blob: d9ac60ffe927c6a0ebced39a5375e8c4444ea3ca [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * net/sched/sch_generic.c Generic packet scheduler routines.
4 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 * Jamal Hadi Salim, <hadi@cyberus.ca> 990601
7 * - Ingress support
8 */
9
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/module.h>
12#include <linux/types.h>
13#include <linux/kernel.h>
14#include <linux/sched.h>
15#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/netdevice.h>
18#include <linux/skbuff.h>
19#include <linux/rtnetlink.h>
20#include <linux/init.h>
21#include <linux/rcupdate.h>
22#include <linux/list.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +020024#include <linux/if_vlan.h>
John Fastabendc5ad1192017-12-07 09:58:19 -080025#include <linux/skb_array.h>
Chris Dion32d3e512017-12-06 10:50:28 -050026#include <linux/if_macvlan.h>
Jiri Pirko292f1c72013-02-12 00:12:03 +000027#include <net/sch_generic.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>
Jesper Dangaard Brouere5430022017-08-15 21:11:03 +020030#include <trace/events/qdisc.h>
Cong Wang141b6b22019-05-01 19:56:59 -070031#include <trace/events/net.h>
Steffen Klassertf53c7232017-12-20 10:41:36 +010032#include <net/xfrm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
stephen hemminger34aedd32013-08-31 10:15:33 -070034/* Qdisc to use by default */
35const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
36EXPORT_SYMBOL(default_qdisc_ops);
37
Yunsheng Lindcad9ee2021-05-14 11:17:01 +080038static void qdisc_maybe_clear_missed(struct Qdisc *q,
39 const struct netdev_queue *txq)
40{
41 clear_bit(__QDISC_STATE_MISSED, &q->state);
42
43 /* Make sure the below netif_xmit_frozen_or_stopped()
44 * checking happens after clearing STATE_MISSED.
45 */
46 smp_mb__after_atomic();
47
48 /* Checking netif_xmit_frozen_or_stopped() again to
49 * make sure STATE_MISSED is set if the STATE_MISSED
50 * set by netif_tx_wake_queue()'s rescheduling of
51 * net_tx_action() is cleared by the above clear_bit().
52 */
53 if (!netif_xmit_frozen_or_stopped(txq))
54 set_bit(__QDISC_STATE_MISSED, &q->state);
Yunsheng Linc4fef012021-06-22 14:49:56 +080055 else
56 set_bit(__QDISC_STATE_DRAINING, &q->state);
Yunsheng Lindcad9ee2021-05-14 11:17:01 +080057}
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/* Main transmission queue. */
60
Patrick McHardy0463d4a2007-04-16 17:02:10 -070061/* Modifications to data participating in scheduling must be protected with
David S. Miller5fb66222008-08-02 20:02:43 -070062 * qdisc_lock(qdisc) spinlock.
Patrick McHardy0463d4a2007-04-16 17:02:10 -070063 *
64 * The idea is the following:
David S. Millerc7e4f3b2008-07-16 03:22:39 -070065 * - enqueue, dequeue are serialized via qdisc root lock
66 * - ingress filtering is also serialized via qdisc root lock
Patrick McHardy0463d4a2007-04-16 17:02:10 -070067 * - updates to tree and tree walking are only done under the rtnl mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 */
John Fastabend70e57d52017-12-07 09:56:23 -080069
Eric Dumazetb88dd522019-09-05 05:20:22 -070070#define SKB_XOFF_MAGIC ((struct sk_buff *)1UL)
71
John Fastabend70e57d52017-12-07 09:56:23 -080072static inline struct sk_buff *__skb_dequeue_bad_txq(struct Qdisc *q)
73{
74 const struct netdev_queue *txq = q->dev_queue;
75 spinlock_t *lock = NULL;
76 struct sk_buff *skb;
77
78 if (q->flags & TCQ_F_NOLOCK) {
79 lock = qdisc_lock(q);
80 spin_lock(lock);
81 }
82
83 skb = skb_peek(&q->skb_bad_txq);
84 if (skb) {
85 /* check the reason of requeuing without tx lock first */
86 txq = skb_get_tx_queue(txq->dev, skb);
87 if (!netif_xmit_frozen_or_stopped(txq)) {
88 skb = __skb_dequeue(&q->skb_bad_txq);
89 if (qdisc_is_percpu_stats(q)) {
90 qdisc_qstats_cpu_backlog_dec(q, skb);
Paolo Abeni73eb6282019-04-10 14:32:41 +020091 qdisc_qstats_cpu_qlen_dec(q);
John Fastabend70e57d52017-12-07 09:56:23 -080092 } else {
93 qdisc_qstats_backlog_dec(q, skb);
94 q->q.qlen--;
95 }
96 } else {
Eric Dumazetb88dd522019-09-05 05:20:22 -070097 skb = SKB_XOFF_MAGIC;
Yunsheng Lindcad9ee2021-05-14 11:17:01 +080098 qdisc_maybe_clear_missed(q, txq);
John Fastabend70e57d52017-12-07 09:56:23 -080099 }
100 }
101
102 if (lock)
103 spin_unlock(lock);
104
105 return skb;
106}
107
108static inline struct sk_buff *qdisc_dequeue_skb_bad_txq(struct Qdisc *q)
109{
110 struct sk_buff *skb = skb_peek(&q->skb_bad_txq);
111
112 if (unlikely(skb))
113 skb = __skb_dequeue_bad_txq(q);
114
115 return skb;
116}
117
118static inline void qdisc_enqueue_skb_bad_txq(struct Qdisc *q,
119 struct sk_buff *skb)
120{
121 spinlock_t *lock = NULL;
122
123 if (q->flags & TCQ_F_NOLOCK) {
124 lock = qdisc_lock(q);
125 spin_lock(lock);
126 }
127
128 __skb_queue_tail(&q->skb_bad_txq, skb);
129
Eric Dumazetcce6294c2018-03-14 18:53:00 -0700130 if (qdisc_is_percpu_stats(q)) {
131 qdisc_qstats_cpu_backlog_inc(q, skb);
Paolo Abeni73eb6282019-04-10 14:32:41 +0200132 qdisc_qstats_cpu_qlen_inc(q);
Eric Dumazetcce6294c2018-03-14 18:53:00 -0700133 } else {
134 qdisc_qstats_backlog_inc(q, skb);
135 q->q.qlen++;
136 }
137
John Fastabend70e57d52017-12-07 09:56:23 -0800138 if (lock)
139 spin_unlock(lock);
140}
141
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200142static inline void dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700143{
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200144 spinlock_t *lock = NULL;
Wei Yongjun9540d972017-12-27 17:05:52 +0800145
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200146 if (q->flags & TCQ_F_NOLOCK) {
147 lock = qdisc_lock(q);
148 spin_lock(lock);
Wei Yongjun9540d972017-12-27 17:05:52 +0800149 }
Jarek Poplawski62523522008-10-06 10:41:50 -0700150
Wei Yongjun9540d972017-12-27 17:05:52 +0800151 while (skb) {
152 struct sk_buff *next = skb->next;
153
154 __skb_queue_tail(&q->gso_skb, skb);
155
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200156 /* it's still part of the queue */
157 if (qdisc_is_percpu_stats(q)) {
158 qdisc_qstats_cpu_requeues_inc(q);
159 qdisc_qstats_cpu_backlog_inc(q, skb);
Paolo Abeni73eb6282019-04-10 14:32:41 +0200160 qdisc_qstats_cpu_qlen_inc(q);
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200161 } else {
162 q->qstats.requeues++;
163 qdisc_qstats_backlog_inc(q, skb);
164 q->q.qlen++;
165 }
Wei Yongjun9540d972017-12-27 17:05:52 +0800166
167 skb = next;
168 }
Yunsheng Linc4fef012021-06-22 14:49:56 +0800169
170 if (lock) {
Paolo Abeni9c01c9f2019-04-10 14:32:39 +0200171 spin_unlock(lock);
Yunsheng Linc4fef012021-06-22 14:49:56 +0800172 set_bit(__QDISC_STATE_MISSED, &q->state);
173 } else {
174 __netif_schedule(q);
175 }
John Fastabenda53851e2017-12-07 09:55:45 -0800176}
177
Eric Dumazet55a93b32014-10-03 15:31:07 -0700178static void try_bulk_dequeue_skb(struct Qdisc *q,
179 struct sk_buff *skb,
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200180 const struct netdev_queue *txq,
181 int *packets)
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200182{
Eric Dumazet55a93b32014-10-03 15:31:07 -0700183 int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200184
185 while (bytelimit > 0) {
Eric Dumazet55a93b32014-10-03 15:31:07 -0700186 struct sk_buff *nskb = q->dequeue(q);
187
188 if (!nskb)
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200189 break;
190
Eric Dumazet55a93b32014-10-03 15:31:07 -0700191 bytelimit -= nskb->len; /* covers GSO len */
192 skb->next = nskb;
193 skb = nskb;
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200194 (*packets)++; /* GSO counts as one pkt */
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200195 }
David S. Millera8305bf2018-07-29 20:42:53 -0700196 skb_mark_not_on_list(skb);
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200197}
198
Eric Dumazet4d202a02016-06-21 23:16:52 -0700199/* This variant of try_bulk_dequeue_skb() makes sure
200 * all skbs in the chain are for the same txq
201 */
202static void try_bulk_dequeue_skb_slow(struct Qdisc *q,
203 struct sk_buff *skb,
204 int *packets)
205{
206 int mapping = skb_get_queue_mapping(skb);
207 struct sk_buff *nskb;
208 int cnt = 0;
209
210 do {
211 nskb = q->dequeue(q);
212 if (!nskb)
213 break;
214 if (unlikely(skb_get_queue_mapping(nskb) != mapping)) {
John Fastabend70e57d52017-12-07 09:56:23 -0800215 qdisc_enqueue_skb_bad_txq(q, nskb);
Eric Dumazet4d202a02016-06-21 23:16:52 -0700216 break;
217 }
218 skb->next = nskb;
219 skb = nskb;
220 } while (++cnt < 8);
221 (*packets) += cnt;
David S. Millera8305bf2018-07-29 20:42:53 -0700222 skb_mark_not_on_list(skb);
Eric Dumazet4d202a02016-06-21 23:16:52 -0700223}
224
Jesper Dangaard Brouer5772e9a2014-10-01 22:35:59 +0200225/* Note that dequeue_skb can possibly return a SKB list (via skb->next).
226 * A requeued skb (via q->gso_skb) can also be a SKB list.
227 */
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200228static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
229 int *packets)
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700230{
Eric Dumazet1abbe132012-12-11 15:54:33 +0000231 const struct netdev_queue *txq = q->dev_queue;
John Fastabendfd8e8d12017-12-07 09:56:42 -0800232 struct sk_buff *skb = NULL;
Jarek Poplawski554794d2008-10-06 09:54:39 -0700233
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200234 *packets = 1;
John Fastabenda53851e2017-12-07 09:55:45 -0800235 if (unlikely(!skb_queue_empty(&q->gso_skb))) {
236 spinlock_t *lock = NULL;
237
238 if (q->flags & TCQ_F_NOLOCK) {
239 lock = qdisc_lock(q);
240 spin_lock(lock);
241 }
242
243 skb = skb_peek(&q->gso_skb);
244
245 /* skb may be null if another cpu pulls gso_skb off in between
246 * empty check and lock.
247 */
248 if (!skb) {
249 if (lock)
250 spin_unlock(lock);
251 goto validate;
252 }
253
Eric Dumazet4d202a02016-06-21 23:16:52 -0700254 /* skb in gso_skb were already validated */
255 *validate = false;
Steffen Klassertf53c7232017-12-20 10:41:36 +0100256 if (xfrm_offload(skb))
257 *validate = true;
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700258 /* check the reason of requeuing without tx lock first */
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200259 txq = skb_get_tx_queue(txq->dev, skb);
Tom Herbert734664982011-11-28 16:32:44 +0000260 if (!netif_xmit_frozen_or_stopped(txq)) {
John Fastabenda53851e2017-12-07 09:55:45 -0800261 skb = __skb_dequeue(&q->gso_skb);
262 if (qdisc_is_percpu_stats(q)) {
263 qdisc_qstats_cpu_backlog_dec(q, skb);
Paolo Abeni73eb6282019-04-10 14:32:41 +0200264 qdisc_qstats_cpu_qlen_dec(q);
John Fastabenda53851e2017-12-07 09:55:45 -0800265 } else {
266 qdisc_qstats_backlog_dec(q, skb);
267 q->q.qlen--;
268 }
269 } else {
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700270 skb = NULL;
Yunsheng Lindcad9ee2021-05-14 11:17:01 +0800271 qdisc_maybe_clear_missed(q, txq);
John Fastabenda53851e2017-12-07 09:55:45 -0800272 }
273 if (lock)
274 spin_unlock(lock);
Jesper Dangaard Brouere5430022017-08-15 21:11:03 +0200275 goto trace;
Eric Dumazet4d202a02016-06-21 23:16:52 -0700276 }
John Fastabenda53851e2017-12-07 09:55:45 -0800277validate:
Eric Dumazet4d202a02016-06-21 23:16:52 -0700278 *validate = true;
John Fastabendfd8e8d12017-12-07 09:56:42 -0800279
280 if ((q->flags & TCQ_F_ONETXQUEUE) &&
Yunsheng Lindcad9ee2021-05-14 11:17:01 +0800281 netif_xmit_frozen_or_stopped(txq)) {
282 qdisc_maybe_clear_missed(q, txq);
John Fastabendfd8e8d12017-12-07 09:56:42 -0800283 return skb;
Yunsheng Lindcad9ee2021-05-14 11:17:01 +0800284 }
John Fastabendfd8e8d12017-12-07 09:56:42 -0800285
John Fastabend70e57d52017-12-07 09:56:23 -0800286 skb = qdisc_dequeue_skb_bad_txq(q);
Eric Dumazetb88dd522019-09-05 05:20:22 -0700287 if (unlikely(skb)) {
288 if (skb == SKB_XOFF_MAGIC)
289 return NULL;
John Fastabend70e57d52017-12-07 09:56:23 -0800290 goto bulk;
Eric Dumazetb88dd522019-09-05 05:20:22 -0700291 }
John Fastabendfd8e8d12017-12-07 09:56:42 -0800292 skb = q->dequeue(q);
Eric Dumazet4d202a02016-06-21 23:16:52 -0700293 if (skb) {
294bulk:
295 if (qdisc_may_bulk(q))
296 try_bulk_dequeue_skb(q, skb, txq, packets);
297 else
298 try_bulk_dequeue_skb_slow(q, skb, packets);
Jarek Poplawskiebf05982008-09-22 22:16:23 -0700299 }
Jesper Dangaard Brouere5430022017-08-15 21:11:03 +0200300trace:
301 trace_qdisc_dequeue(q, txq, *packets, skb);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700302 return skb;
303}
304
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900305/*
Jesper Dangaard Brouer10770bc2014-09-02 16:35:33 +0200306 * Transmit possibly several skbs, and handle the return status as
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700307 * required. Owning running seqcount bit guarantees that
Jesper Dangaard Brouer10770bc2014-09-02 16:35:33 +0200308 * only one CPU can execute this function.
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700309 *
310 * Returns to the caller:
John Fastabend29b86cd2017-12-07 09:54:47 -0800311 * false - hardware queue frozen backoff
312 * true - feel free to send more pkts
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700313 */
John Fastabend29b86cd2017-12-07 09:54:47 -0800314bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
315 struct net_device *dev, struct netdev_queue *txq,
316 spinlock_t *root_lock, bool validate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
Peter P Waskiewicz Jr5f1a4852007-11-13 20:40:55 -0800318 int ret = NETDEV_TX_BUSY;
Steffen Klassertf53c7232017-12-20 10:41:36 +0100319 bool again = false;
David S. Miller7698b4f2008-07-16 01:42:40 -0700320
321 /* And release qdisc */
John Fastabend6b3ba912017-12-07 09:54:25 -0800322 if (root_lock)
323 spin_unlock(root_lock);
Herbert Xud90df3a2007-05-10 04:55:14 -0700324
Eric Dumazet55a93b32014-10-03 15:31:07 -0700325 /* Note that we validate skb (GSO, checksum, ...) outside of locks */
326 if (validate)
Steffen Klassertf53c7232017-12-20 10:41:36 +0100327 skb = validate_xmit_skb_list(skb, dev, &again);
328
329#ifdef CONFIG_XFRM_OFFLOAD
330 if (unlikely(again)) {
331 if (root_lock)
332 spin_lock(root_lock);
333
334 dev_requeue_skb(skb, q);
335 return false;
336 }
337#endif
Patrick McHardy572a9d72009-11-10 06:14:14 +0000338
Lars Persson3dcd493fb2016-04-12 08:45:52 +0200339 if (likely(skb)) {
Eric Dumazet55a93b32014-10-03 15:31:07 -0700340 HARD_TX_LOCK(dev, txq, smp_processor_id());
341 if (!netif_xmit_frozen_or_stopped(txq))
342 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
Yunsheng Lindcad9ee2021-05-14 11:17:01 +0800343 else
344 qdisc_maybe_clear_missed(q, txq);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700345
Eric Dumazet55a93b32014-10-03 15:31:07 -0700346 HARD_TX_UNLOCK(dev, txq);
Lars Persson3dcd493fb2016-04-12 08:45:52 +0200347 } else {
John Fastabend6b3ba912017-12-07 09:54:25 -0800348 if (root_lock)
349 spin_lock(root_lock);
John Fastabend29b86cd2017-12-07 09:54:47 -0800350 return true;
Eric Dumazet55a93b32014-10-03 15:31:07 -0700351 }
John Fastabend6b3ba912017-12-07 09:54:25 -0800352
353 if (root_lock)
354 spin_lock(root_lock);
Jamal Hadi Salimc716a812007-06-10 17:31:24 -0700355
John Fastabend29b86cd2017-12-07 09:54:47 -0800356 if (!dev_xmit_complete(ret)) {
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700357 /* Driver returned NETDEV_TX_BUSY - requeue skb */
Joe Perchese87cc472012-05-13 21:56:26 +0000358 if (unlikely(ret != NETDEV_TX_BUSY))
359 net_warn_ratelimited("BUG %s code %d qlen %d\n",
360 dev->name, ret, q->q.qlen);
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700361
John Fastabend29b86cd2017-12-07 09:54:47 -0800362 dev_requeue_skb(skb, q);
363 return false;
Krishna Kumar6c1361a2007-06-24 19:56:09 -0700364 }
365
John Fastabend29b86cd2017-12-07 09:54:47 -0800366 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000369/*
370 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
371 *
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700372 * running seqcount guarantees only one CPU can process
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000373 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
374 * this queue.
375 *
376 * netif_tx_lock serializes accesses to device driver.
377 *
378 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
379 * if one is grabbed, another must be free.
380 *
381 * Note, that this procedure can be called by a watchdog timer
382 *
383 * Returns to the caller:
384 * 0 - queue is empty or throttled.
385 * >0 - queue is not empty.
386 *
387 */
John Fastabend29b86cd2017-12-07 09:54:47 -0800388static inline bool qdisc_restart(struct Qdisc *q, int *packets)
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000389{
John Fastabend6b3ba912017-12-07 09:54:25 -0800390 spinlock_t *root_lock = NULL;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000391 struct netdev_queue *txq;
392 struct net_device *dev;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000393 struct sk_buff *skb;
Paolo Abeni32f7b442018-05-15 10:50:31 +0200394 bool validate;
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000395
396 /* Dequeue packet */
John Fastabendeb82a992018-03-24 22:25:06 -0700397 skb = dequeue_skb(q, &validate, packets);
Paolo Abeni32f7b442018-05-15 10:50:31 +0200398 if (unlikely(!skb))
John Fastabendeb82a992018-03-24 22:25:06 -0700399 return false;
John Fastabendeb82a992018-03-24 22:25:06 -0700400
Paolo Abeni32f7b442018-05-15 10:50:31 +0200401 if (!(q->flags & TCQ_F_NOLOCK))
John Fastabend6b3ba912017-12-07 09:54:25 -0800402 root_lock = qdisc_lock(q);
403
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000404 dev = qdisc_dev(q);
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200405 txq = skb_get_tx_queue(dev, skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000406
Paolo Abeni32f7b442018-05-15 10:50:31 +0200407 return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000408}
409
David S. Miller37437bb2008-07-16 02:15:04 -0700410void __qdisc_run(struct Qdisc *q)
Herbert Xu48d83322006-06-19 23:57:59 -0700411{
Matthias Tafelmeier3d48b532016-12-29 21:37:21 +0100412 int quota = dev_tx_weight;
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200413 int packets;
Herbert Xu2ba25062008-03-28 16:25:26 -0700414
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200415 while (qdisc_restart(q, &packets)) {
Jesper Dangaard Brouerb8358d72014-10-09 12:18:10 +0200416 quota -= packets;
Eric Dumazetb60fa1c2019-10-01 14:02:36 -0700417 if (quota <= 0) {
Yunsheng Linc4fef012021-06-22 14:49:56 +0800418 if (q->flags & TCQ_F_NOLOCK)
419 set_bit(__QDISC_STATE_MISSED, &q->state);
420 else
421 __netif_schedule(q);
422
Herbert Xu2ba25062008-03-28 16:25:26 -0700423 break;
424 }
425 }
Herbert Xu48d83322006-06-19 23:57:59 -0700426}
427
Eric Dumazet9d214932009-05-17 20:55:16 -0700428unsigned long dev_trans_start(struct net_device *dev)
429{
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200430 unsigned long val, res;
Eric Dumazet9d214932009-05-17 20:55:16 -0700431 unsigned int i;
432
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200433 if (is_vlan_dev(dev))
434 dev = vlan_dev_real_dev(dev);
Chris Dion32d3e512017-12-06 10:50:28 -0500435 else if (netif_is_macvlan(dev))
436 dev = macvlan_dev_real_dev(dev);
Florian Westphal9b366272016-05-03 16:33:14 +0200437 res = netdev_get_tx_queue(dev, 0)->trans_start;
438 for (i = 1; i < dev->num_tx_queues; i++) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700439 val = netdev_get_tx_queue(dev, i)->trans_start;
440 if (val && time_after(val, res))
441 res = val;
442 }
nikolay@redhat.com07ce76a2013-08-03 22:07:47 +0200443
Eric Dumazet9d214932009-05-17 20:55:16 -0700444 return res;
445}
446EXPORT_SYMBOL(dev_trans_start);
447
Kees Cookcdeabbb2017-10-16 17:29:17 -0700448static void dev_watchdog(struct timer_list *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
Kees Cookcdeabbb2017-10-16 17:29:17 -0700450 struct net_device *dev = from_timer(dev, t, watchdog_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Herbert Xu932ff272006-06-09 12:20:56 -0700452 netif_tx_lock(dev);
David S. Millere8a04642008-07-17 00:34:19 -0700453 if (!qdisc_tx_is_noop(dev)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 if (netif_device_present(dev) &&
455 netif_running(dev) &&
456 netif_carrier_ok(dev)) {
Eric Dumazet9d214932009-05-17 20:55:16 -0700457 int some_queue_timedout = 0;
David S. Millere8a04642008-07-17 00:34:19 -0700458 unsigned int i;
Eric Dumazet9d214932009-05-17 20:55:16 -0700459 unsigned long trans_start;
Stephen Hemminger338f7562006-05-16 15:02:12 -0700460
David S. Millere8a04642008-07-17 00:34:19 -0700461 for (i = 0; i < dev->num_tx_queues; i++) {
462 struct netdev_queue *txq;
463
464 txq = netdev_get_tx_queue(dev, i);
Florian Westphal9b366272016-05-03 16:33:14 +0200465 trans_start = txq->trans_start;
Tom Herbert734664982011-11-28 16:32:44 +0000466 if (netif_xmit_stopped(txq) &&
Eric Dumazet9d214932009-05-17 20:55:16 -0700467 time_after(jiffies, (trans_start +
468 dev->watchdog_timeo))) {
469 some_queue_timedout = 1;
david decotignyccf5ff62011-11-16 12:15:10 +0000470 txq->trans_timeout++;
David S. Millere8a04642008-07-17 00:34:19 -0700471 break;
472 }
473 }
474
Eric Dumazet9d214932009-05-17 20:55:16 -0700475 if (some_queue_timedout) {
Cong Wang141b6b22019-05-01 19:56:59 -0700476 trace_net_dev_xmit_timeout(dev, i);
Eric Dumazet9d214932009-05-17 20:55:16 -0700477 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
David S. Miller3019de12011-06-06 16:41:33 -0700478 dev->name, netdev_drivername(dev), i);
Michael S. Tsirkin0290bd22019-12-10 09:23:51 -0500479 dev->netdev_ops->ndo_tx_timeout(dev, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 }
David S. Millere8a04642008-07-17 00:34:19 -0700481 if (!mod_timer(&dev->watchdog_timer,
482 round_jiffies(jiffies +
483 dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 dev_hold(dev);
485 }
486 }
Herbert Xu932ff272006-06-09 12:20:56 -0700487 netif_tx_unlock(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 dev_put(dev);
490}
491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492void __netdev_watchdog_up(struct net_device *dev)
493{
Stephen Hemmingerd3147742008-11-19 21:32:24 -0800494 if (dev->netdev_ops->ndo_tx_timeout) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 if (dev->watchdog_timeo <= 0)
496 dev->watchdog_timeo = 5*HZ;
Venkatesh Pallipadi60468d52007-05-31 21:28:44 -0700497 if (!mod_timer(&dev->watchdog_timer,
498 round_jiffies(jiffies + dev->watchdog_timeo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 dev_hold(dev);
500 }
501}
Valentin Longchamp1a3db272020-06-09 22:11:54 +0200502EXPORT_SYMBOL_GPL(__netdev_watchdog_up);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504static void dev_watchdog_up(struct net_device *dev)
505{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 __netdev_watchdog_up(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
509static void dev_watchdog_down(struct net_device *dev)
510{
Herbert Xu932ff272006-06-09 12:20:56 -0700511 netif_tx_lock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 if (del_timer(&dev->watchdog_timer))
Stephen Hemminger15333062006-03-20 22:32:28 -0800513 dev_put(dev);
Herbert Xu932ff272006-06-09 12:20:56 -0700514 netif_tx_unlock_bh(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700517/**
518 * netif_carrier_on - set carrier
519 * @dev: network device
520 *
Jouke Witteveen989723b2019-02-07 17:14:32 +0100521 * Device has detected acquisition of carrier.
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700522 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700523void netif_carrier_on(struct net_device *dev)
524{
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700525 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
David S. Millerb4730012008-11-19 15:33:54 -0800526 if (dev->reg_state == NETREG_UNINITIALIZED)
527 return;
David Decotignyb2d3bcf2018-01-18 09:59:13 -0800528 atomic_inc(&dev->carrier_up_count);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700529 linkwatch_fire_event(dev);
Jeff Garzikbfaae0f2007-10-17 23:26:43 -0700530 if (netif_running(dev))
531 __netdev_watchdog_up(dev);
532 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700533}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800534EXPORT_SYMBOL(netif_carrier_on);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700535
Stephen Hemmingerbea33482007-10-03 16:41:36 -0700536/**
537 * netif_carrier_off - clear carrier
538 * @dev: network device
539 *
540 * Device has detected loss of carrier.
541 */
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700542void netif_carrier_off(struct net_device *dev)
543{
David S. Millerb4730012008-11-19 15:33:54 -0800544 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
545 if (dev->reg_state == NETREG_UNINITIALIZED)
546 return;
David Decotignyb2d3bcf2018-01-18 09:59:13 -0800547 atomic_inc(&dev->carrier_down_count);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700548 linkwatch_fire_event(dev);
David S. Millerb4730012008-11-19 15:33:54 -0800549 }
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700550}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800551EXPORT_SYMBOL(netif_carrier_off);
Denis Vlasenko0a242ef2005-08-11 15:32:53 -0700552
Jakub Kicinski490dcec2021-05-19 10:18:25 -0700553/**
554 * netif_carrier_event - report carrier state event
555 * @dev: network device
556 *
557 * Device has detected a carrier event but the carrier state wasn't changed.
558 * Use in drivers when querying carrier state asynchronously, to avoid missing
559 * events (link flaps) if link recovers before it's queried.
560 */
561void netif_carrier_event(struct net_device *dev)
562{
563 if (dev->reg_state == NETREG_UNINITIALIZED)
564 return;
565 atomic_inc(&dev->carrier_up_count);
566 atomic_inc(&dev->carrier_down_count);
567 linkwatch_fire_event(dev);
568}
569EXPORT_SYMBOL_GPL(netif_carrier_event);
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/* "NOOP" scheduler: the best scheduler, recommended for all interfaces
572 under all circumstances. It is difficult to invent anything faster or
573 cheaper.
574 */
575
Eric Dumazet520ac302016-06-21 23:16:49 -0700576static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
577 struct sk_buff **to_free)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Eric Dumazet520ac302016-06-21 23:16:49 -0700579 __qdisc_drop(skb, to_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 return NET_XMIT_CN;
581}
582
Yang Yingliang82d567c2013-12-10 20:55:31 +0800583static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
585 return NULL;
586}
587
Eric Dumazet20fea082007-11-14 01:44:41 -0800588struct Qdisc_ops noop_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 .id = "noop",
590 .priv_size = 0,
591 .enqueue = noop_enqueue,
592 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700593 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 .owner = THIS_MODULE,
595};
596
David S. Miller7698b4f2008-07-16 01:42:40 -0700597static struct netdev_queue noop_netdev_queue = {
Li RongQing3b40bf42019-02-25 10:43:06 +0800598 RCU_POINTER_INITIALIZER(qdisc, &noop_qdisc),
Jarek Poplawski9f3ffae2008-10-19 23:37:47 -0700599 .qdisc_sleeping = &noop_qdisc,
David S. Miller7698b4f2008-07-16 01:42:40 -0700600};
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602struct Qdisc noop_qdisc = {
603 .enqueue = noop_enqueue,
604 .dequeue = noop_dequeue,
605 .flags = TCQ_F_BUILTIN,
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900606 .ops = &noop_qdisc_ops,
David S. Miller838740002008-07-17 00:53:03 -0700607 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
David S. Miller7698b4f2008-07-16 01:42:40 -0700608 .dev_queue = &noop_netdev_queue,
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700609 .running = SEQCNT_ZERO(noop_qdisc.running),
Eric Dumazet7b5edbc2010-10-15 19:22:34 +0000610 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
Eric Dumazetf98ebd42018-10-09 15:20:50 -0700611 .gso_skb = {
612 .next = (struct sk_buff *)&noop_qdisc.gso_skb,
613 .prev = (struct sk_buff *)&noop_qdisc.gso_skb,
614 .qlen = 0,
615 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.gso_skb.lock),
616 },
617 .skb_bad_txq = {
618 .next = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
619 .prev = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
620 .qlen = 0,
621 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.skb_bad_txq.lock),
622 },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623};
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800624EXPORT_SYMBOL(noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Alexander Aringe63d7dfd2017-12-20 12:35:13 -0500626static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt,
627 struct netlink_ext_ack *extack)
Phil Sutterd66d6c32015-08-27 21:21:38 +0200628{
629 /* register_qdisc() assigns a default of noop_enqueue if unset,
630 * but __dev_queue_xmit() treats noqueue only as such
631 * if this is NULL - so clear it here. */
632 qdisc->enqueue = NULL;
633 return 0;
634}
635
636struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 .id = "noqueue",
638 .priv_size = 0,
Phil Sutterd66d6c32015-08-27 21:21:38 +0200639 .init = noqueue_init,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 .enqueue = noop_enqueue,
641 .dequeue = noop_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700642 .peek = noop_dequeue,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 .owner = THIS_MODULE,
644};
645
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000646static const u8 prio2band[TC_PRIO_MAX + 1] = {
647 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
648};
Thomas Graf321090e2005-06-18 22:58:35 -0700649
David S. Millerd3678b42008-07-21 09:56:13 -0700650/* 3-band FIFO queue: old style, but should be a bit faster than
651 generic prio+fifo combination.
652 */
653
654#define PFIFO_FAST_BANDS 3
655
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000656/*
657 * Private data for a pfifo_fast scheduler containing:
John Fastabendc5ad1192017-12-07 09:58:19 -0800658 * - rings for priority bands
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000659 */
660struct pfifo_fast_priv {
John Fastabendc5ad1192017-12-07 09:58:19 -0800661 struct skb_array q[PFIFO_FAST_BANDS];
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000662};
663
John Fastabendc5ad1192017-12-07 09:58:19 -0800664static inline struct skb_array *band2list(struct pfifo_fast_priv *priv,
665 int band)
David S. Millerd3678b42008-07-21 09:56:13 -0700666{
John Fastabendc5ad1192017-12-07 09:58:19 -0800667 return &priv->q[band];
David S. Millerd3678b42008-07-21 09:56:13 -0700668}
669
Eric Dumazet520ac302016-06-21 23:16:49 -0700670static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
671 struct sk_buff **to_free)
David S. Millerd3678b42008-07-21 09:56:13 -0700672{
John Fastabendc5ad1192017-12-07 09:58:19 -0800673 int band = prio2band[skb->priority & TC_PRIO_MAX];
674 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
675 struct skb_array *q = band2list(priv, band);
Eric Dumazetcce6294c2018-03-14 18:53:00 -0700676 unsigned int pkt_len = qdisc_pkt_len(skb);
John Fastabendc5ad1192017-12-07 09:58:19 -0800677 int err;
David S. Millerd3678b42008-07-21 09:56:13 -0700678
John Fastabendc5ad1192017-12-07 09:58:19 -0800679 err = skb_array_produce(q, skb);
Thomas Graf821d24ae2005-06-18 22:58:15 -0700680
Davide Caratti092e22e2019-08-27 23:18:53 +0200681 if (unlikely(err)) {
682 if (qdisc_is_percpu_stats(qdisc))
683 return qdisc_drop_cpu(skb, qdisc, to_free);
684 else
685 return qdisc_drop(skb, qdisc, to_free);
686 }
John Fastabendc5ad1192017-12-07 09:58:19 -0800687
Paolo Abeni8a53e612019-04-10 14:32:40 +0200688 qdisc_update_stats_at_enqueue(qdisc, pkt_len);
John Fastabendc5ad1192017-12-07 09:58:19 -0800689 return NET_XMIT_SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000692static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000694 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
John Fastabendc5ad1192017-12-07 09:58:19 -0800695 struct sk_buff *skb = NULL;
Yunsheng Lina90c57f2021-05-14 11:16:59 +0800696 bool need_retry = true;
John Fastabendc5ad1192017-12-07 09:58:19 -0800697 int band;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Yunsheng Lina90c57f2021-05-14 11:16:59 +0800699retry:
John Fastabendc5ad1192017-12-07 09:58:19 -0800700 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
701 struct skb_array *q = band2list(priv, band);
Florian Westphalec323362016-09-18 00:57:32 +0200702
John Fastabendc5ad1192017-12-07 09:58:19 -0800703 if (__skb_array_empty(q))
704 continue;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000705
Paolo Abeni021a17e2018-05-15 16:24:37 +0200706 skb = __skb_array_consume(q);
John Fastabendc5ad1192017-12-07 09:58:19 -0800707 }
708 if (likely(skb)) {
Paolo Abeni8a53e612019-04-10 14:32:40 +0200709 qdisc_update_stats_at_dequeue(qdisc, skb);
Yunsheng Lina90c57f2021-05-14 11:16:59 +0800710 } else if (need_retry &&
Yunsheng Linc4fef012021-06-22 14:49:56 +0800711 READ_ONCE(qdisc->state) & QDISC_STATE_NON_EMPTY) {
Yunsheng Lina90c57f2021-05-14 11:16:59 +0800712 /* Delay clearing the STATE_MISSED here to reduce
713 * the overhead of the second spin_trylock() in
714 * qdisc_run_begin() and __netif_schedule() calling
715 * in qdisc_run_end().
716 */
717 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
Yunsheng Linc4fef012021-06-22 14:49:56 +0800718 clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
Yunsheng Lina90c57f2021-05-14 11:16:59 +0800719
720 /* Make sure dequeuing happens after clearing
721 * STATE_MISSED.
722 */
723 smp_mb__after_atomic();
724
725 need_retry = false;
726
727 goto retry;
David S. Millerd3678b42008-07-21 09:56:13 -0700728 }
Thomas Graff87a9c32005-06-18 22:58:53 -0700729
John Fastabendc5ad1192017-12-07 09:58:19 -0800730 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000733static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700734{
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000735 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
John Fastabendc5ad1192017-12-07 09:58:19 -0800736 struct sk_buff *skb = NULL;
737 int band;
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700738
John Fastabendc5ad1192017-12-07 09:58:19 -0800739 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
740 struct skb_array *q = band2list(priv, band);
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000741
John Fastabendc5ad1192017-12-07 09:58:19 -0800742 skb = __skb_array_peek(q);
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700743 }
744
John Fastabendc5ad1192017-12-07 09:58:19 -0800745 return skb;
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700746}
747
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000748static void pfifo_fast_reset(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
John Fastabendc5ad1192017-12-07 09:58:19 -0800750 int i, band;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000751 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
David S. Millerd3678b42008-07-21 09:56:13 -0700752
John Fastabendc5ad1192017-12-07 09:58:19 -0800753 for (band = 0; band < PFIFO_FAST_BANDS; band++) {
754 struct skb_array *q = band2list(priv, band);
755 struct sk_buff *skb;
David S. Millerd3678b42008-07-21 09:56:13 -0700756
Cong Wang1df94c32017-12-18 14:34:26 -0800757 /* NULL ring is possible if destroy path is due to a failed
758 * skb_array_init() in pfifo_fast_init() case.
759 */
760 if (!q->ring.queue)
761 continue;
762
Paolo Abeni021a17e2018-05-15 16:24:37 +0200763 while ((skb = __skb_array_consume(q)) != NULL)
John Fastabendc5ad1192017-12-07 09:58:19 -0800764 kfree_skb(skb);
765 }
766
Davide Caratti04d37cf2019-08-27 12:29:09 +0200767 if (qdisc_is_percpu_stats(qdisc)) {
768 for_each_possible_cpu(i) {
769 struct gnet_stats_queue *q;
John Fastabendc5ad1192017-12-07 09:58:19 -0800770
Davide Caratti04d37cf2019-08-27 12:29:09 +0200771 q = per_cpu_ptr(qdisc->cpu_qstats, i);
772 q->backlog = 0;
773 q->qlen = 0;
774 }
John Fastabendc5ad1192017-12-07 09:58:19 -0800775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
David S. Millerd3678b42008-07-21 09:56:13 -0700778static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
779{
780 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
781
Eric Dumazetcc7ec452011-01-19 19:26:56 +0000782 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
David S. Miller1b34ec42012-03-29 05:11:39 -0400783 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
784 goto nla_put_failure;
David S. Millerd3678b42008-07-21 09:56:13 -0700785 return skb->len;
786
787nla_put_failure:
788 return -1;
789}
790
Alexander Aringe63d7dfd2017-12-20 12:35:13 -0500791static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt,
792 struct netlink_ext_ack *extack)
David S. Millerd3678b42008-07-21 09:56:13 -0700793{
John Fastabendc5ad1192017-12-07 09:58:19 -0800794 unsigned int qlen = qdisc_dev(qdisc)->tx_queue_len;
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000795 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
John Fastabendc5ad1192017-12-07 09:58:19 -0800796 int prio;
David S. Millerd3678b42008-07-21 09:56:13 -0700797
John Fastabendc5ad1192017-12-07 09:58:19 -0800798 /* guard against zero length rings */
799 if (!qlen)
800 return -EINVAL;
801
802 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
803 struct skb_array *q = band2list(priv, prio);
804 int err;
805
806 err = skb_array_init(q, qlen, GFP_KERNEL);
807 if (err)
808 return -ENOMEM;
809 }
David S. Millerd3678b42008-07-21 09:56:13 -0700810
Eric Dumazet23624932011-01-21 16:26:09 -0800811 /* Can by-pass the queue discipline */
812 qdisc->flags |= TCQ_F_CAN_BYPASS;
David S. Millerd3678b42008-07-21 09:56:13 -0700813 return 0;
814}
815
John Fastabendc5ad1192017-12-07 09:58:19 -0800816static void pfifo_fast_destroy(struct Qdisc *sch)
817{
818 struct pfifo_fast_priv *priv = qdisc_priv(sch);
819 int prio;
820
821 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
822 struct skb_array *q = band2list(priv, prio);
823
824 /* NULL ring is possible if destroy path is due to a failed
825 * skb_array_init() in pfifo_fast_init() case.
826 */
Cong Wang1df94c32017-12-18 14:34:26 -0800827 if (!q->ring.queue)
John Fastabendc5ad1192017-12-07 09:58:19 -0800828 continue;
829 /* Destroy ring but no need to kfree_skb because a call to
830 * pfifo_fast_reset() has already done that work.
831 */
832 ptr_ring_cleanup(&q->ring, NULL);
833 }
834}
835
Cong Wang7007ba62018-01-25 18:26:24 -0800836static int pfifo_fast_change_tx_queue_len(struct Qdisc *sch,
837 unsigned int new_len)
838{
839 struct pfifo_fast_priv *priv = qdisc_priv(sch);
840 struct skb_array *bands[PFIFO_FAST_BANDS];
841 int prio;
842
843 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
844 struct skb_array *q = band2list(priv, prio);
845
846 bands[prio] = q;
847 }
848
849 return skb_array_resize_multiple(bands, PFIFO_FAST_BANDS, new_len,
850 GFP_KERNEL);
851}
852
David S. Miller6ec1c692009-09-06 01:58:51 -0700853struct Qdisc_ops pfifo_fast_ops __read_mostly = {
David S. Millerd3678b42008-07-21 09:56:13 -0700854 .id = "pfifo_fast",
Krishna Kumarfd3ae5e2009-08-18 21:55:59 +0000855 .priv_size = sizeof(struct pfifo_fast_priv),
David S. Millerd3678b42008-07-21 09:56:13 -0700856 .enqueue = pfifo_fast_enqueue,
857 .dequeue = pfifo_fast_dequeue,
Jarek Poplawski99c0db22008-10-31 00:45:27 -0700858 .peek = pfifo_fast_peek,
David S. Millerd3678b42008-07-21 09:56:13 -0700859 .init = pfifo_fast_init,
John Fastabendc5ad1192017-12-07 09:58:19 -0800860 .destroy = pfifo_fast_destroy,
David S. Millerd3678b42008-07-21 09:56:13 -0700861 .reset = pfifo_fast_reset,
862 .dump = pfifo_fast_dump,
Cong Wang7007ba62018-01-25 18:26:24 -0800863 .change_tx_queue_len = pfifo_fast_change_tx_queue_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 .owner = THIS_MODULE,
John Fastabendc5ad1192017-12-07 09:58:19 -0800865 .static_flags = TCQ_F_NOLOCK | TCQ_F_CPUSTATS,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866};
Eric Dumazet1f27cde2016-03-02 08:21:43 -0800867EXPORT_SYMBOL(pfifo_fast_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Cong Wang1a33e102020-05-02 22:22:19 -0700869static struct lock_class_key qdisc_tx_busylock;
870static struct lock_class_key qdisc_running_key;
871
David S. Miller5ce2d482008-07-08 17:06:30 -0700872struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
Alexander Aringd0bd6842017-12-20 12:35:20 -0500873 const struct Qdisc_ops *ops,
874 struct netlink_ext_ack *extack)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 struct Qdisc *sch;
Eric Dumazet846e4632020-10-07 09:51:11 -0700877 unsigned int size = sizeof(*sch) + ops->priv_size;
Thomas Graf3d54b822005-07-05 14:15:09 -0700878 int err = -ENOBUFS;
Jesus Sanchez-Palencia26aa0452017-10-16 18:01:23 -0700879 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Jesus Sanchez-Palencia26aa0452017-10-16 18:01:23 -0700881 if (!dev_queue) {
Alexander Aringd0bd6842017-12-20 12:35:20 -0500882 NL_SET_ERR_MSG(extack, "No device queue given");
Jesus Sanchez-Palencia26aa0452017-10-16 18:01:23 -0700883 err = -EINVAL;
884 goto errout;
885 }
886
887 dev = dev_queue->dev;
Eric Dumazet846e4632020-10-07 09:51:11 -0700888 sch = kzalloc_node(size, GFP_KERNEL, netdev_queue_numa_node_read(dev_queue));
Eric Dumazetf2cd2d32010-11-29 08:14:37 +0000889
Eric Dumazet846e4632020-10-07 09:51:11 -0700890 if (!sch)
Thomas Graf3d54b822005-07-05 14:15:09 -0700891 goto errout;
John Fastabenda53851e2017-12-07 09:55:45 -0800892 __skb_queue_head_init(&sch->gso_skb);
John Fastabend70e57d52017-12-07 09:56:23 -0800893 __skb_queue_head_init(&sch->skb_bad_txq);
Florian Westphal48da34b2016-09-18 00:57:34 +0200894 qdisc_skb_head_init(&sch->q);
895 spin_lock_init(&sch->q.lock);
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000896
John Fastabendd59f5ff2017-12-07 09:55:26 -0800897 if (ops->static_flags & TCQ_F_CPUSTATS) {
898 sch->cpu_bstats =
899 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
900 if (!sch->cpu_bstats)
901 goto errout1;
902
903 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
904 if (!sch->cpu_qstats) {
905 free_percpu(sch->cpu_bstats);
906 goto errout1;
907 }
908 }
909
Eric Dumazet79640a42010-06-02 05:09:29 -0700910 spin_lock_init(&sch->busylock);
Cong Wang1a33e102020-05-02 22:22:19 -0700911 lockdep_set_class(&sch->busylock,
912 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
913
Paolo Abeni96009c72018-05-15 16:24:36 +0200914 /* seqlock has the same scope of busylock, for NOLOCK qdisc */
915 spin_lock_init(&sch->seqlock);
Cong Wang1a33e102020-05-02 22:22:19 -0700916 lockdep_set_class(&sch->busylock,
917 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
918
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700919 seqcount_init(&sch->running);
Cong Wang1a33e102020-05-02 22:22:19 -0700920 lockdep_set_class(&sch->running,
921 dev->qdisc_running_key ?: &qdisc_running_key);
Eric Dumazetf9eb8ae2016-06-06 09:37:15 -0700922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 sch->ops = ops;
John Fastabendd59f5ff2017-12-07 09:55:26 -0800924 sch->flags = ops->static_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 sch->enqueue = ops->enqueue;
926 sch->dequeue = ops->dequeue;
David S. Millerbb949fb2008-07-08 16:55:56 -0700927 sch->dev_queue = dev_queue;
Eric Dumazet23d3b8b2012-09-05 01:02:56 +0000928 dev_hold(dev);
Reshetova, Elena7b936402017-07-04 15:53:07 +0300929 refcount_set(&sch->refcnt, 1);
Thomas Graf3d54b822005-07-05 14:15:09 -0700930
931 return sch;
John Fastabendd59f5ff2017-12-07 09:55:26 -0800932errout1:
Eric Dumazet846e4632020-10-07 09:51:11 -0700933 kfree(sch);
Thomas Graf3d54b822005-07-05 14:15:09 -0700934errout:
WANG Cong01e123d2008-06-27 19:51:35 -0700935 return ERR_PTR(err);
Thomas Graf3d54b822005-07-05 14:15:09 -0700936}
937
Changli Gao3511c912010-10-16 13:04:08 +0000938struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
stephen hemmingerd2a7f262013-08-31 10:15:50 -0700939 const struct Qdisc_ops *ops,
Alexander Aringa38a98822017-12-20 12:35:21 -0500940 unsigned int parentid,
941 struct netlink_ext_ack *extack)
Thomas Graf3d54b822005-07-05 14:15:09 -0700942{
943 struct Qdisc *sch;
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +0900944
Alexander Aringa38a98822017-12-20 12:35:21 -0500945 if (!try_module_get(ops->owner)) {
946 NL_SET_ERR_MSG(extack, "Failed to increase module reference counter");
Eric Dumazet166ee5b2016-08-24 09:39:02 -0700947 return NULL;
Alexander Aringa38a98822017-12-20 12:35:21 -0500948 }
stephen hemminger6da7c8f2013-08-27 16:19:08 -0700949
Alexander Aringa38a98822017-12-20 12:35:21 -0500950 sch = qdisc_alloc(dev_queue, ops, extack);
Eric Dumazet166ee5b2016-08-24 09:39:02 -0700951 if (IS_ERR(sch)) {
952 module_put(ops->owner);
953 return NULL;
954 }
Patrick McHardy9f9afec2006-11-29 17:35:18 -0800955 sch->parent = parentid;
Thomas Graf3d54b822005-07-05 14:15:09 -0700956
Cong Wangf5a78332020-05-26 21:35:25 -0700957 if (!ops->init || ops->init(sch, NULL, extack) == 0) {
958 trace_qdisc_create(ops, dev_queue->dev, parentid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 return sch;
Cong Wangf5a78332020-05-26 21:35:25 -0700960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Vlad Buslov86bd4462018-09-24 19:22:50 +0300962 qdisc_put(sch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 return NULL;
964}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800965EXPORT_SYMBOL(qdisc_create_dflt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
David S. Miller5fb66222008-08-02 20:02:43 -0700967/* Under qdisc_lock(qdisc) and BH! */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969void qdisc_reset(struct Qdisc *qdisc)
970{
Eric Dumazet20fea082007-11-14 01:44:41 -0800971 const struct Qdisc_ops *ops = qdisc->ops;
John Fastabenda53851e2017-12-07 09:55:45 -0800972 struct sk_buff *skb, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Cong Wanga34dac02020-05-26 21:35:24 -0700974 trace_qdisc_reset(qdisc);
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (ops->reset)
977 ops->reset(qdisc);
Jarek Poplawski67305eb2008-11-03 02:52:50 -0800978
John Fastabenda53851e2017-12-07 09:55:45 -0800979 skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) {
980 __skb_unlink(skb, &qdisc->gso_skb);
981 kfree_skb_list(skb);
Krishna Kumarbbd8a0d2009-08-06 01:44:21 +0000982 }
John Fastabenda53851e2017-12-07 09:55:45 -0800983
John Fastabend70e57d52017-12-07 09:56:23 -0800984 skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) {
985 __skb_unlink(skb, &qdisc->skb_bad_txq);
986 kfree_skb_list(skb);
987 }
988
Eric Dumazet4d202a02016-06-21 23:16:52 -0700989 qdisc->q.qlen = 0;
Konstantin Khlebnikovc8e18122017-09-20 15:45:36 +0300990 qdisc->qstats.backlog = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
Patrick McHardy62e3ba12008-01-22 22:10:23 -0800992EXPORT_SYMBOL(qdisc_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Daniel Borkmann81d947e2018-01-15 23:12:09 +0100994void qdisc_free(struct Qdisc *qdisc)
Eric Dumazet5d944c62010-03-31 07:06:04 +0000995{
John Fastabend73c20a82016-01-05 09:11:36 -0800996 if (qdisc_is_percpu_stats(qdisc)) {
John Fastabend22e0f8b2014-09-28 11:52:56 -0700997 free_percpu(qdisc->cpu_bstats);
John Fastabend73c20a82016-01-05 09:11:36 -0800998 free_percpu(qdisc->cpu_qstats);
999 }
John Fastabend22e0f8b2014-09-28 11:52:56 -07001000
Eric Dumazet846e4632020-10-07 09:51:11 -07001001 kfree(qdisc);
Eric Dumazet5d944c62010-03-31 07:06:04 +00001002}
1003
Wei Yongjun53627002018-09-27 14:47:56 +00001004static void qdisc_free_cb(struct rcu_head *head)
Vlad Buslov3a7d0d02018-09-24 19:22:51 +03001005{
1006 struct Qdisc *q = container_of(head, struct Qdisc, rcu);
1007
1008 qdisc_free(q);
1009}
1010
Vlad Buslov86bd4462018-09-24 19:22:50 +03001011static void qdisc_destroy(struct Qdisc *qdisc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012{
Eric Dumazet20fea082007-11-14 01:44:41 -08001013 const struct Qdisc_ops *ops = qdisc->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
David S. Miller3a682fb2008-07-20 18:13:01 -07001015#ifdef CONFIG_NET_SCHED
Jiri Kosina59cc1f62016-08-10 11:05:15 +02001016 qdisc_hash_del(qdisc);
Jarek Poplawskif6e0b232008-08-22 03:24:05 -07001017
Eric Dumazeta2da5702011-01-20 03:48:19 +00001018 qdisc_put_stab(rtnl_dereference(qdisc->stab));
David S. Miller3a682fb2008-07-20 18:13:01 -07001019#endif
Eric Dumazet1c0d32f2016-12-04 09:48:16 -08001020 gen_kill_estimator(&qdisc->rate_est);
Cong Wang4909dab2020-05-26 21:35:23 -07001021
1022 qdisc_reset(qdisc);
1023
Patrick McHardy85670cc2006-09-27 16:45:45 -07001024 if (ops->destroy)
1025 ops->destroy(qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Patrick McHardy85670cc2006-09-27 16:45:45 -07001027 module_put(ops->owner);
David S. Miller5ce2d482008-07-08 17:06:30 -07001028 dev_put(qdisc_dev(qdisc));
David S. Miller8a34c5d2008-07-17 00:47:45 -07001029
Cong Wanga34dac02020-05-26 21:35:24 -07001030 trace_qdisc_destroy(qdisc);
1031
Vlad Buslov3a7d0d02018-09-24 19:22:51 +03001032 call_rcu(&qdisc->rcu, qdisc_free_cb);
David S. Miller8a34c5d2008-07-17 00:47:45 -07001033}
Vlad Buslov86bd4462018-09-24 19:22:50 +03001034
1035void qdisc_put(struct Qdisc *qdisc)
1036{
Cong Wang6efb9712019-09-12 10:22:30 -07001037 if (!qdisc)
1038 return;
1039
Vlad Buslov86bd4462018-09-24 19:22:50 +03001040 if (qdisc->flags & TCQ_F_BUILTIN ||
1041 !refcount_dec_and_test(&qdisc->refcnt))
1042 return;
1043
1044 qdisc_destroy(qdisc);
1045}
1046EXPORT_SYMBOL(qdisc_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Vlad Buslov3a7d0d02018-09-24 19:22:51 +03001048/* Version of qdisc_put() that is called with rtnl mutex unlocked.
1049 * Intended to be used as optimization, this function only takes rtnl lock if
1050 * qdisc reference counter reached zero.
1051 */
1052
1053void qdisc_put_unlocked(struct Qdisc *qdisc)
1054{
1055 if (qdisc->flags & TCQ_F_BUILTIN ||
1056 !refcount_dec_and_rtnl_lock(&qdisc->refcnt))
1057 return;
1058
1059 qdisc_destroy(qdisc);
1060 rtnl_unlock();
1061}
1062EXPORT_SYMBOL(qdisc_put_unlocked);
1063
Patrick McHardy589983c2009-09-04 06:41:20 +00001064/* Attach toplevel qdisc to device queue. */
1065struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
1066 struct Qdisc *qdisc)
1067{
1068 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
1069 spinlock_t *root_lock;
1070
1071 root_lock = qdisc_lock(oqdisc);
1072 spin_lock_bh(root_lock);
1073
Patrick McHardy589983c2009-09-04 06:41:20 +00001074 /* ... and graft new one */
1075 if (qdisc == NULL)
1076 qdisc = &noop_qdisc;
1077 dev_queue->qdisc_sleeping = qdisc;
1078 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
1079
1080 spin_unlock_bh(root_lock);
1081
1082 return oqdisc;
1083}
John Fastabendb8970f02011-01-17 08:06:09 +00001084EXPORT_SYMBOL(dev_graft_qdisc);
Patrick McHardy589983c2009-09-04 06:41:20 +00001085
David S. Millere8a04642008-07-17 00:34:19 -07001086static void attach_one_default_qdisc(struct net_device *dev,
1087 struct netdev_queue *dev_queue,
1088 void *_unused)
1089{
Phil Sutter3e692f22015-08-27 21:21:39 +02001090 struct Qdisc *qdisc;
1091 const struct Qdisc_ops *ops = default_qdisc_ops;
David S. Millere8a04642008-07-17 00:34:19 -07001092
Phil Sutter3e692f22015-08-27 21:21:39 +02001093 if (dev->priv_flags & IFF_NO_QUEUE)
1094 ops = &noqueue_qdisc_ops;
Vincent Prince546b85b2019-10-23 15:44:20 +02001095 else if(dev->type == ARPHRD_CAN)
1096 ops = &pfifo_fast_ops;
Phil Sutter3e692f22015-08-27 21:21:39 +02001097
Alexander Aringa38a98822017-12-20 12:35:21 -05001098 qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT, NULL);
Jesper Dangaard Brouerbf6dba72020-04-30 13:42:22 +02001099 if (!qdisc)
Phil Sutter3e692f22015-08-27 21:21:39 +02001100 return;
Jesper Dangaard Brouerbf6dba72020-04-30 13:42:22 +02001101
Phil Sutter3e692f22015-08-27 21:21:39 +02001102 if (!netif_is_multiqueue(dev))
Eric Dumazet4eaf3b82015-12-01 20:08:51 -08001103 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
David S. Millere8a04642008-07-17 00:34:19 -07001104 dev_queue->qdisc_sleeping = qdisc;
1105}
1106
David S. Miller6ec1c692009-09-06 01:58:51 -07001107static void attach_default_qdiscs(struct net_device *dev)
1108{
1109 struct netdev_queue *txq;
1110 struct Qdisc *qdisc;
1111
1112 txq = netdev_get_tx_queue(dev, 0);
1113
Phil Sutter4b469952015-08-13 19:01:07 +02001114 if (!netif_is_multiqueue(dev) ||
Phil Sutter4b469952015-08-13 19:01:07 +02001115 dev->priv_flags & IFF_NO_QUEUE) {
David S. Miller6ec1c692009-09-06 01:58:51 -07001116 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
1117 dev->qdisc = txq->qdisc_sleeping;
Eric Dumazet551143d2017-08-24 21:12:28 -07001118 qdisc_refcount_inc(dev->qdisc);
David S. Miller6ec1c692009-09-06 01:58:51 -07001119 } else {
Alexander Aringa38a98822017-12-20 12:35:21 -05001120 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT, NULL);
David S. Miller6ec1c692009-09-06 01:58:51 -07001121 if (qdisc) {
David S. Miller6ec1c692009-09-06 01:58:51 -07001122 dev->qdisc = qdisc;
Eric Dumazete57a7842013-12-12 15:41:56 -08001123 qdisc->ops->attach(qdisc);
David S. Miller6ec1c692009-09-06 01:58:51 -07001124 }
1125 }
Jesper Dangaard Brouerbf6dba72020-04-30 13:42:22 +02001126
1127 /* Detect default qdisc setup/init failed and fallback to "noqueue" */
1128 if (dev->qdisc == &noop_qdisc) {
1129 netdev_warn(dev, "default qdisc (%s) fail, fallback to %s\n",
1130 default_qdisc_ops->id, noqueue_qdisc_ops.id);
1131 dev->priv_flags |= IFF_NO_QUEUE;
1132 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
1133 dev->qdisc = txq->qdisc_sleeping;
1134 qdisc_refcount_inc(dev->qdisc);
1135 dev->priv_flags ^= IFF_NO_QUEUE;
1136 }
1137
Jiri Kosina59cc1f62016-08-10 11:05:15 +02001138#ifdef CONFIG_NET_SCHED
WANG Cong92f91702017-04-04 18:51:30 -07001139 if (dev->qdisc != &noop_qdisc)
Jiri Kosina49b49972017-03-08 16:03:32 +01001140 qdisc_hash_add(dev->qdisc, false);
Jiri Kosina59cc1f62016-08-10 11:05:15 +02001141#endif
David S. Miller6ec1c692009-09-06 01:58:51 -07001142}
1143
David S. Millere8a04642008-07-17 00:34:19 -07001144static void transition_one_qdisc(struct net_device *dev,
1145 struct netdev_queue *dev_queue,
1146 void *_need_watchdog)
1147{
David S. Miller838740002008-07-17 00:53:03 -07001148 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -07001149 int *need_watchdog_p = _need_watchdog;
1150
David S. Millera9312ae2008-08-17 21:51:03 -07001151 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
1152 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
1153
David S. Miller838740002008-07-17 00:53:03 -07001154 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
Phil Sutter3e692f22015-08-27 21:21:39 +02001155 if (need_watchdog_p) {
Eric Dumazet9d214932009-05-17 20:55:16 -07001156 dev_queue->trans_start = 0;
David S. Millere8a04642008-07-17 00:34:19 -07001157 *need_watchdog_p = 1;
Eric Dumazet9d214932009-05-17 20:55:16 -07001158 }
David S. Millere8a04642008-07-17 00:34:19 -07001159}
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161void dev_activate(struct net_device *dev)
1162{
David S. Millere8a04642008-07-17 00:34:19 -07001163 int need_watchdog;
David S. Millerb0e1e642008-07-08 17:42:10 -07001164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 /* No queueing discipline is attached to device;
stephen hemminger6da7c8f2013-08-27 16:19:08 -07001166 * create default one for devices, which need queueing
1167 * and noqueue_qdisc for virtual interfaces
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 */
1169
David S. Miller6ec1c692009-09-06 01:58:51 -07001170 if (dev->qdisc == &noop_qdisc)
1171 attach_default_qdiscs(dev);
Patrick McHardyaf356af2009-09-04 06:41:18 +00001172
Tommy S. Christensencacaddf2005-05-03 16:18:52 -07001173 if (!netif_carrier_ok(dev))
1174 /* Delay activation until next carrier-on event */
1175 return;
1176
David S. Millere8a04642008-07-17 00:34:19 -07001177 need_watchdog = 0;
1178 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
Eric Dumazet24824a02010-10-02 06:11:55 +00001179 if (dev_ingress_queue(dev))
1180 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
David S. Millere8a04642008-07-17 00:34:19 -07001181
1182 if (need_watchdog) {
Florian Westphal860e9532016-05-03 16:33:13 +02001183 netif_trans_update(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 dev_watchdog_up(dev);
1185 }
David S. Millerb0e1e642008-07-08 17:42:10 -07001186}
John Fastabendb8970f02011-01-17 08:06:09 +00001187EXPORT_SYMBOL(dev_activate);
David S. Millerb0e1e642008-07-08 17:42:10 -07001188
Cong Wang70f50962020-05-26 21:35:26 -07001189static void qdisc_deactivate(struct Qdisc *qdisc)
1190{
Cong Wang70f50962020-05-26 21:35:26 -07001191 if (qdisc->flags & TCQ_F_BUILTIN)
1192 return;
Cong Wang70f50962020-05-26 21:35:26 -07001193
1194 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
Cong Wang70f50962020-05-26 21:35:26 -07001195}
1196
David S. Millere8a04642008-07-17 00:34:19 -07001197static void dev_deactivate_queue(struct net_device *dev,
1198 struct netdev_queue *dev_queue,
1199 void *_qdisc_default)
David S. Millerb0e1e642008-07-08 17:42:10 -07001200{
David S. Millere8a04642008-07-17 00:34:19 -07001201 struct Qdisc *qdisc_default = _qdisc_default;
David S. Miller970565b2008-07-08 23:10:33 -07001202 struct Qdisc *qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -07001203
John Fastabend46e5da40a2014-09-12 20:04:52 -07001204 qdisc = rtnl_dereference(dev_queue->qdisc);
David S. Millerb0e1e642008-07-08 17:42:10 -07001205 if (qdisc) {
Cong Wang70f50962020-05-26 21:35:26 -07001206 qdisc_deactivate(qdisc);
Jarek Poplawskif7a54c12008-08-27 02:22:07 -07001207 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -07001208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209}
1210
Yunsheng Lin2fb541c2020-09-08 19:02:34 +08001211static void dev_reset_queue(struct net_device *dev,
1212 struct netdev_queue *dev_queue,
1213 void *_unused)
1214{
1215 struct Qdisc *qdisc;
1216 bool nolock;
1217
1218 qdisc = dev_queue->qdisc_sleeping;
1219 if (!qdisc)
1220 return;
1221
1222 nolock = qdisc->flags & TCQ_F_NOLOCK;
1223
1224 if (nolock)
1225 spin_lock_bh(&qdisc->seqlock);
1226 spin_lock_bh(qdisc_lock(qdisc));
1227
1228 qdisc_reset(qdisc);
1229
1230 spin_unlock_bh(qdisc_lock(qdisc));
Yunsheng Lin102b55e2021-05-14 11:17:00 +08001231 if (nolock) {
1232 clear_bit(__QDISC_STATE_MISSED, &qdisc->state);
Yunsheng Linc4fef012021-06-22 14:49:56 +08001233 clear_bit(__QDISC_STATE_DRAINING, &qdisc->state);
Yunsheng Lin2fb541c2020-09-08 19:02:34 +08001234 spin_unlock_bh(&qdisc->seqlock);
Yunsheng Lin102b55e2021-05-14 11:17:00 +08001235 }
Yunsheng Lin2fb541c2020-09-08 19:02:34 +08001236}
1237
David S. Miller4335cd22008-08-17 21:58:07 -07001238static bool some_qdisc_is_busy(struct net_device *dev)
David S. Millere8a04642008-07-17 00:34:19 -07001239{
1240 unsigned int i;
1241
1242 for (i = 0; i < dev->num_tx_queues; i++) {
1243 struct netdev_queue *dev_queue;
David S. Miller7698b4f2008-07-16 01:42:40 -07001244 spinlock_t *root_lock;
David S. Millere2627c82008-07-16 00:56:32 -07001245 struct Qdisc *q;
David S. Millere8a04642008-07-17 00:34:19 -07001246 int val;
1247
1248 dev_queue = netdev_get_tx_queue(dev, i);
David S. Millerb9a3b112008-08-13 15:18:38 -07001249 q = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -07001250
Paolo Abeni32f7b442018-05-15 10:50:31 +02001251 root_lock = qdisc_lock(q);
1252 spin_lock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -07001253
Paolo Abeni32f7b442018-05-15 10:50:31 +02001254 val = (qdisc_is_running(q) ||
1255 test_bit(__QDISC_STATE_SCHED, &q->state));
David S. Millere8a04642008-07-17 00:34:19 -07001256
Paolo Abeni32f7b442018-05-15 10:50:31 +02001257 spin_unlock_bh(root_lock);
David S. Millere8a04642008-07-17 00:34:19 -07001258
1259 if (val)
1260 return true;
1261 }
1262 return false;
1263}
1264
Eric Dumazet31376632011-05-19 23:42:09 +00001265/**
1266 * dev_deactivate_many - deactivate transmissions on several devices
1267 * @head: list of devices to deactivate
1268 *
1269 * This function returns only when all outstanding transmissions
1270 * have completed, unless all devices are in dismantle phase.
1271 */
Octavian Purdila44345722010-12-13 12:44:07 +00001272void dev_deactivate_many(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
Octavian Purdila44345722010-12-13 12:44:07 +00001274 struct net_device *dev;
Herbert Xu41a23b02007-05-10 14:12:47 -07001275
Eric W. Biederman5cde2822013-10-05 19:26:05 -07001276 list_for_each_entry(dev, head, close_list) {
Octavian Purdila44345722010-12-13 12:44:07 +00001277 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
1278 &noop_qdisc);
1279 if (dev_ingress_queue(dev))
1280 dev_deactivate_queue(dev, dev_ingress_queue(dev),
1281 &noop_qdisc);
1282
1283 dev_watchdog_down(dev);
1284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Yunsheng Lin2fb541c2020-09-08 19:02:34 +08001286 /* Wait for outstanding qdisc-less dev_queue_xmit calls or
1287 * outstanding qdisc enqueuing calls.
Eric Dumazet31376632011-05-19 23:42:09 +00001288 * This is avoided if all devices are in dismantle phase :
1289 * Caller will call synchronize_net() for us
1290 */
John Fastabend7bbde832017-12-07 09:56:04 -08001291 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Yunsheng Lin2fb541c2020-09-08 19:02:34 +08001293 list_for_each_entry(dev, head, close_list) {
1294 netdev_for_each_tx_queue(dev, dev_reset_queue, NULL);
1295
1296 if (dev_ingress_queue(dev))
1297 dev_reset_queue(dev, dev_ingress_queue(dev), NULL);
1298 }
1299
Herbert Xud4828d82006-06-22 02:28:18 -07001300 /* Wait for outstanding qdisc_run calls. */
John Fastabend7bbde832017-12-07 09:56:04 -08001301 list_for_each_entry(dev, head, close_list) {
Marc Kleine-Budde4eab4212019-10-16 10:28:33 +02001302 while (some_qdisc_is_busy(dev)) {
1303 /* wait_event() would avoid this sleep-loop but would
1304 * require expensive checks in the fast paths of packet
1305 * processing which isn't worth it.
1306 */
1307 schedule_timeout_uninterruptible(1);
1308 }
John Fastabend7bbde832017-12-07 09:56:04 -08001309 }
Octavian Purdila44345722010-12-13 12:44:07 +00001310}
1311
1312void dev_deactivate(struct net_device *dev)
1313{
1314 LIST_HEAD(single);
1315
Eric W. Biederman5cde2822013-10-05 19:26:05 -07001316 list_add(&dev->close_list, &single);
Octavian Purdila44345722010-12-13 12:44:07 +00001317 dev_deactivate_many(&single);
Eric W. Biederman5f04d502011-02-20 11:49:45 -08001318 list_del(&single);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
John Fastabendb8970f02011-01-17 08:06:09 +00001320EXPORT_SYMBOL(dev_deactivate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Cong Wang48bfd552018-01-25 18:26:23 -08001322static int qdisc_change_tx_queue_len(struct net_device *dev,
1323 struct netdev_queue *dev_queue)
1324{
1325 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1326 const struct Qdisc_ops *ops = qdisc->ops;
1327
1328 if (ops->change_tx_queue_len)
1329 return ops->change_tx_queue_len(qdisc, dev->tx_queue_len);
1330 return 0;
1331}
1332
1333int dev_qdisc_change_tx_queue_len(struct net_device *dev)
1334{
1335 bool up = dev->flags & IFF_UP;
1336 unsigned int i;
1337 int ret = 0;
1338
1339 if (up)
1340 dev_deactivate(dev);
1341
1342 for (i = 0; i < dev->num_tx_queues; i++) {
1343 ret = qdisc_change_tx_queue_len(dev, &dev->_tx[i]);
1344
1345 /* TODO: revert changes on a partial failure */
1346 if (ret)
1347 break;
1348 }
1349
1350 if (up)
1351 dev_activate(dev);
1352 return ret;
1353}
1354
David S. Millerb0e1e642008-07-08 17:42:10 -07001355static void dev_init_scheduler_queue(struct net_device *dev,
1356 struct netdev_queue *dev_queue,
David S. Millere8a04642008-07-17 00:34:19 -07001357 void *_qdisc)
David S. Millerb0e1e642008-07-08 17:42:10 -07001358{
David S. Millere8a04642008-07-17 00:34:19 -07001359 struct Qdisc *qdisc = _qdisc;
1360
John Fastabend46e5da40a2014-09-12 20:04:52 -07001361 rcu_assign_pointer(dev_queue->qdisc, qdisc);
David S. Millerb0e1e642008-07-08 17:42:10 -07001362 dev_queue->qdisc_sleeping = qdisc;
David S. Millerb0e1e642008-07-08 17:42:10 -07001363}
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365void dev_init_scheduler(struct net_device *dev)
1366{
Patrick McHardyaf356af2009-09-04 06:41:18 +00001367 dev->qdisc = &noop_qdisc;
David S. Millere8a04642008-07-17 00:34:19 -07001368 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +00001369 if (dev_ingress_queue(dev))
1370 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Kees Cookcdeabbb2017-10-16 17:29:17 -07001372 timer_setup(&dev->watchdog_timer, dev_watchdog, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
David S. Millere8a04642008-07-17 00:34:19 -07001375static void shutdown_scheduler_queue(struct net_device *dev,
1376 struct netdev_queue *dev_queue,
1377 void *_qdisc_default)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378{
David S. Millerb0e1e642008-07-08 17:42:10 -07001379 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
David S. Millere8a04642008-07-17 00:34:19 -07001380 struct Qdisc *qdisc_default = _qdisc_default;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
David S. Millerb0e1e642008-07-08 17:42:10 -07001382 if (qdisc) {
Jarek Poplawskif7a54c12008-08-27 02:22:07 -07001383 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
David S. Millerb0e1e642008-07-08 17:42:10 -07001384 dev_queue->qdisc_sleeping = qdisc_default;
1385
Vlad Buslov86bd4462018-09-24 19:22:50 +03001386 qdisc_put(qdisc);
YOSHIFUJI Hideaki10297b92007-02-09 23:25:16 +09001387 }
David S. Millerb0e1e642008-07-08 17:42:10 -07001388}
1389
1390void dev_shutdown(struct net_device *dev)
1391{
David S. Millere8a04642008-07-17 00:34:19 -07001392 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
Eric Dumazet24824a02010-10-02 06:11:55 +00001393 if (dev_ingress_queue(dev))
1394 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
Vlad Buslov86bd4462018-09-24 19:22:50 +03001395 qdisc_put(dev->qdisc);
Patrick McHardyaf356af2009-09-04 06:41:18 +00001396 dev->qdisc = &noop_qdisc;
1397
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001398 WARN_ON(timer_pending(&dev->watchdog_timer));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
Jiri Pirko292f1c72013-02-12 00:12:03 +00001400
Baowen Zheng2ffe0392021-03-12 15:08:31 +01001401/**
1402 * psched_ratecfg_precompute__() - Pre-compute values for reciprocal division
1403 * @rate: Rate to compute reciprocal division values of
1404 * @mult: Multiplier for reciprocal division
1405 * @shift: Shift for reciprocal division
1406 *
1407 * The multiplier and shift for reciprocal division by rate are stored
1408 * in mult and shift.
1409 *
1410 * The deal here is to replace a divide by a reciprocal one
1411 * in fast path (a reciprocal divide is a multiply and a shift)
1412 *
1413 * Normal formula would be :
1414 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
1415 *
1416 * We compute mult/shift to use instead :
1417 * time_in_ns = (len * mult) >> shift;
1418 *
1419 * We try to get the highest possible mult value for accuracy,
1420 * but have to make sure no overflows will ever happen.
1421 *
1422 * reciprocal_value() is not used here it doesn't handle 64-bit values.
1423 */
1424static void psched_ratecfg_precompute__(u64 rate, u32 *mult, u8 *shift)
1425{
1426 u64 factor = NSEC_PER_SEC;
1427
1428 *mult = 1;
1429 *shift = 0;
1430
1431 if (rate <= 0)
1432 return;
1433
1434 for (;;) {
1435 *mult = div64_u64(factor, rate);
1436 if (*mult & (1U << 31) || factor & (1ULL << 63))
1437 break;
1438 factor <<= 1;
1439 (*shift)++;
1440 }
1441}
1442
Eric Dumazet01cb71d2013-06-02 13:55:05 +00001443void psched_ratecfg_precompute(struct psched_ratecfg *r,
Eric Dumazet3e1e3aa2013-09-19 09:10:03 -07001444 const struct tc_ratespec *conf,
1445 u64 rate64)
Jiri Pirko292f1c72013-02-12 00:12:03 +00001446{
Eric Dumazet01cb71d2013-06-02 13:55:05 +00001447 memset(r, 0, sizeof(*r));
1448 r->overhead = conf->overhead;
Eric Dumazet3e1e3aa2013-09-19 09:10:03 -07001449 r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
Jesper Dangaard Brouer8a8e3d82013-08-14 23:47:11 +02001450 r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
Baowen Zheng2ffe0392021-03-12 15:08:31 +01001451 psched_ratecfg_precompute__(r->rate_bytes_ps, &r->mult, &r->shift);
Jiri Pirko292f1c72013-02-12 00:12:03 +00001452}
1453EXPORT_SYMBOL(psched_ratecfg_precompute);
Jiri Pirko46209402017-11-03 11:46:25 +01001454
Baowen Zheng2ffe0392021-03-12 15:08:31 +01001455void psched_ppscfg_precompute(struct psched_pktrate *r, u64 pktrate64)
1456{
1457 r->rate_pkts_ps = pktrate64;
1458 psched_ratecfg_precompute__(r->rate_pkts_ps, &r->mult, &r->shift);
1459}
1460EXPORT_SYMBOL(psched_ppscfg_precompute);
1461
Jiri Pirko46209402017-11-03 11:46:25 +01001462static void mini_qdisc_rcu_func(struct rcu_head *head)
1463{
1464}
1465
1466void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1467 struct tcf_proto *tp_head)
1468{
Vlad Busloved76f5e2019-02-11 10:55:38 +02001469 /* Protected with chain0->filter_chain_lock.
1470 * Can't access chain directly because tp_head can be NULL.
1471 */
1472 struct mini_Qdisc *miniq_old =
1473 rcu_dereference_protected(*miniqp->p_miniq, 1);
Jiri Pirko46209402017-11-03 11:46:25 +01001474 struct mini_Qdisc *miniq;
1475
1476 if (!tp_head) {
1477 RCU_INIT_POINTER(*miniqp->p_miniq, NULL);
Cong Wangb2fb01f2017-12-20 23:26:24 -08001478 /* Wait for flying RCU callback before it is freed. */
Paul E. McKenneyae0e3342018-11-06 19:40:39 -08001479 rcu_barrier();
Jiri Pirko46209402017-11-03 11:46:25 +01001480 return;
1481 }
1482
1483 miniq = !miniq_old || miniq_old == &miniqp->miniq2 ?
1484 &miniqp->miniq1 : &miniqp->miniq2;
1485
1486 /* We need to make sure that readers won't see the miniq
Paul E. McKenneyae0e3342018-11-06 19:40:39 -08001487 * we are about to modify. So wait until previous call_rcu callback
Jiri Pirko46209402017-11-03 11:46:25 +01001488 * is done.
1489 */
Paul E. McKenneyae0e3342018-11-06 19:40:39 -08001490 rcu_barrier();
Jiri Pirko46209402017-11-03 11:46:25 +01001491 miniq->filter_list = tp_head;
1492 rcu_assign_pointer(*miniqp->p_miniq, miniq);
1493
1494 if (miniq_old)
Cong Wangb2fb01f2017-12-20 23:26:24 -08001495 /* This is counterpart of the rcu barriers above. We need to
Jiri Pirko46209402017-11-03 11:46:25 +01001496 * block potential new user of miniq_old until all readers
1497 * are not seeing it.
1498 */
Paul E. McKenneyae0e3342018-11-06 19:40:39 -08001499 call_rcu(&miniq_old->rcu, mini_qdisc_rcu_func);
Jiri Pirko46209402017-11-03 11:46:25 +01001500}
1501EXPORT_SYMBOL(mini_qdisc_pair_swap);
1502
Paul Blakey7d17c542020-02-16 12:01:22 +02001503void mini_qdisc_pair_block_init(struct mini_Qdisc_pair *miniqp,
1504 struct tcf_block *block)
1505{
1506 miniqp->miniq1.block = block;
1507 miniqp->miniq2.block = block;
1508}
1509EXPORT_SYMBOL(mini_qdisc_pair_block_init);
1510
Jiri Pirko46209402017-11-03 11:46:25 +01001511void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1512 struct mini_Qdisc __rcu **p_miniq)
1513{
1514 miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats;
1515 miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats;
1516 miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats;
1517 miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats;
1518 miniqp->p_miniq = p_miniq;
1519}
1520EXPORT_SYMBOL(mini_qdisc_pair_init);