1 /* 2 * net/sched/sch_api.c Packet scheduler API. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 * 11 * Fixes: 12 * 13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. 14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support 15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support 16 */ 17 18 #include <linux/module.h> 19 #include <linux/types.h> 20 #include <linux/kernel.h> 21 #include <linux/string.h> 22 #include <linux/errno.h> 23 #include <linux/skbuff.h> 24 #include <linux/init.h> 25 #include <linux/proc_fs.h> 26 #include <linux/seq_file.h> 27 #include <linux/kmod.h> 28 #include <linux/list.h> 29 #include <linux/hrtimer.h> 30 31 #include <net/net_namespace.h> 32 #include <net/sock.h> 33 #include <net/netlink.h> 34 #include <net/pkt_sched.h> 35 36 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, 37 struct Qdisc *old, struct Qdisc *new); 38 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, 39 struct Qdisc *q, unsigned long cl, int event); 40 41 /* 42 43 Short review. 44 ------------- 45 46 This file consists of two interrelated parts: 47 48 1. queueing disciplines manager frontend. 49 2. traffic classes manager frontend. 50 51 Generally, queueing discipline ("qdisc") is a black box, 52 which is able to enqueue packets and to dequeue them (when 53 device is ready to send something) in order and at times 54 determined by algorithm hidden in it. 55 56 qdisc's are divided to two categories: 57 - "queues", which have no internal structure visible from outside. 58 - "schedulers", which split all the packets to "traffic classes", 59 using "packet classifiers" (look at cls_api.c) 60 61 In turn, classes may have child qdiscs (as rule, queues) 62 attached to them etc. etc. etc. 63 64 The goal of the routines in this file is to translate 65 information supplied by user in the form of handles 66 to more intelligible for kernel form, to make some sanity 67 checks and part of work, which is common to all qdiscs 68 and to provide rtnetlink notifications. 69 70 All real intelligent work is done inside qdisc modules. 71 72 73 74 Every discipline has two major routines: enqueue and dequeue. 75 76 ---dequeue 77 78 dequeue usually returns a skb to send. It is allowed to return NULL, 79 but it does not mean that queue is empty, it just means that 80 discipline does not want to send anything this time. 81 Queue is really empty if q->q.qlen == 0. 82 For complicated disciplines with multiple queues q->q is not 83 real packet queue, but however q->q.qlen must be valid. 84 85 ---enqueue 86 87 enqueue returns 0, if packet was enqueued successfully. 88 If packet (this one or another one) was dropped, it returns 89 not zero error code. 90 NET_XMIT_DROP - this packet dropped 91 Expected action: do not backoff, but wait until queue will clear. 92 NET_XMIT_CN - probably this packet enqueued, but another one dropped. 93 Expected action: backoff or ignore 94 NET_XMIT_POLICED - dropped by police. 95 Expected action: backoff or error to real-time apps. 96 97 Auxiliary routines: 98 99 ---requeue 100 101 requeues once dequeued packet. It is used for non-standard or 102 just buggy devices, which can defer output even if dev->tbusy=0. 103 104 ---reset 105 106 returns qdisc to initial state: purge all buffers, clear all 107 timers, counters (except for statistics) etc. 108 109 ---init 110 111 initializes newly created qdisc. 112 113 ---destroy 114 115 destroys resources allocated by init and during lifetime of qdisc. 116 117 ---change 118 119 changes qdisc parameters. 120 */ 121 122 /* Protects list of registered TC modules. It is pure SMP lock. */ 123 static DEFINE_RWLOCK(qdisc_mod_lock); 124 125 126 /************************************************ 127 * Queueing disciplines manipulation. * 128 ************************************************/ 129 130 131 /* The list of all installed queueing disciplines. */ 132 133 static struct Qdisc_ops *qdisc_base; 134 135 /* Register/uregister queueing discipline */ 136 137 int register_qdisc(struct Qdisc_ops *qops) 138 { 139 struct Qdisc_ops *q, **qp; 140 int rc = -EEXIST; 141 142 write_lock(&qdisc_mod_lock); 143 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) 144 if (!strcmp(qops->id, q->id)) 145 goto out; 146 147 if (qops->enqueue == NULL) 148 qops->enqueue = noop_qdisc_ops.enqueue; 149 if (qops->requeue == NULL) 150 qops->requeue = noop_qdisc_ops.requeue; 151 if (qops->dequeue == NULL) 152 qops->dequeue = noop_qdisc_ops.dequeue; 153 154 qops->next = NULL; 155 *qp = qops; 156 rc = 0; 157 out: 158 write_unlock(&qdisc_mod_lock); 159 return rc; 160 } 161 EXPORT_SYMBOL(register_qdisc); 162 163 int unregister_qdisc(struct Qdisc_ops *qops) 164 { 165 struct Qdisc_ops *q, **qp; 166 int err = -ENOENT; 167 168 write_lock(&qdisc_mod_lock); 169 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) 170 if (q == qops) 171 break; 172 if (q) { 173 *qp = q->next; 174 q->next = NULL; 175 err = 0; 176 } 177 write_unlock(&qdisc_mod_lock); 178 return err; 179 } 180 EXPORT_SYMBOL(unregister_qdisc); 181 182 /* We know handle. Find qdisc among all qdisc's attached to device 183 (root qdisc, all its children, children of children etc.) 184 */ 185 186 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) 187 { 188 struct Qdisc *q; 189 190 list_for_each_entry(q, &dev->qdisc_list, list) { 191 if (q->handle == handle) 192 return q; 193 } 194 return NULL; 195 } 196 197 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) 198 { 199 unsigned long cl; 200 struct Qdisc *leaf; 201 const struct Qdisc_class_ops *cops = p->ops->cl_ops; 202 203 if (cops == NULL) 204 return NULL; 205 cl = cops->get(p, classid); 206 207 if (cl == 0) 208 return NULL; 209 leaf = cops->leaf(p, cl); 210 cops->put(p, cl); 211 return leaf; 212 } 213 214 /* Find queueing discipline by name */ 215 216 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) 217 { 218 struct Qdisc_ops *q = NULL; 219 220 if (kind) { 221 read_lock(&qdisc_mod_lock); 222 for (q = qdisc_base; q; q = q->next) { 223 if (nla_strcmp(kind, q->id) == 0) { 224 if (!try_module_get(q->owner)) 225 q = NULL; 226 break; 227 } 228 } 229 read_unlock(&qdisc_mod_lock); 230 } 231 return q; 232 } 233 234 static struct qdisc_rate_table *qdisc_rtab_list; 235 236 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) 237 { 238 struct qdisc_rate_table *rtab; 239 240 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { 241 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { 242 rtab->refcnt++; 243 return rtab; 244 } 245 } 246 247 if (tab == NULL || r->rate == 0 || r->cell_log == 0 || 248 nla_len(tab) != TC_RTAB_SIZE) 249 return NULL; 250 251 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); 252 if (rtab) { 253 rtab->rate = *r; 254 rtab->refcnt = 1; 255 memcpy(rtab->data, nla_data(tab), 1024); 256 rtab->next = qdisc_rtab_list; 257 qdisc_rtab_list = rtab; 258 } 259 return rtab; 260 } 261 EXPORT_SYMBOL(qdisc_get_rtab); 262 263 void qdisc_put_rtab(struct qdisc_rate_table *tab) 264 { 265 struct qdisc_rate_table *rtab, **rtabp; 266 267 if (!tab || --tab->refcnt) 268 return; 269 270 for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { 271 if (rtab == tab) { 272 *rtabp = rtab->next; 273 kfree(rtab); 274 return; 275 } 276 } 277 } 278 EXPORT_SYMBOL(qdisc_put_rtab); 279 280 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) 281 { 282 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, 283 timer); 284 struct net_device *dev = wd->qdisc->dev; 285 286 wd->qdisc->flags &= ~TCQ_F_THROTTLED; 287 smp_wmb(); 288 netif_schedule(dev); 289 290 return HRTIMER_NORESTART; 291 } 292 293 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) 294 { 295 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); 296 wd->timer.function = qdisc_watchdog; 297 wd->qdisc = qdisc; 298 } 299 EXPORT_SYMBOL(qdisc_watchdog_init); 300 301 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) 302 { 303 ktime_t time; 304 305 wd->qdisc->flags |= TCQ_F_THROTTLED; 306 time = ktime_set(0, 0); 307 time = ktime_add_ns(time, PSCHED_US2NS(expires)); 308 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); 309 } 310 EXPORT_SYMBOL(qdisc_watchdog_schedule); 311 312 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) 313 { 314 hrtimer_cancel(&wd->timer); 315 wd->qdisc->flags &= ~TCQ_F_THROTTLED; 316 } 317 EXPORT_SYMBOL(qdisc_watchdog_cancel); 318 319 /* Allocate an unique handle from space managed by kernel */ 320 321 static u32 qdisc_alloc_handle(struct net_device *dev) 322 { 323 int i = 0x10000; 324 static u32 autohandle = TC_H_MAKE(0x80000000U, 0); 325 326 do { 327 autohandle += TC_H_MAKE(0x10000U, 0); 328 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) 329 autohandle = TC_H_MAKE(0x80000000U, 0); 330 } while (qdisc_lookup(dev, autohandle) && --i > 0); 331 332 return i>0 ? autohandle : 0; 333 } 334 335 /* Attach toplevel qdisc to device dev */ 336 337 static struct Qdisc * 338 dev_graft_qdisc(struct net_device *dev, struct Qdisc *qdisc) 339 { 340 struct Qdisc *oqdisc; 341 342 if (dev->flags & IFF_UP) 343 dev_deactivate(dev); 344 345 qdisc_lock_tree(dev); 346 if (qdisc && qdisc->flags&TCQ_F_INGRESS) { 347 oqdisc = dev->qdisc_ingress; 348 /* Prune old scheduler */ 349 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) { 350 /* delete */ 351 qdisc_reset(oqdisc); 352 dev->qdisc_ingress = NULL; 353 } else { /* new */ 354 dev->qdisc_ingress = qdisc; 355 } 356 357 } else { 358 359 oqdisc = dev->qdisc_sleeping; 360 361 /* Prune old scheduler */ 362 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) 363 qdisc_reset(oqdisc); 364 365 /* ... and graft new one */ 366 if (qdisc == NULL) 367 qdisc = &noop_qdisc; 368 dev->qdisc_sleeping = qdisc; 369 dev->qdisc = &noop_qdisc; 370 } 371 372 qdisc_unlock_tree(dev); 373 374 if (dev->flags & IFF_UP) 375 dev_activate(dev); 376 377 return oqdisc; 378 } 379 380 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) 381 { 382 const struct Qdisc_class_ops *cops; 383 unsigned long cl; 384 u32 parentid; 385 386 if (n == 0) 387 return; 388 while ((parentid = sch->parent)) { 389 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) 390 return; 391 392 sch = qdisc_lookup(sch->dev, TC_H_MAJ(parentid)); 393 if (sch == NULL) { 394 WARN_ON(parentid != TC_H_ROOT); 395 return; 396 } 397 cops = sch->ops->cl_ops; 398 if (cops->qlen_notify) { 399 cl = cops->get(sch, parentid); 400 cops->qlen_notify(sch, cl); 401 cops->put(sch, cl); 402 } 403 sch->q.qlen -= n; 404 } 405 } 406 EXPORT_SYMBOL(qdisc_tree_decrease_qlen); 407 408 /* Graft qdisc "new" to class "classid" of qdisc "parent" or 409 to device "dev". 410 411 Old qdisc is not destroyed but returned in *old. 412 */ 413 414 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, 415 u32 classid, 416 struct Qdisc *new, struct Qdisc **old) 417 { 418 int err = 0; 419 struct Qdisc *q = *old; 420 421 422 if (parent == NULL) { 423 if (q && q->flags&TCQ_F_INGRESS) { 424 *old = dev_graft_qdisc(dev, q); 425 } else { 426 *old = dev_graft_qdisc(dev, new); 427 } 428 } else { 429 const struct Qdisc_class_ops *cops = parent->ops->cl_ops; 430 431 err = -EINVAL; 432 433 if (cops) { 434 unsigned long cl = cops->get(parent, classid); 435 if (cl) { 436 err = cops->graft(parent, cl, new, old); 437 cops->put(parent, cl); 438 } 439 } 440 } 441 return err; 442 } 443 444 /* 445 Allocate and initialize new qdisc. 446 447 Parameters are passed via opt. 448 */ 449 450 static struct Qdisc * 451 qdisc_create(struct net_device *dev, u32 parent, u32 handle, 452 struct nlattr **tca, int *errp) 453 { 454 int err; 455 struct nlattr *kind = tca[TCA_KIND]; 456 struct Qdisc *sch; 457 struct Qdisc_ops *ops; 458 459 ops = qdisc_lookup_ops(kind); 460 #ifdef CONFIG_KMOD 461 if (ops == NULL && kind != NULL) { 462 char name[IFNAMSIZ]; 463 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { 464 /* We dropped the RTNL semaphore in order to 465 * perform the module load. So, even if we 466 * succeeded in loading the module we have to 467 * tell the caller to replay the request. We 468 * indicate this using -EAGAIN. 469 * We replay the request because the device may 470 * go away in the mean time. 471 */ 472 rtnl_unlock(); 473 request_module("sch_%s", name); 474 rtnl_lock(); 475 ops = qdisc_lookup_ops(kind); 476 if (ops != NULL) { 477 /* We will try again qdisc_lookup_ops, 478 * so don't keep a reference. 479 */ 480 module_put(ops->owner); 481 err = -EAGAIN; 482 goto err_out; 483 } 484 } 485 } 486 #endif 487 488 err = -ENOENT; 489 if (ops == NULL) 490 goto err_out; 491 492 sch = qdisc_alloc(dev, ops); 493 if (IS_ERR(sch)) { 494 err = PTR_ERR(sch); 495 goto err_out2; 496 } 497 498 sch->parent = parent; 499 500 if (handle == TC_H_INGRESS) { 501 sch->flags |= TCQ_F_INGRESS; 502 sch->stats_lock = &dev->ingress_lock; 503 handle = TC_H_MAKE(TC_H_INGRESS, 0); 504 } else { 505 sch->stats_lock = &dev->queue_lock; 506 if (handle == 0) { 507 handle = qdisc_alloc_handle(dev); 508 err = -ENOMEM; 509 if (handle == 0) 510 goto err_out3; 511 } 512 } 513 514 sch->handle = handle; 515 516 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { 517 if (tca[TCA_RATE]) { 518 err = gen_new_estimator(&sch->bstats, &sch->rate_est, 519 sch->stats_lock, 520 tca[TCA_RATE]); 521 if (err) { 522 /* 523 * Any broken qdiscs that would require 524 * a ops->reset() here? The qdisc was never 525 * in action so it shouldn't be necessary. 526 */ 527 if (ops->destroy) 528 ops->destroy(sch); 529 goto err_out3; 530 } 531 } 532 qdisc_lock_tree(dev); 533 list_add_tail(&sch->list, &dev->qdisc_list); 534 qdisc_unlock_tree(dev); 535 536 return sch; 537 } 538 err_out3: 539 dev_put(dev); 540 kfree((char *) sch - sch->padded); 541 err_out2: 542 module_put(ops->owner); 543 err_out: 544 *errp = err; 545 return NULL; 546 } 547 548 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) 549 { 550 if (tca[TCA_OPTIONS]) { 551 int err; 552 553 if (sch->ops->change == NULL) 554 return -EINVAL; 555 err = sch->ops->change(sch, tca[TCA_OPTIONS]); 556 if (err) 557 return err; 558 } 559 if (tca[TCA_RATE]) 560 gen_replace_estimator(&sch->bstats, &sch->rate_est, 561 sch->stats_lock, tca[TCA_RATE]); 562 return 0; 563 } 564 565 struct check_loop_arg 566 { 567 struct qdisc_walker w; 568 struct Qdisc *p; 569 int depth; 570 }; 571 572 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); 573 574 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) 575 { 576 struct check_loop_arg arg; 577 578 if (q->ops->cl_ops == NULL) 579 return 0; 580 581 arg.w.stop = arg.w.skip = arg.w.count = 0; 582 arg.w.fn = check_loop_fn; 583 arg.depth = depth; 584 arg.p = p; 585 q->ops->cl_ops->walk(q, &arg.w); 586 return arg.w.stop ? -ELOOP : 0; 587 } 588 589 static int 590 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) 591 { 592 struct Qdisc *leaf; 593 const struct Qdisc_class_ops *cops = q->ops->cl_ops; 594 struct check_loop_arg *arg = (struct check_loop_arg *)w; 595 596 leaf = cops->leaf(q, cl); 597 if (leaf) { 598 if (leaf == arg->p || arg->depth > 7) 599 return -ELOOP; 600 return check_loop(leaf, arg->p, arg->depth + 1); 601 } 602 return 0; 603 } 604 605 /* 606 * Delete/get qdisc. 607 */ 608 609 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 610 { 611 struct net *net = sock_net(skb->sk); 612 struct tcmsg *tcm = NLMSG_DATA(n); 613 struct nlattr *tca[TCA_MAX + 1]; 614 struct net_device *dev; 615 u32 clid = tcm->tcm_parent; 616 struct Qdisc *q = NULL; 617 struct Qdisc *p = NULL; 618 int err; 619 620 if (net != &init_net) 621 return -EINVAL; 622 623 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) 624 return -ENODEV; 625 626 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 627 if (err < 0) 628 return err; 629 630 if (clid) { 631 if (clid != TC_H_ROOT) { 632 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { 633 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) 634 return -ENOENT; 635 q = qdisc_leaf(p, clid); 636 } else { /* ingress */ 637 q = dev->qdisc_ingress; 638 } 639 } else { 640 q = dev->qdisc_sleeping; 641 } 642 if (!q) 643 return -ENOENT; 644 645 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) 646 return -EINVAL; 647 } else { 648 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) 649 return -ENOENT; 650 } 651 652 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 653 return -EINVAL; 654 655 if (n->nlmsg_type == RTM_DELQDISC) { 656 if (!clid) 657 return -EINVAL; 658 if (q->handle == 0) 659 return -ENOENT; 660 if ((err = qdisc_graft(dev, p, clid, NULL, &q)) != 0) 661 return err; 662 if (q) { 663 qdisc_notify(skb, n, clid, q, NULL); 664 qdisc_lock_tree(dev); 665 qdisc_destroy(q); 666 qdisc_unlock_tree(dev); 667 } 668 } else { 669 qdisc_notify(skb, n, clid, NULL, q); 670 } 671 return 0; 672 } 673 674 /* 675 Create/change qdisc. 676 */ 677 678 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 679 { 680 struct net *net = sock_net(skb->sk); 681 struct tcmsg *tcm; 682 struct nlattr *tca[TCA_MAX + 1]; 683 struct net_device *dev; 684 u32 clid; 685 struct Qdisc *q, *p; 686 int err; 687 688 if (net != &init_net) 689 return -EINVAL; 690 691 replay: 692 /* Reinit, just in case something touches this. */ 693 tcm = NLMSG_DATA(n); 694 clid = tcm->tcm_parent; 695 q = p = NULL; 696 697 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) 698 return -ENODEV; 699 700 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 701 if (err < 0) 702 return err; 703 704 if (clid) { 705 if (clid != TC_H_ROOT) { 706 if (clid != TC_H_INGRESS) { 707 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) 708 return -ENOENT; 709 q = qdisc_leaf(p, clid); 710 } else { /*ingress */ 711 q = dev->qdisc_ingress; 712 } 713 } else { 714 q = dev->qdisc_sleeping; 715 } 716 717 /* It may be default qdisc, ignore it */ 718 if (q && q->handle == 0) 719 q = NULL; 720 721 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { 722 if (tcm->tcm_handle) { 723 if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) 724 return -EEXIST; 725 if (TC_H_MIN(tcm->tcm_handle)) 726 return -EINVAL; 727 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) 728 goto create_n_graft; 729 if (n->nlmsg_flags&NLM_F_EXCL) 730 return -EEXIST; 731 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 732 return -EINVAL; 733 if (q == p || 734 (p && check_loop(q, p, 0))) 735 return -ELOOP; 736 atomic_inc(&q->refcnt); 737 goto graft; 738 } else { 739 if (q == NULL) 740 goto create_n_graft; 741 742 /* This magic test requires explanation. 743 * 744 * We know, that some child q is already 745 * attached to this parent and have choice: 746 * either to change it or to create/graft new one. 747 * 748 * 1. We are allowed to create/graft only 749 * if CREATE and REPLACE flags are set. 750 * 751 * 2. If EXCL is set, requestor wanted to say, 752 * that qdisc tcm_handle is not expected 753 * to exist, so that we choose create/graft too. 754 * 755 * 3. The last case is when no flags are set. 756 * Alas, it is sort of hole in API, we 757 * cannot decide what to do unambiguously. 758 * For now we select create/graft, if 759 * user gave KIND, which does not match existing. 760 */ 761 if ((n->nlmsg_flags&NLM_F_CREATE) && 762 (n->nlmsg_flags&NLM_F_REPLACE) && 763 ((n->nlmsg_flags&NLM_F_EXCL) || 764 (tca[TCA_KIND] && 765 nla_strcmp(tca[TCA_KIND], q->ops->id)))) 766 goto create_n_graft; 767 } 768 } 769 } else { 770 if (!tcm->tcm_handle) 771 return -EINVAL; 772 q = qdisc_lookup(dev, tcm->tcm_handle); 773 } 774 775 /* Change qdisc parameters */ 776 if (q == NULL) 777 return -ENOENT; 778 if (n->nlmsg_flags&NLM_F_EXCL) 779 return -EEXIST; 780 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) 781 return -EINVAL; 782 err = qdisc_change(q, tca); 783 if (err == 0) 784 qdisc_notify(skb, n, clid, NULL, q); 785 return err; 786 787 create_n_graft: 788 if (!(n->nlmsg_flags&NLM_F_CREATE)) 789 return -ENOENT; 790 if (clid == TC_H_INGRESS) 791 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_parent, 792 tca, &err); 793 else 794 q = qdisc_create(dev, tcm->tcm_parent, tcm->tcm_handle, 795 tca, &err); 796 if (q == NULL) { 797 if (err == -EAGAIN) 798 goto replay; 799 return err; 800 } 801 802 graft: 803 if (1) { 804 struct Qdisc *old_q = NULL; 805 err = qdisc_graft(dev, p, clid, q, &old_q); 806 if (err) { 807 if (q) { 808 qdisc_lock_tree(dev); 809 qdisc_destroy(q); 810 qdisc_unlock_tree(dev); 811 } 812 return err; 813 } 814 qdisc_notify(skb, n, clid, old_q, q); 815 if (old_q) { 816 qdisc_lock_tree(dev); 817 qdisc_destroy(old_q); 818 qdisc_unlock_tree(dev); 819 } 820 } 821 return 0; 822 } 823 824 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, 825 u32 pid, u32 seq, u16 flags, int event) 826 { 827 struct tcmsg *tcm; 828 struct nlmsghdr *nlh; 829 unsigned char *b = skb_tail_pointer(skb); 830 struct gnet_dump d; 831 832 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 833 tcm = NLMSG_DATA(nlh); 834 tcm->tcm_family = AF_UNSPEC; 835 tcm->tcm__pad1 = 0; 836 tcm->tcm__pad2 = 0; 837 tcm->tcm_ifindex = q->dev->ifindex; 838 tcm->tcm_parent = clid; 839 tcm->tcm_handle = q->handle; 840 tcm->tcm_info = atomic_read(&q->refcnt); 841 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); 842 if (q->ops->dump && q->ops->dump(q, skb) < 0) 843 goto nla_put_failure; 844 q->qstats.qlen = q->q.qlen; 845 846 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, 847 TCA_XSTATS, q->stats_lock, &d) < 0) 848 goto nla_put_failure; 849 850 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) 851 goto nla_put_failure; 852 853 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || 854 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || 855 gnet_stats_copy_queue(&d, &q->qstats) < 0) 856 goto nla_put_failure; 857 858 if (gnet_stats_finish_copy(&d) < 0) 859 goto nla_put_failure; 860 861 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 862 return skb->len; 863 864 nlmsg_failure: 865 nla_put_failure: 866 nlmsg_trim(skb, b); 867 return -1; 868 } 869 870 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, 871 u32 clid, struct Qdisc *old, struct Qdisc *new) 872 { 873 struct sk_buff *skb; 874 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 875 876 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 877 if (!skb) 878 return -ENOBUFS; 879 880 if (old && old->handle) { 881 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) 882 goto err_out; 883 } 884 if (new) { 885 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) 886 goto err_out; 887 } 888 889 if (skb->len) 890 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 891 892 err_out: 893 kfree_skb(skb); 894 return -EINVAL; 895 } 896 897 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) 898 { 899 struct net *net = sock_net(skb->sk); 900 int idx, q_idx; 901 int s_idx, s_q_idx; 902 struct net_device *dev; 903 struct Qdisc *q; 904 905 if (net != &init_net) 906 return 0; 907 908 s_idx = cb->args[0]; 909 s_q_idx = q_idx = cb->args[1]; 910 read_lock(&dev_base_lock); 911 idx = 0; 912 for_each_netdev(&init_net, dev) { 913 if (idx < s_idx) 914 goto cont; 915 if (idx > s_idx) 916 s_q_idx = 0; 917 q_idx = 0; 918 list_for_each_entry(q, &dev->qdisc_list, list) { 919 if (q_idx < s_q_idx) { 920 q_idx++; 921 continue; 922 } 923 if (tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, 924 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) 925 goto done; 926 q_idx++; 927 } 928 cont: 929 idx++; 930 } 931 932 done: 933 read_unlock(&dev_base_lock); 934 935 cb->args[0] = idx; 936 cb->args[1] = q_idx; 937 938 return skb->len; 939 } 940 941 942 943 /************************************************ 944 * Traffic classes manipulation. * 945 ************************************************/ 946 947 948 949 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) 950 { 951 struct net *net = sock_net(skb->sk); 952 struct tcmsg *tcm = NLMSG_DATA(n); 953 struct nlattr *tca[TCA_MAX + 1]; 954 struct net_device *dev; 955 struct Qdisc *q = NULL; 956 const struct Qdisc_class_ops *cops; 957 unsigned long cl = 0; 958 unsigned long new_cl; 959 u32 pid = tcm->tcm_parent; 960 u32 clid = tcm->tcm_handle; 961 u32 qid = TC_H_MAJ(clid); 962 int err; 963 964 if (net != &init_net) 965 return -EINVAL; 966 967 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) 968 return -ENODEV; 969 970 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); 971 if (err < 0) 972 return err; 973 974 /* 975 parent == TC_H_UNSPEC - unspecified parent. 976 parent == TC_H_ROOT - class is root, which has no parent. 977 parent == X:0 - parent is root class. 978 parent == X:Y - parent is a node in hierarchy. 979 parent == 0:Y - parent is X:Y, where X:0 is qdisc. 980 981 handle == 0:0 - generate handle from kernel pool. 982 handle == 0:Y - class is X:Y, where X:0 is qdisc. 983 handle == X:Y - clear. 984 handle == X:0 - root class. 985 */ 986 987 /* Step 1. Determine qdisc handle X:0 */ 988 989 if (pid != TC_H_ROOT) { 990 u32 qid1 = TC_H_MAJ(pid); 991 992 if (qid && qid1) { 993 /* If both majors are known, they must be identical. */ 994 if (qid != qid1) 995 return -EINVAL; 996 } else if (qid1) { 997 qid = qid1; 998 } else if (qid == 0) 999 qid = dev->qdisc_sleeping->handle; 1000 1001 /* Now qid is genuine qdisc handle consistent 1002 both with parent and child. 1003 1004 TC_H_MAJ(pid) still may be unspecified, complete it now. 1005 */ 1006 if (pid) 1007 pid = TC_H_MAKE(qid, pid); 1008 } else { 1009 if (qid == 0) 1010 qid = dev->qdisc_sleeping->handle; 1011 } 1012 1013 /* OK. Locate qdisc */ 1014 if ((q = qdisc_lookup(dev, qid)) == NULL) 1015 return -ENOENT; 1016 1017 /* An check that it supports classes */ 1018 cops = q->ops->cl_ops; 1019 if (cops == NULL) 1020 return -EINVAL; 1021 1022 /* Now try to get class */ 1023 if (clid == 0) { 1024 if (pid == TC_H_ROOT) 1025 clid = qid; 1026 } else 1027 clid = TC_H_MAKE(qid, clid); 1028 1029 if (clid) 1030 cl = cops->get(q, clid); 1031 1032 if (cl == 0) { 1033 err = -ENOENT; 1034 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) 1035 goto out; 1036 } else { 1037 switch (n->nlmsg_type) { 1038 case RTM_NEWTCLASS: 1039 err = -EEXIST; 1040 if (n->nlmsg_flags&NLM_F_EXCL) 1041 goto out; 1042 break; 1043 case RTM_DELTCLASS: 1044 err = cops->delete(q, cl); 1045 if (err == 0) 1046 tclass_notify(skb, n, q, cl, RTM_DELTCLASS); 1047 goto out; 1048 case RTM_GETTCLASS: 1049 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); 1050 goto out; 1051 default: 1052 err = -EINVAL; 1053 goto out; 1054 } 1055 } 1056 1057 new_cl = cl; 1058 err = cops->change(q, clid, pid, tca, &new_cl); 1059 if (err == 0) 1060 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); 1061 1062 out: 1063 if (cl) 1064 cops->put(q, cl); 1065 1066 return err; 1067 } 1068 1069 1070 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, 1071 unsigned long cl, 1072 u32 pid, u32 seq, u16 flags, int event) 1073 { 1074 struct tcmsg *tcm; 1075 struct nlmsghdr *nlh; 1076 unsigned char *b = skb_tail_pointer(skb); 1077 struct gnet_dump d; 1078 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; 1079 1080 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); 1081 tcm = NLMSG_DATA(nlh); 1082 tcm->tcm_family = AF_UNSPEC; 1083 tcm->tcm_ifindex = q->dev->ifindex; 1084 tcm->tcm_parent = q->handle; 1085 tcm->tcm_handle = q->handle; 1086 tcm->tcm_info = 0; 1087 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); 1088 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) 1089 goto nla_put_failure; 1090 1091 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, 1092 TCA_XSTATS, q->stats_lock, &d) < 0) 1093 goto nla_put_failure; 1094 1095 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) 1096 goto nla_put_failure; 1097 1098 if (gnet_stats_finish_copy(&d) < 0) 1099 goto nla_put_failure; 1100 1101 nlh->nlmsg_len = skb_tail_pointer(skb) - b; 1102 return skb->len; 1103 1104 nlmsg_failure: 1105 nla_put_failure: 1106 nlmsg_trim(skb, b); 1107 return -1; 1108 } 1109 1110 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, 1111 struct Qdisc *q, unsigned long cl, int event) 1112 { 1113 struct sk_buff *skb; 1114 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; 1115 1116 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1117 if (!skb) 1118 return -ENOBUFS; 1119 1120 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { 1121 kfree_skb(skb); 1122 return -EINVAL; 1123 } 1124 1125 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); 1126 } 1127 1128 struct qdisc_dump_args 1129 { 1130 struct qdisc_walker w; 1131 struct sk_buff *skb; 1132 struct netlink_callback *cb; 1133 }; 1134 1135 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) 1136 { 1137 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; 1138 1139 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, 1140 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); 1141 } 1142 1143 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) 1144 { 1145 struct net *net = sock_net(skb->sk); 1146 int t; 1147 int s_t; 1148 struct net_device *dev; 1149 struct Qdisc *q; 1150 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); 1151 struct qdisc_dump_args arg; 1152 1153 if (net != &init_net) 1154 return 0; 1155 1156 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) 1157 return 0; 1158 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) 1159 return 0; 1160 1161 s_t = cb->args[0]; 1162 t = 0; 1163 1164 list_for_each_entry(q, &dev->qdisc_list, list) { 1165 if (t < s_t || !q->ops->cl_ops || 1166 (tcm->tcm_parent && 1167 TC_H_MAJ(tcm->tcm_parent) != q->handle)) { 1168 t++; 1169 continue; 1170 } 1171 if (t > s_t) 1172 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); 1173 arg.w.fn = qdisc_class_dump; 1174 arg.skb = skb; 1175 arg.cb = cb; 1176 arg.w.stop = 0; 1177 arg.w.skip = cb->args[1]; 1178 arg.w.count = 0; 1179 q->ops->cl_ops->walk(q, &arg.w); 1180 cb->args[1] = arg.w.count; 1181 if (arg.w.stop) 1182 break; 1183 t++; 1184 } 1185 1186 cb->args[0] = t; 1187 1188 dev_put(dev); 1189 return skb->len; 1190 } 1191 1192 /* Main classifier routine: scans classifier chain attached 1193 to this qdisc, (optionally) tests for protocol and asks 1194 specific classifiers. 1195 */ 1196 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, 1197 struct tcf_result *res) 1198 { 1199 __be16 protocol = skb->protocol; 1200 int err = 0; 1201 1202 for (; tp; tp = tp->next) { 1203 if ((tp->protocol == protocol || 1204 tp->protocol == htons(ETH_P_ALL)) && 1205 (err = tp->classify(skb, tp, res)) >= 0) { 1206 #ifdef CONFIG_NET_CLS_ACT 1207 if (err != TC_ACT_RECLASSIFY && skb->tc_verd) 1208 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); 1209 #endif 1210 return err; 1211 } 1212 } 1213 return -1; 1214 } 1215 EXPORT_SYMBOL(tc_classify_compat); 1216 1217 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, 1218 struct tcf_result *res) 1219 { 1220 int err = 0; 1221 __be16 protocol; 1222 #ifdef CONFIG_NET_CLS_ACT 1223 struct tcf_proto *otp = tp; 1224 reclassify: 1225 #endif 1226 protocol = skb->protocol; 1227 1228 err = tc_classify_compat(skb, tp, res); 1229 #ifdef CONFIG_NET_CLS_ACT 1230 if (err == TC_ACT_RECLASSIFY) { 1231 u32 verd = G_TC_VERD(skb->tc_verd); 1232 tp = otp; 1233 1234 if (verd++ >= MAX_REC_LOOP) { 1235 printk("rule prio %u protocol %02x reclassify loop, " 1236 "packet dropped\n", 1237 tp->prio&0xffff, ntohs(tp->protocol)); 1238 return TC_ACT_SHOT; 1239 } 1240 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); 1241 goto reclassify; 1242 } 1243 #endif 1244 return err; 1245 } 1246 EXPORT_SYMBOL(tc_classify); 1247 1248 void tcf_destroy(struct tcf_proto *tp) 1249 { 1250 tp->ops->destroy(tp); 1251 module_put(tp->ops->owner); 1252 kfree(tp); 1253 } 1254 1255 void tcf_destroy_chain(struct tcf_proto *fl) 1256 { 1257 struct tcf_proto *tp; 1258 1259 while ((tp = fl) != NULL) { 1260 fl = tp->next; 1261 tcf_destroy(tp); 1262 } 1263 } 1264 EXPORT_SYMBOL(tcf_destroy_chain); 1265 1266 #ifdef CONFIG_PROC_FS 1267 static int psched_show(struct seq_file *seq, void *v) 1268 { 1269 struct timespec ts; 1270 1271 hrtimer_get_res(CLOCK_MONOTONIC, &ts); 1272 seq_printf(seq, "%08x %08x %08x %08x\n", 1273 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), 1274 1000000, 1275 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); 1276 1277 return 0; 1278 } 1279 1280 static int psched_open(struct inode *inode, struct file *file) 1281 { 1282 return single_open(file, psched_show, PDE(inode)->data); 1283 } 1284 1285 static const struct file_operations psched_fops = { 1286 .owner = THIS_MODULE, 1287 .open = psched_open, 1288 .read = seq_read, 1289 .llseek = seq_lseek, 1290 .release = single_release, 1291 }; 1292 #endif 1293 1294 static int __init pktsched_init(void) 1295 { 1296 register_qdisc(&pfifo_qdisc_ops); 1297 register_qdisc(&bfifo_qdisc_ops); 1298 proc_net_fops_create(&init_net, "psched", 0, &psched_fops); 1299 1300 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); 1301 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); 1302 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); 1303 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); 1304 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); 1305 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); 1306 1307 return 0; 1308 } 1309 1310 subsys_initcall(pktsched_init); 1311