fs/btrfs/reada.c

   1 /*
   2  * Copyright (C) 2011 STRATO.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/sched.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/writeback.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/rbtree.h>
  24 #include <linux/slab.h>
  25 #include <linux/workqueue.h>
  26 #include "ctree.h"
  27 #include "volumes.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30
  31 #undef DEBUG
  32
  33 /*
  34  * This is the implementation for the generic read ahead framework.
  35  *
  36  * To trigger a readahead, btrfs_reada_add must be called. It will start
  37  * a read ahead for the given range [start, end) on tree root. The returned
  38  * handle can either be used to wait on the readahead to finish
  39  * (btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
  40  *
  41  * The read ahead works as follows:
  42  * On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
  43  * reada_start_machine will then search for extents to prefetch and trigger
  44  * some reads. When a read finishes for a node, all contained node/leaf
  45  * pointers that lie in the given range will also be enqueued. The reads will
  46  * be triggered in sequential order, thus giving a big win over a naive
  47  * enumeration. It will also make use of multi-device layouts. Each disk
  48  * will have its on read pointer and all disks will by utilized in parallel.
  49  * Also will no two disks read both sides of a mirror simultaneously, as this
  50  * would waste seeking capacity. Instead both disks will read different parts
  51  * of the filesystem.
  52  * Any number of readaheads can be started in parallel. The read order will be
  53  * determined globally, i.e. 2 parallel readaheads will normally finish faster
  54  * than the 2 started one after another.
  55  */
  56
  57 #define MAX_MIRRORS 2
  58 #define MAX_IN_FLIGHT 6
  59
  60 struct reada_extctl {
  61         struct list_head        list;
  62         struct reada_control    *rc;
  63         u64                     generation;
  64 };
  65
  66 struct reada_extent {
  67         u64                     logical;
  68         struct btrfs_key        top;
  69         u32                     blocksize;
  70         int                     err;
  71         struct list_head        extctl;
  72         struct kref             refcnt;
  73         spinlock_t              lock;
  74         struct reada_zone       *zones[MAX_MIRRORS];
  75         int                     nzones;
  76         struct btrfs_device     *scheduled_for;
  77 };
  78
  79 struct reada_zone {
  80         u64                     start;
  81         u64                     end;
  82         u64                     elems;
  83         struct list_head        list;
  84         spinlock_t              lock;
  85         int                     locked;
  86         struct btrfs_device     *device;
  87         struct btrfs_device     *devs[MAX_MIRRORS]; /* full list, incl self */
  88         int                     ndevs;
  89         struct kref             refcnt;
  90 };
  91
  92 struct reada_machine_work {
  93         struct btrfs_work       work;
  94         struct btrfs_fs_info    *fs_info;
  95 };
  96
  97 static void reada_extent_put(struct btrfs_fs_info *, struct reada_extent *);
  98 static void reada_control_release(struct kref *kref);
  99 static void reada_zone_release(struct kref *kref);
 100 static void reada_start_machine(struct btrfs_fs_info *fs_info);
 101 static void __reada_start_machine(struct btrfs_fs_info *fs_info);
 102
 103 static int reada_add_block(struct reada_control *rc, u64 logical,
 104                            struct btrfs_key *top, int level, u64 generation);
 105
 106 /* recurses */
 107 /* in case of err, eb might be NULL */
 108 static int __readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
 109                             u64 start, int err)
 110 {
 111         int level = 0;
 112         int nritems;
 113         int i;
 114         u64 bytenr;
 115         u64 generation;
 116         struct reada_extent *re;
 117         struct btrfs_fs_info *fs_info = root->fs_info;
 118         struct list_head list;
 119         unsigned long index = start >> PAGE_CACHE_SHIFT;
 120         struct btrfs_device *for_dev;
 121
 122         if (eb)
 123                 level = btrfs_header_level(eb);
 124
 125         /* find extent */
 126         spin_lock(&fs_info->reada_lock);
 127         re = radix_tree_lookup(&fs_info->reada_tree, index);
 128         if (re)
 129                 kref_get(&re->refcnt);
 130         spin_unlock(&fs_info->reada_lock);
 131
 132         if (!re)
 133                 return -1;
 134
 135         spin_lock(&re->lock);
 136         /*
 137          * just take the full list from the extent. afterwards we
 138          * don't need the lock anymore
 139          */
 140         list_replace_init(&re->extctl, &list);
 141         for_dev = re->scheduled_for;
 142         re->scheduled_for = NULL;
 143         spin_unlock(&re->lock);
 144
 145         if (err == 0) {
 146                 nritems = level ? btrfs_header_nritems(eb) : 0;
 147                 generation = btrfs_header_generation(eb);
 148                 /*
 149                  * FIXME: currently we just set nritems to 0 if this is a leaf,
 150                  * effectively ignoring the content. In a next step we could
 151                  * trigger more readahead depending from the content, e.g.
 152                  * fetch the checksums for the extents in the leaf.
 153                  */
 154         } else {
 155                 /*
 156                  * this is the error case, the extent buffer has not been
 157                  * read correctly. We won't access anything from it and
 158                  * just cleanup our data structures. Effectively this will
 159                  * cut the branch below this node from read ahead.
 160                  */
 161                 nritems = 0;
 162                 generation = 0;
 163         }
 164
 165         for (i = 0; i < nritems; i++) {
 166                 struct reada_extctl *rec;
 167                 u64 n_gen;
 168                 struct btrfs_key key;
 169                 struct btrfs_key next_key;
 170
 171                 btrfs_node_key_to_cpu(eb, &key, i);
 172                 if (i + 1 < nritems)
 173                         btrfs_node_key_to_cpu(eb, &next_key, i + 1);
 174                 else
 175                         next_key = re->top;
 176                 bytenr = btrfs_node_blockptr(eb, i);
 177                 n_gen = btrfs_node_ptr_generation(eb, i);
 178
 179                 list_for_each_entry(rec, &list, list) {
 180                         struct reada_control *rc = rec->rc;
 181
 182                         /*
 183                          * if the generation doesn't match, just ignore this
 184                          * extctl. This will probably cut off a branch from
 185                          * prefetch. Alternatively one could start a new (sub-)
 186                          * prefetch for this branch, starting again from root.
 187                          * FIXME: move the generation check out of this loop
 188                          */
 189 #ifdef DEBUG
 190                         if (rec->generation != generation) {
 191                                 printk(KERN_DEBUG "generation mismatch for "
 192                                                 "(%llu,%d,%llu) %llu != %llu\n",
 193                                        key.objectid, key.type, key.offset,
 194                                        rec->generation, generation);
 195                         }
 196 #endif
 197                         if (rec->generation == generation &&
 198                             btrfs_comp_cpu_keys(&key, &rc->key_end) < 0 &&
 199                             btrfs_comp_cpu_keys(&next_key, &rc->key_start) > 0)
 200                                 reada_add_block(rc, bytenr, &next_key,
 201                                                 level - 1, n_gen);
 202                 }
 203         }
 204         /*
 205          * free extctl records
 206          */
 207         while (!list_empty(&list)) {
 208                 struct reada_control *rc;
 209                 struct reada_extctl *rec;
 210
 211                 rec = list_first_entry(&list, struct reada_extctl, list);
 212                 list_del(&rec->list);
 213                 rc = rec->rc;
 214                 kfree(rec);
 215
 216                 kref_get(&rc->refcnt);
 217                 if (atomic_dec_and_test(&rc->elems)) {
 218                         kref_put(&rc->refcnt, reada_control_release);
 219                         wake_up(&rc->wait);
 220                 }
 221                 kref_put(&rc->refcnt, reada_control_release);
 222
 223                 reada_extent_put(fs_info, re);  /* one ref for each entry */
 224         }
 225         reada_extent_put(fs_info, re);  /* our ref */
 226         if (for_dev)
 227                 atomic_dec(&for_dev->reada_in_flight);
 228
 229         return 0;
 230 }
 231
 232 /*
 233  * start is passed separately in case eb in NULL, which may be the case with
 234  * failed I/O
 235  */
 236 int btree_readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
 237                          u64 start, int err)
 238 {
 239         int ret;
 240
 241         ret = __readahead_hook(root, eb, start, err);
 242
 243         reada_start_machine(root->fs_info);
 244
 245         return ret;
 246 }
 247
 248 static struct reada_zone *reada_find_zone(struct btrfs_fs_info *fs_info,
 249                                           struct btrfs_device *dev, u64 logical,
 250                                           struct btrfs_bio *multi)
 251 {
 252         int ret;
 253         int looped = 0;
 254         struct reada_zone *zone;
 255         struct btrfs_block_group_cache *cache = NULL;
 256         u64 start;
 257         u64 end;
 258         int i;
 259
 260 again:
 261         zone = NULL;
 262         spin_lock(&fs_info->reada_lock);
 263         ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
 264                                      logical >> PAGE_CACHE_SHIFT, 1);
 265         if (ret == 1)
 266                 kref_get(&zone->refcnt);
 267         spin_unlock(&fs_info->reada_lock);
 268
 269         if (ret == 1) {
 270                 if (logical >= zone->start && logical < zone->end)
 271                         return zone;
 272                 spin_lock(&fs_info->reada_lock);
 273                 kref_put(&zone->refcnt, reada_zone_release);
 274                 spin_unlock(&fs_info->reada_lock);
 275         }
 276
 277         if (looped)
 278                 return NULL;
 279
 280         cache = btrfs_lookup_block_group(fs_info, logical);
 281         if (!cache)
 282                 return NULL;
 283
 284         start = cache->key.objectid;
 285         end = start + cache->key.offset - 1;
 286         btrfs_put_block_group(cache);
 287
 288         zone = kzalloc(sizeof(*zone), GFP_NOFS);
 289         if (!zone)
 290                 return NULL;
 291
 292         zone->start = start;
 293         zone->end = end;
 294         INIT_LIST_HEAD(&zone->list);
 295         spin_lock_init(&zone->lock);
 296         zone->locked = 0;
 297         kref_init(&zone->refcnt);
 298         zone->elems = 0;
 299         zone->device = dev; /* our device always sits at index 0 */
 300         for (i = 0; i < multi->num_stripes; ++i) {
 301                 /* bounds have already been checked */
 302                 zone->devs[i] = multi->stripes[i].dev;
 303         }
 304         zone->ndevs = multi->num_stripes;
 305
 306         spin_lock(&fs_info->reada_lock);
 307         ret = radix_tree_insert(&dev->reada_zones,
 308                                 (unsigned long)zone->end >> PAGE_CACHE_SHIFT,
 309                                 zone);
 310         spin_unlock(&fs_info->reada_lock);
 311
 312         if (ret) {
 313                 kfree(zone);
 314                 looped = 1;
 315                 goto again;
 316         }
 317
 318         return zone;
 319 }
 320
 321 static struct reada_extent *reada_find_extent(struct btrfs_root *root,
 322                                               u64 logical,
 323                                               struct btrfs_key *top, int level)
 324 {
 325         int ret;
 326         int looped = 0;
 327         struct reada_extent *re = NULL;
 328         struct btrfs_fs_info *fs_info = root->fs_info;
 329         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
 330         struct btrfs_bio *multi = NULL;
 331         struct btrfs_device *dev;
 332         u32 blocksize;
 333         u64 length;
 334         int nzones = 0;
 335         int i;
 336         unsigned long index = logical >> PAGE_CACHE_SHIFT;
 337
 338 again:
 339         spin_lock(&fs_info->reada_lock);
 340         re = radix_tree_lookup(&fs_info->reada_tree, index);
 341         if (re)
 342                 kref_get(&re->refcnt);
 343         spin_unlock(&fs_info->reada_lock);
 344
 345         if (re || looped)
 346                 return re;
 347
 348         re = kzalloc(sizeof(*re), GFP_NOFS);
 349         if (!re)
 350                 return NULL;
 351
 352         blocksize = btrfs_level_size(root, level);
 353         re->logical = logical;
 354         re->blocksize = blocksize;
 355         re->top = *top;
 356         INIT_LIST_HEAD(&re->extctl);
 357         spin_lock_init(&re->lock);
 358         kref_init(&re->refcnt);
 359
 360         /*
 361          * map block
 362          */
 363         length = blocksize;
 364         ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length, &multi, 0);
 365         if (ret || !multi || length < blocksize)
 366                 goto error;
 367
 368         if (multi->num_stripes > MAX_MIRRORS) {
 369                 printk(KERN_ERR "btrfs readahead: more than %d copies not "
 370                                 "supported", MAX_MIRRORS);
 371                 goto error;
 372         }
 373
 374         for (nzones = 0; nzones < multi->num_stripes; ++nzones) {
 375                 struct reada_zone *zone;
 376
 377                 dev = multi->stripes[nzones].dev;
 378                 zone = reada_find_zone(fs_info, dev, logical, multi);
 379                 if (!zone)
 380                         break;
 381
 382                 re->zones[nzones] = zone;
 383                 spin_lock(&zone->lock);
 384                 if (!zone->elems)
 385                         kref_get(&zone->refcnt);
 386                 ++zone->elems;
 387                 spin_unlock(&zone->lock);
 388                 spin_lock(&fs_info->reada_lock);
 389                 kref_put(&zone->refcnt, reada_zone_release);
 390                 spin_unlock(&fs_info->reada_lock);
 391         }
 392         re->nzones = nzones;
 393         if (nzones == 0) {
 394                 /* not a single zone found, error and out */
 395                 goto error;
 396         }
 397
 398         /* insert extent in reada_tree + all per-device trees, all or nothing */
 399         spin_lock(&fs_info->reada_lock);
 400         ret = radix_tree_insert(&fs_info->reada_tree, index, re);
 401         if (ret) {
 402                 spin_unlock(&fs_info->reada_lock);
 403                 if (ret != -ENOMEM) {
 404                         /* someone inserted the extent in the meantime */
 405                         looped = 1;
 406                 }
 407                 goto error;
 408         }
 409         for (i = 0; i < nzones; ++i) {
 410                 dev = multi->stripes[i].dev;
 411                 ret = radix_tree_insert(&dev->reada_extents, index, re);
 412                 if (ret) {
 413                         while (--i >= 0) {
 414                                 dev = multi->stripes[i].dev;
 415                                 BUG_ON(dev == NULL);
 416                                 radix_tree_delete(&dev->reada_extents, index);
 417                         }
 418                         BUG_ON(fs_info == NULL);
 419                         radix_tree_delete(&fs_info->reada_tree, index);
 420                         spin_unlock(&fs_info->reada_lock);
 421                         goto error;
 422                 }
 423         }
 424         spin_unlock(&fs_info->reada_lock);
 425
 426         return re;
 427
 428 error:
 429         while (nzones) {
 430                 struct reada_zone *zone;
 431
 432                 --nzones;
 433                 zone = re->zones[nzones];
 434                 kref_get(&zone->refcnt);
 435                 spin_lock(&zone->lock);
 436                 --zone->elems;
 437                 if (zone->elems == 0) {
 438                         /*
 439                          * no fs_info->reada_lock needed, as this can't be
 440                          * the last ref
 441                          */
 442                         kref_put(&zone->refcnt, reada_zone_release);
 443                 }
 444                 spin_unlock(&zone->lock);
 445
 446                 spin_lock(&fs_info->reada_lock);
 447                 kref_put(&zone->refcnt, reada_zone_release);
 448                 spin_unlock(&fs_info->reada_lock);
 449         }
 450         kfree(re);
 451         if (looped)
 452                 goto again;
 453         return NULL;
 454 }
 455
 456 static void reada_kref_dummy(struct kref *kr)
 457 {
 458 }
 459
 460 static void reada_extent_put(struct btrfs_fs_info *fs_info,
 461                              struct reada_extent *re)
 462 {
 463         int i;
 464         unsigned long index = re->logical >> PAGE_CACHE_SHIFT;
 465
 466         spin_lock(&fs_info->reada_lock);
 467         if (!kref_put(&re->refcnt, reada_kref_dummy)) {
 468                 spin_unlock(&fs_info->reada_lock);
 469                 return;
 470         }
 471
 472         radix_tree_delete(&fs_info->reada_tree, index);
 473         for (i = 0; i < re->nzones; ++i) {
 474                 struct reada_zone *zone = re->zones[i];
 475
 476                 radix_tree_delete(&zone->device->reada_extents, index);
 477         }
 478
 479         spin_unlock(&fs_info->reada_lock);
 480
 481         for (i = 0; i < re->nzones; ++i) {
 482                 struct reada_zone *zone = re->zones[i];
 483
 484                 kref_get(&zone->refcnt);
 485                 spin_lock(&zone->lock);
 486                 --zone->elems;
 487                 if (zone->elems == 0) {
 488                         /* no fs_info->reada_lock needed, as this can't be
 489                          * the last ref */
 490                         kref_put(&zone->refcnt, reada_zone_release);
 491                 }
 492                 spin_unlock(&zone->lock);
 493
 494                 spin_lock(&fs_info->reada_lock);
 495                 kref_put(&zone->refcnt, reada_zone_release);
 496                 spin_unlock(&fs_info->reada_lock);
 497         }
 498         if (re->scheduled_for)
 499                 atomic_dec(&re->scheduled_for->reada_in_flight);
 500
 501         kfree(re);
 502 }
 503
 504 static void reada_zone_release(struct kref *kref)
 505 {
 506         struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);
 507
 508         radix_tree_delete(&zone->device->reada_zones,
 509                           zone->end >> PAGE_CACHE_SHIFT);
 510
 511         kfree(zone);
 512 }
 513
 514 static void reada_control_release(struct kref *kref)
 515 {
 516         struct reada_control *rc = container_of(kref, struct reada_control,
 517                                                 refcnt);
 518
 519         kfree(rc);
 520 }
 521
 522 static int reada_add_block(struct reada_control *rc, u64 logical,
 523                            struct btrfs_key *top, int level, u64 generation)
 524 {
 525         struct btrfs_root *root = rc->root;
 526         struct reada_extent *re;
 527         struct reada_extctl *rec;
 528
 529         re = reada_find_extent(root, logical, top, level); /* takes one ref */
 530         if (!re)
 531                 return -1;
 532
 533         rec = kzalloc(sizeof(*rec), GFP_NOFS);
 534         if (!rec) {
 535                 reada_extent_put(root->fs_info, re);
 536                 return -1;
 537         }
 538
 539         rec->rc = rc;
 540         rec->generation = generation;
 541         atomic_inc(&rc->elems);
 542
 543         spin_lock(&re->lock);
 544         list_add_tail(&rec->list, &re->extctl);
 545         spin_unlock(&re->lock);
 546
 547         /* leave the ref on the extent */
 548
 549         return 0;
 550 }
 551
 552 /*
 553  * called with fs_info->reada_lock held
 554  */
 555 static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
 556 {
 557         int i;
 558         unsigned long index = zone->end >> PAGE_CACHE_SHIFT;
 559
 560         for (i = 0; i < zone->ndevs; ++i) {
 561                 struct reada_zone *peer;
 562                 peer = radix_tree_lookup(&zone->devs[i]->reada_zones, index);
 563                 if (peer && peer->device != zone->device)
 564                         peer->locked = lock;
 565         }
 566 }
 567
 568 /*
 569  * called with fs_info->reada_lock held
 570  */
 571 static int reada_pick_zone(struct btrfs_device *dev)
 572 {
 573         struct reada_zone *top_zone = NULL;
 574         struct reada_zone *top_locked_zone = NULL;
 575         u64 top_elems = 0;
 576         u64 top_locked_elems = 0;
 577         unsigned long index = 0;
 578         int ret;
 579
 580         if (dev->reada_curr_zone) {
 581                 reada_peer_zones_set_lock(dev->reada_curr_zone, 0);
 582                 kref_put(&dev->reada_curr_zone->refcnt, reada_zone_release);
 583                 dev->reada_curr_zone = NULL;
 584         }
 585         /* pick the zone with the most elements */
 586         while (1) {
 587                 struct reada_zone *zone;
 588
 589                 ret = radix_tree_gang_lookup(&dev->reada_zones,
 590                                              (void **)&zone, index, 1);
 591                 if (ret == 0)
 592                         break;
 593                 index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
 594                 if (zone->locked) {
 595                         if (zone->elems > top_locked_elems) {
 596                                 top_locked_elems = zone->elems;
 597                                 top_locked_zone = zone;
 598                         }
 599                 } else {
 600                         if (zone->elems > top_elems) {
 601                                 top_elems = zone->elems;
 602                                 top_zone = zone;
 603                         }
 604                 }
 605         }
 606         if (top_zone)
 607                 dev->reada_curr_zone = top_zone;
 608         else if (top_locked_zone)
 609                 dev->reada_curr_zone = top_locked_zone;
 610         else
 611                 return 0;
 612
 613         dev->reada_next = dev->reada_curr_zone->start;
 614         kref_get(&dev->reada_curr_zone->refcnt);
 615         reada_peer_zones_set_lock(dev->reada_curr_zone, 1);
 616
 617         return 1;
 618 }
 619
 620 static int reada_start_machine_dev(struct btrfs_fs_info *fs_info,
 621                                    struct btrfs_device *dev)
 622 {
 623         struct reada_extent *re = NULL;
 624         int mirror_num = 0;
 625         struct extent_buffer *eb = NULL;
 626         u64 logical;
 627         u32 blocksize;
 628         int ret;
 629         int i;
 630         int need_kick = 0;
 631
 632         spin_lock(&fs_info->reada_lock);
 633         if (dev->reada_curr_zone == NULL) {
 634                 ret = reada_pick_zone(dev);
 635                 if (!ret) {
 636                         spin_unlock(&fs_info->reada_lock);
 637                         return 0;
 638                 }
 639         }
 640         /*
 641          * FIXME currently we issue the reads one extent at a time. If we have
 642          * a contiguous block of extents, we could also coagulate them or use
 643          * plugging to speed things up
 644          */
 645         ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
 646                                      dev->reada_next >> PAGE_CACHE_SHIFT, 1);
 647         if (ret == 0 || re->logical >= dev->reada_curr_zone->end) {
 648                 ret = reada_pick_zone(dev);
 649                 if (!ret) {
 650                         spin_unlock(&fs_info->reada_lock);
 651                         return 0;
 652                 }
 653                 re = NULL;
 654                 ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
 655                                         dev->reada_next >> PAGE_CACHE_SHIFT, 1);
 656         }
 657         if (ret == 0) {
 658                 spin_unlock(&fs_info->reada_lock);
 659                 return 0;
 660         }
 661         dev->reada_next = re->logical + re->blocksize;
 662         kref_get(&re->refcnt);
 663
 664         spin_unlock(&fs_info->reada_lock);
 665
 666         /*
 667          * find mirror num
 668          */
 669         for (i = 0; i < re->nzones; ++i) {
 670                 if (re->zones[i]->device == dev) {
 671                         mirror_num = i + 1;
 672                         break;
 673                 }
 674         }
 675         logical = re->logical;
 676         blocksize = re->blocksize;
 677
 678         spin_lock(&re->lock);
 679         if (re->scheduled_for == NULL) {
 680                 re->scheduled_for = dev;
 681                 need_kick = 1;
 682         }
 683         spin_unlock(&re->lock);
 684
 685         reada_extent_put(fs_info, re);
 686
 687         if (!need_kick)
 688                 return 0;
 689
 690         atomic_inc(&dev->reada_in_flight);
 691         ret = reada_tree_block_flagged(fs_info->extent_root, logical, blocksize,
 692                          mirror_num, &eb);
 693         if (ret)
 694                 __readahead_hook(fs_info->extent_root, NULL, logical, ret);
 695         else if (eb)
 696                 __readahead_hook(fs_info->extent_root, eb, eb->start, ret);
 697
 698         if (eb)
 699                 free_extent_buffer(eb);
 700
 701         return 1;
 702
 703 }
 704
 705 static void reada_start_machine_worker(struct btrfs_work *work)
 706 {
 707         struct reada_machine_work *rmw;
 708         struct btrfs_fs_info *fs_info;
 709
 710         rmw = container_of(work, struct reada_machine_work, work);
 711         fs_info = rmw->fs_info;
 712
 713         kfree(rmw);
 714
 715         __reada_start_machine(fs_info);
 716 }
 717
 718 static void __reada_start_machine(struct btrfs_fs_info *fs_info)
 719 {
 720         struct btrfs_device *device;
 721         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 722         u64 enqueued;
 723         u64 total = 0;
 724         int i;
 725
 726         do {
 727                 enqueued = 0;
 728                 list_for_each_entry(device, &fs_devices->devices, dev_list) {
 729                         if (atomic_read(&device->reada_in_flight) <
 730                             MAX_IN_FLIGHT)
 731                                 enqueued += reada_start_machine_dev(fs_info,
 732                                                                     device);
 733                 }
 734                 total += enqueued;
 735         } while (enqueued && total < 10000);
 736
 737         if (enqueued == 0)
 738                 return;
 739
 740         /*
 741          * If everything is already in the cache, this is effectively single
 742          * threaded. To a) not hold the caller for too long and b) to utilize
 743          * more cores, we broke the loop above after 10000 iterations and now
 744          * enqueue to workers to finish it. This will distribute the load to
 745          * the cores.
 746          */
 747         for (i = 0; i < 2; ++i)
 748                 reada_start_machine(fs_info);
 749 }
 750
 751 static void reada_start_machine(struct btrfs_fs_info *fs_info)
 752 {
 753         struct reada_machine_work *rmw;
 754
 755         rmw = kzalloc(sizeof(*rmw), GFP_NOFS);
 756         if (!rmw) {
 757                 /* FIXME we cannot handle this properly right now */
 758                 BUG();
 759         }
 760         rmw->work.func = reada_start_machine_worker;
 761         rmw->fs_info = fs_info;
 762
 763         btrfs_queue_worker(&fs_info->readahead_workers, &rmw->work);
 764 }
 765
 766 #ifdef DEBUG
 767 static void dump_devs(struct btrfs_fs_info *fs_info, int all)
 768 {
 769         struct btrfs_device *device;
 770         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 771         unsigned long index;
 772         int ret;
 773         int i;
 774         int j;
 775         int cnt;
 776
 777         spin_lock(&fs_info->reada_lock);
 778         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 779                 printk(KERN_DEBUG "dev %lld has %d in flight\n", device->devid,
 780                         atomic_read(&device->reada_in_flight));
 781                 index = 0;
 782                 while (1) {
 783                         struct reada_zone *zone;
 784                         ret = radix_tree_gang_lookup(&device->reada_zones,
 785                                                      (void **)&zone, index, 1);
 786                         if (ret == 0)
 787                                 break;
 788                         printk(KERN_DEBUG "  zone %llu-%llu elems %llu locked "
 789                                 "%d devs", zone->start, zone->end, zone->elems,
 790                                 zone->locked);
 791                         for (j = 0; j < zone->ndevs; ++j) {
 792                                 printk(KERN_CONT " %lld",
 793                                         zone->devs[j]->devid);
 794                         }
 795                         if (device->reada_curr_zone == zone)
 796                                 printk(KERN_CONT " curr off %llu",
 797                                         device->reada_next - zone->start);
 798                         printk(KERN_CONT "\n");
 799                         index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
 800                 }
 801                 cnt = 0;
 802                 index = 0;
 803                 while (all) {
 804                         struct reada_extent *re = NULL;
 805
 806                         ret = radix_tree_gang_lookup(&device->reada_extents,
 807                                                      (void **)&re, index, 1);
 808                         if (ret == 0)
 809                                 break;
 810                         printk(KERN_DEBUG
 811                                 "  re: logical %llu size %u empty %d for %lld",
 812                                 re->logical, re->blocksize,
 813                                 list_empty(&re->extctl), re->scheduled_for ?
 814                                 re->scheduled_for->devid : -1);
 815
 816                         for (i = 0; i < re->nzones; ++i) {
 817                                 printk(KERN_CONT " zone %llu-%llu devs",
 818                                         re->zones[i]->start,
 819                                         re->zones[i]->end);
 820                                 for (j = 0; j < re->zones[i]->ndevs; ++j) {
 821                                         printk(KERN_CONT " %lld",
 822                                                 re->zones[i]->devs[j]->devid);
 823                                 }
 824                         }
 825                         printk(KERN_CONT "\n");
 826                         index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
 827                         if (++cnt > 15)
 828                                 break;
 829                 }
 830         }
 831
 832         index = 0;
 833         cnt = 0;
 834         while (all) {
 835                 struct reada_extent *re = NULL;
 836
 837                 ret = radix_tree_gang_lookup(&fs_info->reada_tree, (void **)&re,
 838                                              index, 1);
 839                 if (ret == 0)
 840                         break;
 841                 if (!re->scheduled_for) {
 842                         index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
 843                         continue;
 844                 }
 845                 printk(KERN_DEBUG
 846                         "re: logical %llu size %u list empty %d for %lld",
 847                         re->logical, re->blocksize, list_empty(&re->extctl),
 848                         re->scheduled_for ? re->scheduled_for->devid : -1);
 849                 for (i = 0; i < re->nzones; ++i) {
 850                         printk(KERN_CONT " zone %llu-%llu devs",
 851                                 re->zones[i]->start,
 852                                 re->zones[i]->end);
 853                         for (i = 0; i < re->nzones; ++i) {
 854                                 printk(KERN_CONT " zone %llu-%llu devs",
 855                                         re->zones[i]->start,
 856                                         re->zones[i]->end);
 857                                 for (j = 0; j < re->zones[i]->ndevs; ++j) {
 858                                         printk(KERN_CONT " %lld",
 859                                                 re->zones[i]->devs[j]->devid);
 860                                 }
 861                         }
 862                 }
 863                 printk(KERN_CONT "\n");
 864                 index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
 865         }
 866         spin_unlock(&fs_info->reada_lock);
 867 }
 868 #endif
 869
 870 /*
 871  * interface
 872  */
 873 struct reada_control *btrfs_reada_add(struct btrfs_root *root,
 874                         struct btrfs_key *key_start, struct btrfs_key *key_end)
 875 {
 876         struct reada_control *rc;
 877         u64 start;
 878         u64 generation;
 879         int level;
 880         struct extent_buffer *node;
 881         static struct btrfs_key max_key = {
 882                 .objectid = (u64)-1,
 883                 .type = (u8)-1,
 884                 .offset = (u64)-1
 885         };
 886
 887         rc = kzalloc(sizeof(*rc), GFP_NOFS);
 888         if (!rc)
 889                 return ERR_PTR(-ENOMEM);
 890
 891         rc->root = root;
 892         rc->key_start = *key_start;
 893         rc->key_end = *key_end;
 894         atomic_set(&rc->elems, 0);
 895         init_waitqueue_head(&rc->wait);
 896         kref_init(&rc->refcnt);
 897         kref_get(&rc->refcnt); /* one ref for having elements */
 898
 899         node = btrfs_root_node(root);
 900         start = node->start;
 901         level = btrfs_header_level(node);
 902         generation = btrfs_header_generation(node);
 903         free_extent_buffer(node);
 904
 905         reada_add_block(rc, start, &max_key, level, generation);
 906
 907         reada_start_machine(root->fs_info);
 908
 909         return rc;
 910 }
 911
 912 #ifdef DEBUG
 913 int btrfs_reada_wait(void *handle)
 914 {
 915         struct reada_control *rc = handle;
 916
 917         while (atomic_read(&rc->elems)) {
 918                 wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
 919                                    5 * HZ);
 920                 dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
 921         }
 922
 923         dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
 924
 925         kref_put(&rc->refcnt, reada_control_release);
 926
 927         return 0;
 928 }
 929 #else
 930 int btrfs_reada_wait(void *handle)
 931 {
 932         struct reada_control *rc = handle;
 933
 934         while (atomic_read(&rc->elems)) {
 935                 wait_event(rc->wait, atomic_read(&rc->elems) == 0);
 936         }
 937
 938         kref_put(&rc->refcnt, reada_control_release);
 939
 940         return 0;
 941 }
 942 #endif
 943
 944 void btrfs_reada_detach(void *handle)
 945 {
 946         struct reada_control *rc = handle;
 947
 948         kref_put(&rc->refcnt, reada_control_release);
 949 }