1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/sch_api.c Packet scheduler API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Fixes:
8 *
9 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
10 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
11 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
12 */
13
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmod.h>
24 #include <linux/list.h>
25 #include <linux/hrtimer.h>
26 #include <linux/slab.h>
27 #include <linux/hashtable.h>
28
29 #include <net/net_namespace.h>
30 #include <net/sock.h>
31 #include <net/netlink.h>
32 #include <net/pkt_sched.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_wrapper.h>
35
36 #include <trace/events/qdisc.h>
37
38 /*
39
40 Short review.
41 -------------
42
43 This file consists of two interrelated parts:
44
45 1. queueing disciplines manager frontend.
46 2. traffic classes manager frontend.
47
48 Generally, queueing discipline ("qdisc") is a black box,
49 which is able to enqueue packets and to dequeue them (when
50 device is ready to send something) in order and at times
51 determined by algorithm hidden in it.
52
53 qdisc's are divided to two categories:
54 - "queues", which have no internal structure visible from outside.
55 - "schedulers", which split all the packets to "traffic classes",
56 using "packet classifiers" (look at cls_api.c)
57
58 In turn, classes may have child qdiscs (as rule, queues)
59 attached to them etc. etc. etc.
60
61 The goal of the routines in this file is to translate
62 information supplied by user in the form of handles
63 to more intelligible for kernel form, to make some sanity
64 checks and part of work, which is common to all qdiscs
65 and to provide rtnetlink notifications.
66
67 All real intelligent work is done inside qdisc modules.
68
69
70
71 Every discipline has two major routines: enqueue and dequeue.
72
73 ---dequeue
74
75 dequeue usually returns a skb to send. It is allowed to return NULL,
76 but it does not mean that queue is empty, it just means that
77 discipline does not want to send anything this time.
78 Queue is really empty if q->q.qlen == 0.
79 For complicated disciplines with multiple queues q->q is not
80 real packet queue, but however q->q.qlen must be valid.
81
82 ---enqueue
83
84 enqueue returns 0, if packet was enqueued successfully.
85 If packet (this one or another one) was dropped, it returns
86 not zero error code.
87 NET_XMIT_DROP - this packet dropped
88 Expected action: do not backoff, but wait until queue will clear.
89 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
90 Expected action: backoff or ignore
91
92 Auxiliary routines:
93
94 ---peek
95
96 like dequeue but without removing a packet from the queue
97
98 ---reset
99
100 returns qdisc to initial state: purge all buffers, clear all
101 timers, counters (except for statistics) etc.
102
103 ---init
104
105 initializes newly created qdisc.
106
107 ---destroy
108
109 destroys resources allocated by init and during lifetime of qdisc.
110
111 ---change
112
113 changes qdisc parameters.
114 */
115
116 /* Protects list of registered TC modules. It is pure SMP lock. */
117 static DEFINE_RWLOCK(qdisc_mod_lock);
118
119
120 /************************************************
121 * Queueing disciplines manipulation. *
122 ************************************************/
123
124
125 /* The list of all installed queueing disciplines. */
126
127 static struct Qdisc_ops *qdisc_base;
128
129 /* Register/unregister queueing discipline */
130
register_qdisc(struct Qdisc_ops * qops)131 int register_qdisc(struct Qdisc_ops *qops)
132 {
133 struct Qdisc_ops *q, **qp;
134 int rc = -EEXIST;
135
136 write_lock(&qdisc_mod_lock);
137 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
138 if (!strcmp(qops->id, q->id))
139 goto out;
140
141 if (qops->enqueue == NULL)
142 qops->enqueue = noop_qdisc_ops.enqueue;
143 if (qops->peek == NULL) {
144 if (qops->dequeue == NULL)
145 qops->peek = noop_qdisc_ops.peek;
146 else
147 goto out_einval;
148 }
149 if (qops->dequeue == NULL)
150 qops->dequeue = noop_qdisc_ops.dequeue;
151
152 if (qops->cl_ops) {
153 const struct Qdisc_class_ops *cops = qops->cl_ops;
154
155 if (!(cops->find && cops->walk && cops->leaf))
156 goto out_einval;
157
158 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
159 goto out_einval;
160 }
161
162 qops->next = NULL;
163 *qp = qops;
164 rc = 0;
165 out:
166 write_unlock(&qdisc_mod_lock);
167 return rc;
168
169 out_einval:
170 rc = -EINVAL;
171 goto out;
172 }
173 EXPORT_SYMBOL(register_qdisc);
174
unregister_qdisc(struct Qdisc_ops * qops)175 void unregister_qdisc(struct Qdisc_ops *qops)
176 {
177 struct Qdisc_ops *q, **qp;
178 int err = -ENOENT;
179
180 write_lock(&qdisc_mod_lock);
181 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
182 if (q == qops)
183 break;
184 if (q) {
185 *qp = q->next;
186 q->next = NULL;
187 err = 0;
188 }
189 write_unlock(&qdisc_mod_lock);
190
191 WARN(err, "unregister qdisc(%s) failed\n", qops->id);
192 }
193 EXPORT_SYMBOL(unregister_qdisc);
194
195 /* Get default qdisc if not otherwise specified */
qdisc_get_default(char * name,size_t len)196 void qdisc_get_default(char *name, size_t len)
197 {
198 read_lock(&qdisc_mod_lock);
199 strscpy(name, default_qdisc_ops->id, len);
200 read_unlock(&qdisc_mod_lock);
201 }
202
qdisc_lookup_default(const char * name)203 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
204 {
205 struct Qdisc_ops *q = NULL;
206
207 for (q = qdisc_base; q; q = q->next) {
208 if (!strcmp(name, q->id)) {
209 if (!try_module_get(q->owner))
210 q = NULL;
211 break;
212 }
213 }
214
215 return q;
216 }
217
218 /* Set new default qdisc to use */
qdisc_set_default(const char * name)219 int qdisc_set_default(const char *name)
220 {
221 const struct Qdisc_ops *ops;
222
223 if (!capable(CAP_NET_ADMIN))
224 return -EPERM;
225
226 write_lock(&qdisc_mod_lock);
227 ops = qdisc_lookup_default(name);
228 if (!ops) {
229 /* Not found, drop lock and try to load module */
230 write_unlock(&qdisc_mod_lock);
231 request_module("sch_%s", name);
232 write_lock(&qdisc_mod_lock);
233
234 ops = qdisc_lookup_default(name);
235 }
236
237 if (ops) {
238 /* Set new default */
239 module_put(default_qdisc_ops->owner);
240 default_qdisc_ops = ops;
241 }
242 write_unlock(&qdisc_mod_lock);
243
244 return ops ? 0 : -ENOENT;
245 }
246
247 #ifdef CONFIG_NET_SCH_DEFAULT
248 /* Set default value from kernel config */
sch_default_qdisc(void)249 static int __init sch_default_qdisc(void)
250 {
251 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
252 }
253 late_initcall(sch_default_qdisc);
254 #endif
255
256 /* We know handle. Find qdisc among all qdisc's attached to device
257 * (root qdisc, all its children, children of children etc.)
258 * Note: caller either uses rtnl or rcu_read_lock()
259 */
260
qdisc_match_from_root(struct Qdisc * root,u32 handle)261 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
262 {
263 struct Qdisc *q;
264
265 if (!qdisc_dev(root))
266 return (root->handle == handle ? root : NULL);
267
268 if (!(root->flags & TCQ_F_BUILTIN) &&
269 root->handle == handle)
270 return root;
271
272 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
273 lockdep_rtnl_is_held()) {
274 if (q->handle == handle)
275 return q;
276 }
277 return NULL;
278 }
279
qdisc_hash_add(struct Qdisc * q,bool invisible)280 void qdisc_hash_add(struct Qdisc *q, bool invisible)
281 {
282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
283 ASSERT_RTNL();
284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
285 if (invisible)
286 q->flags |= TCQ_F_INVISIBLE;
287 }
288 }
289 EXPORT_SYMBOL(qdisc_hash_add);
290
qdisc_hash_del(struct Qdisc * q)291 void qdisc_hash_del(struct Qdisc *q)
292 {
293 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
294 ASSERT_RTNL();
295 hash_del_rcu(&q->hash);
296 }
297 }
298 EXPORT_SYMBOL(qdisc_hash_del);
299
qdisc_lookup(struct net_device * dev,u32 handle)300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
301 {
302 struct Qdisc *q;
303
304 if (!handle)
305 return NULL;
306 q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
307 if (q)
308 goto out;
309
310 if (dev_ingress_queue(dev))
311 q = qdisc_match_from_root(
312 rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping),
313 handle);
314 out:
315 return q;
316 }
317
qdisc_lookup_rcu(struct net_device * dev,u32 handle)318 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
319 {
320 struct netdev_queue *nq;
321 struct Qdisc *q;
322
323 if (!handle)
324 return NULL;
325 q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
326 if (q)
327 goto out;
328
329 nq = dev_ingress_queue_rcu(dev);
330 if (nq)
331 q = qdisc_match_from_root(rcu_dereference(nq->qdisc_sleeping),
332 handle);
333 out:
334 return q;
335 }
336
qdisc_leaf(struct Qdisc * p,u32 classid)337 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
338 {
339 unsigned long cl;
340 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
341
342 if (cops == NULL)
343 return NULL;
344 cl = cops->find(p, classid);
345
346 if (cl == 0)
347 return NULL;
348 return cops->leaf(p, cl);
349 }
350
351 /* Find queueing discipline by name */
352
qdisc_lookup_ops(struct nlattr * kind)353 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
354 {
355 struct Qdisc_ops *q = NULL;
356
357 if (kind) {
358 read_lock(&qdisc_mod_lock);
359 for (q = qdisc_base; q; q = q->next) {
360 if (nla_strcmp(kind, q->id) == 0) {
361 if (!try_module_get(q->owner))
362 q = NULL;
363 break;
364 }
365 }
366 read_unlock(&qdisc_mod_lock);
367 }
368 return q;
369 }
370
371 /* The linklayer setting were not transferred from iproute2, in older
372 * versions, and the rate tables lookup systems have been dropped in
373 * the kernel. To keep backward compatible with older iproute2 tc
374 * utils, we detect the linklayer setting by detecting if the rate
375 * table were modified.
376 *
377 * For linklayer ATM table entries, the rate table will be aligned to
378 * 48 bytes, thus some table entries will contain the same value. The
379 * mpu (min packet unit) is also encoded into the old rate table, thus
380 * starting from the mpu, we find low and high table entries for
381 * mapping this cell. If these entries contain the same value, when
382 * the rate tables have been modified for linklayer ATM.
383 *
384 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
385 * and then roundup to the next cell, calc the table entry one below,
386 * and compare.
387 */
__detect_linklayer(struct tc_ratespec * r,__u32 * rtab)388 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
389 {
390 int low = roundup(r->mpu, 48);
391 int high = roundup(low+1, 48);
392 int cell_low = low >> r->cell_log;
393 int cell_high = (high >> r->cell_log) - 1;
394
395 /* rtab is too inaccurate at rates > 100Mbit/s */
396 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
397 pr_debug("TC linklayer: Giving up ATM detection\n");
398 return TC_LINKLAYER_ETHERNET;
399 }
400
401 if ((cell_high > cell_low) && (cell_high < 256)
402 && (rtab[cell_low] == rtab[cell_high])) {
403 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
404 cell_low, cell_high, rtab[cell_high]);
405 return TC_LINKLAYER_ATM;
406 }
407 return TC_LINKLAYER_ETHERNET;
408 }
409
410 static struct qdisc_rate_table *qdisc_rtab_list;
411
qdisc_get_rtab(struct tc_ratespec * r,struct nlattr * tab,struct netlink_ext_ack * extack)412 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
413 struct nlattr *tab,
414 struct netlink_ext_ack *extack)
415 {
416 struct qdisc_rate_table *rtab;
417
418 if (tab == NULL || r->rate == 0 ||
419 r->cell_log == 0 || r->cell_log >= 32 ||
420 nla_len(tab) != TC_RTAB_SIZE) {
421 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
422 return NULL;
423 }
424
425 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
426 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
427 !memcmp(&rtab->data, nla_data(tab), 1024)) {
428 rtab->refcnt++;
429 return rtab;
430 }
431 }
432
433 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
434 if (rtab) {
435 rtab->rate = *r;
436 rtab->refcnt = 1;
437 memcpy(rtab->data, nla_data(tab), 1024);
438 if (r->linklayer == TC_LINKLAYER_UNAWARE)
439 r->linklayer = __detect_linklayer(r, rtab->data);
440 rtab->next = qdisc_rtab_list;
441 qdisc_rtab_list = rtab;
442 } else {
443 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
444 }
445 return rtab;
446 }
447 EXPORT_SYMBOL(qdisc_get_rtab);
448
qdisc_put_rtab(struct qdisc_rate_table * tab)449 void qdisc_put_rtab(struct qdisc_rate_table *tab)
450 {
451 struct qdisc_rate_table *rtab, **rtabp;
452
453 if (!tab || --tab->refcnt)
454 return;
455
456 for (rtabp = &qdisc_rtab_list;
457 (rtab = *rtabp) != NULL;
458 rtabp = &rtab->next) {
459 if (rtab == tab) {
460 *rtabp = rtab->next;
461 kfree(rtab);
462 return;
463 }
464 }
465 }
466 EXPORT_SYMBOL(qdisc_put_rtab);
467
468 static LIST_HEAD(qdisc_stab_list);
469
470 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
471 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
472 [TCA_STAB_DATA] = { .type = NLA_BINARY },
473 };
474
qdisc_get_stab(struct nlattr * opt,struct netlink_ext_ack * extack)475 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
476 struct netlink_ext_ack *extack)
477 {
478 struct nlattr *tb[TCA_STAB_MAX + 1];
479 struct qdisc_size_table *stab;
480 struct tc_sizespec *s;
481 unsigned int tsize = 0;
482 u16 *tab = NULL;
483 int err;
484
485 err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
486 extack);
487 if (err < 0)
488 return ERR_PTR(err);
489 if (!tb[TCA_STAB_BASE]) {
490 NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
491 return ERR_PTR(-EINVAL);
492 }
493
494 s = nla_data(tb[TCA_STAB_BASE]);
495
496 if (s->tsize > 0) {
497 if (!tb[TCA_STAB_DATA]) {
498 NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
499 return ERR_PTR(-EINVAL);
500 }
501 tab = nla_data(tb[TCA_STAB_DATA]);
502 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
503 }
504
505 if (tsize != s->tsize || (!tab && tsize > 0)) {
506 NL_SET_ERR_MSG(extack, "Invalid size of size table");
507 return ERR_PTR(-EINVAL);
508 }
509
510 list_for_each_entry(stab, &qdisc_stab_list, list) {
511 if (memcmp(&stab->szopts, s, sizeof(*s)))
512 continue;
513 if (tsize > 0 &&
514 memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
515 continue;
516 stab->refcnt++;
517 return stab;
518 }
519
520 if (s->size_log > STAB_SIZE_LOG_MAX ||
521 s->cell_log > STAB_SIZE_LOG_MAX) {
522 NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
523 return ERR_PTR(-EINVAL);
524 }
525
526 stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
527 if (!stab)
528 return ERR_PTR(-ENOMEM);
529
530 stab->refcnt = 1;
531 stab->szopts = *s;
532 if (tsize > 0)
533 memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
534
535 list_add_tail(&stab->list, &qdisc_stab_list);
536
537 return stab;
538 }
539
qdisc_put_stab(struct qdisc_size_table * tab)540 void qdisc_put_stab(struct qdisc_size_table *tab)
541 {
542 if (!tab)
543 return;
544
545 if (--tab->refcnt == 0) {
546 list_del(&tab->list);
547 kfree_rcu(tab, rcu);
548 }
549 }
550 EXPORT_SYMBOL(qdisc_put_stab);
551
qdisc_dump_stab(struct sk_buff * skb,struct qdisc_size_table * stab)552 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
553 {
554 struct nlattr *nest;
555
556 nest = nla_nest_start_noflag(skb, TCA_STAB);
557 if (nest == NULL)
558 goto nla_put_failure;
559 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
560 goto nla_put_failure;
561 nla_nest_end(skb, nest);
562
563 return skb->len;
564
565 nla_put_failure:
566 return -1;
567 }
568
__qdisc_calculate_pkt_len(struct sk_buff * skb,const struct qdisc_size_table * stab)569 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
570 const struct qdisc_size_table *stab)
571 {
572 int pkt_len, slot;
573
574 pkt_len = skb->len + stab->szopts.overhead;
575 if (unlikely(!stab->szopts.tsize))
576 goto out;
577
578 slot = pkt_len + stab->szopts.cell_align;
579 if (unlikely(slot < 0))
580 slot = 0;
581
582 slot >>= stab->szopts.cell_log;
583 if (likely(slot < stab->szopts.tsize))
584 pkt_len = stab->data[slot];
585 else
586 pkt_len = stab->data[stab->szopts.tsize - 1] *
587 (slot / stab->szopts.tsize) +
588 stab->data[slot % stab->szopts.tsize];
589
590 pkt_len <<= stab->szopts.size_log;
591 out:
592 if (unlikely(pkt_len < 1))
593 pkt_len = 1;
594 qdisc_skb_cb(skb)->pkt_len = pkt_len;
595 }
596 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
597
qdisc_warn_nonwc(const char * txt,struct Qdisc * qdisc)598 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
599 {
600 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
601 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
602 txt, qdisc->ops->id, qdisc->handle >> 16);
603 qdisc->flags |= TCQ_F_WARN_NONWC;
604 }
605 }
606 EXPORT_SYMBOL(qdisc_warn_nonwc);
607
qdisc_watchdog(struct hrtimer * timer)608 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
609 {
610 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
611 timer);
612
613 rcu_read_lock();
614 __netif_schedule(qdisc_root(wd->qdisc));
615 rcu_read_unlock();
616
617 return HRTIMER_NORESTART;
618 }
619
qdisc_watchdog_init_clockid(struct qdisc_watchdog * wd,struct Qdisc * qdisc,clockid_t clockid)620 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
621 clockid_t clockid)
622 {
623 hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
624 wd->timer.function = qdisc_watchdog;
625 wd->qdisc = qdisc;
626 }
627 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
628
qdisc_watchdog_init(struct qdisc_watchdog * wd,struct Qdisc * qdisc)629 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
630 {
631 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
632 }
633 EXPORT_SYMBOL(qdisc_watchdog_init);
634
qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog * wd,u64 expires,u64 delta_ns)635 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
636 u64 delta_ns)
637 {
638 bool deactivated;
639
640 rcu_read_lock();
641 deactivated = test_bit(__QDISC_STATE_DEACTIVATED,
642 &qdisc_root_sleeping(wd->qdisc)->state);
643 rcu_read_unlock();
644 if (deactivated)
645 return;
646
647 if (hrtimer_is_queued(&wd->timer)) {
648 u64 softexpires;
649
650 softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer));
651 /* If timer is already set in [expires, expires + delta_ns],
652 * do not reprogram it.
653 */
654 if (softexpires - expires <= delta_ns)
655 return;
656 }
657
658 hrtimer_start_range_ns(&wd->timer,
659 ns_to_ktime(expires),
660 delta_ns,
661 HRTIMER_MODE_ABS_PINNED);
662 }
663 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
664
qdisc_watchdog_cancel(struct qdisc_watchdog * wd)665 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
666 {
667 hrtimer_cancel(&wd->timer);
668 }
669 EXPORT_SYMBOL(qdisc_watchdog_cancel);
670
qdisc_class_hash_alloc(unsigned int n)671 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
672 {
673 struct hlist_head *h;
674 unsigned int i;
675
676 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
677
678 if (h != NULL) {
679 for (i = 0; i < n; i++)
680 INIT_HLIST_HEAD(&h[i]);
681 }
682 return h;
683 }
684
qdisc_class_hash_grow(struct Qdisc * sch,struct Qdisc_class_hash * clhash)685 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
686 {
687 struct Qdisc_class_common *cl;
688 struct hlist_node *next;
689 struct hlist_head *nhash, *ohash;
690 unsigned int nsize, nmask, osize;
691 unsigned int i, h;
692
693 /* Rehash when load factor exceeds 0.75 */
694 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
695 return;
696 nsize = clhash->hashsize * 2;
697 nmask = nsize - 1;
698 nhash = qdisc_class_hash_alloc(nsize);
699 if (nhash == NULL)
700 return;
701
702 ohash = clhash->hash;
703 osize = clhash->hashsize;
704
705 sch_tree_lock(sch);
706 for (i = 0; i < osize; i++) {
707 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
708 h = qdisc_class_hash(cl->classid, nmask);
709 hlist_add_head(&cl->hnode, &nhash[h]);
710 }
711 }
712 clhash->hash = nhash;
713 clhash->hashsize = nsize;
714 clhash->hashmask = nmask;
715 sch_tree_unlock(sch);
716
717 kvfree(ohash);
718 }
719 EXPORT_SYMBOL(qdisc_class_hash_grow);
720
qdisc_class_hash_init(struct Qdisc_class_hash * clhash)721 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
722 {
723 unsigned int size = 4;
724
725 clhash->hash = qdisc_class_hash_alloc(size);
726 if (!clhash->hash)
727 return -ENOMEM;
728 clhash->hashsize = size;
729 clhash->hashmask = size - 1;
730 clhash->hashelems = 0;
731 return 0;
732 }
733 EXPORT_SYMBOL(qdisc_class_hash_init);
734
qdisc_class_hash_destroy(struct Qdisc_class_hash * clhash)735 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
736 {
737 kvfree(clhash->hash);
738 }
739 EXPORT_SYMBOL(qdisc_class_hash_destroy);
740
qdisc_class_hash_insert(struct Qdisc_class_hash * clhash,struct Qdisc_class_common * cl)741 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
742 struct Qdisc_class_common *cl)
743 {
744 unsigned int h;
745
746 INIT_HLIST_NODE(&cl->hnode);
747 h = qdisc_class_hash(cl->classid, clhash->hashmask);
748 hlist_add_head(&cl->hnode, &clhash->hash[h]);
749 clhash->hashelems++;
750 }
751 EXPORT_SYMBOL(qdisc_class_hash_insert);
752
qdisc_class_hash_remove(struct Qdisc_class_hash * clhash,struct Qdisc_class_common * cl)753 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
754 struct Qdisc_class_common *cl)
755 {
756 hlist_del(&cl->hnode);
757 clhash->hashelems--;
758 }
759 EXPORT_SYMBOL(qdisc_class_hash_remove);
760
761 /* Allocate an unique handle from space managed by kernel
762 * Possible range is [8000-FFFF]:0000 (0x8000 values)
763 */
qdisc_alloc_handle(struct net_device * dev)764 static u32 qdisc_alloc_handle(struct net_device *dev)
765 {
766 int i = 0x8000;
767 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
768
769 do {
770 autohandle += TC_H_MAKE(0x10000U, 0);
771 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
772 autohandle = TC_H_MAKE(0x80000000U, 0);
773 if (!qdisc_lookup(dev, autohandle))
774 return autohandle;
775 cond_resched();
776 } while (--i > 0);
777
778 return 0;
779 }
780
qdisc_tree_reduce_backlog(struct Qdisc * sch,int n,int len)781 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
782 {
783 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
784 const struct Qdisc_class_ops *cops;
785 unsigned long cl;
786 u32 parentid;
787 bool notify;
788 int drops;
789
790 if (n == 0 && len == 0)
791 return;
792 drops = max_t(int, n, 0);
793 rcu_read_lock();
794 while ((parentid = sch->parent)) {
795 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
796 break;
797
798 if (sch->flags & TCQ_F_NOPARENT)
799 break;
800 /* Notify parent qdisc only if child qdisc becomes empty.
801 *
802 * If child was empty even before update then backlog
803 * counter is screwed and we skip notification because
804 * parent class is already passive.
805 *
806 * If the original child was offloaded then it is allowed
807 * to be seem as empty, so the parent is notified anyway.
808 */
809 notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
810 !qdisc_is_offloaded);
811 /* TODO: perform the search on a per txq basis */
812 sch = qdisc_lookup_rcu(qdisc_dev(sch), TC_H_MAJ(parentid));
813 if (sch == NULL) {
814 WARN_ON_ONCE(parentid != TC_H_ROOT);
815 break;
816 }
817 cops = sch->ops->cl_ops;
818 if (notify && cops->qlen_notify) {
819 cl = cops->find(sch, parentid);
820 cops->qlen_notify(sch, cl);
821 }
822 sch->q.qlen -= n;
823 sch->qstats.backlog -= len;
824 __qdisc_qstats_drop(sch, drops);
825 }
826 rcu_read_unlock();
827 }
828 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
829
qdisc_offload_dump_helper(struct Qdisc * sch,enum tc_setup_type type,void * type_data)830 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
831 void *type_data)
832 {
833 struct net_device *dev = qdisc_dev(sch);
834 int err;
835
836 sch->flags &= ~TCQ_F_OFFLOADED;
837 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
838 return 0;
839
840 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
841 if (err == -EOPNOTSUPP)
842 return 0;
843
844 if (!err)
845 sch->flags |= TCQ_F_OFFLOADED;
846
847 return err;
848 }
849 EXPORT_SYMBOL(qdisc_offload_dump_helper);
850
qdisc_offload_graft_helper(struct net_device * dev,struct Qdisc * sch,struct Qdisc * new,struct Qdisc * old,enum tc_setup_type type,void * type_data,struct netlink_ext_ack * extack)851 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
852 struct Qdisc *new, struct Qdisc *old,
853 enum tc_setup_type type, void *type_data,
854 struct netlink_ext_ack *extack)
855 {
856 bool any_qdisc_is_offloaded;
857 int err;
858
859 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
860 return;
861
862 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
863
864 /* Don't report error if the graft is part of destroy operation. */
865 if (!err || !new || new == &noop_qdisc)
866 return;
867
868 /* Don't report error if the parent, the old child and the new
869 * one are not offloaded.
870 */
871 any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
872 any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
873 any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
874
875 if (any_qdisc_is_offloaded)
876 NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
877 }
878 EXPORT_SYMBOL(qdisc_offload_graft_helper);
879
qdisc_offload_query_caps(struct net_device * dev,enum tc_setup_type type,void * caps,size_t caps_len)880 void qdisc_offload_query_caps(struct net_device *dev,
881 enum tc_setup_type type,
882 void *caps, size_t caps_len)
883 {
884 const struct net_device_ops *ops = dev->netdev_ops;
885 struct tc_query_caps_base base = {
886 .type = type,
887 .caps = caps,
888 };
889
890 memset(caps, 0, caps_len);
891
892 if (ops->ndo_setup_tc)
893 ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
894 }
895 EXPORT_SYMBOL(qdisc_offload_query_caps);
896
qdisc_offload_graft_root(struct net_device * dev,struct Qdisc * new,struct Qdisc * old,struct netlink_ext_ack * extack)897 static void qdisc_offload_graft_root(struct net_device *dev,
898 struct Qdisc *new, struct Qdisc *old,
899 struct netlink_ext_ack *extack)
900 {
901 struct tc_root_qopt_offload graft_offload = {
902 .command = TC_ROOT_GRAFT,
903 .handle = new ? new->handle : 0,
904 .ingress = (new && new->flags & TCQ_F_INGRESS) ||
905 (old && old->flags & TCQ_F_INGRESS),
906 };
907
908 qdisc_offload_graft_helper(dev, NULL, new, old,
909 TC_SETUP_ROOT_QDISC, &graft_offload, extack);
910 }
911
tc_fill_qdisc(struct sk_buff * skb,struct Qdisc * q,u32 clid,u32 portid,u32 seq,u16 flags,int event,struct netlink_ext_ack * extack)912 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
913 u32 portid, u32 seq, u16 flags, int event,
914 struct netlink_ext_ack *extack)
915 {
916 struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
917 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
918 struct tcmsg *tcm;
919 struct nlmsghdr *nlh;
920 unsigned char *b = skb_tail_pointer(skb);
921 struct gnet_dump d;
922 struct qdisc_size_table *stab;
923 u32 block_index;
924 __u32 qlen;
925
926 cond_resched();
927 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
928 if (!nlh)
929 goto out_nlmsg_trim;
930 tcm = nlmsg_data(nlh);
931 tcm->tcm_family = AF_UNSPEC;
932 tcm->tcm__pad1 = 0;
933 tcm->tcm__pad2 = 0;
934 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
935 tcm->tcm_parent = clid;
936 tcm->tcm_handle = q->handle;
937 tcm->tcm_info = refcount_read(&q->refcnt);
938 if (nla_put_string(skb, TCA_KIND, q->ops->id))
939 goto nla_put_failure;
940 if (q->ops->ingress_block_get) {
941 block_index = q->ops->ingress_block_get(q);
942 if (block_index &&
943 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
944 goto nla_put_failure;
945 }
946 if (q->ops->egress_block_get) {
947 block_index = q->ops->egress_block_get(q);
948 if (block_index &&
949 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
950 goto nla_put_failure;
951 }
952 if (q->ops->dump && q->ops->dump(q, skb) < 0)
953 goto nla_put_failure;
954 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
955 goto nla_put_failure;
956 qlen = qdisc_qlen_sum(q);
957
958 stab = rtnl_dereference(q->stab);
959 if (stab && qdisc_dump_stab(skb, stab) < 0)
960 goto nla_put_failure;
961
962 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
963 NULL, &d, TCA_PAD) < 0)
964 goto nla_put_failure;
965
966 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
967 goto nla_put_failure;
968
969 if (qdisc_is_percpu_stats(q)) {
970 cpu_bstats = q->cpu_bstats;
971 cpu_qstats = q->cpu_qstats;
972 }
973
974 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
975 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
976 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
977 goto nla_put_failure;
978
979 if (gnet_stats_finish_copy(&d) < 0)
980 goto nla_put_failure;
981
982 if (extack && extack->_msg &&
983 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
984 goto out_nlmsg_trim;
985
986 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
987
988 return skb->len;
989
990 out_nlmsg_trim:
991 nla_put_failure:
992 nlmsg_trim(skb, b);
993 return -1;
994 }
995
tc_qdisc_dump_ignore(struct Qdisc * q,bool dump_invisible)996 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
997 {
998 if (q->flags & TCQ_F_BUILTIN)
999 return true;
1000 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
1001 return true;
1002
1003 return false;
1004 }
1005
qdisc_notify(struct net * net,struct sk_buff * oskb,struct nlmsghdr * n,u32 clid,struct Qdisc * old,struct Qdisc * new,struct netlink_ext_ack * extack)1006 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1007 struct nlmsghdr *n, u32 clid,
1008 struct Qdisc *old, struct Qdisc *new,
1009 struct netlink_ext_ack *extack)
1010 {
1011 struct sk_buff *skb;
1012 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1013
1014 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1015 if (!skb)
1016 return -ENOBUFS;
1017
1018 if (old && !tc_qdisc_dump_ignore(old, false)) {
1019 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1020 0, RTM_DELQDISC, extack) < 0)
1021 goto err_out;
1022 }
1023 if (new && !tc_qdisc_dump_ignore(new, false)) {
1024 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1025 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1026 goto err_out;
1027 }
1028
1029 if (skb->len)
1030 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1031 n->nlmsg_flags & NLM_F_ECHO);
1032
1033 err_out:
1034 kfree_skb(skb);
1035 return -EINVAL;
1036 }
1037
notify_and_destroy(struct net * net,struct sk_buff * skb,struct nlmsghdr * n,u32 clid,struct Qdisc * old,struct Qdisc * new,struct netlink_ext_ack * extack)1038 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1039 struct nlmsghdr *n, u32 clid,
1040 struct Qdisc *old, struct Qdisc *new,
1041 struct netlink_ext_ack *extack)
1042 {
1043 if (new || old)
1044 qdisc_notify(net, skb, n, clid, old, new, extack);
1045
1046 if (old)
1047 qdisc_put(old);
1048 }
1049
qdisc_clear_nolock(struct Qdisc * sch)1050 static void qdisc_clear_nolock(struct Qdisc *sch)
1051 {
1052 sch->flags &= ~TCQ_F_NOLOCK;
1053 if (!(sch->flags & TCQ_F_CPUSTATS))
1054 return;
1055
1056 free_percpu(sch->cpu_bstats);
1057 free_percpu(sch->cpu_qstats);
1058 sch->cpu_bstats = NULL;
1059 sch->cpu_qstats = NULL;
1060 sch->flags &= ~TCQ_F_CPUSTATS;
1061 }
1062
1063 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1064 * to device "dev".
1065 *
1066 * When appropriate send a netlink notification using 'skb'
1067 * and "n".
1068 *
1069 * On success, destroy old qdisc.
1070 */
1071
qdisc_graft(struct net_device * dev,struct Qdisc * parent,struct sk_buff * skb,struct nlmsghdr * n,u32 classid,struct Qdisc * new,struct Qdisc * old,struct netlink_ext_ack * extack)1072 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1073 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1074 struct Qdisc *new, struct Qdisc *old,
1075 struct netlink_ext_ack *extack)
1076 {
1077 struct Qdisc *q = old;
1078 struct net *net = dev_net(dev);
1079
1080 if (parent == NULL) {
1081 unsigned int i, num_q, ingress;
1082 struct netdev_queue *dev_queue;
1083
1084 ingress = 0;
1085 num_q = dev->num_tx_queues;
1086 if ((q && q->flags & TCQ_F_INGRESS) ||
1087 (new && new->flags & TCQ_F_INGRESS)) {
1088 ingress = 1;
1089 dev_queue = dev_ingress_queue(dev);
1090 if (!dev_queue) {
1091 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1092 return -ENOENT;
1093 }
1094
1095 q = rtnl_dereference(dev_queue->qdisc_sleeping);
1096
1097 /* This is the counterpart of that qdisc_refcount_inc_nz() call in
1098 * __tcf_qdisc_find() for filter requests.
1099 */
1100 if (!qdisc_refcount_dec_if_one(q)) {
1101 NL_SET_ERR_MSG(extack,
1102 "Current ingress or clsact Qdisc has ongoing filter requests");
1103 return -EBUSY;
1104 }
1105 }
1106
1107 if (dev->flags & IFF_UP)
1108 dev_deactivate(dev);
1109
1110 qdisc_offload_graft_root(dev, new, old, extack);
1111
1112 if (new && new->ops->attach && !ingress)
1113 goto skip;
1114
1115 if (!ingress) {
1116 for (i = 0; i < num_q; i++) {
1117 dev_queue = netdev_get_tx_queue(dev, i);
1118 old = dev_graft_qdisc(dev_queue, new);
1119
1120 if (new && i > 0)
1121 qdisc_refcount_inc(new);
1122 qdisc_put(old);
1123 }
1124 } else {
1125 old = dev_graft_qdisc(dev_queue, NULL);
1126
1127 /* {ingress,clsact}_destroy() @old before grafting @new to avoid
1128 * unprotected concurrent accesses to net_device::miniq_{in,e}gress
1129 * pointer(s) in mini_qdisc_pair_swap().
1130 */
1131 qdisc_notify(net, skb, n, classid, old, new, extack);
1132 qdisc_destroy(old);
1133
1134 dev_graft_qdisc(dev_queue, new);
1135 }
1136
1137 skip:
1138 if (!ingress) {
1139 old = rtnl_dereference(dev->qdisc);
1140 if (new && !new->ops->attach)
1141 qdisc_refcount_inc(new);
1142 rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1143
1144 notify_and_destroy(net, skb, n, classid, old, new, extack);
1145
1146 if (new && new->ops->attach)
1147 new->ops->attach(new);
1148 }
1149
1150 if (dev->flags & IFF_UP)
1151 dev_activate(dev);
1152 } else {
1153 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1154 unsigned long cl;
1155 int err;
1156
1157 /* Only support running class lockless if parent is lockless */
1158 if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1159 qdisc_clear_nolock(new);
1160
1161 if (!cops || !cops->graft)
1162 return -EOPNOTSUPP;
1163
1164 cl = cops->find(parent, classid);
1165 if (!cl) {
1166 NL_SET_ERR_MSG(extack, "Specified class not found");
1167 return -ENOENT;
1168 }
1169
1170 if (new && new->ops == &noqueue_qdisc_ops) {
1171 NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1172 return -EINVAL;
1173 }
1174
1175 err = cops->graft(parent, cl, new, &old, extack);
1176 if (err)
1177 return err;
1178 notify_and_destroy(net, skb, n, classid, old, new, extack);
1179 }
1180 return 0;
1181 }
1182
qdisc_block_indexes_set(struct Qdisc * sch,struct nlattr ** tca,struct netlink_ext_ack * extack)1183 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1184 struct netlink_ext_ack *extack)
1185 {
1186 u32 block_index;
1187
1188 if (tca[TCA_INGRESS_BLOCK]) {
1189 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1190
1191 if (!block_index) {
1192 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1193 return -EINVAL;
1194 }
1195 if (!sch->ops->ingress_block_set) {
1196 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1197 return -EOPNOTSUPP;
1198 }
1199 sch->ops->ingress_block_set(sch, block_index);
1200 }
1201 if (tca[TCA_EGRESS_BLOCK]) {
1202 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1203
1204 if (!block_index) {
1205 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1206 return -EINVAL;
1207 }
1208 if (!sch->ops->egress_block_set) {
1209 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1210 return -EOPNOTSUPP;
1211 }
1212 sch->ops->egress_block_set(sch, block_index);
1213 }
1214 return 0;
1215 }
1216
1217 /*
1218 Allocate and initialize new qdisc.
1219
1220 Parameters are passed via opt.
1221 */
1222
qdisc_create(struct net_device * dev,struct netdev_queue * dev_queue,u32 parent,u32 handle,struct nlattr ** tca,int * errp,struct netlink_ext_ack * extack)1223 static struct Qdisc *qdisc_create(struct net_device *dev,
1224 struct netdev_queue *dev_queue,
1225 u32 parent, u32 handle,
1226 struct nlattr **tca, int *errp,
1227 struct netlink_ext_ack *extack)
1228 {
1229 int err;
1230 struct nlattr *kind = tca[TCA_KIND];
1231 struct Qdisc *sch;
1232 struct Qdisc_ops *ops;
1233 struct qdisc_size_table *stab;
1234
1235 ops = qdisc_lookup_ops(kind);
1236 #ifdef CONFIG_MODULES
1237 if (ops == NULL && kind != NULL) {
1238 char name[IFNAMSIZ];
1239 if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1240 /* We dropped the RTNL semaphore in order to
1241 * perform the module load. So, even if we
1242 * succeeded in loading the module we have to
1243 * tell the caller to replay the request. We
1244 * indicate this using -EAGAIN.
1245 * We replay the request because the device may
1246 * go away in the mean time.
1247 */
1248 rtnl_unlock();
1249 request_module("sch_%s", name);
1250 rtnl_lock();
1251 ops = qdisc_lookup_ops(kind);
1252 if (ops != NULL) {
1253 /* We will try again qdisc_lookup_ops,
1254 * so don't keep a reference.
1255 */
1256 module_put(ops->owner);
1257 err = -EAGAIN;
1258 goto err_out;
1259 }
1260 }
1261 }
1262 #endif
1263
1264 err = -ENOENT;
1265 if (!ops) {
1266 NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1267 goto err_out;
1268 }
1269
1270 sch = qdisc_alloc(dev_queue, ops, extack);
1271 if (IS_ERR(sch)) {
1272 err = PTR_ERR(sch);
1273 goto err_out2;
1274 }
1275
1276 sch->parent = parent;
1277
1278 if (handle == TC_H_INGRESS) {
1279 if (!(sch->flags & TCQ_F_INGRESS)) {
1280 NL_SET_ERR_MSG(extack,
1281 "Specified parent ID is reserved for ingress and clsact Qdiscs");
1282 err = -EINVAL;
1283 goto err_out3;
1284 }
1285 handle = TC_H_MAKE(TC_H_INGRESS, 0);
1286 } else {
1287 if (handle == 0) {
1288 handle = qdisc_alloc_handle(dev);
1289 if (handle == 0) {
1290 NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1291 err = -ENOSPC;
1292 goto err_out3;
1293 }
1294 }
1295 if (!netif_is_multiqueue(dev))
1296 sch->flags |= TCQ_F_ONETXQUEUE;
1297 }
1298
1299 sch->handle = handle;
1300
1301 /* This exist to keep backward compatible with a userspace
1302 * loophole, what allowed userspace to get IFF_NO_QUEUE
1303 * facility on older kernels by setting tx_queue_len=0 (prior
1304 * to qdisc init), and then forgot to reinit tx_queue_len
1305 * before again attaching a qdisc.
1306 */
1307 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1308 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1309 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1310 }
1311
1312 err = qdisc_block_indexes_set(sch, tca, extack);
1313 if (err)
1314 goto err_out3;
1315
1316 if (tca[TCA_STAB]) {
1317 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1318 if (IS_ERR(stab)) {
1319 err = PTR_ERR(stab);
1320 goto err_out3;
1321 }
1322 rcu_assign_pointer(sch->stab, stab);
1323 }
1324
1325 if (ops->init) {
1326 err = ops->init(sch, tca[TCA_OPTIONS], extack);
1327 if (err != 0)
1328 goto err_out4;
1329 }
1330
1331 if (tca[TCA_RATE]) {
1332 err = -EOPNOTSUPP;
1333 if (sch->flags & TCQ_F_MQROOT) {
1334 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1335 goto err_out4;
1336 }
1337
1338 err = gen_new_estimator(&sch->bstats,
1339 sch->cpu_bstats,
1340 &sch->rate_est,
1341 NULL,
1342 true,
1343 tca[TCA_RATE]);
1344 if (err) {
1345 NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1346 goto err_out4;
1347 }
1348 }
1349
1350 qdisc_hash_add(sch, false);
1351 trace_qdisc_create(ops, dev, parent);
1352
1353 return sch;
1354
1355 err_out4:
1356 /* Even if ops->init() failed, we call ops->destroy()
1357 * like qdisc_create_dflt().
1358 */
1359 if (ops->destroy)
1360 ops->destroy(sch);
1361 qdisc_put_stab(rtnl_dereference(sch->stab));
1362 err_out3:
1363 lockdep_unregister_key(&sch->root_lock_key);
1364 netdev_put(dev, &sch->dev_tracker);
1365 qdisc_free(sch);
1366 err_out2:
1367 module_put(ops->owner);
1368 err_out:
1369 *errp = err;
1370 return NULL;
1371 }
1372
qdisc_change(struct Qdisc * sch,struct nlattr ** tca,struct netlink_ext_ack * extack)1373 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1374 struct netlink_ext_ack *extack)
1375 {
1376 struct qdisc_size_table *ostab, *stab = NULL;
1377 int err = 0;
1378
1379 if (tca[TCA_OPTIONS]) {
1380 if (!sch->ops->change) {
1381 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1382 return -EINVAL;
1383 }
1384 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1385 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1386 return -EOPNOTSUPP;
1387 }
1388 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1389 if (err)
1390 return err;
1391 }
1392
1393 if (tca[TCA_STAB]) {
1394 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1395 if (IS_ERR(stab))
1396 return PTR_ERR(stab);
1397 }
1398
1399 ostab = rtnl_dereference(sch->stab);
1400 rcu_assign_pointer(sch->stab, stab);
1401 qdisc_put_stab(ostab);
1402
1403 if (tca[TCA_RATE]) {
1404 /* NB: ignores errors from replace_estimator
1405 because change can't be undone. */
1406 if (sch->flags & TCQ_F_MQROOT)
1407 goto out;
1408 gen_replace_estimator(&sch->bstats,
1409 sch->cpu_bstats,
1410 &sch->rate_est,
1411 NULL,
1412 true,
1413 tca[TCA_RATE]);
1414 }
1415 out:
1416 return 0;
1417 }
1418
1419 struct check_loop_arg {
1420 struct qdisc_walker w;
1421 struct Qdisc *p;
1422 int depth;
1423 };
1424
1425 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1426 struct qdisc_walker *w);
1427
check_loop(struct Qdisc * q,struct Qdisc * p,int depth)1428 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1429 {
1430 struct check_loop_arg arg;
1431
1432 if (q->ops->cl_ops == NULL)
1433 return 0;
1434
1435 arg.w.stop = arg.w.skip = arg.w.count = 0;
1436 arg.w.fn = check_loop_fn;
1437 arg.depth = depth;
1438 arg.p = p;
1439 q->ops->cl_ops->walk(q, &arg.w);
1440 return arg.w.stop ? -ELOOP : 0;
1441 }
1442
1443 static int
check_loop_fn(struct Qdisc * q,unsigned long cl,struct qdisc_walker * w)1444 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1445 {
1446 struct Qdisc *leaf;
1447 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1448 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1449
1450 leaf = cops->leaf(q, cl);
1451 if (leaf) {
1452 if (leaf == arg->p || arg->depth > 7)
1453 return -ELOOP;
1454 return check_loop(leaf, arg->p, arg->depth + 1);
1455 }
1456 return 0;
1457 }
1458
1459 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1460 [TCA_KIND] = { .type = NLA_STRING },
1461 [TCA_RATE] = { .type = NLA_BINARY,
1462 .len = sizeof(struct tc_estimator) },
1463 [TCA_STAB] = { .type = NLA_NESTED },
1464 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
1465 [TCA_CHAIN] = { .type = NLA_U32 },
1466 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
1467 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
1468 };
1469
1470 /*
1471 * Delete/get qdisc.
1472 */
1473
tc_get_qdisc(struct sk_buff * skb,struct nlmsghdr * n,struct netlink_ext_ack * extack)1474 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1475 struct netlink_ext_ack *extack)
1476 {
1477 struct net *net = sock_net(skb->sk);
1478 struct tcmsg *tcm = nlmsg_data(n);
1479 struct nlattr *tca[TCA_MAX + 1];
1480 struct net_device *dev;
1481 u32 clid;
1482 struct Qdisc *q = NULL;
1483 struct Qdisc *p = NULL;
1484 int err;
1485
1486 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1487 rtm_tca_policy, extack);
1488 if (err < 0)
1489 return err;
1490
1491 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1492 if (!dev)
1493 return -ENODEV;
1494
1495 clid = tcm->tcm_parent;
1496 if (clid) {
1497 if (clid != TC_H_ROOT) {
1498 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1499 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1500 if (!p) {
1501 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1502 return -ENOENT;
1503 }
1504 q = qdisc_leaf(p, clid);
1505 } else if (dev_ingress_queue(dev)) {
1506 q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1507 }
1508 } else {
1509 q = rtnl_dereference(dev->qdisc);
1510 }
1511 if (!q) {
1512 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1513 return -ENOENT;
1514 }
1515
1516 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1517 NL_SET_ERR_MSG(extack, "Invalid handle");
1518 return -EINVAL;
1519 }
1520 } else {
1521 q = qdisc_lookup(dev, tcm->tcm_handle);
1522 if (!q) {
1523 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1524 return -ENOENT;
1525 }
1526 }
1527
1528 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1529 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1530 return -EINVAL;
1531 }
1532
1533 if (n->nlmsg_type == RTM_DELQDISC) {
1534 if (!clid) {
1535 NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1536 return -EINVAL;
1537 }
1538 if (q->handle == 0) {
1539 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1540 return -ENOENT;
1541 }
1542 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1543 if (err != 0)
1544 return err;
1545 } else {
1546 qdisc_notify(net, skb, n, clid, NULL, q, NULL);
1547 }
1548 return 0;
1549 }
1550
req_create_or_replace(struct nlmsghdr * n)1551 static bool req_create_or_replace(struct nlmsghdr *n)
1552 {
1553 return (n->nlmsg_flags & NLM_F_CREATE &&
1554 n->nlmsg_flags & NLM_F_REPLACE);
1555 }
1556
req_create_exclusive(struct nlmsghdr * n)1557 static bool req_create_exclusive(struct nlmsghdr *n)
1558 {
1559 return (n->nlmsg_flags & NLM_F_CREATE &&
1560 n->nlmsg_flags & NLM_F_EXCL);
1561 }
1562
req_change(struct nlmsghdr * n)1563 static bool req_change(struct nlmsghdr *n)
1564 {
1565 return (!(n->nlmsg_flags & NLM_F_CREATE) &&
1566 !(n->nlmsg_flags & NLM_F_REPLACE) &&
1567 !(n->nlmsg_flags & NLM_F_EXCL));
1568 }
1569
1570 /*
1571 * Create/change qdisc.
1572 */
tc_modify_qdisc(struct sk_buff * skb,struct nlmsghdr * n,struct netlink_ext_ack * extack)1573 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1574 struct netlink_ext_ack *extack)
1575 {
1576 struct net *net = sock_net(skb->sk);
1577 struct tcmsg *tcm;
1578 struct nlattr *tca[TCA_MAX + 1];
1579 struct net_device *dev;
1580 u32 clid;
1581 struct Qdisc *q, *p;
1582 int err;
1583
1584 replay:
1585 /* Reinit, just in case something touches this. */
1586 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1587 rtm_tca_policy, extack);
1588 if (err < 0)
1589 return err;
1590
1591 tcm = nlmsg_data(n);
1592 clid = tcm->tcm_parent;
1593 q = p = NULL;
1594
1595 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1596 if (!dev)
1597 return -ENODEV;
1598
1599
1600 if (clid) {
1601 if (clid != TC_H_ROOT) {
1602 if (clid != TC_H_INGRESS) {
1603 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1604 if (!p) {
1605 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1606 return -ENOENT;
1607 }
1608 q = qdisc_leaf(p, clid);
1609 } else if (dev_ingress_queue_create(dev)) {
1610 q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1611 }
1612 } else {
1613 q = rtnl_dereference(dev->qdisc);
1614 }
1615
1616 /* It may be default qdisc, ignore it */
1617 if (q && q->handle == 0)
1618 q = NULL;
1619
1620 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1621 if (tcm->tcm_handle) {
1622 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1623 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1624 return -EEXIST;
1625 }
1626 if (TC_H_MIN(tcm->tcm_handle)) {
1627 NL_SET_ERR_MSG(extack, "Invalid minor handle");
1628 return -EINVAL;
1629 }
1630 q = qdisc_lookup(dev, tcm->tcm_handle);
1631 if (!q)
1632 goto create_n_graft;
1633 if (n->nlmsg_flags & NLM_F_EXCL) {
1634 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1635 return -EEXIST;
1636 }
1637 if (tca[TCA_KIND] &&
1638 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1639 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1640 return -EINVAL;
1641 }
1642 if (q->flags & TCQ_F_INGRESS) {
1643 NL_SET_ERR_MSG(extack,
1644 "Cannot regraft ingress or clsact Qdiscs");
1645 return -EINVAL;
1646 }
1647 if (q == p ||
1648 (p && check_loop(q, p, 0))) {
1649 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1650 return -ELOOP;
1651 }
1652 if (clid == TC_H_INGRESS) {
1653 NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
1654 return -EINVAL;
1655 }
1656 qdisc_refcount_inc(q);
1657 goto graft;
1658 } else {
1659 if (!q)
1660 goto create_n_graft;
1661
1662 /* This magic test requires explanation.
1663 *
1664 * We know, that some child q is already
1665 * attached to this parent and have choice:
1666 * 1) change it or 2) create/graft new one.
1667 * If the requested qdisc kind is different
1668 * than the existing one, then we choose graft.
1669 * If they are the same then this is "change"
1670 * operation - just let it fallthrough..
1671 *
1672 * 1. We are allowed to create/graft only
1673 * if the request is explicitly stating
1674 * "please create if it doesn't exist".
1675 *
1676 * 2. If the request is to exclusive create
1677 * then the qdisc tcm_handle is not expected
1678 * to exist, so that we choose create/graft too.
1679 *
1680 * 3. The last case is when no flags are set.
1681 * This will happen when for example tc
1682 * utility issues a "change" command.
1683 * Alas, it is sort of hole in API, we
1684 * cannot decide what to do unambiguously.
1685 * For now we select create/graft.
1686 */
1687 if (tca[TCA_KIND] &&
1688 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1689 if (req_create_or_replace(n) ||
1690 req_create_exclusive(n))
1691 goto create_n_graft;
1692 else if (req_change(n))
1693 goto create_n_graft2;
1694 }
1695 }
1696 }
1697 } else {
1698 if (!tcm->tcm_handle) {
1699 NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1700 return -EINVAL;
1701 }
1702 q = qdisc_lookup(dev, tcm->tcm_handle);
1703 }
1704
1705 /* Change qdisc parameters */
1706 if (!q) {
1707 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1708 return -ENOENT;
1709 }
1710 if (n->nlmsg_flags & NLM_F_EXCL) {
1711 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1712 return -EEXIST;
1713 }
1714 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1715 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1716 return -EINVAL;
1717 }
1718 err = qdisc_change(q, tca, extack);
1719 if (err == 0)
1720 qdisc_notify(net, skb, n, clid, NULL, q, extack);
1721 return err;
1722
1723 create_n_graft:
1724 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1725 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1726 return -ENOENT;
1727 }
1728 create_n_graft2:
1729 if (clid == TC_H_INGRESS) {
1730 if (dev_ingress_queue(dev)) {
1731 q = qdisc_create(dev, dev_ingress_queue(dev),
1732 tcm->tcm_parent, tcm->tcm_parent,
1733 tca, &err, extack);
1734 } else {
1735 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1736 err = -ENOENT;
1737 }
1738 } else {
1739 struct netdev_queue *dev_queue;
1740
1741 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1742 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1743 else if (p)
1744 dev_queue = p->dev_queue;
1745 else
1746 dev_queue = netdev_get_tx_queue(dev, 0);
1747
1748 q = qdisc_create(dev, dev_queue,
1749 tcm->tcm_parent, tcm->tcm_handle,
1750 tca, &err, extack);
1751 }
1752 if (q == NULL) {
1753 if (err == -EAGAIN)
1754 goto replay;
1755 return err;
1756 }
1757
1758 graft:
1759 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1760 if (err) {
1761 if (q)
1762 qdisc_put(q);
1763 return err;
1764 }
1765
1766 return 0;
1767 }
1768
tc_dump_qdisc_root(struct Qdisc * root,struct sk_buff * skb,struct netlink_callback * cb,int * q_idx_p,int s_q_idx,bool recur,bool dump_invisible)1769 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1770 struct netlink_callback *cb,
1771 int *q_idx_p, int s_q_idx, bool recur,
1772 bool dump_invisible)
1773 {
1774 int ret = 0, q_idx = *q_idx_p;
1775 struct Qdisc *q;
1776 int b;
1777
1778 if (!root)
1779 return 0;
1780
1781 q = root;
1782 if (q_idx < s_q_idx) {
1783 q_idx++;
1784 } else {
1785 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1786 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1787 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1788 RTM_NEWQDISC, NULL) <= 0)
1789 goto done;
1790 q_idx++;
1791 }
1792
1793 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1794 * itself has already been dumped.
1795 *
1796 * If we've already dumped the top-level (ingress) qdisc above and the global
1797 * qdisc hashtable, we don't want to hit it again
1798 */
1799 if (!qdisc_dev(root) || !recur)
1800 goto out;
1801
1802 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1803 if (q_idx < s_q_idx) {
1804 q_idx++;
1805 continue;
1806 }
1807 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1808 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1809 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1810 RTM_NEWQDISC, NULL) <= 0)
1811 goto done;
1812 q_idx++;
1813 }
1814
1815 out:
1816 *q_idx_p = q_idx;
1817 return ret;
1818 done:
1819 ret = -1;
1820 goto out;
1821 }
1822
tc_dump_qdisc(struct sk_buff * skb,struct netlink_callback * cb)1823 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1824 {
1825 struct net *net = sock_net(skb->sk);
1826 int idx, q_idx;
1827 int s_idx, s_q_idx;
1828 struct net_device *dev;
1829 const struct nlmsghdr *nlh = cb->nlh;
1830 struct nlattr *tca[TCA_MAX + 1];
1831 int err;
1832
1833 s_idx = cb->args[0];
1834 s_q_idx = q_idx = cb->args[1];
1835
1836 idx = 0;
1837 ASSERT_RTNL();
1838
1839 err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1840 rtm_tca_policy, cb->extack);
1841 if (err < 0)
1842 return err;
1843
1844 for_each_netdev(net, dev) {
1845 struct netdev_queue *dev_queue;
1846
1847 if (idx < s_idx)
1848 goto cont;
1849 if (idx > s_idx)
1850 s_q_idx = 0;
1851 q_idx = 0;
1852
1853 if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1854 skb, cb, &q_idx, s_q_idx,
1855 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1856 goto done;
1857
1858 dev_queue = dev_ingress_queue(dev);
1859 if (dev_queue &&
1860 tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping),
1861 skb, cb, &q_idx, s_q_idx, false,
1862 tca[TCA_DUMP_INVISIBLE]) < 0)
1863 goto done;
1864
1865 cont:
1866 idx++;
1867 }
1868
1869 done:
1870 cb->args[0] = idx;
1871 cb->args[1] = q_idx;
1872
1873 return skb->len;
1874 }
1875
1876
1877
1878 /************************************************
1879 * Traffic classes manipulation. *
1880 ************************************************/
1881
tc_fill_tclass(struct sk_buff * skb,struct Qdisc * q,unsigned long cl,u32 portid,u32 seq,u16 flags,int event,struct netlink_ext_ack * extack)1882 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1883 unsigned long cl, u32 portid, u32 seq, u16 flags,
1884 int event, struct netlink_ext_ack *extack)
1885 {
1886 struct tcmsg *tcm;
1887 struct nlmsghdr *nlh;
1888 unsigned char *b = skb_tail_pointer(skb);
1889 struct gnet_dump d;
1890 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1891
1892 cond_resched();
1893 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1894 if (!nlh)
1895 goto out_nlmsg_trim;
1896 tcm = nlmsg_data(nlh);
1897 tcm->tcm_family = AF_UNSPEC;
1898 tcm->tcm__pad1 = 0;
1899 tcm->tcm__pad2 = 0;
1900 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1901 tcm->tcm_parent = q->handle;
1902 tcm->tcm_handle = q->handle;
1903 tcm->tcm_info = 0;
1904 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1905 goto nla_put_failure;
1906 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1907 goto nla_put_failure;
1908
1909 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1910 NULL, &d, TCA_PAD) < 0)
1911 goto nla_put_failure;
1912
1913 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1914 goto nla_put_failure;
1915
1916 if (gnet_stats_finish_copy(&d) < 0)
1917 goto nla_put_failure;
1918
1919 if (extack && extack->_msg &&
1920 nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1921 goto out_nlmsg_trim;
1922
1923 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1924
1925 return skb->len;
1926
1927 out_nlmsg_trim:
1928 nla_put_failure:
1929 nlmsg_trim(skb, b);
1930 return -1;
1931 }
1932
tclass_notify(struct net * net,struct sk_buff * oskb,struct nlmsghdr * n,struct Qdisc * q,unsigned long cl,int event,struct netlink_ext_ack * extack)1933 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1934 struct nlmsghdr *n, struct Qdisc *q,
1935 unsigned long cl, int event, struct netlink_ext_ack *extack)
1936 {
1937 struct sk_buff *skb;
1938 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1939
1940 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1941 if (!skb)
1942 return -ENOBUFS;
1943
1944 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
1945 kfree_skb(skb);
1946 return -EINVAL;
1947 }
1948
1949 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1950 n->nlmsg_flags & NLM_F_ECHO);
1951 }
1952
tclass_del_notify(struct net * net,const struct Qdisc_class_ops * cops,struct sk_buff * oskb,struct nlmsghdr * n,struct Qdisc * q,unsigned long cl,struct netlink_ext_ack * extack)1953 static int tclass_del_notify(struct net *net,
1954 const struct Qdisc_class_ops *cops,
1955 struct sk_buff *oskb, struct nlmsghdr *n,
1956 struct Qdisc *q, unsigned long cl,
1957 struct netlink_ext_ack *extack)
1958 {
1959 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1960 struct sk_buff *skb;
1961 int err = 0;
1962
1963 if (!cops->delete)
1964 return -EOPNOTSUPP;
1965
1966 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1967 if (!skb)
1968 return -ENOBUFS;
1969
1970 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1971 RTM_DELTCLASS, extack) < 0) {
1972 kfree_skb(skb);
1973 return -EINVAL;
1974 }
1975
1976 err = cops->delete(q, cl, extack);
1977 if (err) {
1978 kfree_skb(skb);
1979 return err;
1980 }
1981
1982 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1983 n->nlmsg_flags & NLM_F_ECHO);
1984 return err;
1985 }
1986
1987 #ifdef CONFIG_NET_CLS
1988
1989 struct tcf_bind_args {
1990 struct tcf_walker w;
1991 unsigned long base;
1992 unsigned long cl;
1993 u32 classid;
1994 };
1995
tcf_node_bind(struct tcf_proto * tp,void * n,struct tcf_walker * arg)1996 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1997 {
1998 struct tcf_bind_args *a = (void *)arg;
1999
2000 if (n && tp->ops->bind_class) {
2001 struct Qdisc *q = tcf_block_q(tp->chain->block);
2002
2003 sch_tree_lock(q);
2004 tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
2005 sch_tree_unlock(q);
2006 }
2007 return 0;
2008 }
2009
2010 struct tc_bind_class_args {
2011 struct qdisc_walker w;
2012 unsigned long new_cl;
2013 u32 portid;
2014 u32 clid;
2015 };
2016
tc_bind_class_walker(struct Qdisc * q,unsigned long cl,struct qdisc_walker * w)2017 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
2018 struct qdisc_walker *w)
2019 {
2020 struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
2021 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2022 struct tcf_block *block;
2023 struct tcf_chain *chain;
2024
2025 block = cops->tcf_block(q, cl, NULL);
2026 if (!block)
2027 return 0;
2028 for (chain = tcf_get_next_chain(block, NULL);
2029 chain;
2030 chain = tcf_get_next_chain(block, chain)) {
2031 struct tcf_proto *tp;
2032
2033 for (tp = tcf_get_next_proto(chain, NULL);
2034 tp; tp = tcf_get_next_proto(chain, tp)) {
2035 struct tcf_bind_args arg = {};
2036
2037 arg.w.fn = tcf_node_bind;
2038 arg.classid = a->clid;
2039 arg.base = cl;
2040 arg.cl = a->new_cl;
2041 tp->ops->walk(tp, &arg.w, true);
2042 }
2043 }
2044
2045 return 0;
2046 }
2047
tc_bind_tclass(struct Qdisc * q,u32 portid,u32 clid,unsigned long new_cl)2048 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2049 unsigned long new_cl)
2050 {
2051 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2052 struct tc_bind_class_args args = {};
2053
2054 if (!cops->tcf_block)
2055 return;
2056 args.portid = portid;
2057 args.clid = clid;
2058 args.new_cl = new_cl;
2059 args.w.fn = tc_bind_class_walker;
2060 q->ops->cl_ops->walk(q, &args.w);
2061 }
2062
2063 #else
2064
tc_bind_tclass(struct Qdisc * q,u32 portid,u32 clid,unsigned long new_cl)2065 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2066 unsigned long new_cl)
2067 {
2068 }
2069
2070 #endif
2071
tc_ctl_tclass(struct sk_buff * skb,struct nlmsghdr * n,struct netlink_ext_ack * extack)2072 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2073 struct netlink_ext_ack *extack)
2074 {
2075 struct net *net = sock_net(skb->sk);
2076 struct tcmsg *tcm = nlmsg_data(n);
2077 struct nlattr *tca[TCA_MAX + 1];
2078 struct net_device *dev;
2079 struct Qdisc *q = NULL;
2080 const struct Qdisc_class_ops *cops;
2081 unsigned long cl = 0;
2082 unsigned long new_cl;
2083 u32 portid;
2084 u32 clid;
2085 u32 qid;
2086 int err;
2087
2088 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2089 rtm_tca_policy, extack);
2090 if (err < 0)
2091 return err;
2092
2093 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2094 if (!dev)
2095 return -ENODEV;
2096
2097 /*
2098 parent == TC_H_UNSPEC - unspecified parent.
2099 parent == TC_H_ROOT - class is root, which has no parent.
2100 parent == X:0 - parent is root class.
2101 parent == X:Y - parent is a node in hierarchy.
2102 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
2103
2104 handle == 0:0 - generate handle from kernel pool.
2105 handle == 0:Y - class is X:Y, where X:0 is qdisc.
2106 handle == X:Y - clear.
2107 handle == X:0 - root class.
2108 */
2109
2110 /* Step 1. Determine qdisc handle X:0 */
2111
2112 portid = tcm->tcm_parent;
2113 clid = tcm->tcm_handle;
2114 qid = TC_H_MAJ(clid);
2115
2116 if (portid != TC_H_ROOT) {
2117 u32 qid1 = TC_H_MAJ(portid);
2118
2119 if (qid && qid1) {
2120 /* If both majors are known, they must be identical. */
2121 if (qid != qid1)
2122 return -EINVAL;
2123 } else if (qid1) {
2124 qid = qid1;
2125 } else if (qid == 0)
2126 qid = rtnl_dereference(dev->qdisc)->handle;
2127
2128 /* Now qid is genuine qdisc handle consistent
2129 * both with parent and child.
2130 *
2131 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2132 */
2133 if (portid)
2134 portid = TC_H_MAKE(qid, portid);
2135 } else {
2136 if (qid == 0)
2137 qid = rtnl_dereference(dev->qdisc)->handle;
2138 }
2139
2140 /* OK. Locate qdisc */
2141 q = qdisc_lookup(dev, qid);
2142 if (!q)
2143 return -ENOENT;
2144
2145 /* An check that it supports classes */
2146 cops = q->ops->cl_ops;
2147 if (cops == NULL)
2148 return -EINVAL;
2149
2150 /* Now try to get class */
2151 if (clid == 0) {
2152 if (portid == TC_H_ROOT)
2153 clid = qid;
2154 } else
2155 clid = TC_H_MAKE(qid, clid);
2156
2157 if (clid)
2158 cl = cops->find(q, clid);
2159
2160 if (cl == 0) {
2161 err = -ENOENT;
2162 if (n->nlmsg_type != RTM_NEWTCLASS ||
2163 !(n->nlmsg_flags & NLM_F_CREATE))
2164 goto out;
2165 } else {
2166 switch (n->nlmsg_type) {
2167 case RTM_NEWTCLASS:
2168 err = -EEXIST;
2169 if (n->nlmsg_flags & NLM_F_EXCL)
2170 goto out;
2171 break;
2172 case RTM_DELTCLASS:
2173 err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2174 /* Unbind the class with flilters with 0 */
2175 tc_bind_tclass(q, portid, clid, 0);
2176 goto out;
2177 case RTM_GETTCLASS:
2178 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS, extack);
2179 goto out;
2180 default:
2181 err = -EINVAL;
2182 goto out;
2183 }
2184 }
2185
2186 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2187 NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2188 return -EOPNOTSUPP;
2189 }
2190
2191 new_cl = cl;
2192 err = -EOPNOTSUPP;
2193 if (cops->change)
2194 err = cops->change(q, clid, portid, tca, &new_cl, extack);
2195 if (err == 0) {
2196 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2197 /* We just create a new class, need to do reverse binding. */
2198 if (cl != new_cl)
2199 tc_bind_tclass(q, portid, clid, new_cl);
2200 }
2201 out:
2202 return err;
2203 }
2204
2205 struct qdisc_dump_args {
2206 struct qdisc_walker w;
2207 struct sk_buff *skb;
2208 struct netlink_callback *cb;
2209 };
2210
qdisc_class_dump(struct Qdisc * q,unsigned long cl,struct qdisc_walker * arg)2211 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2212 struct qdisc_walker *arg)
2213 {
2214 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2215
2216 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2217 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2218 RTM_NEWTCLASS, NULL);
2219 }
2220
tc_dump_tclass_qdisc(struct Qdisc * q,struct sk_buff * skb,struct tcmsg * tcm,struct netlink_callback * cb,int * t_p,int s_t)2221 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2222 struct tcmsg *tcm, struct netlink_callback *cb,
2223 int *t_p, int s_t)
2224 {
2225 struct qdisc_dump_args arg;
2226
2227 if (tc_qdisc_dump_ignore(q, false) ||
2228 *t_p < s_t || !q->ops->cl_ops ||
2229 (tcm->tcm_parent &&
2230 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2231 (*t_p)++;
2232 return 0;
2233 }
2234 if (*t_p > s_t)
2235 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2236 arg.w.fn = qdisc_class_dump;
2237 arg.skb = skb;
2238 arg.cb = cb;
2239 arg.w.stop = 0;
2240 arg.w.skip = cb->args[1];
2241 arg.w.count = 0;
2242 q->ops->cl_ops->walk(q, &arg.w);
2243 cb->args[1] = arg.w.count;
2244 if (arg.w.stop)
2245 return -1;
2246 (*t_p)++;
2247 return 0;
2248 }
2249
tc_dump_tclass_root(struct Qdisc * root,struct sk_buff * skb,struct tcmsg * tcm,struct netlink_callback * cb,int * t_p,int s_t,bool recur)2250 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2251 struct tcmsg *tcm, struct netlink_callback *cb,
2252 int *t_p, int s_t, bool recur)
2253 {
2254 struct Qdisc *q;
2255 int b;
2256
2257 if (!root)
2258 return 0;
2259
2260 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2261 return -1;
2262
2263 if (!qdisc_dev(root) || !recur)
2264 return 0;
2265
2266 if (tcm->tcm_parent) {
2267 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2268 if (q && q != root &&
2269 tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2270 return -1;
2271 return 0;
2272 }
2273 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2274 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2275 return -1;
2276 }
2277
2278 return 0;
2279 }
2280
tc_dump_tclass(struct sk_buff * skb,struct netlink_callback * cb)2281 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2282 {
2283 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2284 struct net *net = sock_net(skb->sk);
2285 struct netdev_queue *dev_queue;
2286 struct net_device *dev;
2287 int t, s_t;
2288
2289 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2290 return 0;
2291 dev = dev_get_by_index(net, tcm->tcm_ifindex);
2292 if (!dev)
2293 return 0;
2294
2295 s_t = cb->args[0];
2296 t = 0;
2297
2298 if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2299 skb, tcm, cb, &t, s_t, true) < 0)
2300 goto done;
2301
2302 dev_queue = dev_ingress_queue(dev);
2303 if (dev_queue &&
2304 tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
2305 skb, tcm, cb, &t, s_t, false) < 0)
2306 goto done;
2307
2308 done:
2309 cb->args[0] = t;
2310
2311 dev_put(dev);
2312 return skb->len;
2313 }
2314
2315 #ifdef CONFIG_PROC_FS
psched_show(struct seq_file * seq,void * v)2316 static int psched_show(struct seq_file *seq, void *v)
2317 {
2318 seq_printf(seq, "%08x %08x %08x %08x\n",
2319 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2320 1000000,
2321 (u32)NSEC_PER_SEC / hrtimer_resolution);
2322
2323 return 0;
2324 }
2325
psched_net_init(struct net * net)2326 static int __net_init psched_net_init(struct net *net)
2327 {
2328 struct proc_dir_entry *e;
2329
2330 e = proc_create_single("psched", 0, net->proc_net, psched_show);
2331 if (e == NULL)
2332 return -ENOMEM;
2333
2334 return 0;
2335 }
2336
psched_net_exit(struct net * net)2337 static void __net_exit psched_net_exit(struct net *net)
2338 {
2339 remove_proc_entry("psched", net->proc_net);
2340 }
2341 #else
psched_net_init(struct net * net)2342 static int __net_init psched_net_init(struct net *net)
2343 {
2344 return 0;
2345 }
2346
psched_net_exit(struct net * net)2347 static void __net_exit psched_net_exit(struct net *net)
2348 {
2349 }
2350 #endif
2351
2352 static struct pernet_operations psched_net_ops = {
2353 .init = psched_net_init,
2354 .exit = psched_net_exit,
2355 };
2356
2357 #if IS_ENABLED(CONFIG_RETPOLINE)
2358 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2359 #endif
2360
pktsched_init(void)2361 static int __init pktsched_init(void)
2362 {
2363 int err;
2364
2365 err = register_pernet_subsys(&psched_net_ops);
2366 if (err) {
2367 pr_err("pktsched_init: "
2368 "cannot initialize per netns operations\n");
2369 return err;
2370 }
2371
2372 register_qdisc(&pfifo_fast_ops);
2373 register_qdisc(&pfifo_qdisc_ops);
2374 register_qdisc(&bfifo_qdisc_ops);
2375 register_qdisc(&pfifo_head_drop_qdisc_ops);
2376 register_qdisc(&mq_qdisc_ops);
2377 register_qdisc(&noqueue_qdisc_ops);
2378
2379 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2380 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2381 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2382 0);
2383 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2384 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2385 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2386 0);
2387
2388 tc_wrapper_init();
2389
2390 return 0;
2391 }
2392
2393 subsys_initcall(pktsched_init);
2394