blob: b4c680900d7a8c9b417d1c976ef838f9eb5eb960 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <net/pkt_sched.h>
Eric Dumazet7fee2262010-05-11 23:19:48 +000029#include <net/dst.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31/* Main transmission queue. */
32
Patrick McHardy0463d4a2007-04-16 17:02:10 -070033/* Modifications to data participating in scheduling must be protected with
David S. Miller5fb66222008-08-02 20:02:43 -070034 * qdisc_lock(qdisc) spinlock.
Patrick McHardy0463d4a2007-04-16 17:02:10 -070035 *
36 * The idea is the following:
David S. Millerc7e4f3b2008-07-16 03:22:39 -070037 * - enqueue, dequeue are serialized via qdisc root lock
38 * - ingress filtering is also serialized via qdisc root lock
Patrick McHardy0463d4a2007-04-16 17:02:10 -070039 * - updates to tree and tree walking are only done under the rtnl mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
David S. Miller37437bb2008-07-16 02:15:04 -070042static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070043{
Eric Dumazet7fee2262010-05-11 23:19:48 +000044 skb_dst_force(skb);
Jarek Poplawski62523522008-10-06 10:41:50 -070045 q->gso_skb = skb;
Jarek Poplawski53e91502008-10-08 11:36:22 -070046 q->qstats.requeues++;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +000047 q->q.qlen++; /* it's still part of the queue */
David S. Miller37437bb2008-07-16 02:15:04 -070048 __netif_schedule(q);
Jarek Poplawski62523522008-10-06 10:41:50 -070049
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070050 return 0;
51}
52
David S. Millerd3b753d2008-07-15 20:14:35 -070053static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070054{
Jarek Poplawski554794d2008-10-06 09:54:39 -070055 struct sk_buff *skb = q->gso_skb;
56
Jarek Poplawskiebf05982008-09-22 22:16:23 -070057 if (unlikely(skb)) {
58 struct net_device *dev = qdisc_dev(q);
59 struct netdev_queue *txq;
60
61 /* check the reason of requeuing without tx lock first */
62 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
Eric Dumazet5a0d2262010-11-23 10:42:02 +000063 if (!netif_tx_queue_frozen_or_stopped(txq)) {
Jarek Poplawski62523522008-10-06 10:41:50 -070064 q->gso_skb = NULL;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +000065 q->q.qlen--;
66 } else
Jarek Poplawskiebf05982008-09-22 22:16:23 -070067 skb = NULL;
68 } else {
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070069 skb = q->dequeue(q);
Jarek Poplawskiebf05982008-09-22 22:16:23 -070070 }
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070071
72 return skb;
73}
74
Krishna Kumar6c1361a2007-06-24 19:56:09 -070075static inline int handle_dev_cpu_collision(struct sk_buff *skb,
David S. Miller970565b2008-07-08 23:10:33 -070076 struct netdev_queue *dev_queue,
Krishna Kumar6c1361a2007-06-24 19:56:09 -070077 struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070078{
Krishna Kumar6c1361a2007-06-24 19:56:09 -070079 int ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070080
David S. Millerc773e842008-07-08 23:13:53 -070081 if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -070082 /*
83 * Same CPU holding the lock. It may be a transient
84 * configuration error, when hard_start_xmit() recurses. We
85 * detect it by checking xmit owner and drop the packet when
86 * deadloop is detected. Return OK to try the next skb.
87 */
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070088 kfree_skb(skb);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070089 if (net_ratelimit())
Eric Dumazetcc7ec452011-01-19 19:26:56 +000090 pr_warning("Dead loop on netdevice %s, fix it urgently!\n",
91 dev_queue->dev->name);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070092 ret = qdisc_qlen(q);
93 } else {
94 /*
95 * Another cpu is holding lock, requeue & delay xmits for
96 * some time.
97 */
Eric Dumazetd6d9ca02010-07-19 10:48:49 +000098 __this_cpu_inc(softnet_data.cpu_collision);
David S. Miller37437bb2008-07-16 02:15:04 -070099 ret = dev_requeue_skb(skb, q);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700100 }
101
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700102 return ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700103}
104
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900105/*
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000106 * Transmit one skb, and handle the return status as required. Holding the
107 * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this
108 * function.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700109 *
110 * Returns to the caller:
111 * 0 - queue is empty or throttled.
112 * >0 - queue is not empty.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700113 */
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000114int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
115 struct net_device *dev, struct netdev_queue *txq,
116 spinlock_t *root_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800118 int ret = NETDEV_TX_BUSY;
David S. Miller7698b4f2008-07-16 01:42:40 -0700119
120 /* And release qdisc */
121 spin_unlock(root_lock);
Herbert Xud90df3a2007-05-10 04:55:14 -0700122
David S. Millerc773e842008-07-08 23:13:53 -0700123 HARD_TX_LOCK(dev, txq, smp_processor_id());
Eric Dumazet5a0d2262010-11-23 10:42:02 +0000124 if (!netif_tx_queue_frozen_or_stopped(txq))
David S. Millerfd2ea0a2008-07-17 01:56:23 -0700125 ret = dev_hard_start_xmit(skb, dev, txq);
Patrick McHardy572a9d72009-11-10 06:14:14 +0000126
David S. Millerc773e842008-07-08 23:13:53 -0700127 HARD_TX_UNLOCK(dev, txq);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700128
David S. Miller7698b4f2008-07-16 01:42:40 -0700129 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700130
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000131 if (dev_xmit_complete(ret)) {
132 /* Driver sent out skb successfully or skb was consumed */
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700133 ret = qdisc_qlen(q);
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000134 } else if (ret == NETDEV_TX_LOCKED) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700135 /* Driver try lock failed */
David S. Millereb6aafe2008-07-08 23:12:38 -0700136 ret = handle_dev_cpu_collision(skb, txq, q);
Jarek Poplawski9a1654b2009-11-15 07:20:12 +0000137 } else {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700138 /* Driver returned NETDEV_TX_BUSY - requeue skb */
139 if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000140 pr_warning("BUG %s code %d qlen %d\n",
141 dev->name, ret, q->q.qlen);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700142
David S. Miller37437bb2008-07-16 02:15:04 -0700143 ret = dev_requeue_skb(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700144 }
145
Eric Dumazet5a0d2262010-11-23 10:42:02 +0000146 if (ret && netif_tx_queue_frozen_or_stopped(txq))
David S. Miller37437bb2008-07-16 02:15:04 -0700147 ret = 0;
148
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700149 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000152/*
153 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
154 *
155 * __QDISC_STATE_RUNNING guarantees only one CPU can process
156 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
157 * this queue.
158 *
159 * netif_tx_lock serializes accesses to device driver.
160 *
161 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
162 * if one is grabbed, another must be free.
163 *
164 * Note, that this procedure can be called by a watchdog timer
165 *
166 * Returns to the caller:
167 * 0 - queue is empty or throttled.
168 * >0 - queue is not empty.
169 *
170 */
171static inline int qdisc_restart(struct Qdisc *q)
172{
173 struct netdev_queue *txq;
174 struct net_device *dev;
175 spinlock_t *root_lock;
176 struct sk_buff *skb;
177
178 /* Dequeue packet */
179 skb = dequeue_skb(q);
180 if (unlikely(!skb))
181 return 0;
Eric Dumazet7fee2262010-05-11 23:19:48 +0000182 WARN_ON_ONCE(skb_dst_is_noref(skb));
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000183 root_lock = qdisc_lock(q);
184 dev = qdisc_dev(q);
185 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
186
187 return sch_direct_xmit(skb, q, dev, txq, root_lock);
188}
189
David S. Miller37437bb2008-07-16 02:15:04 -0700190void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700191{
Herbert Xu2ba25062008-03-28 16:25:26 -0700192 unsigned long start_time = jiffies;
193
David S. Miller37437bb2008-07-16 02:15:04 -0700194 while (qdisc_restart(q)) {
Herbert Xu2ba25062008-03-28 16:25:26 -0700195 /*
196 * Postpone processing if
197 * 1. another process needs the CPU;
198 * 2. we've been doing it for too long.
199 */
200 if (need_resched() || jiffies != start_time) {
David S. Miller37437bb2008-07-16 02:15:04 -0700201 __netif_schedule(q);
Herbert Xu2ba25062008-03-28 16:25:26 -0700202 break;
203 }
204 }
Herbert Xu48d83322006-06-19 23:57:59 -0700205
Eric Dumazetbc135b22010-06-02 03:23:51 -0700206 qdisc_run_end(q);
Herbert Xu48d83322006-06-19 23:57:59 -0700207}
208
Eric Dumazet9d214932009-05-17 20:55:16 -0700209unsigned long dev_trans_start(struct net_device *dev)
210{
211 unsigned long val, res = dev->trans_start;
212 unsigned int i;
213
214 for (i = 0; i < dev->num_tx_queues; i++) {
215 val = netdev_get_tx_queue(dev, i)->trans_start;
216 if (val && time_after(val, res))
217 res = val;
218 }
219 dev->trans_start = res;
220 return res;
221}
222EXPORT_SYMBOL(dev_trans_start);
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224static void dev_watchdog(unsigned long arg)
225{
226 struct net_device *dev = (struct net_device *)arg;
227
Herbert Xu932ff272006-06-09 12:20:56 -0700228 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700229 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 if (netif_device_present(dev) &&
231 netif_running(dev) &&
232 netif_carrier_ok(dev)) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700233 int some_queue_timedout = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700234 unsigned int i;
Eric Dumazet9d214932009-05-17 20:55:16 -0700235 unsigned long trans_start;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700236
David S. Millere8a04642008-07-17 00:34:19 -0700237 for (i = 0; i < dev->num_tx_queues; i++) {
238 struct netdev_queue *txq;
239
240 txq = netdev_get_tx_queue(dev, i);
Eric Dumazet9d214932009-05-17 20:55:16 -0700241 /*
242 * old device drivers set dev->trans_start
243 */
244 trans_start = txq->trans_start ? : dev->trans_start;
245 if (netif_tx_queue_stopped(txq) &&
246 time_after(jiffies, (trans_start +
247 dev->watchdog_timeo))) {
248 some_queue_timedout = 1;
David S. Millere8a04642008-07-17 00:34:19 -0700249 break;
250 }
251 }
252
Eric Dumazet9d214932009-05-17 20:55:16 -0700253 if (some_queue_timedout) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700254 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
David S. Miller3019de12011-06-06 16:41:33 -0700255 dev->name, netdev_drivername(dev), i);
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800256 dev->netdev_ops->ndo_tx_timeout(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 }
David S. Millere8a04642008-07-17 00:34:19 -0700258 if (!mod_timer(&dev->watchdog_timer,
259 round_jiffies(jiffies +
260 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 dev_hold(dev);
262 }
263 }
Herbert Xu932ff272006-06-09 12:20:56 -0700264 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266 dev_put(dev);
267}
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269void __netdev_watchdog_up(struct net_device *dev)
270{
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800271 if (dev->netdev_ops->ndo_tx_timeout) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 if (dev->watchdog_timeo <= 0)
273 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700274 if (!mod_timer(&dev->watchdog_timer,
275 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 dev_hold(dev);
277 }
278}
279
280static void dev_watchdog_up(struct net_device *dev)
281{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283}
284
285static void dev_watchdog_down(struct net_device *dev)
286{
Herbert Xu932ff272006-06-09 12:20:56 -0700287 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800289 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700290 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291}
292
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700293/**
294 * netif_carrier_on - set carrier
295 * @dev: network device
296 *
297 * Device has detected that carrier.
298 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700299void netif_carrier_on(struct net_device *dev)
300{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700301 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
David S. Millerb4730012008-11-19 15:33:54 -0800302 if (dev->reg_state == NETREG_UNINITIALIZED)
303 return;
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700304 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700305 if (netif_running(dev))
306 __netdev_watchdog_up(dev);
307 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700308}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800309EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700310
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700311/**
312 * netif_carrier_off - clear carrier
313 * @dev: network device
314 *
315 * Device has detected loss of carrier.
316 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700317void netif_carrier_off(struct net_device *dev)
318{
David S. Millerb4730012008-11-19 15:33:54 -0800319 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
320 if (dev->reg_state == NETREG_UNINITIALIZED)
321 return;
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700322 linkwatch_fire_event(dev);
David S. Millerb4730012008-11-19 15:33:54 -0800323 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700324}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800325EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700326
Ian Campbell06c46482010-05-26 00:09:42 +0000327/**
328 * netif_notify_peers - notify network peers about existence of @dev
329 * @dev: network device
330 *
331 * Generate traffic such that interested network peers are aware of
332 * @dev, such as by generating a gratuitous ARP. This may be used when
333 * a device wants to inform the rest of the network about some sort of
334 * reconfiguration such as a failover event or virtual machine
335 * migration.
336 */
337void netif_notify_peers(struct net_device *dev)
338{
339 rtnl_lock();
340 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
341 rtnl_unlock();
342}
343EXPORT_SYMBOL(netif_notify_peers);
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
346 under all circumstances. It is difficult to invent anything faster or
347 cheaper.
348 */
349
Thomas Graf94df1092005-06-18 22:59:08 -0700350static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
352 kfree_skb(skb);
353 return NET_XMIT_CN;
354}
355
Thomas Graf94df1092005-06-18 22:59:08 -0700356static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 return NULL;
359}
360
Eric Dumazet20fea082007-11-14 01:44:41 -0800361struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 .id = "noop",
363 .priv_size = 0,
364 .enqueue = noop_enqueue,
365 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700366 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 .owner = THIS_MODULE,
368};
369
David S. Miller7698b4f2008-07-16 01:42:40 -0700370static struct netdev_queue noop_netdev_queue = {
David S. Miller7698b4f2008-07-16 01:42:40 -0700371 .qdisc = &noop_qdisc,
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700372 .qdisc_sleeping = &noop_qdisc,
David S. Miller7698b4f2008-07-16 01:42:40 -0700373};
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375struct Qdisc noop_qdisc = {
376 .enqueue = noop_enqueue,
377 .dequeue = noop_dequeue,
378 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900379 .ops = &noop_qdisc_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 .list = LIST_HEAD_INIT(noop_qdisc.list),
David S. Miller83874002008-07-17 00:53:03 -0700381 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700382 .dev_queue = &noop_netdev_queue,
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000383 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800385EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Eric Dumazet20fea082007-11-14 01:44:41 -0800387static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 .id = "noqueue",
389 .priv_size = 0,
390 .enqueue = noop_enqueue,
391 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700392 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 .owner = THIS_MODULE,
394};
395
David S. Miller30ee42b2008-07-18 23:00:11 -0700396static struct Qdisc noqueue_qdisc;
397static struct netdev_queue noqueue_netdev_queue = {
398 .qdisc = &noqueue_qdisc,
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700399 .qdisc_sleeping = &noqueue_qdisc,
David S. Miller30ee42b2008-07-18 23:00:11 -0700400};
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402static struct Qdisc noqueue_qdisc = {
403 .enqueue = NULL,
404 .dequeue = noop_dequeue,
405 .flags = TCQ_F_BUILTIN,
406 .ops = &noqueue_qdisc_ops,
407 .list = LIST_HEAD_INIT(noqueue_qdisc.list),
David S. Miller30ee42b2008-07-18 23:00:11 -0700408 .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
409 .dev_queue = &noqueue_netdev_queue,
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000410 .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411};
412
413
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000414static const u8 prio2band[TC_PRIO_MAX + 1] = {
415 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
416};
Thomas Graf321090e2005-06-18 22:58:35 -0700417
David S. Millerd3678b42008-07-21 09:56:13 -0700418/* 3-band FIFO queue: old style, but should be a bit faster than
419 generic prio+fifo combination.
420 */
421
422#define PFIFO_FAST_BANDS 3
423
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000424/*
425 * Private data for a pfifo_fast scheduler containing:
426 * - queues for the three band
427 * - bitmap indicating which of the bands contain skbs
428 */
429struct pfifo_fast_priv {
430 u32 bitmap;
431 struct sk_buff_head q[PFIFO_FAST_BANDS];
432};
433
434/*
435 * Convert a bitmap to the first band number where an skb is queued, where:
436 * bitmap=0 means there are no skbs on any band.
437 * bitmap=1 means there is an skb on band 0.
438 * bitmap=7 means there are skbs on all 3 bands, etc.
439 */
440static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
441
442static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
443 int band)
David S. Millerd3678b42008-07-21 09:56:13 -0700444{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000445 return priv->q + band;
David S. Millerd3678b42008-07-21 09:56:13 -0700446}
447
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000448static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
David S. Millerd3678b42008-07-21 09:56:13 -0700449{
Krishna Kumara453e062009-08-30 22:20:28 -0700450 if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
451 int band = prio2band[skb->priority & TC_PRIO_MAX];
452 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
453 struct sk_buff_head *list = band2list(priv, band);
David S. Millerd3678b42008-07-21 09:56:13 -0700454
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000455 priv->bitmap |= (1 << band);
David S. Millerd3678b42008-07-21 09:56:13 -0700456 qdisc->q.qlen++;
Thomas Graf821d24a2005-06-18 22:58:15 -0700457 return __qdisc_enqueue_tail(skb, qdisc, list);
David S. Millerd3678b42008-07-21 09:56:13 -0700458 }
Thomas Graf821d24a2005-06-18 22:58:15 -0700459
460 return qdisc_drop(skb, qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000463static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000465 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
466 int band = bitmap2band[priv->bitmap];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000468 if (likely(band >= 0)) {
469 struct sk_buff_head *list = band2list(priv, band);
470 struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
471
472 qdisc->q.qlen--;
473 if (skb_queue_empty(list))
474 priv->bitmap &= ~(1 << band);
475
476 return skb;
David S. Millerd3678b42008-07-21 09:56:13 -0700477 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return NULL;
480}
481
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000482static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700483{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000484 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
485 int band = bitmap2band[priv->bitmap];
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700486
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000487 if (band >= 0) {
488 struct sk_buff_head *list = band2list(priv, band);
489
490 return skb_peek(list);
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700491 }
492
493 return NULL;
494}
495
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000496static void pfifo_fast_reset(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
David S. Millerd3678b42008-07-21 09:56:13 -0700498 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000499 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700500
501 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000502 __qdisc_reset_queue(qdisc, band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700503
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000504 priv->bitmap = 0;
Thomas Graf821d24a2005-06-18 22:58:15 -0700505 qdisc->qstats.backlog = 0;
David S. Millerd3678b42008-07-21 09:56:13 -0700506 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
David S. Millerd3678b42008-07-21 09:56:13 -0700509static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
510{
511 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
512
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000513 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
David S. Millerd3678b42008-07-21 09:56:13 -0700514 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
515 return skb->len;
516
517nla_put_failure:
518 return -1;
519}
520
521static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
522{
523 int prio;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000524 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700525
526 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000527 skb_queue_head_init(band2list(priv, prio));
David S. Millerd3678b42008-07-21 09:56:13 -0700528
Eric Dumazet23624932011-01-21 16:26:09 -0800529 /* Can by-pass the queue discipline */
530 qdisc->flags |= TCQ_F_CAN_BYPASS;
David S. Millerd3678b42008-07-21 09:56:13 -0700531 return 0;
532}
533
David S. Miller6ec1c692009-09-06 01:58:51 -0700534struct Qdisc_ops pfifo_fast_ops __read_mostly = {
David S. Millerd3678b42008-07-21 09:56:13 -0700535 .id = "pfifo_fast",
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000536 .priv_size = sizeof(struct pfifo_fast_priv),
David S. Millerd3678b42008-07-21 09:56:13 -0700537 .enqueue = pfifo_fast_enqueue,
538 .dequeue = pfifo_fast_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700539 .peek = pfifo_fast_peek,
David S. Millerd3678b42008-07-21 09:56:13 -0700540 .init = pfifo_fast_init,
541 .reset = pfifo_fast_reset,
542 .dump = pfifo_fast_dump,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 .owner = THIS_MODULE,
544};
John Fastabendb8970f02011-01-17 08:06:09 +0000545EXPORT_SYMBOL(pfifo_fast_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
David S. Miller5ce2d482008-07-08 17:06:30 -0700547struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700548 struct Qdisc_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
550 void *p;
551 struct Qdisc *sch;
Eric Dumazetd2760552011-03-03 11:10:02 -0800552 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
Thomas Graf3d54b822005-07-05 14:15:09 -0700553 int err = -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
Eric Dumazetf2cd2d32010-11-29 08:14:37 +0000555 p = kzalloc_node(size, GFP_KERNEL,
556 netdev_queue_numa_node_read(dev_queue));
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (!p)
Thomas Graf3d54b822005-07-05 14:15:09 -0700559 goto errout;
Thomas Graf3d54b822005-07-05 14:15:09 -0700560 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
Eric Dumazetd2760552011-03-03 11:10:02 -0800561 /* if we got non aligned memory, ask more and do alignment ourself */
562 if (sch != p) {
563 kfree(p);
564 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
565 netdev_queue_numa_node_read(dev_queue));
566 if (!p)
567 goto errout;
568 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
569 sch->padded = (char *) sch - (char *) p;
570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 INIT_LIST_HEAD(&sch->list);
572 skb_queue_head_init(&sch->q);
Eric Dumazet79640a42010-06-02 05:09:29 -0700573 spin_lock_init(&sch->busylock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 sch->ops = ops;
575 sch->enqueue = ops->enqueue;
576 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700577 sch->dev_queue = dev_queue;
David S. Miller5ce2d482008-07-08 17:06:30 -0700578 dev_hold(qdisc_dev(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 atomic_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700580
581 return sch;
582errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700583 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700584}
585
Changli Gao3511c912010-10-16 13:04:08 +0000586struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
587 struct Qdisc_ops *ops, unsigned int parentid)
Thomas Graf3d54b822005-07-05 14:15:09 -0700588{
589 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900590
David S. Miller5ce2d482008-07-08 17:06:30 -0700591 sch = qdisc_alloc(dev_queue, ops);
Thomas Graf3d54b822005-07-05 14:15:09 -0700592 if (IS_ERR(sch))
593 goto errout;
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800594 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 if (!ops->init || ops->init(sch, NULL) == 0)
597 return sch;
598
Thomas Graf0fbbeb12005-08-23 10:12:44 -0700599 qdisc_destroy(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700600errout:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 return NULL;
602}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800603EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
David S. Miller5fb66222008-08-02 20:02:43 -0700605/* Under qdisc_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607void qdisc_reset(struct Qdisc *qdisc)
608{
Eric Dumazet20fea082007-11-14 01:44:41 -0800609 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611 if (ops->reset)
612 ops->reset(qdisc);
Jarek Poplawski67305eb2008-11-03 02:52:50 -0800613
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000614 if (qdisc->gso_skb) {
615 kfree_skb(qdisc->gso_skb);
616 qdisc->gso_skb = NULL;
617 qdisc->q.qlen = 0;
618 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800620EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Eric Dumazet5d944c62010-03-31 07:06:04 +0000622static void qdisc_rcu_free(struct rcu_head *head)
623{
624 struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
625
626 kfree((char *) qdisc - qdisc->padded);
627}
628
David S. Miller1e0d5a52008-08-17 22:31:26 -0700629void qdisc_destroy(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
Eric Dumazet20fea082007-11-14 01:44:41 -0800631 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
David S. Miller1e0d5a52008-08-17 22:31:26 -0700633 if (qdisc->flags & TCQ_F_BUILTIN ||
634 !atomic_dec_and_test(&qdisc->refcnt))
635 return;
636
David S. Miller3a682fb2008-07-20 18:13:01 -0700637#ifdef CONFIG_NET_SCHED
Jarek Poplawskif6e0b232008-08-22 03:24:05 -0700638 qdisc_list_del(qdisc);
639
Eric Dumazeta2da5702011-01-20 03:48:19 +0000640 qdisc_put_stab(rtnl_dereference(qdisc->stab));
David S. Miller3a682fb2008-07-20 18:13:01 -0700641#endif
Patrick McHardy85670cc2006-09-27 16:45:45 -0700642 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
Patrick McHardy85670cc2006-09-27 16:45:45 -0700643 if (ops->reset)
644 ops->reset(qdisc);
645 if (ops->destroy)
646 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Patrick McHardy85670cc2006-09-27 16:45:45 -0700648 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -0700649 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -0700650
Jarek Poplawski554794d2008-10-06 09:54:39 -0700651 kfree_skb(qdisc->gso_skb);
Eric Dumazet5d944c62010-03-31 07:06:04 +0000652 /*
653 * gen_estimator est_timer() might access qdisc->q.lock,
654 * wait a RCU grace period before freeing qdisc.
655 */
656 call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
David S. Miller8a34c5d2008-07-17 00:47:45 -0700657}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800658EXPORT_SYMBOL(qdisc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Patrick McHardy589983c2009-09-04 06:41:20 +0000660/* Attach toplevel qdisc to device queue. */
661struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
662 struct Qdisc *qdisc)
663{
664 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
665 spinlock_t *root_lock;
666
667 root_lock = qdisc_lock(oqdisc);
668 spin_lock_bh(root_lock);
669
670 /* Prune old scheduler */
671 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
672 qdisc_reset(oqdisc);
673
674 /* ... and graft new one */
675 if (qdisc == NULL)
676 qdisc = &noop_qdisc;
677 dev_queue->qdisc_sleeping = qdisc;
678 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
679
680 spin_unlock_bh(root_lock);
681
682 return oqdisc;
683}
John Fastabendb8970f02011-01-17 08:06:09 +0000684EXPORT_SYMBOL(dev_graft_qdisc);
Patrick McHardy589983c2009-09-04 06:41:20 +0000685
David S. Millere8a04642008-07-17 00:34:19 -0700686static void attach_one_default_qdisc(struct net_device *dev,
687 struct netdev_queue *dev_queue,
688 void *_unused)
689{
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000690 struct Qdisc *qdisc = &noqueue_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -0700691
692 if (dev->tx_queue_len) {
Changli Gao3511c912010-10-16 13:04:08 +0000693 qdisc = qdisc_create_dflt(dev_queue,
David S. Millerd3678b42008-07-21 09:56:13 -0700694 &pfifo_fast_ops, TC_H_ROOT);
David S. Millere8a04642008-07-17 00:34:19 -0700695 if (!qdisc) {
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000696 netdev_info(dev, "activation failed\n");
David S. Millere8a04642008-07-17 00:34:19 -0700697 return;
698 }
David S. Millere8a04642008-07-17 00:34:19 -0700699 }
700 dev_queue->qdisc_sleeping = qdisc;
701}
702
David S. Miller6ec1c692009-09-06 01:58:51 -0700703static void attach_default_qdiscs(struct net_device *dev)
704{
705 struct netdev_queue *txq;
706 struct Qdisc *qdisc;
707
708 txq = netdev_get_tx_queue(dev, 0);
709
710 if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) {
711 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
712 dev->qdisc = txq->qdisc_sleeping;
713 atomic_inc(&dev->qdisc->refcnt);
714 } else {
Changli Gao3511c912010-10-16 13:04:08 +0000715 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
David S. Miller6ec1c692009-09-06 01:58:51 -0700716 if (qdisc) {
717 qdisc->ops->attach(qdisc);
718 dev->qdisc = qdisc;
719 }
720 }
721}
722
David S. Millere8a04642008-07-17 00:34:19 -0700723static void transition_one_qdisc(struct net_device *dev,
724 struct netdev_queue *dev_queue,
725 void *_need_watchdog)
726{
David S. Miller83874002008-07-17 00:53:03 -0700727 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700728 int *need_watchdog_p = _need_watchdog;
729
David S. Millera9312ae2008-08-17 21:51:03 -0700730 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
731 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
732
David S. Miller83874002008-07-17 00:53:03 -0700733 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
Eric Dumazet9d214932009-05-17 20:55:16 -0700734 if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
735 dev_queue->trans_start = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700736 *need_watchdog_p = 1;
Eric Dumazet9d214932009-05-17 20:55:16 -0700737 }
David S. Millere8a04642008-07-17 00:34:19 -0700738}
739
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740void dev_activate(struct net_device *dev)
741{
David S. Millere8a04642008-07-17 00:34:19 -0700742 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -0700743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /* No queueing discipline is attached to device;
David S. Millerd3678b42008-07-21 09:56:13 -0700745 create default one i.e. pfifo_fast for devices,
746 which need queueing and noqueue_qdisc for
747 virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 */
749
David S. Miller6ec1c692009-09-06 01:58:51 -0700750 if (dev->qdisc == &noop_qdisc)
751 attach_default_qdiscs(dev);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000752
Tommy S. Christensencacaddf2005-05-03 16:18:52 -0700753 if (!netif_carrier_ok(dev))
754 /* Delay activation until next carrier-on event */
755 return;
756
David S. Millere8a04642008-07-17 00:34:19 -0700757 need_watchdog = 0;
758 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
Eric Dumazet24824a02010-10-02 06:11:55 +0000759 if (dev_ingress_queue(dev))
760 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
David S. Millere8a04642008-07-17 00:34:19 -0700761
762 if (need_watchdog) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 dev->trans_start = jiffies;
764 dev_watchdog_up(dev);
765 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700766}
John Fastabendb8970f02011-01-17 08:06:09 +0000767EXPORT_SYMBOL(dev_activate);
David S. Millerb0e1e642008-07-08 17:42:10 -0700768
David S. Millere8a04642008-07-17 00:34:19 -0700769static void dev_deactivate_queue(struct net_device *dev,
770 struct netdev_queue *dev_queue,
771 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -0700772{
David S. Millere8a04642008-07-17 00:34:19 -0700773 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -0700774 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700775
David S. Miller970565b2008-07-08 23:10:33 -0700776 qdisc = dev_queue->qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700777 if (qdisc) {
David S. Miller83874002008-07-17 00:53:03 -0700778 spin_lock_bh(qdisc_lock(qdisc));
779
David S. Millera9312ae2008-08-17 21:51:03 -0700780 if (!(qdisc->flags & TCQ_F_BUILTIN))
781 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
782
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700783 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700784 qdisc_reset(qdisc);
David S. Millerd3b753d2008-07-15 20:14:35 -0700785
David S. Miller83874002008-07-17 00:53:03 -0700786 spin_unlock_bh(qdisc_lock(qdisc));
David S. Millerb0e1e642008-07-08 17:42:10 -0700787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
David S. Miller4335cd22008-08-17 21:58:07 -0700790static bool some_qdisc_is_busy(struct net_device *dev)
David S. Millere8a04642008-07-17 00:34:19 -0700791{
792 unsigned int i;
793
794 for (i = 0; i < dev->num_tx_queues; i++) {
795 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -0700796 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -0700797 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -0700798 int val;
799
800 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millerb9a3b112008-08-13 15:18:38 -0700801 q = dev_queue->qdisc_sleeping;
David S. Miller5fb66222008-08-02 20:02:43 -0700802 root_lock = qdisc_lock(q);
David S. Millere8a04642008-07-17 00:34:19 -0700803
David S. Miller4335cd22008-08-17 21:58:07 -0700804 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700805
Eric Dumazetbc135b22010-06-02 03:23:51 -0700806 val = (qdisc_is_running(q) ||
David S. Millerb9a3b112008-08-13 15:18:38 -0700807 test_bit(__QDISC_STATE_SCHED, &q->state));
David S. Millere8a04642008-07-17 00:34:19 -0700808
David S. Miller4335cd22008-08-17 21:58:07 -0700809 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700810
811 if (val)
812 return true;
813 }
814 return false;
815}
816
Eric Dumazet31376632011-05-19 23:42:09 +0000817/**
818 * dev_deactivate_many - deactivate transmissions on several devices
819 * @head: list of devices to deactivate
820 *
821 * This function returns only when all outstanding transmissions
822 * have completed, unless all devices are in dismantle phase.
823 */
Octavian Purdila44345722010-12-13 12:44:07 +0000824void dev_deactivate_many(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Octavian Purdila44345722010-12-13 12:44:07 +0000826 struct net_device *dev;
Eric Dumazet31376632011-05-19 23:42:09 +0000827 bool sync_needed = false;
Herbert Xu41a23b02007-05-10 14:12:47 -0700828
Octavian Purdila44345722010-12-13 12:44:07 +0000829 list_for_each_entry(dev, head, unreg_list) {
830 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
831 &noop_qdisc);
832 if (dev_ingress_queue(dev))
833 dev_deactivate_queue(dev, dev_ingress_queue(dev),
834 &noop_qdisc);
835
836 dev_watchdog_down(dev);
Eric Dumazet31376632011-05-19 23:42:09 +0000837 sync_needed |= !dev->dismantle;
Octavian Purdila44345722010-12-13 12:44:07 +0000838 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Eric Dumazet31376632011-05-19 23:42:09 +0000840 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
841 * This is avoided if all devices are in dismantle phase :
842 * Caller will call synchronize_net() for us
843 */
844 if (sync_needed)
845 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Herbert Xud4828d82006-06-22 02:28:18 -0700847 /* Wait for outstanding qdisc_run calls. */
Octavian Purdila44345722010-12-13 12:44:07 +0000848 list_for_each_entry(dev, head, unreg_list)
849 while (some_qdisc_is_busy(dev))
850 yield();
851}
852
853void dev_deactivate(struct net_device *dev)
854{
855 LIST_HEAD(single);
856
857 list_add(&dev->unreg_list, &single);
858 dev_deactivate_many(&single);
Eric W. Biederman5f04d502011-02-20 11:49:45 -0800859 list_del(&single);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
John Fastabendb8970f02011-01-17 08:06:09 +0000861EXPORT_SYMBOL(dev_deactivate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
David S. Millerb0e1e642008-07-08 17:42:10 -0700863static void dev_init_scheduler_queue(struct net_device *dev,
864 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -0700865 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -0700866{
David S. Millere8a04642008-07-17 00:34:19 -0700867 struct Qdisc *qdisc = _qdisc;
868
David S. Millerb0e1e642008-07-08 17:42:10 -0700869 dev_queue->qdisc = qdisc;
870 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700871}
872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873void dev_init_scheduler(struct net_device *dev)
874{
Patrick McHardyaf356af2009-09-04 06:41:18 +0000875 dev->qdisc = &noop_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -0700876 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000877 if (dev_ingress_queue(dev))
878 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800880 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
882
David S. Millere8a04642008-07-17 00:34:19 -0700883static void shutdown_scheduler_queue(struct net_device *dev,
884 struct netdev_queue *dev_queue,
885 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
David S. Millerb0e1e642008-07-08 17:42:10 -0700887 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700888 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
David S. Millerb0e1e642008-07-08 17:42:10 -0700890 if (qdisc) {
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700891 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700892 dev_queue->qdisc_sleeping = qdisc_default;
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 qdisc_destroy(qdisc);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900895 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700896}
897
898void dev_shutdown(struct net_device *dev)
899{
David S. Millere8a04642008-07-17 00:34:19 -0700900 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +0000901 if (dev_ingress_queue(dev))
902 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Patrick McHardyaf356af2009-09-04 06:41:18 +0000903 qdisc_destroy(dev->qdisc);
904 dev->qdisc = &noop_qdisc;
905
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700906 WARN_ON(timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}