mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / btrfs / inode-map.c
blob2ae32451fb5b020034cc5cf47699a2004ccb5679
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/delay.h>
20 #include <linux/kthread.h>
21 #include <linux/pagemap.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "free-space-cache.h"
26 #include "inode-map.h"
27 #include "transaction.h"
29 static void fail_caching_thread(struct btrfs_root *root)
31 struct btrfs_fs_info *fs_info = root->fs_info;
33 btrfs_warn(fs_info, "failed to start inode caching task");
34 btrfs_clear_pending_and_info(fs_info, INODE_MAP_CACHE,
35 "disabling inode map caching");
36 spin_lock(&root->ino_cache_lock);
37 root->ino_cache_state = BTRFS_CACHE_ERROR;
38 spin_unlock(&root->ino_cache_lock);
39 wake_up(&root->ino_cache_wait);
42 static int caching_kthread(void *data)
44 struct btrfs_root *root = data;
45 struct btrfs_fs_info *fs_info = root->fs_info;
46 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
47 struct btrfs_key key;
48 struct btrfs_path *path;
49 struct extent_buffer *leaf;
50 u64 last = (u64)-1;
51 int slot;
52 int ret;
54 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
55 return 0;
57 path = btrfs_alloc_path();
58 if (!path) {
59 fail_caching_thread(root);
60 return -ENOMEM;
63 /* Since the commit root is read-only, we can safely skip locking. */
64 path->skip_locking = 1;
65 path->search_commit_root = 1;
66 path->reada = READA_FORWARD;
68 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
69 key.offset = 0;
70 key.type = BTRFS_INODE_ITEM_KEY;
71 again:
72 /* need to make sure the commit_root doesn't disappear */
73 down_read(&fs_info->commit_root_sem);
75 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
76 if (ret < 0)
77 goto out;
79 while (1) {
80 if (btrfs_fs_closing(fs_info))
81 goto out;
83 leaf = path->nodes[0];
84 slot = path->slots[0];
85 if (slot >= btrfs_header_nritems(leaf)) {
86 ret = btrfs_next_leaf(root, path);
87 if (ret < 0)
88 goto out;
89 else if (ret > 0)
90 break;
92 if (need_resched() ||
93 btrfs_transaction_in_commit(fs_info)) {
94 leaf = path->nodes[0];
96 if (WARN_ON(btrfs_header_nritems(leaf) == 0))
97 break;
100 * Save the key so we can advances forward
101 * in the next search.
103 btrfs_item_key_to_cpu(leaf, &key, 0);
104 btrfs_release_path(path);
105 root->ino_cache_progress = last;
106 up_read(&fs_info->commit_root_sem);
107 schedule_timeout(1);
108 goto again;
109 } else
110 continue;
113 btrfs_item_key_to_cpu(leaf, &key, slot);
115 if (key.type != BTRFS_INODE_ITEM_KEY)
116 goto next;
118 if (key.objectid >= root->highest_objectid)
119 break;
121 if (last != (u64)-1 && last + 1 != key.objectid) {
122 __btrfs_add_free_space(fs_info, ctl, last + 1,
123 key.objectid - last - 1);
124 wake_up(&root->ino_cache_wait);
127 last = key.objectid;
128 next:
129 path->slots[0]++;
132 if (last < root->highest_objectid - 1) {
133 __btrfs_add_free_space(fs_info, ctl, last + 1,
134 root->highest_objectid - last - 1);
137 spin_lock(&root->ino_cache_lock);
138 root->ino_cache_state = BTRFS_CACHE_FINISHED;
139 spin_unlock(&root->ino_cache_lock);
141 root->ino_cache_progress = (u64)-1;
142 btrfs_unpin_free_ino(root);
143 out:
144 wake_up(&root->ino_cache_wait);
145 up_read(&fs_info->commit_root_sem);
147 btrfs_free_path(path);
149 return ret;
152 static void start_caching(struct btrfs_root *root)
154 struct btrfs_fs_info *fs_info = root->fs_info;
155 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
156 struct task_struct *tsk;
157 int ret;
158 u64 objectid;
160 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
161 return;
163 spin_lock(&root->ino_cache_lock);
164 if (root->ino_cache_state != BTRFS_CACHE_NO) {
165 spin_unlock(&root->ino_cache_lock);
166 return;
169 root->ino_cache_state = BTRFS_CACHE_STARTED;
170 spin_unlock(&root->ino_cache_lock);
172 ret = load_free_ino_cache(fs_info, root);
173 if (ret == 1) {
174 spin_lock(&root->ino_cache_lock);
175 root->ino_cache_state = BTRFS_CACHE_FINISHED;
176 spin_unlock(&root->ino_cache_lock);
177 wake_up(&root->ino_cache_wait);
178 return;
182 * It can be quite time-consuming to fill the cache by searching
183 * through the extent tree, and this can keep ino allocation path
184 * waiting. Therefore at start we quickly find out the highest
185 * inode number and we know we can use inode numbers which fall in
186 * [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
188 ret = btrfs_find_free_objectid(root, &objectid);
189 if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
190 __btrfs_add_free_space(fs_info, ctl, objectid,
191 BTRFS_LAST_FREE_OBJECTID - objectid + 1);
194 tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu",
195 root->root_key.objectid);
196 if (IS_ERR(tsk))
197 fail_caching_thread(root);
200 int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
202 if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
203 return btrfs_find_free_objectid(root, objectid);
205 again:
206 *objectid = btrfs_find_ino_for_alloc(root);
208 if (*objectid != 0)
209 return 0;
211 start_caching(root);
213 wait_event(root->ino_cache_wait,
214 root->ino_cache_state == BTRFS_CACHE_FINISHED ||
215 root->ino_cache_state == BTRFS_CACHE_ERROR ||
216 root->free_ino_ctl->free_space > 0);
218 if (root->ino_cache_state == BTRFS_CACHE_FINISHED &&
219 root->free_ino_ctl->free_space == 0)
220 return -ENOSPC;
221 else if (root->ino_cache_state == BTRFS_CACHE_ERROR)
222 return btrfs_find_free_objectid(root, objectid);
223 else
224 goto again;
227 void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
229 struct btrfs_fs_info *fs_info = root->fs_info;
230 struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
232 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
233 return;
234 again:
235 if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
236 __btrfs_add_free_space(fs_info, pinned, objectid, 1);
237 } else {
238 down_write(&fs_info->commit_root_sem);
239 spin_lock(&root->ino_cache_lock);
240 if (root->ino_cache_state == BTRFS_CACHE_FINISHED) {
241 spin_unlock(&root->ino_cache_lock);
242 up_write(&fs_info->commit_root_sem);
243 goto again;
245 spin_unlock(&root->ino_cache_lock);
247 start_caching(root);
249 __btrfs_add_free_space(fs_info, pinned, objectid, 1);
251 up_write(&fs_info->commit_root_sem);
256 * When a transaction is committed, we'll move those inode numbers which are
257 * smaller than root->ino_cache_progress from pinned tree to free_ino tree, and
258 * others will just be dropped, because the commit root we were searching has
259 * changed.
261 * Must be called with root->fs_info->commit_root_sem held
263 void btrfs_unpin_free_ino(struct btrfs_root *root)
265 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
266 struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
267 spinlock_t *rbroot_lock = &root->free_ino_pinned->tree_lock;
268 struct btrfs_free_space *info;
269 struct rb_node *n;
270 u64 count;
272 if (!btrfs_test_opt(root->fs_info, INODE_MAP_CACHE))
273 return;
275 while (1) {
276 bool add_to_ctl = true;
278 spin_lock(rbroot_lock);
279 n = rb_first(rbroot);
280 if (!n) {
281 spin_unlock(rbroot_lock);
282 break;
285 info = rb_entry(n, struct btrfs_free_space, offset_index);
286 BUG_ON(info->bitmap); /* Logic error */
288 if (info->offset > root->ino_cache_progress)
289 add_to_ctl = false;
290 else if (info->offset + info->bytes > root->ino_cache_progress)
291 count = root->ino_cache_progress - info->offset + 1;
292 else
293 count = info->bytes;
295 rb_erase(&info->offset_index, rbroot);
296 spin_unlock(rbroot_lock);
297 if (add_to_ctl)
298 __btrfs_add_free_space(root->fs_info, ctl,
299 info->offset, count);
300 kmem_cache_free(btrfs_free_space_cachep, info);
304 #define INIT_THRESHOLD ((SZ_32K / 2) / sizeof(struct btrfs_free_space))
305 #define INODES_PER_BITMAP (PAGE_SIZE * 8)
308 * The goal is to keep the memory used by the free_ino tree won't
309 * exceed the memory if we use bitmaps only.
311 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
313 struct btrfs_free_space *info;
314 struct rb_node *n;
315 int max_ino;
316 int max_bitmaps;
318 n = rb_last(&ctl->free_space_offset);
319 if (!n) {
320 ctl->extents_thresh = INIT_THRESHOLD;
321 return;
323 info = rb_entry(n, struct btrfs_free_space, offset_index);
326 * Find the maximum inode number in the filesystem. Note we
327 * ignore the fact that this can be a bitmap, because we are
328 * not doing precise calculation.
330 max_ino = info->bytes - 1;
332 max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
333 if (max_bitmaps <= ctl->total_bitmaps) {
334 ctl->extents_thresh = 0;
335 return;
338 ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
339 PAGE_SIZE / sizeof(*info);
343 * We don't fall back to bitmap, if we are below the extents threshold
344 * or this chunk of inode numbers is a big one.
346 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
347 struct btrfs_free_space *info)
349 if (ctl->free_extents < ctl->extents_thresh ||
350 info->bytes > INODES_PER_BITMAP / 10)
351 return false;
353 return true;
356 static const struct btrfs_free_space_op free_ino_op = {
357 .recalc_thresholds = recalculate_thresholds,
358 .use_bitmap = use_bitmap,
361 static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
365 static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
366 struct btrfs_free_space *info)
369 * We always use extents for two reasons:
371 * - The pinned tree is only used during the process of caching
372 * work.
373 * - Make code simpler. See btrfs_unpin_free_ino().
375 return false;
378 static const struct btrfs_free_space_op pinned_free_ino_op = {
379 .recalc_thresholds = pinned_recalc_thresholds,
380 .use_bitmap = pinned_use_bitmap,
383 void btrfs_init_free_ino_ctl(struct btrfs_root *root)
385 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
386 struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
388 spin_lock_init(&ctl->tree_lock);
389 ctl->unit = 1;
390 ctl->start = 0;
391 ctl->private = NULL;
392 ctl->op = &free_ino_op;
393 INIT_LIST_HEAD(&ctl->trimming_ranges);
394 mutex_init(&ctl->cache_writeout_mutex);
397 * Initially we allow to use 16K of ram to cache chunks of
398 * inode numbers before we resort to bitmaps. This is somewhat
399 * arbitrary, but it will be adjusted in runtime.
401 ctl->extents_thresh = INIT_THRESHOLD;
403 spin_lock_init(&pinned->tree_lock);
404 pinned->unit = 1;
405 pinned->start = 0;
406 pinned->private = NULL;
407 pinned->extents_thresh = 0;
408 pinned->op = &pinned_free_ino_op;
411 int btrfs_save_ino_cache(struct btrfs_root *root,
412 struct btrfs_trans_handle *trans)
414 struct btrfs_fs_info *fs_info = root->fs_info;
415 struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
416 struct btrfs_path *path;
417 struct inode *inode;
418 struct btrfs_block_rsv *rsv;
419 struct extent_changeset *data_reserved = NULL;
420 u64 num_bytes;
421 u64 alloc_hint = 0;
422 int ret;
423 int prealloc;
424 bool retry = false;
426 /* only fs tree and subvol/snap needs ino cache */
427 if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
428 (root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
429 root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
430 return 0;
432 /* Don't save inode cache if we are deleting this root */
433 if (btrfs_root_refs(&root->root_item) == 0)
434 return 0;
436 if (!btrfs_test_opt(fs_info, INODE_MAP_CACHE))
437 return 0;
439 path = btrfs_alloc_path();
440 if (!path)
441 return -ENOMEM;
443 rsv = trans->block_rsv;
444 trans->block_rsv = &fs_info->trans_block_rsv;
446 num_bytes = trans->bytes_reserved;
448 * 1 item for inode item insertion if need
449 * 4 items for inode item update (in the worst case)
450 * 1 items for slack space if we need do truncation
451 * 1 item for free space object
452 * 3 items for pre-allocation
454 trans->bytes_reserved = btrfs_calc_trans_metadata_size(fs_info, 10);
455 ret = btrfs_block_rsv_add(root, trans->block_rsv,
456 trans->bytes_reserved,
457 BTRFS_RESERVE_NO_FLUSH);
458 if (ret)
459 goto out;
460 trace_btrfs_space_reservation(fs_info, "ino_cache", trans->transid,
461 trans->bytes_reserved, 1);
462 again:
463 inode = lookup_free_ino_inode(root, path);
464 if (IS_ERR(inode) && (PTR_ERR(inode) != -ENOENT || retry)) {
465 ret = PTR_ERR(inode);
466 goto out_release;
469 if (IS_ERR(inode)) {
470 BUG_ON(retry); /* Logic error */
471 retry = true;
473 ret = create_free_ino_inode(root, trans, path);
474 if (ret)
475 goto out_release;
476 goto again;
479 BTRFS_I(inode)->generation = 0;
480 ret = btrfs_update_inode(trans, root, inode);
481 if (ret) {
482 btrfs_abort_transaction(trans, ret);
483 goto out_put;
486 if (i_size_read(inode) > 0) {
487 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
488 if (ret) {
489 if (ret != -ENOSPC)
490 btrfs_abort_transaction(trans, ret);
491 goto out_put;
495 spin_lock(&root->ino_cache_lock);
496 if (root->ino_cache_state != BTRFS_CACHE_FINISHED) {
497 ret = -1;
498 spin_unlock(&root->ino_cache_lock);
499 goto out_put;
501 spin_unlock(&root->ino_cache_lock);
503 spin_lock(&ctl->tree_lock);
504 prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
505 prealloc = ALIGN(prealloc, PAGE_SIZE);
506 prealloc += ctl->total_bitmaps * PAGE_SIZE;
507 spin_unlock(&ctl->tree_lock);
509 /* Just to make sure we have enough space */
510 prealloc += 8 * PAGE_SIZE;
512 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, 0, prealloc);
513 if (ret)
514 goto out_put;
516 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
517 prealloc, prealloc, &alloc_hint);
518 if (ret) {
519 btrfs_delalloc_release_metadata(BTRFS_I(inode), prealloc);
520 goto out_put;
523 ret = btrfs_write_out_ino_cache(root, trans, path, inode);
524 out_put:
525 iput(inode);
526 out_release:
527 trace_btrfs_space_reservation(fs_info, "ino_cache", trans->transid,
528 trans->bytes_reserved, 0);
529 btrfs_block_rsv_release(fs_info, trans->block_rsv,
530 trans->bytes_reserved);
531 out:
532 trans->block_rsv = rsv;
533 trans->bytes_reserved = num_bytes;
535 btrfs_free_path(path);
536 extent_changeset_free(data_reserved);
537 return ret;
540 int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
542 struct btrfs_path *path;
543 int ret;
544 struct extent_buffer *l;
545 struct btrfs_key search_key;
546 struct btrfs_key found_key;
547 int slot;
549 path = btrfs_alloc_path();
550 if (!path)
551 return -ENOMEM;
553 search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
554 search_key.type = -1;
555 search_key.offset = (u64)-1;
556 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
557 if (ret < 0)
558 goto error;
559 BUG_ON(ret == 0); /* Corruption */
560 if (path->slots[0] > 0) {
561 slot = path->slots[0] - 1;
562 l = path->nodes[0];
563 btrfs_item_key_to_cpu(l, &found_key, slot);
564 *objectid = max_t(u64, found_key.objectid,
565 BTRFS_FIRST_FREE_OBJECTID - 1);
566 } else {
567 *objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
569 ret = 0;
570 error:
571 btrfs_free_path(path);
572 return ret;
575 int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
577 int ret;
578 mutex_lock(&root->objectid_mutex);
580 if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
581 btrfs_warn(root->fs_info,
582 "the objectid of root %llu reaches its highest value",
583 root->root_key.objectid);
584 ret = -ENOSPC;
585 goto out;
588 *objectid = ++root->highest_objectid;
589 ret = 0;
590 out:
591 mutex_unlock(&root->objectid_mutex);
592 return ret;