDEV_UP) { |
1808 | update_flags(netdev, 0, NETDEV_UP, &old_flags); | |
1809 | } | |
7eb1bd81 | 1810 | |
86383816 BP |
1811 | exit: |
1812 | ovs_mutex_unlock(&netdev->mutex); | |
8b61709d BP |
1813 | return error; |
1814 | } | |
1815 | ||
44445cac | 1816 | /* Copies 'netdev''s MAC address to 'mac' which is passed as param. */ |
8b61709d | 1817 | static int |
74ff3298 | 1818 | netdev_linux_get_etheraddr(const struct netdev *netdev_, struct eth_addr *mac) |
8b61709d | 1819 | { |
b5d57fc8 | 1820 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 1821 | int error; |
44445cac | 1822 | |
86383816 | 1823 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 | 1824 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { |
756819dd FL |
1825 | netdev_linux_update_via_netlink(netdev); |
1826 | } | |
1827 | ||
1828 | if (!(netdev->cache_valid & VALID_ETHERADDR)) { | |
1829 | /* Fall back to ioctl if netlink fails */ | |
86383816 | 1830 | netdev->ether_addr_error = get_etheraddr(netdev_get_name(netdev_), |
74ff3298 | 1831 | &netdev->etheraddr); |
b5d57fc8 | 1832 | netdev->cache_valid |= VALID_ETHERADDR; |
8b61709d | 1833 | } |
44445cac | 1834 | |
86383816 BP |
1835 | error = netdev->ether_addr_error; |
1836 | if (!error) { | |
74ff3298 | 1837 | *mac = netdev->etheraddr; |
44445cac | 1838 | } |
86383816 | 1839 | ovs_mutex_unlock(&netdev->mutex); |
44445cac | 1840 | |
86383816 | 1841 | return error; |
8b61709d BP |
1842 | } |
1843 | ||
8b61709d | 1844 | static int |
73371c09 | 1845 | netdev_linux_get_mtu__(struct netdev_linux *netdev, int *mtup) |
8b61709d | 1846 | { |
86383816 BP |
1847 | int error; |
1848 | ||
b5d57fc8 | 1849 | if (!(netdev->cache_valid & VALID_MTU)) { |
756819dd FL |
1850 | netdev_linux_update_via_netlink(netdev); |
1851 | } | |
1852 | ||
1853 | if (!(netdev->cache_valid & VALID_MTU)) { | |
1854 | /* Fall back to ioctl if netlink fails */ | |
8b61709d | 1855 | struct ifreq ifr; |
90a6637d | 1856 | |
86383816 | 1857 | netdev->netdev_mtu_error = af_inet_ifreq_ioctl( |
73371c09 | 1858 | netdev_get_name(&netdev->up), &ifr, SIOCGIFMTU, "SIOCGIFMTU"); |
b5d57fc8 BP |
1859 | netdev->mtu = ifr.ifr_mtu; |
1860 | netdev->cache_valid |= VALID_MTU; | |
8b61709d | 1861 | } |
90a6637d | 1862 | |
86383816 BP |
1863 | error = netdev->netdev_mtu_error; |
1864 | if (!error) { | |
b5d57fc8 | 1865 | *mtup = netdev->mtu; |
90a6637d | 1866 | } |
73371c09 BP |
1867 | |
1868 | return error; | |
1869 | } | |
1870 | ||
1871 | /* Returns the maximum size of transmitted (and received) packets on 'netdev', | |
1872 | * in bytes, not including the hardware header; thus, this is typically 1500 | |
1873 | * bytes for Ethernet devices. */ | |
1874 | static int | |
1875 | netdev_linux_get_mtu(const struct netdev *netdev_, int *mtup) | |
1876 | { | |
1877 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
1878 | int error; | |
1879 | ||
1880 | ovs_mutex_lock(&netdev->mutex); | |
1881 | error = netdev_linux_get_mtu__(netdev, mtup); | |
86383816 BP |
1882 | ovs_mutex_unlock(&netdev->mutex); |
1883 | ||
1884 | return error; | |
8b61709d BP |
1885 | } |
1886 | ||
9b020780 PS |
1887 | /* Sets the maximum size of transmitted (MTU) for given device using linux |
1888 | * networking ioctl interface. | |
1889 | */ | |
1890 | static int | |
4124cb12 | 1891 | netdev_linux_set_mtu(struct netdev *netdev_, int mtu) |
9b020780 | 1892 | { |
b5d57fc8 | 1893 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9b020780 PS |
1894 | struct ifreq ifr; |
1895 | int error; | |
1896 | ||
86383816 | 1897 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1898 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1899 | error = EOPNOTSUPP; | |
1900 | goto exit; | |
1901 | } | |
1902 | ||
52b5a5c0 EC |
1903 | #ifdef HAVE_AF_XDP |
1904 | if (netdev_get_class(netdev_) == &netdev_afxdp_class) { | |
1905 | error = netdev_afxdp_verify_mtu_size(netdev_, mtu); | |
1906 | if (error) { | |
1907 | goto exit; | |
1908 | } | |
1909 | } | |
1910 | #endif | |
1911 | ||
b5d57fc8 | 1912 | if (netdev->cache_valid & VALID_MTU) { |
86383816 BP |
1913 | error = netdev->netdev_mtu_error; |
1914 | if (error || netdev->mtu == mtu) { | |
1915 | goto exit; | |
90a6637d | 1916 | } |
b5d57fc8 | 1917 | netdev->cache_valid &= ~VALID_MTU; |
153e5481 | 1918 | } |
9b020780 | 1919 | ifr.ifr_mtu = mtu; |
259e0b1a BP |
1920 | error = af_inet_ifreq_ioctl(netdev_get_name(netdev_), &ifr, |
1921 | SIOCSIFMTU, "SIOCSIFMTU"); | |
90a6637d | 1922 | if (!error || error == ENODEV) { |
b5d57fc8 BP |
1923 | netdev->netdev_mtu_error = error; |
1924 | netdev->mtu = ifr.ifr_mtu; | |
1925 | netdev->cache_valid |= VALID_MTU; | |
9b020780 | 1926 | } |
86383816 BP |
1927 | exit: |
1928 | ovs_mutex_unlock(&netdev->mutex); | |
90a6637d | 1929 | return error; |
9b020780 PS |
1930 | } |
1931 | ||
9ab3d9a3 BP |
1932 | /* Returns the ifindex of 'netdev', if successful, as a positive number. |
1933 | * On failure, returns a negative errno value. */ | |
1934 | static int | |
86383816 | 1935 | netdev_linux_get_ifindex(const struct netdev *netdev_) |
9ab3d9a3 | 1936 | { |
86383816 | 1937 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
9ab3d9a3 BP |
1938 | int ifindex, error; |
1939 | ||
86383816 | 1940 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
1941 | if (netdev_linux_netnsid_is_remote(netdev)) { |
1942 | error = EOPNOTSUPP; | |
1943 | goto exit; | |
1944 | } | |
86383816 | 1945 | error = get_ifindex(netdev_, &ifindex); |
86383816 | 1946 | |
e0e2410d FL |
1947 | exit: |
1948 | ovs_mutex_unlock(&netdev->mutex); | |
9ab3d9a3 BP |
1949 | return error ? -error : ifindex; |
1950 | } | |
1951 | ||
8b61709d BP |
1952 | static int |
1953 | netdev_linux_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1954 | { | |
b5d57fc8 | 1955 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 1956 | |
86383816 | 1957 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
1958 | if (netdev->miimon_interval > 0) { |
1959 | *carrier = netdev->miimon; | |
3a183124 | 1960 | } else { |
b5d57fc8 | 1961 | *carrier = (netdev->ifi_flags & IFF_RUNNING) != 0; |
8b61709d | 1962 | } |
86383816 | 1963 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d | 1964 | |
3a183124 | 1965 | return 0; |
8b61709d BP |
1966 | } |
1967 | ||
65c3058c | 1968 | static long long int |
86383816 | 1969 | netdev_linux_get_carrier_resets(const struct netdev *netdev_) |
65c3058c | 1970 | { |
86383816 BP |
1971 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1972 | long long int carrier_resets; | |
1973 | ||
1974 | ovs_mutex_lock(&netdev->mutex); | |
1975 | carrier_resets = netdev->carrier_resets; | |
1976 | ovs_mutex_unlock(&netdev->mutex); | |
1977 | ||
1978 | return carrier_resets; | |
65c3058c EJ |
1979 | } |
1980 | ||
63331829 | 1981 | static int |
1670c579 EJ |
1982 | netdev_linux_do_miimon(const char *name, int cmd, const char *cmd_name, |
1983 | struct mii_ioctl_data *data) | |
63331829 | 1984 | { |
63331829 | 1985 | struct ifreq ifr; |
782e6111 | 1986 | int error; |
63331829 | 1987 | |
63331829 | 1988 | memset(&ifr, 0, sizeof ifr); |
782e6111 | 1989 | memcpy(&ifr.ifr_data, data, sizeof *data); |
259e0b1a | 1990 | error = af_inet_ifreq_ioctl(name, &ifr, cmd, cmd_name); |
782e6111 | 1991 | memcpy(data, &ifr.ifr_data, sizeof *data); |
63331829 | 1992 | |
782e6111 EJ |
1993 | return error; |
1994 | } | |
1995 | ||
1996 | static int | |
1670c579 | 1997 | netdev_linux_get_miimon(const char *name, bool *miimon) |
782e6111 | 1998 | { |
782e6111 EJ |
1999 | struct mii_ioctl_data data; |
2000 | int error; | |
63331829 | 2001 | |
782e6111 EJ |
2002 | *miimon = false; |
2003 | ||
2004 | memset(&data, 0, sizeof data); | |
1670c579 | 2005 | error = netdev_linux_do_miimon(name, SIOCGMIIPHY, "SIOCGMIIPHY", &data); |
782e6111 EJ |
2006 | if (!error) { |
2007 | /* data.phy_id is filled out by previous SIOCGMIIPHY miimon call. */ | |
2008 | data.reg_num = MII_BMSR; | |
1670c579 | 2009 | error = netdev_linux_do_miimon(name, SIOCGMIIREG, "SIOCGMIIREG", |
782e6111 | 2010 | &data); |
63331829 EJ |
2011 | |
2012 | if (!error) { | |
782e6111 | 2013 | *miimon = !!(data.val_out & BMSR_LSTATUS); |
63331829 | 2014 | } |
9120cfc0 DH |
2015 | } |
2016 | if (error) { | |
63331829 | 2017 | struct ethtool_cmd ecmd; |
63331829 EJ |
2018 | |
2019 | VLOG_DBG_RL(&rl, "%s: failed to query MII, falling back to ethtool", | |
2020 | name); | |
2021 | ||
ab985a77 | 2022 | COVERAGE_INC(netdev_get_ethtool); |
63331829 EJ |
2023 | memset(&ecmd, 0, sizeof ecmd); |
2024 | error = netdev_linux_do_ethtool(name, &ecmd, ETHTOOL_GLINK, | |
2025 | "ETHTOOL_GLINK"); | |
2026 | if (!error) { | |
782e6111 EJ |
2027 | struct ethtool_value eval; |
2028 | ||
2029 | memcpy(&eval, &ecmd, sizeof eval); | |
2030 | *miimon = !!eval.data; | |
63331829 EJ |
2031 | } else { |
2032 | VLOG_WARN_RL(&rl, "%s: ethtool link status failed", name); | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | return error; | |
2037 | } | |
2038 | ||
1670c579 EJ |
2039 | static int |
2040 | netdev_linux_set_miimon_interval(struct netdev *netdev_, | |
2041 | long long int interval) | |
2042 | { | |
b5d57fc8 | 2043 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
1670c579 | 2044 | |
86383816 | 2045 | ovs_mutex_lock(&netdev->mutex); |
1670c579 | 2046 | interval = interval > 0 ? MAX(interval, 100) : 0; |
b5d57fc8 | 2047 | if (netdev->miimon_interval != interval) { |
19c8e9c1 | 2048 | if (interval && !netdev->miimon_interval) { |
812c272c | 2049 | atomic_count_inc(&miimon_cnt); |
19c8e9c1 | 2050 | } else if (!interval && netdev->miimon_interval) { |
812c272c | 2051 | atomic_count_dec(&miimon_cnt); |
19c8e9c1 JS |
2052 | } |
2053 | ||
b5d57fc8 BP |
2054 | netdev->miimon_interval = interval; |
2055 | timer_set_expired(&netdev->miimon_timer); | |
1670c579 | 2056 | } |
86383816 | 2057 | ovs_mutex_unlock(&netdev->mutex); |
1670c579 EJ |
2058 | |
2059 | return 0; | |
2060 | } | |
2061 | ||
2062 | static void | |
2063 | netdev_linux_miimon_run(void) | |
2064 | { | |
2065 | struct shash device_shash; | |
2066 | struct shash_node *node; | |
2067 | ||
2068 | shash_init(&device_shash); | |
b5d57fc8 | 2069 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 2070 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
2071 | struct netdev *netdev = node->data; |
2072 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 EJ |
2073 | bool miimon; |
2074 | ||
86383816 BP |
2075 | ovs_mutex_lock(&dev->mutex); |
2076 | if (dev->miimon_interval > 0 && timer_expired(&dev->miimon_timer)) { | |
2077 | netdev_linux_get_miimon(dev->up.name, &miimon); | |
2078 | if (miimon != dev->miimon) { | |
2079 | dev->miimon = miimon; | |
2080 | netdev_linux_changed(dev, dev->ifi_flags, 0); | |
2081 | } | |
1670c579 | 2082 | |
86383816 | 2083 | timer_set_duration(&dev->miimon_timer, dev->miimon_interval); |
1670c579 | 2084 | } |
86383816 | 2085 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 2086 | netdev_close(netdev); |
1670c579 EJ |
2087 | } |
2088 | ||
2089 | shash_destroy(&device_shash); | |
2090 | } | |
2091 | ||
2092 | static void | |
2093 | netdev_linux_miimon_wait(void) | |
2094 | { | |
2095 | struct shash device_shash; | |
2096 | struct shash_node *node; | |
2097 | ||
2098 | shash_init(&device_shash); | |
b5d57fc8 | 2099 | netdev_get_devices(&netdev_linux_class, &device_shash); |
1670c579 | 2100 | SHASH_FOR_EACH (node, &device_shash) { |
96172faa BP |
2101 | struct netdev *netdev = node->data; |
2102 | struct netdev_linux *dev = netdev_linux_cast(netdev); | |
1670c579 | 2103 | |
86383816 | 2104 | ovs_mutex_lock(&dev->mutex); |
1670c579 EJ |
2105 | if (dev->miimon_interval > 0) { |
2106 | timer_wait(&dev->miimon_timer); | |
2107 | } | |
86383816 | 2108 | ovs_mutex_unlock(&dev->mutex); |
2f980d74 | 2109 | netdev_close(netdev); |
1670c579 EJ |
2110 | } |
2111 | shash_destroy(&device_shash); | |
2112 | } | |
2113 | ||
92df599c JG |
2114 | static void |
2115 | swap_uint64(uint64_t *a, uint64_t *b) | |
2116 | { | |
1de0e8ae BP |
2117 | uint64_t tmp = *a; |
2118 | *a = *b; | |
2119 | *b = tmp; | |
92df599c JG |
2120 | } |
2121 | ||
c060c4cf EJ |
2122 | /* Copies 'src' into 'dst', performing format conversion in the process. |
2123 | * | |
2124 | * 'src' is allowed to be misaligned. */ | |
2125 | static void | |
2126 | netdev_stats_from_ovs_vport_stats(struct netdev_stats *dst, | |
2127 | const struct ovs_vport_stats *src) | |
2128 | { | |
6a54dedc BP |
2129 | dst->rx_packets = get_32aligned_u64(&src->rx_packets); |
2130 | dst->tx_packets = get_32aligned_u64(&src->tx_packets); | |
2131 | dst->rx_bytes = get_32aligned_u64(&src->rx_bytes); | |
2132 | dst->tx_bytes = get_32aligned_u64(&src->tx_bytes); | |
2133 | dst->rx_errors = get_32aligned_u64(&src->rx_errors); | |
2134 | dst->tx_errors = get_32aligned_u64(&src->tx_errors); | |
2135 | dst->rx_dropped = get_32aligned_u64(&src->rx_dropped); | |
2136 | dst->tx_dropped = get_32aligned_u64(&src->tx_dropped); | |
c060c4cf EJ |
2137 | dst->multicast = 0; |
2138 | dst->collisions = 0; | |
2139 | dst->rx_length_errors = 0; | |
2140 | dst->rx_over_errors = 0; | |
2141 | dst->rx_crc_errors = 0; | |
2142 | dst->rx_frame_errors = 0; | |
2143 | dst->rx_fifo_errors = 0; | |
2144 | dst->rx_missed_errors = 0; | |
2145 | dst->tx_aborted_errors = 0; | |
2146 | dst->tx_carrier_errors = 0; | |
2147 | dst->tx_fifo_errors = 0; | |
2148 | dst->tx_heartbeat_errors = 0; | |
2149 | dst->tx_window_errors = 0; | |
2150 | } | |
2151 | ||
2152 | static int | |
2153 | get_stats_via_vport__(const struct netdev *netdev, struct netdev_stats *stats) | |
2154 | { | |
93451a0a | 2155 | struct dpif_netlink_vport reply; |
c060c4cf EJ |
2156 | struct ofpbuf *buf; |
2157 | int error; | |
2158 | ||
93451a0a | 2159 | error = dpif_netlink_vport_get(netdev_get_name(netdev), &reply, &buf); |
c060c4cf EJ |
2160 | if (error) { |
2161 | return error; | |
2162 | } else if (!reply.stats) { | |
2163 | ofpbuf_delete(buf); | |
2164 | return EOPNOTSUPP; | |
2165 | } | |
2166 | ||
2167 | netdev_stats_from_ovs_vport_stats(stats, reply.stats); | |
2168 | ||
2169 | ofpbuf_delete(buf); | |
2170 | ||
2171 | return 0; | |
2172 | } | |
2173 | ||
f613a0d7 PS |
2174 | static void |
2175 | get_stats_via_vport(const struct netdev *netdev_, | |
2176 | struct netdev_stats *stats) | |
8b61709d | 2177 | { |
b5d57fc8 | 2178 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d | 2179 | |
b5d57fc8 BP |
2180 | if (!netdev->vport_stats_error || |
2181 | !(netdev->cache_valid & VALID_VPORT_STAT_ERROR)) { | |
f613a0d7 | 2182 | int error; |
7fbef77a | 2183 | |
c060c4cf | 2184 | error = get_stats_via_vport__(netdev_, stats); |
bb13fe5e | 2185 | if (error && error != ENOENT && error != ENODEV) { |
a57a8488 | 2186 | VLOG_WARN_RL(&rl, "%s: obtaining netdev stats via vport failed " |
10a89ef0 BP |
2187 | "(%s)", |
2188 | netdev_get_name(netdev_), ovs_strerror(error)); | |
f613a0d7 | 2189 | } |
b5d57fc8 BP |
2190 | netdev->vport_stats_error = error; |
2191 | netdev->cache_valid |= VALID_VPORT_STAT_ERROR; | |
8b61709d | 2192 | } |
f613a0d7 | 2193 | } |
8b61709d | 2194 | |
f613a0d7 PS |
2195 | /* Retrieves current device stats for 'netdev-linux'. */ |
2196 | static int | |
2197 | netdev_linux_get_stats(const struct netdev *netdev_, | |
2198 | struct netdev_stats *stats) | |
2199 | { | |
b5d57fc8 | 2200 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
2201 | struct netdev_stats dev_stats; |
2202 | int error; | |
2203 | ||
86383816 | 2204 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 2205 | get_stats_via_vport(netdev_, stats); |
35eef899 | 2206 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 2207 | if (error) { |
86383816 BP |
2208 | if (!netdev->vport_stats_error) { |
2209 | error = 0; | |
f613a0d7 | 2210 | } |
86383816 | 2211 | } else if (netdev->vport_stats_error) { |
04c881eb | 2212 | /* stats not available from OVS then use netdev stats. */ |
f613a0d7 PS |
2213 | *stats = dev_stats; |
2214 | } else { | |
04c881eb AZ |
2215 | /* Use kernel netdev's packet and byte counts since vport's counters |
2216 | * do not reflect packet counts on the wire when GSO, TSO or GRO are | |
2217 | * enabled. */ | |
2218 | stats->rx_packets = dev_stats.rx_packets; | |
2219 | stats->rx_bytes = dev_stats.rx_bytes; | |
2220 | stats->tx_packets = dev_stats.tx_packets; | |
2221 | stats->tx_bytes = dev_stats.tx_bytes; | |
2222 | ||
f613a0d7 PS |
2223 | stats->rx_errors += dev_stats.rx_errors; |
2224 | stats->tx_errors += dev_stats.tx_errors; | |
2225 | stats->rx_dropped += dev_stats.rx_dropped; | |
2226 | stats->tx_dropped += dev_stats.tx_dropped; | |
2227 | stats->multicast += dev_stats.multicast; | |
2228 | stats->collisions += dev_stats.collisions; | |
2229 | stats->rx_length_errors += dev_stats.rx_length_errors; | |
2230 | stats->rx_over_errors += dev_stats.rx_over_errors; | |
2231 | stats->rx_crc_errors += dev_stats.rx_crc_errors; | |
2232 | stats->rx_frame_errors += dev_stats.rx_frame_errors; | |
2233 | stats->rx_fifo_errors += dev_stats.rx_fifo_errors; | |
2234 | stats->rx_missed_errors += dev_stats.rx_missed_errors; | |
2235 | stats->tx_aborted_errors += dev_stats.tx_aborted_errors; | |
2236 | stats->tx_carrier_errors += dev_stats.tx_carrier_errors; | |
2237 | stats->tx_fifo_errors += dev_stats.tx_fifo_errors; | |
2238 | stats->tx_heartbeat_errors += dev_stats.tx_heartbeat_errors; | |
2239 | stats->tx_window_errors += dev_stats.tx_window_errors; | |
2240 | } | |
86383816 BP |
2241 | ovs_mutex_unlock(&netdev->mutex); |
2242 | ||
2243 | return error; | |
f613a0d7 PS |
2244 | } |
2245 | ||
2246 | /* Retrieves current device stats for 'netdev-tap' netdev or | |
2247 | * netdev-internal. */ | |
2248 | static int | |
15aee116 | 2249 | netdev_tap_get_stats(const struct netdev *netdev_, struct netdev_stats *stats) |
f613a0d7 | 2250 | { |
b5d57fc8 | 2251 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
f613a0d7 PS |
2252 | struct netdev_stats dev_stats; |
2253 | int error; | |
2254 | ||
86383816 | 2255 | ovs_mutex_lock(&netdev->mutex); |
f613a0d7 | 2256 | get_stats_via_vport(netdev_, stats); |
35eef899 | 2257 | error = get_stats_via_netlink(netdev_, &dev_stats); |
f613a0d7 | 2258 | if (error) { |
86383816 BP |
2259 | if (!netdev->vport_stats_error) { |
2260 | error = 0; | |
8b61709d | 2261 | } |
86383816 BP |
2262 | } else if (netdev->vport_stats_error) { |
2263 | /* Transmit and receive stats will appear to be swapped relative to the | |
2264 | * other ports since we are the one sending the data, not a remote | |
2265 | * computer. For consistency, we swap them back here. This does not | |
2266 | * apply if we are getting stats from the vport layer because it always | |
2267 | * tracks stats from the perspective of the switch. */ | |
fe6b0e03 | 2268 | |
f613a0d7 | 2269 | *stats = dev_stats; |
92df599c JG |
2270 | swap_uint64(&stats->rx_packets, &stats->tx_packets); |
2271 | swap_uint64(&stats->rx_bytes, &stats->tx_bytes); | |
2272 | swap_uint64(&stats->rx_errors, &stats->tx_errors); | |
2273 | swap_uint64(&stats->rx_dropped, &stats->tx_dropped); | |
fe6b0e03 JG |
2274 | stats->rx_length_errors = 0; |
2275 | stats->rx_over_errors = 0; | |
2276 | stats->rx_crc_errors = 0; | |
2277 | stats->rx_frame_errors = 0; | |
2278 | stats->rx_fifo_errors = 0; | |
2279 | stats->rx_missed_errors = 0; | |
2280 | stats->tx_aborted_errors = 0; | |
2281 | stats->tx_carrier_errors = 0; | |
2282 | stats->tx_fifo_errors = 0; | |
2283 | stats->tx_heartbeat_errors = 0; | |
2284 | stats->tx_window_errors = 0; | |
f613a0d7 | 2285 | } else { |
04c881eb AZ |
2286 | /* Use kernel netdev's packet and byte counts since vport counters |
2287 | * do not reflect packet counts on the wire when GSO, TSO or GRO | |
2288 | * are enabled. */ | |
2289 | stats->rx_packets = dev_stats.tx_packets; | |
2290 | stats->rx_bytes = dev_stats.tx_bytes; | |
2291 | stats->tx_packets = dev_stats.rx_packets; | |
2292 | stats->tx_bytes = dev_stats.rx_bytes; | |
2293 | ||
f613a0d7 PS |
2294 | stats->rx_dropped += dev_stats.tx_dropped; |
2295 | stats->tx_dropped += dev_stats.rx_dropped; | |
fe6b0e03 | 2296 | |
f613a0d7 PS |
2297 | stats->rx_errors += dev_stats.tx_errors; |
2298 | stats->tx_errors += dev_stats.rx_errors; | |
2299 | ||
2300 | stats->multicast += dev_stats.multicast; | |
2301 | stats->collisions += dev_stats.collisions; | |
2302 | } | |
22dcb534 | 2303 | stats->tx_dropped += netdev->tx_dropped; |
29cf9c1b | 2304 | stats->rx_dropped += netdev->rx_dropped; |
86383816 BP |
2305 | ovs_mutex_unlock(&netdev->mutex); |
2306 | ||
2307 | return error; | |
8b61709d BP |
2308 | } |
2309 | ||
bba1e6f3 PS |
2310 | static int |
2311 | netdev_internal_get_stats(const struct netdev *netdev_, | |
2312 | struct netdev_stats *stats) | |
2313 | { | |
b5d57fc8 | 2314 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 2315 | int error; |
bba1e6f3 | 2316 | |
86383816 | 2317 | ovs_mutex_lock(&netdev->mutex); |
bba1e6f3 | 2318 | get_stats_via_vport(netdev_, stats); |
86383816 BP |
2319 | error = netdev->vport_stats_error; |
2320 | ovs_mutex_unlock(&netdev->mutex); | |
2321 | ||
2322 | return error; | |
bba1e6f3 PS |
2323 | } |
2324 | ||
51f87458 | 2325 | static void |
b5d57fc8 | 2326 | netdev_linux_read_features(struct netdev_linux *netdev) |
8b61709d BP |
2327 | { |
2328 | struct ethtool_cmd ecmd; | |
6c038611 | 2329 | uint32_t speed; |
8b61709d BP |
2330 | int error; |
2331 | ||
b5d57fc8 | 2332 | if (netdev->cache_valid & VALID_FEATURES) { |
51f87458 PS |
2333 | return; |
2334 | } | |
2335 | ||
ab985a77 | 2336 | COVERAGE_INC(netdev_get_ethtool); |
8b61709d | 2337 | memset(&ecmd, 0, sizeof ecmd); |
b5d57fc8 | 2338 | error = netdev_linux_do_ethtool(netdev->up.name, &ecmd, |
8b61709d BP |
2339 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
2340 | if (error) { | |
51f87458 | 2341 | goto out; |
8b61709d BP |
2342 | } |
2343 | ||
2344 | /* Supported features. */ | |
b5d57fc8 | 2345 | netdev->supported = 0; |
8b61709d | 2346 | if (ecmd.supported & SUPPORTED_10baseT_Half) { |
b5d57fc8 | 2347 | netdev->supported |= NETDEV_F_10MB_HD; |
8b61709d BP |
2348 | } |
2349 | if (ecmd.supported & SUPPORTED_10baseT_Full) { | |
b5d57fc8 | 2350 | netdev->supported |= NETDEV_F_10MB_FD; |
8b61709d BP |
2351 | } |
2352 | if (ecmd.supported & SUPPORTED_100baseT_Half) { | |
b5d57fc8 | 2353 | netdev->supported |= NETDEV_F_100MB_HD; |
8b61709d BP |
2354 | } |
2355 | if (ecmd.supported & SUPPORTED_100baseT_Full) { | |
b5d57fc8 | 2356 | netdev->supported |= NETDEV_F_100MB_FD; |
8b61709d BP |
2357 | } |
2358 | if (ecmd.supported & SUPPORTED_1000baseT_Half) { | |
b5d57fc8 | 2359 | netdev->supported |= NETDEV_F_1GB_HD; |
8b61709d | 2360 | } |
67bed84c SH |
2361 | if ((ecmd.supported & SUPPORTED_1000baseT_Full) || |
2362 | (ecmd.supported & SUPPORTED_1000baseKX_Full)) { | |
b5d57fc8 | 2363 | netdev->supported |= NETDEV_F_1GB_FD; |
8b61709d | 2364 | } |
67bed84c SH |
2365 | if ((ecmd.supported & SUPPORTED_10000baseT_Full) || |
2366 | (ecmd.supported & SUPPORTED_10000baseKX4_Full) || | |
2367 | (ecmd.supported & SUPPORTED_10000baseKR_Full) || | |
2368 | (ecmd.supported & SUPPORTED_10000baseR_FEC)) { | |
b5d57fc8 | 2369 | netdev->supported |= NETDEV_F_10GB_FD; |
8b61709d | 2370 | } |
67bed84c SH |
2371 | if ((ecmd.supported & SUPPORTED_40000baseKR4_Full) || |
2372 | (ecmd.supported & SUPPORTED_40000baseCR4_Full) || | |
2373 | (ecmd.supported & SUPPORTED_40000baseSR4_Full) || | |
2374 | (ecmd.supported & SUPPORTED_40000baseLR4_Full)) { | |
2375 | netdev->supported |= NETDEV_F_40GB_FD; | |
2376 | } | |
8b61709d | 2377 | if (ecmd.supported & SUPPORTED_TP) { |
b5d57fc8 | 2378 | netdev->supported |= NETDEV_F_COPPER; |
8b61709d BP |
2379 | } |
2380 | if (ecmd.supported & SUPPORTED_FIBRE) { | |
b5d57fc8 | 2381 | netdev->supported |= NETDEV_F_FIBER; |
8b61709d BP |
2382 | } |
2383 | if (ecmd.supported & SUPPORTED_Autoneg) { | |
b5d57fc8 | 2384 | netdev->supported |= NETDEV_F_AUTONEG; |
8b61709d BP |
2385 | } |
2386 | if (ecmd.supported & SUPPORTED_Pause) { | |
b5d57fc8 | 2387 | netdev->supported |= NETDEV_F_PAUSE; |
8b61709d BP |
2388 | } |
2389 | if (ecmd.supported & SUPPORTED_Asym_Pause) { | |
b5d57fc8 | 2390 | netdev->supported |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2391 | } |
2392 | ||
2393 | /* Advertised features. */ | |
b5d57fc8 | 2394 | netdev->advertised = 0; |
8b61709d | 2395 | if (ecmd.advertising & ADVERTISED_10baseT_Half) { |
b5d57fc8 | 2396 | netdev->advertised |= NETDEV_F_10MB_HD; |
8b61709d BP |
2397 | } |
2398 | if (ecmd.advertising & ADVERTISED_10baseT_Full) { | |
b5d57fc8 | 2399 | netdev->advertised |= NETDEV_F_10MB_FD; |
8b61709d BP |
2400 | } |
2401 | if (ecmd.advertising & ADVERTISED_100baseT_Half) { | |
b5d57fc8 | 2402 | netdev->advertised |= NETDEV_F_100MB_HD; |
8b61709d BP |
2403 | } |
2404 | if (ecmd.advertising & ADVERTISED_100baseT_Full) { | |
b5d57fc8 | 2405 | netdev->advertised |= NETDEV_F_100MB_FD; |
8b61709d BP |
2406 | } |
2407 | if (ecmd.advertising & ADVERTISED_1000baseT_Half) { | |
b5d57fc8 | 2408 | netdev->advertised |= NETDEV_F_1GB_HD; |
8b61709d | 2409 | } |
67bed84c SH |
2410 | if ((ecmd.advertising & ADVERTISED_1000baseT_Full) || |
2411 | (ecmd.advertising & ADVERTISED_1000baseKX_Full)) { | |
b5d57fc8 | 2412 | netdev->advertised |= NETDEV_F_1GB_FD; |
8b61709d | 2413 | } |
67bed84c SH |
2414 | if ((ecmd.advertising & ADVERTISED_10000baseT_Full) || |
2415 | (ecmd.advertising & ADVERTISED_10000baseKX4_Full) || | |
2416 | (ecmd.advertising & ADVERTISED_10000baseKR_Full) || | |
2417 | (ecmd.advertising & ADVERTISED_10000baseR_FEC)) { | |
b5d57fc8 | 2418 | netdev->advertised |= NETDEV_F_10GB_FD; |
8b61709d | 2419 | } |
67bed84c SH |
2420 | if ((ecmd.advertising & ADVERTISED_40000baseKR4_Full) || |
2421 | (ecmd.advertising & ADVERTISED_40000baseCR4_Full) || | |
2422 | (ecmd.advertising & ADVERTISED_40000baseSR4_Full) || | |
2423 | (ecmd.advertising & ADVERTISED_40000baseLR4_Full)) { | |
2424 | netdev->advertised |= NETDEV_F_40GB_FD; | |
2425 | } | |
8b61709d | 2426 | if (ecmd.advertising & ADVERTISED_TP) { |
b5d57fc8 | 2427 | netdev->advertised |= NETDEV_F_COPPER; |
8b61709d BP |
2428 | } |
2429 | if (ecmd.advertising & ADVERTISED_FIBRE) { | |
b5d57fc8 | 2430 | netdev->advertised |= NETDEV_F_FIBER; |
8b61709d BP |
2431 | } |
2432 | if (ecmd.advertising & ADVERTISED_Autoneg) { | |
b5d57fc8 | 2433 | netdev->advertised |= NETDEV_F_AUTONEG; |
8b61709d BP |
2434 | } |
2435 | if (ecmd.advertising & ADVERTISED_Pause) { | |
b5d57fc8 | 2436 | netdev->advertised |= NETDEV_F_PAUSE; |
8b61709d BP |
2437 | } |
2438 | if (ecmd.advertising & ADVERTISED_Asym_Pause) { | |
b5d57fc8 | 2439 | netdev->advertised |= NETDEV_F_PAUSE_ASYM; |
8b61709d BP |
2440 | } |
2441 | ||
2442 | /* Current settings. */ | |
0c615356 | 2443 | speed = ethtool_cmd_speed(&ecmd); |
6c038611 | 2444 | if (speed == SPEED_10) { |
b5d57fc8 | 2445 | netdev->current = ecmd.duplex ? NETDEV_F_10MB_FD : NETDEV_F_10MB_HD; |
6c038611 | 2446 | } else if (speed == SPEED_100) { |
b5d57fc8 | 2447 | netdev->current = ecmd.duplex ? NETDEV_F_100MB_FD : NETDEV_F_100MB_HD; |
6c038611 | 2448 | } else if (speed == SPEED_1000) { |
b5d57fc8 | 2449 | netdev->current = ecmd.duplex ? NETDEV_F_1GB_FD : NETDEV_F_1GB_HD; |
6c038611 | 2450 | } else if (speed == SPEED_10000) { |
b5d57fc8 | 2451 | netdev->current = NETDEV_F_10GB_FD; |
6c038611 | 2452 | } else if (speed == 40000) { |
b5d57fc8 | 2453 | netdev->current = NETDEV_F_40GB_FD; |
6c038611 | 2454 | } else if (speed == 100000) { |
b5d57fc8 | 2455 | netdev->current = NETDEV_F_100GB_FD; |
6c038611 | 2456 | } else if (speed == 1000000) { |
b5d57fc8 | 2457 | netdev->current = NETDEV_F_1TB_FD; |
8b61709d | 2458 | } else { |
b5d57fc8 | 2459 | netdev->current = 0; |
8b61709d BP |
2460 | } |
2461 | ||
2462 | if (ecmd.port == PORT_TP) { | |
b5d57fc8 | 2463 | netdev->current |= NETDEV_F_COPPER; |
8b61709d | 2464 | } else if (ecmd.port == PORT_FIBRE) { |
b5d57fc8 | 2465 | netdev->current |= NETDEV_F_FIBER; |
8b61709d BP |
2466 | } |
2467 | ||
2468 | if (ecmd.autoneg) { | |
b5d57fc8 | 2469 | netdev->current |= NETDEV_F_AUTONEG; |
8b61709d BP |
2470 | } |
2471 | ||
51f87458 | 2472 | out: |
b5d57fc8 BP |
2473 | netdev->cache_valid |= VALID_FEATURES; |
2474 | netdev->get_features_error = error; | |
51f87458 PS |
2475 | } |
2476 | ||
887ed8b2 BP |
2477 | /* Stores the features supported by 'netdev' into of '*current', '*advertised', |
2478 | * '*supported', and '*peer'. Each value is a bitmap of NETDEV_* bits. | |
2479 | * Returns 0 if successful, otherwise a positive errno value. */ | |
51f87458 PS |
2480 | static int |
2481 | netdev_linux_get_features(const struct netdev *netdev_, | |
a00ca915 EJ |
2482 | enum netdev_features *current, |
2483 | enum netdev_features *advertised, | |
2484 | enum netdev_features *supported, | |
2485 | enum netdev_features *peer) | |
51f87458 | 2486 | { |
b5d57fc8 | 2487 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
86383816 | 2488 | int error; |
51f87458 | 2489 | |
86383816 | 2490 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2491 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2492 | error = EOPNOTSUPP; | |
2493 | goto exit; | |
2494 | } | |
2495 | ||
b5d57fc8 | 2496 | netdev_linux_read_features(netdev); |
b5d57fc8 BP |
2497 | if (!netdev->get_features_error) { |
2498 | *current = netdev->current; | |
2499 | *advertised = netdev->advertised; | |
2500 | *supported = netdev->supported; | |
887ed8b2 | 2501 | *peer = 0; /* XXX */ |
51f87458 | 2502 | } |
86383816 | 2503 | error = netdev->get_features_error; |
86383816 | 2504 | |
e0e2410d FL |
2505 | exit: |
2506 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2507 | return error; |
8b61709d BP |
2508 | } |
2509 | ||
2510 | /* Set the features advertised by 'netdev' to 'advertise'. */ | |
2511 | static int | |
86383816 | 2512 | netdev_linux_set_advertisements(struct netdev *netdev_, |
6c038611 | 2513 | enum netdev_features advertise) |
8b61709d | 2514 | { |
86383816 | 2515 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
2516 | struct ethtool_cmd ecmd; |
2517 | int error; | |
2518 | ||
86383816 BP |
2519 | ovs_mutex_lock(&netdev->mutex); |
2520 | ||
ab985a77 | 2521 | COVERAGE_INC(netdev_get_ethtool); |
e0e2410d FL |
2522 | |
2523 | if (netdev_linux_netnsid_is_remote(netdev)) { | |
2524 | error = EOPNOTSUPP; | |
2525 | goto exit; | |
2526 | } | |
2527 | ||
8b61709d | 2528 | memset(&ecmd, 0, sizeof ecmd); |
86383816 | 2529 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
8b61709d BP |
2530 | ETHTOOL_GSET, "ETHTOOL_GSET"); |
2531 | if (error) { | |
86383816 | 2532 | goto exit; |
8b61709d BP |
2533 | } |
2534 | ||
2535 | ecmd.advertising = 0; | |
6c038611 | 2536 | if (advertise & NETDEV_F_10MB_HD) { |
8b61709d BP |
2537 | ecmd.advertising |= ADVERTISED_10baseT_Half; |
2538 | } | |
6c038611 | 2539 | if (advertise & NETDEV_F_10MB_FD) { |
8b61709d BP |
2540 | ecmd.advertising |= ADVERTISED_10baseT_Full; |
2541 | } | |
6c038611 | 2542 | if (advertise & NETDEV_F_100MB_HD) { |
8b61709d BP |
2543 | ecmd.advertising |= ADVERTISED_100baseT_Half; |
2544 | } | |
6c038611 | 2545 | if (advertise & NETDEV_F_100MB_FD) { |
8b61709d BP |
2546 | ecmd.advertising |= ADVERTISED_100baseT_Full; |
2547 | } | |
6c038611 | 2548 | if (advertise & NETDEV_F_1GB_HD) { |
8b61709d BP |
2549 | ecmd.advertising |= ADVERTISED_1000baseT_Half; |
2550 | } | |
6c038611 | 2551 | if (advertise & NETDEV_F_1GB_FD) { |
8b61709d BP |
2552 | ecmd.advertising |= ADVERTISED_1000baseT_Full; |
2553 | } | |
6c038611 | 2554 | if (advertise & NETDEV_F_10GB_FD) { |
8b61709d BP |
2555 | ecmd.advertising |= ADVERTISED_10000baseT_Full; |
2556 | } | |
6c038611 | 2557 | if (advertise & NETDEV_F_COPPER) { |
8b61709d BP |
2558 | ecmd.advertising |= ADVERTISED_TP; |
2559 | } | |
6c038611 | 2560 | if (advertise & NETDEV_F_FIBER) { |
8b61709d BP |
2561 | ecmd.advertising |= ADVERTISED_FIBRE; |
2562 | } | |
6c038611 | 2563 | if (advertise & NETDEV_F_AUTONEG) { |
8b61709d BP |
2564 | ecmd.advertising |= ADVERTISED_Autoneg; |
2565 | } | |
6c038611 | 2566 | if (advertise & NETDEV_F_PAUSE) { |
8b61709d BP |
2567 | ecmd.advertising |= ADVERTISED_Pause; |
2568 | } | |
6c038611 | 2569 | if (advertise & NETDEV_F_PAUSE_ASYM) { |
8b61709d BP |
2570 | ecmd.advertising |= ADVERTISED_Asym_Pause; |
2571 | } | |
ab985a77 | 2572 | COVERAGE_INC(netdev_set_ethtool); |
86383816 BP |
2573 | error = netdev_linux_do_ethtool(netdev_get_name(netdev_), &ecmd, |
2574 | ETHTOOL_SSET, "ETHTOOL_SSET"); | |
2575 | ||
2576 | exit: | |
2577 | ovs_mutex_unlock(&netdev->mutex); | |
2578 | return error; | |
8b61709d BP |
2579 | } |
2580 | ||
e7f6ba22 PJV |
2581 | static struct tc_police |
2582 | tc_matchall_fill_police(uint32_t kbits_rate, uint32_t kbits_burst) | |
2583 | { | |
2584 | unsigned int bsize = MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 64; | |
2585 | unsigned int bps = ((uint64_t) kbits_rate * 1000) / 8; | |
2586 | struct tc_police police; | |
2587 | struct tc_ratespec rate; | |
2588 | int mtu = 65535; | |
2589 | ||
2590 | memset(&rate, 0, sizeof rate); | |
2591 | rate.rate = bps; | |
2592 | rate.cell_log = tc_calc_cell_log(mtu); | |
2593 | rate.mpu = ETH_TOTAL_MIN; | |
2594 | ||
2595 | memset(&police, 0, sizeof police); | |
2596 | police.burst = tc_bytes_to_ticks(bps, bsize); | |
2597 | police.action = TC_POLICE_SHOT; | |
2598 | police.rate = rate; | |
2599 | police.mtu = mtu; | |
2600 | ||
2601 | return police; | |
2602 | } | |
2603 | ||
2604 | static void | |
2605 | nl_msg_put_act_police(struct ofpbuf *request, struct tc_police police) | |
2606 | { | |
2607 | size_t offset; | |
2608 | ||
2609 | nl_msg_put_string(request, TCA_ACT_KIND, "police"); | |
2610 | offset = nl_msg_start_nested(request, TCA_ACT_OPTIONS); | |
2611 | nl_msg_put_unspec(request, TCA_POLICE_TBF, &police, sizeof police); | |
2612 | tc_put_rtab(request, TCA_POLICE_RATE, &police.rate); | |
2613 | nl_msg_put_u32(request, TCA_POLICE_RESULT, TC_ACT_UNSPEC); | |
2614 | nl_msg_end_nested(request, offset); | |
2615 | } | |
2616 | ||
2617 | static int | |
2618 | tc_add_matchall_policer(struct netdev *netdev, uint32_t kbits_rate, | |
2619 | uint32_t kbits_burst) | |
2620 | { | |
2621 | uint16_t eth_type = (OVS_FORCE uint16_t) htons(ETH_P_ALL); | |
2622 | size_t basic_offset, action_offset, inner_offset; | |
2623 | uint16_t prio = TC_RESERVED_PRIORITY_POLICE; | |
2624 | int ifindex, index, err = 0; | |
2625 | struct tc_police pol_act; | |
2626 | uint32_t block_id = 0; | |
2627 | struct ofpbuf request; | |
2628 | struct ofpbuf *reply; | |
2629 | struct tcmsg *tcmsg; | |
2630 | uint32_t handle = 1; | |
2631 | ||
2632 | err = get_ifindex(netdev, &ifindex); | |
2633 | if (err) { | |
2634 | return err; | |
2635 | } | |
2636 | ||
2637 | index = block_id ? TCM_IFINDEX_MAGIC_BLOCK : ifindex; | |
2638 | tcmsg = tc_make_request(index, RTM_NEWTFILTER, NLM_F_CREATE | NLM_F_ECHO, | |
2639 | &request); | |
2640 | tcmsg->tcm_parent = block_id ? : TC_INGRESS_PARENT; | |
2641 | tcmsg->tcm_info = tc_make_handle(prio, eth_type); | |
2642 | tcmsg->tcm_handle = handle; | |
2643 | ||
2644 | pol_act = tc_matchall_fill_police(kbits_rate, kbits_burst); | |
2645 | nl_msg_put_string(&request, TCA_KIND, "matchall"); | |
2646 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
2647 | action_offset = nl_msg_start_nested(&request, TCA_MATCHALL_ACT); | |
2648 | inner_offset = nl_msg_start_nested(&request, 1); | |
2649 | nl_msg_put_act_police(&request, pol_act); | |
2650 | nl_msg_end_nested(&request, inner_offset); | |
2651 | nl_msg_end_nested(&request, action_offset); | |
2652 | nl_msg_end_nested(&request, basic_offset); | |
2653 | ||
2654 | err = tc_transact(&request, &reply); | |
2655 | if (!err) { | |
2656 | struct tcmsg *tc = | |
2657 | ofpbuf_at_assert(reply, NLMSG_HDRLEN, sizeof *tc); | |
2658 | ofpbuf_delete(reply); | |
2659 | } | |
2660 | ||
2661 | return err; | |
2662 | } | |
2663 | ||
2664 | static int | |
2665 | tc_del_matchall_policer(struct netdev *netdev) | |
2666 | { | |
acdd544c | 2667 | int prio = TC_RESERVED_PRIORITY_POLICE; |
e7f6ba22 | 2668 | uint32_t block_id = 0; |
acdd544c | 2669 | struct tcf_id id; |
e7f6ba22 PJV |
2670 | int ifindex; |
2671 | int err; | |
2672 | ||
2673 | err = get_ifindex(netdev, &ifindex); | |
2674 | if (err) { | |
2675 | return err; | |
2676 | } | |
2677 | ||
acdd544c PB |
2678 | id = tc_make_tcf_id(ifindex, block_id, prio, TC_INGRESS); |
2679 | err = tc_del_filter(&id); | |
e7f6ba22 PJV |
2680 | if (err) { |
2681 | return err; | |
2682 | } | |
2683 | ||
2684 | return 0; | |
2685 | } | |
2686 | ||
f8500004 JP |
2687 | /* Attempts to set input rate limiting (policing) policy. Returns 0 if |
2688 | * successful, otherwise a positive errno value. */ | |
8b61709d | 2689 | static int |
b5d57fc8 | 2690 | netdev_linux_set_policing(struct netdev *netdev_, |
8b61709d BP |
2691 | uint32_t kbits_rate, uint32_t kbits_burst) |
2692 | { | |
b5d57fc8 BP |
2693 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
2694 | const char *netdev_name = netdev_get_name(netdev_); | |
7874bdff | 2695 | int ifindex; |
f8500004 | 2696 | int error; |
8b61709d | 2697 | |
80a86fbe | 2698 | kbits_burst = (!kbits_rate ? 0 /* Force to 0 if no rate specified. */ |
79abacc8 | 2699 | : !kbits_burst ? 8000 /* Default to 8000 kbits if 0. */ |
80a86fbe BP |
2700 | : kbits_burst); /* Stick with user-specified value. */ |
2701 | ||
86383816 | 2702 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2703 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2704 | error = EOPNOTSUPP; | |
2705 | goto out; | |
2706 | } | |
2707 | ||
b5d57fc8 | 2708 | if (netdev->cache_valid & VALID_POLICING) { |
86383816 BP |
2709 | error = netdev->netdev_policing_error; |
2710 | if (error || (netdev->kbits_rate == kbits_rate && | |
2711 | netdev->kbits_burst == kbits_burst)) { | |
c9f71668 | 2712 | /* Assume that settings haven't changed since we last set them. */ |
86383816 | 2713 | goto out; |
c9f71668 | 2714 | } |
b5d57fc8 | 2715 | netdev->cache_valid &= ~VALID_POLICING; |
80a86fbe BP |
2716 | } |
2717 | ||
718be50d | 2718 | COVERAGE_INC(netdev_set_policing); |
7874bdff | 2719 | |
e7f6ba22 PJV |
2720 | /* Use matchall for policing when offloadling ovs with tc-flower. */ |
2721 | if (netdev_is_flow_api_enabled()) { | |
2722 | error = tc_del_matchall_policer(netdev_); | |
2723 | if (kbits_rate) { | |
2724 | error = tc_add_matchall_policer(netdev_, kbits_rate, kbits_burst); | |
2725 | } | |
2726 | ovs_mutex_unlock(&netdev->mutex); | |
2727 | return error; | |
2728 | } | |
2729 | ||
718be50d TZ |
2730 | error = get_ifindex(netdev_, &ifindex); |
2731 | if (error) { | |
2732 | goto out; | |
2733 | } | |
2734 | ||
f8500004 | 2735 | /* Remove any existing ingress qdisc. */ |
95255018 | 2736 | error = tc_add_del_qdisc(ifindex, false, 0, TC_INGRESS); |
f8500004 JP |
2737 | if (error) { |
2738 | VLOG_WARN_RL(&rl, "%s: removing policing failed: %s", | |
10a89ef0 | 2739 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2740 | goto out; |
f8500004 JP |
2741 | } |
2742 | ||
8b61709d | 2743 | if (kbits_rate) { |
95255018 | 2744 | error = tc_add_del_qdisc(ifindex, true, 0, TC_INGRESS); |
f8500004 JP |
2745 | if (error) { |
2746 | VLOG_WARN_RL(&rl, "%s: adding policing qdisc failed: %s", | |
10a89ef0 | 2747 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2748 | goto out; |
8b61709d BP |
2749 | } |
2750 | ||
b5d57fc8 | 2751 | error = tc_add_policer(netdev_, kbits_rate, kbits_burst); |
f8500004 JP |
2752 | if (error){ |
2753 | VLOG_WARN_RL(&rl, "%s: adding policing action failed: %s", | |
10a89ef0 | 2754 | netdev_name, ovs_strerror(error)); |
c9f71668 | 2755 | goto out; |
8b61709d | 2756 | } |
8b61709d BP |
2757 | } |
2758 | ||
b5d57fc8 BP |
2759 | netdev->kbits_rate = kbits_rate; |
2760 | netdev->kbits_burst = kbits_burst; | |
f8500004 | 2761 | |
c9f71668 PS |
2762 | out: |
2763 | if (!error || error == ENODEV) { | |
b5d57fc8 BP |
2764 | netdev->netdev_policing_error = error; |
2765 | netdev->cache_valid |= VALID_POLICING; | |
c9f71668 | 2766 | } |
86383816 | 2767 | ovs_mutex_unlock(&netdev->mutex); |
c9f71668 | 2768 | return error; |
8b61709d BP |
2769 | } |
2770 | ||
c1c9c9c4 BP |
2771 | static int |
2772 | netdev_linux_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
19993ef3 | 2773 | struct sset *types) |
c1c9c9c4 | 2774 | { |
559eb230 | 2775 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2776 | for (opsp = tcs; *opsp != NULL; opsp++) { |
2777 | const struct tc_ops *ops = *opsp; | |
2778 | if (ops->tc_install && ops->ovs_name[0] != '\0') { | |
19993ef3 | 2779 | sset_add(types, ops->ovs_name); |
c1c9c9c4 BP |
2780 | } |
2781 | } | |
2782 | return 0; | |
2783 | } | |
2784 | ||
2785 | static const struct tc_ops * | |
2786 | tc_lookup_ovs_name(const char *name) | |
2787 | { | |
559eb230 | 2788 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2789 | |
2790 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2791 | const struct tc_ops *ops = *opsp; | |
2792 | if (!strcmp(name, ops->ovs_name)) { | |
2793 | return ops; | |
2794 | } | |
2795 | } | |
2796 | return NULL; | |
2797 | } | |
2798 | ||
2799 | static const struct tc_ops * | |
2800 | tc_lookup_linux_name(const char *name) | |
2801 | { | |
559eb230 | 2802 | const struct tc_ops *const *opsp; |
c1c9c9c4 BP |
2803 | |
2804 | for (opsp = tcs; *opsp != NULL; opsp++) { | |
2805 | const struct tc_ops *ops = *opsp; | |
2806 | if (ops->linux_name && !strcmp(name, ops->linux_name)) { | |
2807 | return ops; | |
2808 | } | |
2809 | } | |
2810 | return NULL; | |
2811 | } | |
2812 | ||
93b13be8 | 2813 | static struct tc_queue * |
b5d57fc8 | 2814 | tc_find_queue__(const struct netdev *netdev_, unsigned int queue_id, |
93b13be8 BP |
2815 | size_t hash) |
2816 | { | |
b5d57fc8 | 2817 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
93b13be8 BP |
2818 | struct tc_queue *queue; |
2819 | ||
b5d57fc8 | 2820 | HMAP_FOR_EACH_IN_BUCKET (queue, hmap_node, hash, &netdev->tc->queues) { |
93b13be8 BP |
2821 | if (queue->queue_id == queue_id) { |
2822 | return queue; | |
2823 | } | |
2824 | } | |
2825 | return NULL; | |
2826 | } | |
2827 | ||
2828 | static struct tc_queue * | |
2829 | tc_find_queue(const struct netdev *netdev, unsigned int queue_id) | |
2830 | { | |
2831 | return tc_find_queue__(netdev, queue_id, hash_int(queue_id, 0)); | |
2832 | } | |
2833 | ||
c1c9c9c4 BP |
2834 | static int |
2835 | netdev_linux_get_qos_capabilities(const struct netdev *netdev OVS_UNUSED, | |
2836 | const char *type, | |
2837 | struct netdev_qos_capabilities *caps) | |
2838 | { | |
2839 | const struct tc_ops *ops = tc_lookup_ovs_name(type); | |
2840 | if (!ops) { | |
2841 | return EOPNOTSUPP; | |
2842 | } | |
2843 | caps->n_queues = ops->n_queues; | |
2844 | return 0; | |
2845 | } | |
2846 | ||
2847 | static int | |
b5d57fc8 | 2848 | netdev_linux_get_qos(const struct netdev *netdev_, |
79f1cbe9 | 2849 | const char **typep, struct smap *details) |
c1c9c9c4 | 2850 | { |
b5d57fc8 | 2851 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2852 | int error; |
2853 | ||
86383816 | 2854 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2855 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2856 | error = EOPNOTSUPP; | |
2857 | goto exit; | |
2858 | } | |
2859 | ||
b5d57fc8 | 2860 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2861 | if (!error) { |
2862 | *typep = netdev->tc->ops->ovs_name; | |
2863 | error = (netdev->tc->ops->qdisc_get | |
2864 | ? netdev->tc->ops->qdisc_get(netdev_, details) | |
2865 | : 0); | |
c1c9c9c4 BP |
2866 | } |
2867 | ||
e0e2410d FL |
2868 | exit: |
2869 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2870 | return error; |
c1c9c9c4 BP |
2871 | } |
2872 | ||
2873 | static int | |
b5d57fc8 | 2874 | netdev_linux_set_qos(struct netdev *netdev_, |
79f1cbe9 | 2875 | const char *type, const struct smap *details) |
c1c9c9c4 | 2876 | { |
b5d57fc8 | 2877 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2878 | const struct tc_ops *new_ops; |
2879 | int error; | |
2880 | ||
2881 | new_ops = tc_lookup_ovs_name(type); | |
2882 | if (!new_ops || !new_ops->tc_install) { | |
2883 | return EOPNOTSUPP; | |
2884 | } | |
2885 | ||
6cf888b8 BS |
2886 | if (new_ops == &tc_ops_noop) { |
2887 | return new_ops->tc_install(netdev_, details); | |
2888 | } | |
2889 | ||
86383816 | 2890 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2891 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2892 | error = EOPNOTSUPP; | |
2893 | goto exit; | |
2894 | } | |
2895 | ||
b5d57fc8 | 2896 | error = tc_query_qdisc(netdev_); |
c1c9c9c4 | 2897 | if (error) { |
86383816 | 2898 | goto exit; |
c1c9c9c4 BP |
2899 | } |
2900 | ||
b5d57fc8 | 2901 | if (new_ops == netdev->tc->ops) { |
86383816 | 2902 | error = new_ops->qdisc_set ? new_ops->qdisc_set(netdev_, details) : 0; |
c1c9c9c4 BP |
2903 | } else { |
2904 | /* Delete existing qdisc. */ | |
b5d57fc8 | 2905 | error = tc_del_qdisc(netdev_); |
c1c9c9c4 | 2906 | if (error) { |
86383816 | 2907 | goto exit; |
c1c9c9c4 | 2908 | } |
b5d57fc8 | 2909 | ovs_assert(netdev->tc == NULL); |
c1c9c9c4 BP |
2910 | |
2911 | /* Install new qdisc. */ | |
b5d57fc8 BP |
2912 | error = new_ops->tc_install(netdev_, details); |
2913 | ovs_assert((error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 | 2914 | } |
86383816 BP |
2915 | |
2916 | exit: | |
2917 | ovs_mutex_unlock(&netdev->mutex); | |
2918 | return error; | |
c1c9c9c4 BP |
2919 | } |
2920 | ||
2921 | static int | |
b5d57fc8 | 2922 | netdev_linux_get_queue(const struct netdev *netdev_, |
79f1cbe9 | 2923 | unsigned int queue_id, struct smap *details) |
c1c9c9c4 | 2924 | { |
b5d57fc8 | 2925 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2926 | int error; |
2927 | ||
86383816 | 2928 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2929 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2930 | error = EOPNOTSUPP; | |
2931 | goto exit; | |
2932 | } | |
2933 | ||
b5d57fc8 | 2934 | error = tc_query_qdisc(netdev_); |
86383816 | 2935 | if (!error) { |
b5d57fc8 | 2936 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); |
86383816 | 2937 | error = (queue |
b5d57fc8 | 2938 | ? netdev->tc->ops->class_get(netdev_, queue, details) |
93b13be8 | 2939 | : ENOENT); |
c1c9c9c4 | 2940 | } |
86383816 | 2941 | |
e0e2410d FL |
2942 | exit: |
2943 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2944 | return error; |
c1c9c9c4 BP |
2945 | } |
2946 | ||
2947 | static int | |
b5d57fc8 | 2948 | netdev_linux_set_queue(struct netdev *netdev_, |
79f1cbe9 | 2949 | unsigned int queue_id, const struct smap *details) |
c1c9c9c4 | 2950 | { |
b5d57fc8 | 2951 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2952 | int error; |
2953 | ||
86383816 | 2954 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2955 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2956 | error = EOPNOTSUPP; | |
2957 | goto exit; | |
2958 | } | |
2959 | ||
b5d57fc8 | 2960 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2961 | if (!error) { |
2962 | error = (queue_id < netdev->tc->ops->n_queues | |
2963 | && netdev->tc->ops->class_set | |
2964 | ? netdev->tc->ops->class_set(netdev_, queue_id, details) | |
2965 | : EINVAL); | |
c1c9c9c4 BP |
2966 | } |
2967 | ||
e0e2410d FL |
2968 | exit: |
2969 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2970 | return error; |
c1c9c9c4 BP |
2971 | } |
2972 | ||
2973 | static int | |
b5d57fc8 | 2974 | netdev_linux_delete_queue(struct netdev *netdev_, unsigned int queue_id) |
c1c9c9c4 | 2975 | { |
b5d57fc8 | 2976 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
2977 | int error; |
2978 | ||
86383816 | 2979 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
2980 | if (netdev_linux_netnsid_is_remote(netdev)) { |
2981 | error = EOPNOTSUPP; | |
2982 | goto exit; | |
2983 | } | |
2984 | ||
b5d57fc8 | 2985 | error = tc_query_qdisc(netdev_); |
86383816 BP |
2986 | if (!error) { |
2987 | if (netdev->tc->ops->class_delete) { | |
2988 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
2989 | error = (queue | |
2990 | ? netdev->tc->ops->class_delete(netdev_, queue) | |
2991 | : ENOENT); | |
2992 | } else { | |
2993 | error = EINVAL; | |
2994 | } | |
c1c9c9c4 | 2995 | } |
86383816 | 2996 | |
e0e2410d FL |
2997 | exit: |
2998 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 2999 | return error; |
c1c9c9c4 BP |
3000 | } |
3001 | ||
3002 | static int | |
b5d57fc8 | 3003 | netdev_linux_get_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
3004 | unsigned int queue_id, |
3005 | struct netdev_queue_stats *stats) | |
3006 | { | |
b5d57fc8 | 3007 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3008 | int error; |
3009 | ||
86383816 | 3010 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
3011 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3012 | error = EOPNOTSUPP; | |
3013 | goto exit; | |
3014 | } | |
3015 | ||
b5d57fc8 | 3016 | error = tc_query_qdisc(netdev_); |
86383816 BP |
3017 | if (!error) { |
3018 | if (netdev->tc->ops->class_get_stats) { | |
3019 | const struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
3020 | if (queue) { | |
3021 | stats->created = queue->created; | |
3022 | error = netdev->tc->ops->class_get_stats(netdev_, queue, | |
3023 | stats); | |
3024 | } else { | |
3025 | error = ENOENT; | |
3026 | } | |
3027 | } else { | |
3028 | error = EOPNOTSUPP; | |
6dc34a0d | 3029 | } |
c1c9c9c4 | 3030 | } |
86383816 | 3031 | |
e0e2410d FL |
3032 | exit: |
3033 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 3034 | return error; |
c1c9c9c4 BP |
3035 | } |
3036 | ||
d57695d7 JS |
3037 | struct queue_dump_state { |
3038 | struct nl_dump dump; | |
3039 | struct ofpbuf buf; | |
3040 | }; | |
3041 | ||
23a98ffe | 3042 | static bool |
d57695d7 | 3043 | start_queue_dump(const struct netdev *netdev, struct queue_dump_state *state) |
c1c9c9c4 BP |
3044 | { |
3045 | struct ofpbuf request; | |
3046 | struct tcmsg *tcmsg; | |
3047 | ||
7874bdff | 3048 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, 0, &request); |
23a98ffe BP |
3049 | if (!tcmsg) { |
3050 | return false; | |
3051 | } | |
3c4de644 | 3052 | tcmsg->tcm_parent = 0; |
d57695d7 | 3053 | nl_dump_start(&state->dump, NETLINK_ROUTE, &request); |
c1c9c9c4 | 3054 | ofpbuf_uninit(&request); |
d57695d7 JS |
3055 | |
3056 | ofpbuf_init(&state->buf, NL_DUMP_BUFSIZE); | |
23a98ffe | 3057 | return true; |
c1c9c9c4 BP |
3058 | } |
3059 | ||
d57695d7 JS |
3060 | static int |
3061 | finish_queue_dump(struct queue_dump_state *state) | |
3062 | { | |
3063 | ofpbuf_uninit(&state->buf); | |
3064 | return nl_dump_done(&state->dump); | |
3065 | } | |
3066 | ||
89454bf4 BP |
3067 | struct netdev_linux_queue_state { |
3068 | unsigned int *queues; | |
3069 | size_t cur_queue; | |
3070 | size_t n_queues; | |
3071 | }; | |
3072 | ||
c1c9c9c4 | 3073 | static int |
89454bf4 | 3074 | netdev_linux_queue_dump_start(const struct netdev *netdev_, void **statep) |
c1c9c9c4 | 3075 | { |
e0e2410d | 3076 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3077 | int error; |
3078 | ||
86383816 | 3079 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
3080 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3081 | error = EOPNOTSUPP; | |
3082 | goto exit; | |
3083 | } | |
3084 | ||
b5d57fc8 | 3085 | error = tc_query_qdisc(netdev_); |
86383816 BP |
3086 | if (!error) { |
3087 | if (netdev->tc->ops->class_get) { | |
89454bf4 BP |
3088 | struct netdev_linux_queue_state *state; |
3089 | struct tc_queue *queue; | |
3090 | size_t i; | |
3091 | ||
3092 | *statep = state = xmalloc(sizeof *state); | |
3093 | state->n_queues = hmap_count(&netdev->tc->queues); | |
3094 | state->cur_queue = 0; | |
3095 | state->queues = xmalloc(state->n_queues * sizeof *state->queues); | |
3096 | ||
3097 | i = 0; | |
3098 | HMAP_FOR_EACH (queue, hmap_node, &netdev->tc->queues) { | |
3099 | state->queues[i++] = queue->queue_id; | |
86383816 | 3100 | } |
c1c9c9c4 | 3101 | } else { |
86383816 | 3102 | error = EOPNOTSUPP; |
c1c9c9c4 BP |
3103 | } |
3104 | } | |
c1c9c9c4 | 3105 | |
e0e2410d FL |
3106 | exit: |
3107 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 3108 | return error; |
c1c9c9c4 BP |
3109 | } |
3110 | ||
89454bf4 BP |
3111 | static int |
3112 | netdev_linux_queue_dump_next(const struct netdev *netdev_, void *state_, | |
3113 | unsigned int *queue_idp, struct smap *details) | |
3114 | { | |
e0e2410d | 3115 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
89454bf4 BP |
3116 | struct netdev_linux_queue_state *state = state_; |
3117 | int error = EOF; | |
3118 | ||
3119 | ovs_mutex_lock(&netdev->mutex); | |
e0e2410d FL |
3120 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3121 | error = EOPNOTSUPP; | |
3122 | goto exit; | |
3123 | } | |
3124 | ||
89454bf4 BP |
3125 | while (state->cur_queue < state->n_queues) { |
3126 | unsigned int queue_id = state->queues[state->cur_queue++]; | |
3127 | struct tc_queue *queue = tc_find_queue(netdev_, queue_id); | |
3128 | ||
3129 | if (queue) { | |
3130 | *queue_idp = queue_id; | |
3131 | error = netdev->tc->ops->class_get(netdev_, queue, details); | |
3132 | break; | |
3133 | } | |
3134 | } | |
89454bf4 | 3135 | |
e0e2410d FL |
3136 | exit: |
3137 | ovs_mutex_unlock(&netdev->mutex); | |
89454bf4 BP |
3138 | return error; |
3139 | } | |
3140 | ||
3141 | static int | |
3142 | netdev_linux_queue_dump_done(const struct netdev *netdev OVS_UNUSED, | |
3143 | void *state_) | |
3144 | { | |
3145 | struct netdev_linux_queue_state *state = state_; | |
3146 | ||
3147 | free(state->queues); | |
3148 | free(state); | |
3149 | return 0; | |
3150 | } | |
3151 | ||
c1c9c9c4 | 3152 | static int |
b5d57fc8 | 3153 | netdev_linux_dump_queue_stats(const struct netdev *netdev_, |
c1c9c9c4 BP |
3154 | netdev_dump_queue_stats_cb *cb, void *aux) |
3155 | { | |
b5d57fc8 | 3156 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
3157 | int error; |
3158 | ||
86383816 | 3159 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
3160 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3161 | error = EOPNOTSUPP; | |
3162 | goto exit; | |
3163 | } | |
3164 | ||
b5d57fc8 | 3165 | error = tc_query_qdisc(netdev_); |
86383816 | 3166 | if (!error) { |
d57695d7 | 3167 | struct queue_dump_state state; |
c1c9c9c4 | 3168 | |
86383816 BP |
3169 | if (!netdev->tc->ops->class_dump_stats) { |
3170 | error = EOPNOTSUPP; | |
d57695d7 | 3171 | } else if (!start_queue_dump(netdev_, &state)) { |
86383816 BP |
3172 | error = ENODEV; |
3173 | } else { | |
3174 | struct ofpbuf msg; | |
3175 | int retval; | |
3176 | ||
d57695d7 | 3177 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
86383816 BP |
3178 | retval = netdev->tc->ops->class_dump_stats(netdev_, &msg, |
3179 | cb, aux); | |
3180 | if (retval) { | |
3181 | error = retval; | |
3182 | } | |
3183 | } | |
3184 | ||
d57695d7 | 3185 | retval = finish_queue_dump(&state); |
86383816 BP |
3186 | if (retval) { |
3187 | error = retval; | |
3188 | } | |
c1c9c9c4 BP |
3189 | } |
3190 | } | |
3191 | ||
e0e2410d FL |
3192 | exit: |
3193 | ovs_mutex_unlock(&netdev->mutex); | |
86383816 | 3194 | return error; |
c1c9c9c4 BP |
3195 | } |
3196 | ||
8b61709d | 3197 | static int |
f1acd62b BP |
3198 | netdev_linux_set_in4(struct netdev *netdev_, struct in_addr address, |
3199 | struct in_addr netmask) | |
8b61709d | 3200 | { |
b5d57fc8 | 3201 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
8b61709d BP |
3202 | int error; |
3203 | ||
86383816 | 3204 | ovs_mutex_lock(&netdev->mutex); |
e0e2410d FL |
3205 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3206 | error = EOPNOTSUPP; | |
3207 | goto exit; | |
3208 | } | |
3209 | ||
f1acd62b | 3210 | error = do_set_addr(netdev_, SIOCSIFADDR, "SIOCSIFADDR", address); |
8b61709d | 3211 | if (!error) { |
f1acd62b | 3212 | if (address.s_addr != INADDR_ANY) { |
8b61709d | 3213 | error = do_set_addr(netdev_, SIOCSIFNETMASK, |
f1acd62b | 3214 | "SIOCSIFNETMASK", netmask); |
8b61709d BP |
3215 | } |
3216 | } | |
49af9a3d | 3217 | |
e0e2410d | 3218 | exit: |
86383816 | 3219 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d BP |
3220 | return error; |
3221 | } | |
3222 | ||
7df6932e AW |
3223 | /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address. |
3224 | * Otherwise, sets '*in6' to 'in6addr_any' and returns the corresponding | |
3225 | * error. */ | |
8b61709d | 3226 | static int |
a8704b50 PS |
3227 | netdev_linux_get_addr_list(const struct netdev *netdev_, |
3228 | struct in6_addr **addr, struct in6_addr **mask, int *n_cnt) | |
8b61709d | 3229 | { |
b5d57fc8 | 3230 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
7df6932e | 3231 | int error; |
86383816 BP |
3232 | |
3233 | ovs_mutex_lock(&netdev->mutex); | |
e0e2410d FL |
3234 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3235 | error = EOPNOTSUPP; | |
3236 | goto exit; | |
3237 | } | |
3238 | ||
a8704b50 | 3239 | error = netdev_get_addrs(netdev_get_name(netdev_), addr, mask, n_cnt); |
86383816 | 3240 | |
e0e2410d FL |
3241 | exit: |
3242 | ovs_mutex_unlock(&netdev->mutex); | |
7df6932e | 3243 | return error; |
8b61709d BP |
3244 | } |
3245 | ||
3246 | static void | |
3247 | make_in4_sockaddr(struct sockaddr *sa, struct in_addr addr) | |
3248 | { | |
3249 | struct sockaddr_in sin; | |
3250 | memset(&sin, 0, sizeof sin); | |
3251 | sin.sin_family = AF_INET; | |
3252 | sin.sin_addr = addr; | |
3253 | sin.sin_port = 0; | |
3254 | ||
3255 | memset(sa, 0, sizeof *sa); | |
3256 | memcpy(sa, &sin, sizeof sin); | |
3257 | } | |
3258 | ||
3259 | static int | |
3260 | do_set_addr(struct netdev *netdev, | |
3261 | int ioctl_nr, const char *ioctl_name, struct in_addr addr) | |
3262 | { | |
3263 | struct ifreq ifr; | |
149f577a | 3264 | |
259e0b1a BP |
3265 | make_in4_sockaddr(&ifr.ifr_addr, addr); |
3266 | return af_inet_ifreq_ioctl(netdev_get_name(netdev), &ifr, ioctl_nr, | |
3267 | ioctl_name); | |
8b61709d BP |
3268 | } |
3269 | ||
3270 | /* Adds 'router' as a default IP gateway. */ | |
3271 | static int | |
67a4917b | 3272 | netdev_linux_add_router(struct netdev *netdev OVS_UNUSED, struct in_addr router) |
8b61709d BP |
3273 | { |
3274 | struct in_addr any = { INADDR_ANY }; | |
3275 | struct rtentry rt; | |
3276 | int error; | |
3277 | ||
3278 | memset(&rt, 0, sizeof rt); | |
3279 | make_in4_sockaddr(&rt.rt_dst, any); | |
3280 | make_in4_sockaddr(&rt.rt_gateway, router); | |
3281 | make_in4_sockaddr(&rt.rt_genmask, any); | |
3282 | rt.rt_flags = RTF_UP | RTF_GATEWAY; | |
259e0b1a | 3283 | error = af_inet_ioctl(SIOCADDRT, &rt); |
8b61709d | 3284 | if (error) { |
10a89ef0 | 3285 | VLOG_WARN("ioctl(SIOCADDRT): %s", ovs_strerror(error)); |
8b61709d BP |
3286 | } |
3287 | return error; | |
3288 | } | |
3289 | ||
f1acd62b BP |
3290 | static int |
3291 | netdev_linux_get_next_hop(const struct in_addr *host, struct in_addr *next_hop, | |
3292 | char **netdev_name) | |
3293 | { | |
3294 | static const char fn[] = "/proc/net/route"; | |
3295 | FILE *stream; | |
3296 | char line[256]; | |
3297 | int ln; | |
3298 | ||
3299 | *netdev_name = NULL; | |
3300 | stream = fopen(fn, "r"); | |
3301 | if (stream == NULL) { | |
10a89ef0 | 3302 | VLOG_WARN_RL(&rl, "%s: open failed: %s", fn, ovs_strerror(errno)); |
f1acd62b BP |
3303 | return errno; |
3304 | } | |
3305 | ||
3306 | ln = 0; | |
3307 | while (fgets(line, sizeof line, stream)) { | |
3308 | if (++ln >= 2) { | |
3309 | char iface[17]; | |
dbba996b | 3310 | ovs_be32 dest, gateway, mask; |
f1acd62b BP |
3311 | int refcnt, metric, mtu; |
3312 | unsigned int flags, use, window, irtt; | |
3313 | ||
c2c28dfd BP |
3314 | if (!ovs_scan(line, |
3315 | "%16s %"SCNx32" %"SCNx32" %04X %d %u %d %"SCNx32 | |
3316 | " %d %u %u\n", | |
3317 | iface, &dest, &gateway, &flags, &refcnt, | |
3318 | &use, &metric, &mask, &mtu, &window, &irtt)) { | |
d295e8e9 | 3319 | VLOG_WARN_RL(&rl, "%s: could not parse line %d: %s", |
f1acd62b BP |
3320 | fn, ln, line); |
3321 | continue; | |
3322 | } | |
3323 | if (!(flags & RTF_UP)) { | |
3324 | /* Skip routes that aren't up. */ | |
3325 | continue; | |
3326 | } | |
3327 | ||
3328 | /* The output of 'dest', 'mask', and 'gateway' were given in | |
d295e8e9 | 3329 | * network byte order, so we don't need need any endian |
f1acd62b BP |
3330 | * conversions here. */ |
3331 | if ((dest & mask) == (host->s_addr & mask)) { | |
3332 | if (!gateway) { | |
3333 | /* The host is directly reachable. */ | |
3334 | next_hop->s_addr = 0; | |
3335 | } else { | |
3336 | /* To reach the host, we must go through a gateway. */ | |
3337 | next_hop->s_addr = gateway; | |
3338 | } | |
3339 | *netdev_name = xstrdup(iface); | |
3340 | fclose(stream); | |
3341 | return 0; | |
3342 | } | |
3343 | } | |
3344 | } | |
3345 | ||
3346 | fclose(stream); | |
3347 | return ENXIO; | |
3348 | } | |
3349 | ||
e210037e | 3350 | static int |
b5d57fc8 | 3351 | netdev_linux_get_status(const struct netdev *netdev_, struct smap *smap) |
e210037e | 3352 | { |
b5d57fc8 | 3353 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
275707c3 EJ |
3354 | int error = 0; |
3355 | ||
86383816 | 3356 | ovs_mutex_lock(&netdev->mutex); |
b5d57fc8 BP |
3357 | if (!(netdev->cache_valid & VALID_DRVINFO)) { |
3358 | struct ethtool_cmd *cmd = (struct ethtool_cmd *) &netdev->drvinfo; | |
275707c3 EJ |
3359 | |
3360 | COVERAGE_INC(netdev_get_ethtool); | |
b5d57fc8 BP |
3361 | memset(&netdev->drvinfo, 0, sizeof netdev->drvinfo); |
3362 | error = netdev_linux_do_ethtool(netdev->up.name, | |
275707c3 EJ |
3363 | cmd, |
3364 | ETHTOOL_GDRVINFO, | |
3365 | "ETHTOOL_GDRVINFO"); | |
3366 | if (!error) { | |
b5d57fc8 | 3367 | netdev->cache_valid |= VALID_DRVINFO; |
275707c3 EJ |
3368 | } |
3369 | } | |
e210037e | 3370 | |
e210037e | 3371 | if (!error) { |
b5d57fc8 BP |
3372 | smap_add(smap, "driver_name", netdev->drvinfo.driver); |
3373 | smap_add(smap, "driver_version", netdev->drvinfo.version); | |
3374 | smap_add(smap, "firmware_version", netdev->drvinfo.fw_version); | |
e210037e | 3375 | } |
86383816 BP |
3376 | ovs_mutex_unlock(&netdev->mutex); |
3377 | ||
e210037e AE |
3378 | return error; |
3379 | } | |
3380 | ||
4f925bd3 | 3381 | static int |
275707c3 EJ |
3382 | netdev_internal_get_status(const struct netdev *netdev OVS_UNUSED, |
3383 | struct smap *smap) | |
4f925bd3 | 3384 | { |
79f1cbe9 | 3385 | smap_add(smap, "driver_name", "openvswitch"); |
4f925bd3 PS |
3386 | return 0; |
3387 | } | |
3388 | ||
25db83be JH |
3389 | static uint32_t |
3390 | netdev_linux_get_block_id(struct netdev *netdev_) | |
3391 | { | |
3392 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3393 | uint32_t block_id = 0; | |
3394 | ||
3395 | ovs_mutex_lock(&netdev->mutex); | |
3396 | /* Ensure the linux netdev has had its fields populated. */ | |
3397 | if (!(netdev->cache_valid & VALID_IFINDEX)) { | |
3398 | netdev_linux_update_via_netlink(netdev); | |
3399 | } | |
3400 | ||
3401 | /* Only assigning block ids to linux netdevs that are LAG masters. */ | |
3402 | if (netdev->is_lag_master) { | |
3403 | block_id = netdev->ifindex; | |
3404 | } | |
3405 | ovs_mutex_unlock(&netdev->mutex); | |
3406 | ||
3407 | return block_id; | |
3408 | } | |
3409 | ||
8b61709d BP |
3410 | /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be |
3411 | * successfully retrieved, it stores the corresponding MAC address in 'mac' and | |
3412 | * returns 0. Otherwise, it returns a positive errno value; in particular, | |
3413 | * ENXIO indicates that there is not ARP table entry for 'ip' on 'netdev'. */ | |
3414 | static int | |
3415 | netdev_linux_arp_lookup(const struct netdev *netdev, | |
74ff3298 | 3416 | ovs_be32 ip, struct eth_addr *mac) |
8b61709d BP |
3417 | { |
3418 | struct arpreq r; | |
c100e025 | 3419 | struct sockaddr_in sin; |
8b61709d BP |
3420 | int retval; |
3421 | ||
3422 | memset(&r, 0, sizeof r); | |
f2cc621b | 3423 | memset(&sin, 0, sizeof sin); |
c100e025 BP |
3424 | sin.sin_family = AF_INET; |
3425 | sin.sin_addr.s_addr = ip; | |
3426 | sin.sin_port = 0; | |
3427 | memcpy(&r.arp_pa, &sin, sizeof sin); | |
8b61709d BP |
3428 | r.arp_ha.sa_family = ARPHRD_ETHER; |
3429 | r.arp_flags = 0; | |
71d7c22f | 3430 | ovs_strzcpy(r.arp_dev, netdev_get_name(netdev), sizeof r.arp_dev); |
8b61709d | 3431 | COVERAGE_INC(netdev_arp_lookup); |
259e0b1a | 3432 | retval = af_inet_ioctl(SIOCGARP, &r); |
8b61709d BP |
3433 | if (!retval) { |
3434 | memcpy(mac, r.arp_ha.sa_data, ETH_ADDR_LEN); | |
3435 | } else if (retval != ENXIO) { | |
3436 | VLOG_WARN_RL(&rl, "%s: could not look up ARP entry for "IP_FMT": %s", | |
10a89ef0 BP |
3437 | netdev_get_name(netdev), IP_ARGS(ip), |
3438 | ovs_strerror(retval)); | |
8b61709d BP |
3439 | } |
3440 | return retval; | |
3441 | } | |
3442 | ||
b24751ff | 3443 | static unsigned int |
8b61709d BP |
3444 | nd_to_iff_flags(enum netdev_flags nd) |
3445 | { | |
b24751ff | 3446 | unsigned int iff = 0; |
8b61709d BP |
3447 | if (nd & NETDEV_UP) { |
3448 | iff |= IFF_UP; | |
3449 | } | |
3450 | if (nd & NETDEV_PROMISC) { | |
3451 | iff |= IFF_PROMISC; | |
3452 | } | |
7ba19d41 AC |
3453 | if (nd & NETDEV_LOOPBACK) { |
3454 | iff |= IFF_LOOPBACK; | |
3455 | } | |
8b61709d BP |
3456 | return iff; |
3457 | } | |
3458 | ||
3459 | static int | |
b24751ff | 3460 | iff_to_nd_flags(unsigned int iff) |
8b61709d BP |
3461 | { |
3462 | enum netdev_flags nd = 0; | |
3463 | if (iff & IFF_UP) { | |
3464 | nd |= NETDEV_UP; | |
3465 | } | |
3466 | if (iff & IFF_PROMISC) { | |
3467 | nd |= NETDEV_PROMISC; | |
3468 | } | |
7ba19d41 AC |
3469 | if (iff & IFF_LOOPBACK) { |
3470 | nd |= NETDEV_LOOPBACK; | |
3471 | } | |
8b61709d BP |
3472 | return nd; |
3473 | } | |
3474 | ||
3475 | static int | |
4f9f3f21 BP |
3476 | update_flags(struct netdev_linux *netdev, enum netdev_flags off, |
3477 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
3478 | OVS_REQUIRES(netdev->mutex) | |
8b61709d | 3479 | { |
b24751ff | 3480 | unsigned int old_flags, new_flags; |
c37d4da4 EJ |
3481 | int error = 0; |
3482 | ||
b5d57fc8 | 3483 | old_flags = netdev->ifi_flags; |
c37d4da4 EJ |
3484 | *old_flagsp = iff_to_nd_flags(old_flags); |
3485 | new_flags = (old_flags & ~nd_to_iff_flags(off)) | nd_to_iff_flags(on); | |
3486 | if (new_flags != old_flags) { | |
4f9f3f21 BP |
3487 | error = set_flags(netdev_get_name(&netdev->up), new_flags); |
3488 | get_flags(&netdev->up, &netdev->ifi_flags); | |
8b61709d | 3489 | } |
4f9f3f21 BP |
3490 | |
3491 | return error; | |
3492 | } | |
3493 | ||
3494 | static int | |
3495 | netdev_linux_update_flags(struct netdev *netdev_, enum netdev_flags off, | |
3496 | enum netdev_flags on, enum netdev_flags *old_flagsp) | |
3497 | { | |
3498 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
756819dd | 3499 | int error = 0; |
4f9f3f21 BP |
3500 | |
3501 | ovs_mutex_lock(&netdev->mutex); | |
756819dd FL |
3502 | if (on || off) { |
3503 | /* Changing flags over netlink isn't support yet. */ | |
e0e2410d FL |
3504 | if (netdev_linux_netnsid_is_remote(netdev)) { |
3505 | error = EOPNOTSUPP; | |
3506 | goto exit; | |
3507 | } | |
756819dd FL |
3508 | error = update_flags(netdev, off, on, old_flagsp); |
3509 | } else { | |
3510 | /* Try reading flags over netlink, or fall back to ioctl. */ | |
3511 | if (!netdev_linux_update_via_netlink(netdev)) { | |
3512 | *old_flagsp = iff_to_nd_flags(netdev->ifi_flags); | |
3513 | } else { | |
3514 | error = update_flags(netdev, off, on, old_flagsp); | |
3515 | } | |
3516 | } | |
e0e2410d FL |
3517 | |
3518 | exit: | |
86383816 | 3519 | ovs_mutex_unlock(&netdev->mutex); |
8b61709d BP |
3520 | return error; |
3521 | } | |
3522 | ||
89c09c1c BP |
3523 | #define NETDEV_LINUX_CLASS_COMMON \ |
3524 | .run = netdev_linux_run, \ | |
3525 | .wait = netdev_linux_wait, \ | |
3526 | .alloc = netdev_linux_alloc, \ | |
89c09c1c | 3527 | .dealloc = netdev_linux_dealloc, \ |
89c09c1c BP |
3528 | .send_wait = netdev_linux_send_wait, \ |
3529 | .set_etheraddr = netdev_linux_set_etheraddr, \ | |
3530 | .get_etheraddr = netdev_linux_get_etheraddr, \ | |
3531 | .get_mtu = netdev_linux_get_mtu, \ | |
3532 | .set_mtu = netdev_linux_set_mtu, \ | |
3533 | .get_ifindex = netdev_linux_get_ifindex, \ | |
3534 | .get_carrier = netdev_linux_get_carrier, \ | |
3535 | .get_carrier_resets = netdev_linux_get_carrier_resets, \ | |
3536 | .set_miimon_interval = netdev_linux_set_miimon_interval, \ | |
3537 | .set_advertisements = netdev_linux_set_advertisements, \ | |
3538 | .set_policing = netdev_linux_set_policing, \ | |
3539 | .get_qos_types = netdev_linux_get_qos_types, \ | |
3540 | .get_qos_capabilities = netdev_linux_get_qos_capabilities, \ | |
3541 | .get_qos = netdev_linux_get_qos, \ | |
3542 | .set_qos = netdev_linux_set_qos, \ | |
3543 | .get_queue = netdev_linux_get_queue, \ | |
3544 | .set_queue = netdev_linux_set_queue, \ | |
3545 | .delete_queue = netdev_linux_delete_queue, \ | |
3546 | .get_queue_stats = netdev_linux_get_queue_stats, \ | |
3547 | .queue_dump_start = netdev_linux_queue_dump_start, \ | |
3548 | .queue_dump_next = netdev_linux_queue_dump_next, \ | |
3549 | .queue_dump_done = netdev_linux_queue_dump_done, \ | |
3550 | .dump_queue_stats = netdev_linux_dump_queue_stats, \ | |
3551 | .set_in4 = netdev_linux_set_in4, \ | |
3552 | .get_addr_list = netdev_linux_get_addr_list, \ | |
3553 | .add_router = netdev_linux_add_router, \ | |
3554 | .get_next_hop = netdev_linux_get_next_hop, \ | |
3555 | .arp_lookup = netdev_linux_arp_lookup, \ | |
3556 | .update_flags = netdev_linux_update_flags, \ | |
3557 | .rxq_alloc = netdev_linux_rxq_alloc, \ | |
89c09c1c | 3558 | .rxq_dealloc = netdev_linux_rxq_dealloc, \ |
89c09c1c BP |
3559 | .rxq_wait = netdev_linux_rxq_wait, \ |
3560 | .rxq_drain = netdev_linux_rxq_drain | |
3561 | ||
3562 | const struct netdev_class netdev_linux_class = { | |
3563 | NETDEV_LINUX_CLASS_COMMON, | |
89c09c1c | 3564 | .type = "system", |
0de1b425 | 3565 | .is_pmd = false, |
89c09c1c | 3566 | .construct = netdev_linux_construct, |
0de1b425 | 3567 | .destruct = netdev_linux_destruct, |
89c09c1c BP |
3568 | .get_stats = netdev_linux_get_stats, |
3569 | .get_features = netdev_linux_get_features, | |
3570 | .get_status = netdev_linux_get_status, | |
0de1b425 WT |
3571 | .get_block_id = netdev_linux_get_block_id, |
3572 | .send = netdev_linux_send, | |
3573 | .rxq_construct = netdev_linux_rxq_construct, | |
3574 | .rxq_destruct = netdev_linux_rxq_destruct, | |
3575 | .rxq_recv = netdev_linux_rxq_recv, | |
89c09c1c BP |
3576 | }; |
3577 | ||
3578 | const struct netdev_class netdev_tap_class = { | |
3579 | NETDEV_LINUX_CLASS_COMMON, | |
3580 | .type = "tap", | |
0de1b425 | 3581 | .is_pmd = false, |
89c09c1c | 3582 | .construct = netdev_linux_construct_tap, |
0de1b425 | 3583 | .destruct = netdev_linux_destruct, |
89c09c1c BP |
3584 | .get_stats = netdev_tap_get_stats, |
3585 | .get_features = netdev_linux_get_features, | |
3586 | .get_status = netdev_linux_get_status, | |
0de1b425 WT |
3587 | .send = netdev_linux_send, |
3588 | .rxq_construct = netdev_linux_rxq_construct, | |
3589 | .rxq_destruct = netdev_linux_rxq_destruct, | |
3590 | .rxq_recv = netdev_linux_rxq_recv, | |
89c09c1c BP |
3591 | }; |
3592 | ||
3593 | const struct netdev_class netdev_internal_class = { | |
3594 | NETDEV_LINUX_CLASS_COMMON, | |
3595 | .type = "internal", | |
0de1b425 | 3596 | .is_pmd = false, |
89c09c1c | 3597 | .construct = netdev_linux_construct, |
0de1b425 | 3598 | .destruct = netdev_linux_destruct, |
89c09c1c BP |
3599 | .get_stats = netdev_internal_get_stats, |
3600 | .get_status = netdev_internal_get_status, | |
0de1b425 WT |
3601 | .send = netdev_linux_send, |
3602 | .rxq_construct = netdev_linux_rxq_construct, | |
3603 | .rxq_destruct = netdev_linux_rxq_destruct, | |
3604 | .rxq_recv = netdev_linux_rxq_recv, | |
89c09c1c | 3605 | }; |
0de1b425 WT |
3606 | |
3607 | #ifdef HAVE_AF_XDP | |
3608 | const struct netdev_class netdev_afxdp_class = { | |
3609 | NETDEV_LINUX_CLASS_COMMON, | |
3610 | .type = "afxdp", | |
3611 | .is_pmd = true, | |
7bf075d9 | 3612 | .init = netdev_afxdp_init, |
f627cf1d | 3613 | .construct = netdev_afxdp_construct, |
0de1b425 WT |
3614 | .destruct = netdev_afxdp_destruct, |
3615 | .get_stats = netdev_afxdp_get_stats, | |
d560bc1b | 3616 | .get_custom_stats = netdev_afxdp_get_custom_stats, |
0de1b425 WT |
3617 | .get_status = netdev_linux_get_status, |
3618 | .set_config = netdev_afxdp_set_config, | |
3619 | .get_config = netdev_afxdp_get_config, | |
3620 | .reconfigure = netdev_afxdp_reconfigure, | |
105cf8df | 3621 | .get_numa_id = netdev_linux_get_numa_id, |
0de1b425 WT |
3622 | .send = netdev_afxdp_batch_send, |
3623 | .rxq_construct = netdev_afxdp_rxq_construct, | |
3624 | .rxq_destruct = netdev_afxdp_rxq_destruct, | |
3625 | .rxq_recv = netdev_afxdp_rxq_recv, | |
3626 | }; | |
3627 | #endif | |
8b61709d | 3628 | \f |
677d9158 JV |
3629 | |
3630 | #define CODEL_N_QUEUES 0x0000 | |
3631 | ||
2f4298ce BP |
3632 | /* In sufficiently new kernel headers these are defined as enums in |
3633 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3634 | * kernels. (This overrides any enum definition in the header file but that's | |
3635 | * harmless.) */ | |
3636 | #define TCA_CODEL_TARGET 1 | |
3637 | #define TCA_CODEL_LIMIT 2 | |
3638 | #define TCA_CODEL_INTERVAL 3 | |
3639 | ||
677d9158 JV |
3640 | struct codel { |
3641 | struct tc tc; | |
3642 | uint32_t target; | |
3643 | uint32_t limit; | |
3644 | uint32_t interval; | |
3645 | }; | |
3646 | ||
3647 | static struct codel * | |
3648 | codel_get__(const struct netdev *netdev_) | |
3649 | { | |
3650 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3651 | return CONTAINER_OF(netdev->tc, struct codel, tc); | |
3652 | } | |
3653 | ||
3654 | static void | |
3655 | codel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3656 | uint32_t interval) | |
3657 | { | |
3658 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3659 | struct codel *codel; | |
3660 | ||
3661 | codel = xmalloc(sizeof *codel); | |
3662 | tc_init(&codel->tc, &tc_ops_codel); | |
3663 | codel->target = target; | |
3664 | codel->limit = limit; | |
3665 | codel->interval = interval; | |
3666 | ||
3667 | netdev->tc = &codel->tc; | |
3668 | } | |
3669 | ||
3670 | static int | |
3671 | codel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3672 | uint32_t interval) | |
3673 | { | |
3674 | size_t opt_offset; | |
3675 | struct ofpbuf request; | |
3676 | struct tcmsg *tcmsg; | |
3677 | uint32_t otarget, olimit, ointerval; | |
3678 | int error; | |
3679 | ||
3680 | tc_del_qdisc(netdev); | |
3681 | ||
7874bdff RD |
3682 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3683 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3684 | if (!tcmsg) { |
3685 | return ENODEV; | |
3686 | } | |
3687 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3688 | tcmsg->tcm_parent = TC_H_ROOT; | |
3689 | ||
3690 | otarget = target ? target : 5000; | |
3691 | olimit = limit ? limit : 10240; | |
3692 | ointerval = interval ? interval : 100000; | |
3693 | ||
3694 | nl_msg_put_string(&request, TCA_KIND, "codel"); | |
3695 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3696 | nl_msg_put_u32(&request, TCA_CODEL_TARGET, otarget); | |
3697 | nl_msg_put_u32(&request, TCA_CODEL_LIMIT, olimit); | |
3698 | nl_msg_put_u32(&request, TCA_CODEL_INTERVAL, ointerval); | |
3699 | nl_msg_end_nested(&request, opt_offset); | |
3700 | ||
3701 | error = tc_transact(&request, NULL); | |
3702 | if (error) { | |
3703 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3704 | "target %u, limit %u, interval %u error %d(%s)", | |
3705 | netdev_get_name(netdev), | |
3706 | otarget, olimit, ointerval, | |
3707 | error, ovs_strerror(error)); | |
3708 | } | |
3709 | return error; | |
3710 | } | |
3711 | ||
3712 | static void | |
3713 | codel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3714 | const struct smap *details, struct codel *codel) | |
3715 | { | |
13c1637f BP |
3716 | codel->target = smap_get_ullong(details, "target", 0); |
3717 | codel->limit = smap_get_ullong(details, "limit", 0); | |
3718 | codel->interval = smap_get_ullong(details, "interval", 0); | |
677d9158 JV |
3719 | |
3720 | if (!codel->target) { | |
3721 | codel->target = 5000; | |
3722 | } | |
3723 | if (!codel->limit) { | |
3724 | codel->limit = 10240; | |
3725 | } | |
3726 | if (!codel->interval) { | |
3727 | codel->interval = 100000; | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | static int | |
3732 | codel_tc_install(struct netdev *netdev, const struct smap *details) | |
3733 | { | |
3734 | int error; | |
3735 | struct codel codel; | |
3736 | ||
3737 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3738 | error = codel_setup_qdisc__(netdev, codel.target, codel.limit, | |
3739 | codel.interval); | |
3740 | if (!error) { | |
3741 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3742 | } | |
3743 | return error; | |
3744 | } | |
3745 | ||
3746 | static int | |
3747 | codel_parse_tca_options__(struct nlattr *nl_options, struct codel *codel) | |
3748 | { | |
3749 | static const struct nl_policy tca_codel_policy[] = { | |
3750 | [TCA_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3751 | [TCA_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3752 | [TCA_CODEL_INTERVAL] = { .type = NL_A_U32 } | |
3753 | }; | |
3754 | ||
3755 | struct nlattr *attrs[ARRAY_SIZE(tca_codel_policy)]; | |
3756 | ||
3757 | if (!nl_parse_nested(nl_options, tca_codel_policy, | |
3758 | attrs, ARRAY_SIZE(tca_codel_policy))) { | |
3759 | VLOG_WARN_RL(&rl, "failed to parse CoDel class options"); | |
3760 | return EPROTO; | |
3761 | } | |
3762 | ||
3763 | codel->target = nl_attr_get_u32(attrs[TCA_CODEL_TARGET]); | |
3764 | codel->limit = nl_attr_get_u32(attrs[TCA_CODEL_LIMIT]); | |
3765 | codel->interval = nl_attr_get_u32(attrs[TCA_CODEL_INTERVAL]); | |
3766 | return 0; | |
3767 | } | |
3768 | ||
3769 | static int | |
3770 | codel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
3771 | { | |
3772 | struct nlattr *nlattr; | |
3773 | const char * kind; | |
3774 | int error; | |
3775 | struct codel codel; | |
3776 | ||
3777 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
3778 | if (error != 0) { | |
3779 | return error; | |
3780 | } | |
3781 | ||
3782 | error = codel_parse_tca_options__(nlattr, &codel); | |
3783 | if (error != 0) { | |
3784 | return error; | |
3785 | } | |
3786 | ||
3787 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3788 | return 0; | |
3789 | } | |
3790 | ||
3791 | ||
3792 | static void | |
3793 | codel_tc_destroy(struct tc *tc) | |
3794 | { | |
3795 | struct codel *codel = CONTAINER_OF(tc, struct codel, tc); | |
3796 | tc_destroy(tc); | |
3797 | free(codel); | |
3798 | } | |
3799 | ||
3800 | static int | |
3801 | codel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
3802 | { | |
3803 | const struct codel *codel = codel_get__(netdev); | |
3804 | smap_add_format(details, "target", "%u", codel->target); | |
3805 | smap_add_format(details, "limit", "%u", codel->limit); | |
3806 | smap_add_format(details, "interval", "%u", codel->interval); | |
3807 | return 0; | |
3808 | } | |
3809 | ||
3810 | static int | |
3811 | codel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
3812 | { | |
3813 | struct codel codel; | |
3814 | ||
3815 | codel_parse_qdisc_details__(netdev, details, &codel); | |
3816 | codel_install__(netdev, codel.target, codel.limit, codel.interval); | |
3817 | codel_get__(netdev)->target = codel.target; | |
3818 | codel_get__(netdev)->limit = codel.limit; | |
3819 | codel_get__(netdev)->interval = codel.interval; | |
3820 | return 0; | |
3821 | } | |
3822 | ||
3823 | static const struct tc_ops tc_ops_codel = { | |
89c09c1c BP |
3824 | .linux_name = "codel", |
3825 | .ovs_name = "linux-codel", | |
3826 | .n_queues = CODEL_N_QUEUES, | |
3827 | .tc_install = codel_tc_install, | |
3828 | .tc_load = codel_tc_load, | |
3829 | .tc_destroy = codel_tc_destroy, | |
3830 | .qdisc_get = codel_qdisc_get, | |
3831 | .qdisc_set = codel_qdisc_set, | |
677d9158 JV |
3832 | }; |
3833 | \f | |
3834 | /* FQ-CoDel traffic control class. */ | |
3835 | ||
3836 | #define FQCODEL_N_QUEUES 0x0000 | |
3837 | ||
2f4298ce BP |
3838 | /* In sufficiently new kernel headers these are defined as enums in |
3839 | * <linux/pkt_sched.h>. Define them here as macros to help out with older | |
3840 | * kernels. (This overrides any enum definition in the header file but that's | |
3841 | * harmless.) */ | |
3842 | #define TCA_FQ_CODEL_TARGET 1 | |
3843 | #define TCA_FQ_CODEL_LIMIT 2 | |
3844 | #define TCA_FQ_CODEL_INTERVAL 3 | |
3845 | #define TCA_FQ_CODEL_ECN 4 | |
3846 | #define TCA_FQ_CODEL_FLOWS 5 | |
3847 | #define TCA_FQ_CODEL_QUANTUM 6 | |
3848 | ||
677d9158 JV |
3849 | struct fqcodel { |
3850 | struct tc tc; | |
3851 | uint32_t target; | |
3852 | uint32_t limit; | |
3853 | uint32_t interval; | |
3854 | uint32_t flows; | |
3855 | uint32_t quantum; | |
3856 | }; | |
3857 | ||
3858 | static struct fqcodel * | |
3859 | fqcodel_get__(const struct netdev *netdev_) | |
3860 | { | |
3861 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3862 | return CONTAINER_OF(netdev->tc, struct fqcodel, tc); | |
3863 | } | |
3864 | ||
3865 | static void | |
3866 | fqcodel_install__(struct netdev *netdev_, uint32_t target, uint32_t limit, | |
3867 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3868 | { | |
3869 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
3870 | struct fqcodel *fqcodel; | |
3871 | ||
3872 | fqcodel = xmalloc(sizeof *fqcodel); | |
3873 | tc_init(&fqcodel->tc, &tc_ops_fqcodel); | |
3874 | fqcodel->target = target; | |
3875 | fqcodel->limit = limit; | |
3876 | fqcodel->interval = interval; | |
3877 | fqcodel->flows = flows; | |
3878 | fqcodel->quantum = quantum; | |
3879 | ||
3880 | netdev->tc = &fqcodel->tc; | |
3881 | } | |
3882 | ||
3883 | static int | |
3884 | fqcodel_setup_qdisc__(struct netdev *netdev, uint32_t target, uint32_t limit, | |
3885 | uint32_t interval, uint32_t flows, uint32_t quantum) | |
3886 | { | |
3887 | size_t opt_offset; | |
3888 | struct ofpbuf request; | |
3889 | struct tcmsg *tcmsg; | |
3890 | uint32_t otarget, olimit, ointerval, oflows, oquantum; | |
3891 | int error; | |
3892 | ||
3893 | tc_del_qdisc(netdev); | |
3894 | ||
7874bdff RD |
3895 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
3896 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
3897 | if (!tcmsg) { |
3898 | return ENODEV; | |
3899 | } | |
3900 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
3901 | tcmsg->tcm_parent = TC_H_ROOT; | |
3902 | ||
3903 | otarget = target ? target : 5000; | |
3904 | olimit = limit ? limit : 10240; | |
3905 | ointerval = interval ? interval : 100000; | |
3906 | oflows = flows ? flows : 1024; | |
3907 | oquantum = quantum ? quantum : 1514; /* fq_codel default quantum is 1514 | |
3908 | not mtu */ | |
3909 | ||
3910 | nl_msg_put_string(&request, TCA_KIND, "fq_codel"); | |
3911 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
3912 | nl_msg_put_u32(&request, TCA_FQ_CODEL_TARGET, otarget); | |
3913 | nl_msg_put_u32(&request, TCA_FQ_CODEL_LIMIT, olimit); | |
3914 | nl_msg_put_u32(&request, TCA_FQ_CODEL_INTERVAL, ointerval); | |
3915 | nl_msg_put_u32(&request, TCA_FQ_CODEL_FLOWS, oflows); | |
3916 | nl_msg_put_u32(&request, TCA_FQ_CODEL_QUANTUM, oquantum); | |
3917 | nl_msg_end_nested(&request, opt_offset); | |
3918 | ||
3919 | error = tc_transact(&request, NULL); | |
3920 | if (error) { | |
3921 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
3922 | "target %u, limit %u, interval %u, flows %u, quantum %u error %d(%s)", | |
3923 | netdev_get_name(netdev), | |
3924 | otarget, olimit, ointerval, oflows, oquantum, | |
3925 | error, ovs_strerror(error)); | |
3926 | } | |
3927 | return error; | |
3928 | } | |
3929 | ||
3930 | static void | |
3931 | fqcodel_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
3932 | const struct smap *details, struct fqcodel *fqcodel) | |
3933 | { | |
13c1637f BP |
3934 | fqcodel->target = smap_get_ullong(details, "target", 0); |
3935 | fqcodel->limit = smap_get_ullong(details, "limit", 0); | |
3936 | fqcodel->interval = smap_get_ullong(details, "interval", 0); | |
3937 | fqcodel->flows = smap_get_ullong(details, "flows", 0); | |
3938 | fqcodel->quantum = smap_get_ullong(details, "quantum", 0); | |
3939 | ||
677d9158 JV |
3940 | if (!fqcodel->target) { |
3941 | fqcodel->target = 5000; | |
3942 | } | |
3943 | if (!fqcodel->limit) { | |
3944 | fqcodel->limit = 10240; | |
3945 | } | |
3946 | if (!fqcodel->interval) { | |
3947 | fqcodel->interval = 1000000; | |
3948 | } | |
3949 | if (!fqcodel->flows) { | |
3950 | fqcodel->flows = 1024; | |
3951 | } | |
3952 | if (!fqcodel->quantum) { | |
3953 | fqcodel->quantum = 1514; | |
3954 | } | |
3955 | } | |
3956 | ||
3957 | static int | |
3958 | fqcodel_tc_install(struct netdev *netdev, const struct smap *details) | |
3959 | { | |
3960 | int error; | |
3961 | struct fqcodel fqcodel; | |
3962 | ||
3963 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
3964 | error = fqcodel_setup_qdisc__(netdev, fqcodel.target, fqcodel.limit, | |
3965 | fqcodel.interval, fqcodel.flows, | |
3966 | fqcodel.quantum); | |
3967 | if (!error) { | |
3968 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, | |
3969 | fqcodel.interval, fqcodel.flows, fqcodel.quantum); | |
3970 | } | |
3971 | return error; | |
3972 | } | |
3973 | ||
3974 | static int | |
3975 | fqcodel_parse_tca_options__(struct nlattr *nl_options, struct fqcodel *fqcodel) | |
3976 | { | |
3977 | static const struct nl_policy tca_fqcodel_policy[] = { | |
3978 | [TCA_FQ_CODEL_TARGET] = { .type = NL_A_U32 }, | |
3979 | [TCA_FQ_CODEL_LIMIT] = { .type = NL_A_U32 }, | |
3980 | [TCA_FQ_CODEL_INTERVAL] = { .type = NL_A_U32 }, | |
3981 | [TCA_FQ_CODEL_FLOWS] = { .type = NL_A_U32 }, | |
3982 | [TCA_FQ_CODEL_QUANTUM] = { .type = NL_A_U32 } | |
3983 | }; | |
3984 | ||
3985 | struct nlattr *attrs[ARRAY_SIZE(tca_fqcodel_policy)]; | |
3986 | ||
3987 | if (!nl_parse_nested(nl_options, tca_fqcodel_policy, | |
3988 | attrs, ARRAY_SIZE(tca_fqcodel_policy))) { | |
3989 | VLOG_WARN_RL(&rl, "failed to parse FQ_CoDel class options"); | |
3990 | return EPROTO; | |
3991 | } | |
3992 | ||
3993 | fqcodel->target = nl_attr_get_u32(attrs[TCA_FQ_CODEL_TARGET]); | |
3994 | fqcodel->limit = nl_attr_get_u32(attrs[TCA_FQ_CODEL_LIMIT]); | |
3995 | fqcodel->interval =nl_attr_get_u32(attrs[TCA_FQ_CODEL_INTERVAL]); | |
3996 | fqcodel->flows = nl_attr_get_u32(attrs[TCA_FQ_CODEL_FLOWS]); | |
3997 | fqcodel->quantum = nl_attr_get_u32(attrs[TCA_FQ_CODEL_QUANTUM]); | |
3998 | return 0; | |
3999 | } | |
4000 | ||
4001 | static int | |
4002 | fqcodel_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
4003 | { | |
4004 | struct nlattr *nlattr; | |
4005 | const char * kind; | |
4006 | int error; | |
4007 | struct fqcodel fqcodel; | |
4008 | ||
4009 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
4010 | if (error != 0) { | |
4011 | return error; | |
4012 | } | |
4013 | ||
4014 | error = fqcodel_parse_tca_options__(nlattr, &fqcodel); | |
4015 | if (error != 0) { | |
4016 | return error; | |
4017 | } | |
4018 | ||
4019 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
4020 | fqcodel.flows, fqcodel.quantum); | |
4021 | return 0; | |
4022 | } | |
4023 | ||
4024 | static void | |
4025 | fqcodel_tc_destroy(struct tc *tc) | |
4026 | { | |
4027 | struct fqcodel *fqcodel = CONTAINER_OF(tc, struct fqcodel, tc); | |
4028 | tc_destroy(tc); | |
4029 | free(fqcodel); | |
4030 | } | |
4031 | ||
4032 | static int | |
4033 | fqcodel_qdisc_get(const struct netdev *netdev, struct smap *details) | |
4034 | { | |
4035 | const struct fqcodel *fqcodel = fqcodel_get__(netdev); | |
4036 | smap_add_format(details, "target", "%u", fqcodel->target); | |
4037 | smap_add_format(details, "limit", "%u", fqcodel->limit); | |
4038 | smap_add_format(details, "interval", "%u", fqcodel->interval); | |
4039 | smap_add_format(details, "flows", "%u", fqcodel->flows); | |
4040 | smap_add_format(details, "quantum", "%u", fqcodel->quantum); | |
4041 | return 0; | |
4042 | } | |
4043 | ||
4044 | static int | |
4045 | fqcodel_qdisc_set(struct netdev *netdev, const struct smap *details) | |
4046 | { | |
4047 | struct fqcodel fqcodel; | |
4048 | ||
4049 | fqcodel_parse_qdisc_details__(netdev, details, &fqcodel); | |
4050 | fqcodel_install__(netdev, fqcodel.target, fqcodel.limit, fqcodel.interval, | |
4051 | fqcodel.flows, fqcodel.quantum); | |
4052 | fqcodel_get__(netdev)->target = fqcodel.target; | |
4053 | fqcodel_get__(netdev)->limit = fqcodel.limit; | |
4054 | fqcodel_get__(netdev)->interval = fqcodel.interval; | |
4055 | fqcodel_get__(netdev)->flows = fqcodel.flows; | |
4056 | fqcodel_get__(netdev)->quantum = fqcodel.quantum; | |
4057 | return 0; | |
4058 | } | |
4059 | ||
4060 | static const struct tc_ops tc_ops_fqcodel = { | |
89c09c1c BP |
4061 | .linux_name = "fq_codel", |
4062 | .ovs_name = "linux-fq_codel", | |
4063 | .n_queues = FQCODEL_N_QUEUES, | |
4064 | .tc_install = fqcodel_tc_install, | |
4065 | .tc_load = fqcodel_tc_load, | |
4066 | .tc_destroy = fqcodel_tc_destroy, | |
4067 | .qdisc_get = fqcodel_qdisc_get, | |
4068 | .qdisc_set = fqcodel_qdisc_set, | |
677d9158 JV |
4069 | }; |
4070 | \f | |
4071 | /* SFQ traffic control class. */ | |
4072 | ||
4073 | #define SFQ_N_QUEUES 0x0000 | |
4074 | ||
4075 | struct sfq { | |
4076 | struct tc tc; | |
4077 | uint32_t quantum; | |
4078 | uint32_t perturb; | |
4079 | }; | |
4080 | ||
4081 | static struct sfq * | |
4082 | sfq_get__(const struct netdev *netdev_) | |
4083 | { | |
4084 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4085 | return CONTAINER_OF(netdev->tc, struct sfq, tc); | |
4086 | } | |
4087 | ||
4088 | static void | |
4089 | sfq_install__(struct netdev *netdev_, uint32_t quantum, uint32_t perturb) | |
4090 | { | |
4091 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4092 | struct sfq *sfq; | |
4093 | ||
4094 | sfq = xmalloc(sizeof *sfq); | |
4095 | tc_init(&sfq->tc, &tc_ops_sfq); | |
4096 | sfq->perturb = perturb; | |
4097 | sfq->quantum = quantum; | |
4098 | ||
4099 | netdev->tc = &sfq->tc; | |
4100 | } | |
4101 | ||
4102 | static int | |
4103 | sfq_setup_qdisc__(struct netdev *netdev, uint32_t quantum, uint32_t perturb) | |
4104 | { | |
4105 | struct tc_sfq_qopt opt; | |
4106 | struct ofpbuf request; | |
4107 | struct tcmsg *tcmsg; | |
4108 | int mtu; | |
4109 | int mtu_error, error; | |
4110 | mtu_error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); | |
4111 | ||
4112 | tc_del_qdisc(netdev); | |
4113 | ||
7874bdff RD |
4114 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
4115 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
677d9158 JV |
4116 | if (!tcmsg) { |
4117 | return ENODEV; | |
4118 | } | |
4119 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
4120 | tcmsg->tcm_parent = TC_H_ROOT; | |
4121 | ||
4122 | memset(&opt, 0, sizeof opt); | |
4123 | if (!quantum) { | |
4124 | if (!mtu_error) { | |
4125 | opt.quantum = mtu; /* if we cannot find mtu, use default */ | |
4126 | } | |
4127 | } else { | |
4128 | opt.quantum = quantum; | |
4129 | } | |
4130 | ||
4131 | if (!perturb) { | |
4132 | opt.perturb_period = 10; | |
4133 | } else { | |
4134 | opt.perturb_period = perturb; | |
4135 | } | |
4136 | ||
4137 | nl_msg_put_string(&request, TCA_KIND, "sfq"); | |
4138 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
4139 | ||
4140 | error = tc_transact(&request, NULL); | |
4141 | if (error) { | |
4142 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
4143 | "quantum %u, perturb %u error %d(%s)", | |
4144 | netdev_get_name(netdev), | |
4145 | opt.quantum, opt.perturb_period, | |
4146 | error, ovs_strerror(error)); | |
4147 | } | |
4148 | return error; | |
4149 | } | |
4150 | ||
4151 | static void | |
4152 | sfq_parse_qdisc_details__(struct netdev *netdev, | |
4153 | const struct smap *details, struct sfq *sfq) | |
4154 | { | |
13c1637f BP |
4155 | sfq->perturb = smap_get_ullong(details, "perturb", 0); |
4156 | sfq->quantum = smap_get_ullong(details, "quantum", 0); | |
677d9158 | 4157 | |
677d9158 JV |
4158 | if (!sfq->perturb) { |
4159 | sfq->perturb = 10; | |
4160 | } | |
4161 | ||
4162 | if (!sfq->quantum) { | |
13c1637f BP |
4163 | int mtu; |
4164 | if (!netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu)) { | |
677d9158 JV |
4165 | sfq->quantum = mtu; |
4166 | } else { | |
4167 | VLOG_WARN_RL(&rl, "when using SFQ, you must specify quantum on a " | |
4168 | "device without mtu"); | |
677d9158 JV |
4169 | } |
4170 | } | |
4171 | } | |
4172 | ||
4173 | static int | |
4174 | sfq_tc_install(struct netdev *netdev, const struct smap *details) | |
4175 | { | |
4176 | int error; | |
4177 | struct sfq sfq; | |
4178 | ||
4179 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
4180 | error = sfq_setup_qdisc__(netdev, sfq.quantum, sfq.perturb); | |
4181 | if (!error) { | |
4182 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
4183 | } | |
4184 | return error; | |
4185 | } | |
4186 | ||
4187 | static int | |
4188 | sfq_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
4189 | { | |
4190 | const struct tc_sfq_qopt *sfq; | |
4191 | struct nlattr *nlattr; | |
4192 | const char * kind; | |
4193 | int error; | |
4194 | ||
4195 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
4196 | if (error == 0) { | |
4197 | sfq = nl_attr_get(nlattr); | |
61265c03 | 4198 | sfq_install__(netdev, sfq->quantum, sfq->perturb_period); |
677d9158 JV |
4199 | return 0; |
4200 | } | |
4201 | ||
4202 | return error; | |
4203 | } | |
4204 | ||
4205 | static void | |
4206 | sfq_tc_destroy(struct tc *tc) | |
4207 | { | |
4208 | struct sfq *sfq = CONTAINER_OF(tc, struct sfq, tc); | |
4209 | tc_destroy(tc); | |
4210 | free(sfq); | |
4211 | } | |
4212 | ||
4213 | static int | |
4214 | sfq_qdisc_get(const struct netdev *netdev, struct smap *details) | |
4215 | { | |
4216 | const struct sfq *sfq = sfq_get__(netdev); | |
4217 | smap_add_format(details, "quantum", "%u", sfq->quantum); | |
4218 | smap_add_format(details, "perturb", "%u", sfq->perturb); | |
4219 | return 0; | |
4220 | } | |
4221 | ||
4222 | static int | |
4223 | sfq_qdisc_set(struct netdev *netdev, const struct smap *details) | |
4224 | { | |
4225 | struct sfq sfq; | |
4226 | ||
4227 | sfq_parse_qdisc_details__(netdev, details, &sfq); | |
4228 | sfq_install__(netdev, sfq.quantum, sfq.perturb); | |
4229 | sfq_get__(netdev)->quantum = sfq.quantum; | |
4230 | sfq_get__(netdev)->perturb = sfq.perturb; | |
4231 | return 0; | |
4232 | } | |
4233 | ||
4234 | static const struct tc_ops tc_ops_sfq = { | |
89c09c1c BP |
4235 | .linux_name = "sfq", |
4236 | .ovs_name = "linux-sfq", | |
4237 | .n_queues = SFQ_N_QUEUES, | |
4238 | .tc_install = sfq_tc_install, | |
4239 | .tc_load = sfq_tc_load, | |
4240 | .tc_destroy = sfq_tc_destroy, | |
4241 | .qdisc_get = sfq_qdisc_get, | |
4242 | .qdisc_set = sfq_qdisc_set, | |
677d9158 JV |
4243 | }; |
4244 | \f | |
2f564bb1 S |
4245 | /* netem traffic control class. */ |
4246 | ||
4247 | struct netem { | |
4248 | struct tc tc; | |
4249 | uint32_t latency; | |
4250 | uint32_t limit; | |
4251 | uint32_t loss; | |
4252 | }; | |
4253 | ||
4254 | static struct netem * | |
4255 | netem_get__(const struct netdev *netdev_) | |
4256 | { | |
4257 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4258 | return CONTAINER_OF(netdev->tc, struct netem, tc); | |
4259 | } | |
4260 | ||
4261 | static void | |
4262 | netem_install__(struct netdev *netdev_, uint32_t latency, | |
4263 | uint32_t limit, uint32_t loss) | |
4264 | { | |
4265 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
4266 | struct netem *netem; | |
4267 | ||
4268 | netem = xmalloc(sizeof *netem); | |
4269 | tc_init(&netem->tc, &tc_ops_netem); | |
4270 | netem->latency = latency; | |
4271 | netem->limit = limit; | |
4272 | netem->loss = loss; | |
4273 | ||
4274 | netdev->tc = &netem->tc; | |
4275 | } | |
4276 | ||
4277 | static int | |
4278 | netem_setup_qdisc__(struct netdev *netdev, uint32_t latency, | |
4279 | uint32_t limit, uint32_t loss) | |
4280 | { | |
4281 | struct tc_netem_qopt opt; | |
4282 | struct ofpbuf request; | |
4283 | struct tcmsg *tcmsg; | |
4284 | int error; | |
4285 | ||
4286 | tc_del_qdisc(netdev); | |
4287 | ||
4288 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, | |
4289 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
4290 | if (!tcmsg) { | |
4291 | return ENODEV; | |
4292 | } | |
4293 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
4294 | tcmsg->tcm_parent = TC_H_ROOT; | |
4295 | ||
4296 | memset(&opt, 0, sizeof opt); | |
4297 | ||
4298 | if (!limit) { | |
4299 | opt.limit = 1000; | |
4300 | } else { | |
4301 | opt.limit = limit; | |
4302 | } | |
4303 | ||
4304 | if (loss) { | |
4305 | if (loss > 100) { | |
4306 | VLOG_WARN_RL(&rl, | |
4307 | "loss should be a percentage value between 0 to 100, " | |
4308 | "loss was %u", loss); | |
4309 | return EINVAL; | |
4310 | } | |
4311 | opt.loss = floor(UINT32_MAX * (loss / 100.0)); | |
4312 | } | |
4313 | ||
4314 | opt.latency = tc_time_to_ticks(latency); | |
4315 | ||
4316 | nl_msg_put_string(&request, TCA_KIND, "netem"); | |
4317 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
4318 | ||
4319 | error = tc_transact(&request, NULL); | |
4320 | if (error) { | |
4321 | VLOG_WARN_RL(&rl, "failed to replace %s qdisc, " | |
4322 | "latency %u, limit %u, loss %u error %d(%s)", | |
4323 | netdev_get_name(netdev), | |
4324 | opt.latency, opt.limit, opt.loss, | |
4325 | error, ovs_strerror(error)); | |
4326 | } | |
4327 | return error; | |
4328 | } | |
4329 | ||
4330 | static void | |
4331 | netem_parse_qdisc_details__(struct netdev *netdev OVS_UNUSED, | |
4332 | const struct smap *details, struct netem *netem) | |
4333 | { | |
4334 | netem->latency = smap_get_ullong(details, "latency", 0); | |
4335 | netem->limit = smap_get_ullong(details, "limit", 0); | |
4336 | netem->loss = smap_get_ullong(details, "loss", 0); | |
4337 | ||
4338 | if (!netem->limit) { | |
4339 | netem->limit = 1000; | |
4340 | } | |
4341 | } | |
4342 | ||
4343 | static int | |
4344 | netem_tc_install(struct netdev *netdev, const struct smap *details) | |
4345 | { | |
4346 | int error; | |
4347 | struct netem netem; | |
4348 | ||
4349 | netem_parse_qdisc_details__(netdev, details, &netem); | |
4350 | error = netem_setup_qdisc__(netdev, netem.latency, | |
4351 | netem.limit, netem.loss); | |
4352 | if (!error) { | |
4353 | netem_install__(netdev, netem.latency, netem.limit, netem.loss); | |
4354 | } | |
4355 | return error; | |
4356 | } | |
4357 | ||
4358 | static int | |
4359 | netem_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg) | |
4360 | { | |
4361 | const struct tc_netem_qopt *netem; | |
4362 | struct nlattr *nlattr; | |
4363 | const char *kind; | |
4364 | int error; | |
4365 | ||
4366 | error = tc_parse_qdisc(nlmsg, &kind, &nlattr); | |
4367 | if (error == 0) { | |
4368 | netem = nl_attr_get(nlattr); | |
4369 | netem_install__(netdev, netem->latency, netem->limit, netem->loss); | |
4370 | return 0; | |
4371 | } | |
4372 | ||
4373 | return error; | |
4374 | } | |
4375 | ||
4376 | static void | |
4377 | netem_tc_destroy(struct tc *tc) | |
4378 | { | |
4379 | struct netem *netem = CONTAINER_OF(tc, struct netem, tc); | |
4380 | tc_destroy(tc); | |
4381 | free(netem); | |
4382 | } | |
4383 | ||
4384 | static int | |
4385 | netem_qdisc_get(const struct netdev *netdev, struct smap *details) | |
4386 | { | |
4387 | const struct netem *netem = netem_get__(netdev); | |
4388 | smap_add_format(details, "latency", "%u", netem->latency); | |
4389 | smap_add_format(details, "limit", "%u", netem->limit); | |
4390 | smap_add_format(details, "loss", "%u", netem->loss); | |
4391 | return 0; | |
4392 | } | |
4393 | ||
4394 | static int | |
4395 | netem_qdisc_set(struct netdev *netdev, const struct smap *details) | |
4396 | { | |
4397 | struct netem netem; | |
4398 | ||
4399 | netem_parse_qdisc_details__(netdev, details, &netem); | |
4400 | netem_install__(netdev, netem.latency, netem.limit, netem.loss); | |
4401 | netem_get__(netdev)->latency = netem.latency; | |
4402 | netem_get__(netdev)->limit = netem.limit; | |
4403 | netem_get__(netdev)->loss = netem.loss; | |
4404 | return 0; | |
4405 | } | |
4406 | ||
4407 | static const struct tc_ops tc_ops_netem = { | |
4408 | .linux_name = "netem", | |
4409 | .ovs_name = "linux-netem", | |
4410 | .n_queues = 0, | |
4411 | .tc_install = netem_tc_install, | |
4412 | .tc_load = netem_tc_load, | |
4413 | .tc_destroy = netem_tc_destroy, | |
4414 | .qdisc_get = netem_qdisc_get, | |
4415 | .qdisc_set = netem_qdisc_set, | |
4416 | }; | |
4417 | \f | |
c1c9c9c4 | 4418 | /* HTB traffic control class. */ |
559843ed | 4419 | |
c1c9c9c4 | 4420 | #define HTB_N_QUEUES 0xf000 |
4f631ccd | 4421 | #define HTB_RATE2QUANTUM 10 |
8b61709d | 4422 | |
c1c9c9c4 BP |
4423 | struct htb { |
4424 | struct tc tc; | |
4425 | unsigned int max_rate; /* In bytes/s. */ | |
4426 | }; | |
8b61709d | 4427 | |
c1c9c9c4 | 4428 | struct htb_class { |
93b13be8 | 4429 | struct tc_queue tc_queue; |
c1c9c9c4 BP |
4430 | unsigned int min_rate; /* In bytes/s. */ |
4431 | unsigned int max_rate; /* In bytes/s. */ | |
4432 | unsigned int burst; /* In bytes. */ | |
4433 | unsigned int priority; /* Lower values are higher priorities. */ | |
4434 | }; | |
8b61709d | 4435 | |
c1c9c9c4 | 4436 | static struct htb * |
b5d57fc8 | 4437 | htb_get__(const struct netdev *netdev_) |
c1c9c9c4 | 4438 | { |
b5d57fc8 BP |
4439 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4440 | return CONTAINER_OF(netdev->tc, struct htb, tc); | |
c1c9c9c4 BP |
4441 | } |
4442 | ||
24045e35 | 4443 | static void |
b5d57fc8 | 4444 | htb_install__(struct netdev *netdev_, uint64_t max_rate) |
c1c9c9c4 | 4445 | { |
b5d57fc8 | 4446 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
4447 | struct htb *htb; |
4448 | ||
4449 | htb = xmalloc(sizeof *htb); | |
4450 | tc_init(&htb->tc, &tc_ops_htb); | |
4451 | htb->max_rate = max_rate; | |
4452 | ||
b5d57fc8 | 4453 | netdev->tc = &htb->tc; |
c1c9c9c4 BP |
4454 | } |
4455 | ||
4456 | /* Create an HTB qdisc. | |
4457 | * | |
a339aa81 | 4458 | * Equivalent to "tc qdisc add dev <dev> root handle 1: htb default 1". */ |
c1c9c9c4 BP |
4459 | static int |
4460 | htb_setup_qdisc__(struct netdev *netdev) | |
4461 | { | |
4462 | size_t opt_offset; | |
4463 | struct tc_htb_glob opt; | |
4464 | struct ofpbuf request; | |
4465 | struct tcmsg *tcmsg; | |
4466 | ||
4467 | tc_del_qdisc(netdev); | |
4468 | ||
7874bdff RD |
4469 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
4470 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
23a98ffe BP |
4471 | if (!tcmsg) { |
4472 | return ENODEV; | |
4473 | } | |
c1c9c9c4 BP |
4474 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
4475 | tcmsg->tcm_parent = TC_H_ROOT; | |
4476 | ||
4477 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
4478 | ||
4479 | memset(&opt, 0, sizeof opt); | |
4f631ccd | 4480 | opt.rate2quantum = HTB_RATE2QUANTUM; |
c1c9c9c4 | 4481 | opt.version = 3; |
4ecf12d5 | 4482 | opt.defcls = 1; |
c1c9c9c4 BP |
4483 | |
4484 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4485 | nl_msg_put_unspec(&request, TCA_HTB_INIT, &opt, sizeof opt); | |
4486 | nl_msg_end_nested(&request, opt_offset); | |
4487 | ||
4488 | return tc_transact(&request, NULL); | |
4489 | } | |
4490 | ||
4491 | /* Equivalent to "tc class replace <dev> classid <handle> parent <parent> htb | |
4492 | * rate <min_rate>bps ceil <max_rate>bps burst <burst>b prio <priority>". */ | |
4493 | static int | |
4494 | htb_setup_class__(struct netdev *netdev, unsigned int handle, | |
4495 | unsigned int parent, struct htb_class *class) | |
4496 | { | |
4497 | size_t opt_offset; | |
4498 | struct tc_htb_opt opt; | |
4499 | struct ofpbuf request; | |
4500 | struct tcmsg *tcmsg; | |
4501 | int error; | |
4502 | int mtu; | |
4503 | ||
73371c09 | 4504 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 4505 | if (error) { |
f915f1a8 BP |
4506 | VLOG_WARN_RL(&rl, "cannot set up HTB on device %s that lacks MTU", |
4507 | netdev_get_name(netdev)); | |
9b020780 | 4508 | return error; |
f915f1a8 | 4509 | } |
c1c9c9c4 BP |
4510 | |
4511 | memset(&opt, 0, sizeof opt); | |
4512 | tc_fill_rate(&opt.rate, class->min_rate, mtu); | |
4513 | tc_fill_rate(&opt.ceil, class->max_rate, mtu); | |
4f631ccd AW |
4514 | /* Makes sure the quantum is at least MTU. Setting quantum will |
4515 | * make htb ignore the r2q for this class. */ | |
4516 | if ((class->min_rate / HTB_RATE2QUANTUM) < mtu) { | |
4517 | opt.quantum = mtu; | |
4518 | } | |
c1c9c9c4 BP |
4519 | opt.buffer = tc_calc_buffer(opt.rate.rate, mtu, class->burst); |
4520 | opt.cbuffer = tc_calc_buffer(opt.ceil.rate, mtu, class->burst); | |
4521 | opt.prio = class->priority; | |
4522 | ||
7874bdff RD |
4523 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
4524 | &request); | |
23a98ffe BP |
4525 | if (!tcmsg) { |
4526 | return ENODEV; | |
4527 | } | |
c1c9c9c4 BP |
4528 | tcmsg->tcm_handle = handle; |
4529 | tcmsg->tcm_parent = parent; | |
4530 | ||
4531 | nl_msg_put_string(&request, TCA_KIND, "htb"); | |
4532 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
4533 | nl_msg_put_unspec(&request, TCA_HTB_PARMS, &opt, sizeof opt); | |
4534 | tc_put_rtab(&request, TCA_HTB_RTAB, &opt.rate); | |
4535 | tc_put_rtab(&request, TCA_HTB_CTAB, &opt.ceil); | |
4536 | nl_msg_end_nested(&request, opt_offset); | |
4537 | ||
4538 | error = tc_transact(&request, NULL); | |
4539 | if (error) { | |
4540 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
4541 | "min_rate=%u max_rate=%u burst=%u prio=%u (%s)", | |
4542 | netdev_get_name(netdev), | |
4543 | tc_get_major(handle), tc_get_minor(handle), | |
4544 | tc_get_major(parent), tc_get_minor(parent), | |
4545 | class->min_rate, class->max_rate, | |
10a89ef0 | 4546 | class->burst, class->priority, ovs_strerror(error)); |
c1c9c9c4 BP |
4547 | } |
4548 | return error; | |
4549 | } | |
4550 | ||
4551 | /* Parses Netlink attributes in 'options' for HTB parameters and stores a | |
4552 | * description of them into 'details'. The description complies with the | |
4553 | * specification given in the vswitch database documentation for linux-htb | |
4554 | * queue details. */ | |
4555 | static int | |
4556 | htb_parse_tca_options__(struct nlattr *nl_options, struct htb_class *class) | |
4557 | { | |
4558 | static const struct nl_policy tca_htb_policy[] = { | |
4559 | [TCA_HTB_PARMS] = { .type = NL_A_UNSPEC, .optional = false, | |
4560 | .min_len = sizeof(struct tc_htb_opt) }, | |
4561 | }; | |
4562 | ||
4563 | struct nlattr *attrs[ARRAY_SIZE(tca_htb_policy)]; | |
4564 | const struct tc_htb_opt *htb; | |
4565 | ||
4566 | if (!nl_parse_nested(nl_options, tca_htb_policy, | |
4567 | attrs, ARRAY_SIZE(tca_htb_policy))) { | |
4568 | VLOG_WARN_RL(&rl, "failed to parse HTB class options"); | |
4569 | return EPROTO; | |
4570 | } | |
4571 | ||
4572 | htb = nl_attr_get(attrs[TCA_HTB_PARMS]); | |
4573 | class->min_rate = htb->rate.rate; | |
4574 | class->max_rate = htb->ceil.rate; | |
4575 | class->burst = tc_ticks_to_bytes(htb->rate.rate, htb->buffer); | |
4576 | class->priority = htb->prio; | |
4577 | return 0; | |
4578 | } | |
4579 | ||
4580 | static int | |
4581 | htb_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
4582 | struct htb_class *options, | |
4583 | struct netdev_queue_stats *stats) | |
4584 | { | |
4585 | struct nlattr *nl_options; | |
4586 | unsigned int handle; | |
4587 | int error; | |
4588 | ||
4589 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
4590 | if (!error && queue_id) { | |
17ee3c1f BP |
4591 | unsigned int major = tc_get_major(handle); |
4592 | unsigned int minor = tc_get_minor(handle); | |
4593 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
4594 | *queue_id = minor - 1; | |
c1c9c9c4 BP |
4595 | } else { |
4596 | error = EPROTO; | |
4597 | } | |
4598 | } | |
4599 | if (!error && options) { | |
4600 | error = htb_parse_tca_options__(nl_options, options); | |
4601 | } | |
4602 | return error; | |
4603 | } | |
4604 | ||
4605 | static void | |
73371c09 | 4606 | htb_parse_qdisc_details__(struct netdev *netdev_, |
79f1cbe9 | 4607 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 | 4608 | { |
73371c09 | 4609 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 | 4610 | |
13c1637f | 4611 | hc->max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
c1c9c9c4 | 4612 | if (!hc->max_rate) { |
a00ca915 | 4613 | enum netdev_features current; |
c1c9c9c4 | 4614 | |
73371c09 BP |
4615 | netdev_linux_read_features(netdev); |
4616 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 4617 | hc->max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
c1c9c9c4 BP |
4618 | } |
4619 | hc->min_rate = hc->max_rate; | |
4620 | hc->burst = 0; | |
4621 | hc->priority = 0; | |
4622 | } | |
4623 | ||
4624 | static int | |
4625 | htb_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 4626 | const struct smap *details, struct htb_class *hc) |
c1c9c9c4 BP |
4627 | { |
4628 | const struct htb *htb = htb_get__(netdev); | |
9b020780 | 4629 | int mtu, error; |
214117fd | 4630 | unsigned long long int max_rate_bit; |
c1c9c9c4 | 4631 | |
73371c09 | 4632 | error = netdev_linux_get_mtu__(netdev_linux_cast(netdev), &mtu); |
9b020780 | 4633 | if (error) { |
f915f1a8 BP |
4634 | VLOG_WARN_RL(&rl, "cannot parse HTB class on device %s that lacks MTU", |
4635 | netdev_get_name(netdev)); | |
9b020780 | 4636 | return error; |
f915f1a8 BP |
4637 | } |
4638 | ||
4f104611 EJ |
4639 | /* HTB requires at least an mtu sized min-rate to send any traffic even |
4640 | * on uncongested links. */ | |
13c1637f | 4641 | hc->min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
4f104611 | 4642 | hc->min_rate = MAX(hc->min_rate, mtu); |
c1c9c9c4 BP |
4643 | hc->min_rate = MIN(hc->min_rate, htb->max_rate); |
4644 | ||
4645 | /* max-rate */ | |
214117fd KF |
4646 | max_rate_bit = smap_get_ullong(details, "max-rate", 0); |
4647 | hc->max_rate = max_rate_bit ? max_rate_bit / 8 : htb->max_rate; | |
c1c9c9c4 BP |
4648 | hc->max_rate = MAX(hc->max_rate, hc->min_rate); |
4649 | hc->max_rate = MIN(hc->max_rate, htb->max_rate); | |
4650 | ||
4651 | /* burst | |
4652 | * | |
4653 | * According to hints in the documentation that I've read, it is important | |
4654 | * that 'burst' be at least as big as the largest frame that might be | |
4655 | * transmitted. Also, making 'burst' a bit bigger than necessary is OK, | |
4656 | * but having it a bit too small is a problem. Since netdev_get_mtu() | |
4657 | * doesn't include the Ethernet header, we need to add at least 14 (18?) to | |
4658 | * the MTU. We actually add 64, instead of 14, as a guard against | |
4659 | * additional headers get tacked on somewhere that we're not aware of. */ | |
13c1637f | 4660 | hc->burst = smap_get_ullong(details, "burst", 0) / 8; |
c1c9c9c4 BP |
4661 | hc->burst = MAX(hc->burst, mtu + 64); |
4662 | ||
4663 | /* priority */ | |
13c1637f | 4664 | hc->priority = smap_get_ullong(details, "priority", 0); |
c1c9c9c4 BP |
4665 | |
4666 | return 0; | |
4667 | } | |
4668 | ||
4669 | static int | |
4670 | htb_query_class__(const struct netdev *netdev, unsigned int handle, | |
4671 | unsigned int parent, struct htb_class *options, | |
4672 | struct netdev_queue_stats *stats) | |
4673 | { | |
4674 | struct ofpbuf *reply; | |
4675 | int error; | |
4676 | ||
4677 | error = tc_query_class(netdev, handle, parent, &reply); | |
4678 | if (!error) { | |
4679 | error = htb_parse_tcmsg__(reply, NULL, options, stats); | |
4680 | ofpbuf_delete(reply); | |
4681 | } | |
4682 | return error; | |
4683 | } | |
4684 | ||
4685 | static int | |
79f1cbe9 | 4686 | htb_tc_install(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
4687 | { |
4688 | int error; | |
4689 | ||
4690 | error = htb_setup_qdisc__(netdev); | |
4691 | if (!error) { | |
4692 | struct htb_class hc; | |
4693 | ||
4694 | htb_parse_qdisc_details__(netdev, details, &hc); | |
4695 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4696 | tc_make_handle(1, 0), &hc); | |
4697 | if (!error) { | |
4698 | htb_install__(netdev, hc.max_rate); | |
4699 | } | |
4700 | } | |
4701 | return error; | |
4702 | } | |
4703 | ||
93b13be8 BP |
4704 | static struct htb_class * |
4705 | htb_class_cast__(const struct tc_queue *queue) | |
4706 | { | |
4707 | return CONTAINER_OF(queue, struct htb_class, tc_queue); | |
4708 | } | |
4709 | ||
c1c9c9c4 BP |
4710 | static void |
4711 | htb_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4712 | const struct htb_class *hc) | |
4713 | { | |
4714 | struct htb *htb = htb_get__(netdev); | |
93b13be8 BP |
4715 | size_t hash = hash_int(queue_id, 0); |
4716 | struct tc_queue *queue; | |
c1c9c9c4 BP |
4717 | struct htb_class *hcp; |
4718 | ||
93b13be8 BP |
4719 | queue = tc_find_queue__(netdev, queue_id, hash); |
4720 | if (queue) { | |
4721 | hcp = htb_class_cast__(queue); | |
4722 | } else { | |
c1c9c9c4 | 4723 | hcp = xmalloc(sizeof *hcp); |
93b13be8 BP |
4724 | queue = &hcp->tc_queue; |
4725 | queue->queue_id = queue_id; | |
6dc34a0d | 4726 | queue->created = time_msec(); |
93b13be8 | 4727 | hmap_insert(&htb->tc.queues, &queue->hmap_node, hash); |
c1c9c9c4 | 4728 | } |
93b13be8 BP |
4729 | |
4730 | hcp->min_rate = hc->min_rate; | |
4731 | hcp->max_rate = hc->max_rate; | |
4732 | hcp->burst = hc->burst; | |
4733 | hcp->priority = hc->priority; | |
c1c9c9c4 BP |
4734 | } |
4735 | ||
4736 | static int | |
4737 | htb_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
4738 | { | |
c1c9c9c4 | 4739 | struct ofpbuf msg; |
d57695d7 | 4740 | struct queue_dump_state state; |
c1c9c9c4 | 4741 | struct htb_class hc; |
c1c9c9c4 BP |
4742 | |
4743 | /* Get qdisc options. */ | |
4744 | hc.max_rate = 0; | |
4745 | htb_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 4746 | htb_install__(netdev, hc.max_rate); |
c1c9c9c4 BP |
4747 | |
4748 | /* Get queues. */ | |
d57695d7 | 4749 | if (!start_queue_dump(netdev, &state)) { |
23a98ffe BP |
4750 | return ENODEV; |
4751 | } | |
d57695d7 | 4752 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
c1c9c9c4 BP |
4753 | unsigned int queue_id; |
4754 | ||
4755 | if (!htb_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
4756 | htb_update_queue__(netdev, queue_id, &hc); | |
4757 | } | |
4758 | } | |
d57695d7 | 4759 | finish_queue_dump(&state); |
c1c9c9c4 BP |
4760 | |
4761 | return 0; | |
4762 | } | |
4763 | ||
4764 | static void | |
4765 | htb_tc_destroy(struct tc *tc) | |
4766 | { | |
4767 | struct htb *htb = CONTAINER_OF(tc, struct htb, tc); | |
4ec3d7c7 | 4768 | struct htb_class *hc; |
c1c9c9c4 | 4769 | |
4ec3d7c7 | 4770 | HMAP_FOR_EACH_POP (hc, tc_queue.hmap_node, &htb->tc.queues) { |
c1c9c9c4 BP |
4771 | free(hc); |
4772 | } | |
4773 | tc_destroy(tc); | |
4774 | free(htb); | |
4775 | } | |
4776 | ||
4777 | static int | |
79f1cbe9 | 4778 | htb_qdisc_get(const struct netdev *netdev, struct smap *details) |
c1c9c9c4 BP |
4779 | { |
4780 | const struct htb *htb = htb_get__(netdev); | |
79f1cbe9 | 4781 | smap_add_format(details, "max-rate", "%llu", 8ULL * htb->max_rate); |
c1c9c9c4 BP |
4782 | return 0; |
4783 | } | |
4784 | ||
4785 | static int | |
79f1cbe9 | 4786 | htb_qdisc_set(struct netdev *netdev, const struct smap *details) |
c1c9c9c4 BP |
4787 | { |
4788 | struct htb_class hc; | |
4789 | int error; | |
4790 | ||
4791 | htb_parse_qdisc_details__(netdev, details, &hc); | |
4792 | error = htb_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
4793 | tc_make_handle(1, 0), &hc); | |
4794 | if (!error) { | |
4795 | htb_get__(netdev)->max_rate = hc.max_rate; | |
4796 | } | |
4797 | return error; | |
4798 | } | |
4799 | ||
4800 | static int | |
93b13be8 | 4801 | htb_class_get(const struct netdev *netdev OVS_UNUSED, |
79f1cbe9 | 4802 | const struct tc_queue *queue, struct smap *details) |
c1c9c9c4 | 4803 | { |
93b13be8 | 4804 | const struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4805 | |
79f1cbe9 | 4806 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
c1c9c9c4 | 4807 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 4808 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
c1c9c9c4 | 4809 | } |
79f1cbe9 | 4810 | smap_add_format(details, "burst", "%llu", 8ULL * hc->burst); |
c1c9c9c4 | 4811 | if (hc->priority) { |
79f1cbe9 | 4812 | smap_add_format(details, "priority", "%u", hc->priority); |
c1c9c9c4 BP |
4813 | } |
4814 | return 0; | |
4815 | } | |
4816 | ||
4817 | static int | |
4818 | htb_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 4819 | const struct smap *details) |
c1c9c9c4 BP |
4820 | { |
4821 | struct htb_class hc; | |
4822 | int error; | |
4823 | ||
4824 | error = htb_parse_class_details__(netdev, details, &hc); | |
4825 | if (error) { | |
4826 | return error; | |
4827 | } | |
4828 | ||
17ee3c1f | 4829 | error = htb_setup_class__(netdev, tc_make_handle(1, queue_id + 1), |
c1c9c9c4 BP |
4830 | tc_make_handle(1, 0xfffe), &hc); |
4831 | if (error) { | |
4832 | return error; | |
4833 | } | |
4834 | ||
4835 | htb_update_queue__(netdev, queue_id, &hc); | |
4836 | return 0; | |
4837 | } | |
4838 | ||
4839 | static int | |
93b13be8 | 4840 | htb_class_delete(struct netdev *netdev, struct tc_queue *queue) |
c1c9c9c4 | 4841 | { |
93b13be8 | 4842 | struct htb_class *hc = htb_class_cast__(queue); |
c1c9c9c4 | 4843 | struct htb *htb = htb_get__(netdev); |
c1c9c9c4 BP |
4844 | int error; |
4845 | ||
93b13be8 | 4846 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); |
c1c9c9c4 | 4847 | if (!error) { |
93b13be8 | 4848 | hmap_remove(&htb->tc.queues, &hc->tc_queue.hmap_node); |
c1c9c9c4 | 4849 | free(hc); |
c1c9c9c4 BP |
4850 | } |
4851 | return error; | |
4852 | } | |
4853 | ||
4854 | static int | |
93b13be8 | 4855 | htb_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, |
c1c9c9c4 BP |
4856 | struct netdev_queue_stats *stats) |
4857 | { | |
93b13be8 | 4858 | return htb_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), |
c1c9c9c4 BP |
4859 | tc_make_handle(1, 0xfffe), NULL, stats); |
4860 | } | |
4861 | ||
4862 | static int | |
4863 | htb_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
4864 | const struct ofpbuf *nlmsg, | |
4865 | netdev_dump_queue_stats_cb *cb, void *aux) | |
4866 | { | |
4867 | struct netdev_queue_stats stats; | |
17ee3c1f | 4868 | unsigned int handle, major, minor; |
c1c9c9c4 BP |
4869 | int error; |
4870 | ||
4871 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
4872 | if (error) { | |
4873 | return error; | |
4874 | } | |
4875 | ||
17ee3c1f BP |
4876 | major = tc_get_major(handle); |
4877 | minor = tc_get_minor(handle); | |
4878 | if (major == 1 && minor > 0 && minor <= HTB_N_QUEUES) { | |
d5590e7e | 4879 | (*cb)(minor - 1, &stats, aux); |
c1c9c9c4 BP |
4880 | } |
4881 | return 0; | |
4882 | } | |
4883 | ||
4884 | static const struct tc_ops tc_ops_htb = { | |
89c09c1c BP |
4885 | .linux_name = "htb", |
4886 | .ovs_name = "linux-htb", | |
4887 | .n_queues = HTB_N_QUEUES, | |
4888 | .tc_install = htb_tc_install, | |
4889 | .tc_load = htb_tc_load, | |
4890 | .tc_destroy = htb_tc_destroy, | |
4891 | .qdisc_get = htb_qdisc_get, | |
4892 | .qdisc_set = htb_qdisc_set, | |
4893 | .class_get = htb_class_get, | |
4894 | .class_set = htb_class_set, | |
4895 | .class_delete = htb_class_delete, | |
4896 | .class_get_stats = htb_class_get_stats, | |
4897 | .class_dump_stats = htb_class_dump_stats | |
c1c9c9c4 BP |
4898 | }; |
4899 | \f | |
a339aa81 EJ |
4900 | /* "linux-hfsc" traffic control class. */ |
4901 | ||
4902 | #define HFSC_N_QUEUES 0xf000 | |
4903 | ||
4904 | struct hfsc { | |
4905 | struct tc tc; | |
4906 | uint32_t max_rate; | |
4907 | }; | |
4908 | ||
4909 | struct hfsc_class { | |
4910 | struct tc_queue tc_queue; | |
4911 | uint32_t min_rate; | |
4912 | uint32_t max_rate; | |
4913 | }; | |
4914 | ||
4915 | static struct hfsc * | |
b5d57fc8 | 4916 | hfsc_get__(const struct netdev *netdev_) |
a339aa81 | 4917 | { |
b5d57fc8 BP |
4918 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
4919 | return CONTAINER_OF(netdev->tc, struct hfsc, tc); | |
a339aa81 EJ |
4920 | } |
4921 | ||
4922 | static struct hfsc_class * | |
4923 | hfsc_class_cast__(const struct tc_queue *queue) | |
4924 | { | |
4925 | return CONTAINER_OF(queue, struct hfsc_class, tc_queue); | |
4926 | } | |
4927 | ||
24045e35 | 4928 | static void |
b5d57fc8 | 4929 | hfsc_install__(struct netdev *netdev_, uint32_t max_rate) |
a339aa81 | 4930 | { |
b5d57fc8 | 4931 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 EJ |
4932 | struct hfsc *hfsc; |
4933 | ||
a339aa81 EJ |
4934 | hfsc = xmalloc(sizeof *hfsc); |
4935 | tc_init(&hfsc->tc, &tc_ops_hfsc); | |
4936 | hfsc->max_rate = max_rate; | |
b5d57fc8 | 4937 | netdev->tc = &hfsc->tc; |
a339aa81 EJ |
4938 | } |
4939 | ||
4940 | static void | |
4941 | hfsc_update_queue__(struct netdev *netdev, unsigned int queue_id, | |
4942 | const struct hfsc_class *hc) | |
4943 | { | |
4944 | size_t hash; | |
4945 | struct hfsc *hfsc; | |
4946 | struct hfsc_class *hcp; | |
4947 | struct tc_queue *queue; | |
4948 | ||
4949 | hfsc = hfsc_get__(netdev); | |
4950 | hash = hash_int(queue_id, 0); | |
4951 | ||
4952 | queue = tc_find_queue__(netdev, queue_id, hash); | |
4953 | if (queue) { | |
4954 | hcp = hfsc_class_cast__(queue); | |
4955 | } else { | |
4956 | hcp = xmalloc(sizeof *hcp); | |
4957 | queue = &hcp->tc_queue; | |
4958 | queue->queue_id = queue_id; | |
6dc34a0d | 4959 | queue->created = time_msec(); |
a339aa81 EJ |
4960 | hmap_insert(&hfsc->tc.queues, &queue->hmap_node, hash); |
4961 | } | |
4962 | ||
4963 | hcp->min_rate = hc->min_rate; | |
4964 | hcp->max_rate = hc->max_rate; | |
4965 | } | |
4966 | ||
4967 | static int | |
4968 | hfsc_parse_tca_options__(struct nlattr *nl_options, struct hfsc_class *class) | |
4969 | { | |
4970 | const struct tc_service_curve *rsc, *fsc, *usc; | |
4971 | static const struct nl_policy tca_hfsc_policy[] = { | |
4972 | [TCA_HFSC_RSC] = { | |
4973 | .type = NL_A_UNSPEC, | |
4974 | .optional = false, | |
4975 | .min_len = sizeof(struct tc_service_curve), | |
4976 | }, | |
4977 | [TCA_HFSC_FSC] = { | |
4978 | .type = NL_A_UNSPEC, | |
4979 | .optional = false, | |
4980 | .min_len = sizeof(struct tc_service_curve), | |
4981 | }, | |
4982 | [TCA_HFSC_USC] = { | |
4983 | .type = NL_A_UNSPEC, | |
4984 | .optional = false, | |
4985 | .min_len = sizeof(struct tc_service_curve), | |
4986 | }, | |
4987 | }; | |
4988 | struct nlattr *attrs[ARRAY_SIZE(tca_hfsc_policy)]; | |
4989 | ||
4990 | if (!nl_parse_nested(nl_options, tca_hfsc_policy, | |
4991 | attrs, ARRAY_SIZE(tca_hfsc_policy))) { | |
4992 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options"); | |
4993 | return EPROTO; | |
4994 | } | |
4995 | ||
4996 | rsc = nl_attr_get(attrs[TCA_HFSC_RSC]); | |
4997 | fsc = nl_attr_get(attrs[TCA_HFSC_FSC]); | |
4998 | usc = nl_attr_get(attrs[TCA_HFSC_USC]); | |
4999 | ||
5000 | if (rsc->m1 != 0 || rsc->d != 0 || | |
5001 | fsc->m1 != 0 || fsc->d != 0 || | |
5002 | usc->m1 != 0 || usc->d != 0) { | |
5003 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
5004 | "Non-linear service curves are not supported."); | |
5005 | return EPROTO; | |
5006 | } | |
5007 | ||
5008 | if (rsc->m2 != fsc->m2) { | |
5009 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
5010 | "Real-time service curves are not supported "); | |
5011 | return EPROTO; | |
5012 | } | |
5013 | ||
5014 | if (rsc->m2 > usc->m2) { | |
5015 | VLOG_WARN_RL(&rl, "failed to parse HFSC class options. " | |
5016 | "Min-rate service curve is greater than " | |
5017 | "the max-rate service curve."); | |
5018 | return EPROTO; | |
5019 | } | |
5020 | ||
5021 | class->min_rate = fsc->m2; | |
5022 | class->max_rate = usc->m2; | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | static int | |
5027 | hfsc_parse_tcmsg__(struct ofpbuf *tcmsg, unsigned int *queue_id, | |
5028 | struct hfsc_class *options, | |
5029 | struct netdev_queue_stats *stats) | |
5030 | { | |
5031 | int error; | |
5032 | unsigned int handle; | |
5033 | struct nlattr *nl_options; | |
5034 | ||
5035 | error = tc_parse_class(tcmsg, &handle, &nl_options, stats); | |
5036 | if (error) { | |
5037 | return error; | |
5038 | } | |
5039 | ||
5040 | if (queue_id) { | |
5041 | unsigned int major, minor; | |
5042 | ||
5043 | major = tc_get_major(handle); | |
5044 | minor = tc_get_minor(handle); | |
5045 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
5046 | *queue_id = minor - 1; | |
5047 | } else { | |
5048 | return EPROTO; | |
5049 | } | |
5050 | } | |
5051 | ||
5052 | if (options) { | |
5053 | error = hfsc_parse_tca_options__(nl_options, options); | |
5054 | } | |
5055 | ||
5056 | return error; | |
5057 | } | |
5058 | ||
5059 | static int | |
5060 | hfsc_query_class__(const struct netdev *netdev, unsigned int handle, | |
5061 | unsigned int parent, struct hfsc_class *options, | |
5062 | struct netdev_queue_stats *stats) | |
5063 | { | |
5064 | int error; | |
5065 | struct ofpbuf *reply; | |
5066 | ||
5067 | error = tc_query_class(netdev, handle, parent, &reply); | |
5068 | if (error) { | |
5069 | return error; | |
5070 | } | |
5071 | ||
5072 | error = hfsc_parse_tcmsg__(reply, NULL, options, stats); | |
5073 | ofpbuf_delete(reply); | |
5074 | return error; | |
5075 | } | |
5076 | ||
5077 | static void | |
73371c09 | 5078 | hfsc_parse_qdisc_details__(struct netdev *netdev_, const struct smap *details, |
a339aa81 EJ |
5079 | struct hfsc_class *class) |
5080 | { | |
73371c09 | 5081 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
a339aa81 | 5082 | |
13c1637f | 5083 | uint32_t max_rate = smap_get_ullong(details, "max-rate", 0) / 8; |
a339aa81 | 5084 | if (!max_rate) { |
a00ca915 | 5085 | enum netdev_features current; |
a339aa81 | 5086 | |
73371c09 BP |
5087 | netdev_linux_read_features(netdev); |
5088 | current = !netdev->get_features_error ? netdev->current : 0; | |
d02a5f8e | 5089 | max_rate = netdev_features_to_bps(current, 100 * 1000 * 1000) / 8; |
a339aa81 EJ |
5090 | } |
5091 | ||
5092 | class->min_rate = max_rate; | |
5093 | class->max_rate = max_rate; | |
5094 | } | |
5095 | ||
5096 | static int | |
5097 | hfsc_parse_class_details__(struct netdev *netdev, | |
79f1cbe9 | 5098 | const struct smap *details, |
a339aa81 EJ |
5099 | struct hfsc_class * class) |
5100 | { | |
5101 | const struct hfsc *hfsc; | |
5102 | uint32_t min_rate, max_rate; | |
a339aa81 EJ |
5103 | |
5104 | hfsc = hfsc_get__(netdev); | |
a339aa81 | 5105 | |
13c1637f | 5106 | min_rate = smap_get_ullong(details, "min-rate", 0) / 8; |
79398bad | 5107 | min_rate = MAX(min_rate, 1); |
a339aa81 EJ |
5108 | min_rate = MIN(min_rate, hfsc->max_rate); |
5109 | ||
13c1637f | 5110 | max_rate = smap_get_ullong(details, "max-rate", hfsc->max_rate * 8) / 8; |
a339aa81 EJ |
5111 | max_rate = MAX(max_rate, min_rate); |
5112 | max_rate = MIN(max_rate, hfsc->max_rate); | |
5113 | ||
5114 | class->min_rate = min_rate; | |
5115 | class->max_rate = max_rate; | |
5116 | ||
5117 | return 0; | |
5118 | } | |
5119 | ||
5120 | /* Create an HFSC qdisc. | |
5121 | * | |
5122 | * Equivalent to "tc qdisc add dev <dev> root handle 1: hfsc default 1". */ | |
5123 | static int | |
5124 | hfsc_setup_qdisc__(struct netdev * netdev) | |
5125 | { | |
5126 | struct tcmsg *tcmsg; | |
5127 | struct ofpbuf request; | |
5128 | struct tc_hfsc_qopt opt; | |
5129 | ||
5130 | tc_del_qdisc(netdev); | |
5131 | ||
7874bdff RD |
5132 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWQDISC, |
5133 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
a339aa81 EJ |
5134 | |
5135 | if (!tcmsg) { | |
5136 | return ENODEV; | |
5137 | } | |
5138 | ||
5139 | tcmsg->tcm_handle = tc_make_handle(1, 0); | |
5140 | tcmsg->tcm_parent = TC_H_ROOT; | |
5141 | ||
5142 | memset(&opt, 0, sizeof opt); | |
5143 | opt.defcls = 1; | |
5144 | ||
5145 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
5146 | nl_msg_put_unspec(&request, TCA_OPTIONS, &opt, sizeof opt); | |
5147 | ||
5148 | return tc_transact(&request, NULL); | |
5149 | } | |
5150 | ||
5151 | /* Create an HFSC class. | |
5152 | * | |
5153 | * Equivalent to "tc class add <dev> parent <parent> classid <handle> hfsc | |
5154 | * sc rate <min_rate> ul rate <max_rate>" */ | |
5155 | static int | |
5156 | hfsc_setup_class__(struct netdev *netdev, unsigned int handle, | |
5157 | unsigned int parent, struct hfsc_class *class) | |
5158 | { | |
5159 | int error; | |
5160 | size_t opt_offset; | |
5161 | struct tcmsg *tcmsg; | |
5162 | struct ofpbuf request; | |
5163 | struct tc_service_curve min, max; | |
5164 | ||
7874bdff RD |
5165 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTCLASS, NLM_F_CREATE, |
5166 | &request); | |
a339aa81 EJ |
5167 | |
5168 | if (!tcmsg) { | |
5169 | return ENODEV; | |
5170 | } | |
5171 | ||
5172 | tcmsg->tcm_handle = handle; | |
5173 | tcmsg->tcm_parent = parent; | |
5174 | ||
5175 | min.m1 = 0; | |
5176 | min.d = 0; | |
5177 | min.m2 = class->min_rate; | |
5178 | ||
5179 | max.m1 = 0; | |
5180 | max.d = 0; | |
5181 | max.m2 = class->max_rate; | |
5182 | ||
5183 | nl_msg_put_string(&request, TCA_KIND, "hfsc"); | |
5184 | opt_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
5185 | nl_msg_put_unspec(&request, TCA_HFSC_RSC, &min, sizeof min); | |
5186 | nl_msg_put_unspec(&request, TCA_HFSC_FSC, &min, sizeof min); | |
5187 | nl_msg_put_unspec(&request, TCA_HFSC_USC, &max, sizeof max); | |
5188 | nl_msg_end_nested(&request, opt_offset); | |
5189 | ||
5190 | error = tc_transact(&request, NULL); | |
5191 | if (error) { | |
5192 | VLOG_WARN_RL(&rl, "failed to replace %s class %u:%u, parent %u:%u, " | |
5193 | "min-rate %ubps, max-rate %ubps (%s)", | |
5194 | netdev_get_name(netdev), | |
5195 | tc_get_major(handle), tc_get_minor(handle), | |
5196 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 5197 | class->min_rate, class->max_rate, ovs_strerror(error)); |
a339aa81 EJ |
5198 | } |
5199 | ||
5200 | return error; | |
5201 | } | |
5202 | ||
5203 | static int | |
79f1cbe9 | 5204 | hfsc_tc_install(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
5205 | { |
5206 | int error; | |
5207 | struct hfsc_class class; | |
5208 | ||
5209 | error = hfsc_setup_qdisc__(netdev); | |
5210 | ||
5211 | if (error) { | |
5212 | return error; | |
5213 | } | |
5214 | ||
5215 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
5216 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
5217 | tc_make_handle(1, 0), &class); | |
5218 | ||
5219 | if (error) { | |
5220 | return error; | |
5221 | } | |
5222 | ||
5223 | hfsc_install__(netdev, class.max_rate); | |
5224 | return 0; | |
5225 | } | |
5226 | ||
5227 | static int | |
5228 | hfsc_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
5229 | { | |
5230 | struct ofpbuf msg; | |
d57695d7 | 5231 | struct queue_dump_state state; |
a339aa81 EJ |
5232 | struct hfsc_class hc; |
5233 | ||
5234 | hc.max_rate = 0; | |
5235 | hfsc_query_class__(netdev, tc_make_handle(1, 0xfffe), 0, &hc, NULL); | |
24045e35 | 5236 | hfsc_install__(netdev, hc.max_rate); |
a339aa81 | 5237 | |
d57695d7 | 5238 | if (!start_queue_dump(netdev, &state)) { |
a339aa81 EJ |
5239 | return ENODEV; |
5240 | } | |
5241 | ||
d57695d7 | 5242 | while (nl_dump_next(&state.dump, &msg, &state.buf)) { |
a339aa81 EJ |
5243 | unsigned int queue_id; |
5244 | ||
5245 | if (!hfsc_parse_tcmsg__(&msg, &queue_id, &hc, NULL)) { | |
5246 | hfsc_update_queue__(netdev, queue_id, &hc); | |
5247 | } | |
5248 | } | |
5249 | ||
d57695d7 | 5250 | finish_queue_dump(&state); |
a339aa81 EJ |
5251 | return 0; |
5252 | } | |
5253 | ||
5254 | static void | |
5255 | hfsc_tc_destroy(struct tc *tc) | |
5256 | { | |
5257 | struct hfsc *hfsc; | |
5258 | struct hfsc_class *hc, *next; | |
5259 | ||
5260 | hfsc = CONTAINER_OF(tc, struct hfsc, tc); | |
5261 | ||
5262 | HMAP_FOR_EACH_SAFE (hc, next, tc_queue.hmap_node, &hfsc->tc.queues) { | |
5263 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
5264 | free(hc); | |
5265 | } | |
5266 | ||
5267 | tc_destroy(tc); | |
5268 | free(hfsc); | |
5269 | } | |
5270 | ||
5271 | static int | |
79f1cbe9 | 5272 | hfsc_qdisc_get(const struct netdev *netdev, struct smap *details) |
a339aa81 EJ |
5273 | { |
5274 | const struct hfsc *hfsc; | |
5275 | hfsc = hfsc_get__(netdev); | |
79f1cbe9 | 5276 | smap_add_format(details, "max-rate", "%llu", 8ULL * hfsc->max_rate); |
a339aa81 EJ |
5277 | return 0; |
5278 | } | |
5279 | ||
5280 | static int | |
79f1cbe9 | 5281 | hfsc_qdisc_set(struct netdev *netdev, const struct smap *details) |
a339aa81 EJ |
5282 | { |
5283 | int error; | |
5284 | struct hfsc_class class; | |
5285 | ||
5286 | hfsc_parse_qdisc_details__(netdev, details, &class); | |
5287 | error = hfsc_setup_class__(netdev, tc_make_handle(1, 0xfffe), | |
5288 | tc_make_handle(1, 0), &class); | |
5289 | ||
5290 | if (!error) { | |
5291 | hfsc_get__(netdev)->max_rate = class.max_rate; | |
5292 | } | |
5293 | ||
5294 | return error; | |
5295 | } | |
5296 | ||
5297 | static int | |
5298 | hfsc_class_get(const struct netdev *netdev OVS_UNUSED, | |
79f1cbe9 | 5299 | const struct tc_queue *queue, struct smap *details) |
a339aa81 EJ |
5300 | { |
5301 | const struct hfsc_class *hc; | |
5302 | ||
5303 | hc = hfsc_class_cast__(queue); | |
79f1cbe9 | 5304 | smap_add_format(details, "min-rate", "%llu", 8ULL * hc->min_rate); |
a339aa81 | 5305 | if (hc->min_rate != hc->max_rate) { |
79f1cbe9 | 5306 | smap_add_format(details, "max-rate", "%llu", 8ULL * hc->max_rate); |
a339aa81 EJ |
5307 | } |
5308 | return 0; | |
5309 | } | |
5310 | ||
5311 | static int | |
5312 | hfsc_class_set(struct netdev *netdev, unsigned int queue_id, | |
79f1cbe9 | 5313 | const struct smap *details) |
a339aa81 EJ |
5314 | { |
5315 | int error; | |
5316 | struct hfsc_class class; | |
5317 | ||
5318 | error = hfsc_parse_class_details__(netdev, details, &class); | |
5319 | if (error) { | |
5320 | return error; | |
5321 | } | |
5322 | ||
5323 | error = hfsc_setup_class__(netdev, tc_make_handle(1, queue_id + 1), | |
5324 | tc_make_handle(1, 0xfffe), &class); | |
5325 | if (error) { | |
5326 | return error; | |
5327 | } | |
5328 | ||
5329 | hfsc_update_queue__(netdev, queue_id, &class); | |
5330 | return 0; | |
5331 | } | |
5332 | ||
5333 | static int | |
5334 | hfsc_class_delete(struct netdev *netdev, struct tc_queue *queue) | |
5335 | { | |
5336 | int error; | |
5337 | struct hfsc *hfsc; | |
5338 | struct hfsc_class *hc; | |
5339 | ||
5340 | hc = hfsc_class_cast__(queue); | |
5341 | hfsc = hfsc_get__(netdev); | |
5342 | ||
5343 | error = tc_delete_class(netdev, tc_make_handle(1, queue->queue_id + 1)); | |
5344 | if (!error) { | |
5345 | hmap_remove(&hfsc->tc.queues, &hc->tc_queue.hmap_node); | |
5346 | free(hc); | |
5347 | } | |
5348 | return error; | |
5349 | } | |
5350 | ||
5351 | static int | |
5352 | hfsc_class_get_stats(const struct netdev *netdev, const struct tc_queue *queue, | |
5353 | struct netdev_queue_stats *stats) | |
5354 | { | |
5355 | return hfsc_query_class__(netdev, tc_make_handle(1, queue->queue_id + 1), | |
5356 | tc_make_handle(1, 0xfffe), NULL, stats); | |
5357 | } | |
5358 | ||
5359 | static int | |
5360 | hfsc_class_dump_stats(const struct netdev *netdev OVS_UNUSED, | |
5361 | const struct ofpbuf *nlmsg, | |
5362 | netdev_dump_queue_stats_cb *cb, void *aux) | |
5363 | { | |
5364 | struct netdev_queue_stats stats; | |
5365 | unsigned int handle, major, minor; | |
5366 | int error; | |
5367 | ||
5368 | error = tc_parse_class(nlmsg, &handle, NULL, &stats); | |
5369 | if (error) { | |
5370 | return error; | |
5371 | } | |
5372 | ||
5373 | major = tc_get_major(handle); | |
5374 | minor = tc_get_minor(handle); | |
5375 | if (major == 1 && minor > 0 && minor <= HFSC_N_QUEUES) { | |
5376 | (*cb)(minor - 1, &stats, aux); | |
5377 | } | |
5378 | return 0; | |
5379 | } | |
5380 | ||
5381 | static const struct tc_ops tc_ops_hfsc = { | |
89c09c1c BP |
5382 | .linux_name = "hfsc", |
5383 | .ovs_name = "linux-hfsc", | |
5384 | .n_queues = HFSC_N_QUEUES, /* n_queues */ | |
5385 | .tc_install = hfsc_tc_install, | |
5386 | .tc_load = hfsc_tc_load, | |
5387 | .tc_destroy = hfsc_tc_destroy, | |
5388 | .qdisc_get = hfsc_qdisc_get, | |
5389 | .qdisc_set = hfsc_qdisc_set, | |
5390 | .class_get = hfsc_class_get, | |
5391 | .class_set = hfsc_class_set, | |
5392 | .class_delete = hfsc_class_delete, | |
5393 | .class_get_stats = hfsc_class_get_stats, | |
5394 | .class_dump_stats = hfsc_class_dump_stats, | |
a339aa81 EJ |
5395 | }; |
5396 | \f | |
6cf888b8 BS |
5397 | /* "linux-noop" traffic control class. */ |
5398 | ||
5399 | static void | |
5400 | noop_install__(struct netdev *netdev_) | |
5401 | { | |
5402 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); | |
5403 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); | |
5404 | ||
5405 | netdev->tc = CONST_CAST(struct tc *, &tc); | |
5406 | } | |
5407 | ||
5408 | static int | |
5409 | noop_tc_install(struct netdev *netdev, | |
5410 | const struct smap *details OVS_UNUSED) | |
5411 | { | |
5412 | noop_install__(netdev); | |
5413 | return 0; | |
5414 | } | |
5415 | ||
5416 | static int | |
5417 | noop_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
5418 | { | |
5419 | noop_install__(netdev); | |
5420 | return 0; | |
5421 | } | |
5422 | ||
5423 | static const struct tc_ops tc_ops_noop = { | |
89c09c1c BP |
5424 | .ovs_name = "linux-noop", /* ovs_name */ |
5425 | .tc_install = noop_tc_install, | |
5426 | .tc_load = noop_tc_load, | |
6cf888b8 BS |
5427 | }; |
5428 | \f | |
c1c9c9c4 BP |
5429 | /* "linux-default" traffic control class. |
5430 | * | |
5431 | * This class represents the default, unnamed Linux qdisc. It corresponds to | |
5432 | * the "" (empty string) QoS type in the OVS database. */ | |
5433 | ||
5434 | static void | |
b5d57fc8 | 5435 | default_install__(struct netdev *netdev_) |
c1c9c9c4 | 5436 | { |
b5d57fc8 | 5437 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 5438 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_default); |
c1c9c9c4 | 5439 | |
559eb230 BP |
5440 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
5441 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 5442 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
5443 | } |
5444 | ||
5445 | static int | |
5446 | default_tc_install(struct netdev *netdev, | |
79f1cbe9 | 5447 | const struct smap *details OVS_UNUSED) |
c1c9c9c4 BP |
5448 | { |
5449 | default_install__(netdev); | |
5450 | return 0; | |
5451 | } | |
5452 | ||
5453 | static int | |
5454 | default_tc_load(struct netdev *netdev, struct ofpbuf *nlmsg OVS_UNUSED) | |
5455 | { | |
5456 | default_install__(netdev); | |
5457 | return 0; | |
5458 | } | |
5459 | ||
5460 | static const struct tc_ops tc_ops_default = { | |
89c09c1c BP |
5461 | .ovs_name = "", /* ovs_name */ |
5462 | .tc_install = default_tc_install, | |
5463 | .tc_load = default_tc_load, | |
c1c9c9c4 BP |
5464 | }; |
5465 | \f | |
5466 | /* "linux-other" traffic control class. | |
5467 | * | |
5468 | * */ | |
5469 | ||
5470 | static int | |
b5d57fc8 | 5471 | other_tc_load(struct netdev *netdev_, struct ofpbuf *nlmsg OVS_UNUSED) |
c1c9c9c4 | 5472 | { |
b5d57fc8 | 5473 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
559eb230 | 5474 | static const struct tc tc = TC_INITIALIZER(&tc, &tc_ops_other); |
c1c9c9c4 | 5475 | |
559eb230 BP |
5476 | /* Nothing but a tc class implementation is allowed to write to a tc. This |
5477 | * class never does that, so we can legitimately use a const tc object. */ | |
b5d57fc8 | 5478 | netdev->tc = CONST_CAST(struct tc *, &tc); |
c1c9c9c4 BP |
5479 | return 0; |
5480 | } | |
5481 | ||
5482 | static const struct tc_ops tc_ops_other = { | |
89c09c1c BP |
5483 | .ovs_name = "linux-other", |
5484 | .tc_load = other_tc_load, | |
c1c9c9c4 BP |
5485 | }; |
5486 | \f | |
5487 | /* Traffic control. */ | |
5488 | ||
5489 | /* Number of kernel "tc" ticks per second. */ | |
5490 | static double ticks_per_s; | |
5491 | ||
5492 | /* Number of kernel "jiffies" per second. This is used for the purpose of | |
5493 | * computing buffer sizes. Generally kernel qdiscs need to be able to buffer | |
5494 | * one jiffy's worth of data. | |
5495 | * | |
5496 | * There are two possibilities here: | |
5497 | * | |
5498 | * - 'buffer_hz' is the kernel's real timer tick rate, a small number in the | |
5499 | * approximate range of 100 to 1024. That means that we really need to | |
5500 | * make sure that the qdisc can buffer that much data. | |
5501 | * | |
5502 | * - 'buffer_hz' is an absurdly large number. That means that the kernel | |
5503 | * has finely granular timers and there's no need to fudge additional room | |
5504 | * for buffers. (There's no extra effort needed to implement that: the | |
5505 | * large 'buffer_hz' is used as a divisor, so practically any number will | |
5506 | * come out as 0 in the division. Small integer results in the case of | |
5507 | * really high dividends won't have any real effect anyhow.) | |
5508 | */ | |
5509 | static unsigned int buffer_hz; | |
5510 | ||
7874bdff RD |
5511 | static struct tcmsg * |
5512 | netdev_linux_tc_make_request(const struct netdev *netdev, int type, | |
5513 | unsigned int flags, struct ofpbuf *request) | |
5514 | { | |
5515 | int ifindex; | |
5516 | int error; | |
5517 | ||
5518 | error = get_ifindex(netdev, &ifindex); | |
5519 | if (error) { | |
5520 | return NULL; | |
5521 | } | |
5522 | ||
5523 | return tc_make_request(ifindex, type, flags, request); | |
5524 | } | |
5525 | ||
f8500004 JP |
5526 | /* Adds a policer to 'netdev' with a rate of 'kbits_rate' and a burst size |
5527 | * of 'kbits_burst'. | |
5528 | * | |
5529 | * This function is equivalent to running: | |
5530 | * /sbin/tc filter add dev <devname> parent ffff: protocol all prio 49 | |
5531 | * basic police rate <kbits_rate>kbit burst <kbits_burst>k | |
5532 | * mtu 65535 drop | |
5533 | * | |
5534 | * The configuration and stats may be seen with the following command: | |
c7952afb | 5535 | * /sbin/tc -s filter show dev <devname> parent ffff: |
f8500004 JP |
5536 | * |
5537 | * Returns 0 if successful, otherwise a positive errno value. | |
5538 | */ | |
5539 | static int | |
c7952afb BP |
5540 | tc_add_policer(struct netdev *netdev, |
5541 | uint32_t kbits_rate, uint32_t kbits_burst) | |
f8500004 JP |
5542 | { |
5543 | struct tc_police tc_police; | |
5544 | struct ofpbuf request; | |
5545 | struct tcmsg *tcmsg; | |
5546 | size_t basic_offset; | |
5547 | size_t police_offset; | |
5548 | int error; | |
5549 | int mtu = 65535; | |
5550 | ||
5551 | memset(&tc_police, 0, sizeof tc_police); | |
5552 | tc_police.action = TC_POLICE_SHOT; | |
5553 | tc_police.mtu = mtu; | |
1aca400c | 5554 | tc_fill_rate(&tc_police.rate, ((uint64_t) kbits_rate * 1000)/8, mtu); |
c7952afb | 5555 | |
79abacc8 MAA |
5556 | /* The following appears wrong in one way: In networking a kilobit is |
5557 | * usually 1000 bits but this uses 1024 bits. | |
c7952afb BP |
5558 | * |
5559 | * However if you "fix" those problems then "tc filter show ..." shows | |
5560 | * "125000b", meaning 125,000 bits, when OVS configures it for 1000 kbit == | |
5561 | * 1,000,000 bits, whereas this actually ends up doing the right thing from | |
5562 | * tc's point of view. Whatever. */ | |
5563 | tc_police.burst = tc_bytes_to_ticks( | |
79abacc8 | 5564 | tc_police.rate.rate, MIN(UINT32_MAX / 1024, kbits_burst) * 1024 / 8); |
f8500004 | 5565 | |
7874bdff RD |
5566 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_NEWTFILTER, |
5567 | NLM_F_EXCL | NLM_F_CREATE, &request); | |
f8500004 JP |
5568 | if (!tcmsg) { |
5569 | return ENODEV; | |
5570 | } | |
5571 | tcmsg->tcm_parent = tc_make_handle(0xffff, 0); | |
5572 | tcmsg->tcm_info = tc_make_handle(49, | |
5573 | (OVS_FORCE uint16_t) htons(ETH_P_ALL)); | |
5574 | ||
5575 | nl_msg_put_string(&request, TCA_KIND, "basic"); | |
5576 | basic_offset = nl_msg_start_nested(&request, TCA_OPTIONS); | |
5577 | police_offset = nl_msg_start_nested(&request, TCA_BASIC_POLICE); | |
5578 | nl_msg_put_unspec(&request, TCA_POLICE_TBF, &tc_police, sizeof tc_police); | |
5579 | tc_put_rtab(&request, TCA_POLICE_RATE, &tc_police.rate); | |
5580 | nl_msg_end_nested(&request, police_offset); | |
5581 | nl_msg_end_nested(&request, basic_offset); | |
5582 | ||
5583 | error = tc_transact(&request, NULL); | |
5584 | if (error) { | |
5585 | return error; | |
5586 | } | |
5587 | ||
5588 | return 0; | |
5589 | } | |
5590 | ||
c1c9c9c4 BP |
5591 | static void |
5592 | read_psched(void) | |
5593 | { | |
5594 | /* The values in psched are not individually very meaningful, but they are | |
5595 | * important. The tables below show some values seen in the wild. | |
5596 | * | |
5597 | * Some notes: | |
5598 | * | |
5599 | * - "c" has always been a constant 1000000 since at least Linux 2.4.14. | |
5600 | * (Before that, there are hints that it was 1000000000.) | |
5601 | * | |
5602 | * - "d" can be unrealistically large, see the comment on 'buffer_hz' | |
5603 | * above. | |
5604 | * | |
5605 | * /proc/net/psched | |
5606 | * ----------------------------------- | |
5607 | * [1] 000c8000 000f4240 000f4240 00000064 | |
5608 | * [2] 000003e8 00000400 000f4240 3b9aca00 | |
5609 | * [3] 000003e8 00000400 000f4240 3b9aca00 | |
5610 | * [4] 000003e8 00000400 000f4240 00000064 | |
5611 | * [5] 000003e8 00000040 000f4240 3b9aca00 | |
5612 | * [6] 000003e8 00000040 000f4240 000000f9 | |
5613 | * | |
5614 | * a b c d ticks_per_s buffer_hz | |
5615 | * ------- --------- ---------- ------------- ----------- ------------- | |
5616 | * [1] 819,200 1,000,000 1,000,000 100 819,200 100 | |
5617 | * [2] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
5618 | * [3] 1,000 1,024 1,000,000 1,000,000,000 976,562 1,000,000,000 | |
5619 | * [4] 1,000 1,024 1,000,000 100 976,562 100 | |
5620 | * [5] 1,000 64 1,000,000 1,000,000,000 15,625,000 1,000,000,000 | |
5621 | * [6] 1,000 64 1,000,000 249 15,625,000 249 | |
5622 | * | |
5623 | * [1] 2.6.18-128.1.6.el5.xs5.5.0.505.1024xen from XenServer 5.5.0-24648p | |
5624 | * [2] 2.6.26-1-686-bigmem from Debian lenny | |
5625 | * [3] 2.6.26-2-sparc64 from Debian lenny | |
5626 | * [4] 2.6.27.42-0.1.1.xs5.6.810.44.111163xen from XenServer 5.6.810-31078p | |
5627 | * [5] 2.6.32.21.22 (approx.) from Ubuntu 10.04 on VMware Fusion | |
5628 | * [6] 2.6.34 from kernel.org on KVM | |
5629 | */ | |
23882115 | 5630 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
c1c9c9c4 BP |
5631 | static const char fn[] = "/proc/net/psched"; |
5632 | unsigned int a, b, c, d; | |
5633 | FILE *stream; | |
5634 | ||
23882115 BP |
5635 | if (!ovsthread_once_start(&once)) { |
5636 | return; | |
5637 | } | |
5638 | ||
c1c9c9c4 BP |
5639 | ticks_per_s = 1.0; |
5640 | buffer_hz = 100; | |
5641 | ||
5642 | stream = fopen(fn, "r"); | |
5643 | if (!stream) { | |
10a89ef0 | 5644 | VLOG_WARN("%s: open failed: %s", fn, ovs_strerror(errno)); |
23882115 | 5645 | goto exit; |
c1c9c9c4 BP |
5646 | } |
5647 | ||
5648 | if (fscanf(stream, "%x %x %x %x", &a, &b, &c, &d) != 4) { | |
5649 | VLOG_WARN("%s: read failed", fn); | |
5650 | fclose(stream); | |
23882115 | 5651 | goto exit; |
c1c9c9c4 BP |
5652 | } |
5653 | VLOG_DBG("%s: psched parameters are: %u %u %u %u", fn, a, b, c, d); | |
5654 | fclose(stream); | |
5655 | ||
1bab4901 | 5656 | if (!a || !b || !c) { |
c1c9c9c4 | 5657 | VLOG_WARN("%s: invalid scheduler parameters", fn); |
23882115 | 5658 | goto exit; |
c1c9c9c4 BP |
5659 | } |
5660 | ||
5661 | ticks_per_s = (double) a * c / b; | |
5662 | if (c == 1000000) { | |
5663 | buffer_hz = d; | |
5664 | } else { | |
5665 | VLOG_WARN("%s: unexpected psched parameters: %u %u %u %u", | |
5666 | fn, a, b, c, d); | |
5667 | } | |
5668 | VLOG_DBG("%s: ticks_per_s=%f buffer_hz=%u", fn, ticks_per_s, buffer_hz); | |
23882115 BP |
5669 | |
5670 | exit: | |
5671 | ovsthread_once_done(&once); | |
c1c9c9c4 BP |
5672 | } |
5673 | ||
5674 | /* Returns the number of bytes that can be transmitted in 'ticks' ticks at a | |
5675 | * rate of 'rate' bytes per second. */ | |
5676 | static unsigned int | |
5677 | tc_ticks_to_bytes(unsigned int rate, unsigned int ticks) | |
5678 | { | |
23882115 | 5679 | read_psched(); |
c1c9c9c4 BP |
5680 | return (rate * ticks) / ticks_per_s; |
5681 | } | |
5682 | ||
5683 | /* Returns the number of ticks that it would take to transmit 'size' bytes at a | |
5684 | * rate of 'rate' bytes per second. */ | |
5685 | static unsigned int | |
5686 | tc_bytes_to_ticks(unsigned int rate, unsigned int size) | |
5687 | { | |
23882115 | 5688 | read_psched(); |
015c93a4 | 5689 | return rate ? ((unsigned long long int) ticks_per_s * size) / rate : 0; |
c1c9c9c4 BP |
5690 | } |
5691 | ||
5692 | /* Returns the number of bytes that need to be reserved for qdisc buffering at | |
5693 | * a transmission rate of 'rate' bytes per second. */ | |
5694 | static unsigned int | |
5695 | tc_buffer_per_jiffy(unsigned int rate) | |
5696 | { | |
23882115 | 5697 | read_psched(); |
c1c9c9c4 BP |
5698 | return rate / buffer_hz; |
5699 | } | |
5700 | ||
2f564bb1 S |
5701 | static uint32_t |
5702 | tc_time_to_ticks(uint32_t time) { | |
5703 | read_psched(); | |
5704 | return time * (ticks_per_s / 1000000); | |
5705 | } | |
5706 | ||
c1c9c9c4 BP |
5707 | /* Given Netlink 'msg' that describes a qdisc, extracts the name of the qdisc, |
5708 | * e.g. "htb", into '*kind' (if it is nonnull). If 'options' is nonnull, | |
5709 | * extracts 'msg''s TCA_OPTIONS attributes into '*options' if it is present or | |
5710 | * stores NULL into it if it is absent. | |
5711 | * | |
5712 | * '*kind' and '*options' point into 'msg', so they are owned by whoever owns | |
5713 | * 'msg'. | |
5714 | * | |
5715 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5716 | static int | |
5717 | tc_parse_qdisc(const struct ofpbuf *msg, const char **kind, | |
5718 | struct nlattr **options) | |
5719 | { | |
5720 | static const struct nl_policy tca_policy[] = { | |
5721 | [TCA_KIND] = { .type = NL_A_STRING, .optional = false }, | |
5722 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = true }, | |
5723 | }; | |
5724 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
5725 | ||
5726 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
5727 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
5728 | VLOG_WARN_RL(&rl, "failed to parse qdisc message"); | |
5729 | goto error; | |
5730 | } | |
5731 | ||
5732 | if (kind) { | |
5733 | *kind = nl_attr_get_string(ta[TCA_KIND]); | |
5734 | } | |
5735 | ||
5736 | if (options) { | |
5737 | *options = ta[TCA_OPTIONS]; | |
5738 | } | |
5739 | ||
5740 | return 0; | |
5741 | ||
5742 | error: | |
5743 | if (kind) { | |
5744 | *kind = NULL; | |
5745 | } | |
5746 | if (options) { | |
5747 | *options = NULL; | |
5748 | } | |
5749 | return EPROTO; | |
5750 | } | |
5751 | ||
5752 | /* Given Netlink 'msg' that describes a class, extracts the queue ID (e.g. the | |
5753 | * minor number of its class ID) into '*queue_id', its TCA_OPTIONS attribute | |
5754 | * into '*options', and its queue statistics into '*stats'. Any of the output | |
5755 | * arguments may be null. | |
5756 | * | |
5757 | * Returns 0 if successful, otherwise a positive errno value. */ | |
5758 | static int | |
5759 | tc_parse_class(const struct ofpbuf *msg, unsigned int *handlep, | |
5760 | struct nlattr **options, struct netdev_queue_stats *stats) | |
5761 | { | |
5762 | static const struct nl_policy tca_policy[] = { | |
5763 | [TCA_OPTIONS] = { .type = NL_A_NESTED, .optional = false }, | |
5764 | [TCA_STATS2] = { .type = NL_A_NESTED, .optional = false }, | |
5765 | }; | |
5766 | struct nlattr *ta[ARRAY_SIZE(tca_policy)]; | |
5767 | ||
5768 | if (!nl_policy_parse(msg, NLMSG_HDRLEN + sizeof(struct tcmsg), | |
5769 | tca_policy, ta, ARRAY_SIZE(ta))) { | |
5770 | VLOG_WARN_RL(&rl, "failed to parse class message"); | |
5771 | goto error; | |
5772 | } | |
5773 | ||
5774 | if (handlep) { | |
5775 | struct tcmsg *tc = ofpbuf_at_assert(msg, NLMSG_HDRLEN, sizeof *tc); | |
5776 | *handlep = tc->tcm_handle; | |
5777 | } | |
5778 | ||
5779 | if (options) { | |
5780 | *options = ta[TCA_OPTIONS]; | |
5781 | } | |
5782 | ||
5783 | if (stats) { | |
5784 | const struct gnet_stats_queue *gsq; | |
5785 | struct gnet_stats_basic gsb; | |
5786 | ||
5787 | static const struct nl_policy stats_policy[] = { | |
5788 | [TCA_STATS_BASIC] = { .type = NL_A_UNSPEC, .optional = false, | |
5789 | .min_len = sizeof gsb }, | |
5790 | [TCA_STATS_QUEUE] = { .type = NL_A_UNSPEC, .optional = false, | |
5791 | .min_len = sizeof *gsq }, | |
5792 | }; | |
5793 | struct nlattr *sa[ARRAY_SIZE(stats_policy)]; | |
5794 | ||
5795 | if (!nl_parse_nested(ta[TCA_STATS2], stats_policy, | |
5796 | sa, ARRAY_SIZE(sa))) { | |
5797 | VLOG_WARN_RL(&rl, "failed to parse class stats"); | |
5798 | goto error; | |
5799 | } | |
5800 | ||
5801 | /* Alignment issues screw up the length of struct gnet_stats_basic on | |
5802 | * some arch/bitsize combinations. Newer versions of Linux have a | |
5803 | * struct gnet_stats_basic_packed, but we can't depend on that. The | |
5804 | * easiest thing to do is just to make a copy. */ | |
5805 | memset(&gsb, 0, sizeof gsb); | |
5806 | memcpy(&gsb, nl_attr_get(sa[TCA_STATS_BASIC]), | |
5807 | MIN(nl_attr_get_size(sa[TCA_STATS_BASIC]), sizeof gsb)); | |
5808 | stats->tx_bytes = gsb.bytes; | |
5809 | stats->tx_packets = gsb.packets; | |
5810 | ||
5811 | gsq = nl_attr_get(sa[TCA_STATS_QUEUE]); | |
5812 | stats->tx_errors = gsq->drops; | |
5813 | } | |
5814 | ||
5815 | return 0; | |
5816 | ||
5817 | error: | |
5818 | if (options) { | |
5819 | *options = NULL; | |
5820 | } | |
5821 | if (stats) { | |
5822 | memset(stats, 0, sizeof *stats); | |
5823 | } | |
5824 | return EPROTO; | |
5825 | } | |
5826 | ||
5827 | /* Queries the kernel for class with identifier 'handle' and parent 'parent' | |
5828 | * on 'netdev'. */ | |
5829 | static int | |
5830 | tc_query_class(const struct netdev *netdev, | |
5831 | unsigned int handle, unsigned int parent, | |
5832 | struct ofpbuf **replyp) | |
5833 | { | |
5834 | struct ofpbuf request; | |
5835 | struct tcmsg *tcmsg; | |
5836 | int error; | |
5837 | ||
7874bdff RD |
5838 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_GETTCLASS, NLM_F_ECHO, |
5839 | &request); | |
23a98ffe BP |
5840 | if (!tcmsg) { |
5841 | return ENODEV; | |
5842 | } | |
c1c9c9c4 BP |
5843 | tcmsg->tcm_handle = handle; |
5844 | tcmsg->tcm_parent = parent; | |
5845 | ||
5846 | error = tc_transact(&request, replyp); | |
5847 | if (error) { | |
5848 | VLOG_WARN_RL(&rl, "query %s class %u:%u (parent %u:%u) failed (%s)", | |
5849 | netdev_get_name(netdev), | |
5850 | tc_get_major(handle), tc_get_minor(handle), | |
5851 | tc_get_major(parent), tc_get_minor(parent), | |
10a89ef0 | 5852 | ovs_strerror(error)); |
c1c9c9c4 BP |
5853 | } |
5854 | return error; | |
5855 | } | |
5856 | ||
5857 | /* Equivalent to "tc class del dev <name> handle <handle>". */ | |
5858 | static int | |
5859 | tc_delete_class(const struct netdev *netdev, unsigned int handle) | |
5860 | { | |
5861 | struct ofpbuf request; | |
5862 | struct tcmsg *tcmsg; | |
5863 | int error; | |
5864 | ||
7874bdff | 5865 | tcmsg = netdev_linux_tc_make_request(netdev, RTM_DELTCLASS, 0, &request); |
23a98ffe BP |
5866 | if (!tcmsg) { |
5867 | return ENODEV; | |
5868 | } | |
c1c9c9c4 BP |
5869 | tcmsg->tcm_handle = handle; |
5870 | tcmsg->tcm_parent = 0; | |
5871 | ||
5872 | error = tc_transact(&request, NULL); | |
5873 | if (error) { | |
5874 | VLOG_WARN_RL(&rl, "delete %s class %u:%u failed (%s)", | |
5875 | netdev_get_name(netdev), | |
5876 | tc_get_major(handle), tc_get_minor(handle), | |
10a89ef0 | 5877 | ovs_strerror(error)); |
c1c9c9c4 BP |
5878 | } |
5879 | return error; | |
5880 | } | |
5881 | ||
5882 | /* Equivalent to "tc qdisc del dev <name> root". */ | |
5883 | static int | |
b5d57fc8 | 5884 | tc_del_qdisc(struct netdev *netdev_) |
c1c9c9c4 | 5885 | { |
b5d57fc8 | 5886 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5887 | struct ofpbuf request; |
5888 | struct tcmsg *tcmsg; | |
5889 | int error; | |
5890 | ||
7874bdff | 5891 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_DELQDISC, 0, &request); |
23a98ffe BP |
5892 | if (!tcmsg) { |
5893 | return ENODEV; | |
5894 | } | |
c1c9c9c4 BP |
5895 | tcmsg->tcm_handle = tc_make_handle(1, 0); |
5896 | tcmsg->tcm_parent = TC_H_ROOT; | |
5897 | ||
5898 | error = tc_transact(&request, NULL); | |
5899 | if (error == EINVAL) { | |
5900 | /* EINVAL probably means that the default qdisc was in use, in which | |
5901 | * case we've accomplished our purpose. */ | |
5902 | error = 0; | |
5903 | } | |
b5d57fc8 BP |
5904 | if (!error && netdev->tc) { |
5905 | if (netdev->tc->ops->tc_destroy) { | |
5906 | netdev->tc->ops->tc_destroy(netdev->tc); | |
c1c9c9c4 | 5907 | } |
b5d57fc8 | 5908 | netdev->tc = NULL; |
c1c9c9c4 BP |
5909 | } |
5910 | return error; | |
5911 | } | |
5912 | ||
ac3e3aaa BP |
5913 | static bool |
5914 | getqdisc_is_safe(void) | |
5915 | { | |
5916 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
5917 | static bool safe = false; | |
5918 | ||
5919 | if (ovsthread_once_start(&once)) { | |
5920 | struct utsname utsname; | |
5921 | int major, minor; | |
5922 | ||
5923 | if (uname(&utsname) == -1) { | |
5924 | VLOG_WARN("uname failed (%s)", ovs_strerror(errno)); | |
5925 | } else if (!ovs_scan(utsname.release, "%d.%d", &major, &minor)) { | |
5926 | VLOG_WARN("uname reported bad OS release (%s)", utsname.release); | |
5927 | } else if (major < 2 || (major == 2 && minor < 35)) { | |
5928 | VLOG_INFO("disabling unsafe RTM_GETQDISC in Linux kernel %s", | |
5929 | utsname.release); | |
5930 | } else { | |
5931 | safe = true; | |
5932 | } | |
5933 | ovsthread_once_done(&once); | |
5934 | } | |
5935 | return safe; | |
5936 | } | |
5937 | ||
c1c9c9c4 BP |
5938 | /* If 'netdev''s qdisc type and parameters are not yet known, queries the |
5939 | * kernel to determine what they are. Returns 0 if successful, otherwise a | |
5940 | * positive errno value. */ | |
5941 | static int | |
b5d57fc8 | 5942 | tc_query_qdisc(const struct netdev *netdev_) |
c1c9c9c4 | 5943 | { |
b5d57fc8 | 5944 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c1c9c9c4 BP |
5945 | struct ofpbuf request, *qdisc; |
5946 | const struct tc_ops *ops; | |
5947 | struct tcmsg *tcmsg; | |
5948 | int load_error; | |
5949 | int error; | |
5950 | ||
b5d57fc8 | 5951 | if (netdev->tc) { |
c1c9c9c4 BP |
5952 | return 0; |
5953 | } | |
5954 | ||
5955 | /* This RTM_GETQDISC is crafted to avoid OOPSing kernels that do not have | |
5956 | * commit 53b0f08 "net_sched: Fix qdisc_notify()", which is anything before | |
5957 | * 2.6.35 without that fix backported to it. | |
5958 | * | |
5959 | * To avoid the OOPS, we must not make a request that would attempt to dump | |
5960 | * a "built-in" qdisc, that is, the default pfifo_fast qdisc or one of a | |
5961 | * few others. There are a few ways that I can see to do this, but most of | |
5962 | * them seem to be racy (and if you lose the race the kernel OOPSes). The | |
5963 | * technique chosen here is to assume that any non-default qdisc that we | |
5964 | * create will have a class with handle 1:0. The built-in qdiscs only have | |
5965 | * a class with handle 0:0. | |
5966 | * | |
ac3e3aaa BP |
5967 | * On Linux 2.6.35+ we use the straightforward method because it allows us |
5968 | * to handle non-builtin qdiscs without handle 1:0 (e.g. codel). However, | |
5969 | * in such a case we get no response at all from the kernel (!) if a | |
5970 | * builtin qdisc is in use (which is later caught by "!error && | |
5971 | * !qdisc->size"). */ | |
7874bdff RD |
5972 | tcmsg = netdev_linux_tc_make_request(netdev_, RTM_GETQDISC, NLM_F_ECHO, |
5973 | &request); | |
23a98ffe BP |
5974 | if (!tcmsg) { |
5975 | return ENODEV; | |
5976 | } | |
ac3e3aaa BP |
5977 | tcmsg->tcm_handle = tc_make_handle(getqdisc_is_safe() ? 0 : 1, 0); |
5978 | tcmsg->tcm_parent = getqdisc_is_safe() ? TC_H_ROOT : 0; | |
c1c9c9c4 BP |
5979 | |
5980 | /* Figure out what tc class to instantiate. */ | |
5981 | error = tc_transact(&request, &qdisc); | |
ac3e3aaa | 5982 | if (!error && qdisc->size) { |
c1c9c9c4 BP |
5983 | const char *kind; |
5984 | ||
5985 | error = tc_parse_qdisc(qdisc, &kind, NULL); | |
5986 | if (error) { | |
5987 | ops = &tc_ops_other; | |
5988 | } else { | |
5989 | ops = tc_lookup_linux_name(kind); | |
5990 | if (!ops) { | |
5991 | static struct vlog_rate_limit rl2 = VLOG_RATE_LIMIT_INIT(1, 1); | |
ac3e3aaa | 5992 | VLOG_DBG_RL(&rl2, "unknown qdisc \"%s\"", kind); |
c1c9c9c4 BP |
5993 | |
5994 | ops = &tc_ops_other; | |
5995 | } | |
5996 | } | |
ac3e3aaa BP |
5997 | } else if ((!error && !qdisc->size) || error == ENOENT) { |
5998 | /* Either it's a built-in qdisc, or (on Linux pre-2.6.35) it's a qdisc | |
5999 | * set up by some other entity that doesn't have a handle 1:0. We will | |
6000 | * assume that it's the system default qdisc. */ | |
c1c9c9c4 BP |
6001 | ops = &tc_ops_default; |
6002 | error = 0; | |
6003 | } else { | |
6004 | /* Who knows? Maybe the device got deleted. */ | |
6005 | VLOG_WARN_RL(&rl, "query %s qdisc failed (%s)", | |
10a89ef0 | 6006 | netdev_get_name(netdev_), ovs_strerror(error)); |
c1c9c9c4 BP |
6007 | ops = &tc_ops_other; |
6008 | } | |
6009 | ||
6010 | /* Instantiate it. */ | |
b5d57fc8 BP |
6011 | load_error = ops->tc_load(CONST_CAST(struct netdev *, netdev_), qdisc); |
6012 | ovs_assert((load_error == 0) == (netdev->tc != NULL)); | |
c1c9c9c4 BP |
6013 | ofpbuf_delete(qdisc); |
6014 | ||
6015 | return error ? error : load_error; | |
6016 | } | |
6017 | ||
6018 | /* Linux traffic control uses tables with 256 entries ("rtab" tables) to | |
6019 | approximate the time to transmit packets of various lengths. For an MTU of | |
6020 | 256 or less, each entry is exact; for an MTU of 257 through 512, each entry | |
6021 | represents two possible packet lengths; for a MTU of 513 through 1024, four | |
6022 | possible lengths; and so on. | |
6023 | ||
6024 | Returns, for the specified 'mtu', the number of bits that packet lengths | |
6025 | need to be shifted right to fit within such a 256-entry table. */ | |
6026 | static int | |
6027 | tc_calc_cell_log(unsigned int mtu) | |
6028 | { | |
6029 | int cell_log; | |
6030 | ||
6031 | if (!mtu) { | |
6032 | mtu = ETH_PAYLOAD_MAX; | |
6033 | } | |
6034 | mtu += ETH_HEADER_LEN + VLAN_HEADER_LEN; | |
6035 | ||
6036 | for (cell_log = 0; mtu >= 256; cell_log++) { | |
6037 | mtu >>= 1; | |
6038 | } | |
6039 | ||
6040 | return cell_log; | |
6041 | } | |
6042 | ||
6043 | /* Initializes 'rate' properly for a rate of 'Bps' bytes per second with an MTU | |
6044 | * of 'mtu'. */ | |
6045 | static void | |
6046 | tc_fill_rate(struct tc_ratespec *rate, uint64_t Bps, int mtu) | |
6047 | { | |
6048 | memset(rate, 0, sizeof *rate); | |
6049 | rate->cell_log = tc_calc_cell_log(mtu); | |
6050 | /* rate->overhead = 0; */ /* New in 2.6.24, not yet in some */ | |
6051 | /* rate->cell_align = 0; */ /* distro headers. */ | |
6052 | rate->mpu = ETH_TOTAL_MIN; | |
6053 | rate->rate = Bps; | |
6054 | } | |
6055 | ||
6056 | /* Appends to 'msg' an "rtab" table for the specified 'rate' as a Netlink | |
6057 | * attribute of the specified "type". | |
6058 | * | |
6059 | * See tc_calc_cell_log() above for a description of "rtab"s. */ | |
e7f6ba22 | 6060 | void |
c1c9c9c4 BP |
6061 | tc_put_rtab(struct ofpbuf *msg, uint16_t type, const struct tc_ratespec *rate) |
6062 | { | |
6063 | uint32_t *rtab; | |
6064 | unsigned int i; | |
6065 | ||
6066 | rtab = nl_msg_put_unspec_uninit(msg, type, TC_RTAB_SIZE); | |
6067 | for (i = 0; i < TC_RTAB_SIZE / sizeof *rtab; i++) { | |
6068 | unsigned packet_size = (i + 1) << rate->cell_log; | |
6069 | if (packet_size < rate->mpu) { | |
6070 | packet_size = rate->mpu; | |
6071 | } | |
6072 | rtab[i] = tc_bytes_to_ticks(rate->rate, packet_size); | |
6073 | } | |
6074 | } | |
6075 | ||
6076 | /* Calculates the proper value of 'buffer' or 'cbuffer' in HTB options given a | |
6077 | * rate of 'Bps' bytes per second, the specified 'mtu', and a user-requested | |
6078 | * burst size of 'burst_bytes'. (If no value was requested, a 'burst_bytes' of | |
015c93a4 | 6079 | * 0 is fine.) */ |
c1c9c9c4 BP |
6080 | static int |
6081 | tc_calc_buffer(unsigned int Bps, int mtu, uint64_t burst_bytes) | |
6082 | { | |
6083 | unsigned int min_burst = tc_buffer_per_jiffy(Bps) + mtu; | |
6084 | return tc_bytes_to_ticks(Bps, MAX(burst_bytes, min_burst)); | |
6085 | } | |
d3980822 | 6086 | \f |
aaf2fb1a BP |
6087 | /* Linux-only functions declared in netdev-linux.h */ |
6088 | ||
6089 | /* Modifies the 'flag' bit in ethtool's flags field for 'netdev'. If | |
6090 | * 'enable' is true, the bit is set. Otherwise, it is cleared. */ | |
6091 | int | |
6092 | netdev_linux_ethtool_set_flag(struct netdev *netdev, uint32_t flag, | |
6093 | const char *flag_name, bool enable) | |
6094 | { | |
6095 | const char *netdev_name = netdev_get_name(netdev); | |
6096 | struct ethtool_value evalue; | |
6097 | uint32_t new_flags; | |
6098 | int error; | |
6099 | ||
ab985a77 | 6100 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
6101 | memset(&evalue, 0, sizeof evalue); |
6102 | error = netdev_linux_do_ethtool(netdev_name, | |
6103 | (struct ethtool_cmd *)&evalue, | |
6104 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
6105 | if (error) { | |
6106 | return error; | |
6107 | } | |
6108 | ||
ab985a77 | 6109 | COVERAGE_INC(netdev_set_ethtool); |
ad2dade5 AS |
6110 | new_flags = (evalue.data & ~flag) | (enable ? flag : 0); |
6111 | if (new_flags == evalue.data) { | |
6112 | return 0; | |
6113 | } | |
6114 | evalue.data = new_flags; | |
aaf2fb1a BP |
6115 | error = netdev_linux_do_ethtool(netdev_name, |
6116 | (struct ethtool_cmd *)&evalue, | |
6117 | ETHTOOL_SFLAGS, "ETHTOOL_SFLAGS"); | |
6118 | if (error) { | |
6119 | return error; | |
6120 | } | |
6121 | ||
ab985a77 | 6122 | COVERAGE_INC(netdev_get_ethtool); |
aaf2fb1a BP |
6123 | memset(&evalue, 0, sizeof evalue); |
6124 | error = netdev_linux_do_ethtool(netdev_name, | |
6125 | (struct ethtool_cmd *)&evalue, | |
6126 | ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); | |
6127 | if (error) { | |
6128 | return error; | |
6129 | } | |
6130 | ||
6131 | if (new_flags != evalue.data) { | |
6132 | VLOG_WARN_RL(&rl, "attempt to %s ethtool %s flag on network " | |
6133 | "device %s failed", enable ? "enable" : "disable", | |
6134 | flag_name, netdev_name); | |
6135 | return EOPNOTSUPP; | |
6136 | } | |
6137 | ||
6138 | return 0; | |
6139 | } | |
6140 | \f | |
6141 | /* Utility functions. */ | |
6142 | ||
d3980822 | 6143 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
f613a0d7 | 6144 | static void |
d3980822 BP |
6145 | netdev_stats_from_rtnl_link_stats(struct netdev_stats *dst, |
6146 | const struct rtnl_link_stats *src) | |
6147 | { | |
f613a0d7 PS |
6148 | dst->rx_packets = src->rx_packets; |
6149 | dst->tx_packets = src->tx_packets; | |
6150 | dst->rx_bytes = src->rx_bytes; | |
6151 | dst->tx_bytes = src->tx_bytes; | |
6152 | dst->rx_errors = src->rx_errors; | |
6153 | dst->tx_errors = src->tx_errors; | |
6154 | dst->rx_dropped = src->rx_dropped; | |
6155 | dst->tx_dropped = src->tx_dropped; | |
6156 | dst->multicast = src->multicast; | |
6157 | dst->collisions = src->collisions; | |
6158 | dst->rx_length_errors = src->rx_length_errors; | |
6159 | dst->rx_over_errors = src->rx_over_errors; | |
6160 | dst->rx_crc_errors = src->rx_crc_errors; | |
6161 | dst->rx_frame_errors = src->rx_frame_errors; | |
6162 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
6163 | dst->rx_missed_errors = src->rx_missed_errors; | |
6164 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
6165 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
6166 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
6167 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
6168 | dst->tx_window_errors = src->tx_window_errors; | |
d3980822 BP |
6169 | } |
6170 | ||
337c9b99 BP |
6171 | /* Copies 'src' into 'dst', performing format conversion in the process. */ |
6172 | static void | |
6173 | netdev_stats_from_rtnl_link_stats64(struct netdev_stats *dst, | |
6174 | const struct rtnl_link_stats64 *src) | |
6175 | { | |
6176 | dst->rx_packets = src->rx_packets; | |
6177 | dst->tx_packets = src->tx_packets; | |
6178 | dst->rx_bytes = src->rx_bytes; | |
6179 | dst->tx_bytes = src->tx_bytes; | |
6180 | dst->rx_errors = src->rx_errors; | |
6181 | dst->tx_errors = src->tx_errors; | |
6182 | dst->rx_dropped = src->rx_dropped; | |
6183 | dst->tx_dropped = src->tx_dropped; | |
6184 | dst->multicast = src->multicast; | |
6185 | dst->collisions = src->collisions; | |
6186 | dst->rx_length_errors = src->rx_length_errors; | |
6187 | dst->rx_over_errors = src->rx_over_errors; | |
6188 | dst->rx_crc_errors = src->rx_crc_errors; | |
6189 | dst->rx_frame_errors = src->rx_frame_errors; | |
6190 | dst->rx_fifo_errors = src->rx_fifo_errors; | |
6191 | dst->rx_missed_errors = src->rx_missed_errors; | |
6192 | dst->tx_aborted_errors = src->tx_aborted_errors; | |
6193 | dst->tx_carrier_errors = src->tx_carrier_errors; | |
6194 | dst->tx_fifo_errors = src->tx_fifo_errors; | |
6195 | dst->tx_heartbeat_errors = src->tx_heartbeat_errors; | |
6196 | dst->tx_window_errors = src->tx_window_errors; | |
6197 | } | |
6198 | ||
0de1b425 | 6199 | int |
35eef899 | 6200 | get_stats_via_netlink(const struct netdev *netdev_, struct netdev_stats *stats) |
c1c9c9c4 | 6201 | { |
c1c9c9c4 BP |
6202 | struct ofpbuf request; |
6203 | struct ofpbuf *reply; | |
c1c9c9c4 BP |
6204 | int error; |
6205 | ||
d6e3feb5 | 6206 | /* Filtering all counters by default */ |
6207 | memset(stats, 0xFF, sizeof(struct netdev_stats)); | |
6208 | ||
c1c9c9c4 | 6209 | ofpbuf_init(&request, 0); |
13a24df8 BP |
6210 | nl_msg_put_nlmsghdr(&request, |
6211 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ), | |
6212 | RTM_GETLINK, NLM_F_REQUEST); | |
6213 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); | |
6214 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(netdev_)); | |
a88b4e04 | 6215 | error = nl_transact(NETLINK_ROUTE, &request, &reply); |
c1c9c9c4 BP |
6216 | ofpbuf_uninit(&request); |
6217 | if (error) { | |
6218 | return error; | |
6219 | } | |
6220 | ||
13a24df8 | 6221 | if (ofpbuf_try_pull(reply, NLMSG_HDRLEN + sizeof(struct ifinfomsg))) { |
337c9b99 BP |
6222 | const struct nlattr *a = nl_attr_find(reply, 0, IFLA_STATS64); |
6223 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats64)) { | |
6224 | netdev_stats_from_rtnl_link_stats64(stats, nl_attr_get(a)); | |
13a24df8 BP |
6225 | error = 0; |
6226 | } else { | |
71f21279 | 6227 | a = nl_attr_find(reply, 0, IFLA_STATS); |
337c9b99 BP |
6228 | if (a && nl_attr_get_size(a) >= sizeof(struct rtnl_link_stats)) { |
6229 | netdev_stats_from_rtnl_link_stats(stats, nl_attr_get(a)); | |
6230 | error = 0; | |
6231 | } else { | |
6232 | VLOG_WARN_RL(&rl, "RTM_GETLINK reply lacks stats"); | |
6233 | error = EPROTO; | |
6234 | } | |
13a24df8 BP |
6235 | } |
6236 | } else { | |
6237 | VLOG_WARN_RL(&rl, "short RTM_GETLINK reply"); | |
6238 | error = EPROTO; | |
c1c9c9c4 | 6239 | } |
8b61709d | 6240 | |
8b61709d | 6241 | |
576e26d7 | 6242 | ofpbuf_delete(reply); |
35eef899 | 6243 | return error; |
8b61709d | 6244 | } |
c1c9c9c4 | 6245 | |
3a183124 | 6246 | static int |
b5d57fc8 | 6247 | get_flags(const struct netdev *dev, unsigned int *flags) |
8b61709d BP |
6248 | { |
6249 | struct ifreq ifr; | |
6250 | int error; | |
6251 | ||
755be9ea | 6252 | *flags = 0; |
259e0b1a | 6253 | error = af_inet_ifreq_ioctl(dev->name, &ifr, SIOCGIFFLAGS, "SIOCGIFFLAGS"); |
755be9ea EJ |
6254 | if (!error) { |
6255 | *flags = ifr.ifr_flags; | |
6256 | } | |
8b61709d BP |
6257 | return error; |
6258 | } | |
6259 | ||
6260 | static int | |
4b609110 | 6261 | set_flags(const char *name, unsigned int flags) |
8b61709d BP |
6262 | { |
6263 | struct ifreq ifr; | |
6264 | ||
6265 | ifr.ifr_flags = flags; | |
259e0b1a | 6266 | return af_inet_ifreq_ioctl(name, &ifr, SIOCSIFFLAGS, "SIOCSIFFLAGS"); |
8b61709d BP |
6267 | } |
6268 | ||
01b25786 PB |
6269 | int |
6270 | linux_get_ifindex(const char *netdev_name) | |
8b61709d BP |
6271 | { |
6272 | struct ifreq ifr; | |
259e0b1a | 6273 | int error; |
8b61709d | 6274 | |
71d7c22f | 6275 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 6276 | COVERAGE_INC(netdev_get_ifindex); |
259e0b1a BP |
6277 | |
6278 | error = af_inet_ioctl(SIOCGIFINDEX, &ifr); | |
6279 | if (error) { | |
580e1152 RD |
6280 | /* ENODEV probably means that a vif disappeared asynchronously and |
6281 | * hasn't been removed from the database yet, so reduce the log level | |
6282 | * to INFO for that case. */ | |
6283 | VLOG_RL(&rl, error == ENODEV ? VLL_INFO : VLL_ERR, | |
6284 | "ioctl(SIOCGIFINDEX) on %s device failed: %s", | |
6285 | netdev_name, ovs_strerror(error)); | |
259e0b1a | 6286 | return -error; |
8b61709d BP |
6287 | } |
6288 | return ifr.ifr_ifindex; | |
6289 | } | |
6290 | ||
6291 | static int | |
6292 | get_ifindex(const struct netdev *netdev_, int *ifindexp) | |
6293 | { | |
b5d57fc8 | 6294 | struct netdev_linux *netdev = netdev_linux_cast(netdev_); |
c7b1b0a5 | 6295 | |
b5d57fc8 | 6296 | if (!(netdev->cache_valid & VALID_IFINDEX)) { |
756819dd FL |
6297 | netdev_linux_update_via_netlink(netdev); |
6298 | } | |
6299 | ||
6300 | if (!(netdev->cache_valid & VALID_IFINDEX)) { | |
6301 | /* Fall back to ioctl if netlink fails */ | |
01b25786 | 6302 | int ifindex = linux_get_ifindex(netdev_get_name(netdev_)); |
c7b1b0a5 | 6303 | |
8b61709d | 6304 | if (ifindex < 0) { |
b5d57fc8 BP |
6305 | netdev->get_ifindex_error = -ifindex; |
6306 | netdev->ifindex = 0; | |
c7b1b0a5 | 6307 | } else { |
b5d57fc8 BP |
6308 | netdev->get_ifindex_error = 0; |
6309 | netdev->ifindex = ifindex; | |
8b61709d | 6310 | } |
b5d57fc8 | 6311 | netdev->cache_valid |= VALID_IFINDEX; |
8b61709d | 6312 | } |
c7b1b0a5 | 6313 | |
b5d57fc8 BP |
6314 | *ifindexp = netdev->ifindex; |
6315 | return netdev->get_ifindex_error; | |
8b61709d BP |
6316 | } |
6317 | ||
6318 | static int | |
756819dd FL |
6319 | netdev_linux_update_via_netlink(struct netdev_linux *netdev) |
6320 | { | |
6321 | struct ofpbuf request; | |
6322 | struct ofpbuf *reply; | |
6323 | struct rtnetlink_change chg; | |
6324 | struct rtnetlink_change *change = &chg; | |
6325 | int error; | |
6326 | ||
6327 | ofpbuf_init(&request, 0); | |
6328 | nl_msg_put_nlmsghdr(&request, | |
b43762a5 FL |
6329 | sizeof(struct ifinfomsg) + NL_ATTR_SIZE(IFNAMSIZ) + |
6330 | NL_A_U32_SIZE, RTM_GETLINK, NLM_F_REQUEST); | |
756819dd FL |
6331 | ofpbuf_put_zeros(&request, sizeof(struct ifinfomsg)); |
6332 | ||
6333 | /* The correct identifiers for a Linux device are netnsid and ifindex, | |
6334 | * but ifindex changes as the port is moved to another network namespace | |
6335 | * and the interface name statically stored in ovsdb. */ | |
6336 | nl_msg_put_string(&request, IFLA_IFNAME, netdev_get_name(&netdev->up)); | |
6337 | if (netdev_linux_netnsid_is_remote(netdev)) { | |
23fa50f6 | 6338 | nl_msg_put_u32(&request, IFLA_IF_NETNSID, netdev->netnsid); |
756819dd FL |
6339 | } |
6340 | error = nl_transact(NETLINK_ROUTE, &request, &reply); | |
6341 | ofpbuf_uninit(&request); | |
6342 | if (error) { | |
6343 | ofpbuf_delete(reply); | |
6344 | return error; | |
6345 | } | |
6346 | ||
6347 | if (rtnetlink_parse(reply, change) | |
6348 | && change->nlmsg_type == RTM_NEWLINK) { | |
6349 | bool changed = false; | |
6350 | error = 0; | |
6351 | ||
6352 | /* Update netdev from rtnl msg and increment its seq if needed. */ | |
6353 | if ((change->ifi_flags ^ netdev->ifi_flags) & IFF_RUNNING) { | |
6354 | netdev->carrier_resets++; | |
6355 | changed = true; | |
6356 | } | |
6357 | if (change->ifi_flags != netdev->ifi_flags) { | |
6358 | netdev->ifi_flags = change->ifi_flags; | |
6359 | changed = true; | |
6360 | } | |
6361 | if (change->mtu && change->mtu != netdev->mtu) { | |
6362 | netdev->mtu = change->mtu; | |
6363 | netdev->cache_valid |= VALID_MTU; | |
6364 | netdev->netdev_mtu_error = 0; | |
6365 | changed = true; | |
6366 | } | |
6367 | if (!eth_addr_is_zero(change->mac) | |
6368 | && !eth_addr_equals(change->mac, netdev->etheraddr)) { | |
6369 | netdev->etheraddr = change->mac; | |
6370 | netdev->cache_valid |= VALID_ETHERADDR; | |
6371 | netdev->ether_addr_error = 0; | |
6372 | changed = true; | |
6373 | } | |
6374 | if (change->if_index != netdev->ifindex) { | |
6375 | netdev->ifindex = change->if_index; | |
6376 | netdev->cache_valid |= VALID_IFINDEX; | |
6377 | netdev->get_ifindex_error = 0; | |
6378 | changed = true; | |
6379 | } | |
3d9c99ab JH |
6380 | if (change->master && netdev_linux_kind_is_lag(change->master)) { |
6381 | netdev->is_lag_master = true; | |
6382 | } | |
756819dd FL |
6383 | if (changed) { |
6384 | netdev_change_seq_changed(&netdev->up); | |
6385 | } | |
6386 | } else { | |
6387 | error = EINVAL; | |
6388 | } | |
6389 | ||
6390 | ofpbuf_delete(reply); | |
6391 | return error; | |
6392 | } | |
6393 | ||
6394 | static int | |
74ff3298 | 6395 | get_etheraddr(const char *netdev_name, struct eth_addr *ea) |
8b61709d BP |
6396 | { |
6397 | struct ifreq ifr; | |
6398 | int hwaddr_family; | |
259e0b1a | 6399 | int error; |
8b61709d BP |
6400 | |
6401 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 6402 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
8b61709d | 6403 | COVERAGE_INC(netdev_get_hwaddr); |
259e0b1a BP |
6404 | error = af_inet_ioctl(SIOCGIFHWADDR, &ifr); |
6405 | if (error) { | |
78857dfb BP |
6406 | /* ENODEV probably means that a vif disappeared asynchronously and |
6407 | * hasn't been removed from the database yet, so reduce the log level | |
6408 | * to INFO for that case. */ | |
259e0b1a | 6409 | VLOG(error == ENODEV ? VLL_INFO : VLL_ERR, |
78857dfb | 6410 | "ioctl(SIOCGIFHWADDR) on %s device failed: %s", |
259e0b1a BP |
6411 | netdev_name, ovs_strerror(error)); |
6412 | return error; | |
8b61709d BP |
6413 | } |
6414 | hwaddr_family = ifr.ifr_hwaddr.sa_family; | |
2482b0b0 JS |
6415 | if (hwaddr_family != AF_UNSPEC && hwaddr_family != ARPHRD_ETHER && |
6416 | hwaddr_family != ARPHRD_NONE) { | |
c9697f35 | 6417 | VLOG_INFO("%s device has unknown hardware address family %d", |
8b61709d | 6418 | netdev_name, hwaddr_family); |
c9697f35 | 6419 | return EINVAL; |
8b61709d BP |
6420 | } |
6421 | memcpy(ea, ifr.ifr_hwaddr.sa_data, ETH_ADDR_LEN); | |
6422 | return 0; | |
6423 | } | |
6424 | ||
6425 | static int | |
74ff3298 | 6426 | set_etheraddr(const char *netdev_name, const struct eth_addr mac) |
8b61709d BP |
6427 | { |
6428 | struct ifreq ifr; | |
259e0b1a | 6429 | int error; |
8b61709d BP |
6430 | |
6431 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 6432 | ovs_strzcpy(ifr.ifr_name, netdev_name, sizeof ifr.ifr_name); |
44445cac | 6433 | ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER; |
74ff3298 | 6434 | memcpy(ifr.ifr_hwaddr.sa_data, &mac, ETH_ADDR_LEN); |
8b61709d | 6435 | COVERAGE_INC(netdev_set_hwaddr); |
259e0b1a BP |
6436 | error = af_inet_ioctl(SIOCSIFHWADDR, &ifr); |
6437 | if (error) { | |
8b61709d | 6438 | VLOG_ERR("ioctl(SIOCSIFHWADDR) on %s device failed: %s", |
259e0b1a | 6439 | netdev_name, ovs_strerror(error)); |
8b61709d | 6440 | } |
259e0b1a | 6441 | return error; |
8b61709d BP |
6442 | } |
6443 | ||
6444 | static int | |
0b0544d7 | 6445 | netdev_linux_do_ethtool(const char *name, struct ethtool_cmd *ecmd, |
8b61709d BP |
6446 | int cmd, const char *cmd_name) |
6447 | { | |
6448 | struct ifreq ifr; | |
259e0b1a | 6449 | int error; |
8b61709d BP |
6450 | |
6451 | memset(&ifr, 0, sizeof ifr); | |
71d7c22f | 6452 | ovs_strzcpy(ifr.ifr_name, name, sizeof ifr.ifr_name); |
8b61709d BP |
6453 | ifr.ifr_data = (caddr_t) ecmd; |
6454 | ||
6455 | ecmd->cmd = cmd; | |
259e0b1a BP |
6456 | error = af_inet_ioctl(SIOCETHTOOL, &ifr); |
6457 | if (error) { | |
6458 | if (error != EOPNOTSUPP) { | |
8b61709d | 6459 | VLOG_WARN_RL(&rl, "ethtool command %s on network device %s " |
259e0b1a | 6460 | "failed: %s", cmd_name, name, ovs_strerror(error)); |
8b61709d BP |
6461 | } else { |
6462 | /* The device doesn't support this operation. That's pretty | |
6463 | * common, so there's no point in logging anything. */ | |
6464 | } | |
8b61709d | 6465 | } |
259e0b1a | 6466 | return error; |
8b61709d | 6467 | } |
f1acd62b | 6468 | |
488d734d BP |
6469 | /* Returns an AF_PACKET raw socket or a negative errno value. */ |
6470 | static int | |
6471 | af_packet_sock(void) | |
6472 | { | |
23882115 BP |
6473 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
6474 | static int sock; | |
488d734d | 6475 | |
23882115 | 6476 | if (ovsthread_once_start(&once)) { |
488d734d BP |
6477 | sock = socket(AF_PACKET, SOCK_RAW, 0); |
6478 | if (sock >= 0) { | |
8450059e BP |
6479 | int error = set_nonblocking(sock); |
6480 | if (error) { | |
6481 | close(sock); | |
6482 | sock = -error; | |
29cf9c1b FL |
6483 | } else if (userspace_tso_enabled()) { |
6484 | int val = 1; | |
6485 | error = setsockopt(sock, SOL_PACKET, PACKET_VNET_HDR, &val, | |
6486 | sizeof val); | |
6487 | if (error) { | |
6488 | error = errno; | |
6489 | VLOG_ERR("failed to enable vnet hdr in raw socket: %s", | |
6490 | ovs_strerror(errno)); | |
6491 | close(sock); | |
6492 | sock = -error; | |
6493 | } | |
8450059e | 6494 | } |
488d734d BP |
6495 | } else { |
6496 | sock = -errno; | |
10a89ef0 BP |
6497 | VLOG_ERR("failed to create packet socket: %s", |
6498 | ovs_strerror(errno)); | |
488d734d | 6499 | } |
23882115 | 6500 | ovsthread_once_done(&once); |
488d734d BP |
6501 | } |
6502 | ||
6503 | return sock; | |
6504 | } | |
29cf9c1b FL |
6505 | |
6506 | static int | |
6507 | netdev_linux_parse_l2(struct dp_packet *b, uint16_t *l4proto) | |
6508 | { | |
6509 | struct eth_header *eth_hdr; | |
6510 | ovs_be16 eth_type; | |
6511 | int l2_len; | |
6512 | ||
6513 | eth_hdr = dp_packet_at(b, 0, ETH_HEADER_LEN); | |
6514 | if (!eth_hdr) { | |
6515 | return -EINVAL; | |
6516 | } | |
6517 | ||
6518 | l2_len = ETH_HEADER_LEN; | |
6519 | eth_type = eth_hdr->eth_type; | |
6520 | if (eth_type_vlan(eth_type)) { | |
6521 | struct vlan_header *vlan = dp_packet_at(b, l2_len, VLAN_HEADER_LEN); | |
6522 | ||
6523 | if (!vlan) { | |
6524 | return -EINVAL; | |
6525 | } | |
6526 | ||
6527 | eth_type = vlan->vlan_next_type; | |
6528 | l2_len += VLAN_HEADER_LEN; | |
6529 | } | |
6530 | ||
6531 | if (eth_type == htons(ETH_TYPE_IP)) { | |
6532 | struct ip_header *ip_hdr = dp_packet_at(b, l2_len, IP_HEADER_LEN); | |
6533 | ||
6534 | if (!ip_hdr) { | |
6535 | return -EINVAL; | |
6536 | } | |
6537 | ||
6538 | *l4proto = ip_hdr->ip_proto; | |
6539 | dp_packet_hwol_set_tx_ipv4(b); | |
6540 | } else if (eth_type == htons(ETH_TYPE_IPV6)) { | |
6541 | struct ovs_16aligned_ip6_hdr *nh6; | |
6542 | ||
6543 | nh6 = dp_packet_at(b, l2_len, IPV6_HEADER_LEN); | |
6544 | if (!nh6) { | |
6545 | return -EINVAL; | |
6546 | } | |
6547 | ||
6548 | *l4proto = nh6->ip6_ctlun.ip6_un1.ip6_un1_nxt; | |
6549 | dp_packet_hwol_set_tx_ipv6(b); | |
6550 | } | |
6551 | ||
6552 | return 0; | |
6553 | } | |
6554 | ||
6555 | static int | |
6556 | netdev_linux_parse_vnet_hdr(struct dp_packet *b) | |
6557 | { | |
6558 | struct virtio_net_hdr *vnet = dp_packet_pull(b, sizeof *vnet); | |
6559 | uint16_t l4proto = 0; | |
6560 | ||
6561 | if (OVS_UNLIKELY(!vnet)) { | |
6562 | return -EINVAL; | |
6563 | } | |
6564 | ||
6565 | if (vnet->flags == 0 && vnet->gso_type == VIRTIO_NET_HDR_GSO_NONE) { | |
6566 | return 0; | |
6567 | } | |
6568 | ||
6569 | if (netdev_linux_parse_l2(b, &l4proto)) { | |
6570 | return -EINVAL; | |
6571 | } | |
6572 | ||
6573 | if (vnet->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
6574 | if (l4proto == IPPROTO_TCP) { | |
6575 | dp_packet_hwol_set_csum_tcp(b); | |
6576 | } else if (l4proto == IPPROTO_UDP) { | |
6577 | dp_packet_hwol_set_csum_udp(b); | |
6578 | } else if (l4proto == IPPROTO_SCTP) { | |
6579 | dp_packet_hwol_set_csum_sctp(b); | |
6580 | } | |
6581 | } | |
6582 | ||
6583 | if (l4proto && vnet->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
6584 | uint8_t allowed_mask = VIRTIO_NET_HDR_GSO_TCPV4 | |
6585 | | VIRTIO_NET_HDR_GSO_TCPV6 | |
6586 | | VIRTIO_NET_HDR_GSO_UDP; | |
6587 | uint8_t type = vnet->gso_type & allowed_mask; | |
6588 | ||
6589 | if (type == VIRTIO_NET_HDR_GSO_TCPV4 | |
6590 | || type == VIRTIO_NET_HDR_GSO_TCPV6) { | |
6591 | dp_packet_hwol_set_tcp_seg(b); | |
6592 | } | |
6593 | } | |
6594 | ||
6595 | return 0; | |
6596 | } | |
6597 | ||
6598 | static void | |
6599 | netdev_linux_prepend_vnet_hdr(struct dp_packet *b, int mtu) | |
6600 | { | |
6601 | struct virtio_net_hdr *vnet = dp_packet_push_zeros(b, sizeof *vnet); | |
6602 | ||
6603 | if (dp_packet_hwol_is_tso(b)) { | |
6604 | uint16_t hdr_len = ((char *)dp_packet_l4(b) - (char *)dp_packet_eth(b)) | |
6605 | + TCP_HEADER_LEN; | |
6606 | ||
6607 | vnet->hdr_len = (OVS_FORCE __virtio16)hdr_len; | |
6608 | vnet->gso_size = (OVS_FORCE __virtio16)(mtu - hdr_len); | |
6609 | if (dp_packet_hwol_is_ipv4(b)) { | |
6610 | vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
6611 | } else { | |
6612 | vnet->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
6613 | } | |
6614 | ||
6615 | } else { | |
6616 | vnet->flags = VIRTIO_NET_HDR_GSO_NONE; | |
6617 | } | |
6618 | ||
6619 | if (dp_packet_hwol_l4_mask(b)) { | |
6620 | vnet->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
6621 | vnet->csum_start = (OVS_FORCE __virtio16)((char *)dp_packet_l4(b) | |
6622 | - (char *)dp_packet_eth(b)); | |
6623 | ||
6624 | if (dp_packet_hwol_l4_is_tcp(b)) { | |
6625 | vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof( | |
6626 | struct tcp_header, tcp_csum); | |
6627 | } else if (dp_packet_hwol_l4_is_udp(b)) { | |
6628 | vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof( | |
6629 | struct udp_header, udp_csum); | |
6630 | } else if (dp_packet_hwol_l4_is_sctp(b)) { | |
6631 | vnet->csum_offset = (OVS_FORCE __virtio16) __builtin_offsetof( | |
6632 | struct sctp_header, sctp_csum); | |
6633 | } else { | |
6634 | VLOG_WARN_RL(&rl, "Unsupported L4 protocol"); | |
6635 | } | |
6636 | } | |
6637 | } |