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/delay.h>
  35 #include <linux/blktrace_api.h>
  36 #include <linux/hash.h>
  37
  38 #include <asm/uaccess.h>
  39
  40 static DEFINE_SPINLOCK(elv_list_lock);
  41 static LIST_HEAD(elv_list);
  42
  43 /*
  44  * Merge hash stuff.
  45  */
  46 static const int elv_hash_shift = 6;
  47 #define ELV_HASH_BLOCK(sec)     ((sec) >> 3)
  48 #define ELV_HASH_FN(sec)        (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
  49 #define ELV_HASH_ENTRIES        (1 << elv_hash_shift)
  50 #define rq_hash_key(rq)         ((rq)->sector + (rq)->nr_sectors)
  51 #define ELV_ON_HASH(rq)         (!hlist_unhashed(&(rq)->hash))
  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_merge(struct request *rq, struct bio *bio)
  58 {
  59         request_queue_t *q = rq->q;
  60         elevator_t *e = q->elevator;
  61
  62         if (e->ops->elevator_allow_merge_fn)
  63                 return e->ops->elevator_allow_merge_fn(q, rq, bio);
  64
  65         return 1;
  66 }
  67
  68 /*
  69  * can we safely merge with this request?
  70  */
  71 inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
  72 {
  73         if (!rq_mergeable(rq))
  74                 return 0;
  75
  76         /*
  77          * different data direction or already started, don't merge
  78          */
  79         if (bio_data_dir(bio) != rq_data_dir(rq))
  80                 return 0;
  81
  82         /*
  83          * must be same device and not a special request
  84          */
  85         if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
  86                 return 0;
  87
  88         if (!elv_iosched_allow_merge(rq, bio))
  89                 return 0;
  90
  91         return 1;
  92 }
  93 EXPORT_SYMBOL(elv_rq_merge_ok);
  94
  95 static inline int elv_try_merge(struct request *__rq, struct bio *bio)
  96 {
  97         int ret = ELEVATOR_NO_MERGE;
  98
  99         /*
 100          * we can merge and sequence is ok, check if it's possible
 101          */
 102         if (elv_rq_merge_ok(__rq, bio)) {
 103                 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
 104                         ret = ELEVATOR_BACK_MERGE;
 105                 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
 106                         ret = ELEVATOR_FRONT_MERGE;
 107         }
 108
 109         return ret;
 110 }
 111
 112 static struct elevator_type *elevator_find(const char *name)
 113 {
 114         struct elevator_type *e;
 115         struct list_head *entry;
 116
 117         list_for_each(entry, &elv_list) {
 118
 119                 e = list_entry(entry, struct elevator_type, list);
 120
 121                 if (!strcmp(e->elevator_name, name))
 122                         return e;
 123         }
 124
 125         return NULL;
 126 }
 127
 128 static void elevator_put(struct elevator_type *e)
 129 {
 130         module_put(e->elevator_owner);
 131 }
 132
 133 static struct elevator_type *elevator_get(const char *name)
 134 {
 135         struct elevator_type *e;
 136
 137         spin_lock_irq(&elv_list_lock);
 138
 139         e = elevator_find(name);
 140         if (e && !try_module_get(e->elevator_owner))
 141                 e = NULL;
 142
 143         spin_unlock_irq(&elv_list_lock);
 144
 145         return e;
 146 }
 147
 148 static void *elevator_init_queue(request_queue_t *q, struct elevator_queue *eq)
 149 {
 150         return eq->ops->elevator_init_fn(q);
 151 }
 152
 153 static void elevator_attach(request_queue_t *q, struct elevator_queue *eq,
 154                            void *data)
 155 {
 156         q->elevator = eq;
 157         eq->elevator_data = data;
 158 }
 159
 160 static char chosen_elevator[16];
 161
 162 static int __init elevator_setup(char *str)
 163 {
 164         /*
 165          * Be backwards-compatible with previous kernels, so users
 166          * won't get the wrong elevator.
 167          */
 168         if (!strcmp(str, "as"))
 169                 strcpy(chosen_elevator, "anticipatory");
 170         else
 171                 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
 172         return 1;
 173 }
 174
 175 __setup("elevator=", elevator_setup);
 176
 177 static struct kobj_type elv_ktype;
 178
 179 static elevator_t *elevator_alloc(request_queue_t *q, struct elevator_type *e)
 180 {
 181         elevator_t *eq;
 182         int i;
 183
 184         eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL, q->node);
 185         if (unlikely(!eq))
 186                 goto err;
 187
 188         memset(eq, 0, sizeof(*eq));
 189         eq->ops = &e->ops;
 190         eq->elevator_type = e;
 191         kobject_init(&eq->kobj);
 192         snprintf(eq->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
 193         eq->kobj.ktype = &elv_ktype;
 194         mutex_init(&eq->sysfs_lock);
 195
 196         eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
 197                                         GFP_KERNEL, q->node);
 198         if (!eq->hash)
 199                 goto err;
 200
 201         for (i = 0; i < ELV_HASH_ENTRIES; i++)
 202                 INIT_HLIST_HEAD(&eq->hash[i]);
 203
 204         return eq;
 205 err:
 206         kfree(eq);
 207         elevator_put(e);
 208         return NULL;
 209 }
 210
 211 static void elevator_release(struct kobject *kobj)
 212 {
 213         elevator_t *e = container_of(kobj, elevator_t, kobj);
 214
 215         elevator_put(e->elevator_type);
 216         kfree(e->hash);
 217         kfree(e);
 218 }
 219
 220 int elevator_init(request_queue_t *q, char *name)
 221 {
 222         struct elevator_type *e = NULL;
 223         struct elevator_queue *eq;
 224         int ret = 0;
 225         void *data;
 226
 227         INIT_LIST_HEAD(&q->queue_head);
 228         q->last_merge = NULL;
 229         q->end_sector = 0;
 230         q->boundary_rq = NULL;
 231
 232         if (name && !(e = elevator_get(name)))
 233                 return -EINVAL;
 234
 235         if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator)))
 236                 printk("I/O scheduler %s not found\n", chosen_elevator);
 237
 238         if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) {
 239                 printk("Default I/O scheduler not found, using no-op\n");
 240                 e = elevator_get("noop");
 241         }
 242
 243         eq = elevator_alloc(q, e);
 244         if (!eq)
 245                 return -ENOMEM;
 246
 247         data = elevator_init_queue(q, eq);
 248         if (!data) {
 249                 kobject_put(&eq->kobj);
 250                 return -ENOMEM;
 251         }
 252
 253         elevator_attach(q, eq, data);
 254         return ret;
 255 }
 256
 257 EXPORT_SYMBOL(elevator_init);
 258
 259 void elevator_exit(elevator_t *e)
 260 {
 261         mutex_lock(&e->sysfs_lock);
 262         if (e->ops->elevator_exit_fn)
 263                 e->ops->elevator_exit_fn(e);
 264         e->ops = NULL;
 265         mutex_unlock(&e->sysfs_lock);
 266
 267         kobject_put(&e->kobj);
 268 }
 269
 270 EXPORT_SYMBOL(elevator_exit);
 271
 272 static void elv_activate_rq(request_queue_t *q, struct request *rq)
 273 {
 274         elevator_t *e = q->elevator;
 275
 276         if (e->ops->elevator_activate_req_fn)
 277                 e->ops->elevator_activate_req_fn(q, rq);
 278 }
 279
 280 static void elv_deactivate_rq(request_queue_t *q, struct request *rq)
 281 {
 282         elevator_t *e = q->elevator;
 283
 284         if (e->ops->elevator_deactivate_req_fn)
 285                 e->ops->elevator_deactivate_req_fn(q, rq);
 286 }
 287
 288 static inline void __elv_rqhash_del(struct request *rq)
 289 {
 290         hlist_del_init(&rq->hash);
 291 }
 292
 293 static void elv_rqhash_del(request_queue_t *q, struct request *rq)
 294 {
 295         if (ELV_ON_HASH(rq))
 296                 __elv_rqhash_del(rq);
 297 }
 298
 299 static void elv_rqhash_add(request_queue_t *q, struct request *rq)
 300 {
 301         elevator_t *e = q->elevator;
 302
 303         BUG_ON(ELV_ON_HASH(rq));
 304         hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
 305 }
 306
 307 static void elv_rqhash_reposition(request_queue_t *q, struct request *rq)
 308 {
 309         __elv_rqhash_del(rq);
 310         elv_rqhash_add(q, rq);
 311 }
 312
 313 static struct request *elv_rqhash_find(request_queue_t *q, sector_t offset)
 314 {
 315         elevator_t *e = q->elevator;
 316         struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
 317         struct hlist_node *entry, *next;
 318         struct request *rq;
 319
 320         hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
 321                 BUG_ON(!ELV_ON_HASH(rq));
 322
 323                 if (unlikely(!rq_mergeable(rq))) {
 324                         __elv_rqhash_del(rq);
 325                         continue;
 326                 }
 327
 328                 if (rq_hash_key(rq) == offset)
 329                         return rq;
 330         }
 331
 332         return NULL;
 333 }
 334
 335 /*
 336  * RB-tree support functions for inserting/lookup/removal of requests
 337  * in a sorted RB tree.
 338  */
 339 struct request *elv_rb_add(struct rb_root *root, struct request *rq)
 340 {
 341         struct rb_node **p = &root->rb_node;
 342         struct rb_node *parent = NULL;
 343         struct request *__rq;
 344
 345         while (*p) {
 346                 parent = *p;
 347                 __rq = rb_entry(parent, struct request, rb_node);
 348
 349                 if (rq->sector < __rq->sector)
 350                         p = &(*p)->rb_left;
 351                 else if (rq->sector > __rq->sector)
 352                         p = &(*p)->rb_right;
 353                 else
 354                         return __rq;
 355         }
 356
 357         rb_link_node(&rq->rb_node, parent, p);
 358         rb_insert_color(&rq->rb_node, root);
 359         return NULL;
 360 }
 361
 362 EXPORT_SYMBOL(elv_rb_add);
 363
 364 void elv_rb_del(struct rb_root *root, struct request *rq)
 365 {
 366         BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
 367         rb_erase(&rq->rb_node, root);
 368         RB_CLEAR_NODE(&rq->rb_node);
 369 }
 370
 371 EXPORT_SYMBOL(elv_rb_del);
 372
 373 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
 374 {
 375         struct rb_node *n = root->rb_node;
 376         struct request *rq;
 377
 378         while (n) {
 379                 rq = rb_entry(n, struct request, rb_node);
 380
 381                 if (sector < rq->sector)
 382                         n = n->rb_left;
 383                 else if (sector > rq->sector)
 384                         n = n->rb_right;
 385                 else
 386                         return rq;
 387         }
 388
 389         return NULL;
 390 }
 391
 392 EXPORT_SYMBOL(elv_rb_find);
 393
 394 /*
 395  * Insert rq into dispatch queue of q.  Queue lock must be held on
 396  * entry.  rq is sort insted into the dispatch queue. To be used by
 397  * specific elevators.
 398  */
 399 void elv_dispatch_sort(request_queue_t *q, struct request *rq)
 400 {
 401         sector_t boundary;
 402         struct list_head *entry;
 403
 404         if (q->last_merge == rq)
 405                 q->last_merge = NULL;
 406
 407         elv_rqhash_del(q, rq);
 408
 409         q->nr_sorted--;
 410
 411         boundary = q->end_sector;
 412
 413         list_for_each_prev(entry, &q->queue_head) {
 414                 struct request *pos = list_entry_rq(entry);
 415
 416                 if (rq_data_dir(rq) != rq_data_dir(pos))
 417                         break;
 418                 if (pos->cmd_flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
 419                         break;
 420                 if (rq->sector >= boundary) {
 421                         if (pos->sector < boundary)
 422                                 continue;
 423                 } else {
 424                         if (pos->sector >= boundary)
 425                                 break;
 426                 }
 427                 if (rq->sector >= pos->sector)
 428                         break;
 429         }
 430
 431         list_add(&rq->queuelist, entry);
 432 }
 433
 434 EXPORT_SYMBOL(elv_dispatch_sort);
 435
 436 /*
 437  * Insert rq into dispatch queue of q.  Queue lock must be held on
 438  * entry.  rq is added to the back of the dispatch queue. To be used by
 439  * specific elevators.
 440  */
 441 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
 442 {
 443         if (q->last_merge == rq)
 444                 q->last_merge = NULL;
 445
 446         elv_rqhash_del(q, rq);
 447
 448         q->nr_sorted--;
 449
 450         q->end_sector = rq_end_sector(rq);
 451         q->boundary_rq = rq;
 452         list_add_tail(&rq->queuelist, &q->queue_head);
 453 }
 454
 455 EXPORT_SYMBOL(elv_dispatch_add_tail);
 456
 457 int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
 458 {
 459         elevator_t *e = q->elevator;
 460         struct request *__rq;
 461         int ret;
 462
 463         /*
 464          * First try one-hit cache.
 465          */
 466         if (q->last_merge) {
 467                 ret = elv_try_merge(q->last_merge, bio);
 468                 if (ret != ELEVATOR_NO_MERGE) {
 469                         *req = q->last_merge;
 470                         return ret;
 471                 }
 472         }
 473
 474         /*
 475          * See if our hash lookup can find a potential backmerge.
 476          */
 477         __rq = elv_rqhash_find(q, bio->bi_sector);
 478         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 479                 *req = __rq;
 480                 return ELEVATOR_BACK_MERGE;
 481         }
 482
 483         if (e->ops->elevator_merge_fn)
 484                 return e->ops->elevator_merge_fn(q, req, bio);
 485
 486         return ELEVATOR_NO_MERGE;
 487 }
 488
 489 void elv_merged_request(request_queue_t *q, struct request *rq, int type)
 490 {
 491         elevator_t *e = q->elevator;
 492
 493         if (e->ops->elevator_merged_fn)
 494                 e->ops->elevator_merged_fn(q, rq, type);
 495
 496         if (type == ELEVATOR_BACK_MERGE)
 497                 elv_rqhash_reposition(q, rq);
 498
 499         q->last_merge = rq;
 500 }
 501
 502 void elv_merge_requests(request_queue_t *q, struct request *rq,
 503                              struct request *next)
 504 {
 505         elevator_t *e = q->elevator;
 506
 507         if (e->ops->elevator_merge_req_fn)
 508                 e->ops->elevator_merge_req_fn(q, rq, next);
 509
 510         elv_rqhash_reposition(q, rq);
 511         elv_rqhash_del(q, next);
 512
 513         q->nr_sorted--;
 514         q->last_merge = rq;
 515 }
 516
 517 void elv_requeue_request(request_queue_t *q, struct request *rq)
 518 {
 519         /*
 520          * it already went through dequeue, we need to decrement the
 521          * in_flight count again
 522          */
 523         if (blk_account_rq(rq)) {
 524                 q->in_flight--;
 525                 if (blk_sorted_rq(rq))
 526                         elv_deactivate_rq(q, rq);
 527         }
 528
 529         rq->cmd_flags &= ~REQ_STARTED;
 530
 531         elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
 532 }
 533
 534 static void elv_drain_elevator(request_queue_t *q)
 535 {
 536         static int printed;
 537         while (q->elevator->ops->elevator_dispatch_fn(q, 1))
 538                 ;
 539         if (q->nr_sorted == 0)
 540                 return;
 541         if (printed++ < 10) {
 542                 printk(KERN_ERR "%s: forced dispatching is broken "
 543                        "(nr_sorted=%u), please report this\n",
 544                        q->elevator->elevator_type->elevator_name, q->nr_sorted);
 545         }
 546 }
 547
 548 void elv_insert(request_queue_t *q, struct request *rq, int where)
 549 {
 550         struct list_head *pos;
 551         unsigned ordseq;
 552         int unplug_it = 1;
 553
 554         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 555
 556         rq->q = q;
 557
 558         switch (where) {
 559         case ELEVATOR_INSERT_FRONT:
 560                 rq->cmd_flags |= REQ_SOFTBARRIER;
 561
 562                 list_add(&rq->queuelist, &q->queue_head);
 563                 break;
 564
 565         case ELEVATOR_INSERT_BACK:
 566                 rq->cmd_flags |= REQ_SOFTBARRIER;
 567                 elv_drain_elevator(q);
 568                 list_add_tail(&rq->queuelist, &q->queue_head);
 569                 /*
 570                  * We kick the queue here for the following reasons.
 571                  * - The elevator might have returned NULL previously
 572                  *   to delay requests and returned them now.  As the
 573                  *   queue wasn't empty before this request, ll_rw_blk
 574                  *   won't run the queue on return, resulting in hang.
 575                  * - Usually, back inserted requests won't be merged
 576                  *   with anything.  There's no point in delaying queue
 577                  *   processing.
 578                  */
 579                 blk_remove_plug(q);
 580                 q->request_fn(q);
 581                 break;
 582
 583         case ELEVATOR_INSERT_SORT:
 584                 BUG_ON(!blk_fs_request(rq));
 585                 rq->cmd_flags |= REQ_SORTED;
 586                 q->nr_sorted++;
 587                 if (rq_mergeable(rq)) {
 588                         elv_rqhash_add(q, rq);
 589                         if (!q->last_merge)
 590                                 q->last_merge = rq;
 591                 }
 592
 593                 /*
 594                  * Some ioscheds (cfq) run q->request_fn directly, so
 595                  * rq cannot be accessed after calling
 596                  * elevator_add_req_fn.
 597                  */
 598                 q->elevator->ops->elevator_add_req_fn(q, rq);
 599                 break;
 600
 601         case ELEVATOR_INSERT_REQUEUE:
 602                 /*
 603                  * If ordered flush isn't in progress, we do front
 604                  * insertion; otherwise, requests should be requeued
 605                  * in ordseq order.
 606                  */
 607                 rq->cmd_flags |= REQ_SOFTBARRIER;
 608
 609                 /*
 610                  * Most requeues happen because of a busy condition,
 611                  * don't force unplug of the queue for that case.
 612                  */
 613                 unplug_it = 0;
 614
 615                 if (q->ordseq == 0) {
 616                         list_add(&rq->queuelist, &q->queue_head);
 617                         break;
 618                 }
 619
 620                 ordseq = blk_ordered_req_seq(rq);
 621
 622                 list_for_each(pos, &q->queue_head) {
 623                         struct request *pos_rq = list_entry_rq(pos);
 624                         if (ordseq <= blk_ordered_req_seq(pos_rq))
 625                                 break;
 626                 }
 627
 628                 list_add_tail(&rq->queuelist, pos);
 629                 break;
 630
 631         default:
 632                 printk(KERN_ERR "%s: bad insertion point %d\n",
 633                        __FUNCTION__, where);
 634                 BUG();
 635         }
 636
 637         if (unplug_it && blk_queue_plugged(q)) {
 638                 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
 639                         - q->in_flight;
 640
 641                 if (nrq >= q->unplug_thresh)
 642                         __generic_unplug_device(q);
 643         }
 644 }
 645
 646 void __elv_add_request(request_queue_t *q, struct request *rq, int where,
 647                        int plug)
 648 {
 649         if (q->ordcolor)
 650                 rq->cmd_flags |= REQ_ORDERED_COLOR;
 651
 652         if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
 653                 /*
 654                  * toggle ordered color
 655                  */
 656                 if (blk_barrier_rq(rq))
 657                         q->ordcolor ^= 1;
 658
 659                 /*
 660                  * barriers implicitly indicate back insertion
 661                  */
 662                 if (where == ELEVATOR_INSERT_SORT)
 663                         where = ELEVATOR_INSERT_BACK;
 664
 665                 /*
 666                  * this request is scheduling boundary, update
 667                  * end_sector
 668                  */
 669                 if (blk_fs_request(rq)) {
 670                         q->end_sector = rq_end_sector(rq);
 671                         q->boundary_rq = rq;
 672                 }
 673         } else if (!(rq->cmd_flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
 674                 where = ELEVATOR_INSERT_BACK;
 675
 676         if (plug)
 677                 blk_plug_device(q);
 678
 679         elv_insert(q, rq, where);
 680 }
 681
 682 EXPORT_SYMBOL(__elv_add_request);
 683
 684 void elv_add_request(request_queue_t *q, struct request *rq, int where,
 685                      int plug)
 686 {
 687         unsigned long flags;
 688
 689         spin_lock_irqsave(q->queue_lock, flags);
 690         __elv_add_request(q, rq, where, plug);
 691         spin_unlock_irqrestore(q->queue_lock, flags);
 692 }
 693
 694 EXPORT_SYMBOL(elv_add_request);
 695
 696 static inline struct request *__elv_next_request(request_queue_t *q)
 697 {
 698         struct request *rq;
 699
 700         while (1) {
 701                 while (!list_empty(&q->queue_head)) {
 702                         rq = list_entry_rq(q->queue_head.next);
 703                         if (blk_do_ordered(q, &rq))
 704                                 return rq;
 705                 }
 706
 707                 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
 708                         return NULL;
 709         }
 710 }
 711
 712 struct request *elv_next_request(request_queue_t *q)
 713 {
 714         struct request *rq;
 715         int ret;
 716
 717         while ((rq = __elv_next_request(q)) != NULL) {
 718                 if (!(rq->cmd_flags & REQ_STARTED)) {
 719                         /*
 720                          * This is the first time the device driver
 721                          * sees this request (possibly after
 722                          * requeueing).  Notify IO scheduler.
 723                          */
 724                         if (blk_sorted_rq(rq))
 725                                 elv_activate_rq(q, rq);
 726
 727                         /*
 728                          * just mark as started even if we don't start
 729                          * it, a request that has been delayed should
 730                          * not be passed by new incoming requests
 731                          */
 732                         rq->cmd_flags |= REQ_STARTED;
 733                         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 734                 }
 735
 736                 if (!q->boundary_rq || q->boundary_rq == rq) {
 737                         q->end_sector = rq_end_sector(rq);
 738                         q->boundary_rq = NULL;
 739                 }
 740
 741                 if ((rq->cmd_flags & REQ_DONTPREP) || !q->prep_rq_fn)
 742                         break;
 743
 744                 ret = q->prep_rq_fn(q, rq);
 745                 if (ret == BLKPREP_OK) {
 746                         break;
 747                 } else if (ret == BLKPREP_DEFER) {
 748                         /*
 749                          * the request may have been (partially) prepped.
 750                          * we need to keep this request in the front to
 751                          * avoid resource deadlock.  REQ_STARTED will
 752                          * prevent other fs requests from passing this one.
 753                          */
 754                         rq = NULL;
 755                         break;
 756                 } else if (ret == BLKPREP_KILL) {
 757                         int nr_bytes = rq->hard_nr_sectors << 9;
 758
 759                         if (!nr_bytes)
 760                                 nr_bytes = rq->data_len;
 761
 762                         blkdev_dequeue_request(rq);
 763                         rq->cmd_flags |= REQ_QUIET;
 764                         end_that_request_chunk(rq, 0, nr_bytes);
 765                         end_that_request_last(rq, 0);
 766                 } else {
 767                         printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
 768                                                                 ret);
 769                         break;
 770                 }
 771         }
 772
 773         return rq;
 774 }
 775
 776 EXPORT_SYMBOL(elv_next_request);
 777
 778 void elv_dequeue_request(request_queue_t *q, struct request *rq)
 779 {
 780         BUG_ON(list_empty(&rq->queuelist));
 781         BUG_ON(ELV_ON_HASH(rq));
 782
 783         list_del_init(&rq->queuelist);
 784
 785         /*
 786          * the time frame between a request being removed from the lists
 787          * and to it is freed is accounted as io that is in progress at
 788          * the driver side.
 789          */
 790         if (blk_account_rq(rq))
 791                 q->in_flight++;
 792 }
 793
 794 EXPORT_SYMBOL(elv_dequeue_request);
 795
 796 int elv_queue_empty(request_queue_t *q)
 797 {
 798         elevator_t *e = q->elevator;
 799
 800         if (!list_empty(&q->queue_head))
 801                 return 0;
 802
 803         if (e->ops->elevator_queue_empty_fn)
 804                 return e->ops->elevator_queue_empty_fn(q);
 805
 806         return 1;
 807 }
 808
 809 EXPORT_SYMBOL(elv_queue_empty);
 810
 811 struct request *elv_latter_request(request_queue_t *q, struct request *rq)
 812 {
 813         elevator_t *e = q->elevator;
 814
 815         if (e->ops->elevator_latter_req_fn)
 816                 return e->ops->elevator_latter_req_fn(q, rq);
 817         return NULL;
 818 }
 819
 820 struct request *elv_former_request(request_queue_t *q, struct request *rq)
 821 {
 822         elevator_t *e = q->elevator;
 823
 824         if (e->ops->elevator_former_req_fn)
 825                 return e->ops->elevator_former_req_fn(q, rq);
 826         return NULL;
 827 }
 828
 829 int elv_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask)
 830 {
 831         elevator_t *e = q->elevator;
 832
 833         if (e->ops->elevator_set_req_fn)
 834                 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
 835
 836         rq->elevator_private = NULL;
 837         return 0;
 838 }
 839
 840 void elv_put_request(request_queue_t *q, struct request *rq)
 841 {
 842         elevator_t *e = q->elevator;
 843
 844         if (e->ops->elevator_put_req_fn)
 845                 e->ops->elevator_put_req_fn(rq);
 846 }
 847
 848 int elv_may_queue(request_queue_t *q, int rw)
 849 {
 850         elevator_t *e = q->elevator;
 851
 852         if (e->ops->elevator_may_queue_fn)
 853                 return e->ops->elevator_may_queue_fn(q, rw);
 854
 855         return ELV_MQUEUE_MAY;
 856 }
 857
 858 void elv_completed_request(request_queue_t *q, struct request *rq)
 859 {
 860         elevator_t *e = q->elevator;
 861
 862         /*
 863          * request is released from the driver, io must be done
 864          */
 865         if (blk_account_rq(rq)) {
 866                 q->in_flight--;
 867                 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
 868                         e->ops->elevator_completed_req_fn(q, rq);
 869         }
 870
 871         /*
 872          * Check if the queue is waiting for fs requests to be
 873          * drained for flush sequence.
 874          */
 875         if (unlikely(q->ordseq)) {
 876                 struct request *first_rq = list_entry_rq(q->queue_head.next);
 877                 if (q->in_flight == 0 &&
 878                     blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
 879                     blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
 880                         blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
 881                         q->request_fn(q);
 882                 }
 883         }
 884 }
 885
 886 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
 887
 888 static ssize_t
 889 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 890 {
 891         elevator_t *e = container_of(kobj, elevator_t, kobj);
 892         struct elv_fs_entry *entry = to_elv(attr);
 893         ssize_t error;
 894
 895         if (!entry->show)
 896                 return -EIO;
 897
 898         mutex_lock(&e->sysfs_lock);
 899         error = e->ops ? entry->show(e, page) : -ENOENT;
 900         mutex_unlock(&e->sysfs_lock);
 901         return error;
 902 }
 903
 904 static ssize_t
 905 elv_attr_store(struct kobject *kobj, struct attribute *attr,
 906                const char *page, size_t length)
 907 {
 908         elevator_t *e = container_of(kobj, elevator_t, kobj);
 909         struct elv_fs_entry *entry = to_elv(attr);
 910         ssize_t error;
 911
 912         if (!entry->store)
 913                 return -EIO;
 914
 915         mutex_lock(&e->sysfs_lock);
 916         error = e->ops ? entry->store(e, page, length) : -ENOENT;
 917         mutex_unlock(&e->sysfs_lock);
 918         return error;
 919 }
 920
 921 static struct sysfs_ops elv_sysfs_ops = {
 922         .show   = elv_attr_show,
 923         .store  = elv_attr_store,
 924 };
 925
 926 static struct kobj_type elv_ktype = {
 927         .sysfs_ops      = &elv_sysfs_ops,
 928         .release        = elevator_release,
 929 };
 930
 931 int elv_register_queue(struct request_queue *q)
 932 {
 933         elevator_t *e = q->elevator;
 934         int error;
 935
 936         e->kobj.parent = &q->kobj;
 937
 938         error = kobject_add(&e->kobj);
 939         if (!error) {
 940                 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
 941                 if (attr) {
 942                         while (attr->attr.name) {
 943                                 if (sysfs_create_file(&e->kobj, &attr->attr))
 944                                         break;
 945                                 attr++;
 946                         }
 947                 }
 948                 kobject_uevent(&e->kobj, KOBJ_ADD);
 949         }
 950         return error;
 951 }
 952
 953 static void __elv_unregister_queue(elevator_t *e)
 954 {
 955         kobject_uevent(&e->kobj, KOBJ_REMOVE);
 956         kobject_del(&e->kobj);
 957 }
 958
 959 void elv_unregister_queue(struct request_queue *q)
 960 {
 961         if (q)
 962                 __elv_unregister_queue(q->elevator);
 963 }
 964
 965 int elv_register(struct elevator_type *e)
 966 {
 967         char *def = "";
 968         spin_lock_irq(&elv_list_lock);
 969         BUG_ON(elevator_find(e->elevator_name));
 970         list_add_tail(&e->list, &elv_list);
 971         spin_unlock_irq(&elv_list_lock);
 972
 973         if (!strcmp(e->elevator_name, chosen_elevator) ||
 974                         (!*chosen_elevator &&
 975                          !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
 976                                 def = " (default)";
 977
 978         printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name, def);
 979         return 0;
 980 }
 981 EXPORT_SYMBOL_GPL(elv_register);
 982
 983 void elv_unregister(struct elevator_type *e)
 984 {
 985         struct task_struct *g, *p;
 986
 987         /*
 988          * Iterate every thread in the process to remove the io contexts.
 989          */
 990         if (e->ops.trim) {
 991                 read_lock(&tasklist_lock);
 992                 do_each_thread(g, p) {
 993                         task_lock(p);
 994                         if (p->io_context)
 995                                 e->ops.trim(p->io_context);
 996                         task_unlock(p);
 997                 } while_each_thread(g, p);
 998                 read_unlock(&tasklist_lock);
 999         }
1000
1001         spin_lock_irq(&elv_list_lock);
1002         list_del_init(&e->list);
1003         spin_unlock_irq(&elv_list_lock);
1004 }
1005 EXPORT_SYMBOL_GPL(elv_unregister);
1006
1007 /*
1008  * switch to new_e io scheduler. be careful not to introduce deadlocks -
1009  * we don't free the old io scheduler, before we have allocated what we
1010  * need for the new one. this way we have a chance of going back to the old
1011  * one, if the new one fails init for some reason.
1012  */
1013 static int elevator_switch(request_queue_t *q, struct elevator_type *new_e)
1014 {
1015         elevator_t *old_elevator, *e;
1016         void *data;
1017
1018         /*
1019          * Allocate new elevator
1020          */
1021         e = elevator_alloc(q, new_e);
1022         if (!e)
1023                 return 0;
1024
1025         data = elevator_init_queue(q, e);
1026         if (!data) {
1027                 kobject_put(&e->kobj);
1028                 return 0;
1029         }
1030
1031         /*
1032          * Turn on BYPASS and drain all requests w/ elevator private data
1033          */
1034         spin_lock_irq(q->queue_lock);
1035
1036         set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1037
1038         elv_drain_elevator(q);
1039
1040         while (q->rq.elvpriv) {
1041                 blk_remove_plug(q);
1042                 q->request_fn(q);
1043                 spin_unlock_irq(q->queue_lock);
1044                 msleep(10);
1045                 spin_lock_irq(q->queue_lock);
1046                 elv_drain_elevator(q);
1047         }
1048
1049         /*
1050          * Remember old elevator.
1051          */
1052         old_elevator = q->elevator;
1053
1054         /*
1055          * attach and start new elevator
1056          */
1057         elevator_attach(q, e, data);
1058
1059         spin_unlock_irq(q->queue_lock);
1060
1061         __elv_unregister_queue(old_elevator);
1062
1063         if (elv_register_queue(q))
1064                 goto fail_register;
1065
1066         /*
1067          * finally exit old elevator and turn off BYPASS.
1068          */
1069         elevator_exit(old_elevator);
1070         clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1071         return 1;
1072
1073 fail_register:
1074         /*
1075          * switch failed, exit the new io scheduler and reattach the old
1076          * one again (along with re-adding the sysfs dir)
1077          */
1078         elevator_exit(e);
1079         q->elevator = old_elevator;
1080         elv_register_queue(q);
1081         clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1082         return 0;
1083 }
1084
1085 ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
1086 {
1087         char elevator_name[ELV_NAME_MAX];
1088         size_t len;
1089         struct elevator_type *e;
1090
1091         elevator_name[sizeof(elevator_name) - 1] = '\0';
1092         strncpy(elevator_name, name, sizeof(elevator_name) - 1);
1093         len = strlen(elevator_name);
1094
1095         if (len && elevator_name[len - 1] == '\n')
1096                 elevator_name[len - 1] = '\0';
1097
1098         e = elevator_get(elevator_name);
1099         if (!e) {
1100                 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1101                 return -EINVAL;
1102         }
1103
1104         if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1105                 elevator_put(e);
1106                 return count;
1107         }
1108
1109         if (!elevator_switch(q, e))
1110                 printk(KERN_ERR "elevator: switch to %s failed\n",elevator_name);
1111         return count;
1112 }
1113
1114 ssize_t elv_iosched_show(request_queue_t *q, char *name)
1115 {
1116         elevator_t *e = q->elevator;
1117         struct elevator_type *elv = e->elevator_type;
1118         struct list_head *entry;
1119         int len = 0;
1120
1121         spin_lock_irq(&elv_list_lock);
1122         list_for_each(entry, &elv_list) {
1123                 struct elevator_type *__e;
1124
1125                 __e = list_entry(entry, struct elevator_type, list);
1126                 if (!strcmp(elv->elevator_name, __e->elevator_name))
1127                         len += sprintf(name+len, "[%s] ", elv->elevator_name);
1128                 else
1129                         len += sprintf(name+len, "%s ", __e->elevator_name);
1130         }
1131         spin_unlock_irq(&elv_list_lock);
1132
1133         len += sprintf(len+name, "\n");
1134         return len;
1135 }
1136
1137 struct request *elv_rb_former_request(request_queue_t *q, struct request *rq)
1138 {
1139         struct rb_node *rbprev = rb_prev(&rq->rb_node);
1140
1141         if (rbprev)
1142                 return rb_entry_rq(rbprev);
1143
1144         return NULL;
1145 }
1146
1147 EXPORT_SYMBOL(elv_rb_former_request);
1148
1149 struct request *elv_rb_latter_request(request_queue_t *q, struct request *rq)
1150 {
1151         struct rb_node *rbnext = rb_next(&rq->rb_node);
1152
1153         if (rbnext)
1154                 return rb_entry_rq(rbnext);
1155
1156         return NULL;
1157 }
1158
1159 EXPORT_SYMBOL(elv_rb_latter_request);