block/elevator.c

   1 /*
   2  *  Block device elevator/IO-scheduler.
   3  *
   4  *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
   5  *
   6  * 30042000 Jens Axboe <axboe@kernel.dk> :
   7  *
   8  * Split the elevator a bit so that it is possible to choose a different
   9  * one or even write a new "plug in". There are three pieces:
  10  * - elevator_fn, inserts a new request in the queue list
  11  * - elevator_merge_fn, decides whether a new buffer can be merged with
  12  *   an existing request
  13  * - elevator_dequeue_fn, called when a request is taken off the active list
  14  *
  15  * 20082000 Dave Jones <davej@suse.de> :
  16  * Removed tests for max-bomb-segments, which was breaking elvtune
  17  *  when run without -bN
  18  *
  19  * Jens:
  20  * - Rework again to work with bio instead of buffer_heads
  21  * - loose bi_dev comparisons, partition handling is right now
  22  * - completely modularize elevator setup and teardown
  23  *
  24  */
  25 #include <linux/kernel.h>
  26 #include <linux/fs.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/elevator.h>
  29 #include <linux/bio.h>
  30 #include <linux/module.h>
  31 #include <linux/slab.h>
  32 #include <linux/init.h>
  33 #include <linux/compiler.h>
  34 #include <linux/blktrace_api.h>
  35 #include <linux/hash.h>
  36 #include <linux/uaccess.h>
  37 #include <linux/pm_runtime.h>
  38 #include <linux/blk-cgroup.h>
  39
  40 #include <trace/events/block.h>
  41
  42 #include "blk.h"
  43 #include "blk-mq-sched.h"
  44
  45 static DEFINE_SPINLOCK(elv_list_lock);
  46 static LIST_HEAD(elv_list);
  47
  48 /*
  49  * Merge hash stuff.
  50  */
  51 #define rq_hash_key(rq)         (blk_rq_pos(rq) + blk_rq_sectors(rq))
  52
  53 /*
  54  * Query io scheduler to see if the current process issuing bio may be
  55  * merged with rq.
  56  */
  57 static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio)
  58 {
  59         struct request_queue *q = rq->q;
  60         struct elevator_queue *e = q->elevator;
  61
  62         if (e->uses_mq && e->type->ops.mq.allow_merge)
  63                 return e->type->ops.mq.allow_merge(q, rq, bio);
  64         else if (!e->uses_mq && e->type->ops.sq.elevator_allow_bio_merge_fn)
  65                 return e->type->ops.sq.elevator_allow_bio_merge_fn(q, rq, bio);
  66
  67         return 1;
  68 }
  69
  70 /*
  71  * can we safely merge with this request?
  72  */
  73 bool elv_bio_merge_ok(struct request *rq, struct bio *bio)
  74 {
  75         if (!blk_rq_merge_ok(rq, bio))
  76                 return false;
  77
  78         if (!elv_iosched_allow_bio_merge(rq, bio))
  79                 return false;
  80
  81         return true;
  82 }
  83 EXPORT_SYMBOL(elv_bio_merge_ok);
  84
  85 static struct elevator_type *elevator_find(const char *name)
  86 {
  87         struct elevator_type *e;
  88
  89         list_for_each_entry(e, &elv_list, list) {
  90                 if (!strcmp(e->elevator_name, name))
  91                         return e;
  92         }
  93
  94         return NULL;
  95 }
  96
  97 static void elevator_put(struct elevator_type *e)
  98 {
  99         module_put(e->elevator_owner);
 100 }
 101
 102 static struct elevator_type *elevator_get(const char *name, bool try_loading)
 103 {
 104         struct elevator_type *e;
 105
 106         spin_lock(&elv_list_lock);
 107
 108         e = elevator_find(name);
 109         if (!e && try_loading) {
 110                 spin_unlock(&elv_list_lock);
 111                 request_module("%s-iosched", name);
 112                 spin_lock(&elv_list_lock);
 113                 e = elevator_find(name);
 114         }
 115
 116         if (e && !try_module_get(e->elevator_owner))
 117                 e = NULL;
 118
 119         spin_unlock(&elv_list_lock);
 120
 121         return e;
 122 }
 123
 124 static char chosen_elevator[ELV_NAME_MAX];
 125
 126 static int __init elevator_setup(char *str)
 127 {
 128         /*
 129          * Be backwards-compatible with previous kernels, so users
 130          * won't get the wrong elevator.
 131          */
 132         strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
 133         return 1;
 134 }
 135
 136 __setup("elevator=", elevator_setup);
 137
 138 /* called during boot to load the elevator chosen by the elevator param */
 139 void __init load_default_elevator_module(void)
 140 {
 141         struct elevator_type *e;
 142
 143         if (!chosen_elevator[0])
 144                 return;
 145
 146         spin_lock(&elv_list_lock);
 147         e = elevator_find(chosen_elevator);
 148         spin_unlock(&elv_list_lock);
 149
 150         if (!e)
 151                 request_module("%s-iosched", chosen_elevator);
 152 }
 153
 154 static struct kobj_type elv_ktype;
 155
 156 struct elevator_queue *elevator_alloc(struct request_queue *q,
 157                                   struct elevator_type *e)
 158 {
 159         struct elevator_queue *eq;
 160
 161         eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
 162         if (unlikely(!eq))
 163                 return NULL;
 164
 165         eq->type = e;
 166         kobject_init(&eq->kobj, &elv_ktype);
 167         mutex_init(&eq->sysfs_lock);
 168         hash_init(eq->hash);
 169         eq->uses_mq = e->uses_mq;
 170
 171         return eq;
 172 }
 173 EXPORT_SYMBOL(elevator_alloc);
 174
 175 static void elevator_release(struct kobject *kobj)
 176 {
 177         struct elevator_queue *e;
 178
 179         e = container_of(kobj, struct elevator_queue, kobj);
 180         elevator_put(e->type);
 181         kfree(e);
 182 }
 183
 184 int elevator_init(struct request_queue *q, char *name)
 185 {
 186         struct elevator_type *e = NULL;
 187         int err;
 188
 189         /*
 190          * q->sysfs_lock must be held to provide mutual exclusion between
 191          * elevator_switch() and here.
 192          */
 193         lockdep_assert_held(&q->sysfs_lock);
 194
 195         if (unlikely(q->elevator))
 196                 return 0;
 197
 198         INIT_LIST_HEAD(&q->queue_head);
 199         q->last_merge = NULL;
 200         q->end_sector = 0;
 201         q->boundary_rq = NULL;
 202
 203         if (name) {
 204                 e = elevator_get(name, true);
 205                 if (!e)
 206                         return -EINVAL;
 207         }
 208
 209         /*
 210          * Use the default elevator specified by config boot param for
 211          * non-mq devices, or by config option. Don't try to load modules
 212          * as we could be running off async and request_module() isn't
 213          * allowed from async.
 214          */
 215         if (!e && !q->mq_ops && *chosen_elevator) {
 216                 e = elevator_get(chosen_elevator, false);
 217                 if (!e)
 218                         printk(KERN_ERR "I/O scheduler %s not found\n",
 219                                                         chosen_elevator);
 220         }
 221
 222         if (!e) {
 223                 /*
 224                  * For blk-mq devices, we default to using mq-deadline,
 225                  * if available, for single queue devices. If deadline
 226                  * isn't available OR we have multiple queues, default
 227                  * to "none".
 228                  */
 229                 if (q->mq_ops) {
 230                         if (q->nr_hw_queues == 1)
 231                                 e = elevator_get("mq-deadline", false);
 232                         if (!e)
 233                                 return 0;
 234                 } else
 235                         e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
 236
 237                 if (!e) {
 238                         printk(KERN_ERR
 239                                 "Default I/O scheduler not found. " \
 240                                 "Using noop.\n");
 241                         e = elevator_get("noop", false);
 242                 }
 243         }
 244
 245         if (e->uses_mq)
 246                 err = blk_mq_init_sched(q, e);
 247         else
 248                 err = e->ops.sq.elevator_init_fn(q, e);
 249         if (err)
 250                 elevator_put(e);
 251         return err;
 252 }
 253 EXPORT_SYMBOL(elevator_init);
 254
 255 void elevator_exit(struct request_queue *q, struct elevator_queue *e)
 256 {
 257         mutex_lock(&e->sysfs_lock);
 258         if (e->uses_mq && e->type->ops.mq.exit_sched)
 259                 blk_mq_exit_sched(q, e);
 260         else if (!e->uses_mq && e->type->ops.sq.elevator_exit_fn)
 261                 e->type->ops.sq.elevator_exit_fn(e);
 262         mutex_unlock(&e->sysfs_lock);
 263
 264         kobject_put(&e->kobj);
 265 }
 266 EXPORT_SYMBOL(elevator_exit);
 267
 268 static inline void __elv_rqhash_del(struct request *rq)
 269 {
 270         hash_del(&rq->hash);
 271         rq->rq_flags &= ~RQF_HASHED;
 272 }
 273
 274 void elv_rqhash_del(struct request_queue *q, struct request *rq)
 275 {
 276         if (ELV_ON_HASH(rq))
 277                 __elv_rqhash_del(rq);
 278 }
 279 EXPORT_SYMBOL_GPL(elv_rqhash_del);
 280
 281 void elv_rqhash_add(struct request_queue *q, struct request *rq)
 282 {
 283         struct elevator_queue *e = q->elevator;
 284
 285         BUG_ON(ELV_ON_HASH(rq));
 286         hash_add(e->hash, &rq->hash, rq_hash_key(rq));
 287         rq->rq_flags |= RQF_HASHED;
 288 }
 289 EXPORT_SYMBOL_GPL(elv_rqhash_add);
 290
 291 void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
 292 {
 293         __elv_rqhash_del(rq);
 294         elv_rqhash_add(q, rq);
 295 }
 296
 297 struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
 298 {
 299         struct elevator_queue *e = q->elevator;
 300         struct hlist_node *next;
 301         struct request *rq;
 302
 303         hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
 304                 BUG_ON(!ELV_ON_HASH(rq));
 305
 306                 if (unlikely(!rq_mergeable(rq))) {
 307                         __elv_rqhash_del(rq);
 308                         continue;
 309                 }
 310
 311                 if (rq_hash_key(rq) == offset)
 312                         return rq;
 313         }
 314
 315         return NULL;
 316 }
 317
 318 /*
 319  * RB-tree support functions for inserting/lookup/removal of requests
 320  * in a sorted RB tree.
 321  */
 322 void elv_rb_add(struct rb_root *root, struct request *rq)
 323 {
 324         struct rb_node **p = &root->rb_node;
 325         struct rb_node *parent = NULL;
 326         struct request *__rq;
 327
 328         while (*p) {
 329                 parent = *p;
 330                 __rq = rb_entry(parent, struct request, rb_node);
 331
 332                 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
 333                         p = &(*p)->rb_left;
 334                 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
 335                         p = &(*p)->rb_right;
 336         }
 337
 338         rb_link_node(&rq->rb_node, parent, p);
 339         rb_insert_color(&rq->rb_node, root);
 340 }
 341 EXPORT_SYMBOL(elv_rb_add);
 342
 343 void elv_rb_del(struct rb_root *root, struct request *rq)
 344 {
 345         BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
 346         rb_erase(&rq->rb_node, root);
 347         RB_CLEAR_NODE(&rq->rb_node);
 348 }
 349 EXPORT_SYMBOL(elv_rb_del);
 350
 351 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
 352 {
 353         struct rb_node *n = root->rb_node;
 354         struct request *rq;
 355
 356         while (n) {
 357                 rq = rb_entry(n, struct request, rb_node);
 358
 359                 if (sector < blk_rq_pos(rq))
 360                         n = n->rb_left;
 361                 else if (sector > blk_rq_pos(rq))
 362                         n = n->rb_right;
 363                 else
 364                         return rq;
 365         }
 366
 367         return NULL;
 368 }
 369 EXPORT_SYMBOL(elv_rb_find);
 370
 371 /*
 372  * Insert rq into dispatch queue of q.  Queue lock must be held on
 373  * entry.  rq is sort instead into the dispatch queue. To be used by
 374  * specific elevators.
 375  */
 376 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
 377 {
 378         sector_t boundary;
 379         struct list_head *entry;
 380
 381         if (q->last_merge == rq)
 382                 q->last_merge = NULL;
 383
 384         elv_rqhash_del(q, rq);
 385
 386         q->nr_sorted--;
 387
 388         boundary = q->end_sector;
 389         list_for_each_prev(entry, &q->queue_head) {
 390                 struct request *pos = list_entry_rq(entry);
 391
 392                 if (req_op(rq) != req_op(pos))
 393                         break;
 394                 if (rq_data_dir(rq) != rq_data_dir(pos))
 395                         break;
 396                 if (pos->rq_flags & (RQF_STARTED | RQF_SOFTBARRIER))
 397                         break;
 398                 if (blk_rq_pos(rq) >= boundary) {
 399                         if (blk_rq_pos(pos) < boundary)
 400                                 continue;
 401                 } else {
 402                         if (blk_rq_pos(pos) >= boundary)
 403                                 break;
 404                 }
 405                 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
 406                         break;
 407         }
 408
 409         list_add(&rq->queuelist, entry);
 410 }
 411 EXPORT_SYMBOL(elv_dispatch_sort);
 412
 413 /*
 414  * Insert rq into dispatch queue of q.  Queue lock must be held on
 415  * entry.  rq is added to the back of the dispatch queue. To be used by
 416  * specific elevators.
 417  */
 418 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
 419 {
 420         if (q->last_merge == rq)
 421                 q->last_merge = NULL;
 422
 423         elv_rqhash_del(q, rq);
 424
 425         q->nr_sorted--;
 426
 427         q->end_sector = rq_end_sector(rq);
 428         q->boundary_rq = rq;
 429         list_add_tail(&rq->queuelist, &q->queue_head);
 430 }
 431 EXPORT_SYMBOL(elv_dispatch_add_tail);
 432
 433 enum elv_merge elv_merge(struct request_queue *q, struct request **req,
 434                 struct bio *bio)
 435 {
 436         struct elevator_queue *e = q->elevator;
 437         struct request *__rq;
 438
 439         /*
 440          * Levels of merges:
 441          *      nomerges:  No merges at all attempted
 442          *      noxmerges: Only simple one-hit cache try
 443          *      merges:    All merge tries attempted
 444          */
 445         if (blk_queue_nomerges(q) || !bio_mergeable(bio))
 446                 return ELEVATOR_NO_MERGE;
 447
 448         /*
 449          * First try one-hit cache.
 450          */
 451         if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) {
 452                 enum elv_merge ret = blk_try_merge(q->last_merge, bio);
 453
 454                 if (ret != ELEVATOR_NO_MERGE) {
 455                         *req = q->last_merge;
 456                         return ret;
 457                 }
 458         }
 459
 460         if (blk_queue_noxmerges(q))
 461                 return ELEVATOR_NO_MERGE;
 462
 463         /*
 464          * See if our hash lookup can find a potential backmerge.
 465          */
 466         __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
 467         if (__rq && elv_bio_merge_ok(__rq, bio)) {
 468                 *req = __rq;
 469                 return ELEVATOR_BACK_MERGE;
 470         }
 471
 472         if (e->uses_mq && e->type->ops.mq.request_merge)
 473                 return e->type->ops.mq.request_merge(q, req, bio);
 474         else if (!e->uses_mq && e->type->ops.sq.elevator_merge_fn)
 475                 return e->type->ops.sq.elevator_merge_fn(q, req, bio);
 476
 477         return ELEVATOR_NO_MERGE;
 478 }
 479
 480 /*
 481  * Attempt to do an insertion back merge. Only check for the case where
 482  * we can append 'rq' to an existing request, so we can throw 'rq' away
 483  * afterwards.
 484  *
 485  * Returns true if we merged, false otherwise
 486  */
 487 bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq)
 488 {
 489         struct request *__rq;
 490         bool ret;
 491
 492         if (blk_queue_nomerges(q))
 493                 return false;
 494
 495         /*
 496          * First try one-hit cache.
 497          */
 498         if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
 499                 return true;
 500
 501         if (blk_queue_noxmerges(q))
 502                 return false;
 503
 504         ret = false;
 505         /*
 506          * See if our hash lookup can find a potential backmerge.
 507          */
 508         while (1) {
 509                 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
 510                 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
 511                         break;
 512
 513                 /* The merged request could be merged with others, try again */
 514                 ret = true;
 515                 rq = __rq;
 516         }
 517
 518         return ret;
 519 }
 520
 521 void elv_merged_request(struct request_queue *q, struct request *rq,
 522                 enum elv_merge type)
 523 {
 524         struct elevator_queue *e = q->elevator;
 525
 526         if (e->uses_mq && e->type->ops.mq.request_merged)
 527                 e->type->ops.mq.request_merged(q, rq, type);
 528         else if (!e->uses_mq && e->type->ops.sq.elevator_merged_fn)
 529                 e->type->ops.sq.elevator_merged_fn(q, rq, type);
 530
 531         if (type == ELEVATOR_BACK_MERGE)
 532                 elv_rqhash_reposition(q, rq);
 533
 534         q->last_merge = rq;
 535 }
 536
 537 void elv_merge_requests(struct request_queue *q, struct request *rq,
 538                              struct request *next)
 539 {
 540         struct elevator_queue *e = q->elevator;
 541         bool next_sorted = false;
 542
 543         if (e->uses_mq && e->type->ops.mq.requests_merged)
 544                 e->type->ops.mq.requests_merged(q, rq, next);
 545         else if (e->type->ops.sq.elevator_merge_req_fn) {
 546                 next_sorted = (__force bool)(next->rq_flags & RQF_SORTED);
 547                 if (next_sorted)
 548                         e->type->ops.sq.elevator_merge_req_fn(q, rq, next);
 549         }
 550
 551         elv_rqhash_reposition(q, rq);
 552
 553         if (next_sorted) {
 554                 elv_rqhash_del(q, next);
 555                 q->nr_sorted--;
 556         }
 557
 558         q->last_merge = rq;
 559 }
 560
 561 void elv_bio_merged(struct request_queue *q, struct request *rq,
 562                         struct bio *bio)
 563 {
 564         struct elevator_queue *e = q->elevator;
 565
 566         if (WARN_ON_ONCE(e->uses_mq))
 567                 return;
 568
 569         if (e->type->ops.sq.elevator_bio_merged_fn)
 570                 e->type->ops.sq.elevator_bio_merged_fn(q, rq, bio);
 571 }
 572
 573 #ifdef CONFIG_PM
 574 static void blk_pm_requeue_request(struct request *rq)
 575 {
 576         if (rq->q->dev && !(rq->rq_flags & RQF_PM))
 577                 rq->q->nr_pending--;
 578 }
 579
 580 static void blk_pm_add_request(struct request_queue *q, struct request *rq)
 581 {
 582         if (q->dev && !(rq->rq_flags & RQF_PM) && q->nr_pending++ == 0 &&
 583             (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
 584                 pm_request_resume(q->dev);
 585 }
 586 #else
 587 static inline void blk_pm_requeue_request(struct request *rq) {}
 588 static inline void blk_pm_add_request(struct request_queue *q,
 589                                       struct request *rq)
 590 {
 591 }
 592 #endif
 593
 594 void elv_requeue_request(struct request_queue *q, struct request *rq)
 595 {
 596         /*
 597          * it already went through dequeue, we need to decrement the
 598          * in_flight count again
 599          */
 600         if (blk_account_rq(rq)) {
 601                 q->in_flight[rq_is_sync(rq)]--;
 602                 if (rq->rq_flags & RQF_SORTED)
 603                         elv_deactivate_rq(q, rq);
 604         }
 605
 606         rq->rq_flags &= ~RQF_STARTED;
 607
 608         blk_pm_requeue_request(rq);
 609
 610         __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
 611 }
 612
 613 void elv_drain_elevator(struct request_queue *q)
 614 {
 615         struct elevator_queue *e = q->elevator;
 616         static int printed;
 617
 618         if (WARN_ON_ONCE(e->uses_mq))
 619                 return;
 620
 621         lockdep_assert_held(q->queue_lock);
 622
 623         while (e->type->ops.sq.elevator_dispatch_fn(q, 1))
 624                 ;
 625         if (q->nr_sorted && printed++ < 10) {
 626                 printk(KERN_ERR "%s: forced dispatching is broken "
 627                        "(nr_sorted=%u), please report this\n",
 628                        q->elevator->type->elevator_name, q->nr_sorted);
 629         }
 630 }
 631
 632 void __elv_add_request(struct request_queue *q, struct request *rq, int where)
 633 {
 634         trace_block_rq_insert(q, rq);
 635
 636         blk_pm_add_request(q, rq);
 637
 638         rq->q = q;
 639
 640         if (rq->rq_flags & RQF_SOFTBARRIER) {
 641                 /* barriers are scheduling boundary, update end_sector */
 642                 if (!blk_rq_is_passthrough(rq)) {
 643                         q->end_sector = rq_end_sector(rq);
 644                         q->boundary_rq = rq;
 645                 }
 646         } else if (!(rq->rq_flags & RQF_ELVPRIV) &&
 647                     (where == ELEVATOR_INSERT_SORT ||
 648                      where == ELEVATOR_INSERT_SORT_MERGE))
 649                 where = ELEVATOR_INSERT_BACK;
 650
 651         switch (where) {
 652         case ELEVATOR_INSERT_REQUEUE:
 653         case ELEVATOR_INSERT_FRONT:
 654                 rq->rq_flags |= RQF_SOFTBARRIER;
 655                 list_add(&rq->queuelist, &q->queue_head);
 656                 break;
 657
 658         case ELEVATOR_INSERT_BACK:
 659                 rq->rq_flags |= RQF_SOFTBARRIER;
 660                 elv_drain_elevator(q);
 661                 list_add_tail(&rq->queuelist, &q->queue_head);
 662                 /*
 663                  * We kick the queue here for the following reasons.
 664                  * - The elevator might have returned NULL previously
 665                  *   to delay requests and returned them now.  As the
 666                  *   queue wasn't empty before this request, ll_rw_blk
 667                  *   won't run the queue on return, resulting in hang.
 668                  * - Usually, back inserted requests won't be merged
 669                  *   with anything.  There's no point in delaying queue
 670                  *   processing.
 671                  */
 672                 __blk_run_queue(q);
 673                 break;
 674
 675         case ELEVATOR_INSERT_SORT_MERGE:
 676                 /*
 677                  * If we succeed in merging this request with one in the
 678                  * queue already, we are done - rq has now been freed,
 679                  * so no need to do anything further.
 680                  */
 681                 if (elv_attempt_insert_merge(q, rq))
 682                         break;
 683         case ELEVATOR_INSERT_SORT:
 684                 BUG_ON(blk_rq_is_passthrough(rq));
 685                 rq->rq_flags |= RQF_SORTED;
 686                 q->nr_sorted++;
 687                 if (rq_mergeable(rq)) {
 688                         elv_rqhash_add(q, rq);
 689                         if (!q->last_merge)
 690                                 q->last_merge = rq;
 691                 }
 692
 693                 /*
 694                  * Some ioscheds (cfq) run q->request_fn directly, so
 695                  * rq cannot be accessed after calling
 696                  * elevator_add_req_fn.
 697                  */
 698                 q->elevator->type->ops.sq.elevator_add_req_fn(q, rq);
 699                 break;
 700
 701         case ELEVATOR_INSERT_FLUSH:
 702                 rq->rq_flags |= RQF_SOFTBARRIER;
 703                 blk_insert_flush(rq);
 704                 break;
 705         default:
 706                 printk(KERN_ERR "%s: bad insertion point %d\n",
 707                        __func__, where);
 708                 BUG();
 709         }
 710 }
 711 EXPORT_SYMBOL(__elv_add_request);
 712
 713 void elv_add_request(struct request_queue *q, struct request *rq, int where)
 714 {
 715         unsigned long flags;
 716
 717         spin_lock_irqsave(q->queue_lock, flags);
 718         __elv_add_request(q, rq, where);
 719         spin_unlock_irqrestore(q->queue_lock, flags);
 720 }
 721 EXPORT_SYMBOL(elv_add_request);
 722
 723 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
 724 {
 725         struct elevator_queue *e = q->elevator;
 726
 727         if (e->uses_mq && e->type->ops.mq.next_request)
 728                 return e->type->ops.mq.next_request(q, rq);
 729         else if (!e->uses_mq && e->type->ops.sq.elevator_latter_req_fn)
 730                 return e->type->ops.sq.elevator_latter_req_fn(q, rq);
 731
 732         return NULL;
 733 }
 734
 735 struct request *elv_former_request(struct request_queue *q, struct request *rq)
 736 {
 737         struct elevator_queue *e = q->elevator;
 738
 739         if (e->uses_mq && e->type->ops.mq.former_request)
 740                 return e->type->ops.mq.former_request(q, rq);
 741         if (!e->uses_mq && e->type->ops.sq.elevator_former_req_fn)
 742                 return e->type->ops.sq.elevator_former_req_fn(q, rq);
 743         return NULL;
 744 }
 745
 746 int elv_set_request(struct request_queue *q, struct request *rq,
 747                     struct bio *bio, gfp_t gfp_mask)
 748 {
 749         struct elevator_queue *e = q->elevator;
 750
 751         if (WARN_ON_ONCE(e->uses_mq))
 752                 return 0;
 753
 754         if (e->type->ops.sq.elevator_set_req_fn)
 755                 return e->type->ops.sq.elevator_set_req_fn(q, rq, bio, gfp_mask);
 756         return 0;
 757 }
 758
 759 void elv_put_request(struct request_queue *q, struct request *rq)
 760 {
 761         struct elevator_queue *e = q->elevator;
 762
 763         if (WARN_ON_ONCE(e->uses_mq))
 764                 return;
 765
 766         if (e->type->ops.sq.elevator_put_req_fn)
 767                 e->type->ops.sq.elevator_put_req_fn(rq);
 768 }
 769
 770 int elv_may_queue(struct request_queue *q, unsigned int op)
 771 {
 772         struct elevator_queue *e = q->elevator;
 773
 774         if (WARN_ON_ONCE(e->uses_mq))
 775                 return 0;
 776
 777         if (e->type->ops.sq.elevator_may_queue_fn)
 778                 return e->type->ops.sq.elevator_may_queue_fn(q, op);
 779
 780         return ELV_MQUEUE_MAY;
 781 }
 782
 783 void elv_completed_request(struct request_queue *q, struct request *rq)
 784 {
 785         struct elevator_queue *e = q->elevator;
 786
 787         if (WARN_ON_ONCE(e->uses_mq))
 788                 return;
 789
 790         /*
 791          * request is released from the driver, io must be done
 792          */
 793         if (blk_account_rq(rq)) {
 794                 q->in_flight[rq_is_sync(rq)]--;
 795                 if ((rq->rq_flags & RQF_SORTED) &&
 796                     e->type->ops.sq.elevator_completed_req_fn)
 797                         e->type->ops.sq.elevator_completed_req_fn(q, rq);
 798         }
 799 }
 800
 801 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
 802
 803 static ssize_t
 804 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 805 {
 806         struct elv_fs_entry *entry = to_elv(attr);
 807         struct elevator_queue *e;
 808         ssize_t error;
 809
 810         if (!entry->show)
 811                 return -EIO;
 812
 813         e = container_of(kobj, struct elevator_queue, kobj);
 814         mutex_lock(&e->sysfs_lock);
 815         error = e->type ? entry->show(e, page) : -ENOENT;
 816         mutex_unlock(&e->sysfs_lock);
 817         return error;
 818 }
 819
 820 static ssize_t
 821 elv_attr_store(struct kobject *kobj, struct attribute *attr,
 822                const char *page, size_t length)
 823 {
 824         struct elv_fs_entry *entry = to_elv(attr);
 825         struct elevator_queue *e;
 826         ssize_t error;
 827
 828         if (!entry->store)
 829                 return -EIO;
 830
 831         e = container_of(kobj, struct elevator_queue, kobj);
 832         mutex_lock(&e->sysfs_lock);
 833         error = e->type ? entry->store(e, page, length) : -ENOENT;
 834         mutex_unlock(&e->sysfs_lock);
 835         return error;
 836 }
 837
 838 static const struct sysfs_ops elv_sysfs_ops = {
 839         .show   = elv_attr_show,
 840         .store  = elv_attr_store,
 841 };
 842
 843 static struct kobj_type elv_ktype = {
 844         .sysfs_ops      = &elv_sysfs_ops,
 845         .release        = elevator_release,
 846 };
 847
 848 int elv_register_queue(struct request_queue *q)
 849 {
 850         struct elevator_queue *e = q->elevator;
 851         int error;
 852
 853         error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
 854         if (!error) {
 855                 struct elv_fs_entry *attr = e->type->elevator_attrs;
 856                 if (attr) {
 857                         while (attr->attr.name) {
 858                                 if (sysfs_create_file(&e->kobj, &attr->attr))
 859                                         break;
 860                                 attr++;
 861                         }
 862                 }
 863                 kobject_uevent(&e->kobj, KOBJ_ADD);
 864                 e->registered = 1;
 865                 if (!e->uses_mq && e->type->ops.sq.elevator_registered_fn)
 866                         e->type->ops.sq.elevator_registered_fn(q);
 867         }
 868         return error;
 869 }
 870 EXPORT_SYMBOL(elv_register_queue);
 871
 872 void elv_unregister_queue(struct request_queue *q)
 873 {
 874         if (q) {
 875                 struct elevator_queue *e = q->elevator;
 876
 877                 kobject_uevent(&e->kobj, KOBJ_REMOVE);
 878                 kobject_del(&e->kobj);
 879                 e->registered = 0;
 880         }
 881 }
 882 EXPORT_SYMBOL(elv_unregister_queue);
 883
 884 int elv_register(struct elevator_type *e)
 885 {
 886         char *def = "";
 887
 888         /* create icq_cache if requested */
 889         if (e->icq_size) {
 890                 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
 891                     WARN_ON(e->icq_align < __alignof__(struct io_cq)))
 892                         return -EINVAL;
 893
 894                 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
 895                          "%s_io_cq", e->elevator_name);
 896                 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
 897                                                  e->icq_align, 0, NULL);
 898                 if (!e->icq_cache)
 899                         return -ENOMEM;
 900         }
 901
 902         /* register, don't allow duplicate names */
 903         spin_lock(&elv_list_lock);
 904         if (elevator_find(e->elevator_name)) {
 905                 spin_unlock(&elv_list_lock);
 906                 if (e->icq_cache)
 907                         kmem_cache_destroy(e->icq_cache);
 908                 return -EBUSY;
 909         }
 910         list_add_tail(&e->list, &elv_list);
 911         spin_unlock(&elv_list_lock);
 912
 913         /* print pretty message */
 914         if (!strcmp(e->elevator_name, chosen_elevator) ||
 915                         (!*chosen_elevator &&
 916                          !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
 917                                 def = " (default)";
 918
 919         printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
 920                                                                 def);
 921         return 0;
 922 }
 923 EXPORT_SYMBOL_GPL(elv_register);
 924
 925 void elv_unregister(struct elevator_type *e)
 926 {
 927         /* unregister */
 928         spin_lock(&elv_list_lock);
 929         list_del_init(&e->list);
 930         spin_unlock(&elv_list_lock);
 931
 932         /*
 933          * Destroy icq_cache if it exists.  icq's are RCU managed.  Make
 934          * sure all RCU operations are complete before proceeding.
 935          */
 936         if (e->icq_cache) {
 937                 rcu_barrier();
 938                 kmem_cache_destroy(e->icq_cache);
 939                 e->icq_cache = NULL;
 940         }
 941 }
 942 EXPORT_SYMBOL_GPL(elv_unregister);
 943
 944 static int elevator_switch_mq(struct request_queue *q,
 945                               struct elevator_type *new_e)
 946 {
 947         int ret;
 948
 949         blk_mq_freeze_queue(q);
 950         blk_mq_quiesce_queue(q);
 951
 952         if (q->elevator) {
 953                 if (q->elevator->registered)
 954                         elv_unregister_queue(q);
 955                 ioc_clear_queue(q);
 956                 elevator_exit(q, q->elevator);
 957         }
 958
 959         ret = blk_mq_init_sched(q, new_e);
 960         if (ret)
 961                 goto out;
 962
 963         if (new_e) {
 964                 ret = elv_register_queue(q);
 965                 if (ret) {
 966                         elevator_exit(q, q->elevator);
 967                         goto out;
 968                 }
 969         }
 970
 971         if (new_e)
 972                 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
 973         else
 974                 blk_add_trace_msg(q, "elv switch: none");
 975
 976 out:
 977         blk_mq_unfreeze_queue(q);
 978         blk_mq_start_stopped_hw_queues(q, true);
 979         return ret;
 980
 981 }
 982
 983 /*
 984  * switch to new_e io scheduler. be careful not to introduce deadlocks -
 985  * we don't free the old io scheduler, before we have allocated what we
 986  * need for the new one. this way we have a chance of going back to the old
 987  * one, if the new one fails init for some reason.
 988  */
 989 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
 990 {
 991         struct elevator_queue *old = q->elevator;
 992         bool old_registered = false;
 993         int err;
 994
 995         if (q->mq_ops)
 996                 return elevator_switch_mq(q, new_e);
 997
 998         /*
 999          * Turn on BYPASS and drain all requests w/ elevator private data.
1000          * Block layer doesn't call into a quiesced elevator - all requests
1001          * are directly put on the dispatch list without elevator data
1002          * using INSERT_BACK.  All requests have SOFTBARRIER set and no
1003          * merge happens either.
1004          */
1005         if (old) {
1006                 old_registered = old->registered;
1007
1008                 blk_queue_bypass_start(q);
1009
1010                 /* unregister and clear all auxiliary data of the old elevator */
1011                 if (old_registered)
1012                         elv_unregister_queue(q);
1013
1014                 ioc_clear_queue(q);
1015         }
1016
1017         /* allocate, init and register new elevator */
1018         err = new_e->ops.sq.elevator_init_fn(q, new_e);
1019         if (err)
1020                 goto fail_init;
1021
1022         err = elv_register_queue(q);
1023         if (err)
1024                 goto fail_register;
1025
1026         /* done, kill the old one and finish */
1027         if (old) {
1028                 elevator_exit(q, old);
1029                 blk_queue_bypass_end(q);
1030         }
1031
1032         blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
1033
1034         return 0;
1035
1036 fail_register:
1037         elevator_exit(q, q->elevator);
1038 fail_init:
1039         /* switch failed, restore and re-register old elevator */
1040         if (old) {
1041                 q->elevator = old;
1042                 elv_register_queue(q);
1043                 blk_queue_bypass_end(q);
1044         }
1045
1046         return err;
1047 }
1048
1049 /*
1050  * Switch this queue to the given IO scheduler.
1051  */
1052 static int __elevator_change(struct request_queue *q, const char *name)
1053 {
1054         char elevator_name[ELV_NAME_MAX];
1055         struct elevator_type *e;
1056
1057         /*
1058          * Special case for mq, turn off scheduling
1059          */
1060         if (q->mq_ops && !strncmp(name, "none", 4))
1061                 return elevator_switch(q, NULL);
1062
1063         strlcpy(elevator_name, name, sizeof(elevator_name));
1064         e = elevator_get(strstrip(elevator_name), true);
1065         if (!e) {
1066                 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1067                 return -EINVAL;
1068         }
1069
1070         if (q->elevator &&
1071             !strcmp(elevator_name, q->elevator->type->elevator_name)) {
1072                 elevator_put(e);
1073                 return 0;
1074         }
1075
1076         if (!e->uses_mq && q->mq_ops) {
1077                 elevator_put(e);
1078                 return -EINVAL;
1079         }
1080         if (e->uses_mq && !q->mq_ops) {
1081                 elevator_put(e);
1082                 return -EINVAL;
1083         }
1084
1085         return elevator_switch(q, e);
1086 }
1087
1088 int elevator_change(struct request_queue *q, const char *name)
1089 {
1090         int ret;
1091
1092         /* Protect q->elevator from elevator_init() */
1093         mutex_lock(&q->sysfs_lock);
1094         ret = __elevator_change(q, name);
1095         mutex_unlock(&q->sysfs_lock);
1096
1097         return ret;
1098 }
1099 EXPORT_SYMBOL(elevator_change);
1100
1101 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1102                           size_t count)
1103 {
1104         int ret;
1105
1106         if (!(q->mq_ops || q->request_fn))
1107                 return count;
1108
1109         ret = __elevator_change(q, name);
1110         if (!ret)
1111                 return count;
1112
1113         printk(KERN_ERR "elevator: switch to %s failed\n", name);
1114         return ret;
1115 }
1116
1117 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1118 {
1119         struct elevator_queue *e = q->elevator;
1120         struct elevator_type *elv = NULL;
1121         struct elevator_type *__e;
1122         int len = 0;
1123
1124         if (!blk_queue_stackable(q))
1125                 return sprintf(name, "none\n");
1126
1127         if (!q->elevator)
1128                 len += sprintf(name+len, "[none] ");
1129         else
1130                 elv = e->type;
1131
1132         spin_lock(&elv_list_lock);
1133         list_for_each_entry(__e, &elv_list, list) {
1134                 if (elv && !strcmp(elv->elevator_name, __e->elevator_name)) {
1135                         len += sprintf(name+len, "[%s] ", elv->elevator_name);
1136                         continue;
1137                 }
1138                 if (__e->uses_mq && q->mq_ops)
1139                         len += sprintf(name+len, "%s ", __e->elevator_name);
1140                 else if (!__e->uses_mq && !q->mq_ops)
1141                         len += sprintf(name+len, "%s ", __e->elevator_name);
1142         }
1143         spin_unlock(&elv_list_lock);
1144
1145         if (q->mq_ops && q->elevator)
1146                 len += sprintf(name+len, "none");
1147
1148         len += sprintf(len+name, "\n");
1149         return len;
1150 }
1151
1152 struct request *elv_rb_former_request(struct request_queue *q,
1153                                       struct request *rq)
1154 {
1155         struct rb_node *rbprev = rb_prev(&rq->rb_node);
1156
1157         if (rbprev)
1158                 return rb_entry_rq(rbprev);
1159
1160         return NULL;
1161 }
1162 EXPORT_SYMBOL(elv_rb_former_request);
1163
1164 struct request *elv_rb_latter_request(struct request_queue *q,
1165                                       struct request *rq)
1166 {
1167         struct rb_node *rbnext = rb_next(&rq->rb_node);
1168
1169         if (rbnext)
1170                 return rb_entry_rq(rbnext);
1171
1172         return NULL;
1173 }
1174 EXPORT_SYMBOL(elv_rb_latter_request);