blob: e95b67cd571844798de98dc4a97743ed89e0d37b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/sch_generic.c Generic packet scheduler routines.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Jamal Hadi Salim, <hadi@cyberus.ca> 990601
11 * - Ingress support
12 */
13
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/module.h>
16#include <linux/types.h>
17#include <linux/kernel.h>
18#include <linux/sched.h>
19#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/netdevice.h>
22#include <linux/skbuff.h>
23#include <linux/rtnetlink.h>
24#include <linux/init.h>
25#include <linux/rcupdate.h>
26#include <linux/list.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +020028#include <linux/if_vlan.h>
Jiri Pirko292f1c72013-02-12 00:12:03 +000029#include <net/sch_generic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <net/pkt_sched.h>
Eric Dumazet7fee2262010-05-11 23:19:48 +000031#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
stephen hemminger34aedd32013-08-31 10:15:33 -070033/* Qdisc to use by default */
34const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
35EXPORT_SYMBOL(default_qdisc_ops);
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/* Main transmission queue. */
38
Patrick McHardy0463d4a2007-04-16 17:02:10 -070039/* Modifications to data participating in scheduling must be protected with
David S. Miller5fb66222008-08-02 20:02:43 -070040 * qdisc_lock(qdisc) spinlock.
Patrick McHardy0463d4a2007-04-16 17:02:10 -070041 *
42 * The idea is the following:
David S. Millerc7e4f3b2008-07-16 03:22:39 -070043 * - enqueue, dequeue are serialized via qdisc root lock
44 * - ingress filtering is also serialized via qdisc root lock
Patrick McHardy0463d4a2007-04-16 17:02:10 -070045 * - updates to tree and tree walking are only done under the rtnl mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David S. Miller37437bb2008-07-16 02:15:04 -070048static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070049{
Jarek Poplawski62523522008-10-06 10:41:50 -070050 q->gso_skb = skb;
Jarek Poplawski53e91502008-10-08 11:36:22 -070051 q->qstats.requeues++;
WANG Conga27758f2016-06-03 15:05:57 -070052 qdisc_qstats_backlog_inc(q, skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +000053 q->q.qlen++; /* it's still part of the queue */
David S. Miller37437bb2008-07-16 02:15:04 -070054 __netif_schedule(q);
Jarek Poplawski62523522008-10-06 10:41:50 -070055
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070056 return 0;
57}
58
Eric Dumazet55a93b32014-10-03 15:31:07 -070059static void try_bulk_dequeue_skb(struct Qdisc *q,
60 struct sk_buff *skb,
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +020061 const struct netdev_queue *txq,
62 int *packets)
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +020063{
Eric Dumazet55a93b32014-10-03 15:31:07 -070064 int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +020065
66 while (bytelimit > 0) {
Eric Dumazet55a93b32014-10-03 15:31:07 -070067 struct sk_buff *nskb = q->dequeue(q);
68
69 if (!nskb)
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +020070 break;
71
Eric Dumazet55a93b32014-10-03 15:31:07 -070072 bytelimit -= nskb->len; /* covers GSO len */
73 skb->next = nskb;
74 skb = nskb;
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +020075 (*packets)++; /* GSO counts as one pkt */
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +020076 }
Eric Dumazet55a93b32014-10-03 15:31:07 -070077 skb->next = NULL;
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +020078}
79
Eric Dumazet4d202a02016-06-21 23:16:52 -070080/* This variant of try_bulk_dequeue_skb() makes sure
81 * all skbs in the chain are for the same txq
82 */
83static void try_bulk_dequeue_skb_slow(struct Qdisc *q,
84 struct sk_buff *skb,
85 int *packets)
86{
87 int mapping = skb_get_queue_mapping(skb);
88 struct sk_buff *nskb;
89 int cnt = 0;
90
91 do {
92 nskb = q->dequeue(q);
93 if (!nskb)
94 break;
95 if (unlikely(skb_get_queue_mapping(nskb) != mapping)) {
96 q->skb_bad_txq = nskb;
97 qdisc_qstats_backlog_inc(q, nskb);
98 q->q.qlen++;
99 break;
100 }
101 skb->next = nskb;
102 skb = nskb;
103 } while (++cnt < 8);
104 (*packets) += cnt;
105 skb->next = NULL;
106}
107
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200108/* Note that dequeue_skb can possibly return a SKB list (via skb->next).
109 * A requeued skb (via q->gso_skb) can also be a SKB list.
110 */
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200111static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
112 int *packets)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700113{
Jarek Poplawski554794d2008-10-06 09:54:39 -0700114 struct sk_buff *skb = q->gso_skb;
Eric Dumazet1abbe132012-12-11 15:54:33 +0000115 const struct netdev_queue *txq = q->dev_queue;
Jarek Poplawski554794d2008-10-06 09:54:39 -0700116
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200117 *packets = 1;
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700118 if (unlikely(skb)) {
Eric Dumazet4d202a02016-06-21 23:16:52 -0700119 /* skb in gso_skb were already validated */
120 *validate = false;
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700121 /* check the reason of requeuing without tx lock first */
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200122 txq = skb_get_tx_queue(txq->dev, skb);
Tom Herbert734664982011-11-28 16:32:44 +0000123 if (!netif_xmit_frozen_or_stopped(txq)) {
Jarek Poplawski62523522008-10-06 10:41:50 -0700124 q->gso_skb = NULL;
WANG Conga27758f2016-06-03 15:05:57 -0700125 qdisc_qstats_backlog_dec(q, skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000126 q->q.qlen--;
127 } else
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700128 skb = NULL;
Eric Dumazet4d202a02016-06-21 23:16:52 -0700129 return skb;
130 }
131 *validate = true;
132 skb = q->skb_bad_txq;
133 if (unlikely(skb)) {
134 /* check the reason of requeuing without tx lock first */
135 txq = skb_get_tx_queue(txq->dev, skb);
136 if (!netif_xmit_frozen_or_stopped(txq)) {
137 q->skb_bad_txq = NULL;
138 qdisc_qstats_backlog_dec(q, skb);
139 q->q.qlen--;
140 goto bulk;
David S. Miller50cbe9a2014-08-30 19:13:51 -0700141 }
Eric Dumazet4d202a02016-06-21 23:16:52 -0700142 return NULL;
143 }
144 if (!(q->flags & TCQ_F_ONETXQUEUE) ||
145 !netif_xmit_frozen_or_stopped(txq))
146 skb = q->dequeue(q);
147 if (skb) {
148bulk:
149 if (qdisc_may_bulk(q))
150 try_bulk_dequeue_skb(q, skb, txq, packets);
151 else
152 try_bulk_dequeue_skb_slow(q, skb, packets);
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700153 }
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700154 return skb;
155}
156
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900157/*
Jesper Dangaard Brouer10770bc2014-09-02 16:35:33 +0200158 * Transmit possibly several skbs, and handle the return status as
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700159 * required. Owning running seqcount bit guarantees that
Jesper Dangaard Brouer10770bc2014-09-02 16:35:33 +0200160 * only one CPU can execute this function.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700161 *
162 * Returns to the caller:
163 * 0 - queue is empty or throttled.
164 * >0 - queue is not empty.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700165 */
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000166int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
167 struct net_device *dev, struct netdev_queue *txq,
Eric Dumazet55a93b32014-10-03 15:31:07 -0700168 spinlock_t *root_lock, bool validate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800170 int ret = NETDEV_TX_BUSY;
David S. Miller7698b4f2008-07-16 01:42:40 -0700171
172 /* And release qdisc */
173 spin_unlock(root_lock);
Herbert Xud90df3a2007-05-10 04:55:14 -0700174
Eric Dumazet55a93b32014-10-03 15:31:07 -0700175 /* Note that we validate skb (GSO, checksum, ...) outside of locks */
176 if (validate)
177 skb = validate_xmit_skb_list(skb, dev);
Patrick McHardy572a9d72009-11-10 06:14:14 +0000178
Lars Persson3dcd493fb2016-04-12 08:45:52 +0200179 if (likely(skb)) {
Eric Dumazet55a93b32014-10-03 15:31:07 -0700180 HARD_TX_LOCK(dev, txq, smp_processor_id());
181 if (!netif_xmit_frozen_or_stopped(txq))
182 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700183
Eric Dumazet55a93b32014-10-03 15:31:07 -0700184 HARD_TX_UNLOCK(dev, txq);
Lars Persson3dcd493fb2016-04-12 08:45:52 +0200185 } else {
Eric Dumazet52fbb292016-06-09 07:45:11 -0700186 spin_lock(root_lock);
Lars Persson3dcd493fb2016-04-12 08:45:52 +0200187 return qdisc_qlen(q);
Eric Dumazet55a93b32014-10-03 15:31:07 -0700188 }
Eric Dumazet52fbb292016-06-09 07:45:11 -0700189 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700190
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000191 if (dev_xmit_complete(ret)) {
192 /* Driver sent out skb successfully or skb was consumed */
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700193 ret = qdisc_qlen(q);
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000194 } else {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700195 /* Driver returned NETDEV_TX_BUSY - requeue skb */
Joe Perchese87cc472012-05-13 21:56:26 +0000196 if (unlikely(ret != NETDEV_TX_BUSY))
197 net_warn_ratelimited("BUG %s code %d qlen %d\n",
198 dev->name, ret, q->q.qlen);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700199
David S. Miller37437bb2008-07-16 02:15:04 -0700200 ret = dev_requeue_skb(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700201 }
202
Tom Herbert734664982011-11-28 16:32:44 +0000203 if (ret && netif_xmit_frozen_or_stopped(txq))
David S. Miller37437bb2008-07-16 02:15:04 -0700204 ret = 0;
205
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700206 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000209/*
210 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
211 *
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700212 * running seqcount guarantees only one CPU can process
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000213 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
214 * this queue.
215 *
216 * netif_tx_lock serializes accesses to device driver.
217 *
218 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
219 * if one is grabbed, another must be free.
220 *
221 * Note, that this procedure can be called by a watchdog timer
222 *
223 * Returns to the caller:
224 * 0 - queue is empty or throttled.
225 * >0 - queue is not empty.
226 *
227 */
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200228static inline int qdisc_restart(struct Qdisc *q, int *packets)
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000229{
230 struct netdev_queue *txq;
231 struct net_device *dev;
232 spinlock_t *root_lock;
233 struct sk_buff *skb;
Eric Dumazet55a93b32014-10-03 15:31:07 -0700234 bool validate;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000235
236 /* Dequeue packet */
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200237 skb = dequeue_skb(q, &validate, packets);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000238 if (unlikely(!skb))
239 return 0;
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200240
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000241 root_lock = qdisc_lock(q);
242 dev = qdisc_dev(q);
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200243 txq = skb_get_tx_queue(dev, skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000244
Eric Dumazet55a93b32014-10-03 15:31:07 -0700245 return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000246}
247
David S. Miller37437bb2008-07-16 02:15:04 -0700248void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700249{
jamald5b8aa12011-06-26 08:13:54 +0000250 int quota = weight_p;
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200251 int packets;
Herbert Xu2ba25062008-03-28 16:25:26 -0700252
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200253 while (qdisc_restart(q, &packets)) {
Herbert Xu2ba25062008-03-28 16:25:26 -0700254 /*
jamald5b8aa12011-06-26 08:13:54 +0000255 * Ordered by possible occurrence: Postpone processing if
256 * 1. we've exceeded packet quota
257 * 2. another process needs the CPU;
Herbert Xu2ba25062008-03-28 16:25:26 -0700258 */
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200259 quota -= packets;
260 if (quota <= 0 || need_resched()) {
David S. Miller37437bb2008-07-16 02:15:04 -0700261 __netif_schedule(q);
Herbert Xu2ba25062008-03-28 16:25:26 -0700262 break;
263 }
264 }
Herbert Xu48d83322006-06-19 23:57:59 -0700265
Eric Dumazetbc135b22010-06-02 03:23:51 -0700266 qdisc_run_end(q);
Herbert Xu48d83322006-06-19 23:57:59 -0700267}
268
Eric Dumazet9d214932009-05-17 20:55:16 -0700269unsigned long dev_trans_start(struct net_device *dev)
270{
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200271 unsigned long val, res;
Eric Dumazet9d214932009-05-17 20:55:16 -0700272 unsigned int i;
273
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200274 if (is_vlan_dev(dev))
275 dev = vlan_dev_real_dev(dev);
Florian Westphal9b366272016-05-03 16:33:14 +0200276 res = netdev_get_tx_queue(dev, 0)->trans_start;
277 for (i = 1; i < dev->num_tx_queues; i++) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700278 val = netdev_get_tx_queue(dev, i)->trans_start;
279 if (val && time_after(val, res))
280 res = val;
281 }
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200282
Eric Dumazet9d214932009-05-17 20:55:16 -0700283 return res;
284}
285EXPORT_SYMBOL(dev_trans_start);
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287static void dev_watchdog(unsigned long arg)
288{
289 struct net_device *dev = (struct net_device *)arg;
290
Herbert Xu932ff272006-06-09 12:20:56 -0700291 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700292 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 if (netif_device_present(dev) &&
294 netif_running(dev) &&
295 netif_carrier_ok(dev)) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700296 int some_queue_timedout = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700297 unsigned int i;
Eric Dumazet9d214932009-05-17 20:55:16 -0700298 unsigned long trans_start;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700299
David S. Millere8a04642008-07-17 00:34:19 -0700300 for (i = 0; i < dev->num_tx_queues; i++) {
301 struct netdev_queue *txq;
302
303 txq = netdev_get_tx_queue(dev, i);
Florian Westphal9b366272016-05-03 16:33:14 +0200304 trans_start = txq->trans_start;
Tom Herbert734664982011-11-28 16:32:44 +0000305 if (netif_xmit_stopped(txq) &&
Eric Dumazet9d214932009-05-17 20:55:16 -0700306 time_after(jiffies, (trans_start +
307 dev->watchdog_timeo))) {
308 some_queue_timedout = 1;
david decotignyccf5ff62011-11-16 12:15:10 +0000309 txq->trans_timeout++;
David S. Millere8a04642008-07-17 00:34:19 -0700310 break;
311 }
312 }
313
Eric Dumazet9d214932009-05-17 20:55:16 -0700314 if (some_queue_timedout) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700315 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
David S. Miller3019de12011-06-06 16:41:33 -0700316 dev->name, netdev_drivername(dev), i);
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800317 dev->netdev_ops->ndo_tx_timeout(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 }
David S. Millere8a04642008-07-17 00:34:19 -0700319 if (!mod_timer(&dev->watchdog_timer,
320 round_jiffies(jiffies +
321 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 dev_hold(dev);
323 }
324 }
Herbert Xu932ff272006-06-09 12:20:56 -0700325 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 dev_put(dev);
328}
329
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330void __netdev_watchdog_up(struct net_device *dev)
331{
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800332 if (dev->netdev_ops->ndo_tx_timeout) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (dev->watchdog_timeo <= 0)
334 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700335 if (!mod_timer(&dev->watchdog_timer,
336 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 dev_hold(dev);
338 }
339}
340
341static void dev_watchdog_up(struct net_device *dev)
342{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346static void dev_watchdog_down(struct net_device *dev)
347{
Herbert Xu932ff272006-06-09 12:20:56 -0700348 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800350 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700351 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352}
353
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700354/**
355 * netif_carrier_on - set carrier
356 * @dev: network device
357 *
358 * Device has detected that carrier.
359 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700360void netif_carrier_on(struct net_device *dev)
361{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700362 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
David S. Millerb4730012008-11-19 15:33:54 -0800363 if (dev->reg_state == NETREG_UNINITIALIZED)
364 return;
david decotigny2d3b4792014-03-29 09:48:35 -0700365 atomic_inc(&dev->carrier_changes);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700366 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700367 if (netif_running(dev))
368 __netdev_watchdog_up(dev);
369 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700370}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800371EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700372
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700373/**
374 * netif_carrier_off - clear carrier
375 * @dev: network device
376 *
377 * Device has detected loss of carrier.
378 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700379void netif_carrier_off(struct net_device *dev)
380{
David S. Millerb4730012008-11-19 15:33:54 -0800381 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
382 if (dev->reg_state == NETREG_UNINITIALIZED)
383 return;
david decotigny2d3b4792014-03-29 09:48:35 -0700384 atomic_inc(&dev->carrier_changes);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700385 linkwatch_fire_event(dev);
David S. Millerb4730012008-11-19 15:33:54 -0800386 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700387}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800388EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
391 under all circumstances. It is difficult to invent anything faster or
392 cheaper.
393 */
394
Eric Dumazet520ac302016-06-21 23:16:49 -0700395static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
396 struct sk_buff **to_free)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397{
Eric Dumazet520ac302016-06-21 23:16:49 -0700398 __qdisc_drop(skb, to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 return NET_XMIT_CN;
400}
401
Yang Yingliang82d567c2013-12-10 20:55:31 +0800402static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
404 return NULL;
405}
406
Eric Dumazet20fea082007-11-14 01:44:41 -0800407struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 .id = "noop",
409 .priv_size = 0,
410 .enqueue = noop_enqueue,
411 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700412 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 .owner = THIS_MODULE,
414};
415
David S. Miller7698b4f2008-07-16 01:42:40 -0700416static struct netdev_queue noop_netdev_queue = {
David S. Miller7698b4f2008-07-16 01:42:40 -0700417 .qdisc = &noop_qdisc,
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700418 .qdisc_sleeping = &noop_qdisc,
David S. Miller7698b4f2008-07-16 01:42:40 -0700419};
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421struct Qdisc noop_qdisc = {
422 .enqueue = noop_enqueue,
423 .dequeue = noop_dequeue,
424 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900425 .ops = &noop_qdisc_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 .list = LIST_HEAD_INIT(noop_qdisc.list),
David S. Miller83874002008-07-17 00:53:03 -0700427 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700428 .dev_queue = &noop_netdev_queue,
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700429 .running = SEQCNT_ZERO(noop_qdisc.running),
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000430 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800432EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Phil Sutterd66d6c32015-08-27 21:21:38 +0200434static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt)
435{
436 /* register_qdisc() assigns a default of noop_enqueue if unset,
437 * but __dev_queue_xmit() treats noqueue only as such
438 * if this is NULL - so clear it here. */
439 qdisc->enqueue = NULL;
440 return 0;
441}
442
443struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 .id = "noqueue",
445 .priv_size = 0,
Phil Sutterd66d6c32015-08-27 21:21:38 +0200446 .init = noqueue_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 .enqueue = noop_enqueue,
448 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700449 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 .owner = THIS_MODULE,
451};
452
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000453static const u8 prio2band[TC_PRIO_MAX + 1] = {
454 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
455};
Thomas Graf321090e2005-06-18 22:58:35 -0700456
David S. Millerd3678b42008-07-21 09:56:13 -0700457/* 3-band FIFO queue: old style, but should be a bit faster than
458 generic prio+fifo combination.
459 */
460
461#define PFIFO_FAST_BANDS 3
462
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000463/*
464 * Private data for a pfifo_fast scheduler containing:
465 * - queues for the three band
466 * - bitmap indicating which of the bands contain skbs
467 */
468struct pfifo_fast_priv {
469 u32 bitmap;
470 struct sk_buff_head q[PFIFO_FAST_BANDS];
471};
472
473/*
474 * Convert a bitmap to the first band number where an skb is queued, where:
475 * bitmap=0 means there are no skbs on any band.
476 * bitmap=1 means there is an skb on band 0.
477 * bitmap=7 means there are skbs on all 3 bands, etc.
478 */
479static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
480
481static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
482 int band)
David S. Millerd3678b42008-07-21 09:56:13 -0700483{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000484 return priv->q + band;
David S. Millerd3678b42008-07-21 09:56:13 -0700485}
486
Eric Dumazet520ac302016-06-21 23:16:49 -0700487static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
488 struct sk_buff **to_free)
David S. Millerd3678b42008-07-21 09:56:13 -0700489{
Krishna Kumara453e062009-08-30 22:20:28 -0700490 if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
491 int band = prio2band[skb->priority & TC_PRIO_MAX];
492 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
493 struct sk_buff_head *list = band2list(priv, band);
David S. Millerd3678b42008-07-21 09:56:13 -0700494
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000495 priv->bitmap |= (1 << band);
David S. Millerd3678b42008-07-21 09:56:13 -0700496 qdisc->q.qlen++;
Thomas Graf821d24a2005-06-18 22:58:15 -0700497 return __qdisc_enqueue_tail(skb, qdisc, list);
David S. Millerd3678b42008-07-21 09:56:13 -0700498 }
Thomas Graf821d24a2005-06-18 22:58:15 -0700499
Eric Dumazet520ac302016-06-21 23:16:49 -0700500 return qdisc_drop(skb, qdisc, to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501}
502
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000503static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000505 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
506 int band = bitmap2band[priv->bitmap];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000508 if (likely(band >= 0)) {
509 struct sk_buff_head *list = band2list(priv, band);
510 struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
511
512 qdisc->q.qlen--;
513 if (skb_queue_empty(list))
514 priv->bitmap &= ~(1 << band);
515
516 return skb;
David S. Millerd3678b42008-07-21 09:56:13 -0700517 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700518
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return NULL;
520}
521
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000522static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700523{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000524 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
525 int band = bitmap2band[priv->bitmap];
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700526
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000527 if (band >= 0) {
528 struct sk_buff_head *list = band2list(priv, band);
529
530 return skb_peek(list);
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700531 }
532
533 return NULL;
534}
535
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000536static void pfifo_fast_reset(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
David S. Millerd3678b42008-07-21 09:56:13 -0700538 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000539 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700540
541 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Eric Dumazet1b5c5492016-06-13 20:21:50 -0700542 __qdisc_reset_queue(band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700543
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000544 priv->bitmap = 0;
Thomas Graf821d24a2005-06-18 22:58:15 -0700545 qdisc->qstats.backlog = 0;
David S. Millerd3678b42008-07-21 09:56:13 -0700546 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548
David S. Millerd3678b42008-07-21 09:56:13 -0700549static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
550{
551 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
552
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000553 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
David S. Miller1b34ec42012-03-29 05:11:39 -0400554 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
555 goto nla_put_failure;
David S. Millerd3678b42008-07-21 09:56:13 -0700556 return skb->len;
557
558nla_put_failure:
559 return -1;
560}
561
562static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
563{
564 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000565 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700566
567 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Eric Dumazetab34f642014-09-17 08:05:05 -0700568 __skb_queue_head_init(band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700569
Eric Dumazet23624932011-01-21 16:26:09 -0800570 /* Can by-pass the queue discipline */
571 qdisc->flags |= TCQ_F_CAN_BYPASS;
David S. Millerd3678b42008-07-21 09:56:13 -0700572 return 0;
573}
574
David S. Miller6ec1c692009-09-06 01:58:51 -0700575struct Qdisc_ops pfifo_fast_ops __read_mostly = {
David S. Millerd3678b42008-07-21 09:56:13 -0700576 .id = "pfifo_fast",
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000577 .priv_size = sizeof(struct pfifo_fast_priv),
David S. Millerd3678b42008-07-21 09:56:13 -0700578 .enqueue = pfifo_fast_enqueue,
579 .dequeue = pfifo_fast_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700580 .peek = pfifo_fast_peek,
David S. Millerd3678b42008-07-21 09:56:13 -0700581 .init = pfifo_fast_init,
582 .reset = pfifo_fast_reset,
583 .dump = pfifo_fast_dump,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 .owner = THIS_MODULE,
585};
Eric Dumazet1f27cde2016-03-02 08:21:43 -0800586EXPORT_SYMBOL(pfifo_fast_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000588static struct lock_class_key qdisc_tx_busylock;
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700589static struct lock_class_key qdisc_running_key;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000590
David S. Miller5ce2d482008-07-08 17:06:30 -0700591struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
stephen hemmingerd2a7f262013-08-31 10:15:50 -0700592 const struct Qdisc_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
594 void *p;
595 struct Qdisc *sch;
Eric Dumazetd2760552011-03-03 11:10:02 -0800596 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
Thomas Graf3d54b822005-07-05 14:15:09 -0700597 int err = -ENOBUFS;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000598 struct net_device *dev = dev_queue->dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Eric Dumazetf2cd2d32010-11-29 08:14:37 +0000600 p = kzalloc_node(size, GFP_KERNEL,
601 netdev_queue_numa_node_read(dev_queue));
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (!p)
Thomas Graf3d54b822005-07-05 14:15:09 -0700604 goto errout;
Thomas Graf3d54b822005-07-05 14:15:09 -0700605 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
Eric Dumazetd2760552011-03-03 11:10:02 -0800606 /* if we got non aligned memory, ask more and do alignment ourself */
607 if (sch != p) {
608 kfree(p);
609 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
610 netdev_queue_numa_node_read(dev_queue));
611 if (!p)
612 goto errout;
613 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
614 sch->padded = (char *) sch - (char *) p;
615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 INIT_LIST_HEAD(&sch->list);
617 skb_queue_head_init(&sch->q);
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000618
Eric Dumazet79640a42010-06-02 05:09:29 -0700619 spin_lock_init(&sch->busylock);
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000620 lockdep_set_class(&sch->busylock,
621 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
622
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700623 seqcount_init(&sch->running);
624 lockdep_set_class(&sch->running,
625 dev->qdisc_running_key ?: &qdisc_running_key);
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 sch->ops = ops;
628 sch->enqueue = ops->enqueue;
629 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700630 sch->dev_queue = dev_queue;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000631 dev_hold(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 atomic_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700633
634 return sch;
635errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700636 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700637}
638
Changli Gao3511c912010-10-16 13:04:08 +0000639struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
stephen hemmingerd2a7f262013-08-31 10:15:50 -0700640 const struct Qdisc_ops *ops,
641 unsigned int parentid)
Thomas Graf3d54b822005-07-05 14:15:09 -0700642{
643 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900644
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700645 if (!try_module_get(ops->owner))
646 goto errout;
647
David S. Miller5ce2d482008-07-08 17:06:30 -0700648 sch = qdisc_alloc(dev_queue, ops);
Thomas Graf3d54b822005-07-05 14:15:09 -0700649 if (IS_ERR(sch))
650 goto errout;
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800651 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (!ops->init || ops->init(sch, NULL) == 0)
654 return sch;
655
Thomas Graf0fbbeb12005-08-23 10:12:44 -0700656 qdisc_destroy(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700657errout:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return NULL;
659}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800660EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
David S. Miller5fb66222008-08-02 20:02:43 -0700662/* Under qdisc_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664void qdisc_reset(struct Qdisc *qdisc)
665{
Eric Dumazet20fea082007-11-14 01:44:41 -0800666 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 if (ops->reset)
669 ops->reset(qdisc);
Jarek Poplawski67305eb2008-11-03 02:52:50 -0800670
Eric Dumazet4d202a02016-06-21 23:16:52 -0700671 kfree_skb(qdisc->skb_bad_txq);
672 qdisc->skb_bad_txq = NULL;
673
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000674 if (qdisc->gso_skb) {
Jesper Dangaard Brouer3f3c7ee2014-09-03 12:12:50 +0200675 kfree_skb_list(qdisc->gso_skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000676 qdisc->gso_skb = NULL;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000677 }
Eric Dumazet4d202a02016-06-21 23:16:52 -0700678 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800680EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Eric Dumazet5d944c62010-03-31 07:06:04 +0000682static void qdisc_rcu_free(struct rcu_head *head)
683{
684 struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
685
John Fastabend73c20a82016-01-05 09:11:36 -0800686 if (qdisc_is_percpu_stats(qdisc)) {
John Fastabend22e0f8b2014-09-28 11:52:56 -0700687 free_percpu(qdisc->cpu_bstats);
John Fastabend73c20a82016-01-05 09:11:36 -0800688 free_percpu(qdisc->cpu_qstats);
689 }
John Fastabend22e0f8b2014-09-28 11:52:56 -0700690
Eric Dumazet5d944c62010-03-31 07:06:04 +0000691 kfree((char *) qdisc - qdisc->padded);
692}
693
David S. Miller1e0d5a52008-08-17 22:31:26 -0700694void qdisc_destroy(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Eric Dumazet20fea082007-11-14 01:44:41 -0800696 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
David S. Miller1e0d5a52008-08-17 22:31:26 -0700698 if (qdisc->flags & TCQ_F_BUILTIN ||
699 !atomic_dec_and_test(&qdisc->refcnt))
700 return;
701
David S. Miller3a682fb2008-07-20 18:13:01 -0700702#ifdef CONFIG_NET_SCHED
Jarek Poplawskif6e0b232008-08-22 03:24:05 -0700703 qdisc_list_del(qdisc);
704
Eric Dumazeta2da5702011-01-20 03:48:19 +0000705 qdisc_put_stab(rtnl_dereference(qdisc->stab));
David S. Miller3a682fb2008-07-20 18:13:01 -0700706#endif
Patrick McHardy85670cc2006-09-27 16:45:45 -0700707 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
Patrick McHardy85670cc2006-09-27 16:45:45 -0700708 if (ops->reset)
709 ops->reset(qdisc);
710 if (ops->destroy)
711 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Patrick McHardy85670cc2006-09-27 16:45:45 -0700713 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -0700714 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -0700715
Jesper Dangaard Brouer3f3c7ee2014-09-03 12:12:50 +0200716 kfree_skb_list(qdisc->gso_skb);
Eric Dumazet4d202a02016-06-21 23:16:52 -0700717 kfree_skb(qdisc->skb_bad_txq);
Eric Dumazet5d944c62010-03-31 07:06:04 +0000718 /*
719 * gen_estimator est_timer() might access qdisc->q.lock,
720 * wait a RCU grace period before freeing qdisc.
721 */
722 call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
David S. Miller8a34c5d2008-07-17 00:47:45 -0700723}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800724EXPORT_SYMBOL(qdisc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Patrick McHardy589983c2009-09-04 06:41:20 +0000726/* Attach toplevel qdisc to device queue. */
727struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
728 struct Qdisc *qdisc)
729{
730 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
731 spinlock_t *root_lock;
732
733 root_lock = qdisc_lock(oqdisc);
734 spin_lock_bh(root_lock);
735
736 /* Prune old scheduler */
737 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
738 qdisc_reset(oqdisc);
739
740 /* ... and graft new one */
741 if (qdisc == NULL)
742 qdisc = &noop_qdisc;
743 dev_queue->qdisc_sleeping = qdisc;
744 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
745
746 spin_unlock_bh(root_lock);
747
748 return oqdisc;
749}
John Fastabendb8970f02011-01-17 08:06:09 +0000750EXPORT_SYMBOL(dev_graft_qdisc);
Patrick McHardy589983c2009-09-04 06:41:20 +0000751
David S. Millere8a04642008-07-17 00:34:19 -0700752static void attach_one_default_qdisc(struct net_device *dev,
753 struct netdev_queue *dev_queue,
754 void *_unused)
755{
Phil Sutter3e692f22015-08-27 21:21:39 +0200756 struct Qdisc *qdisc;
757 const struct Qdisc_ops *ops = default_qdisc_ops;
David S. Millere8a04642008-07-17 00:34:19 -0700758
Phil Sutter3e692f22015-08-27 21:21:39 +0200759 if (dev->priv_flags & IFF_NO_QUEUE)
760 ops = &noqueue_qdisc_ops;
761
762 qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT);
763 if (!qdisc) {
764 netdev_info(dev, "activation failed\n");
765 return;
David S. Millere8a04642008-07-17 00:34:19 -0700766 }
Phil Sutter3e692f22015-08-27 21:21:39 +0200767 if (!netif_is_multiqueue(dev))
Eric Dumazet4eaf3b82015-12-01 20:08:51 -0800768 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
David S. Millere8a04642008-07-17 00:34:19 -0700769 dev_queue->qdisc_sleeping = qdisc;
770}
771
David S. Miller6ec1c692009-09-06 01:58:51 -0700772static void attach_default_qdiscs(struct net_device *dev)
773{
774 struct netdev_queue *txq;
775 struct Qdisc *qdisc;
776
777 txq = netdev_get_tx_queue(dev, 0);
778
Phil Sutter4b469952015-08-13 19:01:07 +0200779 if (!netif_is_multiqueue(dev) ||
Phil Sutter4b469952015-08-13 19:01:07 +0200780 dev->priv_flags & IFF_NO_QUEUE) {
David S. Miller6ec1c692009-09-06 01:58:51 -0700781 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
782 dev->qdisc = txq->qdisc_sleeping;
783 atomic_inc(&dev->qdisc->refcnt);
784 } else {
Changli Gao3511c912010-10-16 13:04:08 +0000785 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
David S. Miller6ec1c692009-09-06 01:58:51 -0700786 if (qdisc) {
David S. Miller6ec1c692009-09-06 01:58:51 -0700787 dev->qdisc = qdisc;
Eric Dumazete57a7842013-12-12 15:41:56 -0800788 qdisc->ops->attach(qdisc);
David S. Miller6ec1c692009-09-06 01:58:51 -0700789 }
790 }
791}
792
David S. Millere8a04642008-07-17 00:34:19 -0700793static void transition_one_qdisc(struct net_device *dev,
794 struct netdev_queue *dev_queue,
795 void *_need_watchdog)
796{
David S. Miller83874002008-07-17 00:53:03 -0700797 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700798 int *need_watchdog_p = _need_watchdog;
799
David S. Millera9312ae2008-08-17 21:51:03 -0700800 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
801 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
802
David S. Miller83874002008-07-17 00:53:03 -0700803 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
Phil Sutter3e692f22015-08-27 21:21:39 +0200804 if (need_watchdog_p) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700805 dev_queue->trans_start = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700806 *need_watchdog_p = 1;
Eric Dumazet9d214932009-05-17 20:55:16 -0700807 }
David S. Millere8a04642008-07-17 00:34:19 -0700808}
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810void dev_activate(struct net_device *dev)
811{
David S. Millere8a04642008-07-17 00:34:19 -0700812 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -0700813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 /* No queueing discipline is attached to device;
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700815 * create default one for devices, which need queueing
816 * and noqueue_qdisc for virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 */
818
David S. Miller6ec1c692009-09-06 01:58:51 -0700819 if (dev->qdisc == &noop_qdisc)
820 attach_default_qdiscs(dev);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000821
Tommy S. Christensencacaddf2005-05-03 16:18:52 -0700822 if (!netif_carrier_ok(dev))
823 /* Delay activation until next carrier-on event */
824 return;
825
David S. Millere8a04642008-07-17 00:34:19 -0700826 need_watchdog = 0;
827 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
Eric Dumazet24824a02010-10-02 06:11:55 +0000828 if (dev_ingress_queue(dev))
829 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
David S. Millere8a04642008-07-17 00:34:19 -0700830
831 if (need_watchdog) {
Florian Westphal860e9532016-05-03 16:33:13 +0200832 netif_trans_update(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 dev_watchdog_up(dev);
834 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700835}
John Fastabendb8970f02011-01-17 08:06:09 +0000836EXPORT_SYMBOL(dev_activate);
David S. Millerb0e1e642008-07-08 17:42:10 -0700837
David S. Millere8a04642008-07-17 00:34:19 -0700838static void dev_deactivate_queue(struct net_device *dev,
839 struct netdev_queue *dev_queue,
840 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -0700841{
David S. Millere8a04642008-07-17 00:34:19 -0700842 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -0700843 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700844
John Fastabend46e5da42014-09-12 20:04:52 -0700845 qdisc = rtnl_dereference(dev_queue->qdisc);
David S. Millerb0e1e642008-07-08 17:42:10 -0700846 if (qdisc) {
David S. Miller83874002008-07-17 00:53:03 -0700847 spin_lock_bh(qdisc_lock(qdisc));
848
David S. Millera9312ae2008-08-17 21:51:03 -0700849 if (!(qdisc->flags & TCQ_F_BUILTIN))
850 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
851
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700852 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700853 qdisc_reset(qdisc);
David S. Millerd3b753d2008-07-15 20:14:35 -0700854
David S. Miller83874002008-07-17 00:53:03 -0700855 spin_unlock_bh(qdisc_lock(qdisc));
David S. Millerb0e1e642008-07-08 17:42:10 -0700856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857}
858
David S. Miller4335cd22008-08-17 21:58:07 -0700859static bool some_qdisc_is_busy(struct net_device *dev)
David S. Millere8a04642008-07-17 00:34:19 -0700860{
861 unsigned int i;
862
863 for (i = 0; i < dev->num_tx_queues; i++) {
864 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -0700865 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -0700866 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -0700867 int val;
868
869 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millerb9a3b112008-08-13 15:18:38 -0700870 q = dev_queue->qdisc_sleeping;
David S. Miller5fb66222008-08-02 20:02:43 -0700871 root_lock = qdisc_lock(q);
David S. Millere8a04642008-07-17 00:34:19 -0700872
David S. Miller4335cd22008-08-17 21:58:07 -0700873 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700874
Eric Dumazetbc135b22010-06-02 03:23:51 -0700875 val = (qdisc_is_running(q) ||
David S. Millerb9a3b112008-08-13 15:18:38 -0700876 test_bit(__QDISC_STATE_SCHED, &q->state));
David S. Millere8a04642008-07-17 00:34:19 -0700877
David S. Miller4335cd22008-08-17 21:58:07 -0700878 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700879
880 if (val)
881 return true;
882 }
883 return false;
884}
885
Eric Dumazet31376632011-05-19 23:42:09 +0000886/**
887 * dev_deactivate_many - deactivate transmissions on several devices
888 * @head: list of devices to deactivate
889 *
890 * This function returns only when all outstanding transmissions
891 * have completed, unless all devices are in dismantle phase.
892 */
Octavian Purdila44345722010-12-13 12:44:07 +0000893void dev_deactivate_many(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
Octavian Purdila44345722010-12-13 12:44:07 +0000895 struct net_device *dev;
Eric Dumazet31376632011-05-19 23:42:09 +0000896 bool sync_needed = false;
Herbert Xu41a23b02007-05-10 14:12:47 -0700897
Eric W. Biederman5cde2822013-10-05 19:26:05 -0700898 list_for_each_entry(dev, head, close_list) {
Octavian Purdila44345722010-12-13 12:44:07 +0000899 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
900 &noop_qdisc);
901 if (dev_ingress_queue(dev))
902 dev_deactivate_queue(dev, dev_ingress_queue(dev),
903 &noop_qdisc);
904
905 dev_watchdog_down(dev);
Eric Dumazet31376632011-05-19 23:42:09 +0000906 sync_needed |= !dev->dismantle;
Octavian Purdila44345722010-12-13 12:44:07 +0000907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Eric Dumazet31376632011-05-19 23:42:09 +0000909 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
910 * This is avoided if all devices are in dismantle phase :
911 * Caller will call synchronize_net() for us
912 */
913 if (sync_needed)
914 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Herbert Xud4828d82006-06-22 02:28:18 -0700916 /* Wait for outstanding qdisc_run calls. */
Eric W. Biederman5cde2822013-10-05 19:26:05 -0700917 list_for_each_entry(dev, head, close_list)
Octavian Purdila44345722010-12-13 12:44:07 +0000918 while (some_qdisc_is_busy(dev))
919 yield();
920}
921
922void dev_deactivate(struct net_device *dev)
923{
924 LIST_HEAD(single);
925
Eric W. Biederman5cde2822013-10-05 19:26:05 -0700926 list_add(&dev->close_list, &single);
Octavian Purdila44345722010-12-13 12:44:07 +0000927 dev_deactivate_many(&single);
Eric W. Biederman5f04d502011-02-20 11:49:45 -0800928 list_del(&single);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
John Fastabendb8970f02011-01-17 08:06:09 +0000930EXPORT_SYMBOL(dev_deactivate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
David S. Millerb0e1e642008-07-08 17:42:10 -0700932static void dev_init_scheduler_queue(struct net_device *dev,
933 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -0700934 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -0700935{
David S. Millere8a04642008-07-17 00:34:19 -0700936 struct Qdisc *qdisc = _qdisc;
937
John Fastabend46e5da42014-09-12 20:04:52 -0700938 rcu_assign_pointer(dev_queue->qdisc, qdisc);
David S. Millerb0e1e642008-07-08 17:42:10 -0700939 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700940}
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942void dev_init_scheduler(struct net_device *dev)
943{
Patrick McHardyaf356af2009-09-04 06:41:18 +0000944 dev->qdisc = &noop_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -0700945 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000946 if (dev_ingress_queue(dev))
947 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800949 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950}
951
David S. Millere8a04642008-07-17 00:34:19 -0700952static void shutdown_scheduler_queue(struct net_device *dev,
953 struct netdev_queue *dev_queue,
954 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
David S. Millerb0e1e642008-07-08 17:42:10 -0700956 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700957 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
David S. Millerb0e1e642008-07-08 17:42:10 -0700959 if (qdisc) {
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700960 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700961 dev_queue->qdisc_sleeping = qdisc_default;
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 qdisc_destroy(qdisc);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900964 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700965}
966
967void dev_shutdown(struct net_device *dev)
968{
David S. Millere8a04642008-07-17 00:34:19 -0700969 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000970 if (dev_ingress_queue(dev))
971 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000972 qdisc_destroy(dev->qdisc);
973 dev->qdisc = &noop_qdisc;
974
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700975 WARN_ON(timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976}
Jiri Pirko292f1c72013-02-12 00:12:03 +0000977
Eric Dumazet01cb71d2013-06-02 13:55:05 +0000978void psched_ratecfg_precompute(struct psched_ratecfg *r,
Eric Dumazet3e1e3aa2013-09-19 09:10:03 -0700979 const struct tc_ratespec *conf,
980 u64 rate64)
Jiri Pirko292f1c72013-02-12 00:12:03 +0000981{
Eric Dumazet01cb71d2013-06-02 13:55:05 +0000982 memset(r, 0, sizeof(*r));
983 r->overhead = conf->overhead;
Eric Dumazet3e1e3aa2013-09-19 09:10:03 -0700984 r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
Jesper Dangaard Brouer8a8e3d82013-08-14 23:47:11 +0200985 r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
Jiri Pirko292f1c72013-02-12 00:12:03 +0000986 r->mult = 1;
987 /*
Eric Dumazet130d3d62013-06-06 13:56:19 -0700988 * The deal here is to replace a divide by a reciprocal one
989 * in fast path (a reciprocal divide is a multiply and a shift)
990 *
991 * Normal formula would be :
992 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
993 *
994 * We compute mult/shift to use instead :
995 * time_in_ns = (len * mult) >> shift;
996 *
997 * We try to get the highest possible mult value for accuracy,
998 * but have to make sure no overflows will ever happen.
Jiri Pirko292f1c72013-02-12 00:12:03 +0000999 */
Eric Dumazet130d3d62013-06-06 13:56:19 -07001000 if (r->rate_bytes_ps > 0) {
1001 u64 factor = NSEC_PER_SEC;
Jiri Pirko292f1c72013-02-12 00:12:03 +00001002
Eric Dumazet130d3d62013-06-06 13:56:19 -07001003 for (;;) {
1004 r->mult = div64_u64(factor, r->rate_bytes_ps);
1005 if (r->mult & (1U << 31) || factor & (1ULL << 63))
1006 break;
1007 factor <<= 1;
1008 r->shift++;
1009 }
Jiri Pirko292f1c72013-02-12 00:12:03 +00001010 }
1011}
1012EXPORT_SYMBOL(psched_ratecfg_precompute);