fs/btrfs/inode-map.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  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/delay.h>
  20 #include <linux/kthread.h>
  21 #include <linux/pagemap.h>
  22
  23 #include "ctree.h"
  24 #include "disk-io.h"
  25 #include "free-space-cache.h"
  26 #include "inode-map.h"
  27 #include "transaction.h"
  28
  29 static int caching_kthread(void *data)
  30 {
  31         struct btrfs_root *root = data;
  32         struct btrfs_fs_info *fs_info = root->fs_info;
  33         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
  34         struct btrfs_key key;
  35         struct btrfs_path *path;
  36         struct extent_buffer *leaf;
  37         u64 last = (u64)-1;
  38         int slot;
  39         int ret;
  40
  41         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
  42                 return 0;
  43
  44         path = btrfs_alloc_path();
  45         if (!path)
  46                 return -ENOMEM;
  47
  48         /* Since the commit root is read-only, we can safely skip locking. */
  49         path->skip_locking = 1;
  50         path->search_commit_root = 1;
  51         path->reada = 2;
  52
  53         key.objectid = BTRFS_FIRST_FREE_OBJECTID;
  54         key.offset = 0;
  55         key.type = BTRFS_INODE_ITEM_KEY;
  56 again:
  57         /* need to make sure the commit_root doesn't disappear */
  58         down_read(&fs_info->commit_root_sem);
  59
  60         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
  61         if (ret < 0)
  62                 goto out;
  63
  64         while (1) {
  65                 if (btrfs_fs_closing(fs_info))
  66                         goto out;
  67
  68                 leaf = path->nodes[0];
  69                 slot = path->slots[0];
  70                 if (slot >= btrfs_header_nritems(leaf)) {
  71                         ret = btrfs_next_leaf(root, path);
  72                         if (ret < 0)
  73                                 goto out;
  74                         else if (ret > 0)
  75                                 break;
  76
  77                         if (need_resched() ||
  78                             btrfs_transaction_in_commit(fs_info)) {
  79                                 leaf = path->nodes[0];
  80
  81                                 if (WARN_ON(btrfs_header_nritems(leaf) == 0))
  82                                         break;
  83
  84                                 /*
  85                                  * Save the key so we can advances forward
  86                                  * in the next search.
  87                                  */
  88                                 btrfs_item_key_to_cpu(leaf, &key, 0);
  89                                 btrfs_release_path(path);
  90                                 root->ino_cache_progress = last;
  91                                 up_read(&fs_info->commit_root_sem);
  92                                 schedule_timeout(1);
  93                                 goto again;
  94                         } else
  95                                 continue;
  96                 }
  97
  98                 btrfs_item_key_to_cpu(leaf, &key, slot);
  99
 100                 if (key.type != BTRFS_INODE_ITEM_KEY)
 101                         goto next;
 102
 103                 if (key.objectid >= root->highest_objectid)
 104                         break;
 105
 106                 if (last != (u64)-1 && last + 1 != key.objectid) {
 107                         __btrfs_add_free_space(ctl, last + 1,
 108                                                key.objectid - last - 1);
 109                         wake_up(&root->ino_cache_wait);
 110                 }
 111
 112                 last = key.objectid;
 113 next:
 114                 path->slots[0]++;
 115         }
 116
 117         if (last < root->highest_objectid - 1) {
 118                 __btrfs_add_free_space(ctl, last + 1,
 119                                        root->highest_objectid - last - 1);
 120         }
 121
 122         spin_lock(&root->ino_cache_lock);
 123         root->ino_cache_state = BTRFS_CACHE_FINISHED;
 124         spin_unlock(&root->ino_cache_lock);
 125
 126         root->ino_cache_progress = (u64)-1;
 127         btrfs_unpin_free_ino(root);
 128 out:
 129         wake_up(&root->ino_cache_wait);
 130         up_read(&fs_info->commit_root_sem);
 131
 132         btrfs_free_path(path);
 133
 134         return ret;
 135 }
 136
 137 static void start_caching(struct btrfs_root *root)
 138 {
 139         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
 140         struct task_struct *tsk;
 141         int ret;
 142         u64 objectid;
 143
 144         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 145                 return;
 146
 147         spin_lock(&root->ino_cache_lock);
 148         if (root->ino_cache_state != BTRFS_CACHE_NO) {
 149                 spin_unlock(&root->ino_cache_lock);
 150                 return;
 151         }
 152
 153         root->ino_cache_state = BTRFS_CACHE_STARTED;
 154         spin_unlock(&root->ino_cache_lock);
 155
 156         ret = load_free_ino_cache(root->fs_info, root);
 157         if (ret == 1) {
 158                 spin_lock(&root->ino_cache_lock);
 159                 root->ino_cache_state = BTRFS_CACHE_FINISHED;
 160                 spin_unlock(&root->ino_cache_lock);
 161                 return;
 162         }
 163
 164         /*
 165          * It can be quite time-consuming to fill the cache by searching
 166          * through the extent tree, and this can keep ino allocation path
 167          * waiting. Therefore at start we quickly find out the highest
 168          * inode number and we know we can use inode numbers which fall in
 169          * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
 170          */
 171         ret = btrfs_find_free_objectid(root, &objectid);
 172         if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
 173                 __btrfs_add_free_space(ctl, objectid,
 174                                        BTRFS_LAST_FREE_OBJECTID - objectid + 1);
 175         }
 176
 177         tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu",
 178                           root->root_key.objectid);
 179         if (IS_ERR(tsk)) {
 180                 btrfs_warn(root->fs_info, "failed to start inode caching task");
 181                 btrfs_clear_pending_and_info(root->fs_info, INODE_MAP_CACHE,
 182                                 "disabling inode map caching");
 183         }
 184 }
 185
 186 int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
 187 {
 188         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 189                 return btrfs_find_free_objectid(root, objectid);
 190
 191 again:
 192         *objectid = btrfs_find_ino_for_alloc(root);
 193
 194         if (*objectid != 0)
 195                 return 0;
 196
 197         start_caching(root);
 198
 199         wait_event(root->ino_cache_wait,
 200                    root->ino_cache_state == BTRFS_CACHE_FINISHED ||
 201                    root->free_ino_ctl->free_space > 0);
 202
 203         if (root->ino_cache_state == BTRFS_CACHE_FINISHED &&
 204             root->free_ino_ctl->free_space == 0)
 205                 return -ENOSPC;
 206         else
 207                 goto again;
 208 }
 209
 210 void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
 211 {
 212         struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
 213
 214         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 215                 return;
 216 again:
 217         if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
 218                 __btrfs_add_free_space(pinned, objectid, 1);
 219         } else {
 220                 down_write(&root->fs_info->commit_root_sem);
 221                 spin_lock(&root->ino_cache_lock);
 222                 if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
 223                         spin_unlock(&root->ino_cache_lock);
 224                         up_write(&root->fs_info->commit_root_sem);
 225                         goto again;
 226                 }
 227                 spin_unlock(&root->ino_cache_lock);
 228
 229                 start_caching(root);
 230
 231                 __btrfs_add_free_space(pinned, objectid, 1);
 232
 233                 up_write(&root->fs_info->commit_root_sem);
 234         }
 235 }
 236
 237 /*
 238  * When a transaction is committed, we'll move those inode numbers which are
 239  * smaller than root->ino_cache_progress from pinned tree to free_ino tree, and
 240  * others will just be dropped, because the commit root we were searching has
 241  * changed.
 242  *
 243  * Must be called with root->fs_info->commit_root_sem held
 244  */
 245 void btrfs_unpin_free_ino(struct btrfs_root *root)
 246 {
 247         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
 248         struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
 249         spinlock_t *rbroot_lock = &root->free_ino_pinned->tree_lock;
 250         struct btrfs_free_space *info;
 251         struct rb_node *n;
 252         u64 count;
 253
 254         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 255                 return;
 256
 257         while (1) {
 258                 bool add_to_ctl = true;
 259
 260                 spin_lock(rbroot_lock);
 261                 n = rb_first(rbroot);
 262                 if (!n) {
 263                         spin_unlock(rbroot_lock);
 264                         break;
 265                 }
 266
 267                 info = rb_entry(n, struct btrfs_free_space, offset_index);
 268                 BUG_ON(info->bitmap); /* Logic error */
 269
 270                 if (info->offset > root->ino_cache_progress)
 271                         add_to_ctl = false;
 272                 else if (info->offset + info->bytes > root->ino_cache_progress)
 273                         count = root->ino_cache_progress - info->offset + 1;
 274                 else
 275                         count = info->bytes;
 276
 277                 rb_erase(&info->offset_index, rbroot);
 278                 spin_unlock(rbroot_lock);
 279                 if (add_to_ctl)
 280                         __btrfs_add_free_space(ctl, info->offset, count);
 281                 kmem_cache_free(btrfs_free_space_cachep, info);
 282         }
 283 }
 284
 285 #define INIT_THRESHOLD  (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
 286 #define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
 287
 288 /*
 289  * The goal is to keep the memory used by the free_ino tree won't
 290  * exceed the memory if we use bitmaps only.
 291  */
 292 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
 293 {
 294         struct btrfs_free_space *info;
 295         struct rb_node *n;
 296         int max_ino;
 297         int max_bitmaps;
 298
 299         n = rb_last(&ctl->free_space_offset);
 300         if (!n) {
 301                 ctl->extents_thresh = INIT_THRESHOLD;
 302                 return;
 303         }
 304         info = rb_entry(n, struct btrfs_free_space, offset_index);
 305
 306         /*
 307          * Find the maximum inode number in the filesystem. Note we
 308          * ignore the fact that this can be a bitmap, because we are
 309          * not doing precise calculation.
 310          */
 311         max_ino = info->bytes - 1;
 312
 313         max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
 314         if (max_bitmaps <= ctl->total_bitmaps) {
 315                 ctl->extents_thresh = 0;
 316                 return;
 317         }
 318
 319         ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
 320                                 PAGE_CACHE_SIZE / sizeof(*info);
 321 }
 322
 323 /*
 324  * We don't fall back to bitmap, if we are below the extents threshold
 325  * or this chunk of inode numbers is a big one.
 326  */
 327 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
 328                        struct btrfs_free_space *info)
 329 {
 330         if (ctl->free_extents < ctl->extents_thresh ||
 331             info->bytes > INODES_PER_BITMAP / 10)
 332                 return false;
 333
 334         return true;
 335 }
 336
 337 static struct btrfs_free_space_op free_ino_op = {
 338         .recalc_thresholds      = recalculate_thresholds,
 339         .use_bitmap             = use_bitmap,
 340 };
 341
 342 static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
 343 {
 344 }
 345
 346 static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
 347                               struct btrfs_free_space *info)
 348 {
 349         /*
 350          * We always use extents for two reasons:
 351          *
 352          * - The pinned tree is only used during the process of caching
 353          *   work.
 354          * - Make code simpler. See btrfs_unpin_free_ino().
 355          */
 356         return false;
 357 }
 358
 359 static struct btrfs_free_space_op pinned_free_ino_op = {
 360         .recalc_thresholds      = pinned_recalc_thresholds,
 361         .use_bitmap             = pinned_use_bitmap,
 362 };
 363
 364 void btrfs_init_free_ino_ctl(struct btrfs_root *root)
 365 {
 366         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
 367         struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
 368
 369         spin_lock_init(&ctl->tree_lock);
 370         ctl->unit = 1;
 371         ctl->start = 0;
 372         ctl->private = NULL;
 373         ctl->op = &free_ino_op;
 374         INIT_LIST_HEAD(&ctl->trimming_ranges);
 375         mutex_init(&ctl->cache_writeout_mutex);
 376
 377         /*
 378          * Initially we allow to use 16K of ram to cache chunks of
 379          * inode numbers before we resort to bitmaps. This is somewhat
 380          * arbitrary, but it will be adjusted in runtime.
 381          */
 382         ctl->extents_thresh = INIT_THRESHOLD;
 383
 384         spin_lock_init(&pinned->tree_lock);
 385         pinned->unit = 1;
 386         pinned->start = 0;
 387         pinned->private = NULL;
 388         pinned->extents_thresh = 0;
 389         pinned->op = &pinned_free_ino_op;
 390 }
 391
 392 int btrfs_save_ino_cache(struct btrfs_root *root,
 393                          struct btrfs_trans_handle *trans)
 394 {
 395         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
 396         struct btrfs_path *path;
 397         struct inode *inode;
 398         struct btrfs_block_rsv *rsv;
 399         u64 num_bytes;
 400         u64 alloc_hint = 0;
 401         int ret;
 402         int prealloc;
 403         bool retry = false;
 404
 405         /* only fs tree and subvol/snap needs ino cache */
 406         if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
 407             (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
 408              root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
 409                 return 0;
 410
 411         /* Don't save inode cache if we are deleting this root */
 412         if (btrfs_root_refs(&root->root_item) == 0)
 413                 return 0;
 414
 415         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 416                 return 0;
 417
 418         path = btrfs_alloc_path();
 419         if (!path)
 420                 return -ENOMEM;
 421
 422         rsv = trans->block_rsv;
 423         trans->block_rsv = &root->fs_info->trans_block_rsv;
 424
 425         num_bytes = trans->bytes_reserved;
 426         /*
 427          * 1 item for inode item insertion if need
 428          * 4 items for inode item update (in the worst case)
 429          * 1 items for slack space if we need do truncation
 430          * 1 item for free space object
 431          * 3 items for pre-allocation
 432          */
 433         trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 10);
 434         ret = btrfs_block_rsv_add(root, trans->block_rsv,
 435                                   trans->bytes_reserved,
 436                                   BTRFS_RESERVE_NO_FLUSH);
 437         if (ret)
 438                 goto out;
 439         trace_btrfs_space_reservation(root->fs_info, "ino_cache",
 440                                       trans->transid, trans->bytes_reserved, 1);
 441 again:
 442         inode = lookup_free_ino_inode(root, path);
 443         if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
 444                 ret = PTR_ERR(inode);
 445                 goto out_release;
 446         }
 447
 448         if (IS_ERR(inode)) {
 449                 BUG_ON(retry); /* Logic error */
 450                 retry = true;
 451
 452                 ret = create_free_ino_inode(root, trans, path);
 453                 if (ret)
 454                         goto out_release;
 455                 goto again;
 456         }
 457
 458         BTRFS_I(inode)->generation = 0;
 459         ret = btrfs_update_inode(trans, root, inode);
 460         if (ret) {
 461                 btrfs_abort_transaction(trans, root, ret);
 462                 goto out_put;
 463         }
 464
 465         if (i_size_read(inode) > 0) {
 466                 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
 467                 if (ret) {
 468                         if (ret != -ENOSPC)
 469                                 btrfs_abort_transaction(trans, root, ret);
 470                         goto out_put;
 471                 }
 472         }
 473
 474         spin_lock(&root->ino_cache_lock);
 475         if (root->ino_cache_state != BTRFS_CACHE_FINISHED) {
 476                 ret = -1;
 477                 spin_unlock(&root->ino_cache_lock);
 478                 goto out_put;
 479         }
 480         spin_unlock(&root->ino_cache_lock);
 481
 482         spin_lock(&ctl->tree_lock);
 483         prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
 484         prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
 485         prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
 486         spin_unlock(&ctl->tree_lock);
 487
 488         /* Just to make sure we have enough space */
 489         prealloc += 8 * PAGE_CACHE_SIZE;
 490
 491         ret = btrfs_delalloc_reserve_space(inode, prealloc);
 492         if (ret)
 493                 goto out_put;
 494
 495         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
 496                                               prealloc, prealloc, &alloc_hint);
 497         if (ret) {
 498                 btrfs_delalloc_release_space(inode, prealloc);
 499                 goto out_put;
 500         }
 501         btrfs_free_reserved_data_space(inode, prealloc);
 502
 503         ret = btrfs_write_out_ino_cache(root, trans, path, inode);
 504 out_put:
 505         iput(inode);
 506 out_release:
 507         trace_btrfs_space_reservation(root->fs_info, "ino_cache",
 508                                       trans->transid, trans->bytes_reserved, 0);
 509         btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
 510 out:
 511         trans->block_rsv = rsv;
 512         trans->bytes_reserved = num_bytes;
 513
 514         btrfs_free_path(path);
 515         return ret;
 516 }
 517
 518 static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
 519 {
 520         struct btrfs_path *path;
 521         int ret;
 522         struct extent_buffer *l;
 523         struct btrfs_key search_key;
 524         struct btrfs_key found_key;
 525         int slot;
 526
 527         path = btrfs_alloc_path();
 528         if (!path)
 529                 return -ENOMEM;
 530
 531         search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
 532         search_key.type = -1;
 533         search_key.offset = (u64)-1;
 534         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
 535         if (ret < 0)
 536                 goto error;
 537         BUG_ON(ret == 0); /* Corruption */
 538         if (path->slots[0] > 0) {
 539                 slot = path->slots[0] - 1;
 540                 l = path->nodes[0];
 541                 btrfs_item_key_to_cpu(l, &found_key, slot);
 542                 *objectid = max_t(u64, found_key.objectid,
 543                                   BTRFS_FIRST_FREE_OBJECTID - 1);
 544         } else {
 545                 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
 546         }
 547         ret = 0;
 548 error:
 549         btrfs_free_path(path);
 550         return ret;
 551 }
 552
 553 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
 554 {
 555         int ret;
 556         mutex_lock(&root->objectid_mutex);
 557
 558         if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
 559                 ret = btrfs_find_highest_objectid(root,
 560                                                   &root->highest_objectid);
 561                 if (ret)
 562                         goto out;
 563         }
 564
 565         if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
 566                 ret = -ENOSPC;
 567                 goto out;
 568         }
 569
 570         *objectid = ++root->highest_objectid;
 571         ret = 0;
 572 out:
 573         mutex_unlock(&root->objectid_mutex);
 574         return ret;
 575 }