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