blob: 43abd4d27ea6d767ae9c2227510b2628bb3d2a90 [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. Millerc7e4f3b2008-07-16 03:22:39 -070032 * qdisc_root_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. Miller7698b4f2008-07-16 01:42:40 -0700129 root_lock = qdisc_root_lock(q);
130
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());
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800138 if (!netif_subqueue_stopped(dev, skb))
David S. Millerfd2ea0a2008-07-17 01:56:23 -0700139 ret = dev_hard_start_xmit(skb, dev, txq);
David S. Millerc773e842008-07-08 23:13:53 -0700140 HARD_TX_UNLOCK(dev, txq);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700141
David S. Miller7698b4f2008-07-16 01:42:40 -0700142 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700143
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700144 switch (ret) {
145 case NETDEV_TX_OK:
146 /* Driver sent out skb successfully */
147 ret = qdisc_qlen(q);
148 break;
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700149
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700150 case NETDEV_TX_LOCKED:
151 /* Driver try lock failed */
David S. Millereb6aafe2008-07-08 23:12:38 -0700152 ret = handle_dev_cpu_collision(skb, txq, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700153 break;
154
155 default:
156 /* Driver returned NETDEV_TX_BUSY - requeue skb */
157 if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
158 printk(KERN_WARNING "BUG %s code %d qlen %d\n",
159 dev->name, ret, q->q.qlen);
160
David S. Miller37437bb2008-07-16 02:15:04 -0700161 ret = dev_requeue_skb(skb, q);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700162 break;
163 }
164
David S. Miller37437bb2008-07-16 02:15:04 -0700165 if (ret && netif_tx_queue_stopped(txq))
166 ret = 0;
167
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700168 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
170
David S. Miller37437bb2008-07-16 02:15:04 -0700171void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700172{
Herbert Xu2ba25062008-03-28 16:25:26 -0700173 unsigned long start_time = jiffies;
174
David S. Miller37437bb2008-07-16 02:15:04 -0700175 while (qdisc_restart(q)) {
Herbert Xu2ba25062008-03-28 16:25:26 -0700176 /*
177 * Postpone processing if
178 * 1. another process needs the CPU;
179 * 2. we've been doing it for too long.
180 */
181 if (need_resched() || jiffies != start_time) {
David S. Miller37437bb2008-07-16 02:15:04 -0700182 __netif_schedule(q);
Herbert Xu2ba25062008-03-28 16:25:26 -0700183 break;
184 }
185 }
Herbert Xu48d83322006-06-19 23:57:59 -0700186
David S. Millere2627c82008-07-16 00:56:32 -0700187 clear_bit(__QDISC_STATE_RUNNING, &q->state);
Herbert Xu48d83322006-06-19 23:57:59 -0700188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190static void dev_watchdog(unsigned long arg)
191{
192 struct net_device *dev = (struct net_device *)arg;
193
Herbert Xu932ff272006-06-09 12:20:56 -0700194 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700195 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 if (netif_device_present(dev) &&
197 netif_running(dev) &&
198 netif_carrier_ok(dev)) {
David S. Millere8a04642008-07-17 00:34:19 -0700199 int some_queue_stopped = 0;
200 unsigned int i;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700201
David S. Millere8a04642008-07-17 00:34:19 -0700202 for (i = 0; i < dev->num_tx_queues; i++) {
203 struct netdev_queue *txq;
204
205 txq = netdev_get_tx_queue(dev, i);
206 if (netif_tx_queue_stopped(txq)) {
207 some_queue_stopped = 1;
208 break;
209 }
210 }
211
212 if (some_queue_stopped &&
213 time_after(jiffies, (dev->trans_start +
214 dev->watchdog_timeo))) {
Arjan van de Ven6579e572008-07-21 13:31:48 -0700215 char drivername[64];
216 printk(KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit timed out\n",
217 dev->name, netdev_drivername(dev, drivername, 64));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 dev->tx_timeout(dev);
Arjan van de Venb4192bb2008-05-02 16:21:07 -0700219 WARN_ON_ONCE(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 }
David S. Millere8a04642008-07-17 00:34:19 -0700221 if (!mod_timer(&dev->watchdog_timer,
222 round_jiffies(jiffies +
223 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 dev_hold(dev);
225 }
226 }
Herbert Xu932ff272006-06-09 12:20:56 -0700227 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 dev_put(dev);
230}
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232void __netdev_watchdog_up(struct net_device *dev)
233{
234 if (dev->tx_timeout) {
235 if (dev->watchdog_timeo <= 0)
236 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700237 if (!mod_timer(&dev->watchdog_timer,
238 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 dev_hold(dev);
240 }
241}
242
243static void dev_watchdog_up(struct net_device *dev)
244{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
248static void dev_watchdog_down(struct net_device *dev)
249{
Herbert Xu932ff272006-06-09 12:20:56 -0700250 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800252 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700253 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700256/**
257 * netif_carrier_on - set carrier
258 * @dev: network device
259 *
260 * Device has detected that carrier.
261 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700262void netif_carrier_on(struct net_device *dev)
263{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700264 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700265 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700266 if (netif_running(dev))
267 __netdev_watchdog_up(dev);
268 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700269}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800270EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700271
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700272/**
273 * netif_carrier_off - clear carrier
274 * @dev: network device
275 *
276 * Device has detected loss of carrier.
277 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700278void netif_carrier_off(struct net_device *dev)
279{
280 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state))
281 linkwatch_fire_event(dev);
282}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800283EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
286 under all circumstances. It is difficult to invent anything faster or
287 cheaper.
288 */
289
Thomas Graf94df1092005-06-18 22:59:08 -0700290static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291{
292 kfree_skb(skb);
293 return NET_XMIT_CN;
294}
295
Thomas Graf94df1092005-06-18 22:59:08 -0700296static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
298 return NULL;
299}
300
Thomas Graf94df1092005-06-18 22:59:08 -0700301static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 if (net_ratelimit())
Thomas Graf94df1092005-06-18 22:59:08 -0700304 printk(KERN_DEBUG "%s deferred output. It is buggy.\n",
305 skb->dev->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 kfree_skb(skb);
307 return NET_XMIT_CN;
308}
309
Eric Dumazet20fea082007-11-14 01:44:41 -0800310struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 .id = "noop",
312 .priv_size = 0,
313 .enqueue = noop_enqueue,
314 .dequeue = noop_dequeue,
315 .requeue = noop_requeue,
316 .owner = THIS_MODULE,
317};
318
David S. Miller7698b4f2008-07-16 01:42:40 -0700319static struct netdev_queue noop_netdev_queue = {
David S. Miller7698b4f2008-07-16 01:42:40 -0700320 .qdisc = &noop_qdisc,
321};
322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323struct Qdisc noop_qdisc = {
324 .enqueue = noop_enqueue,
325 .dequeue = noop_dequeue,
326 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900327 .ops = &noop_qdisc_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 .list = LIST_HEAD_INIT(noop_qdisc.list),
David S. Miller83874002008-07-17 00:53:03 -0700329 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700330 .dev_queue = &noop_netdev_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800332EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Eric Dumazet20fea082007-11-14 01:44:41 -0800334static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 .id = "noqueue",
336 .priv_size = 0,
337 .enqueue = noop_enqueue,
338 .dequeue = noop_dequeue,
339 .requeue = noop_requeue,
340 .owner = THIS_MODULE,
341};
342
David S. Miller30ee42b2008-07-18 23:00:11 -0700343static struct Qdisc noqueue_qdisc;
344static struct netdev_queue noqueue_netdev_queue = {
345 .qdisc = &noqueue_qdisc,
346};
347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348static struct Qdisc noqueue_qdisc = {
349 .enqueue = NULL,
350 .dequeue = noop_dequeue,
351 .flags = TCQ_F_BUILTIN,
352 .ops = &noqueue_qdisc_ops,
353 .list = LIST_HEAD_INIT(noqueue_qdisc.list),
David S. Miller30ee42b2008-07-18 23:00:11 -0700354 .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
355 .dev_queue = &noqueue_netdev_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356};
357
358
David S. Millerd3678b42008-07-21 09:56:13 -0700359static const u8 prio2band[TC_PRIO_MAX+1] =
360 { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 };
Thomas Graf321090e2005-06-18 22:58:35 -0700361
David S. Millerd3678b42008-07-21 09:56:13 -0700362/* 3-band FIFO queue: old style, but should be a bit faster than
363 generic prio+fifo combination.
364 */
365
366#define PFIFO_FAST_BANDS 3
367
368static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
369 struct Qdisc *qdisc)
370{
371 struct sk_buff_head *list = qdisc_priv(qdisc);
372 return list + prio2band[skb->priority & TC_PRIO_MAX];
373}
374
375static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
376{
377 struct sk_buff_head *list = prio2list(skb, qdisc);
378
379 if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
380 qdisc->q.qlen++;
Thomas Graf821d24a2005-06-18 22:58:15 -0700381 return __qdisc_enqueue_tail(skb, qdisc, list);
David S. Millerd3678b42008-07-21 09:56:13 -0700382 }
Thomas Graf821d24a2005-06-18 22:58:15 -0700383
384 return qdisc_drop(skb, qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386
David S. Millerd3678b42008-07-21 09:56:13 -0700387static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
David S. Millerd3678b42008-07-21 09:56:13 -0700389 int prio;
390 struct sk_buff_head *list = qdisc_priv(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
David S. Millerd3678b42008-07-21 09:56:13 -0700392 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
393 if (!skb_queue_empty(list + prio)) {
394 qdisc->q.qlen--;
395 return __qdisc_dequeue_head(qdisc, list + prio);
396 }
397 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 return NULL;
400}
401
David S. Millerd3678b42008-07-21 09:56:13 -0700402static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
David S. Millerd3678b42008-07-21 09:56:13 -0700404 qdisc->q.qlen++;
405 return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
David S. Millerd3678b42008-07-21 09:56:13 -0700408static void pfifo_fast_reset(struct Qdisc* qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
David S. Millerd3678b42008-07-21 09:56:13 -0700410 int prio;
411 struct sk_buff_head *list = qdisc_priv(qdisc);
412
413 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
414 __qdisc_reset_queue(qdisc, list + prio);
415
Thomas Graf821d24a2005-06-18 22:58:15 -0700416 qdisc->qstats.backlog = 0;
David S. Millerd3678b42008-07-21 09:56:13 -0700417 qdisc->q.qlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
David S. Millerd3678b42008-07-21 09:56:13 -0700420static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
421{
422 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
423
424 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1);
425 NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
426 return skb->len;
427
428nla_put_failure:
429 return -1;
430}
431
432static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
433{
434 int prio;
435 struct sk_buff_head *list = qdisc_priv(qdisc);
436
437 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
438 skb_queue_head_init(list + prio);
439
440 return 0;
441}
442
443static struct Qdisc_ops pfifo_fast_ops __read_mostly = {
444 .id = "pfifo_fast",
445 .priv_size = PFIFO_FAST_BANDS * sizeof(struct sk_buff_head),
446 .enqueue = pfifo_fast_enqueue,
447 .dequeue = pfifo_fast_dequeue,
448 .requeue = pfifo_fast_requeue,
449 .init = pfifo_fast_init,
450 .reset = pfifo_fast_reset,
451 .dump = pfifo_fast_dump,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 .owner = THIS_MODULE,
453};
454
David S. Miller5ce2d482008-07-08 17:06:30 -0700455struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
David S. Millerbb949fb2008-07-08 16:55:56 -0700456 struct Qdisc_ops *ops)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 void *p;
459 struct Qdisc *sch;
Thomas Graf3d54b822005-07-05 14:15:09 -0700460 unsigned int size;
461 int err = -ENOBUFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 /* ensure that the Qdisc and the private data are 32-byte aligned */
Thomas Graf3d54b822005-07-05 14:15:09 -0700464 size = QDISC_ALIGN(sizeof(*sch));
465 size += ops->priv_size + (QDISC_ALIGNTO - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Panagiotis Issaris0da974f2006-07-21 14:51:30 -0700467 p = kzalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (!p)
Thomas Graf3d54b822005-07-05 14:15:09 -0700469 goto errout;
Thomas Graf3d54b822005-07-05 14:15:09 -0700470 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
471 sch->padded = (char *) sch - (char *) p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 INIT_LIST_HEAD(&sch->list);
474 skb_queue_head_init(&sch->q);
475 sch->ops = ops;
476 sch->enqueue = ops->enqueue;
477 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700478 sch->dev_queue = dev_queue;
David S. Miller5ce2d482008-07-08 17:06:30 -0700479 dev_hold(qdisc_dev(sch));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 atomic_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700481
482 return sch;
483errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700484 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700485}
486
David S. Millerbb949fb2008-07-08 16:55:56 -0700487struct Qdisc * qdisc_create_dflt(struct net_device *dev,
488 struct netdev_queue *dev_queue,
489 struct Qdisc_ops *ops,
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800490 unsigned int parentid)
Thomas Graf3d54b822005-07-05 14:15:09 -0700491{
492 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900493
David S. Miller5ce2d482008-07-08 17:06:30 -0700494 sch = qdisc_alloc(dev_queue, ops);
Thomas Graf3d54b822005-07-05 14:15:09 -0700495 if (IS_ERR(sch))
496 goto errout;
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800497 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (!ops->init || ops->init(sch, NULL) == 0)
500 return sch;
501
Thomas Graf0fbbeb12005-08-23 10:12:44 -0700502 qdisc_destroy(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700503errout:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 return NULL;
505}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800506EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
David S. Miller83874002008-07-17 00:53:03 -0700508/* Under qdisc_root_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510void qdisc_reset(struct Qdisc *qdisc)
511{
Eric Dumazet20fea082007-11-14 01:44:41 -0800512 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 if (ops->reset)
515 ops->reset(qdisc);
516}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800517EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900519/* this is the rcu callback function to clean up a qdisc when there
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 * are no further references to it */
521
522static void __qdisc_destroy(struct rcu_head *head)
523{
524 struct Qdisc *qdisc = container_of(head, struct Qdisc, q_rcu);
Eric Dumazet20fea082007-11-14 01:44:41 -0800525 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
David S. Miller3a682fb2008-07-20 18:13:01 -0700527#ifdef CONFIG_NET_SCHED
Jussi Kivilinna175f9c12008-07-20 00:08:47 -0700528 qdisc_put_stab(qdisc->stab);
David S. Miller3a682fb2008-07-20 18:13:01 -0700529#endif
Patrick McHardy85670cc2006-09-27 16:45:45 -0700530 gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
Patrick McHardy85670cc2006-09-27 16:45:45 -0700531 if (ops->reset)
532 ops->reset(qdisc);
533 if (ops->destroy)
534 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Patrick McHardy85670cc2006-09-27 16:45:45 -0700536 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -0700537 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -0700538
David S. Miller83874002008-07-17 00:53:03 -0700539 kfree_skb(qdisc->gso_skb);
540
David S. Miller8a34c5d2008-07-17 00:47:45 -0700541 kfree((char *) qdisc - qdisc->padded);
542}
543
David S. Miller83874002008-07-17 00:53:03 -0700544/* Under qdisc_root_lock(qdisc) and BH! */
David S. Miller8a34c5d2008-07-17 00:47:45 -0700545
546void qdisc_destroy(struct Qdisc *qdisc)
547{
548 if (qdisc->flags & TCQ_F_BUILTIN ||
549 !atomic_dec_and_test(&qdisc->refcnt))
550 return;
551
David S. Miller30723672008-07-18 22:50:15 -0700552 if (qdisc->parent)
553 list_del(&qdisc->list);
David S. Miller8a34c5d2008-07-17 00:47:45 -0700554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 call_rcu(&qdisc->q_rcu, __qdisc_destroy);
556}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800557EXPORT_SYMBOL(qdisc_destroy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
David S. Millere8a04642008-07-17 00:34:19 -0700559static bool dev_all_qdisc_sleeping_noop(struct net_device *dev)
560{
561 unsigned int i;
562
563 for (i = 0; i < dev->num_tx_queues; i++) {
564 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
565
566 if (txq->qdisc_sleeping != &noop_qdisc)
567 return false;
568 }
569 return true;
570}
571
572static void attach_one_default_qdisc(struct net_device *dev,
573 struct netdev_queue *dev_queue,
574 void *_unused)
575{
576 struct Qdisc *qdisc;
577
578 if (dev->tx_queue_len) {
579 qdisc = qdisc_create_dflt(dev, dev_queue,
David S. Millerd3678b42008-07-21 09:56:13 -0700580 &pfifo_fast_ops, TC_H_ROOT);
David S. Millere8a04642008-07-17 00:34:19 -0700581 if (!qdisc) {
582 printk(KERN_INFO "%s: activation failed\n", dev->name);
583 return;
584 }
David S. Millere8a04642008-07-17 00:34:19 -0700585 } else {
586 qdisc = &noqueue_qdisc;
587 }
588 dev_queue->qdisc_sleeping = qdisc;
589}
590
591static void transition_one_qdisc(struct net_device *dev,
592 struct netdev_queue *dev_queue,
593 void *_need_watchdog)
594{
David S. Miller83874002008-07-17 00:53:03 -0700595 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700596 int *need_watchdog_p = _need_watchdog;
597
David S. Miller83874002008-07-17 00:53:03 -0700598 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
599 if (new_qdisc != &noqueue_qdisc)
David S. Millere8a04642008-07-17 00:34:19 -0700600 *need_watchdog_p = 1;
David S. Millere8a04642008-07-17 00:34:19 -0700601}
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603void dev_activate(struct net_device *dev)
604{
David S. Millere8a04642008-07-17 00:34:19 -0700605 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -0700606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 /* No queueing discipline is attached to device;
David S. Millerd3678b42008-07-21 09:56:13 -0700608 create default one i.e. pfifo_fast for devices,
609 which need queueing and noqueue_qdisc for
610 virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
612
David S. Millere8a04642008-07-17 00:34:19 -0700613 if (dev_all_qdisc_sleeping_noop(dev))
614 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Tommy S. Christensencacaddf2005-05-03 16:18:52 -0700616 if (!netif_carrier_ok(dev))
617 /* Delay activation until next carrier-on event */
618 return;
619
David S. Millere8a04642008-07-17 00:34:19 -0700620 need_watchdog = 0;
621 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
622
623 if (need_watchdog) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 dev->trans_start = jiffies;
625 dev_watchdog_up(dev);
626 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700627}
628
David S. Millere8a04642008-07-17 00:34:19 -0700629static void dev_deactivate_queue(struct net_device *dev,
630 struct netdev_queue *dev_queue,
631 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -0700632{
David S. Millere8a04642008-07-17 00:34:19 -0700633 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -0700634 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700635
David S. Miller970565b2008-07-08 23:10:33 -0700636 qdisc = dev_queue->qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700637 if (qdisc) {
David S. Miller83874002008-07-17 00:53:03 -0700638 spin_lock_bh(qdisc_lock(qdisc));
639
David S. Millerb0e1e642008-07-08 17:42:10 -0700640 dev_queue->qdisc = qdisc_default;
641 qdisc_reset(qdisc);
David S. Millerd3b753d2008-07-15 20:14:35 -0700642
David S. Miller83874002008-07-17 00:53:03 -0700643 spin_unlock_bh(qdisc_lock(qdisc));
David S. Millerb0e1e642008-07-08 17:42:10 -0700644 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
David S. Millere8a04642008-07-17 00:34:19 -0700647static bool some_qdisc_is_running(struct net_device *dev, int lock)
648{
649 unsigned int i;
650
651 for (i = 0; i < dev->num_tx_queues; i++) {
652 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -0700653 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -0700654 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -0700655 int val;
656
657 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millere2627c82008-07-16 00:56:32 -0700658 q = dev_queue->qdisc;
David S. Miller7698b4f2008-07-16 01:42:40 -0700659 root_lock = qdisc_root_lock(q);
David S. Millere8a04642008-07-17 00:34:19 -0700660
661 if (lock)
David S. Miller7698b4f2008-07-16 01:42:40 -0700662 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700663
David S. Millere2627c82008-07-16 00:56:32 -0700664 val = test_bit(__QDISC_STATE_RUNNING, &q->state);
David S. Millere8a04642008-07-17 00:34:19 -0700665
666 if (lock)
David S. Miller7698b4f2008-07-16 01:42:40 -0700667 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -0700668
669 if (val)
670 return true;
671 }
672 return false;
673}
674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675void dev_deactivate(struct net_device *dev)
676{
David S. Millere8a04642008-07-17 00:34:19 -0700677 bool running;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
David S. Millere8a04642008-07-17 00:34:19 -0700679 netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc);
Herbert Xu41a23b02007-05-10 14:12:47 -0700680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 dev_watchdog_down(dev);
682
Herbert Xuce0e32e2007-10-18 22:37:58 -0700683 /* Wait for outstanding qdisc-less dev_queue_xmit calls. */
Herbert Xud4828d82006-06-22 02:28:18 -0700684 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Herbert Xud4828d82006-06-22 02:28:18 -0700686 /* Wait for outstanding qdisc_run calls. */
Herbert Xuce0e32e2007-10-18 22:37:58 -0700687 do {
David S. Millere8a04642008-07-17 00:34:19 -0700688 while (some_qdisc_is_running(dev, 0))
Herbert Xuce0e32e2007-10-18 22:37:58 -0700689 yield();
690
691 /*
692 * Double-check inside queue lock to ensure that all effects
693 * of the queue run are visible when we return.
694 */
David S. Millere8a04642008-07-17 00:34:19 -0700695 running = some_qdisc_is_running(dev, 1);
Herbert Xuce0e32e2007-10-18 22:37:58 -0700696
697 /*
698 * The running flag should never be set at this point because
699 * we've already set dev->qdisc to noop_qdisc *inside* the same
700 * pair of spin locks. That is, if any qdisc_run starts after
701 * our initial test it should see the noop_qdisc and then
702 * clear the RUNNING bit before dropping the queue lock. So
703 * if it is set here then we've found a bug.
704 */
705 } while (WARN_ON_ONCE(running));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706}
707
David S. Millerb0e1e642008-07-08 17:42:10 -0700708static void dev_init_scheduler_queue(struct net_device *dev,
709 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -0700710 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -0700711{
David S. Millere8a04642008-07-17 00:34:19 -0700712 struct Qdisc *qdisc = _qdisc;
713
David S. Millerb0e1e642008-07-08 17:42:10 -0700714 dev_queue->qdisc = qdisc;
715 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -0700716}
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718void dev_init_scheduler(struct net_device *dev)
719{
David S. Millere8a04642008-07-17 00:34:19 -0700720 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
David S. Millerb0e1e642008-07-08 17:42:10 -0700721 dev_init_scheduler_queue(dev, &dev->rx_queue, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800723 setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
David S. Millere8a04642008-07-17 00:34:19 -0700726static void shutdown_scheduler_queue(struct net_device *dev,
727 struct netdev_queue *dev_queue,
728 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
David S. Millerb0e1e642008-07-08 17:42:10 -0700730 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -0700731 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
David S. Millerb0e1e642008-07-08 17:42:10 -0700733 if (qdisc) {
David S. Miller17715e62008-07-16 02:36:04 -0700734 spinlock_t *root_lock = qdisc_root_lock(qdisc);
735
David S. Millerb0e1e642008-07-08 17:42:10 -0700736 dev_queue->qdisc = qdisc_default;
737 dev_queue->qdisc_sleeping = qdisc_default;
738
David S. Millercffe1c52008-07-25 01:25:04 -0700739 spin_lock_bh(root_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 qdisc_destroy(qdisc);
David S. Millercffe1c52008-07-25 01:25:04 -0700741 spin_unlock_bh(root_lock);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900742 }
David S. Millerb0e1e642008-07-08 17:42:10 -0700743}
744
745void dev_shutdown(struct net_device *dev)
746{
David S. Millere8a04642008-07-17 00:34:19 -0700747 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
748 shutdown_scheduler_queue(dev, &dev->rx_queue, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 BUG_TRAP(!timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}