blob: 9634091ee2f0f7f1c65059d6ae454281d2db0adf [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <net/pkt_sched.h>
28
29/* Main transmission queue. */
30
Patrick McHardy0463d4a2007-04-16 17:02:10 -070031/* Modifications to data participating in scheduling must be protected with
David S. Miller5fb66222008-08-02 20:02:43 -070032 * qdisc_lock(qdisc) spinlock.
Patrick McHardy0463d4a2007-04-16 17:02:10 -070033 *
34 * The idea is the following:
David S. Millerc7e4f3b2008-07-16 03:22:39 -070035 * - enqueue, dequeue are serialized via qdisc root lock
36 * - ingress filtering is also serialized via qdisc root lock
Patrick McHardy0463d4a2007-04-16 17:02:10 -070037 * - updates to tree and tree walking are only done under the rtnl mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070040static inline int qdisc_qlen(struct Qdisc *q)
41{
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070042 return q->q.qlen;
43}
44
David S. Miller37437bb2008-07-16 02:15:04 -070045static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070046{
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070047 if (unlikely(skb->next))
David S. Millerd3b753d2008-07-15 20:14:35 -070048 q->gso_skb = skb;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070049 else
50 q->ops->requeue(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070051
David S. Miller37437bb2008-07-16 02:15:04 -070052 __netif_schedule(q);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070053 return 0;
54}
55
David S. Millerd3b753d2008-07-15 20:14:35 -070056static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070057{
Krishna Kumar6c1361a2007-06-24 19:56:09 -070058 struct sk_buff *skb;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070059
David S. Millerd3b753d2008-07-15 20:14:35 -070060 if ((skb = q->gso_skb))
61 q->gso_skb = NULL;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070062 else
63 skb = q->dequeue(q);
64
65 return skb;
66}
67
Krishna Kumar6c1361a2007-06-24 19:56:09 -070068static inline int handle_dev_cpu_collision(struct sk_buff *skb,
David S. Miller970565b2008-07-08 23:10:33 -070069 struct netdev_queue *dev_queue,
Krishna Kumar6c1361a2007-06-24 19:56:09 -070070 struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070071{
Krishna Kumar6c1361a2007-06-24 19:56:09 -070072 int ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070073
David S. Millerc773e842008-07-08 23:13:53 -070074 if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -070075 /*
76 * Same CPU holding the lock. It may be a transient
77 * configuration error, when hard_start_xmit() recurses. We
78 * detect it by checking xmit owner and drop the packet when
79 * deadloop is detected. Return OK to try the next skb.
80 */
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070081 kfree_skb(skb);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070082 if (net_ratelimit())
83 printk(KERN_WARNING "Dead loop on netdevice %s, "
David S. Millerc773e842008-07-08 23:13:53 -070084 "fix it urgently!\n", dev_queue->dev->name);
Krishna Kumar6c1361a2007-06-24 19:56:09 -070085 ret = qdisc_qlen(q);
86 } else {
87 /*
88 * Another cpu is holding lock, requeue & delay xmits for
89 * some time.
90 */
91 __get_cpu_var(netdev_rx_stat).cpu_collision++;
David S. Miller37437bb2008-07-16 02:15:04 -070092 ret = dev_requeue_skb(skb, q);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070093 }
94
Krishna Kumar6c1361a2007-06-24 19:56:09 -070095 return ret;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -070096}
97
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +090098/*
David S. Miller83874002008-07-17 00:53:03 -070099 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700100 *
David S. Millere2627c82008-07-16 00:56:32 -0700101 * __QDISC_STATE_RUNNING guarantees only one CPU can process
David S. Miller83874002008-07-17 00:53:03 -0700102 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
103 * this queue.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700104 *
105 * netif_tx_lock serializes accesses to device driver.
106 *
David S. Miller83874002008-07-17 00:53:03 -0700107 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700108 * if one is grabbed, another must be free.
109 *
110 * Note, that this procedure can be called by a watchdog timer
111 *
112 * Returns to the caller:
113 * 0 - queue is empty or throttled.
114 * >0 - queue is not empty.
115 *
116 */
David S. Miller37437bb2008-07-16 02:15:04 -0700117static inline int qdisc_restart(struct Qdisc *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
David S. Miller37437bb2008-07-16 02:15:04 -0700119 struct netdev_queue *txq;
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800120 int ret = NETDEV_TX_BUSY;
David S. Millereb6aafe2008-07-08 23:12:38 -0700121 struct net_device *dev;
David S. Miller7698b4f2008-07-16 01:42:40 -0700122 spinlock_t *root_lock;
David S. Millereb6aafe2008-07-08 23:12:38 -0700123 struct sk_buff *skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700125 /* Dequeue packet */
David S. Millerd3b753d2008-07-15 20:14:35 -0700126 if (unlikely((skb = dequeue_skb(q)) == NULL))
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700127 return 0;
Herbert Xuf6a78bf2006-06-22 02:57:17 -0700128
David S. Miller5fb66222008-08-02 20:02:43 -0700129 root_lock = qdisc_lock(q);
David S. Miller7698b4f2008-07-16 01:42:40 -0700130
131 /* And release qdisc */
132 spin_unlock(root_lock);
Herbert Xud90df3a2007-05-10 04:55:14 -0700133
David S. Miller37437bb2008-07-16 02:15:04 -0700134 dev = qdisc_dev(q);
135 txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
David S. Millereb6aafe2008-07-08 23:12:38 -0700136
David S. Millerc773e842008-07-08 23:13:53 -0700137 HARD_TX_LOCK(dev, txq, smp_processor_id());
David S. Millerc3f26a22008-07-31 16:58:50 -0700138 if (!netif_tx_queue_stopped(txq) &&
139 !netif_tx_queue_frozen(txq))
David S. Millerfd2ea0a2008-07-17 01:56:23 -0700140 ret = dev_hard_start_xmit(skb, dev, txq);
David S. Millerc773e842008-07-08 23:13:53 -0700141 HARD_TX_UNLOCK(dev, txq);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700142
David S. Miller7698b4f2008-07-16 01:42:40 -0700143 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700144
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700145 switch (ret) {
146 case NETDEV_TX_OK:
147 /* Driver sent out skb successfully */
148 ret = qdisc_qlen(q);
149 break;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700150
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700151 case NETDEV_TX_LOCKED:
152 /* Driver try lock failed */
David S. Millereb6aafe2008-07-08 23:12:38 -0700153 ret = handle_dev_cpu_collision(skb, txq, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700154 break;
155
156 default:
157 /* Driver returned NETDEV_TX_BUSY - requeue skb */
158 if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
159 printk(KERN_WARNING "BUG %s code %d qlen %d\n",
160 dev->name, ret, q->q.qlen);
161
David S. Miller37437bb2008-07-16 02:15:04 -0700162 ret = dev_requeue_skb(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700163 break;
164 }
165
David S. Millerc3f26a22008-07-31 16:58:50 -0700166 if (ret && (netif_tx_queue_stopped(txq) ||
167 netif_tx_queue_frozen(txq)))
David S. Miller37437bb2008-07-16 02:15:04 -0700168 ret = 0;
169
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700170 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
David S. Miller37437bb2008-07-16 02:15:04 -0700173void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700174{
Herbert Xu2ba25062008-03-28 16:25:26 -0700175 unsigned long start_time = jiffies;
176
David S. Miller37437bb2008-07-16 02:15:04 -0700177 while (qdisc_restart(q)) {
Herbert Xu2ba25062008-03-28 16:25:26 -0700178 /*
179 * Postpone processing if
180 * 1. another process needs the CPU;
181 * 2. we've been doing it for too long.
182 */
183 if (need_resched() || jiffies != start_time) {
David S. Miller37437bb2008-07-16 02:15:04 -0700184 __netif_schedule(q);
Herbert Xu2ba25062008-03-28 16:25:26 -0700185 break;
186 }
187 }
Herbert Xu48d83322006-06-19 23:57:59 -0700188
David S. Millere2627c82008-07-16 00:56:32 -0700189 clear_bit(__QDISC_STATE_RUNNING, &q->state);
Herbert Xu48d83322006-06-19 23:57:59 -0700190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192static void dev_watchdog(unsigned long arg)
193{
194 struct net_device *dev = (struct net_device *)arg;
195
Herbert Xu932ff272006-06-09 12:20:56 -0700196 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700197 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (netif_device_present(dev) &&
199 netif_running(dev) &&
200 netif_carrier_ok(dev)) {
David S. Millere8a04642008-07-17 00:34:19 -0700201 int some_queue_stopped = 0;
202 unsigned int i;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700203
David S. Millere8a04642008-07-17 00:34:19 -0700204 for (i = 0; i < dev->num_tx_queues; i++) {
205 struct netdev_queue *txq;
206
207 txq = netdev_get_tx_queue(dev, i);
208 if (netif_tx_queue_stopped(txq)) {
209 some_queue_stopped = 1;
210 break;
211 }
212 }
213
214 if (some_queue_stopped &&
215 time_after(jiffies, (dev->trans_start +
216 dev->watchdog_timeo))) {
Arjan van de Ven6579e572008-07-21 13:31:48 -0700217 char drivername[64];
218 printk(KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit timed out\n",
219 dev->name, netdev_drivername(dev, drivername, 64));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 dev->tx_timeout(dev);
Arjan van de Venb4192bb2008-05-02 16:21:07 -0700221 WARN_ON_ONCE(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
David S. Millere8a04642008-07-17 00:34:19 -0700223 if (!mod_timer(&dev->watchdog_timer,
224 round_jiffies(jiffies +
225 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 dev_hold(dev);
227 }
228 }
Herbert Xu932ff272006-06-09 12:20:56 -0700229 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
231 dev_put(dev);
232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234void __netdev_watchdog_up(struct net_device *dev)
235{
236 if (dev->tx_timeout) {
237 if (dev->watchdog_timeo <= 0)
238 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700239 if (!mod_timer(&dev->watchdog_timer,
240 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 dev_hold(dev);
242 }
243}
244
245static void dev_watchdog_up(struct net_device *dev)
246{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
250static void dev_watchdog_down(struct net_device *dev)
251{
Herbert Xu932ff272006-06-09 12:20:56 -0700252 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800254 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700255 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700258/**
259 * netif_carrier_on - set carrier
260 * @dev: network device
261 *
262 * Device has detected that carrier.
263 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700264void netif_carrier_on(struct net_device *dev)
265{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700266 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700267 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700268 if (netif_running(dev))
269 __netdev_watchdog_up(dev);
270 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700271}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800272EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700273
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700274/**
275 * netif_carrier_off - clear carrier
276 * @dev: network device
277 *
278 * Device has detected loss of carrier.
279 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700280void netif_carrier_off(struct net_device *dev)
281{
282 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state))
283 linkwatch_fire_event(dev);
284}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800285EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
288 under all circumstances. It is difficult to invent anything faster or
289 cheaper.
290 */
291
Thomas Graf94df1092005-06-18 22:59:08 -0700292static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
294 kfree_skb(skb);
295 return NET_XMIT_CN;
296}
297
Thomas Graf94df1092005-06-18 22:59:08 -0700298static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 return NULL;
301}
302
Thomas Graf94df1092005-06-18 22:59:08 -0700303static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 if (net_ratelimit())
Thomas Graf94df1092005-06-18 22:59:08 -0700306 printk(KERN_DEBUG "%s deferred output. It is buggy.\n",
307 skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 kfree_skb(skb);
309 return NET_XMIT_CN;
310}
311
Eric Dumazet20fea082007-11-14 01:44:41 -0800312struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 .id = "noop",
314 .priv_size = 0,
315 .enqueue = noop_enqueue,
316 .dequeue = noop_dequeue,
317 .requeue = noop_requeue,
318 .owner = THIS_MODULE,
319};
320
David S. Miller7698b4f2008-07-16 01:42:40 -0700321static struct netdev_queue noop_netdev_queue = {
David S. Miller7698b4f2008-07-16 01:42:40 -0700322 .qdisc = &noop_qdisc,
323};
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325struct Qdisc noop_qdisc = {
326 .enqueue = noop_enqueue,
327 .dequeue = noop_dequeue,
328 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900329 .ops = &noop_qdisc_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 .list = LIST_HEAD_INIT(noop_qdisc.list),
David S. Miller83874002008-07-17 00:53:03 -0700331 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700332 .dev_queue = &noop_netdev_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800334EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Eric Dumazet20fea082007-11-14 01:44:41 -0800336static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 .id = "noqueue",
338 .priv_size = 0,
339 .enqueue = noop_enqueue,
340 .dequeue = noop_dequeue,
341 .requeue = noop_requeue,
342 .owner = THIS_MODULE,
343};
344
David S. Miller30ee42b2008-07-18 23:00:11 -0700345static struct Qdisc noqueue_qdisc;
346static struct netdev_queue noqueue_netdev_queue = {
347 .qdisc = &noqueue_qdisc,
348};
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350static struct Qdisc noqueue_qdisc = {
351 .enqueue = NULL,
352 .dequeue = noop_dequeue,
353 .flags = TCQ_F_BUILTIN,
354 .ops = &noqueue_qdisc_ops,
355 .list = LIST_HEAD_INIT(noqueue_qdisc.list),
David S. Miller30ee42b2008-07-18 23:00:11 -0700356 .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
357 .dev_queue = &noqueue_netdev_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358};
359
360
David S. Millerd3678b42008-07-21 09:56:13 -0700361static const u8 prio2band[TC_PRIO_MAX+1] =
362 { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 };
Thomas Graf321090e2005-06-18 22:58:35 -0700363
David S. Millerd3678b42008-07-21 09:56:13 -0700364/* 3-band FIFO queue: old style, but should be a bit faster than
365 generic prio+fifo combination.
366 */
367
368#define PFIFO_FAST_BANDS 3
369
370static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
371 struct Qdisc *qdisc)
372{
373 struct sk_buff_head *list = qdisc_priv(qdisc);
374 return list + prio2band[skb->priority & TC_PRIO_MAX];
375}
376
377static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
378{
379 struct sk_buff_head *list = prio2list(skb, qdisc);
380
381 if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
382 qdisc->q.qlen++;
Thomas Graf821d24a2005-06-18 22:58:15 -0700383 return __qdisc_enqueue_tail(skb, qdisc, list);
David S. Millerd3678b42008-07-21 09:56:13 -0700384 }
Thomas Graf821d24a2005-06-18 22:58:15 -0700385
386 return qdisc_drop(skb, qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387}
388
David S. Millerd3678b42008-07-21 09:56:13 -0700389static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
David S. Millerd3678b42008-07-21 09:56:13 -0700391 int prio;
392 struct sk_buff_head *list = qdisc_priv(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
David S. Millerd3678b42008-07-21 09:56:13 -0700394 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
395 if (!skb_queue_empty(list + prio)) {
396 qdisc->q.qlen--;
397 return __qdisc_dequeue_head(qdisc, list + prio);
398 }
399 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 return NULL;
402}
403
David S. Millerd3678b42008-07-21 09:56:13 -0700404static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
David S. Millerd3678b42008-07-21 09:56:13 -0700406 qdisc->q.qlen++;
407 return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408}
409
David S. Millerd3678b42008-07-21 09:56:13 -0700410static void pfifo_fast_reset(struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
David S. Millerd3678b42008-07-21 09:56:13 -0700412 int prio;
413 struct sk_buff_head *list = qdisc_priv(qdisc);
414
415 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
416 __qdisc_reset_queue(qdisc, list + prio);
417
Thomas Graf821d24a2005-06-18 22:58:15 -0700418 qdisc->qstats.backlog = 0;
David S. Millerd3678b42008-07-21 09:56:13 -0700419 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
David S. Millerd3678b42008-07-21 09:56:13 -0700422static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
423{
424 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
425
426 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1);
427 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
428 return skb->len;
429
430nla_put_failure:
431 return -1;
432}
433
434static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
435{
436 int prio;
437 struct sk_buff_head *list = qdisc_priv(qdisc);
438
439 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
440 skb_queue_head_init(list + prio);
441
442 return 0;
443}
444
445static struct Qdisc_ops pfifo_fast_ops __read_mostly = {
446 .id = "pfifo_fast",
447 .priv_size = PFIFO_FAST_BANDS * sizeof(struct sk_buff_head),
448 .enqueue = pfifo_fast_enqueue,
449 .dequeue = pfifo_fast_dequeue,
450 .requeue = pfifo_fast_requeue,
451 .init = pfifo_fast_init,
452 .reset = pfifo_fast_reset,
453 .dump = pfifo_fast_dump,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 .owner = THIS_MODULE,
455};
456
David S. Miller5ce2d482008-07-08 17:06:30 -0700457struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700458 struct Qdisc_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
460 void *p;
461 struct Qdisc *sch;
Thomas Graf3d54b822005-07-05 14:15:09 -0700462 unsigned int size;
463 int err = -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 /* ensure that the Qdisc and the private data are 32-byte aligned */
Thomas Graf3d54b822005-07-05 14:15:09 -0700466 size = QDISC_ALIGN(sizeof(*sch));
467 size += ops->priv_size + (QDISC_ALIGNTO - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700469 p = kzalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (!p)
Thomas Graf3d54b822005-07-05 14:15:09 -0700471 goto errout;
Thomas Graf3d54b822005-07-05 14:15:09 -0700472 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
473 sch->padded = (char *) sch - (char *) p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 INIT_LIST_HEAD(&sch->list);
476 skb_queue_head_init(&sch->q);
477 sch->ops = ops;
478 sch->enqueue = ops->enqueue;
479 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700480 sch->dev_queue = dev_queue;
David S. Miller5ce2d482008-07-08 17:06:30 -0700481 dev_hold(qdisc_dev(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 atomic_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700483
484 return sch;
485errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700486 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700487}
488
David S. Millerbb949fb2008-07-08 16:55:56 -0700489struct Qdisc * qdisc_create_dflt(struct net_device *dev,
490 struct netdev_queue *dev_queue,
491 struct Qdisc_ops *ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800492 unsigned int parentid)
Thomas Graf3d54b822005-07-05 14:15:09 -0700493{
494 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900495
David S. Miller5ce2d482008-07-08 17:06:30 -0700496 sch = qdisc_alloc(dev_queue, ops);
Thomas Graf3d54b822005-07-05 14:15:09 -0700497 if (IS_ERR(sch))
498 goto errout;
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800499 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if (!ops->init || ops->init(sch, NULL) == 0)
502 return sch;
503
Thomas Graf0fbbeb12005-08-23 10:12:44 -0700504 qdisc_destroy(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700505errout:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 return NULL;
507}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800508EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
David S. Miller5fb66222008-08-02 20:02:43 -0700510/* Under qdisc_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
512void qdisc_reset(struct Qdisc *qdisc)
513{
Eric Dumazet20fea082007-11-14 01:44:41 -0800514 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
516 if (ops->reset)
517 ops->reset(qdisc);
518}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800519EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
David S. Miller1e0d5a52008-08-17 22:31:26 -0700521void qdisc_destroy(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
Eric Dumazet20fea082007-11-14 01:44:41 -0800523 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
David S. Miller1e0d5a52008-08-17 22:31:26 -0700525 if (qdisc->flags & TCQ_F_BUILTIN ||
526 !atomic_dec_and_test(&qdisc->refcnt))
527 return;
528
David S. Miller3a682fb2008-07-20 18:13:01 -0700529#ifdef CONFIG_NET_SCHED
Jarek Poplawskif6e0b232008-08-22 03:24:05 -0700530 qdisc_list_del(qdisc);
531
Jussi Kivilinna175f9c12008-07-20 00:08:47 -0700532 qdisc_put_stab(qdisc->stab);
David S. Miller3a682fb2008-07-20 18:13:01 -0700533#endif
Patrick McHardy85670cc2006-09-27 16:45:45 -0700534 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
Patrick McHardy85670cc2006-09-27 16:45:45 -0700535 if (ops->reset)
536 ops->reset(qdisc);
537 if (ops->destroy)
538 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
Patrick McHardy85670cc2006-09-27 16:45:45 -0700540 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -0700541 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -0700542
David S. Miller83874002008-07-17 00:53:03 -0700543 kfree_skb(qdisc->gso_skb);
544
David S. Miller8a34c5d2008-07-17 00:47:45 -0700545 kfree((char *) qdisc - qdisc->padded);
546}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800547EXPORT_SYMBOL(qdisc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
David S. Millere8a04642008-07-17 00:34:19 -0700549static bool dev_all_qdisc_sleeping_noop(struct net_device *dev)
550{
551 unsigned int i;
552
553 for (i = 0; i < dev->num_tx_queues; i++) {
554 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
555
556 if (txq->qdisc_sleeping != &noop_qdisc)
557 return false;
558 }
559 return true;
560}
561
562static void attach_one_default_qdisc(struct net_device *dev,
563 struct netdev_queue *dev_queue,
564 void *_unused)
565{
566 struct Qdisc *qdisc;
567
568 if (dev->tx_queue_len) {
569 qdisc = qdisc_create_dflt(dev, dev_queue,
David S. Millerd3678b42008-07-21 09:56:13 -0700570 &pfifo_fast_ops, TC_H_ROOT);
David S. Millere8a04642008-07-17 00:34:19 -0700571 if (!qdisc) {
572 printk(KERN_INFO "%s: activation failed\n", dev->name);
573 return;
574 }
David S. Millere8a04642008-07-17 00:34:19 -0700575 } else {
576 qdisc = &noqueue_qdisc;
577 }
578 dev_queue->qdisc_sleeping = qdisc;
579}
580
581static void transition_one_qdisc(struct net_device *dev,
582 struct netdev_queue *dev_queue,
583 void *_need_watchdog)
584{
David S. Miller83874002008-07-17 00:53:03 -0700585 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700586 int *need_watchdog_p = _need_watchdog;
587
David S. Millera9312ae2008-08-17 21:51:03 -0700588 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
589 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
590
David S. Miller83874002008-07-17 00:53:03 -0700591 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
David S. Miller8d50b532008-07-30 02:37:46 -0700592 if (need_watchdog_p && new_qdisc != &noqueue_qdisc)
David S. Millere8a04642008-07-17 00:34:19 -0700593 *need_watchdog_p = 1;
David S. Millere8a04642008-07-17 00:34:19 -0700594}
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596void dev_activate(struct net_device *dev)
597{
David S. Millere8a04642008-07-17 00:34:19 -0700598 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -0700599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 /* No queueing discipline is attached to device;
David S. Millerd3678b42008-07-21 09:56:13 -0700601 create default one i.e. pfifo_fast for devices,
602 which need queueing and noqueue_qdisc for
603 virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
605
David S. Millere8a04642008-07-17 00:34:19 -0700606 if (dev_all_qdisc_sleeping_noop(dev))
607 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Tommy S. Christensencacaddf2005-05-03 16:18:52 -0700609 if (!netif_carrier_ok(dev))
610 /* Delay activation until next carrier-on event */
611 return;
612
David S. Millere8a04642008-07-17 00:34:19 -0700613 need_watchdog = 0;
614 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
David S. Miller8d50b532008-07-30 02:37:46 -0700615 transition_one_qdisc(dev, &dev->rx_queue, NULL);
David S. Millere8a04642008-07-17 00:34:19 -0700616
617 if (need_watchdog) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 dev->trans_start = jiffies;
619 dev_watchdog_up(dev);
620 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700621}
622
David S. Millere8a04642008-07-17 00:34:19 -0700623static void dev_deactivate_queue(struct net_device *dev,
624 struct netdev_queue *dev_queue,
625 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -0700626{
David S. Millere8a04642008-07-17 00:34:19 -0700627 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -0700628 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700629
David S. Miller970565b2008-07-08 23:10:33 -0700630 qdisc = dev_queue->qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700631 if (qdisc) {
David S. Miller83874002008-07-17 00:53:03 -0700632 spin_lock_bh(qdisc_lock(qdisc));
633
David S. Millera9312ae2008-08-17 21:51:03 -0700634 if (!(qdisc->flags & TCQ_F_BUILTIN))
635 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
636
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700637 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700638 qdisc_reset(qdisc);
David S. Millerd3b753d2008-07-15 20:14:35 -0700639
David S. Miller83874002008-07-17 00:53:03 -0700640 spin_unlock_bh(qdisc_lock(qdisc));
David S. Millerb0e1e642008-07-08 17:42:10 -0700641 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642}
643
David S. Miller4335cd22008-08-17 21:58:07 -0700644static bool some_qdisc_is_busy(struct net_device *dev)
David S. Millere8a04642008-07-17 00:34:19 -0700645{
646 unsigned int i;
647
648 for (i = 0; i < dev->num_tx_queues; i++) {
649 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -0700650 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -0700651 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -0700652 int val;
653
654 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millerb9a3b112008-08-13 15:18:38 -0700655 q = dev_queue->qdisc_sleeping;
David S. Miller5fb66222008-08-02 20:02:43 -0700656 root_lock = qdisc_lock(q);
David S. Millere8a04642008-07-17 00:34:19 -0700657
David S. Miller4335cd22008-08-17 21:58:07 -0700658 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700659
David S. Millerb9a3b112008-08-13 15:18:38 -0700660 val = (test_bit(__QDISC_STATE_RUNNING, &q->state) ||
661 test_bit(__QDISC_STATE_SCHED, &q->state));
David S. Millere8a04642008-07-17 00:34:19 -0700662
David S. Miller4335cd22008-08-17 21:58:07 -0700663 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700664
665 if (val)
666 return true;
667 }
668 return false;
669}
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671void dev_deactivate(struct net_device *dev)
672{
David S. Millere8a04642008-07-17 00:34:19 -0700673 netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc);
David S. Miller8d50b532008-07-30 02:37:46 -0700674 dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc);
Herbert Xu41a23b02007-05-10 14:12:47 -0700675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 dev_watchdog_down(dev);
677
Herbert Xuce0e32e2007-10-18 22:37:58 -0700678 /* Wait for outstanding qdisc-less dev_queue_xmit calls. */
Herbert Xud4828d82006-06-22 02:28:18 -0700679 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Herbert Xud4828d82006-06-22 02:28:18 -0700681 /* Wait for outstanding qdisc_run calls. */
David S. Miller4335cd22008-08-17 21:58:07 -0700682 while (some_qdisc_is_busy(dev))
683 yield();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
David S. Millerb0e1e642008-07-08 17:42:10 -0700686static void dev_init_scheduler_queue(struct net_device *dev,
687 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -0700688 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -0700689{
David S. Millere8a04642008-07-17 00:34:19 -0700690 struct Qdisc *qdisc = _qdisc;
691
David S. Millerb0e1e642008-07-08 17:42:10 -0700692 dev_queue->qdisc = qdisc;
693 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700694}
695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696void dev_init_scheduler(struct net_device *dev)
697{
David S. Millere8a04642008-07-17 00:34:19 -0700698 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
David S. Miller8d50b532008-07-30 02:37:46 -0700699 dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800701 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703
David S. Millere8a04642008-07-17 00:34:19 -0700704static void shutdown_scheduler_queue(struct net_device *dev,
705 struct netdev_queue *dev_queue,
706 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
David S. Millerb0e1e642008-07-08 17:42:10 -0700708 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700709 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
David S. Millerb0e1e642008-07-08 17:42:10 -0700711 if (qdisc) {
Jarek Poplawskif7a54c12008-08-27 02:22:07 -0700712 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -0700713 dev_queue->qdisc_sleeping = qdisc_default;
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 qdisc_destroy(qdisc);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900716 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700717}
718
719void dev_shutdown(struct net_device *dev)
720{
David S. Millere8a04642008-07-17 00:34:19 -0700721 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
David S. Miller8d50b532008-07-30 02:37:46 -0700722 shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700723 WARN_ON(timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}