| 1 | |
| 2 | |
| 3 | * net/sched/sch_api.c Packet scheduler API.
|
| 4 | *
|
| 5 | * Authors: Alexey Kuznetsov, <[email protected]>
|
| 6 | *
|
| 7 | * Fixes:
|
| 8 | *
|
| 9 | * Rani Assaf <[email protected]> :980802: JIFFIES and CPU clock sources are repaired.
|
| 10 | * Eduardo J. Blanco <[email protected]> :990222: kmod support
|
| 11 | * Jamal Hadi Salim <[email protected]>: 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 | #include <linux/bpf.h>
|
| 29 |
|
| 30 | #include <net/netdev_lock.h>
|
| 31 | #include <net/net_namespace.h>
|
| 32 | #include <net/sock.h>
|
| 33 | #include <net/netlink.h>
|
| 34 | #include <net/pkt_sched.h>
|
| 35 | #include <net/pkt_cls.h>
|
| 36 | #include <net/tc_wrapper.h>
|
| 37 |
|
| 38 | #include <trace/events/qdisc.h>
|
| 39 |
|
| 40 | |
| 41 |
|
| 42 | Short review.
|
| 43 | -------------
|
| 44 |
|
| 45 | This file consists of two interrelated parts:
|
| 46 |
|
| 47 | 1. queueing disciplines manager frontend.
|
| 48 | 2. traffic classes manager frontend.
|
| 49 |
|
| 50 | Generally, queueing discipline ("qdisc") is a black box,
|
| 51 | which is able to enqueue packets and to dequeue them (when
|
| 52 | device is ready to send something) in order and at times
|
| 53 | determined by algorithm hidden in it.
|
| 54 |
|
| 55 | qdisc's are divided to two categories:
|
| 56 | - "queues", which have no internal structure visible from outside.
|
| 57 | - "schedulers", which split all the packets to "traffic classes",
|
| 58 | using "packet classifiers" (look at cls_api.c)
|
| 59 |
|
| 60 | In turn, classes may have child qdiscs (as rule, queues)
|
| 61 | attached to them etc. etc. etc.
|
| 62 |
|
| 63 | The goal of the routines in this file is to translate
|
| 64 | information supplied by user in the form of handles
|
| 65 | to more intelligible for kernel form, to make some sanity
|
| 66 | checks and part of work, which is common to all qdiscs
|
| 67 | and to provide rtnetlink notifications.
|
| 68 |
|
| 69 | All real intelligent work is done inside qdisc modules.
|
| 70 |
|
| 71 |
|
| 72 |
|
| 73 | Every discipline has two major routines: enqueue and dequeue.
|
| 74 |
|
| 75 | ---dequeue
|
| 76 |
|
| 77 | dequeue usually returns a skb to send. It is allowed to return NULL,
|
| 78 | but it does not mean that queue is empty, it just means that
|
| 79 | discipline does not want to send anything this time.
|
| 80 | Queue is really empty if q->q.qlen == 0.
|
| 81 | For complicated disciplines with multiple queues q->q is not
|
| 82 | real packet queue, but however q->q.qlen must be valid.
|
| 83 |
|
| 84 | ---enqueue
|
| 85 |
|
| 86 | enqueue returns 0, if packet was enqueued successfully.
|
| 87 | If packet (this one or another one) was dropped, it returns
|
| 88 | not zero error code.
|
| 89 | NET_XMIT_DROP - this packet dropped
|
| 90 | Expected action: do not backoff, but wait until queue will clear.
|
| 91 | NET_XMIT_CN - probably this packet enqueued, but another one dropped.
|
| 92 | Expected action: backoff or ignore
|
| 93 |
|
| 94 | Auxiliary routines:
|
| 95 |
|
| 96 | ---peek
|
| 97 |
|
| 98 | like dequeue but without removing a packet from the queue
|
| 99 |
|
| 100 | ---reset
|
| 101 |
|
| 102 | returns qdisc to initial state: purge all buffers, clear all
|
| 103 | timers, counters (except for statistics) etc.
|
| 104 |
|
| 105 | ---init
|
| 106 |
|
| 107 | initializes newly created qdisc.
|
| 108 |
|
| 109 | ---destroy
|
| 110 |
|
| 111 | destroys resources allocated by init and during lifetime of qdisc.
|
| 112 |
|
| 113 | ---change
|
| 114 |
|
| 115 | changes qdisc parameters.
|
| 116 |
|
| 117 |
|
| 118 |
|
| 119 | static DEFINE_RWLOCK(qdisc_mod_lock);
|
| 120 |
|
| 121 |
|
| 122 | |
| 123 | * Queueing disciplines manipulation. *
|
| 124 | ***********************************************
|
| 125 |
|
| 126 |
|
| 127 |
|
| 128 |
|
| 129 | static struct Qdisc_ops *qdisc_base;
|
| 130 |
|
| 131 |
|
| 132 |
|
| 133 | int register_qdisc(struct Qdisc_ops *qops)
|
| 134 | {
|
| 135 | struct Qdisc_ops *q, **qp;
|
| 136 | int rc = -EEXIST;
|
| 137 |
|
| 138 | write_lock(&qdisc_mod_lock);
|
| 139 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
|
| 140 | if (!strcmp(qops->id, q->id))
|
| 141 | goto out;
|
| 142 |
|
| 143 | if (qops->enqueue == NULL)
|
| 144 | qops->enqueue = noop_qdisc_ops.enqueue;
|
| 145 | if (qops->peek == NULL) {
|
| 146 | if (qops->dequeue == NULL)
|
| 147 | qops->peek = noop_qdisc_ops.peek;
|
| 148 | else
|
| 149 | goto out_einval;
|
| 150 | }
|
| 151 | if (qops->dequeue == NULL)
|
| 152 | qops->dequeue = noop_qdisc_ops.dequeue;
|
| 153 |
|
| 154 | if (qops->cl_ops) {
|
| 155 | const struct Qdisc_class_ops *cops = qops->cl_ops;
|
| 156 |
|
| 157 | if (!(cops->find && cops->walk && cops->leaf))
|
| 158 | goto out_einval;
|
| 159 |
|
| 160 | if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
|
| 161 | goto out_einval;
|
| 162 | }
|
| 163 |
|
| 164 | qops->next = NULL;
|
| 165 | *qp = qops;
|
| 166 | rc = 0;
|
| 167 | out:
|
| 168 | write_unlock(&qdisc_mod_lock);
|
| 169 | return rc;
|
| 170 |
|
| 171 | out_einval:
|
| 172 | rc = -EINVAL;
|
| 173 | goto out;
|
| 174 | }
|
| 175 | EXPORT_SYMBOL(register_qdisc);
|
| 176 |
|
| 177 | void unregister_qdisc(struct Qdisc_ops *qops)
|
| 178 | {
|
| 179 | struct Qdisc_ops *q, **qp;
|
| 180 | int err = -ENOENT;
|
| 181 |
|
| 182 | write_lock(&qdisc_mod_lock);
|
| 183 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
|
| 184 | if (q == qops)
|
| 185 | break;
|
| 186 | if (q) {
|
| 187 | *qp = q->next;
|
| 188 | q->next = NULL;
|
| 189 | err = 0;
|
| 190 | }
|
| 191 | write_unlock(&qdisc_mod_lock);
|
| 192 |
|
| 193 | WARN(err, "unregister qdisc(%s) failed\n", qops->id);
|
| 194 | }
|
| 195 | EXPORT_SYMBOL(unregister_qdisc);
|
| 196 |
|
| 197 |
|
| 198 | void qdisc_get_default(char *name, size_t len)
|
| 199 | {
|
| 200 | read_lock(&qdisc_mod_lock);
|
| 201 | strscpy(name, default_qdisc_ops->id, len);
|
| 202 | read_unlock(&qdisc_mod_lock);
|
| 203 | }
|
| 204 |
|
| 205 | static struct Qdisc_ops *qdisc_lookup_default(const char *name)
|
| 206 | {
|
| 207 | struct Qdisc_ops *q = NULL;
|
| 208 |
|
| 209 | for (q = qdisc_base; q; q = q->next) {
|
| 210 | if (!strcmp(name, q->id)) {
|
| 211 | if (!bpf_try_module_get(q, q->owner))
|
| 212 | q = NULL;
|
| 213 | break;
|
| 214 | }
|
| 215 | }
|
| 216 |
|
| 217 | return q;
|
| 218 | }
|
| 219 |
|
| 220 |
|
| 221 | int qdisc_set_default(const char *name)
|
| 222 | {
|
| 223 | const struct Qdisc_ops *ops;
|
| 224 |
|
| 225 | if (!capable(CAP_NET_ADMIN))
|
| 226 | return -EPERM;
|
| 227 |
|
| 228 | write_lock(&qdisc_mod_lock);
|
| 229 | ops = qdisc_lookup_default(name);
|
| 230 | if (!ops) {
|
| 231 |
|
| 232 | write_unlock(&qdisc_mod_lock);
|
| 233 | request_module(NET_SCH_ALIAS_PREFIX "%s", name);
|
| 234 | write_lock(&qdisc_mod_lock);
|
| 235 |
|
| 236 | ops = qdisc_lookup_default(name);
|
| 237 | }
|
| 238 |
|
| 239 | if (ops) {
|
| 240 |
|
| 241 | bpf_module_put(default_qdisc_ops, default_qdisc_ops->owner);
|
| 242 | default_qdisc_ops = ops;
|
| 243 | }
|
| 244 | write_unlock(&qdisc_mod_lock);
|
| 245 |
|
| 246 | return ops ? 0 : -ENOENT;
|
| 247 | }
|
| 248 |
|
| 249 | #ifdef CONFIG_NET_SCH_DEFAULT
|
| 250 |
|
| 251 | static int __init sch_default_qdisc(void)
|
| 252 | {
|
| 253 | return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
|
| 254 | }
|
| 255 | late_initcall(sch_default_qdisc);
|
| 256 | #endif
|
| 257 |
|
| 258 | |
| 259 | * (root qdisc, all its children, children of children etc.)
|
| 260 | * Note: caller either uses rtnl or rcu_read_lock()
|
| 261 |
|
| 262 |
|
| 263 | static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
|
| 264 | {
|
| 265 | struct Qdisc *q;
|
| 266 |
|
| 267 | if (!qdisc_dev(root))
|
| 268 | return (root->handle == handle ? root : NULL);
|
| 269 |
|
| 270 | if (!(root->flags & TCQ_F_BUILTIN) &&
|
| 271 | root->handle == handle)
|
| 272 | return root;
|
| 273 |
|
| 274 | hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
|
| 275 | lockdep_rtnl_is_held()) {
|
| 276 | if (q->handle == handle)
|
| 277 | return q;
|
| 278 | }
|
| 279 | return NULL;
|
| 280 | }
|
| 281 |
|
| 282 | void qdisc_hash_add(struct Qdisc *q, bool invisible)
|
| 283 | {
|
| 284 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
|
| 285 | ASSERT_RTNL();
|
| 286 | hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
|
| 287 | if (invisible)
|
| 288 | q->flags |= TCQ_F_INVISIBLE;
|
| 289 | }
|
| 290 | }
|
| 291 | EXPORT_SYMBOL(qdisc_hash_add);
|
| 292 |
|
| 293 | void qdisc_hash_del(struct Qdisc *q)
|
| 294 | {
|
| 295 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
|
| 296 | ASSERT_RTNL();
|
| 297 | hash_del_rcu(&q->hash);
|
| 298 | }
|
| 299 | }
|
| 300 | EXPORT_SYMBOL(qdisc_hash_del);
|
| 301 |
|
| 302 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
|
| 303 | {
|
| 304 | struct Qdisc *q;
|
| 305 |
|
| 306 | if (!handle)
|
| 307 | return NULL;
|
| 308 | q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
|
| 309 | if (q)
|
| 310 | goto out;
|
| 311 |
|
| 312 | if (dev_ingress_queue(dev))
|
| 313 | q = qdisc_match_from_root(
|
| 314 | rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping),
|
| 315 | handle);
|
| 316 | out:
|
| 317 | return q;
|
| 318 | }
|
| 319 |
|
| 320 | struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
|
| 321 | {
|
| 322 | struct netdev_queue *nq;
|
| 323 | struct Qdisc *q;
|
| 324 |
|
| 325 | if (!handle)
|
| 326 | return NULL;
|
| 327 | q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
|
| 328 | if (q)
|
| 329 | goto out;
|
| 330 |
|
| 331 | nq = dev_ingress_queue_rcu(dev);
|
| 332 | if (nq)
|
| 333 | q = qdisc_match_from_root(rcu_dereference(nq->qdisc_sleeping),
|
| 334 | handle);
|
| 335 | out:
|
| 336 | return q;
|
| 337 | }
|
| 338 |
|
| 339 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid,
|
| 340 | struct netlink_ext_ack *extack)
|
| 341 | {
|
| 342 | unsigned long cl;
|
| 343 | const struct Qdisc_class_ops *cops = p->ops->cl_ops;
|
| 344 |
|
| 345 | if (cops == NULL) {
|
| 346 | NL_SET_ERR_MSG(extack, "Parent qdisc is not classful");
|
| 347 | return ERR_PTR(-EOPNOTSUPP);
|
| 348 | }
|
| 349 | cl = cops->find(p, classid);
|
| 350 |
|
| 351 | if (cl == 0) {
|
| 352 | NL_SET_ERR_MSG(extack, "Specified class not found");
|
| 353 | return ERR_PTR(-ENOENT);
|
| 354 | }
|
| 355 | return cops->leaf(p, cl);
|
| 356 | }
|
| 357 |
|
| 358 |
|
| 359 |
|
| 360 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
|
| 361 | {
|
| 362 | struct Qdisc_ops *q = NULL;
|
| 363 |
|
| 364 | if (kind) {
|
| 365 | read_lock(&qdisc_mod_lock);
|
| 366 | for (q = qdisc_base; q; q = q->next) {
|
| 367 | if (nla_strcmp(kind, q->id) == 0) {
|
| 368 | if (!bpf_try_module_get(q, q->owner))
|
| 369 | q = NULL;
|
| 370 | break;
|
| 371 | }
|
| 372 | }
|
| 373 | read_unlock(&qdisc_mod_lock);
|
| 374 | }
|
| 375 | return q;
|
| 376 | }
|
| 377 |
|
| 378 | |
| 379 | * versions, and the rate tables lookup systems have been dropped in
|
| 380 | * the kernel. To keep backward compatible with older iproute2 tc
|
| 381 | * utils, we detect the linklayer setting by detecting if the rate
|
| 382 | * table were modified.
|
| 383 | *
|
| 384 | * For linklayer ATM table entries, the rate table will be aligned to
|
| 385 | * 48 bytes, thus some table entries will contain the same value. The
|
| 386 | * mpu (min packet unit) is also encoded into the old rate table, thus
|
| 387 | * starting from the mpu, we find low and high table entries for
|
| 388 | * mapping this cell. If these entries contain the same value, when
|
| 389 | * the rate tables have been modified for linklayer ATM.
|
| 390 | *
|
| 391 | * This is done by rounding mpu to the nearest 48 bytes cell/entry,
|
| 392 | * and then roundup to the next cell, calc the table entry one below,
|
| 393 | * and compare.
|
| 394 |
|
| 395 | static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
|
| 396 | {
|
| 397 | int low = roundup(r->mpu, 48);
|
| 398 | int high = roundup(low+1, 48);
|
| 399 | int cell_low = low >> r->cell_log;
|
| 400 | int cell_high = (high >> r->cell_log) - 1;
|
| 401 |
|
| 402 |
|
| 403 | if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
|
| 404 | pr_debug("TC linklayer: Giving up ATM detection\n");
|
| 405 | return TC_LINKLAYER_ETHERNET;
|
| 406 | }
|
| 407 |
|
| 408 | if ((cell_high > cell_low) && (cell_high < 256)
|
| 409 | && (rtab[cell_low] == rtab[cell_high])) {
|
| 410 | pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
|
| 411 | cell_low, cell_high, rtab[cell_high]);
|
| 412 | return TC_LINKLAYER_ATM;
|
| 413 | }
|
| 414 | return TC_LINKLAYER_ETHERNET;
|
| 415 | }
|
| 416 |
|
| 417 | static struct qdisc_rate_table *qdisc_rtab_list;
|
| 418 | static DEFINE_SPINLOCK(qdisc_rtab_lock);
|
| 419 |
|
| 420 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
|
| 421 | struct nlattr *tab,
|
| 422 | struct netlink_ext_ack *extack)
|
| 423 | {
|
| 424 | struct qdisc_rate_table *rtab, *new_rtab;
|
| 425 |
|
| 426 | if (tab == NULL || r->rate == 0 ||
|
| 427 | r->cell_log == 0 || r->cell_log >= 32 ||
|
| 428 | nla_len(tab) != TC_RTAB_SIZE) {
|
| 429 | NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
|
| 430 | return NULL;
|
| 431 | }
|
| 432 |
|
| 433 | new_rtab = kmalloc_obj(*new_rtab);
|
| 434 |
|
| 435 | spin_lock(&qdisc_rtab_lock);
|
| 436 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
|
| 437 | if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
|
| 438 | !memcmp(&rtab->data, nla_data(tab), TC_RTAB_SIZE)) {
|
| 439 | rtab->refcnt++;
|
| 440 | spin_unlock(&qdisc_rtab_lock);
|
| 441 | kfree(new_rtab);
|
| 442 | return rtab;
|
| 443 | }
|
| 444 | }
|
| 445 |
|
| 446 | rtab = new_rtab;
|
| 447 | if (rtab) {
|
| 448 | rtab->rate = *r;
|
| 449 | rtab->refcnt = 1;
|
| 450 | memcpy(rtab->data, nla_data(tab), TC_RTAB_SIZE);
|
| 451 | if (r->linklayer == TC_LINKLAYER_UNAWARE)
|
| 452 | r->linklayer = __detect_linklayer(r, rtab->data);
|
| 453 | rtab->next = qdisc_rtab_list;
|
| 454 | qdisc_rtab_list = rtab;
|
| 455 | } else {
|
| 456 | NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
|
| 457 | }
|
| 458 | spin_unlock(&qdisc_rtab_lock);
|
| 459 | return rtab;
|
| 460 | }
|
| 461 | EXPORT_SYMBOL(qdisc_get_rtab);
|
| 462 |
|
| 463 | void qdisc_put_rtab(struct qdisc_rate_table *tab)
|
| 464 | {
|
| 465 | struct qdisc_rate_table *rtab, **rtabp;
|
| 466 |
|
| 467 | if (!tab)
|
| 468 | return;
|
| 469 |
|
| 470 | spin_lock(&qdisc_rtab_lock);
|
| 471 | if (--tab->refcnt) {
|
| 472 | spin_unlock(&qdisc_rtab_lock);
|
| 473 | return;
|
| 474 | }
|
| 475 |
|
| 476 | for (rtabp = &qdisc_rtab_list;
|
| 477 | (rtab = *rtabp) != NULL;
|
| 478 | rtabp = &rtab->next) {
|
| 479 | if (rtab == tab) {
|
| 480 | *rtabp = rtab->next;
|
| 481 | break;
|
| 482 | }
|
| 483 | }
|
| 484 | spin_unlock(&qdisc_rtab_lock);
|
| 485 | kfree(tab);
|
| 486 | }
|
| 487 | EXPORT_SYMBOL(qdisc_put_rtab);
|
| 488 |
|
| 489 | static LIST_HEAD(qdisc_stab_list);
|
| 490 |
|
| 491 | static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
|
| 492 | [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
|
| 493 | [TCA_STAB_DATA] = { .type = NLA_BINARY },
|
| 494 | };
|
| 495 |
|
| 496 | static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
|
| 497 | struct netlink_ext_ack *extack)
|
| 498 | {
|
| 499 | struct nlattr *tb[TCA_STAB_MAX + 1];
|
| 500 | struct qdisc_size_table *stab;
|
| 501 | struct tc_sizespec *s;
|
| 502 | unsigned int tsize = 0;
|
| 503 | u16 *tab = NULL;
|
| 504 | int err;
|
| 505 |
|
| 506 | err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
|
| 507 | extack);
|
| 508 | if (err < 0)
|
| 509 | return ERR_PTR(err);
|
| 510 | if (!tb[TCA_STAB_BASE]) {
|
| 511 | NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
|
| 512 | return ERR_PTR(-EINVAL);
|
| 513 | }
|
| 514 |
|
| 515 | s = nla_data(tb[TCA_STAB_BASE]);
|
| 516 |
|
| 517 | if (s->tsize > 0) {
|
| 518 | if (!tb[TCA_STAB_DATA]) {
|
| 519 | NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
|
| 520 | return ERR_PTR(-EINVAL);
|
| 521 | }
|
| 522 | tab = nla_data(tb[TCA_STAB_DATA]);
|
| 523 | tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
|
| 524 | }
|
| 525 |
|
| 526 | if (tsize != s->tsize || (!tab && tsize > 0)) {
|
| 527 | NL_SET_ERR_MSG(extack, "Invalid size of size table");
|
| 528 | return ERR_PTR(-EINVAL);
|
| 529 | }
|
| 530 |
|
| 531 | list_for_each_entry(stab, &qdisc_stab_list, list) {
|
| 532 | if (memcmp(&stab->szopts, s, sizeof(*s)))
|
| 533 | continue;
|
| 534 | if (tsize > 0 &&
|
| 535 | memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
|
| 536 | continue;
|
| 537 | stab->refcnt++;
|
| 538 | return stab;
|
| 539 | }
|
| 540 |
|
| 541 | if (s->size_log > STAB_SIZE_LOG_MAX ||
|
| 542 | s->cell_log > STAB_SIZE_LOG_MAX) {
|
| 543 | NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
|
| 544 | return ERR_PTR(-EINVAL);
|
| 545 | }
|
| 546 |
|
| 547 | stab = kmalloc_flex(*stab, data, tsize);
|
| 548 | if (!stab)
|
| 549 | return ERR_PTR(-ENOMEM);
|
| 550 |
|
| 551 | stab->refcnt = 1;
|
| 552 | stab->szopts = *s;
|
| 553 | if (tsize > 0)
|
| 554 | memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
|
| 555 |
|
| 556 | list_add_tail(&stab->list, &qdisc_stab_list);
|
| 557 |
|
| 558 | return stab;
|
| 559 | }
|
| 560 |
|
| 561 | void qdisc_put_stab(struct qdisc_size_table *tab)
|
| 562 | {
|
| 563 | if (!tab)
|
| 564 | return;
|
| 565 |
|
| 566 | if (--tab->refcnt == 0) {
|
| 567 | list_del(&tab->list);
|
| 568 | kfree_rcu(tab, rcu);
|
| 569 | }
|
| 570 | }
|
| 571 | EXPORT_SYMBOL(qdisc_put_stab);
|
| 572 |
|
| 573 | static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
|
| 574 | {
|
| 575 | struct nlattr *nest;
|
| 576 |
|
| 577 | nest = nla_nest_start_noflag(skb, TCA_STAB);
|
| 578 | if (nest == NULL)
|
| 579 | goto nla_put_failure;
|
| 580 | if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
|
| 581 | goto nla_put_failure;
|
| 582 | nla_nest_end(skb, nest);
|
| 583 |
|
| 584 | return skb->len;
|
| 585 |
|
| 586 | nla_put_failure:
|
| 587 | return -1;
|
| 588 | }
|
| 589 |
|
| 590 | void __qdisc_calculate_pkt_len(struct sk_buff *skb,
|
| 591 | const struct qdisc_size_table *stab)
|
| 592 | {
|
| 593 | int pkt_len, slot;
|
| 594 |
|
| 595 | pkt_len = skb->len + stab->szopts.overhead;
|
| 596 | if (unlikely(!stab->szopts.tsize))
|
| 597 | goto out;
|
| 598 |
|
| 599 | slot = pkt_len + stab->szopts.cell_align;
|
| 600 | if (unlikely(slot < 0))
|
| 601 | slot = 0;
|
| 602 |
|
| 603 | slot >>= stab->szopts.cell_log;
|
| 604 | if (likely(slot < stab->szopts.tsize))
|
| 605 | pkt_len = stab->data[slot];
|
| 606 | else
|
| 607 | pkt_len = stab->data[stab->szopts.tsize - 1] *
|
| 608 | (slot / stab->szopts.tsize) +
|
| 609 | stab->data[slot % stab->szopts.tsize];
|
| 610 |
|
| 611 | pkt_len <<= stab->szopts.size_log;
|
| 612 | out:
|
| 613 | |
| 614 | * (via overhead, the data table, or size_log); cap it so deficit
|
| 615 | * schedulers such as DRR/ETS terminate their refill loops.
|
| 616 |
|
| 617 | pkt_len = clamp_t(int, pkt_len, 1, QDISC_PKT_LEN_MAX);
|
| 618 | qdisc_skb_cb(skb)->pkt_len = pkt_len;
|
| 619 | }
|
| 620 |
|
| 621 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
|
| 622 | {
|
| 623 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
|
| 624 | timer);
|
| 625 |
|
| 626 | rcu_read_lock();
|
| 627 | __netif_schedule(qdisc_root(wd->qdisc));
|
| 628 | rcu_read_unlock();
|
| 629 |
|
| 630 | return HRTIMER_NORESTART;
|
| 631 | }
|
| 632 |
|
| 633 | void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
|
| 634 | clockid_t clockid)
|
| 635 | {
|
| 636 | hrtimer_setup(&wd->timer, qdisc_watchdog, clockid, HRTIMER_MODE_ABS_PINNED);
|
| 637 | wd->qdisc = qdisc;
|
| 638 | }
|
| 639 | EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
|
| 640 |
|
| 641 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
|
| 642 | {
|
| 643 | qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
|
| 644 | }
|
| 645 | EXPORT_SYMBOL(qdisc_watchdog_init);
|
| 646 |
|
| 647 | void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
|
| 648 | u64 delta_ns)
|
| 649 | {
|
| 650 | bool deactivated;
|
| 651 |
|
| 652 | rcu_read_lock();
|
| 653 | deactivated = test_bit(__QDISC_STATE_DEACTIVATED,
|
| 654 | &qdisc_root_sleeping(wd->qdisc)->state);
|
| 655 | rcu_read_unlock();
|
| 656 | if (deactivated)
|
| 657 | return;
|
| 658 |
|
| 659 | if (hrtimer_is_queued(&wd->timer)) {
|
| 660 | u64 softexpires;
|
| 661 |
|
| 662 | softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer));
|
| 663 | |
| 664 | * do not reprogram it.
|
| 665 |
|
| 666 | if (softexpires - expires <= delta_ns)
|
| 667 | return;
|
| 668 | }
|
| 669 |
|
| 670 | hrtimer_start_range_ns(&wd->timer,
|
| 671 | ns_to_ktime(expires),
|
| 672 | delta_ns,
|
| 673 | HRTIMER_MODE_ABS_PINNED);
|
| 674 | }
|
| 675 | EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
|
| 676 |
|
| 677 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
|
| 678 | {
|
| 679 | hrtimer_cancel(&wd->timer);
|
| 680 | }
|
| 681 | EXPORT_SYMBOL(qdisc_watchdog_cancel);
|
| 682 |
|
| 683 | static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
|
| 684 | {
|
| 685 | struct hlist_head *h;
|
| 686 | unsigned int i;
|
| 687 |
|
| 688 | h = kvmalloc_objs(struct hlist_head, n);
|
| 689 |
|
| 690 | if (h != NULL) {
|
| 691 | for (i = 0; i < n; i++)
|
| 692 | INIT_HLIST_HEAD(&h[i]);
|
| 693 | }
|
| 694 | return h;
|
| 695 | }
|
| 696 |
|
| 697 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
|
| 698 | {
|
| 699 | struct Qdisc_class_common *cl;
|
| 700 | struct hlist_node *next;
|
| 701 | struct hlist_head *nhash, *ohash;
|
| 702 | unsigned int nsize, nmask, osize;
|
| 703 | unsigned int i, h;
|
| 704 |
|
| 705 |
|
| 706 | if (clhash->hashelems * 4 <= clhash->hashsize * 3)
|
| 707 | return;
|
| 708 | nsize = clhash->hashsize * 2;
|
| 709 | nmask = nsize - 1;
|
| 710 | nhash = qdisc_class_hash_alloc(nsize);
|
| 711 | if (nhash == NULL)
|
| 712 | return;
|
| 713 |
|
| 714 | ohash = clhash->hash;
|
| 715 | osize = clhash->hashsize;
|
| 716 |
|
| 717 | sch_tree_lock(sch);
|
| 718 | for (i = 0; i < osize; i++) {
|
| 719 | hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
|
| 720 | h = qdisc_class_hash(cl->classid, nmask);
|
| 721 | hlist_add_head(&cl->hnode, &nhash[h]);
|
| 722 | }
|
| 723 | }
|
| 724 | clhash->hash = nhash;
|
| 725 | clhash->hashsize = nsize;
|
| 726 | clhash->hashmask = nmask;
|
| 727 | sch_tree_unlock(sch);
|
| 728 |
|
| 729 | kvfree(ohash);
|
| 730 | }
|
| 731 | EXPORT_SYMBOL(qdisc_class_hash_grow);
|
| 732 |
|
| 733 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
|
| 734 | {
|
| 735 | unsigned int size = 4;
|
| 736 |
|
| 737 | clhash->hash = qdisc_class_hash_alloc(size);
|
| 738 | if (!clhash->hash)
|
| 739 | return -ENOMEM;
|
| 740 | clhash->hashsize = size;
|
| 741 | clhash->hashmask = size - 1;
|
| 742 | clhash->hashelems = 0;
|
| 743 | return 0;
|
| 744 | }
|
| 745 | EXPORT_SYMBOL(qdisc_class_hash_init);
|
| 746 |
|
| 747 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
|
| 748 | {
|
| 749 | kvfree(clhash->hash);
|
| 750 | }
|
| 751 | EXPORT_SYMBOL(qdisc_class_hash_destroy);
|
| 752 |
|
| 753 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
|
| 754 | struct Qdisc_class_common *cl)
|
| 755 | {
|
| 756 | unsigned int h;
|
| 757 |
|
| 758 | INIT_HLIST_NODE(&cl->hnode);
|
| 759 | h = qdisc_class_hash(cl->classid, clhash->hashmask);
|
| 760 | hlist_add_head(&cl->hnode, &clhash->hash[h]);
|
| 761 | clhash->hashelems++;
|
| 762 | }
|
| 763 | EXPORT_SYMBOL(qdisc_class_hash_insert);
|
| 764 |
|
| 765 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
|
| 766 | struct Qdisc_class_common *cl)
|
| 767 | {
|
| 768 | hlist_del(&cl->hnode);
|
| 769 | clhash->hashelems--;
|
| 770 | }
|
| 771 | EXPORT_SYMBOL(qdisc_class_hash_remove);
|
| 772 |
|
| 773 | |
| 774 | * Possible range is [8000-FFFF]:0000 (0x8000 values)
|
| 775 |
|
| 776 | static u32 qdisc_alloc_handle(struct net_device *dev)
|
| 777 | {
|
| 778 | int i = 0x8000;
|
| 779 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
|
| 780 |
|
| 781 | do {
|
| 782 | autohandle += TC_H_MAKE(0x10000U, 0);
|
| 783 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
|
| 784 | autohandle = TC_H_MAKE(0x80000000U, 0);
|
| 785 | if (!qdisc_lookup(dev, autohandle))
|
| 786 | return autohandle;
|
| 787 | cond_resched();
|
| 788 | } while (--i > 0);
|
| 789 |
|
| 790 | return 0;
|
| 791 | }
|
| 792 |
|
| 793 | void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
|
| 794 | {
|
| 795 | const struct Qdisc_class_ops *cops;
|
| 796 | unsigned long cl;
|
| 797 | u32 parentid;
|
| 798 | bool notify;
|
| 799 | int drops;
|
| 800 |
|
| 801 | drops = max_t(int, n, 0);
|
| 802 | rcu_read_lock();
|
| 803 | while ((parentid = sch->parent)) {
|
| 804 | if (parentid == TC_H_ROOT)
|
| 805 | break;
|
| 806 |
|
| 807 | if (sch->flags & TCQ_F_NOPARENT)
|
| 808 | break;
|
| 809 |
|
| 810 | notify = !sch->q.qlen;
|
| 811 |
|
| 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 | |
| 820 | * multiple times.
|
| 821 |
|
| 822 | cl = cops->find(sch, parentid);
|
| 823 | cops->qlen_notify(sch, cl);
|
| 824 | }
|
| 825 | WRITE_ONCE(sch->q.qlen, sch->q.qlen - n);
|
| 826 | qstats_backlog_sub(sch, len);
|
| 827 | __qdisc_qstats_drop(sch, drops);
|
| 828 | }
|
| 829 | rcu_read_unlock();
|
| 830 | }
|
| 831 | EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
|
| 832 |
|
| 833 | int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
|
| 834 | void *type_data)
|
| 835 | {
|
| 836 | struct net_device *dev = qdisc_dev(sch);
|
| 837 | int err;
|
| 838 |
|
| 839 | sch->flags &= ~TCQ_F_OFFLOADED;
|
| 840 | if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
|
| 841 | return 0;
|
| 842 |
|
| 843 | err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
|
| 844 | if (err == -EOPNOTSUPP)
|
| 845 | return 0;
|
| 846 |
|
| 847 | if (!err)
|
| 848 | sch->flags |= TCQ_F_OFFLOADED;
|
| 849 |
|
| 850 | return err;
|
| 851 | }
|
| 852 | EXPORT_SYMBOL(qdisc_offload_dump_helper);
|
| 853 |
|
| 854 | void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
|
| 855 | struct Qdisc *new, struct Qdisc *old,
|
| 856 | enum tc_setup_type type, void *type_data,
|
| 857 | struct netlink_ext_ack *extack)
|
| 858 | {
|
| 859 | bool any_qdisc_is_offloaded;
|
| 860 | int err;
|
| 861 |
|
| 862 | if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
|
| 863 | return;
|
| 864 |
|
| 865 | err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
|
| 866 |
|
| 867 |
|
| 868 | if (!err || !new || new == &noop_qdisc)
|
| 869 | return;
|
| 870 |
|
| 871 | |
| 872 | * one are not offloaded.
|
| 873 |
|
| 874 | any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
|
| 875 | any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
|
| 876 | any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
|
| 877 |
|
| 878 | if (any_qdisc_is_offloaded)
|
| 879 | NL_SET_ERR_MSG_WEAK(extack, "Offloading graft operation failed.");
|
| 880 | }
|
| 881 | EXPORT_SYMBOL(qdisc_offload_graft_helper);
|
| 882 |
|
| 883 | void qdisc_offload_query_caps(struct net_device *dev,
|
| 884 | enum tc_setup_type type,
|
| 885 | void *caps, size_t caps_len)
|
| 886 | {
|
| 887 | const struct net_device_ops *ops = dev->netdev_ops;
|
| 888 | struct tc_query_caps_base base = {
|
| 889 | .type = type,
|
| 890 | .caps = caps,
|
| 891 | };
|
| 892 |
|
| 893 | memset(caps, 0, caps_len);
|
| 894 |
|
| 895 | if (ops->ndo_setup_tc)
|
| 896 | ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
|
| 897 | }
|
| 898 | EXPORT_SYMBOL(qdisc_offload_query_caps);
|
| 899 |
|
| 900 | static void qdisc_offload_graft_root(struct net_device *dev,
|
| 901 | struct Qdisc *new, struct Qdisc *old,
|
| 902 | struct netlink_ext_ack *extack)
|
| 903 | {
|
| 904 | struct tc_root_qopt_offload graft_offload = {
|
| 905 | .command = TC_ROOT_GRAFT,
|
| 906 | .handle = new ? new->handle : 0,
|
| 907 | .ingress = (new && new->flags & TCQ_F_INGRESS) ||
|
| 908 | (old && old->flags & TCQ_F_INGRESS),
|
| 909 | };
|
| 910 |
|
| 911 | qdisc_offload_graft_helper(dev, NULL, new, old,
|
| 912 | TC_SETUP_ROOT_QDISC, &graft_offload, extack);
|
| 913 | }
|
| 914 |
|
| 915 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
|
| 916 | u32 portid, u32 seq, u16 flags, int event,
|
| 917 | struct netlink_ext_ack *extack)
|
| 918 | {
|
| 919 | struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
|
| 920 | struct gnet_stats_queue __percpu *cpu_qstats = NULL;
|
| 921 | struct tcmsg *tcm;
|
| 922 | struct nlmsghdr *nlh;
|
| 923 | unsigned char *b = skb_tail_pointer(skb);
|
| 924 | struct gnet_dump d;
|
| 925 | struct qdisc_size_table *stab;
|
| 926 | u32 block_index;
|
| 927 | __u32 qlen;
|
| 928 |
|
| 929 | cond_resched();
|
| 930 | nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
|
| 931 | if (!nlh)
|
| 932 | goto out_nlmsg_trim;
|
| 933 | tcm = nlmsg_data(nlh);
|
| 934 | tcm->tcm_family = AF_UNSPEC;
|
| 935 | tcm->tcm__pad1 = 0;
|
| 936 | tcm->tcm__pad2 = 0;
|
| 937 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
|
| 938 | tcm->tcm_parent = clid;
|
| 939 | tcm->tcm_handle = q->handle;
|
| 940 | tcm->tcm_info = refcount_read(&q->refcnt);
|
| 941 | if (nla_put_string(skb, TCA_KIND, q->ops->id))
|
| 942 | goto nla_put_failure;
|
| 943 | if (q->ops->ingress_block_get) {
|
| 944 | block_index = q->ops->ingress_block_get(q);
|
| 945 | if (block_index &&
|
| 946 | nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
|
| 947 | goto nla_put_failure;
|
| 948 | }
|
| 949 | if (q->ops->egress_block_get) {
|
| 950 | block_index = q->ops->egress_block_get(q);
|
| 951 | if (block_index &&
|
| 952 | nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
|
| 953 | goto nla_put_failure;
|
| 954 | }
|
| 955 | if (q->ops->dump && q->ops->dump(q, skb) < 0)
|
| 956 | goto nla_put_failure;
|
| 957 | if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
|
| 958 | goto nla_put_failure;
|
| 959 | qlen = qdisc_qlen_sum(q);
|
| 960 |
|
| 961 | stab = rtnl_dereference(q->stab);
|
| 962 | if (stab && qdisc_dump_stab(skb, stab) < 0)
|
| 963 | goto nla_put_failure;
|
| 964 |
|
| 965 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
|
| 966 | NULL, &d, TCA_PAD) < 0)
|
| 967 | goto nla_put_failure;
|
| 968 |
|
| 969 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
|
| 970 | goto nla_put_failure;
|
| 971 |
|
| 972 | if (qdisc_is_percpu_stats(q)) {
|
| 973 | cpu_bstats = q->cpu_bstats;
|
| 974 | cpu_qstats = q->cpu_qstats;
|
| 975 | }
|
| 976 |
|
| 977 | if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
|
| 978 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
|
| 979 | gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
|
| 980 | goto nla_put_failure;
|
| 981 |
|
| 982 | if (gnet_stats_finish_copy(&d) < 0)
|
| 983 | goto nla_put_failure;
|
| 984 |
|
| 985 | if (extack && extack->_msg &&
|
| 986 | nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
|
| 987 | goto out_nlmsg_trim;
|
| 988 |
|
| 989 | nlh->nlmsg_len = skb_tail_pointer(skb) - b;
|
| 990 |
|
| 991 | return skb->len;
|
| 992 |
|
| 993 | out_nlmsg_trim:
|
| 994 | nla_put_failure:
|
| 995 | nlmsg_trim(skb, b);
|
| 996 | return -EMSGSIZE;
|
| 997 | }
|
| 998 |
|
| 999 | static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible,
|
| 1000 | const struct tcmsg *tcm)
|
| 1001 | {
|
| 1002 | if (q->flags & TCQ_F_BUILTIN)
|
| 1003 | return true;
|
| 1004 | if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
|
| 1005 | return true;
|
| 1006 | if (tcm) {
|
| 1007 | if (tcm->tcm_handle && tcm->tcm_handle != q->handle)
|
| 1008 | return true;
|
| 1009 | }
|
| 1010 | return false;
|
| 1011 | }
|
| 1012 |
|
| 1013 | static int qdisc_get_notify(struct net *net, struct sk_buff *oskb,
|
| 1014 | struct nlmsghdr *n, u32 clid, struct Qdisc *q,
|
| 1015 | struct netlink_ext_ack *extack)
|
| 1016 | {
|
| 1017 | struct sk_buff *skb;
|
| 1018 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
|
| 1019 |
|
| 1020 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
| 1021 | if (!skb)
|
| 1022 | return -ENOBUFS;
|
| 1023 |
|
| 1024 | if (!tc_qdisc_dump_ignore(q, false, NULL)) {
|
| 1025 | if (tc_fill_qdisc(skb, q, clid, portid, n->nlmsg_seq, 0,
|
| 1026 | RTM_NEWQDISC, extack) < 0)
|
| 1027 | goto err_out;
|
| 1028 | }
|
| 1029 |
|
| 1030 | if (skb->len)
|
| 1031 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
|
| 1032 | n->nlmsg_flags & NLM_F_ECHO);
|
| 1033 |
|
| 1034 | err_out:
|
| 1035 | kfree_skb(skb);
|
| 1036 | return -EINVAL;
|
| 1037 | }
|
| 1038 |
|
| 1039 | static int qdisc_notify(struct net *net, struct sk_buff *oskb,
|
| 1040 | struct nlmsghdr *n, u32 clid,
|
| 1041 | struct Qdisc *old, struct Qdisc *new,
|
| 1042 | struct netlink_ext_ack *extack)
|
| 1043 | {
|
| 1044 | struct sk_buff *skb;
|
| 1045 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
|
| 1046 |
|
| 1047 | if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
|
| 1048 | return 0;
|
| 1049 |
|
| 1050 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
| 1051 | if (!skb)
|
| 1052 | return -ENOBUFS;
|
| 1053 |
|
| 1054 | if (old && !tc_qdisc_dump_ignore(old, false, NULL)) {
|
| 1055 | if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
|
| 1056 | 0, RTM_DELQDISC, extack) < 0)
|
| 1057 | goto err_out;
|
| 1058 | }
|
| 1059 | if (new && !tc_qdisc_dump_ignore(new, false, NULL)) {
|
| 1060 | if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
|
| 1061 | old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
|
| 1062 | goto err_out;
|
| 1063 | }
|
| 1064 |
|
| 1065 | if (skb->len)
|
| 1066 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
|
| 1067 | n->nlmsg_flags & NLM_F_ECHO);
|
| 1068 |
|
| 1069 | err_out:
|
| 1070 | kfree_skb(skb);
|
| 1071 | return -EINVAL;
|
| 1072 | }
|
| 1073 |
|
| 1074 | static void notify_and_destroy(struct net *net, struct sk_buff *skb,
|
| 1075 | struct nlmsghdr *n, u32 clid,
|
| 1076 | struct Qdisc *old, struct Qdisc *new,
|
| 1077 | struct netlink_ext_ack *extack)
|
| 1078 | {
|
| 1079 | if (new || old)
|
| 1080 | qdisc_notify(net, skb, n, clid, old, new, extack);
|
| 1081 |
|
| 1082 | if (old)
|
| 1083 | qdisc_put(old);
|
| 1084 | }
|
| 1085 |
|
| 1086 | static void qdisc_clear_nolock(struct Qdisc *sch)
|
| 1087 | {
|
| 1088 | sch->flags &= ~TCQ_F_NOLOCK;
|
| 1089 | if (!(sch->flags & TCQ_F_CPUSTATS))
|
| 1090 | return;
|
| 1091 |
|
| 1092 | free_percpu(sch->cpu_bstats);
|
| 1093 | free_percpu(sch->cpu_qstats);
|
| 1094 | sch->cpu_bstats = NULL;
|
| 1095 | sch->cpu_qstats = NULL;
|
| 1096 | sch->flags &= ~TCQ_F_CPUSTATS;
|
| 1097 | }
|
| 1098 |
|
| 1099 | |
| 1100 | * to device "dev".
|
| 1101 | *
|
| 1102 | * When appropriate send a netlink notification using 'skb'
|
| 1103 | * and "n".
|
| 1104 | *
|
| 1105 | * On success, destroy old qdisc.
|
| 1106 |
|
| 1107 |
|
| 1108 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
|
| 1109 | struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
|
| 1110 | struct Qdisc *new, struct Qdisc *old,
|
| 1111 | struct netlink_ext_ack *extack)
|
| 1112 | {
|
| 1113 | struct Qdisc *q = old;
|
| 1114 | struct net *net = dev_net(dev);
|
| 1115 |
|
| 1116 | if (parent == NULL) {
|
| 1117 | unsigned int i, num_q, ingress;
|
| 1118 | struct netdev_queue *dev_queue;
|
| 1119 |
|
| 1120 | if (new)
|
| 1121 | new->depth = 0;
|
| 1122 |
|
| 1123 | ingress = 0;
|
| 1124 | num_q = dev->num_tx_queues;
|
| 1125 | if ((q && q->flags & TCQ_F_INGRESS) ||
|
| 1126 | (new && new->flags & TCQ_F_INGRESS)) {
|
| 1127 | ingress = 1;
|
| 1128 | dev_queue = dev_ingress_queue(dev);
|
| 1129 | if (!dev_queue) {
|
| 1130 | NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
|
| 1131 | return -ENOENT;
|
| 1132 | }
|
| 1133 |
|
| 1134 | q = rtnl_dereference(dev_queue->qdisc_sleeping);
|
| 1135 |
|
| 1136 | |
| 1137 | * __tcf_qdisc_find() for filter requests.
|
| 1138 |
|
| 1139 | if (!qdisc_refcount_dec_if_one(q)) {
|
| 1140 | NL_SET_ERR_MSG(extack,
|
| 1141 | "Current ingress or clsact Qdisc has ongoing filter requests");
|
| 1142 | return -EBUSY;
|
| 1143 | }
|
| 1144 | }
|
| 1145 |
|
| 1146 | if (dev->flags & IFF_UP)
|
| 1147 | dev_deactivate(dev, false);
|
| 1148 |
|
| 1149 | qdisc_offload_graft_root(dev, new, old, extack);
|
| 1150 |
|
| 1151 | if (new && new->ops->attach && !ingress)
|
| 1152 | goto skip;
|
| 1153 |
|
| 1154 | if (!ingress) {
|
| 1155 | for (i = 0; i < num_q; i++) {
|
| 1156 | dev_queue = netdev_get_tx_queue(dev, i);
|
| 1157 | old = dev_graft_qdisc(dev_queue, new);
|
| 1158 |
|
| 1159 | if (new && i > 0)
|
| 1160 | qdisc_refcount_inc(new);
|
| 1161 | qdisc_put(old);
|
| 1162 | }
|
| 1163 | } else {
|
| 1164 | old = dev_graft_qdisc(dev_queue, NULL);
|
| 1165 |
|
| 1166 | |
| 1167 | * unprotected concurrent accesses to net_device::miniq_{in,e}gress
|
| 1168 | * pointer(s) in mini_qdisc_pair_swap().
|
| 1169 |
|
| 1170 | qdisc_notify(net, skb, n, classid, old, new, extack);
|
| 1171 | qdisc_destroy(old);
|
| 1172 |
|
| 1173 | dev_graft_qdisc(dev_queue, new);
|
| 1174 | }
|
| 1175 |
|
| 1176 | skip:
|
| 1177 | if (!ingress) {
|
| 1178 | old = rtnl_dereference(dev->qdisc);
|
| 1179 | if (new && !new->ops->attach)
|
| 1180 | qdisc_refcount_inc(new);
|
| 1181 | rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
|
| 1182 |
|
| 1183 | notify_and_destroy(net, skb, n, classid, old, new, extack);
|
| 1184 |
|
| 1185 | if (new && new->ops->attach)
|
| 1186 | new->ops->attach(new);
|
| 1187 | }
|
| 1188 |
|
| 1189 | if (dev->flags & IFF_UP)
|
| 1190 | dev_activate(dev);
|
| 1191 | } else {
|
| 1192 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
|
| 1193 | unsigned long cl;
|
| 1194 | int err;
|
| 1195 |
|
| 1196 |
|
| 1197 | if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
|
| 1198 | qdisc_clear_nolock(new);
|
| 1199 |
|
| 1200 | if (!cops || !cops->graft)
|
| 1201 | return -EOPNOTSUPP;
|
| 1202 |
|
| 1203 | cl = cops->find(parent, classid);
|
| 1204 | if (!cl) {
|
| 1205 | NL_SET_ERR_MSG(extack, "Specified class not found");
|
| 1206 | return -ENOENT;
|
| 1207 | }
|
| 1208 |
|
| 1209 | if (new && new->ops == &noqueue_qdisc_ops) {
|
| 1210 | NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
|
| 1211 | return -EINVAL;
|
| 1212 | }
|
| 1213 |
|
| 1214 | if (new &&
|
| 1215 | !(parent->flags & TCQ_F_MQROOT) &&
|
| 1216 | rcu_access_pointer(new->stab)) {
|
| 1217 | NL_SET_ERR_MSG(extack, "STAB not supported on a non root");
|
| 1218 | return -EINVAL;
|
| 1219 | }
|
| 1220 | if (new && parent->depth >= 7) {
|
| 1221 | NL_SET_ERR_MSG(extack, "Qdisc hierarchy is too deep");
|
| 1222 | return -E2BIG;
|
| 1223 | }
|
| 1224 | err = cops->graft(parent, cl, new, &old, extack);
|
| 1225 | if (err)
|
| 1226 | return err;
|
| 1227 | if (new)
|
| 1228 | new->depth = parent->depth + 1;
|
| 1229 | notify_and_destroy(net, skb, n, classid, old, new, extack);
|
| 1230 | }
|
| 1231 | return 0;
|
| 1232 | }
|
| 1233 |
|
| 1234 | static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
|
| 1235 | struct netlink_ext_ack *extack)
|
| 1236 | {
|
| 1237 | u32 block_index;
|
| 1238 |
|
| 1239 | if (tca[TCA_INGRESS_BLOCK]) {
|
| 1240 | block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
|
| 1241 |
|
| 1242 | if (!block_index) {
|
| 1243 | NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
|
| 1244 | return -EINVAL;
|
| 1245 | }
|
| 1246 | if (!sch->ops->ingress_block_set) {
|
| 1247 | NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
|
| 1248 | return -EOPNOTSUPP;
|
| 1249 | }
|
| 1250 | sch->ops->ingress_block_set(sch, block_index);
|
| 1251 | }
|
| 1252 | if (tca[TCA_EGRESS_BLOCK]) {
|
| 1253 | block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
|
| 1254 |
|
| 1255 | if (!block_index) {
|
| 1256 | NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
|
| 1257 | return -EINVAL;
|
| 1258 | }
|
| 1259 | if (!sch->ops->egress_block_set) {
|
| 1260 | NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
|
| 1261 | return -EOPNOTSUPP;
|
| 1262 | }
|
| 1263 | sch->ops->egress_block_set(sch, block_index);
|
| 1264 | }
|
| 1265 | return 0;
|
| 1266 | }
|
| 1267 |
|
| 1268 | |
| 1269 | Allocate and initialize new qdisc.
|
| 1270 |
|
| 1271 | Parameters are passed via opt.
|
| 1272 |
|
| 1273 |
|
| 1274 | static struct Qdisc *qdisc_create(struct net_device *dev,
|
| 1275 | struct netdev_queue *dev_queue,
|
| 1276 | u32 parent, u32 handle,
|
| 1277 | struct nlattr **tca, int *errp,
|
| 1278 | struct netlink_ext_ack *extack)
|
| 1279 | {
|
| 1280 | int err;
|
| 1281 | struct nlattr *kind = tca[TCA_KIND];
|
| 1282 | struct Qdisc *sch;
|
| 1283 | struct Qdisc_ops *ops;
|
| 1284 | struct qdisc_size_table *stab;
|
| 1285 |
|
| 1286 | ops = qdisc_lookup_ops(kind);
|
| 1287 | if (!ops) {
|
| 1288 | err = -ENOENT;
|
| 1289 | NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
|
| 1290 | goto err_out;
|
| 1291 | }
|
| 1292 |
|
| 1293 | sch = qdisc_alloc(dev_queue, ops, extack);
|
| 1294 | if (IS_ERR(sch)) {
|
| 1295 | err = PTR_ERR(sch);
|
| 1296 | goto err_out2;
|
| 1297 | }
|
| 1298 |
|
| 1299 | sch->parent = parent;
|
| 1300 |
|
| 1301 | if (handle == TC_H_INGRESS) {
|
| 1302 | if (!(sch->flags & TCQ_F_INGRESS)) {
|
| 1303 | NL_SET_ERR_MSG(extack,
|
| 1304 | "Specified parent ID is reserved for ingress and clsact Qdiscs");
|
| 1305 | err = -EINVAL;
|
| 1306 | goto err_out3;
|
| 1307 | }
|
| 1308 | handle = TC_H_MAKE(TC_H_INGRESS, 0);
|
| 1309 | } else {
|
| 1310 | if (handle == 0) {
|
| 1311 | handle = qdisc_alloc_handle(dev);
|
| 1312 | if (handle == 0) {
|
| 1313 | NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
|
| 1314 | err = -ENOSPC;
|
| 1315 | goto err_out3;
|
| 1316 | }
|
| 1317 | }
|
| 1318 | if (!netif_is_multiqueue(dev))
|
| 1319 | sch->flags |= TCQ_F_ONETXQUEUE;
|
| 1320 | }
|
| 1321 |
|
| 1322 | sch->handle = handle;
|
| 1323 |
|
| 1324 | |
| 1325 | * loophole, what allowed userspace to get IFF_NO_QUEUE
|
| 1326 | * facility on older kernels by setting tx_queue_len=0 (prior
|
| 1327 | * to qdisc init), and then forgot to reinit tx_queue_len
|
| 1328 | * before again attaching a qdisc.
|
| 1329 |
|
| 1330 | if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
|
| 1331 | WRITE_ONCE(dev->tx_queue_len, DEFAULT_TX_QUEUE_LEN);
|
| 1332 | netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
|
| 1333 | }
|
| 1334 |
|
| 1335 | err = qdisc_block_indexes_set(sch, tca, extack);
|
| 1336 | if (err)
|
| 1337 | goto err_out3;
|
| 1338 |
|
| 1339 | if (tca[TCA_STAB]) {
|
| 1340 | stab = qdisc_get_stab(tca[TCA_STAB], extack);
|
| 1341 | if (IS_ERR(stab)) {
|
| 1342 | err = PTR_ERR(stab);
|
| 1343 | goto err_out3;
|
| 1344 | }
|
| 1345 | rcu_assign_pointer(sch->stab, stab);
|
| 1346 | }
|
| 1347 |
|
| 1348 | if (ops->init) {
|
| 1349 | err = ops->init(sch, tca[TCA_OPTIONS], extack);
|
| 1350 | if (err != 0)
|
| 1351 | goto err_out4;
|
| 1352 | }
|
| 1353 |
|
| 1354 | if (tca[TCA_RATE]) {
|
| 1355 | err = -EOPNOTSUPP;
|
| 1356 | if (sch->flags & TCQ_F_MQROOT) {
|
| 1357 | NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
|
| 1358 | goto err_out4;
|
| 1359 | }
|
| 1360 |
|
| 1361 | err = gen_new_estimator(&sch->bstats,
|
| 1362 | sch->cpu_bstats,
|
| 1363 | &sch->rate_est,
|
| 1364 | NULL,
|
| 1365 | true,
|
| 1366 | tca[TCA_RATE]);
|
| 1367 | if (err) {
|
| 1368 | NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
|
| 1369 | goto err_out4;
|
| 1370 | }
|
| 1371 | }
|
| 1372 |
|
| 1373 | qdisc_hash_add(sch, false);
|
| 1374 | trace_qdisc_create(ops, dev, parent);
|
| 1375 |
|
| 1376 | return sch;
|
| 1377 |
|
| 1378 | err_out4:
|
| 1379 | |
| 1380 | * like qdisc_create_dflt().
|
| 1381 |
|
| 1382 | if (ops->destroy)
|
| 1383 | ops->destroy(sch);
|
| 1384 | qdisc_put_stab(rtnl_dereference(sch->stab));
|
| 1385 | err_out3:
|
| 1386 | qdisc_lock_uninit(sch, ops);
|
| 1387 | netdev_put(dev, &sch->dev_tracker);
|
| 1388 | qdisc_free_rcu(sch);
|
| 1389 | err_out2:
|
| 1390 | bpf_module_put(ops, ops->owner);
|
| 1391 | err_out:
|
| 1392 | *errp = err;
|
| 1393 | return NULL;
|
| 1394 | }
|
| 1395 |
|
| 1396 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
|
| 1397 | struct netlink_ext_ack *extack)
|
| 1398 | {
|
| 1399 | struct qdisc_size_table *ostab, *stab = NULL;
|
| 1400 | int err = 0;
|
| 1401 |
|
| 1402 | if (tca[TCA_OPTIONS]) {
|
| 1403 | if (!sch->ops->change) {
|
| 1404 | NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
|
| 1405 | return -EINVAL;
|
| 1406 | }
|
| 1407 | if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
|
| 1408 | NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
|
| 1409 | return -EOPNOTSUPP;
|
| 1410 | }
|
| 1411 | err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
|
| 1412 | if (err)
|
| 1413 | return err;
|
| 1414 | }
|
| 1415 |
|
| 1416 | if (tca[TCA_STAB]) {
|
| 1417 | stab = qdisc_get_stab(tca[TCA_STAB], extack);
|
| 1418 | if (IS_ERR(stab))
|
| 1419 | return PTR_ERR(stab);
|
| 1420 | }
|
| 1421 |
|
| 1422 | ostab = rtnl_dereference(sch->stab);
|
| 1423 | rcu_assign_pointer(sch->stab, stab);
|
| 1424 | qdisc_put_stab(ostab);
|
| 1425 |
|
| 1426 | if (tca[TCA_RATE]) {
|
| 1427 | |
| 1428 | because change can't be undone.
|
| 1429 | if (sch->flags & TCQ_F_MQROOT)
|
| 1430 | goto out;
|
| 1431 | gen_replace_estimator(&sch->bstats,
|
| 1432 | sch->cpu_bstats,
|
| 1433 | &sch->rate_est,
|
| 1434 | NULL,
|
| 1435 | true,
|
| 1436 | tca[TCA_RATE]);
|
| 1437 | }
|
| 1438 | out:
|
| 1439 | return 0;
|
| 1440 | }
|
| 1441 |
|
| 1442 | struct check_loop_arg {
|
| 1443 | struct qdisc_walker w;
|
| 1444 | struct Qdisc *p;
|
| 1445 | int depth;
|
| 1446 | };
|
| 1447 |
|
| 1448 | static int check_loop_fn(struct Qdisc *q, unsigned long cl,
|
| 1449 | struct qdisc_walker *w);
|
| 1450 |
|
| 1451 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
|
| 1452 | {
|
| 1453 | struct check_loop_arg arg;
|
| 1454 |
|
| 1455 | if (q->ops->cl_ops == NULL)
|
| 1456 | return 0;
|
| 1457 |
|
| 1458 | arg.w.stop = arg.w.skip = arg.w.count = 0;
|
| 1459 | arg.w.fn = check_loop_fn;
|
| 1460 | arg.depth = depth;
|
| 1461 | arg.p = p;
|
| 1462 | q->ops->cl_ops->walk(q, &arg.w);
|
| 1463 | return arg.w.stop ? -ELOOP : 0;
|
| 1464 | }
|
| 1465 |
|
| 1466 | static int
|
| 1467 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
|
| 1468 | {
|
| 1469 | struct Qdisc *leaf;
|
| 1470 | const struct Qdisc_class_ops *cops = q->ops->cl_ops;
|
| 1471 | struct check_loop_arg *arg = (struct check_loop_arg *)w;
|
| 1472 |
|
| 1473 | leaf = cops->leaf(q, cl);
|
| 1474 | if (leaf) {
|
| 1475 | if (leaf == arg->p || arg->depth > 7)
|
| 1476 | return -ELOOP;
|
| 1477 | return check_loop(leaf, arg->p, arg->depth + 1);
|
| 1478 | }
|
| 1479 | return 0;
|
| 1480 | }
|
| 1481 |
|
| 1482 | const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
|
| 1483 | [TCA_KIND] = { .type = NLA_STRING },
|
| 1484 | [TCA_RATE] = { .type = NLA_BINARY,
|
| 1485 | .len = sizeof(struct tc_estimator) },
|
| 1486 | [TCA_STAB] = { .type = NLA_NESTED },
|
| 1487 | [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
|
| 1488 | [TCA_CHAIN] = { .type = NLA_U32 },
|
| 1489 | [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
|
| 1490 | [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
|
| 1491 | };
|
| 1492 |
|
| 1493 | |
| 1494 | * Delete/get qdisc.
|
| 1495 |
|
| 1496 |
|
| 1497 | static int __tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
|
| 1498 | struct netlink_ext_ack *extack,
|
| 1499 | struct net_device *dev,
|
| 1500 | struct nlattr *tca[TCA_MAX + 1],
|
| 1501 | struct tcmsg *tcm)
|
| 1502 | {
|
| 1503 | struct net *net = sock_net(skb->sk);
|
| 1504 | struct Qdisc *q = NULL;
|
| 1505 | struct Qdisc *p = NULL;
|
| 1506 | u32 clid;
|
| 1507 | int err;
|
| 1508 |
|
| 1509 | clid = tcm->tcm_parent;
|
| 1510 | if (clid) {
|
| 1511 | if (clid != TC_H_ROOT) {
|
| 1512 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
|
| 1513 | p = qdisc_lookup(dev, TC_H_MAJ(clid));
|
| 1514 | if (!p) {
|
| 1515 | NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
|
| 1516 | return -ENOENT;
|
| 1517 | }
|
| 1518 | q = qdisc_leaf(p, clid, extack);
|
| 1519 | } else if (dev_ingress_queue(dev)) {
|
| 1520 | q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
|
| 1521 | }
|
| 1522 | } else {
|
| 1523 | q = rtnl_dereference(dev->qdisc);
|
| 1524 | }
|
| 1525 | if (!q) {
|
| 1526 | NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
|
| 1527 | return -ENOENT;
|
| 1528 | }
|
| 1529 | if (IS_ERR(q))
|
| 1530 | return PTR_ERR(q);
|
| 1531 |
|
| 1532 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
|
| 1533 | NL_SET_ERR_MSG(extack, "Invalid handle");
|
| 1534 | return -EINVAL;
|
| 1535 | }
|
| 1536 | } else {
|
| 1537 | q = qdisc_lookup(dev, tcm->tcm_handle);
|
| 1538 | if (!q) {
|
| 1539 | NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
|
| 1540 | return -ENOENT;
|
| 1541 | }
|
| 1542 | }
|
| 1543 |
|
| 1544 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
|
| 1545 | NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
|
| 1546 | return -EINVAL;
|
| 1547 | }
|
| 1548 |
|
| 1549 | if (n->nlmsg_type == RTM_DELQDISC) {
|
| 1550 | if (!clid) {
|
| 1551 | NL_SET_ERR_MSG(extack, "Classid cannot be zero");
|
| 1552 | return -EINVAL;
|
| 1553 | }
|
| 1554 | if (q->handle == 0) {
|
| 1555 | NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
|
| 1556 | return -ENOENT;
|
| 1557 | }
|
| 1558 | err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
|
| 1559 | if (err != 0)
|
| 1560 | return err;
|
| 1561 | } else {
|
| 1562 | qdisc_get_notify(net, skb, n, clid, q, NULL);
|
| 1563 | }
|
| 1564 | return 0;
|
| 1565 | }
|
| 1566 |
|
| 1567 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
|
| 1568 | struct netlink_ext_ack *extack)
|
| 1569 | {
|
| 1570 | struct net *net = sock_net(skb->sk);
|
| 1571 | struct tcmsg *tcm = nlmsg_data(n);
|
| 1572 | struct nlattr *tca[TCA_MAX + 1];
|
| 1573 | struct net_device *dev;
|
| 1574 | int err;
|
| 1575 |
|
| 1576 | err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
|
| 1577 | rtm_tca_policy, extack);
|
| 1578 | if (err < 0)
|
| 1579 | return err;
|
| 1580 |
|
| 1581 | dev = __dev_get_by_index(net, tcm->tcm_ifindex);
|
| 1582 | if (!dev)
|
| 1583 | return -ENODEV;
|
| 1584 |
|
| 1585 | netdev_lock_ops(dev);
|
| 1586 | err = __tc_get_qdisc(skb, n, extack, dev, tca, tcm);
|
| 1587 | netdev_unlock_ops(dev);
|
| 1588 |
|
| 1589 | return err;
|
| 1590 | }
|
| 1591 |
|
| 1592 | static bool req_create_or_replace(struct nlmsghdr *n)
|
| 1593 | {
|
| 1594 | return (n->nlmsg_flags & NLM_F_CREATE &&
|
| 1595 | n->nlmsg_flags & NLM_F_REPLACE);
|
| 1596 | }
|
| 1597 |
|
| 1598 | static bool req_create_exclusive(struct nlmsghdr *n)
|
| 1599 | {
|
| 1600 | return (n->nlmsg_flags & NLM_F_CREATE &&
|
| 1601 | n->nlmsg_flags & NLM_F_EXCL);
|
| 1602 | }
|
| 1603 |
|
| 1604 | static bool req_change(struct nlmsghdr *n)
|
| 1605 | {
|
| 1606 | return (!(n->nlmsg_flags & NLM_F_CREATE) &&
|
| 1607 | !(n->nlmsg_flags & NLM_F_REPLACE) &&
|
| 1608 | !(n->nlmsg_flags & NLM_F_EXCL));
|
| 1609 | }
|
| 1610 |
|
| 1611 | static int __tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
|
| 1612 | struct netlink_ext_ack *extack,
|
| 1613 | struct net_device *dev,
|
| 1614 | struct nlattr *tca[TCA_MAX + 1],
|
| 1615 | struct tcmsg *tcm)
|
| 1616 | {
|
| 1617 | struct Qdisc *q = NULL;
|
| 1618 | struct Qdisc *p = NULL;
|
| 1619 | u32 clid;
|
| 1620 | int err;
|
| 1621 |
|
| 1622 | clid = tcm->tcm_parent;
|
| 1623 |
|
| 1624 | if (clid) {
|
| 1625 | if (clid != TC_H_ROOT) {
|
| 1626 | if (clid != TC_H_INGRESS) {
|
| 1627 | p = qdisc_lookup(dev, TC_H_MAJ(clid));
|
| 1628 | if (!p) {
|
| 1629 | NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
|
| 1630 | return -ENOENT;
|
| 1631 | }
|
| 1632 | if (p->flags & TCQ_F_INGRESS) {
|
| 1633 | NL_SET_ERR_MSG(extack,
|
| 1634 | "Cannot add children to ingress/clsact qdisc");
|
| 1635 | return -EOPNOTSUPP;
|
| 1636 | }
|
| 1637 | q = qdisc_leaf(p, clid, extack);
|
| 1638 | if (IS_ERR(q))
|
| 1639 | return PTR_ERR(q);
|
| 1640 | } else if (dev_ingress_queue_create(dev)) {
|
| 1641 | q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
|
| 1642 | }
|
| 1643 | } else {
|
| 1644 | q = rtnl_dereference(dev->qdisc);
|
| 1645 | }
|
| 1646 |
|
| 1647 |
|
| 1648 | if (q && q->handle == 0)
|
| 1649 | q = NULL;
|
| 1650 |
|
| 1651 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
|
| 1652 | if (tcm->tcm_handle) {
|
| 1653 | if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
|
| 1654 | NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
|
| 1655 | return -EEXIST;
|
| 1656 | }
|
| 1657 | if (TC_H_MIN(tcm->tcm_handle)) {
|
| 1658 | NL_SET_ERR_MSG(extack, "Invalid minor handle");
|
| 1659 | return -EINVAL;
|
| 1660 | }
|
| 1661 | q = qdisc_lookup(dev, tcm->tcm_handle);
|
| 1662 | if (!q)
|
| 1663 | goto create_n_graft;
|
| 1664 | if (q->parent != tcm->tcm_parent) {
|
| 1665 | NL_SET_ERR_MSG(extack, "Cannot move an existing qdisc to a different parent");
|
| 1666 | return -EINVAL;
|
| 1667 | }
|
| 1668 | if (n->nlmsg_flags & NLM_F_EXCL) {
|
| 1669 | NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
|
| 1670 | return -EEXIST;
|
| 1671 | }
|
| 1672 | if (tca[TCA_KIND] &&
|
| 1673 | nla_strcmp(tca[TCA_KIND], q->ops->id)) {
|
| 1674 | NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
|
| 1675 | return -EINVAL;
|
| 1676 | }
|
| 1677 | if (q->flags & TCQ_F_INGRESS) {
|
| 1678 | NL_SET_ERR_MSG(extack,
|
| 1679 | "Cannot regraft ingress or clsact Qdiscs");
|
| 1680 | return -EINVAL;
|
| 1681 | }
|
| 1682 | if (q == p ||
|
| 1683 | (p && check_loop(q, p, 0))) {
|
| 1684 | NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
|
| 1685 | return -ELOOP;
|
| 1686 | }
|
| 1687 | if (clid == TC_H_INGRESS) {
|
| 1688 | NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
|
| 1689 | return -EINVAL;
|
| 1690 | }
|
| 1691 | qdisc_refcount_inc(q);
|
| 1692 | goto graft;
|
| 1693 | } else {
|
| 1694 | if (!q)
|
| 1695 | goto create_n_graft;
|
| 1696 |
|
| 1697 | |
| 1698 | *
|
| 1699 | * We know, that some child q is already
|
| 1700 | * attached to this parent and have choice:
|
| 1701 | * 1) change it or 2) create/graft new one.
|
| 1702 | * If the requested qdisc kind is different
|
| 1703 | * than the existing one, then we choose graft.
|
| 1704 | * If they are the same then this is "change"
|
| 1705 | * operation - just let it fallthrough..
|
| 1706 | *
|
| 1707 | * 1. We are allowed to create/graft only
|
| 1708 | * if the request is explicitly stating
|
| 1709 | * "please create if it doesn't exist".
|
| 1710 | *
|
| 1711 | * 2. If the request is to exclusive create
|
| 1712 | * then the qdisc tcm_handle is not expected
|
| 1713 | * to exist, so that we choose create/graft too.
|
| 1714 | *
|
| 1715 | * 3. The last case is when no flags are set.
|
| 1716 | * This will happen when for example tc
|
| 1717 | * utility issues a "change" command.
|
| 1718 | * Alas, it is sort of hole in API, we
|
| 1719 | * cannot decide what to do unambiguously.
|
| 1720 | * For now we select create/graft.
|
| 1721 |
|
| 1722 | if (tca[TCA_KIND] &&
|
| 1723 | nla_strcmp(tca[TCA_KIND], q->ops->id)) {
|
| 1724 | if (req_create_or_replace(n) ||
|
| 1725 | req_create_exclusive(n))
|
| 1726 | goto create_n_graft;
|
| 1727 | else if (req_change(n))
|
| 1728 | goto create_n_graft2;
|
| 1729 | }
|
| 1730 | }
|
| 1731 | }
|
| 1732 | } else {
|
| 1733 | if (!tcm->tcm_handle) {
|
| 1734 | NL_SET_ERR_MSG(extack, "Handle cannot be zero");
|
| 1735 | return -EINVAL;
|
| 1736 | }
|
| 1737 | q = qdisc_lookup(dev, tcm->tcm_handle);
|
| 1738 | }
|
| 1739 |
|
| 1740 |
|
| 1741 | if (!q) {
|
| 1742 | NL_SET_ERR_MSG(extack, "Specified qdisc not found");
|
| 1743 | return -ENOENT;
|
| 1744 | }
|
| 1745 | if (n->nlmsg_flags & NLM_F_EXCL) {
|
| 1746 | NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
|
| 1747 | return -EEXIST;
|
| 1748 | }
|
| 1749 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
|
| 1750 | NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
|
| 1751 | return -EINVAL;
|
| 1752 | }
|
| 1753 | err = qdisc_change(q, tca, extack);
|
| 1754 | if (err == 0)
|
| 1755 | qdisc_notify(sock_net(skb->sk), skb, n, clid, NULL, q, extack);
|
| 1756 | return err;
|
| 1757 |
|
| 1758 | create_n_graft:
|
| 1759 | if (!(n->nlmsg_flags & NLM_F_CREATE)) {
|
| 1760 | NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
|
| 1761 | return -ENOENT;
|
| 1762 | }
|
| 1763 | create_n_graft2:
|
| 1764 | if (clid == TC_H_INGRESS) {
|
| 1765 | if (dev_ingress_queue(dev)) {
|
| 1766 | q = qdisc_create(dev, dev_ingress_queue(dev),
|
| 1767 | tcm->tcm_parent, tcm->tcm_parent,
|
| 1768 | tca, &err, extack);
|
| 1769 | } else {
|
| 1770 | NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
|
| 1771 | err = -ENOENT;
|
| 1772 | }
|
| 1773 | } else {
|
| 1774 | struct netdev_queue *dev_queue;
|
| 1775 |
|
| 1776 | if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
|
| 1777 | dev_queue = p->ops->cl_ops->select_queue(p, tcm);
|
| 1778 | else if (p)
|
| 1779 | dev_queue = p->dev_queue;
|
| 1780 | else
|
| 1781 | dev_queue = netdev_get_tx_queue(dev, 0);
|
| 1782 |
|
| 1783 | q = qdisc_create(dev, dev_queue,
|
| 1784 | tcm->tcm_parent, tcm->tcm_handle,
|
| 1785 | tca, &err, extack);
|
| 1786 | }
|
| 1787 | if (!q)
|
| 1788 | return err;
|
| 1789 |
|
| 1790 | graft:
|
| 1791 | err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
|
| 1792 | if (err) {
|
| 1793 | if (q)
|
| 1794 | qdisc_put(q);
|
| 1795 | return err;
|
| 1796 | }
|
| 1797 |
|
| 1798 | return 0;
|
| 1799 | }
|
| 1800 |
|
| 1801 | static void request_qdisc_module(struct nlattr *kind)
|
| 1802 | {
|
| 1803 | struct Qdisc_ops *ops;
|
| 1804 | char name[IFNAMSIZ];
|
| 1805 |
|
| 1806 | if (!kind)
|
| 1807 | return;
|
| 1808 |
|
| 1809 | ops = qdisc_lookup_ops(kind);
|
| 1810 | if (ops) {
|
| 1811 | bpf_module_put(ops, ops->owner);
|
| 1812 | return;
|
| 1813 | }
|
| 1814 |
|
| 1815 | if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
|
| 1816 | rtnl_unlock();
|
| 1817 | request_module(NET_SCH_ALIAS_PREFIX "%s", name);
|
| 1818 | rtnl_lock();
|
| 1819 | }
|
| 1820 | }
|
| 1821 |
|
| 1822 | |
| 1823 | * Create/change qdisc.
|
| 1824 |
|
| 1825 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
|
| 1826 | struct netlink_ext_ack *extack)
|
| 1827 | {
|
| 1828 | struct net *net = sock_net(skb->sk);
|
| 1829 | struct nlattr *tca[TCA_MAX + 1];
|
| 1830 | struct net_device *dev;
|
| 1831 | struct tcmsg *tcm;
|
| 1832 | int err;
|
| 1833 |
|
| 1834 | err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
|
| 1835 | rtm_tca_policy, extack);
|
| 1836 | if (err < 0)
|
| 1837 | return err;
|
| 1838 |
|
| 1839 | request_qdisc_module(tca[TCA_KIND]);
|
| 1840 |
|
| 1841 | tcm = nlmsg_data(n);
|
| 1842 | dev = __dev_get_by_index(net, tcm->tcm_ifindex);
|
| 1843 | if (!dev)
|
| 1844 | return -ENODEV;
|
| 1845 |
|
| 1846 | netdev_lock_ops(dev);
|
| 1847 | err = __tc_modify_qdisc(skb, n, extack, dev, tca, tcm);
|
| 1848 | netdev_unlock_ops(dev);
|
| 1849 |
|
| 1850 | return err;
|
| 1851 | }
|
| 1852 |
|
| 1853 | static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
|
| 1854 | struct netlink_callback *cb,
|
| 1855 | int *q_idx_p, int s_q_idx, bool recur,
|
| 1856 | bool dump_invisible)
|
| 1857 | {
|
| 1858 | const struct nlmsghdr *nlh = cb->nlh;
|
| 1859 | int ret = 0, q_idx = *q_idx_p;
|
| 1860 | const struct tcmsg *tcm;
|
| 1861 | struct Qdisc *q;
|
| 1862 | int b;
|
| 1863 |
|
| 1864 | if (!root)
|
| 1865 | return 0;
|
| 1866 |
|
| 1867 | tcm = nlmsg_data(nlh);
|
| 1868 | q = root;
|
| 1869 | if (q_idx < s_q_idx) {
|
| 1870 | q_idx++;
|
| 1871 | } else {
|
| 1872 | if (!tc_qdisc_dump_ignore(q, dump_invisible, tcm))
|
| 1873 | ret = tc_fill_qdisc(skb, q, q->parent,
|
| 1874 | NETLINK_CB(cb->skb).portid,
|
| 1875 | nlh->nlmsg_seq, NLM_F_MULTI,
|
| 1876 | RTM_NEWQDISC, NULL);
|
| 1877 | if (ret < 0)
|
| 1878 | goto out;
|
| 1879 | q_idx++;
|
| 1880 | }
|
| 1881 |
|
| 1882 | |
| 1883 | * itself has already been dumped.
|
| 1884 | *
|
| 1885 | * If we've already dumped the top-level (ingress) qdisc above and the global
|
| 1886 | * qdisc hashtable, we don't want to hit it again
|
| 1887 |
|
| 1888 | if (!qdisc_dev(root) || !recur)
|
| 1889 | goto out;
|
| 1890 |
|
| 1891 | hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
|
| 1892 | if (q_idx < s_q_idx) {
|
| 1893 | q_idx++;
|
| 1894 | continue;
|
| 1895 | }
|
| 1896 | if (!tc_qdisc_dump_ignore(q, dump_invisible, tcm))
|
| 1897 | ret = tc_fill_qdisc(skb, q, q->parent,
|
| 1898 | NETLINK_CB(cb->skb).portid,
|
| 1899 | nlh->nlmsg_seq, NLM_F_MULTI,
|
| 1900 | RTM_NEWQDISC, NULL);
|
| 1901 | if (ret < 0)
|
| 1902 | goto out;
|
| 1903 | q_idx++;
|
| 1904 | }
|
| 1905 |
|
| 1906 | out:
|
| 1907 | *q_idx_p = q_idx;
|
| 1908 | return ret;
|
| 1909 | }
|
| 1910 |
|
| 1911 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
|
| 1912 | {
|
| 1913 | const struct nlmsghdr *nlh = cb->nlh;
|
| 1914 | struct net *net = sock_net(skb->sk);
|
| 1915 | struct nlattr *tca[TCA_MAX + 1];
|
| 1916 | struct {
|
| 1917 | unsigned long ifindex;
|
| 1918 | int q_idx;
|
| 1919 | } *ctx = (void *)cb->ctx;
|
| 1920 | const struct tcmsg *tcm;
|
| 1921 | struct net_device *dev;
|
| 1922 | int s_q_idx, q_idx;
|
| 1923 | int err;
|
| 1924 |
|
| 1925 | ASSERT_RTNL();
|
| 1926 |
|
| 1927 | err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
|
| 1928 | rtm_tca_policy, cb->extack);
|
| 1929 | if (err < 0)
|
| 1930 | return err;
|
| 1931 | tcm = nlmsg_data(nlh);
|
| 1932 | if (tcm->tcm_ifindex && !ctx->ifindex)
|
| 1933 | ctx->ifindex = tcm->tcm_ifindex;
|
| 1934 |
|
| 1935 | s_q_idx = ctx->q_idx;
|
| 1936 |
|
| 1937 | for_each_netdev_dump(net, dev, ctx->ifindex) {
|
| 1938 | struct netdev_queue *dev_queue;
|
| 1939 | struct Qdisc *q;
|
| 1940 |
|
| 1941 | if (tcm->tcm_ifindex && ctx->ifindex != tcm->tcm_ifindex)
|
| 1942 | break;
|
| 1943 |
|
| 1944 | q_idx = 0;
|
| 1945 |
|
| 1946 | netdev_lock_ops(dev);
|
| 1947 | q = rtnl_dereference(dev->qdisc);
|
| 1948 | err = tc_dump_qdisc_root(q, skb, cb, &q_idx, s_q_idx,
|
| 1949 | true, tca[TCA_DUMP_INVISIBLE]);
|
| 1950 | if (err < 0)
|
| 1951 | goto error_unlock;
|
| 1952 |
|
| 1953 | dev_queue = dev_ingress_queue(dev);
|
| 1954 | if (dev_queue) {
|
| 1955 | q = rtnl_dereference(dev_queue->qdisc_sleeping);
|
| 1956 | err = tc_dump_qdisc_root(q, skb, cb, &q_idx, s_q_idx,
|
| 1957 | false, tca[TCA_DUMP_INVISIBLE]);
|
| 1958 | if (err < 0)
|
| 1959 | goto error_unlock;
|
| 1960 | }
|
| 1961 | netdev_unlock_ops(dev);
|
| 1962 | s_q_idx = 0;
|
| 1963 | }
|
| 1964 | return skb->len;
|
| 1965 |
|
| 1966 | error_unlock:
|
| 1967 | netdev_unlock_ops(dev);
|
| 1968 | ctx->q_idx = q_idx;
|
| 1969 |
|
| 1970 | return err;
|
| 1971 | }
|
| 1972 |
|
| 1973 |
|
| 1974 |
|
| 1975 | |
| 1976 | * Traffic classes manipulation. *
|
| 1977 | ***********************************************
|
| 1978 |
|
| 1979 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
|
| 1980 | unsigned long cl, u32 portid, u32 seq, u16 flags,
|
| 1981 | int event, struct netlink_ext_ack *extack)
|
| 1982 | {
|
| 1983 | struct tcmsg *tcm;
|
| 1984 | struct nlmsghdr *nlh;
|
| 1985 | unsigned char *b = skb_tail_pointer(skb);
|
| 1986 | struct gnet_dump d;
|
| 1987 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
|
| 1988 |
|
| 1989 | cond_resched();
|
| 1990 | nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
|
| 1991 | if (!nlh)
|
| 1992 | goto out_nlmsg_trim;
|
| 1993 | tcm = nlmsg_data(nlh);
|
| 1994 | tcm->tcm_family = AF_UNSPEC;
|
| 1995 | tcm->tcm__pad1 = 0;
|
| 1996 | tcm->tcm__pad2 = 0;
|
| 1997 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
|
| 1998 | tcm->tcm_parent = q->handle;
|
| 1999 | tcm->tcm_handle = q->handle;
|
| 2000 | tcm->tcm_info = 0;
|
| 2001 | if (nla_put_string(skb, TCA_KIND, q->ops->id))
|
| 2002 | goto nla_put_failure;
|
| 2003 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
|
| 2004 | goto nla_put_failure;
|
| 2005 |
|
| 2006 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
|
| 2007 | NULL, &d, TCA_PAD) < 0)
|
| 2008 | goto nla_put_failure;
|
| 2009 |
|
| 2010 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
|
| 2011 | goto nla_put_failure;
|
| 2012 |
|
| 2013 | if (gnet_stats_finish_copy(&d) < 0)
|
| 2014 | goto nla_put_failure;
|
| 2015 |
|
| 2016 | if (extack && extack->_msg &&
|
| 2017 | nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
|
| 2018 | goto out_nlmsg_trim;
|
| 2019 |
|
| 2020 | nlh->nlmsg_len = skb_tail_pointer(skb) - b;
|
| 2021 |
|
| 2022 | return skb->len;
|
| 2023 |
|
| 2024 | out_nlmsg_trim:
|
| 2025 | nla_put_failure:
|
| 2026 | nlmsg_trim(skb, b);
|
| 2027 | return -EMSGSIZE;
|
| 2028 | }
|
| 2029 |
|
| 2030 | static int tclass_notify(struct net *net, struct sk_buff *oskb,
|
| 2031 | struct nlmsghdr *n, struct Qdisc *q,
|
| 2032 | unsigned long cl, int event, struct netlink_ext_ack *extack)
|
| 2033 | {
|
| 2034 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
|
| 2035 | struct sk_buff *skb;
|
| 2036 | int ret;
|
| 2037 |
|
| 2038 | if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
|
| 2039 | return 0;
|
| 2040 |
|
| 2041 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
| 2042 | if (!skb)
|
| 2043 | return -ENOBUFS;
|
| 2044 |
|
| 2045 | ret = tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack);
|
| 2046 | if (ret < 0) {
|
| 2047 | kfree_skb(skb);
|
| 2048 | return ret;
|
| 2049 | }
|
| 2050 |
|
| 2051 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
|
| 2052 | n->nlmsg_flags & NLM_F_ECHO);
|
| 2053 | }
|
| 2054 |
|
| 2055 | static int tclass_get_notify(struct net *net, struct sk_buff *oskb,
|
| 2056 | struct nlmsghdr *n, struct Qdisc *q,
|
| 2057 | unsigned long cl, struct netlink_ext_ack *extack)
|
| 2058 | {
|
| 2059 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
|
| 2060 | struct sk_buff *skb;
|
| 2061 | int ret;
|
| 2062 |
|
| 2063 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
| 2064 | if (!skb)
|
| 2065 | return -ENOBUFS;
|
| 2066 |
|
| 2067 | ret = tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
|
| 2068 | RTM_NEWTCLASS, extack);
|
| 2069 | if (ret < 0) {
|
| 2070 | kfree_skb(skb);
|
| 2071 | return ret;
|
| 2072 | }
|
| 2073 |
|
| 2074 | return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
|
| 2075 | n->nlmsg_flags & NLM_F_ECHO);
|
| 2076 | }
|
| 2077 |
|
| 2078 | static int tclass_del_notify(struct net *net,
|
| 2079 | const struct Qdisc_class_ops *cops,
|
| 2080 | struct sk_buff *oskb, struct nlmsghdr *n,
|
| 2081 | struct Qdisc *q, unsigned long cl,
|
| 2082 | struct netlink_ext_ack *extack)
|
| 2083 | {
|
| 2084 | u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
|
| 2085 | struct sk_buff *skb = NULL;
|
| 2086 | int err = 0;
|
| 2087 |
|
| 2088 | if (!cops->delete)
|
| 2089 | return -EOPNOTSUPP;
|
| 2090 |
|
| 2091 | if (rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) {
|
| 2092 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
|
| 2093 | if (!skb)
|
| 2094 | return -ENOBUFS;
|
| 2095 |
|
| 2096 | err = tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
|
| 2097 | RTM_DELTCLASS, extack);
|
| 2098 | if (err < 0) {
|
| 2099 | kfree_skb(skb);
|
| 2100 | return err;
|
| 2101 | }
|
| 2102 | }
|
| 2103 |
|
| 2104 | err = cops->delete(q, cl, extack);
|
| 2105 | if (err) {
|
| 2106 | kfree_skb(skb);
|
| 2107 | return err;
|
| 2108 | }
|
| 2109 |
|
| 2110 | err = rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC,
|
| 2111 | n->nlmsg_flags & NLM_F_ECHO);
|
| 2112 | return err;
|
| 2113 | }
|
| 2114 |
|
| 2115 | #ifdef CONFIG_NET_CLS
|
| 2116 |
|
| 2117 | struct tcf_bind_args {
|
| 2118 | struct tcf_walker w;
|
| 2119 | unsigned long base;
|
| 2120 | unsigned long cl;
|
| 2121 | u32 classid;
|
| 2122 | };
|
| 2123 |
|
| 2124 | static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
|
| 2125 | {
|
| 2126 | struct tcf_bind_args *a = (void *)arg;
|
| 2127 |
|
| 2128 | if (n && tp->ops->bind_class) {
|
| 2129 | struct Qdisc *q = tcf_block_q(tp->chain->block);
|
| 2130 |
|
| 2131 | sch_tree_lock(q);
|
| 2132 | tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
|
| 2133 | sch_tree_unlock(q);
|
| 2134 | }
|
| 2135 | return 0;
|
| 2136 | }
|
| 2137 |
|
| 2138 | struct tc_bind_class_args {
|
| 2139 | struct qdisc_walker w;
|
| 2140 | unsigned long new_cl;
|
| 2141 | u32 portid;
|
| 2142 | u32 clid;
|
| 2143 | };
|
| 2144 |
|
| 2145 | static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
|
| 2146 | struct qdisc_walker *w)
|
| 2147 | {
|
| 2148 | struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
|
| 2149 | const struct Qdisc_class_ops *cops = q->ops->cl_ops;
|
| 2150 | struct tcf_block *block;
|
| 2151 | struct tcf_chain *chain;
|
| 2152 |
|
| 2153 | block = cops->tcf_block(q, cl, NULL);
|
| 2154 | if (!block)
|
| 2155 | return 0;
|
| 2156 | for (chain = tcf_get_next_chain(block, NULL);
|
| 2157 | chain;
|
| 2158 | chain = tcf_get_next_chain(block, chain)) {
|
| 2159 | struct tcf_proto *tp;
|
| 2160 |
|
| 2161 | for (tp = tcf_get_next_proto(chain, NULL);
|
| 2162 | tp; tp = tcf_get_next_proto(chain, tp)) {
|
| 2163 | struct tcf_bind_args arg = {};
|
| 2164 |
|
| 2165 | arg.w.fn = tcf_node_bind;
|
| 2166 | arg.classid = a->clid;
|
| 2167 | arg.base = cl;
|
| 2168 | arg.cl = a->new_cl;
|
| 2169 | tp->ops->walk(tp, &arg.w, true);
|
| 2170 | }
|
| 2171 | }
|
| 2172 |
|
| 2173 | return 0;
|
| 2174 | }
|
| 2175 |
|
| 2176 | static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
|
| 2177 | unsigned long new_cl)
|
| 2178 | {
|
| 2179 | const struct Qdisc_class_ops *cops = q->ops->cl_ops;
|
| 2180 | struct tc_bind_class_args args = {};
|
| 2181 |
|
| 2182 | if (!cops->tcf_block)
|
| 2183 | return;
|
| 2184 | args.portid = portid;
|
| 2185 | args.clid = clid;
|
| 2186 | args.new_cl = new_cl;
|
| 2187 | args.w.fn = tc_bind_class_walker;
|
| 2188 | q->ops->cl_ops->walk(q, &args.w);
|
| 2189 | }
|
| 2190 |
|
| 2191 | #else
|
| 2192 |
|
| 2193 | static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
|
| 2194 | unsigned long new_cl)
|
| 2195 | {
|
| 2196 | }
|
| 2197 |
|
| 2198 | #endif
|
| 2199 |
|
| 2200 | static int __tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
|
| 2201 | struct netlink_ext_ack *extack,
|
| 2202 | struct net_device *dev,
|
| 2203 | struct nlattr *tca[TCA_MAX + 1],
|
| 2204 | struct tcmsg *tcm)
|
| 2205 | {
|
| 2206 | struct net *net = sock_net(skb->sk);
|
| 2207 | const struct Qdisc_class_ops *cops;
|
| 2208 | struct Qdisc *q = NULL;
|
| 2209 | unsigned long cl = 0;
|
| 2210 | unsigned long new_cl;
|
| 2211 | u32 portid;
|
| 2212 | u32 clid;
|
| 2213 | u32 qid;
|
| 2214 | int err;
|
| 2215 |
|
| 2216 | |
| 2217 | parent == TC_H_UNSPEC - unspecified parent.
|
| 2218 | parent == TC_H_ROOT - class is root, which has no parent.
|
| 2219 | parent == X:0 - parent is root class.
|
| 2220 | parent == X:Y - parent is a node in hierarchy.
|
| 2221 | parent == 0:Y - parent is X:Y, where X:0 is qdisc.
|
| 2222 |
|
| 2223 | handle == 0:0 - generate handle from kernel pool.
|
| 2224 | handle == 0:Y - class is X:Y, where X:0 is qdisc.
|
| 2225 | handle == X:Y - clear.
|
| 2226 | handle == X:0 - root class.
|
| 2227 |
|
| 2228 |
|
| 2229 |
|
| 2230 |
|
| 2231 | portid = tcm->tcm_parent;
|
| 2232 | clid = tcm->tcm_handle;
|
| 2233 | qid = TC_H_MAJ(clid);
|
| 2234 |
|
| 2235 | if (portid != TC_H_ROOT) {
|
| 2236 | u32 qid1 = TC_H_MAJ(portid);
|
| 2237 |
|
| 2238 | if (qid && qid1) {
|
| 2239 |
|
| 2240 | if (qid != qid1)
|
| 2241 | return -EINVAL;
|
| 2242 | } else if (qid1) {
|
| 2243 | qid = qid1;
|
| 2244 | } else if (qid == 0)
|
| 2245 | qid = rtnl_dereference(dev->qdisc)->handle;
|
| 2246 |
|
| 2247 | |
| 2248 | * both with parent and child.
|
| 2249 | *
|
| 2250 | * TC_H_MAJ(portid) still may be unspecified, complete it now.
|
| 2251 |
|
| 2252 | if (portid)
|
| 2253 | portid = TC_H_MAKE(qid, portid);
|
| 2254 | } else {
|
| 2255 | if (qid == 0)
|
| 2256 | qid = rtnl_dereference(dev->qdisc)->handle;
|
| 2257 | }
|
| 2258 |
|
| 2259 |
|
| 2260 | q = qdisc_lookup(dev, qid);
|
| 2261 | if (!q)
|
| 2262 | return -ENOENT;
|
| 2263 |
|
| 2264 |
|
| 2265 | cops = q->ops->cl_ops;
|
| 2266 | if (cops == NULL)
|
| 2267 | return -EINVAL;
|
| 2268 |
|
| 2269 |
|
| 2270 | if (clid == 0) {
|
| 2271 | if (portid == TC_H_ROOT)
|
| 2272 | clid = qid;
|
| 2273 | } else
|
| 2274 | clid = TC_H_MAKE(qid, clid);
|
| 2275 |
|
| 2276 | if (clid)
|
| 2277 | cl = cops->find(q, clid);
|
| 2278 |
|
| 2279 | if (cl == 0) {
|
| 2280 | err = -ENOENT;
|
| 2281 | if (n->nlmsg_type != RTM_NEWTCLASS ||
|
| 2282 | !(n->nlmsg_flags & NLM_F_CREATE))
|
| 2283 | goto out;
|
| 2284 | } else {
|
| 2285 | switch (n->nlmsg_type) {
|
| 2286 | case RTM_NEWTCLASS:
|
| 2287 | err = -EEXIST;
|
| 2288 | if (n->nlmsg_flags & NLM_F_EXCL)
|
| 2289 | goto out;
|
| 2290 | break;
|
| 2291 | case RTM_DELTCLASS:
|
| 2292 | err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
|
| 2293 |
|
| 2294 | tc_bind_tclass(q, portid, clid, 0);
|
| 2295 | goto out;
|
| 2296 | case RTM_GETTCLASS:
|
| 2297 | err = tclass_get_notify(net, skb, n, q, cl, extack);
|
| 2298 | goto out;
|
| 2299 | default:
|
| 2300 | err = -EINVAL;
|
| 2301 | goto out;
|
| 2302 | }
|
| 2303 | }
|
| 2304 |
|
| 2305 | if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
|
| 2306 | NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
|
| 2307 | return -EOPNOTSUPP;
|
| 2308 | }
|
| 2309 |
|
| 2310 |
|
| 2311 | if (clid == TC_H_ROOT) {
|
| 2312 | NL_SET_ERR_MSG(extack, "Cannot create traffic class with classid TC_H_ROOT");
|
| 2313 | return -EINVAL;
|
| 2314 | }
|
| 2315 |
|
| 2316 | new_cl = cl;
|
| 2317 | err = -EOPNOTSUPP;
|
| 2318 | if (cops->change)
|
| 2319 | err = cops->change(q, clid, portid, tca, &new_cl, extack);
|
| 2320 | if (err == 0) {
|
| 2321 | tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
|
| 2322 |
|
| 2323 | if (cl != new_cl)
|
| 2324 | tc_bind_tclass(q, portid, clid, new_cl);
|
| 2325 | }
|
| 2326 | out:
|
| 2327 | return err;
|
| 2328 | }
|
| 2329 |
|
| 2330 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
|
| 2331 | struct netlink_ext_ack *extack)
|
| 2332 | {
|
| 2333 | struct net *net = sock_net(skb->sk);
|
| 2334 | struct tcmsg *tcm = nlmsg_data(n);
|
| 2335 | struct nlattr *tca[TCA_MAX + 1];
|
| 2336 | struct net_device *dev;
|
| 2337 | int err;
|
| 2338 |
|
| 2339 | err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
|
| 2340 | rtm_tca_policy, extack);
|
| 2341 | if (err < 0)
|
| 2342 | return err;
|
| 2343 |
|
| 2344 | dev = __dev_get_by_index(net, tcm->tcm_ifindex);
|
| 2345 | if (!dev)
|
| 2346 | return -ENODEV;
|
| 2347 |
|
| 2348 | netdev_lock_ops(dev);
|
| 2349 | err = __tc_ctl_tclass(skb, n, extack, dev, tca, tcm);
|
| 2350 | netdev_unlock_ops(dev);
|
| 2351 |
|
| 2352 | return err;
|
| 2353 | }
|
| 2354 |
|
| 2355 | struct qdisc_dump_args {
|
| 2356 | struct qdisc_walker w;
|
| 2357 | struct sk_buff *skb;
|
| 2358 | struct netlink_callback *cb;
|
| 2359 | };
|
| 2360 |
|
| 2361 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
|
| 2362 | struct qdisc_walker *arg)
|
| 2363 | {
|
| 2364 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
|
| 2365 |
|
| 2366 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
|
| 2367 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
|
| 2368 | RTM_NEWTCLASS, NULL);
|
| 2369 | }
|
| 2370 |
|
| 2371 | static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
|
| 2372 | struct tcmsg *tcm, struct netlink_callback *cb,
|
| 2373 | int *t_p, int s_t)
|
| 2374 | {
|
| 2375 | struct qdisc_dump_args arg;
|
| 2376 |
|
| 2377 | if (tc_qdisc_dump_ignore(q, false, NULL) ||
|
| 2378 | *t_p < s_t || !q->ops->cl_ops ||
|
| 2379 | (tcm->tcm_parent &&
|
| 2380 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
|
| 2381 | (*t_p)++;
|
| 2382 | return 0;
|
| 2383 | }
|
| 2384 | if (*t_p > s_t)
|
| 2385 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
|
| 2386 | arg.w.fn = qdisc_class_dump;
|
| 2387 | arg.skb = skb;
|
| 2388 | arg.cb = cb;
|
| 2389 | arg.w.stop = 0;
|
| 2390 | arg.w.skip = cb->args[1];
|
| 2391 | arg.w.count = 0;
|
| 2392 | q->ops->cl_ops->walk(q, &arg.w);
|
| 2393 | cb->args[1] = arg.w.count;
|
| 2394 | if (arg.w.stop)
|
| 2395 | return -1;
|
| 2396 | (*t_p)++;
|
| 2397 | return 0;
|
| 2398 | }
|
| 2399 |
|
| 2400 | static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
|
| 2401 | struct tcmsg *tcm, struct netlink_callback *cb,
|
| 2402 | int *t_p, int s_t, bool recur)
|
| 2403 | {
|
| 2404 | struct Qdisc *q;
|
| 2405 | int b;
|
| 2406 |
|
| 2407 | if (!root)
|
| 2408 | return 0;
|
| 2409 |
|
| 2410 | if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
|
| 2411 | return -1;
|
| 2412 |
|
| 2413 | if (!qdisc_dev(root) || !recur)
|
| 2414 | return 0;
|
| 2415 |
|
| 2416 | if (tcm->tcm_parent) {
|
| 2417 | q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
|
| 2418 | if (q && q != root &&
|
| 2419 | tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
|
| 2420 | return -1;
|
| 2421 | return 0;
|
| 2422 | }
|
| 2423 | hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
|
| 2424 | if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
|
| 2425 | return -1;
|
| 2426 | }
|
| 2427 |
|
| 2428 | return 0;
|
| 2429 | }
|
| 2430 |
|
| 2431 | static int __tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb,
|
| 2432 | struct tcmsg *tcm, struct net_device *dev)
|
| 2433 | {
|
| 2434 | struct netdev_queue *dev_queue;
|
| 2435 | int t, s_t;
|
| 2436 |
|
| 2437 | s_t = cb->args[0];
|
| 2438 | t = 0;
|
| 2439 |
|
| 2440 | if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
|
| 2441 | skb, tcm, cb, &t, s_t, true) < 0)
|
| 2442 | goto done;
|
| 2443 |
|
| 2444 | dev_queue = dev_ingress_queue(dev);
|
| 2445 | if (dev_queue &&
|
| 2446 | tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
|
| 2447 | skb, tcm, cb, &t, s_t, false) < 0)
|
| 2448 | goto done;
|
| 2449 |
|
| 2450 | done:
|
| 2451 | cb->args[0] = t;
|
| 2452 |
|
| 2453 | return skb->len;
|
| 2454 | }
|
| 2455 |
|
| 2456 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
|
| 2457 | {
|
| 2458 | struct tcmsg *tcm = nlmsg_data(cb->nlh);
|
| 2459 | struct net *net = sock_net(skb->sk);
|
| 2460 | struct net_device *dev;
|
| 2461 | int err;
|
| 2462 |
|
| 2463 | if (nlmsg_len(cb->nlh) < sizeof(*tcm))
|
| 2464 | return 0;
|
| 2465 |
|
| 2466 | dev = dev_get_by_index(net, tcm->tcm_ifindex);
|
| 2467 | if (!dev)
|
| 2468 | return 0;
|
| 2469 |
|
| 2470 | netdev_lock_ops(dev);
|
| 2471 | err = __tc_dump_tclass(skb, cb, tcm, dev);
|
| 2472 | netdev_unlock_ops(dev);
|
| 2473 |
|
| 2474 | dev_put(dev);
|
| 2475 |
|
| 2476 | return err;
|
| 2477 | }
|
| 2478 |
|
| 2479 | #ifdef CONFIG_PROC_FS
|
| 2480 | static int psched_show(struct seq_file *seq, void *v)
|
| 2481 | {
|
| 2482 | seq_printf(seq, "%08x %08x %08x %08x\n",
|
| 2483 | (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
|
| 2484 | 1000000,
|
| 2485 | (u32)NSEC_PER_SEC / hrtimer_resolution);
|
| 2486 |
|
| 2487 | return 0;
|
| 2488 | }
|
| 2489 |
|
| 2490 | static int __net_init psched_net_init(struct net *net)
|
| 2491 | {
|
| 2492 | struct proc_dir_entry *e;
|
| 2493 |
|
| 2494 | e = proc_create_single("psched", 0, net->proc_net, psched_show);
|
| 2495 | if (e == NULL)
|
| 2496 | return -ENOMEM;
|
| 2497 |
|
| 2498 | return 0;
|
| 2499 | }
|
| 2500 |
|
| 2501 | static void __net_exit psched_net_exit(struct net *net)
|
| 2502 | {
|
| 2503 | remove_proc_entry("psched", net->proc_net);
|
| 2504 | }
|
| 2505 | #else
|
| 2506 | static int __net_init psched_net_init(struct net *net)
|
| 2507 | {
|
| 2508 | return 0;
|
| 2509 | }
|
| 2510 |
|
| 2511 | static void __net_exit psched_net_exit(struct net *net)
|
| 2512 | {
|
| 2513 | }
|
| 2514 | #endif
|
| 2515 |
|
| 2516 | static struct pernet_operations psched_net_ops = {
|
| 2517 | .init = psched_net_init,
|
| 2518 | .exit = psched_net_exit,
|
| 2519 | };
|
| 2520 |
|
| 2521 | #if IS_ENABLED(CONFIG_MITIGATION_RETPOLINE)
|
| 2522 | DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper_act);
|
| 2523 | DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper_cls);
|
| 2524 | #endif
|
| 2525 |
|
| 2526 | static const struct rtnl_msg_handler psched_rtnl_msg_handlers[] __initconst = {
|
| 2527 | {.msgtype = RTM_NEWQDISC, .doit = tc_modify_qdisc},
|
| 2528 | {.msgtype = RTM_DELQDISC, .doit = tc_get_qdisc},
|
| 2529 | {.msgtype = RTM_GETQDISC, .doit = tc_get_qdisc,
|
| 2530 | .dumpit = tc_dump_qdisc},
|
| 2531 | {.msgtype = RTM_NEWTCLASS, .doit = tc_ctl_tclass},
|
| 2532 | {.msgtype = RTM_DELTCLASS, .doit = tc_ctl_tclass},
|
| 2533 | {.msgtype = RTM_GETTCLASS, .doit = tc_ctl_tclass,
|
| 2534 | .dumpit = tc_dump_tclass},
|
| 2535 | };
|
| 2536 |
|
| 2537 | static int __init pktsched_init(void)
|
| 2538 | {
|
| 2539 | int err;
|
| 2540 |
|
| 2541 | err = register_pernet_subsys(&psched_net_ops);
|
| 2542 | if (err) {
|
| 2543 | pr_err("pktsched_init: "
|
| 2544 | "cannot initialize per netns operations\n");
|
| 2545 | return err;
|
| 2546 | }
|
| 2547 |
|
| 2548 | register_qdisc(&pfifo_fast_ops);
|
| 2549 | register_qdisc(&pfifo_qdisc_ops);
|
| 2550 | register_qdisc(&bfifo_qdisc_ops);
|
| 2551 | register_qdisc(&pfifo_head_drop_qdisc_ops);
|
| 2552 | register_qdisc(&mq_qdisc_ops);
|
| 2553 | register_qdisc(&noqueue_qdisc_ops);
|
| 2554 |
|
| 2555 | rtnl_register_many(psched_rtnl_msg_handlers);
|
| 2556 |
|
| 2557 | tc_wrapper_init();
|
| 2558 |
|
| 2559 | return 0;
|
| 2560 | }
|
| 2561 |
|
| 2562 | subsys_initcall(pktsched_init);
|