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
  38 #include <trace/events/block.h>
  39
  40 #include "blk.h"
  41
  42 static DEFINE_SPINLOCK(elv_list_lock);
  43 static LIST_HEAD(elv_list);
  44
  45 /*
  46  * Merge hash stuff.
  47  */
  48 static const int elv_hash_shift = 6;
  49 #define ELV_HASH_BLOCK(sec)     ((sec) >> 3)
  50 #define ELV_HASH_FN(sec)        \
  51                 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
  52 #define ELV_HASH_ENTRIES        (1 << elv_hash_shift)
  53 #define rq_hash_key(rq)         (blk_rq_pos(rq) + blk_rq_sectors(rq))
  54
  55 /*
  56  * Query io scheduler to see if the current process issuing bio may be
  57  * merged with rq.
  58  */
  59 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
  60 {
  61         struct request_queue *q = rq->q;
  62         struct elevator_queue *e = q->elevator;
  63
  64         if (e->type->ops.elevator_allow_merge_fn)
  65                 return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
  66
  67         return 1;
  68 }
  69
  70 /*
  71  * can we safely merge with this request?
  72  */
  73 bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
  74 {
  75         if (!blk_rq_merge_ok(rq, bio))
  76                 return 0;
  77
  78         if (!elv_iosched_allow_merge(rq, bio))
  79                 return 0;
  80
  81         return 1;
  82 }
  83 EXPORT_SYMBOL(elv_rq_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)
 103 {
 104         struct elevator_type *e;
 105
 106         spin_lock(&elv_list_lock);
 107
 108         e = elevator_find(name);
 109         if (!e) {
 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 int elevator_init_queue(struct request_queue *q,
 125                                struct elevator_queue *eq)
 126 {
 127         eq->elevator_data = eq->type->ops.elevator_init_fn(q);
 128         if (eq->elevator_data)
 129                 return 0;
 130         return -ENOMEM;
 131 }
 132
 133 static char chosen_elevator[ELV_NAME_MAX];
 134
 135 static int __init elevator_setup(char *str)
 136 {
 137         /*
 138          * Be backwards-compatible with previous kernels, so users
 139          * won't get the wrong elevator.
 140          */
 141         strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
 142         return 1;
 143 }
 144
 145 __setup("elevator=", elevator_setup);
 146
 147 static struct kobj_type elv_ktype;
 148
 149 static struct elevator_queue *elevator_alloc(struct request_queue *q,
 150                                   struct elevator_type *e)
 151 {
 152         struct elevator_queue *eq;
 153         int i;
 154
 155         eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
 156         if (unlikely(!eq))
 157                 goto err;
 158
 159         eq->type = e;
 160         kobject_init(&eq->kobj, &elv_ktype);
 161         mutex_init(&eq->sysfs_lock);
 162
 163         eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
 164                                         GFP_KERNEL, q->node);
 165         if (!eq->hash)
 166                 goto err;
 167
 168         for (i = 0; i < ELV_HASH_ENTRIES; i++)
 169                 INIT_HLIST_HEAD(&eq->hash[i]);
 170
 171         return eq;
 172 err:
 173         kfree(eq);
 174         elevator_put(e);
 175         return NULL;
 176 }
 177
 178 static void elevator_release(struct kobject *kobj)
 179 {
 180         struct elevator_queue *e;
 181
 182         e = container_of(kobj, struct elevator_queue, kobj);
 183         elevator_put(e->type);
 184         kfree(e->hash);
 185         kfree(e);
 186 }
 187
 188 int elevator_init(struct request_queue *q, char *name)
 189 {
 190         struct elevator_type *e = NULL;
 191         struct elevator_queue *eq;
 192         int err;
 193
 194         if (unlikely(q->elevator))
 195                 return 0;
 196
 197         INIT_LIST_HEAD(&q->queue_head);
 198         q->last_merge = NULL;
 199         q->end_sector = 0;
 200         q->boundary_rq = NULL;
 201
 202         if (name) {
 203                 e = elevator_get(name);
 204                 if (!e)
 205                         return -EINVAL;
 206         }
 207
 208         if (!e && *chosen_elevator) {
 209                 e = elevator_get(chosen_elevator);
 210                 if (!e)
 211                         printk(KERN_ERR "I/O scheduler %s not found\n",
 212                                                         chosen_elevator);
 213         }
 214
 215         if (!e) {
 216                 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
 217                 if (!e) {
 218                         printk(KERN_ERR
 219                                 "Default I/O scheduler not found. " \
 220                                 "Using noop.\n");
 221                         e = elevator_get("noop");
 222                 }
 223         }
 224
 225         eq = elevator_alloc(q, e);
 226         if (!eq)
 227                 return -ENOMEM;
 228
 229         err = elevator_init_queue(q, eq);
 230         if (err) {
 231                 kobject_put(&eq->kobj);
 232                 return err;
 233         }
 234
 235         q->elevator = eq;
 236         return 0;
 237 }
 238 EXPORT_SYMBOL(elevator_init);
 239
 240 void elevator_exit(struct elevator_queue *e)
 241 {
 242         mutex_lock(&e->sysfs_lock);
 243         if (e->type->ops.elevator_exit_fn)
 244                 e->type->ops.elevator_exit_fn(e);
 245         mutex_unlock(&e->sysfs_lock);
 246
 247         kobject_put(&e->kobj);
 248 }
 249 EXPORT_SYMBOL(elevator_exit);
 250
 251 static inline void __elv_rqhash_del(struct request *rq)
 252 {
 253         hlist_del_init(&rq->hash);
 254 }
 255
 256 static void elv_rqhash_del(struct request_queue *q, struct request *rq)
 257 {
 258         if (ELV_ON_HASH(rq))
 259                 __elv_rqhash_del(rq);
 260 }
 261
 262 static void elv_rqhash_add(struct request_queue *q, struct request *rq)
 263 {
 264         struct elevator_queue *e = q->elevator;
 265
 266         BUG_ON(ELV_ON_HASH(rq));
 267         hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
 268 }
 269
 270 static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
 271 {
 272         __elv_rqhash_del(rq);
 273         elv_rqhash_add(q, rq);
 274 }
 275
 276 static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
 277 {
 278         struct elevator_queue *e = q->elevator;
 279         struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
 280         struct hlist_node *entry, *next;
 281         struct request *rq;
 282
 283         hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
 284                 BUG_ON(!ELV_ON_HASH(rq));
 285
 286                 if (unlikely(!rq_mergeable(rq))) {
 287                         __elv_rqhash_del(rq);
 288                         continue;
 289                 }
 290
 291                 if (rq_hash_key(rq) == offset)
 292                         return rq;
 293         }
 294
 295         return NULL;
 296 }
 297
 298 /*
 299  * RB-tree support functions for inserting/lookup/removal of requests
 300  * in a sorted RB tree.
 301  */
 302 void elv_rb_add(struct rb_root *root, struct request *rq)
 303 {
 304         struct rb_node **p = &root->rb_node;
 305         struct rb_node *parent = NULL;
 306         struct request *__rq;
 307
 308         while (*p) {
 309                 parent = *p;
 310                 __rq = rb_entry(parent, struct request, rb_node);
 311
 312                 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
 313                         p = &(*p)->rb_left;
 314                 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
 315                         p = &(*p)->rb_right;
 316         }
 317
 318         rb_link_node(&rq->rb_node, parent, p);
 319         rb_insert_color(&rq->rb_node, root);
 320 }
 321 EXPORT_SYMBOL(elv_rb_add);
 322
 323 void elv_rb_del(struct rb_root *root, struct request *rq)
 324 {
 325         BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
 326         rb_erase(&rq->rb_node, root);
 327         RB_CLEAR_NODE(&rq->rb_node);
 328 }
 329 EXPORT_SYMBOL(elv_rb_del);
 330
 331 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
 332 {
 333         struct rb_node *n = root->rb_node;
 334         struct request *rq;
 335
 336         while (n) {
 337                 rq = rb_entry(n, struct request, rb_node);
 338
 339                 if (sector < blk_rq_pos(rq))
 340                         n = n->rb_left;
 341                 else if (sector > blk_rq_pos(rq))
 342                         n = n->rb_right;
 343                 else
 344                         return rq;
 345         }
 346
 347         return NULL;
 348 }
 349 EXPORT_SYMBOL(elv_rb_find);
 350
 351 /*
 352  * Insert rq into dispatch queue of q.  Queue lock must be held on
 353  * entry.  rq is sort instead into the dispatch queue. To be used by
 354  * specific elevators.
 355  */
 356 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
 357 {
 358         sector_t boundary;
 359         struct list_head *entry;
 360         int stop_flags;
 361
 362         if (q->last_merge == rq)
 363                 q->last_merge = NULL;
 364
 365         elv_rqhash_del(q, rq);
 366
 367         q->nr_sorted--;
 368
 369         boundary = q->end_sector;
 370         stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
 371         list_for_each_prev(entry, &q->queue_head) {
 372                 struct request *pos = list_entry_rq(entry);
 373
 374                 if ((rq->cmd_flags & REQ_DISCARD) !=
 375                     (pos->cmd_flags & REQ_DISCARD))
 376                         break;
 377                 if (rq_data_dir(rq) != rq_data_dir(pos))
 378                         break;
 379                 if (pos->cmd_flags & stop_flags)
 380                         break;
 381                 if (blk_rq_pos(rq) >= boundary) {
 382                         if (blk_rq_pos(pos) < boundary)
 383                                 continue;
 384                 } else {
 385                         if (blk_rq_pos(pos) >= boundary)
 386                                 break;
 387                 }
 388                 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
 389                         break;
 390         }
 391
 392         list_add(&rq->queuelist, entry);
 393 }
 394 EXPORT_SYMBOL(elv_dispatch_sort);
 395
 396 /*
 397  * Insert rq into dispatch queue of q.  Queue lock must be held on
 398  * entry.  rq is added to the back of the dispatch queue. To be used by
 399  * specific elevators.
 400  */
 401 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
 402 {
 403         if (q->last_merge == rq)
 404                 q->last_merge = NULL;
 405
 406         elv_rqhash_del(q, rq);
 407
 408         q->nr_sorted--;
 409
 410         q->end_sector = rq_end_sector(rq);
 411         q->boundary_rq = rq;
 412         list_add_tail(&rq->queuelist, &q->queue_head);
 413 }
 414 EXPORT_SYMBOL(elv_dispatch_add_tail);
 415
 416 int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
 417 {
 418         struct elevator_queue *e = q->elevator;
 419         struct request *__rq;
 420         int ret;
 421
 422         /*
 423          * Levels of merges:
 424          *      nomerges:  No merges at all attempted
 425          *      noxmerges: Only simple one-hit cache try
 426          *      merges:    All merge tries attempted
 427          */
 428         if (blk_queue_nomerges(q))
 429                 return ELEVATOR_NO_MERGE;
 430
 431         /*
 432          * First try one-hit cache.
 433          */
 434         if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
 435                 ret = blk_try_merge(q->last_merge, bio);
 436                 if (ret != ELEVATOR_NO_MERGE) {
 437                         *req = q->last_merge;
 438                         return ret;
 439                 }
 440         }
 441
 442         if (blk_queue_noxmerges(q))
 443                 return ELEVATOR_NO_MERGE;
 444
 445         /*
 446          * See if our hash lookup can find a potential backmerge.
 447          */
 448         __rq = elv_rqhash_find(q, bio->bi_sector);
 449         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 450                 *req = __rq;
 451                 return ELEVATOR_BACK_MERGE;
 452         }
 453
 454         if (e->type->ops.elevator_merge_fn)
 455                 return e->type->ops.elevator_merge_fn(q, req, bio);
 456
 457         return ELEVATOR_NO_MERGE;
 458 }
 459
 460 /*
 461  * Attempt to do an insertion back merge. Only check for the case where
 462  * we can append 'rq' to an existing request, so we can throw 'rq' away
 463  * afterwards.
 464  *
 465  * Returns true if we merged, false otherwise
 466  */
 467 static bool elv_attempt_insert_merge(struct request_queue *q,
 468                                      struct request *rq)
 469 {
 470         struct request *__rq;
 471
 472         if (blk_queue_nomerges(q))
 473                 return false;
 474
 475         /*
 476          * First try one-hit cache.
 477          */
 478         if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
 479                 return true;
 480
 481         if (blk_queue_noxmerges(q))
 482                 return false;
 483
 484         /*
 485          * See if our hash lookup can find a potential backmerge.
 486          */
 487         __rq = elv_rqhash_find(q, blk_rq_pos(rq));
 488         if (__rq && blk_attempt_req_merge(q, __rq, rq))
 489                 return true;
 490
 491         return false;
 492 }
 493
 494 void elv_merged_request(struct request_queue *q, struct request *rq, int type)
 495 {
 496         struct elevator_queue *e = q->elevator;
 497
 498         if (e->type->ops.elevator_merged_fn)
 499                 e->type->ops.elevator_merged_fn(q, rq, type);
 500
 501         if (type == ELEVATOR_BACK_MERGE)
 502                 elv_rqhash_reposition(q, rq);
 503
 504         q->last_merge = rq;
 505 }
 506
 507 void elv_merge_requests(struct request_queue *q, struct request *rq,
 508                              struct request *next)
 509 {
 510         struct elevator_queue *e = q->elevator;
 511         const int next_sorted = next->cmd_flags & REQ_SORTED;
 512
 513         if (next_sorted && e->type->ops.elevator_merge_req_fn)
 514                 e->type->ops.elevator_merge_req_fn(q, rq, next);
 515
 516         elv_rqhash_reposition(q, rq);
 517
 518         if (next_sorted) {
 519                 elv_rqhash_del(q, next);
 520                 q->nr_sorted--;
 521         }
 522
 523         q->last_merge = rq;
 524 }
 525
 526 void elv_bio_merged(struct request_queue *q, struct request *rq,
 527                         struct bio *bio)
 528 {
 529         struct elevator_queue *e = q->elevator;
 530
 531         if (e->type->ops.elevator_bio_merged_fn)
 532                 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
 533 }
 534
 535 void elv_requeue_request(struct request_queue *q, struct request *rq)
 536 {
 537         /*
 538          * it already went through dequeue, we need to decrement the
 539          * in_flight count again
 540          */
 541         if (blk_account_rq(rq)) {
 542                 q->in_flight[rq_is_sync(rq)]--;
 543                 if (rq->cmd_flags & REQ_SORTED)
 544                         elv_deactivate_rq(q, rq);
 545         }
 546
 547         rq->cmd_flags &= ~REQ_STARTED;
 548
 549         __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
 550 }
 551
 552 void elv_drain_elevator(struct request_queue *q)
 553 {
 554         static int printed;
 555
 556         lockdep_assert_held(q->queue_lock);
 557
 558         while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
 559                 ;
 560         if (q->nr_sorted && printed++ < 10) {
 561                 printk(KERN_ERR "%s: forced dispatching is broken "
 562                        "(nr_sorted=%u), please report this\n",
 563                        q->elevator->type->elevator_name, q->nr_sorted);
 564         }
 565 }
 566
 567 void elv_quiesce_start(struct request_queue *q)
 568 {
 569         if (!q->elevator)
 570                 return;
 571
 572         spin_lock_irq(q->queue_lock);
 573         queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
 574         spin_unlock_irq(q->queue_lock);
 575
 576         blk_drain_queue(q, false);
 577 }
 578
 579 void elv_quiesce_end(struct request_queue *q)
 580 {
 581         spin_lock_irq(q->queue_lock);
 582         queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
 583         spin_unlock_irq(q->queue_lock);
 584 }
 585
 586 void __elv_add_request(struct request_queue *q, struct request *rq, int where)
 587 {
 588         trace_block_rq_insert(q, rq);
 589
 590         rq->q = q;
 591
 592         if (rq->cmd_flags & REQ_SOFTBARRIER) {
 593                 /* barriers are scheduling boundary, update end_sector */
 594                 if (rq->cmd_type == REQ_TYPE_FS ||
 595                     (rq->cmd_flags & REQ_DISCARD)) {
 596                         q->end_sector = rq_end_sector(rq);
 597                         q->boundary_rq = rq;
 598                 }
 599         } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
 600                     (where == ELEVATOR_INSERT_SORT ||
 601                      where == ELEVATOR_INSERT_SORT_MERGE))
 602                 where = ELEVATOR_INSERT_BACK;
 603
 604         switch (where) {
 605         case ELEVATOR_INSERT_REQUEUE:
 606         case ELEVATOR_INSERT_FRONT:
 607                 rq->cmd_flags |= REQ_SOFTBARRIER;
 608                 list_add(&rq->queuelist, &q->queue_head);
 609                 break;
 610
 611         case ELEVATOR_INSERT_BACK:
 612                 rq->cmd_flags |= REQ_SOFTBARRIER;
 613                 elv_drain_elevator(q);
 614                 list_add_tail(&rq->queuelist, &q->queue_head);
 615                 /*
 616                  * We kick the queue here for the following reasons.
 617                  * - The elevator might have returned NULL previously
 618                  *   to delay requests and returned them now.  As the
 619                  *   queue wasn't empty before this request, ll_rw_blk
 620                  *   won't run the queue on return, resulting in hang.
 621                  * - Usually, back inserted requests won't be merged
 622                  *   with anything.  There's no point in delaying queue
 623                  *   processing.
 624                  */
 625                 __blk_run_queue(q);
 626                 break;
 627
 628         case ELEVATOR_INSERT_SORT_MERGE:
 629                 /*
 630                  * If we succeed in merging this request with one in the
 631                  * queue already, we are done - rq has now been freed,
 632                  * so no need to do anything further.
 633                  */
 634                 if (elv_attempt_insert_merge(q, rq))
 635                         break;
 636         case ELEVATOR_INSERT_SORT:
 637                 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
 638                        !(rq->cmd_flags & REQ_DISCARD));
 639                 rq->cmd_flags |= REQ_SORTED;
 640                 q->nr_sorted++;
 641                 if (rq_mergeable(rq)) {
 642                         elv_rqhash_add(q, rq);
 643                         if (!q->last_merge)
 644                                 q->last_merge = rq;
 645                 }
 646
 647                 /*
 648                  * Some ioscheds (cfq) run q->request_fn directly, so
 649                  * rq cannot be accessed after calling
 650                  * elevator_add_req_fn.
 651                  */
 652                 q->elevator->type->ops.elevator_add_req_fn(q, rq);
 653                 break;
 654
 655         case ELEVATOR_INSERT_FLUSH:
 656                 rq->cmd_flags |= REQ_SOFTBARRIER;
 657                 blk_insert_flush(rq);
 658                 break;
 659         default:
 660                 printk(KERN_ERR "%s: bad insertion point %d\n",
 661                        __func__, where);
 662                 BUG();
 663         }
 664 }
 665 EXPORT_SYMBOL(__elv_add_request);
 666
 667 void elv_add_request(struct request_queue *q, struct request *rq, int where)
 668 {
 669         unsigned long flags;
 670
 671         spin_lock_irqsave(q->queue_lock, flags);
 672         __elv_add_request(q, rq, where);
 673         spin_unlock_irqrestore(q->queue_lock, flags);
 674 }
 675 EXPORT_SYMBOL(elv_add_request);
 676
 677 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
 678 {
 679         struct elevator_queue *e = q->elevator;
 680
 681         if (e->type->ops.elevator_latter_req_fn)
 682                 return e->type->ops.elevator_latter_req_fn(q, rq);
 683         return NULL;
 684 }
 685
 686 struct request *elv_former_request(struct request_queue *q, struct request *rq)
 687 {
 688         struct elevator_queue *e = q->elevator;
 689
 690         if (e->type->ops.elevator_former_req_fn)
 691                 return e->type->ops.elevator_former_req_fn(q, rq);
 692         return NULL;
 693 }
 694
 695 int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
 696 {
 697         struct elevator_queue *e = q->elevator;
 698
 699         if (e->type->ops.elevator_set_req_fn)
 700                 return e->type->ops.elevator_set_req_fn(q, rq, gfp_mask);
 701         return 0;
 702 }
 703
 704 void elv_put_request(struct request_queue *q, struct request *rq)
 705 {
 706         struct elevator_queue *e = q->elevator;
 707
 708         if (e->type->ops.elevator_put_req_fn)
 709                 e->type->ops.elevator_put_req_fn(rq);
 710 }
 711
 712 int elv_may_queue(struct request_queue *q, int rw)
 713 {
 714         struct elevator_queue *e = q->elevator;
 715
 716         if (e->type->ops.elevator_may_queue_fn)
 717                 return e->type->ops.elevator_may_queue_fn(q, rw);
 718
 719         return ELV_MQUEUE_MAY;
 720 }
 721
 722 void elv_abort_queue(struct request_queue *q)
 723 {
 724         struct request *rq;
 725
 726         blk_abort_flushes(q);
 727
 728         while (!list_empty(&q->queue_head)) {
 729                 rq = list_entry_rq(q->queue_head.next);
 730                 rq->cmd_flags |= REQ_QUIET;
 731                 trace_block_rq_abort(q, rq);
 732                 /*
 733                  * Mark this request as started so we don't trigger
 734                  * any debug logic in the end I/O path.
 735                  */
 736                 blk_start_request(rq);
 737                 __blk_end_request_all(rq, -EIO);
 738         }
 739 }
 740 EXPORT_SYMBOL(elv_abort_queue);
 741
 742 void elv_completed_request(struct request_queue *q, struct request *rq)
 743 {
 744         struct elevator_queue *e = q->elevator;
 745
 746         /*
 747          * request is released from the driver, io must be done
 748          */
 749         if (blk_account_rq(rq)) {
 750                 q->in_flight[rq_is_sync(rq)]--;
 751                 if ((rq->cmd_flags & REQ_SORTED) &&
 752                     e->type->ops.elevator_completed_req_fn)
 753                         e->type->ops.elevator_completed_req_fn(q, rq);
 754         }
 755 }
 756
 757 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
 758
 759 static ssize_t
 760 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 761 {
 762         struct elv_fs_entry *entry = to_elv(attr);
 763         struct elevator_queue *e;
 764         ssize_t error;
 765
 766         if (!entry->show)
 767                 return -EIO;
 768
 769         e = container_of(kobj, struct elevator_queue, kobj);
 770         mutex_lock(&e->sysfs_lock);
 771         error = e->type ? entry->show(e, page) : -ENOENT;
 772         mutex_unlock(&e->sysfs_lock);
 773         return error;
 774 }
 775
 776 static ssize_t
 777 elv_attr_store(struct kobject *kobj, struct attribute *attr,
 778                const char *page, size_t length)
 779 {
 780         struct elv_fs_entry *entry = to_elv(attr);
 781         struct elevator_queue *e;
 782         ssize_t error;
 783
 784         if (!entry->store)
 785                 return -EIO;
 786
 787         e = container_of(kobj, struct elevator_queue, kobj);
 788         mutex_lock(&e->sysfs_lock);
 789         error = e->type ? entry->store(e, page, length) : -ENOENT;
 790         mutex_unlock(&e->sysfs_lock);
 791         return error;
 792 }
 793
 794 static const struct sysfs_ops elv_sysfs_ops = {
 795         .show   = elv_attr_show,
 796         .store  = elv_attr_store,
 797 };
 798
 799 static struct kobj_type elv_ktype = {
 800         .sysfs_ops      = &elv_sysfs_ops,
 801         .release        = elevator_release,
 802 };
 803
 804 int __elv_register_queue(struct request_queue *q, struct elevator_queue *e)
 805 {
 806         int error;
 807
 808         error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
 809         if (!error) {
 810                 struct elv_fs_entry *attr = e->type->elevator_attrs;
 811                 if (attr) {
 812                         while (attr->attr.name) {
 813                                 if (sysfs_create_file(&e->kobj, &attr->attr))
 814                                         break;
 815                                 attr++;
 816                         }
 817                 }
 818                 kobject_uevent(&e->kobj, KOBJ_ADD);
 819                 e->registered = 1;
 820         }
 821         return error;
 822 }
 823
 824 int elv_register_queue(struct request_queue *q)
 825 {
 826         return __elv_register_queue(q, q->elevator);
 827 }
 828 EXPORT_SYMBOL(elv_register_queue);
 829
 830 void elv_unregister_queue(struct request_queue *q)
 831 {
 832         if (q) {
 833                 struct elevator_queue *e = q->elevator;
 834
 835                 kobject_uevent(&e->kobj, KOBJ_REMOVE);
 836                 kobject_del(&e->kobj);
 837                 e->registered = 0;
 838         }
 839 }
 840 EXPORT_SYMBOL(elv_unregister_queue);
 841
 842 int elv_register(struct elevator_type *e)
 843 {
 844         char *def = "";
 845
 846         /* create icq_cache if requested */
 847         if (e->icq_size) {
 848                 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
 849                     WARN_ON(e->icq_align < __alignof__(struct io_cq)))
 850                         return -EINVAL;
 851
 852                 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
 853                          "%s_io_cq", e->elevator_name);
 854                 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
 855                                                  e->icq_align, 0, NULL);
 856                 if (!e->icq_cache)
 857                         return -ENOMEM;
 858         }
 859
 860         /* register, don't allow duplicate names */
 861         spin_lock(&elv_list_lock);
 862         if (elevator_find(e->elevator_name)) {
 863                 spin_unlock(&elv_list_lock);
 864                 if (e->icq_cache)
 865                         kmem_cache_destroy(e->icq_cache);
 866                 return -EBUSY;
 867         }
 868         list_add_tail(&e->list, &elv_list);
 869         spin_unlock(&elv_list_lock);
 870
 871         /* print pretty message */
 872         if (!strcmp(e->elevator_name, chosen_elevator) ||
 873                         (!*chosen_elevator &&
 874                          !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
 875                                 def = " (default)";
 876
 877         printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
 878                                                                 def);
 879         return 0;
 880 }
 881 EXPORT_SYMBOL_GPL(elv_register);
 882
 883 void elv_unregister(struct elevator_type *e)
 884 {
 885         /* unregister */
 886         spin_lock(&elv_list_lock);
 887         list_del_init(&e->list);
 888         spin_unlock(&elv_list_lock);
 889
 890         /*
 891          * Destroy icq_cache if it exists.  icq's are RCU managed.  Make
 892          * sure all RCU operations are complete before proceeding.
 893          */
 894         if (e->icq_cache) {
 895                 rcu_barrier();
 896                 kmem_cache_destroy(e->icq_cache);
 897                 e->icq_cache = NULL;
 898         }
 899 }
 900 EXPORT_SYMBOL_GPL(elv_unregister);
 901
 902 /*
 903  * switch to new_e io scheduler. be careful not to introduce deadlocks -
 904  * we don't free the old io scheduler, before we have allocated what we
 905  * need for the new one. this way we have a chance of going back to the old
 906  * one, if the new one fails init for some reason.
 907  */
 908 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
 909 {
 910         struct elevator_queue *old_elevator, *e;
 911         int err;
 912
 913         /* allocate new elevator */
 914         e = elevator_alloc(q, new_e);
 915         if (!e)
 916                 return -ENOMEM;
 917
 918         err = elevator_init_queue(q, e);
 919         if (err) {
 920                 kobject_put(&e->kobj);
 921                 return err;
 922         }
 923
 924         /* turn on BYPASS and drain all requests w/ elevator private data */
 925         elv_quiesce_start(q);
 926
 927         /* unregister old queue, register new one and kill old elevator */
 928         if (q->elevator->registered) {
 929                 elv_unregister_queue(q);
 930                 err = __elv_register_queue(q, e);
 931                 if (err)
 932                         goto fail_register;
 933         }
 934
 935         /* done, clear io_cq's, switch elevators and turn off BYPASS */
 936         spin_lock_irq(q->queue_lock);
 937         ioc_clear_queue(q);
 938         old_elevator = q->elevator;
 939         q->elevator = e;
 940         spin_unlock_irq(q->queue_lock);
 941
 942         elevator_exit(old_elevator);
 943         elv_quiesce_end(q);
 944
 945         blk_add_trace_msg(q, "elv switch: %s", e->type->elevator_name);
 946
 947         return 0;
 948
 949 fail_register:
 950         /*
 951          * switch failed, exit the new io scheduler and reattach the old
 952          * one again (along with re-adding the sysfs dir)
 953          */
 954         elevator_exit(e);
 955         elv_register_queue(q);
 956         elv_quiesce_end(q);
 957
 958         return err;
 959 }
 960
 961 /*
 962  * Switch this queue to the given IO scheduler.
 963  */
 964 int elevator_change(struct request_queue *q, const char *name)
 965 {
 966         char elevator_name[ELV_NAME_MAX];
 967         struct elevator_type *e;
 968
 969         if (!q->elevator)
 970                 return -ENXIO;
 971
 972         strlcpy(elevator_name, name, sizeof(elevator_name));
 973         e = elevator_get(strstrip(elevator_name));
 974         if (!e) {
 975                 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
 976                 return -EINVAL;
 977         }
 978
 979         if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
 980                 elevator_put(e);
 981                 return 0;
 982         }
 983
 984         return elevator_switch(q, e);
 985 }
 986 EXPORT_SYMBOL(elevator_change);
 987
 988 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
 989                           size_t count)
 990 {
 991         int ret;
 992
 993         if (!q->elevator)
 994                 return count;
 995
 996         ret = elevator_change(q, name);
 997         if (!ret)
 998                 return count;
 999
1000         printk(KERN_ERR "elevator: switch to %s failed\n", name);
1001         return ret;
1002 }
1003
1004 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1005 {
1006         struct elevator_queue *e = q->elevator;
1007         struct elevator_type *elv;
1008         struct elevator_type *__e;
1009         int len = 0;
1010
1011         if (!q->elevator || !blk_queue_stackable(q))
1012                 return sprintf(name, "none\n");
1013
1014         elv = e->type;
1015
1016         spin_lock(&elv_list_lock);
1017         list_for_each_entry(__e, &elv_list, list) {
1018                 if (!strcmp(elv->elevator_name, __e->elevator_name))
1019                         len += sprintf(name+len, "[%s] ", elv->elevator_name);
1020                 else
1021                         len += sprintf(name+len, "%s ", __e->elevator_name);
1022         }
1023         spin_unlock(&elv_list_lock);
1024
1025         len += sprintf(len+name, "\n");
1026         return len;
1027 }
1028
1029 struct request *elv_rb_former_request(struct request_queue *q,
1030                                       struct request *rq)
1031 {
1032         struct rb_node *rbprev = rb_prev(&rq->rb_node);
1033
1034         if (rbprev)
1035                 return rb_entry_rq(rbprev);
1036
1037         return NULL;
1038 }
1039 EXPORT_SYMBOL(elv_rb_former_request);
1040
1041 struct request *elv_rb_latter_request(struct request_queue *q,
1042                                       struct request *rq)
1043 {
1044         struct rb_node *rbnext = rb_next(&rq->rb_node);
1045
1046         if (rbnext)
1047                 return rb_entry_rq(rbnext);
1048
1049         return NULL;
1050 }
1051 EXPORT_SYMBOL(elv_rb_latter_request);