Btrfs: rename the scrub context structure
[linux-3.10.git] / fs / btrfs / free-space-cache.c
1 /*
2  * Copyright (C) 2008 Red Hat.  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/pagemap.h>
20 #include <linux/sched.h>
21 #include <linux/slab.h>
22 #include <linux/math64.h>
23 #include <linux/ratelimit.h>
24 #include "ctree.h"
25 #include "free-space-cache.h"
26 #include "transaction.h"
27 #include "disk-io.h"
28 #include "extent_io.h"
29 #include "inode-map.h"
30
31 #define BITS_PER_BITMAP         (PAGE_CACHE_SIZE * 8)
32 #define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
33
34 static int link_free_space(struct btrfs_free_space_ctl *ctl,
35                            struct btrfs_free_space *info);
36 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
37                               struct btrfs_free_space *info);
38
39 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
40                                                struct btrfs_path *path,
41                                                u64 offset)
42 {
43         struct btrfs_key key;
44         struct btrfs_key location;
45         struct btrfs_disk_key disk_key;
46         struct btrfs_free_space_header *header;
47         struct extent_buffer *leaf;
48         struct inode *inode = NULL;
49         int ret;
50
51         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
52         key.offset = offset;
53         key.type = 0;
54
55         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
56         if (ret < 0)
57                 return ERR_PTR(ret);
58         if (ret > 0) {
59                 btrfs_release_path(path);
60                 return ERR_PTR(-ENOENT);
61         }
62
63         leaf = path->nodes[0];
64         header = btrfs_item_ptr(leaf, path->slots[0],
65                                 struct btrfs_free_space_header);
66         btrfs_free_space_key(leaf, header, &disk_key);
67         btrfs_disk_key_to_cpu(&location, &disk_key);
68         btrfs_release_path(path);
69
70         inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
71         if (!inode)
72                 return ERR_PTR(-ENOENT);
73         if (IS_ERR(inode))
74                 return inode;
75         if (is_bad_inode(inode)) {
76                 iput(inode);
77                 return ERR_PTR(-ENOENT);
78         }
79
80         mapping_set_gfp_mask(inode->i_mapping,
81                         mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
82
83         return inode;
84 }
85
86 struct inode *lookup_free_space_inode(struct btrfs_root *root,
87                                       struct btrfs_block_group_cache
88                                       *block_group, struct btrfs_path *path)
89 {
90         struct inode *inode = NULL;
91         u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
92
93         spin_lock(&block_group->lock);
94         if (block_group->inode)
95                 inode = igrab(block_group->inode);
96         spin_unlock(&block_group->lock);
97         if (inode)
98                 return inode;
99
100         inode = __lookup_free_space_inode(root, path,
101                                           block_group->key.objectid);
102         if (IS_ERR(inode))
103                 return inode;
104
105         spin_lock(&block_group->lock);
106         if (!((BTRFS_I(inode)->flags & flags) == flags)) {
107                 printk(KERN_INFO "Old style space inode found, converting.\n");
108                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
109                         BTRFS_INODE_NODATACOW;
110                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
111         }
112
113         if (!block_group->iref) {
114                 block_group->inode = igrab(inode);
115                 block_group->iref = 1;
116         }
117         spin_unlock(&block_group->lock);
118
119         return inode;
120 }
121
122 int __create_free_space_inode(struct btrfs_root *root,
123                               struct btrfs_trans_handle *trans,
124                               struct btrfs_path *path, u64 ino, u64 offset)
125 {
126         struct btrfs_key key;
127         struct btrfs_disk_key disk_key;
128         struct btrfs_free_space_header *header;
129         struct btrfs_inode_item *inode_item;
130         struct extent_buffer *leaf;
131         u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
132         int ret;
133
134         ret = btrfs_insert_empty_inode(trans, root, path, ino);
135         if (ret)
136                 return ret;
137
138         /* We inline crc's for the free disk space cache */
139         if (ino != BTRFS_FREE_INO_OBJECTID)
140                 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
141
142         leaf = path->nodes[0];
143         inode_item = btrfs_item_ptr(leaf, path->slots[0],
144                                     struct btrfs_inode_item);
145         btrfs_item_key(leaf, &disk_key, path->slots[0]);
146         memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
147                              sizeof(*inode_item));
148         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
149         btrfs_set_inode_size(leaf, inode_item, 0);
150         btrfs_set_inode_nbytes(leaf, inode_item, 0);
151         btrfs_set_inode_uid(leaf, inode_item, 0);
152         btrfs_set_inode_gid(leaf, inode_item, 0);
153         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
154         btrfs_set_inode_flags(leaf, inode_item, flags);
155         btrfs_set_inode_nlink(leaf, inode_item, 1);
156         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
157         btrfs_set_inode_block_group(leaf, inode_item, offset);
158         btrfs_mark_buffer_dirty(leaf);
159         btrfs_release_path(path);
160
161         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
162         key.offset = offset;
163         key.type = 0;
164
165         ret = btrfs_insert_empty_item(trans, root, path, &key,
166                                       sizeof(struct btrfs_free_space_header));
167         if (ret < 0) {
168                 btrfs_release_path(path);
169                 return ret;
170         }
171         leaf = path->nodes[0];
172         header = btrfs_item_ptr(leaf, path->slots[0],
173                                 struct btrfs_free_space_header);
174         memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
175         btrfs_set_free_space_key(leaf, header, &disk_key);
176         btrfs_mark_buffer_dirty(leaf);
177         btrfs_release_path(path);
178
179         return 0;
180 }
181
182 int create_free_space_inode(struct btrfs_root *root,
183                             struct btrfs_trans_handle *trans,
184                             struct btrfs_block_group_cache *block_group,
185                             struct btrfs_path *path)
186 {
187         int ret;
188         u64 ino;
189
190         ret = btrfs_find_free_objectid(root, &ino);
191         if (ret < 0)
192                 return ret;
193
194         return __create_free_space_inode(root, trans, path, ino,
195                                          block_group->key.objectid);
196 }
197
198 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
199                                     struct btrfs_trans_handle *trans,
200                                     struct btrfs_path *path,
201                                     struct inode *inode)
202 {
203         struct btrfs_block_rsv *rsv;
204         u64 needed_bytes;
205         loff_t oldsize;
206         int ret = 0;
207
208         rsv = trans->block_rsv;
209         trans->block_rsv = &root->fs_info->global_block_rsv;
210
211         /* 1 for slack space, 1 for updating the inode */
212         needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
213                 btrfs_calc_trans_metadata_size(root, 1);
214
215         spin_lock(&trans->block_rsv->lock);
216         if (trans->block_rsv->reserved < needed_bytes) {
217                 spin_unlock(&trans->block_rsv->lock);
218                 trans->block_rsv = rsv;
219                 return -ENOSPC;
220         }
221         spin_unlock(&trans->block_rsv->lock);
222
223         oldsize = i_size_read(inode);
224         btrfs_i_size_write(inode, 0);
225         truncate_pagecache(inode, oldsize, 0);
226
227         /*
228          * We don't need an orphan item because truncating the free space cache
229          * will never be split across transactions.
230          */
231         ret = btrfs_truncate_inode_items(trans, root, inode,
232                                          0, BTRFS_EXTENT_DATA_KEY);
233
234         if (ret) {
235                 trans->block_rsv = rsv;
236                 btrfs_abort_transaction(trans, root, ret);
237                 return ret;
238         }
239
240         ret = btrfs_update_inode(trans, root, inode);
241         if (ret)
242                 btrfs_abort_transaction(trans, root, ret);
243         trans->block_rsv = rsv;
244
245         return ret;
246 }
247
248 static int readahead_cache(struct inode *inode)
249 {
250         struct file_ra_state *ra;
251         unsigned long last_index;
252
253         ra = kzalloc(sizeof(*ra), GFP_NOFS);
254         if (!ra)
255                 return -ENOMEM;
256
257         file_ra_state_init(ra, inode->i_mapping);
258         last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
259
260         page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
261
262         kfree(ra);
263
264         return 0;
265 }
266
267 struct io_ctl {
268         void *cur, *orig;
269         struct page *page;
270         struct page **pages;
271         struct btrfs_root *root;
272         unsigned long size;
273         int index;
274         int num_pages;
275         unsigned check_crcs:1;
276 };
277
278 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
279                        struct btrfs_root *root)
280 {
281         memset(io_ctl, 0, sizeof(struct io_ctl));
282         io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
283                 PAGE_CACHE_SHIFT;
284         io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
285                                 GFP_NOFS);
286         if (!io_ctl->pages)
287                 return -ENOMEM;
288         io_ctl->root = root;
289         if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
290                 io_ctl->check_crcs = 1;
291         return 0;
292 }
293
294 static void io_ctl_free(struct io_ctl *io_ctl)
295 {
296         kfree(io_ctl->pages);
297 }
298
299 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
300 {
301         if (io_ctl->cur) {
302                 kunmap(io_ctl->page);
303                 io_ctl->cur = NULL;
304                 io_ctl->orig = NULL;
305         }
306 }
307
308 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
309 {
310         WARN_ON(io_ctl->cur);
311         BUG_ON(io_ctl->index >= io_ctl->num_pages);
312         io_ctl->page = io_ctl->pages[io_ctl->index++];
313         io_ctl->cur = kmap(io_ctl->page);
314         io_ctl->orig = io_ctl->cur;
315         io_ctl->size = PAGE_CACHE_SIZE;
316         if (clear)
317                 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
318 }
319
320 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
321 {
322         int i;
323
324         io_ctl_unmap_page(io_ctl);
325
326         for (i = 0; i < io_ctl->num_pages; i++) {
327                 if (io_ctl->pages[i]) {
328                         ClearPageChecked(io_ctl->pages[i]);
329                         unlock_page(io_ctl->pages[i]);
330                         page_cache_release(io_ctl->pages[i]);
331                 }
332         }
333 }
334
335 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
336                                 int uptodate)
337 {
338         struct page *page;
339         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
340         int i;
341
342         for (i = 0; i < io_ctl->num_pages; i++) {
343                 page = find_or_create_page(inode->i_mapping, i, mask);
344                 if (!page) {
345                         io_ctl_drop_pages(io_ctl);
346                         return -ENOMEM;
347                 }
348                 io_ctl->pages[i] = page;
349                 if (uptodate && !PageUptodate(page)) {
350                         btrfs_readpage(NULL, page);
351                         lock_page(page);
352                         if (!PageUptodate(page)) {
353                                 printk(KERN_ERR "btrfs: error reading free "
354                                        "space cache\n");
355                                 io_ctl_drop_pages(io_ctl);
356                                 return -EIO;
357                         }
358                 }
359         }
360
361         for (i = 0; i < io_ctl->num_pages; i++) {
362                 clear_page_dirty_for_io(io_ctl->pages[i]);
363                 set_page_extent_mapped(io_ctl->pages[i]);
364         }
365
366         return 0;
367 }
368
369 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
370 {
371         __le64 *val;
372
373         io_ctl_map_page(io_ctl, 1);
374
375         /*
376          * Skip the csum areas.  If we don't check crcs then we just have a
377          * 64bit chunk at the front of the first page.
378          */
379         if (io_ctl->check_crcs) {
380                 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
381                 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
382         } else {
383                 io_ctl->cur += sizeof(u64);
384                 io_ctl->size -= sizeof(u64) * 2;
385         }
386
387         val = io_ctl->cur;
388         *val = cpu_to_le64(generation);
389         io_ctl->cur += sizeof(u64);
390 }
391
392 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
393 {
394         __le64 *gen;
395
396         /*
397          * Skip the crc area.  If we don't check crcs then we just have a 64bit
398          * chunk at the front of the first page.
399          */
400         if (io_ctl->check_crcs) {
401                 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
402                 io_ctl->size -= sizeof(u64) +
403                         (sizeof(u32) * io_ctl->num_pages);
404         } else {
405                 io_ctl->cur += sizeof(u64);
406                 io_ctl->size -= sizeof(u64) * 2;
407         }
408
409         gen = io_ctl->cur;
410         if (le64_to_cpu(*gen) != generation) {
411                 printk_ratelimited(KERN_ERR "btrfs: space cache generation "
412                                    "(%Lu) does not match inode (%Lu)\n", *gen,
413                                    generation);
414                 io_ctl_unmap_page(io_ctl);
415                 return -EIO;
416         }
417         io_ctl->cur += sizeof(u64);
418         return 0;
419 }
420
421 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
422 {
423         u32 *tmp;
424         u32 crc = ~(u32)0;
425         unsigned offset = 0;
426
427         if (!io_ctl->check_crcs) {
428                 io_ctl_unmap_page(io_ctl);
429                 return;
430         }
431
432         if (index == 0)
433                 offset = sizeof(u32) * io_ctl->num_pages;
434
435         crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
436                               PAGE_CACHE_SIZE - offset);
437         btrfs_csum_final(crc, (char *)&crc);
438         io_ctl_unmap_page(io_ctl);
439         tmp = kmap(io_ctl->pages[0]);
440         tmp += index;
441         *tmp = crc;
442         kunmap(io_ctl->pages[0]);
443 }
444
445 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
446 {
447         u32 *tmp, val;
448         u32 crc = ~(u32)0;
449         unsigned offset = 0;
450
451         if (!io_ctl->check_crcs) {
452                 io_ctl_map_page(io_ctl, 0);
453                 return 0;
454         }
455
456         if (index == 0)
457                 offset = sizeof(u32) * io_ctl->num_pages;
458
459         tmp = kmap(io_ctl->pages[0]);
460         tmp += index;
461         val = *tmp;
462         kunmap(io_ctl->pages[0]);
463
464         io_ctl_map_page(io_ctl, 0);
465         crc = btrfs_csum_data(io_ctl->root, io_ctl->orig + offset, crc,
466                               PAGE_CACHE_SIZE - offset);
467         btrfs_csum_final(crc, (char *)&crc);
468         if (val != crc) {
469                 printk_ratelimited(KERN_ERR "btrfs: csum mismatch on free "
470                                    "space cache\n");
471                 io_ctl_unmap_page(io_ctl);
472                 return -EIO;
473         }
474
475         return 0;
476 }
477
478 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
479                             void *bitmap)
480 {
481         struct btrfs_free_space_entry *entry;
482
483         if (!io_ctl->cur)
484                 return -ENOSPC;
485
486         entry = io_ctl->cur;
487         entry->offset = cpu_to_le64(offset);
488         entry->bytes = cpu_to_le64(bytes);
489         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
490                 BTRFS_FREE_SPACE_EXTENT;
491         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
492         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
493
494         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
495                 return 0;
496
497         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
498
499         /* No more pages to map */
500         if (io_ctl->index >= io_ctl->num_pages)
501                 return 0;
502
503         /* map the next page */
504         io_ctl_map_page(io_ctl, 1);
505         return 0;
506 }
507
508 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
509 {
510         if (!io_ctl->cur)
511                 return -ENOSPC;
512
513         /*
514          * If we aren't at the start of the current page, unmap this one and
515          * map the next one if there is any left.
516          */
517         if (io_ctl->cur != io_ctl->orig) {
518                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
519                 if (io_ctl->index >= io_ctl->num_pages)
520                         return -ENOSPC;
521                 io_ctl_map_page(io_ctl, 0);
522         }
523
524         memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
525         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
526         if (io_ctl->index < io_ctl->num_pages)
527                 io_ctl_map_page(io_ctl, 0);
528         return 0;
529 }
530
531 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
532 {
533         /*
534          * If we're not on the boundary we know we've modified the page and we
535          * need to crc the page.
536          */
537         if (io_ctl->cur != io_ctl->orig)
538                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
539         else
540                 io_ctl_unmap_page(io_ctl);
541
542         while (io_ctl->index < io_ctl->num_pages) {
543                 io_ctl_map_page(io_ctl, 1);
544                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
545         }
546 }
547
548 static int io_ctl_read_entry(struct io_ctl *io_ctl,
549                             struct btrfs_free_space *entry, u8 *type)
550 {
551         struct btrfs_free_space_entry *e;
552         int ret;
553
554         if (!io_ctl->cur) {
555                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
556                 if (ret)
557                         return ret;
558         }
559
560         e = io_ctl->cur;
561         entry->offset = le64_to_cpu(e->offset);
562         entry->bytes = le64_to_cpu(e->bytes);
563         *type = e->type;
564         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
565         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
566
567         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
568                 return 0;
569
570         io_ctl_unmap_page(io_ctl);
571
572         return 0;
573 }
574
575 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
576                               struct btrfs_free_space *entry)
577 {
578         int ret;
579
580         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
581         if (ret)
582                 return ret;
583
584         memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
585         io_ctl_unmap_page(io_ctl);
586
587         return 0;
588 }
589
590 /*
591  * Since we attach pinned extents after the fact we can have contiguous sections
592  * of free space that are split up in entries.  This poses a problem with the
593  * tree logging stuff since it could have allocated across what appears to be 2
594  * entries since we would have merged the entries when adding the pinned extents
595  * back to the free space cache.  So run through the space cache that we just
596  * loaded and merge contiguous entries.  This will make the log replay stuff not
597  * blow up and it will make for nicer allocator behavior.
598  */
599 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
600 {
601         struct btrfs_free_space *e, *prev = NULL;
602         struct rb_node *n;
603
604 again:
605         spin_lock(&ctl->tree_lock);
606         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
607                 e = rb_entry(n, struct btrfs_free_space, offset_index);
608                 if (!prev)
609                         goto next;
610                 if (e->bitmap || prev->bitmap)
611                         goto next;
612                 if (prev->offset + prev->bytes == e->offset) {
613                         unlink_free_space(ctl, prev);
614                         unlink_free_space(ctl, e);
615                         prev->bytes += e->bytes;
616                         kmem_cache_free(btrfs_free_space_cachep, e);
617                         link_free_space(ctl, prev);
618                         prev = NULL;
619                         spin_unlock(&ctl->tree_lock);
620                         goto again;
621                 }
622 next:
623                 prev = e;
624         }
625         spin_unlock(&ctl->tree_lock);
626 }
627
628 int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
629                             struct btrfs_free_space_ctl *ctl,
630                             struct btrfs_path *path, u64 offset)
631 {
632         struct btrfs_free_space_header *header;
633         struct extent_buffer *leaf;
634         struct io_ctl io_ctl;
635         struct btrfs_key key;
636         struct btrfs_free_space *e, *n;
637         struct list_head bitmaps;
638         u64 num_entries;
639         u64 num_bitmaps;
640         u64 generation;
641         u8 type;
642         int ret = 0;
643
644         INIT_LIST_HEAD(&bitmaps);
645
646         /* Nothing in the space cache, goodbye */
647         if (!i_size_read(inode))
648                 return 0;
649
650         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
651         key.offset = offset;
652         key.type = 0;
653
654         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
655         if (ret < 0)
656                 return 0;
657         else if (ret > 0) {
658                 btrfs_release_path(path);
659                 return 0;
660         }
661
662         ret = -1;
663
664         leaf = path->nodes[0];
665         header = btrfs_item_ptr(leaf, path->slots[0],
666                                 struct btrfs_free_space_header);
667         num_entries = btrfs_free_space_entries(leaf, header);
668         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
669         generation = btrfs_free_space_generation(leaf, header);
670         btrfs_release_path(path);
671
672         if (BTRFS_I(inode)->generation != generation) {
673                 printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
674                        " not match free space cache generation (%llu)\n",
675                        (unsigned long long)BTRFS_I(inode)->generation,
676                        (unsigned long long)generation);
677                 return 0;
678         }
679
680         if (!num_entries)
681                 return 0;
682
683         ret = io_ctl_init(&io_ctl, inode, root);
684         if (ret)
685                 return ret;
686
687         ret = readahead_cache(inode);
688         if (ret)
689                 goto out;
690
691         ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
692         if (ret)
693                 goto out;
694
695         ret = io_ctl_check_crc(&io_ctl, 0);
696         if (ret)
697                 goto free_cache;
698
699         ret = io_ctl_check_generation(&io_ctl, generation);
700         if (ret)
701                 goto free_cache;
702
703         while (num_entries) {
704                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
705                                       GFP_NOFS);
706                 if (!e)
707                         goto free_cache;
708
709                 ret = io_ctl_read_entry(&io_ctl, e, &type);
710                 if (ret) {
711                         kmem_cache_free(btrfs_free_space_cachep, e);
712                         goto free_cache;
713                 }
714
715                 if (!e->bytes) {
716                         kmem_cache_free(btrfs_free_space_cachep, e);
717                         goto free_cache;
718                 }
719
720                 if (type == BTRFS_FREE_SPACE_EXTENT) {
721                         spin_lock(&ctl->tree_lock);
722                         ret = link_free_space(ctl, e);
723                         spin_unlock(&ctl->tree_lock);
724                         if (ret) {
725                                 printk(KERN_ERR "Duplicate entries in "
726                                        "free space cache, dumping\n");
727                                 kmem_cache_free(btrfs_free_space_cachep, e);
728                                 goto free_cache;
729                         }
730                 } else {
731                         BUG_ON(!num_bitmaps);
732                         num_bitmaps--;
733                         e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
734                         if (!e->bitmap) {
735                                 kmem_cache_free(
736                                         btrfs_free_space_cachep, e);
737                                 goto free_cache;
738                         }
739                         spin_lock(&ctl->tree_lock);
740                         ret = link_free_space(ctl, e);
741                         ctl->total_bitmaps++;
742                         ctl->op->recalc_thresholds(ctl);
743                         spin_unlock(&ctl->tree_lock);
744                         if (ret) {
745                                 printk(KERN_ERR "Duplicate entries in "
746                                        "free space cache, dumping\n");
747                                 kmem_cache_free(btrfs_free_space_cachep, e);
748                                 goto free_cache;
749                         }
750                         list_add_tail(&e->list, &bitmaps);
751                 }
752
753                 num_entries--;
754         }
755
756         io_ctl_unmap_page(&io_ctl);
757
758         /*
759          * We add the bitmaps at the end of the entries in order that
760          * the bitmap entries are added to the cache.
761          */
762         list_for_each_entry_safe(e, n, &bitmaps, list) {
763                 list_del_init(&e->list);
764                 ret = io_ctl_read_bitmap(&io_ctl, e);
765                 if (ret)
766                         goto free_cache;
767         }
768
769         io_ctl_drop_pages(&io_ctl);
770         merge_space_tree(ctl);
771         ret = 1;
772 out:
773         io_ctl_free(&io_ctl);
774         return ret;
775 free_cache:
776         io_ctl_drop_pages(&io_ctl);
777         __btrfs_remove_free_space_cache(ctl);
778         goto out;
779 }
780
781 int load_free_space_cache(struct btrfs_fs_info *fs_info,
782                           struct btrfs_block_group_cache *block_group)
783 {
784         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
785         struct btrfs_root *root = fs_info->tree_root;
786         struct inode *inode;
787         struct btrfs_path *path;
788         int ret = 0;
789         bool matched;
790         u64 used = btrfs_block_group_used(&block_group->item);
791
792         /*
793          * If this block group has been marked to be cleared for one reason or
794          * another then we can't trust the on disk cache, so just return.
795          */
796         spin_lock(&block_group->lock);
797         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
798                 spin_unlock(&block_group->lock);
799                 return 0;
800         }
801         spin_unlock(&block_group->lock);
802
803         path = btrfs_alloc_path();
804         if (!path)
805                 return 0;
806         path->search_commit_root = 1;
807         path->skip_locking = 1;
808
809         inode = lookup_free_space_inode(root, block_group, path);
810         if (IS_ERR(inode)) {
811                 btrfs_free_path(path);
812                 return 0;
813         }
814
815         /* We may have converted the inode and made the cache invalid. */
816         spin_lock(&block_group->lock);
817         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
818                 spin_unlock(&block_group->lock);
819                 btrfs_free_path(path);
820                 goto out;
821         }
822         spin_unlock(&block_group->lock);
823
824         ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
825                                       path, block_group->key.objectid);
826         btrfs_free_path(path);
827         if (ret <= 0)
828                 goto out;
829
830         spin_lock(&ctl->tree_lock);
831         matched = (ctl->free_space == (block_group->key.offset - used -
832                                        block_group->bytes_super));
833         spin_unlock(&ctl->tree_lock);
834
835         if (!matched) {
836                 __btrfs_remove_free_space_cache(ctl);
837                 printk(KERN_ERR "block group %llu has an wrong amount of free "
838                        "space\n", block_group->key.objectid);
839                 ret = -1;
840         }
841 out:
842         if (ret < 0) {
843                 /* This cache is bogus, make sure it gets cleared */
844                 spin_lock(&block_group->lock);
845                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
846                 spin_unlock(&block_group->lock);
847                 ret = 0;
848
849                 printk(KERN_ERR "btrfs: failed to load free space cache "
850                        "for block group %llu\n", block_group->key.objectid);
851         }
852
853         iput(inode);
854         return ret;
855 }
856
857 /**
858  * __btrfs_write_out_cache - write out cached info to an inode
859  * @root - the root the inode belongs to
860  * @ctl - the free space cache we are going to write out
861  * @block_group - the block_group for this cache if it belongs to a block_group
862  * @trans - the trans handle
863  * @path - the path to use
864  * @offset - the offset for the key we'll insert
865  *
866  * This function writes out a free space cache struct to disk for quick recovery
867  * on mount.  This will return 0 if it was successfull in writing the cache out,
868  * and -1 if it was not.
869  */
870 int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
871                             struct btrfs_free_space_ctl *ctl,
872                             struct btrfs_block_group_cache *block_group,
873                             struct btrfs_trans_handle *trans,
874                             struct btrfs_path *path, u64 offset)
875 {
876         struct btrfs_free_space_header *header;
877         struct extent_buffer *leaf;
878         struct rb_node *node;
879         struct list_head *pos, *n;
880         struct extent_state *cached_state = NULL;
881         struct btrfs_free_cluster *cluster = NULL;
882         struct extent_io_tree *unpin = NULL;
883         struct io_ctl io_ctl;
884         struct list_head bitmap_list;
885         struct btrfs_key key;
886         u64 start, extent_start, extent_end, len;
887         int entries = 0;
888         int bitmaps = 0;
889         int ret;
890         int err = -1;
891
892         INIT_LIST_HEAD(&bitmap_list);
893
894         if (!i_size_read(inode))
895                 return -1;
896
897         ret = io_ctl_init(&io_ctl, inode, root);
898         if (ret)
899                 return -1;
900
901         /* Get the cluster for this block_group if it exists */
902         if (block_group && !list_empty(&block_group->cluster_list))
903                 cluster = list_entry(block_group->cluster_list.next,
904                                      struct btrfs_free_cluster,
905                                      block_group_list);
906
907         /* Lock all pages first so we can lock the extent safely. */
908         io_ctl_prepare_pages(&io_ctl, inode, 0);
909
910         lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
911                          0, &cached_state);
912
913         node = rb_first(&ctl->free_space_offset);
914         if (!node && cluster) {
915                 node = rb_first(&cluster->root);
916                 cluster = NULL;
917         }
918
919         /* Make sure we can fit our crcs into the first page */
920         if (io_ctl.check_crcs &&
921             (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE) {
922                 WARN_ON(1);
923                 goto out_nospc;
924         }
925
926         io_ctl_set_generation(&io_ctl, trans->transid);
927
928         /* Write out the extent entries */
929         while (node) {
930                 struct btrfs_free_space *e;
931
932                 e = rb_entry(node, struct btrfs_free_space, offset_index);
933                 entries++;
934
935                 ret = io_ctl_add_entry(&io_ctl, e->offset, e->bytes,
936                                        e->bitmap);
937                 if (ret)
938                         goto out_nospc;
939
940                 if (e->bitmap) {
941                         list_add_tail(&e->list, &bitmap_list);
942                         bitmaps++;
943                 }
944                 node = rb_next(node);
945                 if (!node && cluster) {
946                         node = rb_first(&cluster->root);
947                         cluster = NULL;
948                 }
949         }
950
951         /*
952          * We want to add any pinned extents to our free space cache
953          * so we don't leak the space
954          */
955
956         /*
957          * We shouldn't have switched the pinned extents yet so this is the
958          * right one
959          */
960         unpin = root->fs_info->pinned_extents;
961
962         if (block_group)
963                 start = block_group->key.objectid;
964
965         while (block_group && (start < block_group->key.objectid +
966                                block_group->key.offset)) {
967                 ret = find_first_extent_bit(unpin, start,
968                                             &extent_start, &extent_end,
969                                             EXTENT_DIRTY, NULL);
970                 if (ret) {
971                         ret = 0;
972                         break;
973                 }
974
975                 /* This pinned extent is out of our range */
976                 if (extent_start >= block_group->key.objectid +
977                     block_group->key.offset)
978                         break;
979
980                 extent_start = max(extent_start, start);
981                 extent_end = min(block_group->key.objectid +
982                                  block_group->key.offset, extent_end + 1);
983                 len = extent_end - extent_start;
984
985                 entries++;
986                 ret = io_ctl_add_entry(&io_ctl, extent_start, len, NULL);
987                 if (ret)
988                         goto out_nospc;
989
990                 start = extent_end;
991         }
992
993         /* Write out the bitmaps */
994         list_for_each_safe(pos, n, &bitmap_list) {
995                 struct btrfs_free_space *entry =
996                         list_entry(pos, struct btrfs_free_space, list);
997
998                 ret = io_ctl_add_bitmap(&io_ctl, entry->bitmap);
999                 if (ret)
1000                         goto out_nospc;
1001                 list_del_init(&entry->list);
1002         }
1003
1004         /* Zero out the rest of the pages just to make sure */
1005         io_ctl_zero_remaining_pages(&io_ctl);
1006
1007         ret = btrfs_dirty_pages(root, inode, io_ctl.pages, io_ctl.num_pages,
1008                                 0, i_size_read(inode), &cached_state);
1009         io_ctl_drop_pages(&io_ctl);
1010         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1011                              i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1012
1013         if (ret)
1014                 goto out;
1015
1016
1017         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1018
1019         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1020         key.offset = offset;
1021         key.type = 0;
1022
1023         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1024         if (ret < 0) {
1025                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1026                                  EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1027                                  GFP_NOFS);
1028                 goto out;
1029         }
1030         leaf = path->nodes[0];
1031         if (ret > 0) {
1032                 struct btrfs_key found_key;
1033                 BUG_ON(!path->slots[0]);
1034                 path->slots[0]--;
1035                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1036                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1037                     found_key.offset != offset) {
1038                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1039                                          inode->i_size - 1,
1040                                          EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1041                                          NULL, GFP_NOFS);
1042                         btrfs_release_path(path);
1043                         goto out;
1044                 }
1045         }
1046
1047         BTRFS_I(inode)->generation = trans->transid;
1048         header = btrfs_item_ptr(leaf, path->slots[0],
1049                                 struct btrfs_free_space_header);
1050         btrfs_set_free_space_entries(leaf, header, entries);
1051         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1052         btrfs_set_free_space_generation(leaf, header, trans->transid);
1053         btrfs_mark_buffer_dirty(leaf);
1054         btrfs_release_path(path);
1055
1056         err = 0;
1057 out:
1058         io_ctl_free(&io_ctl);
1059         if (err) {
1060                 invalidate_inode_pages2(inode->i_mapping);
1061                 BTRFS_I(inode)->generation = 0;
1062         }
1063         btrfs_update_inode(trans, root, inode);
1064         return err;
1065
1066 out_nospc:
1067         list_for_each_safe(pos, n, &bitmap_list) {
1068                 struct btrfs_free_space *entry =
1069                         list_entry(pos, struct btrfs_free_space, list);
1070                 list_del_init(&entry->list);
1071         }
1072         io_ctl_drop_pages(&io_ctl);
1073         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1074                              i_size_read(inode) - 1, &cached_state, GFP_NOFS);
1075         goto out;
1076 }
1077
1078 int btrfs_write_out_cache(struct btrfs_root *root,
1079                           struct btrfs_trans_handle *trans,
1080                           struct btrfs_block_group_cache *block_group,
1081                           struct btrfs_path *path)
1082 {
1083         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1084         struct inode *inode;
1085         int ret = 0;
1086
1087         root = root->fs_info->tree_root;
1088
1089         spin_lock(&block_group->lock);
1090         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1091                 spin_unlock(&block_group->lock);
1092                 return 0;
1093         }
1094         spin_unlock(&block_group->lock);
1095
1096         inode = lookup_free_space_inode(root, block_group, path);
1097         if (IS_ERR(inode))
1098                 return 0;
1099
1100         ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1101                                       path, block_group->key.objectid);
1102         if (ret) {
1103                 spin_lock(&block_group->lock);
1104                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1105                 spin_unlock(&block_group->lock);
1106                 ret = 0;
1107 #ifdef DEBUG
1108                 printk(KERN_ERR "btrfs: failed to write free space cache "
1109                        "for block group %llu\n", block_group->key.objectid);
1110 #endif
1111         }
1112
1113         iput(inode);
1114         return ret;
1115 }
1116
1117 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1118                                           u64 offset)
1119 {
1120         BUG_ON(offset < bitmap_start);
1121         offset -= bitmap_start;
1122         return (unsigned long)(div_u64(offset, unit));
1123 }
1124
1125 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1126 {
1127         return (unsigned long)(div_u64(bytes, unit));
1128 }
1129
1130 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1131                                    u64 offset)
1132 {
1133         u64 bitmap_start;
1134         u64 bytes_per_bitmap;
1135
1136         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1137         bitmap_start = offset - ctl->start;
1138         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1139         bitmap_start *= bytes_per_bitmap;
1140         bitmap_start += ctl->start;
1141
1142         return bitmap_start;
1143 }
1144
1145 static int tree_insert_offset(struct rb_root *root, u64 offset,
1146                               struct rb_node *node, int bitmap)
1147 {
1148         struct rb_node **p = &root->rb_node;
1149         struct rb_node *parent = NULL;
1150         struct btrfs_free_space *info;
1151
1152         while (*p) {
1153                 parent = *p;
1154                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1155
1156                 if (offset < info->offset) {
1157                         p = &(*p)->rb_left;
1158                 } else if (offset > info->offset) {
1159                         p = &(*p)->rb_right;
1160                 } else {
1161                         /*
1162                          * we could have a bitmap entry and an extent entry
1163                          * share the same offset.  If this is the case, we want
1164                          * the extent entry to always be found first if we do a
1165                          * linear search through the tree, since we want to have
1166                          * the quickest allocation time, and allocating from an
1167                          * extent is faster than allocating from a bitmap.  So
1168                          * if we're inserting a bitmap and we find an entry at
1169                          * this offset, we want to go right, or after this entry
1170                          * logically.  If we are inserting an extent and we've
1171                          * found a bitmap, we want to go left, or before
1172                          * logically.
1173                          */
1174                         if (bitmap) {
1175                                 if (info->bitmap) {
1176                                         WARN_ON_ONCE(1);
1177                                         return -EEXIST;
1178                                 }
1179                                 p = &(*p)->rb_right;
1180                         } else {
1181                                 if (!info->bitmap) {
1182                                         WARN_ON_ONCE(1);
1183                                         return -EEXIST;
1184                                 }
1185                                 p = &(*p)->rb_left;
1186                         }
1187                 }
1188         }
1189
1190         rb_link_node(node, parent, p);
1191         rb_insert_color(node, root);
1192
1193         return 0;
1194 }
1195
1196 /*
1197  * searches the tree for the given offset.
1198  *
1199  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1200  * want a section that has at least bytes size and comes at or after the given
1201  * offset.
1202  */
1203 static struct btrfs_free_space *
1204 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1205                    u64 offset, int bitmap_only, int fuzzy)
1206 {
1207         struct rb_node *n = ctl->free_space_offset.rb_node;
1208         struct btrfs_free_space *entry, *prev = NULL;
1209
1210         /* find entry that is closest to the 'offset' */
1211         while (1) {
1212                 if (!n) {
1213                         entry = NULL;
1214                         break;
1215                 }
1216
1217                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1218                 prev = entry;
1219
1220                 if (offset < entry->offset)
1221                         n = n->rb_left;
1222                 else if (offset > entry->offset)
1223                         n = n->rb_right;
1224                 else
1225                         break;
1226         }
1227
1228         if (bitmap_only) {
1229                 if (!entry)
1230                         return NULL;
1231                 if (entry->bitmap)
1232                         return entry;
1233
1234                 /*
1235                  * bitmap entry and extent entry may share same offset,
1236                  * in that case, bitmap entry comes after extent entry.
1237                  */
1238                 n = rb_next(n);
1239                 if (!n)
1240                         return NULL;
1241                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1242                 if (entry->offset != offset)
1243                         return NULL;
1244
1245                 WARN_ON(!entry->bitmap);
1246                 return entry;
1247         } else if (entry) {
1248                 if (entry->bitmap) {
1249                         /*
1250                          * if previous extent entry covers the offset,
1251                          * we should return it instead of the bitmap entry
1252                          */
1253                         n = rb_prev(&entry->offset_index);
1254                         if (n) {
1255                                 prev = rb_entry(n, struct btrfs_free_space,
1256                                                 offset_index);
1257                                 if (!prev->bitmap &&
1258                                     prev->offset + prev->bytes > offset)
1259                                         entry = prev;
1260                         }
1261                 }
1262                 return entry;
1263         }
1264
1265         if (!prev)
1266                 return NULL;
1267
1268         /* find last entry before the 'offset' */
1269         entry = prev;
1270         if (entry->offset > offset) {
1271                 n = rb_prev(&entry->offset_index);
1272                 if (n) {
1273                         entry = rb_entry(n, struct btrfs_free_space,
1274                                         offset_index);
1275                         BUG_ON(entry->offset > offset);
1276                 } else {
1277                         if (fuzzy)
1278                                 return entry;
1279                         else
1280                                 return NULL;
1281                 }
1282         }
1283
1284         if (entry->bitmap) {
1285                 n = rb_prev(&entry->offset_index);
1286                 if (n) {
1287                         prev = rb_entry(n, struct btrfs_free_space,
1288                                         offset_index);
1289                         if (!prev->bitmap &&
1290                             prev->offset + prev->bytes > offset)
1291                                 return prev;
1292                 }
1293                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1294                         return entry;
1295         } else if (entry->offset + entry->bytes > offset)
1296                 return entry;
1297
1298         if (!fuzzy)
1299                 return NULL;
1300
1301         while (1) {
1302                 if (entry->bitmap) {
1303                         if (entry->offset + BITS_PER_BITMAP *
1304                             ctl->unit > offset)
1305                                 break;
1306                 } else {
1307                         if (entry->offset + entry->bytes > offset)
1308                                 break;
1309                 }
1310
1311                 n = rb_next(&entry->offset_index);
1312                 if (!n)
1313                         return NULL;
1314                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1315         }
1316         return entry;
1317 }
1318
1319 static inline void
1320 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1321                     struct btrfs_free_space *info)
1322 {
1323         rb_erase(&info->offset_index, &ctl->free_space_offset);
1324         ctl->free_extents--;
1325 }
1326
1327 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1328                               struct btrfs_free_space *info)
1329 {
1330         __unlink_free_space(ctl, info);
1331         ctl->free_space -= info->bytes;
1332 }
1333
1334 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1335                            struct btrfs_free_space *info)
1336 {
1337         int ret = 0;
1338
1339         BUG_ON(!info->bitmap && !info->bytes);
1340         ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1341                                  &info->offset_index, (info->bitmap != NULL));
1342         if (ret)
1343                 return ret;
1344
1345         ctl->free_space += info->bytes;
1346         ctl->free_extents++;
1347         return ret;
1348 }
1349
1350 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1351 {
1352         struct btrfs_block_group_cache *block_group = ctl->private;
1353         u64 max_bytes;
1354         u64 bitmap_bytes;
1355         u64 extent_bytes;
1356         u64 size = block_group->key.offset;
1357         u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
1358         int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1359
1360         BUG_ON(ctl->total_bitmaps > max_bitmaps);
1361
1362         /*
1363          * The goal is to keep the total amount of memory used per 1gb of space
1364          * at or below 32k, so we need to adjust how much memory we allow to be
1365          * used by extent based free space tracking
1366          */
1367         if (size < 1024 * 1024 * 1024)
1368                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1369         else
1370                 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1371                         div64_u64(size, 1024 * 1024 * 1024);
1372
1373         /*
1374          * we want to account for 1 more bitmap than what we have so we can make
1375          * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1376          * we add more bitmaps.
1377          */
1378         bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1379
1380         if (bitmap_bytes >= max_bytes) {
1381                 ctl->extents_thresh = 0;
1382                 return;
1383         }
1384
1385         /*
1386          * we want the extent entry threshold to always be at most 1/2 the maxw
1387          * bytes we can have, or whatever is less than that.
1388          */
1389         extent_bytes = max_bytes - bitmap_bytes;
1390         extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1391
1392         ctl->extents_thresh =
1393                 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1394 }
1395
1396 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1397                                        struct btrfs_free_space *info,
1398                                        u64 offset, u64 bytes)
1399 {
1400         unsigned long start, count;
1401
1402         start = offset_to_bit(info->offset, ctl->unit, offset);
1403         count = bytes_to_bits(bytes, ctl->unit);
1404         BUG_ON(start + count > BITS_PER_BITMAP);
1405
1406         bitmap_clear(info->bitmap, start, count);
1407
1408         info->bytes -= bytes;
1409 }
1410
1411 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1412                               struct btrfs_free_space *info, u64 offset,
1413                               u64 bytes)
1414 {
1415         __bitmap_clear_bits(ctl, info, offset, bytes);
1416         ctl->free_space -= bytes;
1417 }
1418
1419 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1420                             struct btrfs_free_space *info, u64 offset,
1421                             u64 bytes)
1422 {
1423         unsigned long start, count;
1424
1425         start = offset_to_bit(info->offset, ctl->unit, offset);
1426         count = bytes_to_bits(bytes, ctl->unit);
1427         BUG_ON(start + count > BITS_PER_BITMAP);
1428
1429         bitmap_set(info->bitmap, start, count);
1430
1431         info->bytes += bytes;
1432         ctl->free_space += bytes;
1433 }
1434
1435 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1436                          struct btrfs_free_space *bitmap_info, u64 *offset,
1437                          u64 *bytes)
1438 {
1439         unsigned long found_bits = 0;
1440         unsigned long bits, i;
1441         unsigned long next_zero;
1442
1443         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1444                           max_t(u64, *offset, bitmap_info->offset));
1445         bits = bytes_to_bits(*bytes, ctl->unit);
1446
1447         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1448                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1449                                                BITS_PER_BITMAP, i);
1450                 if ((next_zero - i) >= bits) {
1451                         found_bits = next_zero - i;
1452                         break;
1453                 }
1454                 i = next_zero;
1455         }
1456
1457         if (found_bits) {
1458                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1459                 *bytes = (u64)(found_bits) * ctl->unit;
1460                 return 0;
1461         }
1462
1463         return -1;
1464 }
1465
1466 static struct btrfs_free_space *
1467 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes)
1468 {
1469         struct btrfs_free_space *entry;
1470         struct rb_node *node;
1471         int ret;
1472
1473         if (!ctl->free_space_offset.rb_node)
1474                 return NULL;
1475
1476         entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1477         if (!entry)
1478                 return NULL;
1479
1480         for (node = &entry->offset_index; node; node = rb_next(node)) {
1481                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1482                 if (entry->bytes < *bytes)
1483                         continue;
1484
1485                 if (entry->bitmap) {
1486                         ret = search_bitmap(ctl, entry, offset, bytes);
1487                         if (!ret)
1488                                 return entry;
1489                         continue;
1490                 }
1491
1492                 *offset = entry->offset;
1493                 *bytes = entry->bytes;
1494                 return entry;
1495         }
1496
1497         return NULL;
1498 }
1499
1500 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1501                            struct btrfs_free_space *info, u64 offset)
1502 {
1503         info->offset = offset_to_bitmap(ctl, offset);
1504         info->bytes = 0;
1505         INIT_LIST_HEAD(&info->list);
1506         link_free_space(ctl, info);
1507         ctl->total_bitmaps++;
1508
1509         ctl->op->recalc_thresholds(ctl);
1510 }
1511
1512 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1513                         struct btrfs_free_space *bitmap_info)
1514 {
1515         unlink_free_space(ctl, bitmap_info);
1516         kfree(bitmap_info->bitmap);
1517         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1518         ctl->total_bitmaps--;
1519         ctl->op->recalc_thresholds(ctl);
1520 }
1521
1522 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1523                               struct btrfs_free_space *bitmap_info,
1524                               u64 *offset, u64 *bytes)
1525 {
1526         u64 end;
1527         u64 search_start, search_bytes;
1528         int ret;
1529
1530 again:
1531         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1532
1533         /*
1534          * We need to search for bits in this bitmap.  We could only cover some
1535          * of the extent in this bitmap thanks to how we add space, so we need
1536          * to search for as much as it as we can and clear that amount, and then
1537          * go searching for the next bit.
1538          */
1539         search_start = *offset;
1540         search_bytes = ctl->unit;
1541         search_bytes = min(search_bytes, end - search_start + 1);
1542         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1543         BUG_ON(ret < 0 || search_start != *offset);
1544
1545         /* We may have found more bits than what we need */
1546         search_bytes = min(search_bytes, *bytes);
1547
1548         /* Cannot clear past the end of the bitmap */
1549         search_bytes = min(search_bytes, end - search_start + 1);
1550
1551         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1552         *offset += search_bytes;
1553         *bytes -= search_bytes;
1554
1555         if (*bytes) {
1556                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1557                 if (!bitmap_info->bytes)
1558                         free_bitmap(ctl, bitmap_info);
1559
1560                 /*
1561                  * no entry after this bitmap, but we still have bytes to
1562                  * remove, so something has gone wrong.
1563                  */
1564                 if (!next)
1565                         return -EINVAL;
1566
1567                 bitmap_info = rb_entry(next, struct btrfs_free_space,
1568                                        offset_index);
1569
1570                 /*
1571                  * if the next entry isn't a bitmap we need to return to let the
1572                  * extent stuff do its work.
1573                  */
1574                 if (!bitmap_info->bitmap)
1575                         return -EAGAIN;
1576
1577                 /*
1578                  * Ok the next item is a bitmap, but it may not actually hold
1579                  * the information for the rest of this free space stuff, so
1580                  * look for it, and if we don't find it return so we can try
1581                  * everything over again.
1582                  */
1583                 search_start = *offset;
1584                 search_bytes = ctl->unit;
1585                 ret = search_bitmap(ctl, bitmap_info, &search_start,
1586                                     &search_bytes);
1587                 if (ret < 0 || search_start != *offset)
1588                         return -EAGAIN;
1589
1590                 goto again;
1591         } else if (!bitmap_info->bytes)
1592                 free_bitmap(ctl, bitmap_info);
1593
1594         return 0;
1595 }
1596
1597 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1598                                struct btrfs_free_space *info, u64 offset,
1599                                u64 bytes)
1600 {
1601         u64 bytes_to_set = 0;
1602         u64 end;
1603
1604         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1605
1606         bytes_to_set = min(end - offset, bytes);
1607
1608         bitmap_set_bits(ctl, info, offset, bytes_to_set);
1609
1610         return bytes_to_set;
1611
1612 }
1613
1614 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1615                       struct btrfs_free_space *info)
1616 {
1617         struct btrfs_block_group_cache *block_group = ctl->private;
1618
1619         /*
1620          * If we are below the extents threshold then we can add this as an
1621          * extent, and don't have to deal with the bitmap
1622          */
1623         if (ctl->free_extents < ctl->extents_thresh) {
1624                 /*
1625                  * If this block group has some small extents we don't want to
1626                  * use up all of our free slots in the cache with them, we want
1627                  * to reserve them to larger extents, however if we have plent
1628                  * of cache left then go ahead an dadd them, no sense in adding
1629                  * the overhead of a bitmap if we don't have to.
1630                  */
1631                 if (info->bytes <= block_group->sectorsize * 4) {
1632                         if (ctl->free_extents * 2 <= ctl->extents_thresh)
1633                                 return false;
1634                 } else {
1635                         return false;
1636                 }
1637         }
1638
1639         /*
1640          * some block groups are so tiny they can't be enveloped by a bitmap, so
1641          * don't even bother to create a bitmap for this
1642          */
1643         if (BITS_PER_BITMAP * block_group->sectorsize >
1644             block_group->key.offset)
1645                 return false;
1646
1647         return true;
1648 }
1649
1650 static struct btrfs_free_space_op free_space_op = {
1651         .recalc_thresholds      = recalculate_thresholds,
1652         .use_bitmap             = use_bitmap,
1653 };
1654
1655 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1656                               struct btrfs_free_space *info)
1657 {
1658         struct btrfs_free_space *bitmap_info;
1659         struct btrfs_block_group_cache *block_group = NULL;
1660         int added = 0;
1661         u64 bytes, offset, bytes_added;
1662         int ret;
1663
1664         bytes = info->bytes;
1665         offset = info->offset;
1666
1667         if (!ctl->op->use_bitmap(ctl, info))
1668                 return 0;
1669
1670         if (ctl->op == &free_space_op)
1671                 block_group = ctl->private;
1672 again:
1673         /*
1674          * Since we link bitmaps right into the cluster we need to see if we
1675          * have a cluster here, and if so and it has our bitmap we need to add
1676          * the free space to that bitmap.
1677          */
1678         if (block_group && !list_empty(&block_group->cluster_list)) {
1679                 struct btrfs_free_cluster *cluster;
1680                 struct rb_node *node;
1681                 struct btrfs_free_space *entry;
1682
1683                 cluster = list_entry(block_group->cluster_list.next,
1684                                      struct btrfs_free_cluster,
1685                                      block_group_list);
1686                 spin_lock(&cluster->lock);
1687                 node = rb_first(&cluster->root);
1688                 if (!node) {
1689                         spin_unlock(&cluster->lock);
1690                         goto no_cluster_bitmap;
1691                 }
1692
1693                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1694                 if (!entry->bitmap) {
1695                         spin_unlock(&cluster->lock);
1696                         goto no_cluster_bitmap;
1697                 }
1698
1699                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1700                         bytes_added = add_bytes_to_bitmap(ctl, entry,
1701                                                           offset, bytes);
1702                         bytes -= bytes_added;
1703                         offset += bytes_added;
1704                 }
1705                 spin_unlock(&cluster->lock);
1706                 if (!bytes) {
1707                         ret = 1;
1708                         goto out;
1709                 }
1710         }
1711
1712 no_cluster_bitmap:
1713         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1714                                          1, 0);
1715         if (!bitmap_info) {
1716                 BUG_ON(added);
1717                 goto new_bitmap;
1718         }
1719
1720         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1721         bytes -= bytes_added;
1722         offset += bytes_added;
1723         added = 0;
1724
1725         if (!bytes) {
1726                 ret = 1;
1727                 goto out;
1728         } else
1729                 goto again;
1730
1731 new_bitmap:
1732         if (info && info->bitmap) {
1733                 add_new_bitmap(ctl, info, offset);
1734                 added = 1;
1735                 info = NULL;
1736                 goto again;
1737         } else {
1738                 spin_unlock(&ctl->tree_lock);
1739
1740                 /* no pre-allocated info, allocate a new one */
1741                 if (!info) {
1742                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
1743                                                  GFP_NOFS);
1744                         if (!info) {
1745                                 spin_lock(&ctl->tree_lock);
1746                                 ret = -ENOMEM;
1747                                 goto out;
1748                         }
1749                 }
1750
1751                 /* allocate the bitmap */
1752                 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1753                 spin_lock(&ctl->tree_lock);
1754                 if (!info->bitmap) {
1755                         ret = -ENOMEM;
1756                         goto out;
1757                 }
1758                 goto again;
1759         }
1760
1761 out:
1762         if (info) {
1763                 if (info->bitmap)
1764                         kfree(info->bitmap);
1765                 kmem_cache_free(btrfs_free_space_cachep, info);
1766         }
1767
1768         return ret;
1769 }
1770
1771 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1772                           struct btrfs_free_space *info, bool update_stat)
1773 {
1774         struct btrfs_free_space *left_info;
1775         struct btrfs_free_space *right_info;
1776         bool merged = false;
1777         u64 offset = info->offset;
1778         u64 bytes = info->bytes;
1779
1780         /*
1781          * first we want to see if there is free space adjacent to the range we
1782          * are adding, if there is remove that struct and add a new one to
1783          * cover the entire range
1784          */
1785         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1786         if (right_info && rb_prev(&right_info->offset_index))
1787                 left_info = rb_entry(rb_prev(&right_info->offset_index),
1788                                      struct btrfs_free_space, offset_index);
1789         else
1790                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1791
1792         if (right_info && !right_info->bitmap) {
1793                 if (update_stat)
1794                         unlink_free_space(ctl, right_info);
1795                 else
1796                         __unlink_free_space(ctl, right_info);
1797                 info->bytes += right_info->bytes;
1798                 kmem_cache_free(btrfs_free_space_cachep, right_info);
1799                 merged = true;
1800         }
1801
1802         if (left_info && !left_info->bitmap &&
1803             left_info->offset + left_info->bytes == offset) {
1804                 if (update_stat)
1805                         unlink_free_space(ctl, left_info);
1806                 else
1807                         __unlink_free_space(ctl, left_info);
1808                 info->offset = left_info->offset;
1809                 info->bytes += left_info->bytes;
1810                 kmem_cache_free(btrfs_free_space_cachep, left_info);
1811                 merged = true;
1812         }
1813
1814         return merged;
1815 }
1816
1817 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1818                            u64 offset, u64 bytes)
1819 {
1820         struct btrfs_free_space *info;
1821         int ret = 0;
1822
1823         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1824         if (!info)
1825                 return -ENOMEM;
1826
1827         info->offset = offset;
1828         info->bytes = bytes;
1829
1830         spin_lock(&ctl->tree_lock);
1831
1832         if (try_merge_free_space(ctl, info, true))
1833                 goto link;
1834
1835         /*
1836          * There was no extent directly to the left or right of this new
1837          * extent then we know we're going to have to allocate a new extent, so
1838          * before we do that see if we need to drop this into a bitmap
1839          */
1840         ret = insert_into_bitmap(ctl, info);
1841         if (ret < 0) {
1842                 goto out;
1843         } else if (ret) {
1844                 ret = 0;
1845                 goto out;
1846         }
1847 link:
1848         ret = link_free_space(ctl, info);
1849         if (ret)
1850                 kmem_cache_free(btrfs_free_space_cachep, info);
1851 out:
1852         spin_unlock(&ctl->tree_lock);
1853
1854         if (ret) {
1855                 printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
1856                 BUG_ON(ret == -EEXIST);
1857         }
1858
1859         return ret;
1860 }
1861
1862 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1863                             u64 offset, u64 bytes)
1864 {
1865         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1866         struct btrfs_free_space *info;
1867         int ret = 0;
1868
1869         spin_lock(&ctl->tree_lock);
1870
1871 again:
1872         if (!bytes)
1873                 goto out_lock;
1874
1875         info = tree_search_offset(ctl, offset, 0, 0);
1876         if (!info) {
1877                 /*
1878                  * oops didn't find an extent that matched the space we wanted
1879                  * to remove, look for a bitmap instead
1880                  */
1881                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1882                                           1, 0);
1883                 if (!info) {
1884                         /* the tree logging code might be calling us before we
1885                          * have fully loaded the free space rbtree for this
1886                          * block group.  So it is possible the entry won't
1887                          * be in the rbtree yet at all.  The caching code
1888                          * will make sure not to put it in the rbtree if
1889                          * the logging code has pinned it.
1890                          */
1891                         goto out_lock;
1892                 }
1893         }
1894
1895         if (!info->bitmap) {
1896                 unlink_free_space(ctl, info);
1897                 if (offset == info->offset) {
1898                         u64 to_free = min(bytes, info->bytes);
1899
1900                         info->bytes -= to_free;
1901                         info->offset += to_free;
1902                         if (info->bytes) {
1903                                 ret = link_free_space(ctl, info);
1904                                 WARN_ON(ret);
1905                         } else {
1906                                 kmem_cache_free(btrfs_free_space_cachep, info);
1907                         }
1908
1909                         offset += to_free;
1910                         bytes -= to_free;
1911                         goto again;
1912                 } else {
1913                         u64 old_end = info->bytes + info->offset;
1914
1915                         info->bytes = offset - info->offset;
1916                         ret = link_free_space(ctl, info);
1917                         WARN_ON(ret);
1918                         if (ret)
1919                                 goto out_lock;
1920
1921                         /* Not enough bytes in this entry to satisfy us */
1922                         if (old_end < offset + bytes) {
1923                                 bytes -= old_end - offset;
1924                                 offset = old_end;
1925                                 goto again;
1926                         } else if (old_end == offset + bytes) {
1927                                 /* all done */
1928                                 goto out_lock;
1929                         }
1930                         spin_unlock(&ctl->tree_lock);
1931
1932                         ret = btrfs_add_free_space(block_group, offset + bytes,
1933                                                    old_end - (offset + bytes));
1934                         WARN_ON(ret);
1935                         goto out;
1936                 }
1937         }
1938
1939         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
1940         if (ret == -EAGAIN)
1941                 goto again;
1942         BUG_ON(ret); /* logic error */
1943 out_lock:
1944         spin_unlock(&ctl->tree_lock);
1945 out:
1946         return ret;
1947 }
1948
1949 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
1950                            u64 bytes)
1951 {
1952         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1953         struct btrfs_free_space *info;
1954         struct rb_node *n;
1955         int count = 0;
1956
1957         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
1958                 info = rb_entry(n, struct btrfs_free_space, offset_index);
1959                 if (info->bytes >= bytes && !block_group->ro)
1960                         count++;
1961                 printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
1962                        (unsigned long long)info->offset,
1963                        (unsigned long long)info->bytes,
1964                        (info->bitmap) ? "yes" : "no");
1965         }
1966         printk(KERN_INFO "block group has cluster?: %s\n",
1967                list_empty(&block_group->cluster_list) ? "no" : "yes");
1968         printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
1969                "\n", count);
1970 }
1971
1972 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
1973 {
1974         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1975
1976         spin_lock_init(&ctl->tree_lock);
1977         ctl->unit = block_group->sectorsize;
1978         ctl->start = block_group->key.objectid;
1979         ctl->private = block_group;
1980         ctl->op = &free_space_op;
1981
1982         /*
1983          * we only want to have 32k of ram per block group for keeping
1984          * track of free space, and if we pass 1/2 of that we want to
1985          * start converting things over to using bitmaps
1986          */
1987         ctl->extents_thresh = ((1024 * 32) / 2) /
1988                                 sizeof(struct btrfs_free_space);
1989 }
1990
1991 /*
1992  * for a given cluster, put all of its extents back into the free
1993  * space cache.  If the block group passed doesn't match the block group
1994  * pointed to by the cluster, someone else raced in and freed the
1995  * cluster already.  In that case, we just return without changing anything
1996  */
1997 static int
1998 __btrfs_return_cluster_to_free_space(
1999                              struct btrfs_block_group_cache *block_group,
2000                              struct btrfs_free_cluster *cluster)
2001 {
2002         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2003         struct btrfs_free_space *entry;
2004         struct rb_node *node;
2005
2006         spin_lock(&cluster->lock);
2007         if (cluster->block_group != block_group)
2008                 goto out;
2009
2010         cluster->block_group = NULL;
2011         cluster->window_start = 0;
2012         list_del_init(&cluster->block_group_list);
2013
2014         node = rb_first(&cluster->root);
2015         while (node) {
2016                 bool bitmap;
2017
2018                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2019                 node = rb_next(&entry->offset_index);
2020                 rb_erase(&entry->offset_index, &cluster->root);
2021
2022                 bitmap = (entry->bitmap != NULL);
2023                 if (!bitmap)
2024                         try_merge_free_space(ctl, entry, false);
2025                 tree_insert_offset(&ctl->free_space_offset,
2026                                    entry->offset, &entry->offset_index, bitmap);
2027         }
2028         cluster->root = RB_ROOT;
2029
2030 out:
2031         spin_unlock(&cluster->lock);
2032         btrfs_put_block_group(block_group);
2033         return 0;
2034 }
2035
2036 void __btrfs_remove_free_space_cache_locked(struct btrfs_free_space_ctl *ctl)
2037 {
2038         struct btrfs_free_space *info;
2039         struct rb_node *node;
2040
2041         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2042                 info = rb_entry(node, struct btrfs_free_space, offset_index);
2043                 if (!info->bitmap) {
2044                         unlink_free_space(ctl, info);
2045                         kmem_cache_free(btrfs_free_space_cachep, info);
2046                 } else {
2047                         free_bitmap(ctl, info);
2048                 }
2049                 if (need_resched()) {
2050                         spin_unlock(&ctl->tree_lock);
2051                         cond_resched();
2052                         spin_lock(&ctl->tree_lock);
2053                 }
2054         }
2055 }
2056
2057 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2058 {
2059         spin_lock(&ctl->tree_lock);
2060         __btrfs_remove_free_space_cache_locked(ctl);
2061         spin_unlock(&ctl->tree_lock);
2062 }
2063
2064 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2065 {
2066         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2067         struct btrfs_free_cluster *cluster;
2068         struct list_head *head;
2069
2070         spin_lock(&ctl->tree_lock);
2071         while ((head = block_group->cluster_list.next) !=
2072                &block_group->cluster_list) {
2073                 cluster = list_entry(head, struct btrfs_free_cluster,
2074                                      block_group_list);
2075
2076                 WARN_ON(cluster->block_group != block_group);
2077                 __btrfs_return_cluster_to_free_space(block_group, cluster);
2078                 if (need_resched()) {
2079                         spin_unlock(&ctl->tree_lock);
2080                         cond_resched();
2081                         spin_lock(&ctl->tree_lock);
2082                 }
2083         }
2084         __btrfs_remove_free_space_cache_locked(ctl);
2085         spin_unlock(&ctl->tree_lock);
2086
2087 }
2088
2089 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2090                                u64 offset, u64 bytes, u64 empty_size)
2091 {
2092         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2093         struct btrfs_free_space *entry = NULL;
2094         u64 bytes_search = bytes + empty_size;
2095         u64 ret = 0;
2096
2097         spin_lock(&ctl->tree_lock);
2098         entry = find_free_space(ctl, &offset, &bytes_search);
2099         if (!entry)
2100                 goto out;
2101
2102         ret = offset;
2103         if (entry->bitmap) {
2104                 bitmap_clear_bits(ctl, entry, offset, bytes);
2105                 if (!entry->bytes)
2106                         free_bitmap(ctl, entry);
2107         } else {
2108                 unlink_free_space(ctl, entry);
2109                 entry->offset += bytes;
2110                 entry->bytes -= bytes;
2111                 if (!entry->bytes)
2112                         kmem_cache_free(btrfs_free_space_cachep, entry);
2113                 else
2114                         link_free_space(ctl, entry);
2115         }
2116
2117 out:
2118         spin_unlock(&ctl->tree_lock);
2119
2120         return ret;
2121 }
2122
2123 /*
2124  * given a cluster, put all of its extents back into the free space
2125  * cache.  If a block group is passed, this function will only free
2126  * a cluster that belongs to the passed block group.
2127  *
2128  * Otherwise, it'll get a reference on the block group pointed to by the
2129  * cluster and remove the cluster from it.
2130  */
2131 int btrfs_return_cluster_to_free_space(
2132                                struct btrfs_block_group_cache *block_group,
2133                                struct btrfs_free_cluster *cluster)
2134 {
2135         struct btrfs_free_space_ctl *ctl;
2136         int ret;
2137
2138         /* first, get a safe pointer to the block group */
2139         spin_lock(&cluster->lock);
2140         if (!block_group) {
2141                 block_group = cluster->block_group;
2142                 if (!block_group) {
2143                         spin_unlock(&cluster->lock);
2144                         return 0;
2145                 }
2146         } else if (cluster->block_group != block_group) {
2147                 /* someone else has already freed it don't redo their work */
2148                 spin_unlock(&cluster->lock);
2149                 return 0;
2150         }
2151         atomic_inc(&block_group->count);
2152         spin_unlock(&cluster->lock);
2153
2154         ctl = block_group->free_space_ctl;
2155
2156         /* now return any extents the cluster had on it */
2157         spin_lock(&ctl->tree_lock);
2158         ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2159         spin_unlock(&ctl->tree_lock);
2160
2161         /* finally drop our ref */
2162         btrfs_put_block_group(block_group);
2163         return ret;
2164 }
2165
2166 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2167                                    struct btrfs_free_cluster *cluster,
2168                                    struct btrfs_free_space *entry,
2169                                    u64 bytes, u64 min_start)
2170 {
2171         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2172         int err;
2173         u64 search_start = cluster->window_start;
2174         u64 search_bytes = bytes;
2175         u64 ret = 0;
2176
2177         search_start = min_start;
2178         search_bytes = bytes;
2179
2180         err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2181         if (err)
2182                 return 0;
2183
2184         ret = search_start;
2185         __bitmap_clear_bits(ctl, entry, ret, bytes);
2186
2187         return ret;
2188 }
2189
2190 /*
2191  * given a cluster, try to allocate 'bytes' from it, returns 0
2192  * if it couldn't find anything suitably large, or a logical disk offset
2193  * if things worked out
2194  */
2195 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2196                              struct btrfs_free_cluster *cluster, u64 bytes,
2197                              u64 min_start)
2198 {
2199         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2200         struct btrfs_free_space *entry = NULL;
2201         struct rb_node *node;
2202         u64 ret = 0;
2203
2204         spin_lock(&cluster->lock);
2205         if (bytes > cluster->max_size)
2206                 goto out;
2207
2208         if (cluster->block_group != block_group)
2209                 goto out;
2210
2211         node = rb_first(&cluster->root);
2212         if (!node)
2213                 goto out;
2214
2215         entry = rb_entry(node, struct btrfs_free_space, offset_index);
2216         while(1) {
2217                 if (entry->bytes < bytes ||
2218                     (!entry->bitmap && entry->offset < min_start)) {
2219                         node = rb_next(&entry->offset_index);
2220                         if (!node)
2221                                 break;
2222                         entry = rb_entry(node, struct btrfs_free_space,
2223                                          offset_index);
2224                         continue;
2225                 }
2226
2227                 if (entry->bitmap) {
2228                         ret = btrfs_alloc_from_bitmap(block_group,
2229                                                       cluster, entry, bytes,
2230                                                       cluster->window_start);
2231                         if (ret == 0) {
2232                                 node = rb_next(&entry->offset_index);
2233                                 if (!node)
2234                                         break;
2235                                 entry = rb_entry(node, struct btrfs_free_space,
2236                                                  offset_index);
2237                                 continue;
2238                         }
2239                         cluster->window_start += bytes;
2240                 } else {
2241                         ret = entry->offset;
2242
2243                         entry->offset += bytes;
2244                         entry->bytes -= bytes;
2245                 }
2246
2247                 if (entry->bytes == 0)
2248                         rb_erase(&entry->offset_index, &cluster->root);
2249                 break;
2250         }
2251 out:
2252         spin_unlock(&cluster->lock);
2253
2254         if (!ret)
2255                 return 0;
2256
2257         spin_lock(&ctl->tree_lock);
2258
2259         ctl->free_space -= bytes;
2260         if (entry->bytes == 0) {
2261                 ctl->free_extents--;
2262                 if (entry->bitmap) {
2263                         kfree(entry->bitmap);
2264                         ctl->total_bitmaps--;
2265                         ctl->op->recalc_thresholds(ctl);
2266                 }
2267                 kmem_cache_free(btrfs_free_space_cachep, entry);
2268         }
2269
2270         spin_unlock(&ctl->tree_lock);
2271
2272         return ret;
2273 }
2274
2275 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2276                                 struct btrfs_free_space *entry,
2277                                 struct btrfs_free_cluster *cluster,
2278                                 u64 offset, u64 bytes,
2279                                 u64 cont1_bytes, u64 min_bytes)
2280 {
2281         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2282         unsigned long next_zero;
2283         unsigned long i;
2284         unsigned long want_bits;
2285         unsigned long min_bits;
2286         unsigned long found_bits;
2287         unsigned long start = 0;
2288         unsigned long total_found = 0;
2289         int ret;
2290
2291         i = offset_to_bit(entry->offset, block_group->sectorsize,
2292                           max_t(u64, offset, entry->offset));
2293         want_bits = bytes_to_bits(bytes, block_group->sectorsize);
2294         min_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
2295
2296 again:
2297         found_bits = 0;
2298         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2299                 next_zero = find_next_zero_bit(entry->bitmap,
2300                                                BITS_PER_BITMAP, i);
2301                 if (next_zero - i >= min_bits) {
2302                         found_bits = next_zero - i;
2303                         break;
2304                 }
2305                 i = next_zero;
2306         }
2307
2308         if (!found_bits)
2309                 return -ENOSPC;
2310
2311         if (!total_found) {
2312                 start = i;
2313                 cluster->max_size = 0;
2314         }
2315
2316         total_found += found_bits;
2317
2318         if (cluster->max_size < found_bits * block_group->sectorsize)
2319                 cluster->max_size = found_bits * block_group->sectorsize;
2320
2321         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2322                 i = next_zero + 1;
2323                 goto again;
2324         }
2325
2326         cluster->window_start = start * block_group->sectorsize +
2327                 entry->offset;
2328         rb_erase(&entry->offset_index, &ctl->free_space_offset);
2329         ret = tree_insert_offset(&cluster->root, entry->offset,
2330                                  &entry->offset_index, 1);
2331         BUG_ON(ret); /* -EEXIST; Logic error */
2332
2333         trace_btrfs_setup_cluster(block_group, cluster,
2334                                   total_found * block_group->sectorsize, 1);
2335         return 0;
2336 }
2337
2338 /*
2339  * This searches the block group for just extents to fill the cluster with.
2340  * Try to find a cluster with at least bytes total bytes, at least one
2341  * extent of cont1_bytes, and other clusters of at least min_bytes.
2342  */
2343 static noinline int
2344 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2345                         struct btrfs_free_cluster *cluster,
2346                         struct list_head *bitmaps, u64 offset, u64 bytes,
2347                         u64 cont1_bytes, u64 min_bytes)
2348 {
2349         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2350         struct btrfs_free_space *first = NULL;
2351         struct btrfs_free_space *entry = NULL;
2352         struct btrfs_free_space *last;
2353         struct rb_node *node;
2354         u64 window_start;
2355         u64 window_free;
2356         u64 max_extent;
2357         u64 total_size = 0;
2358
2359         entry = tree_search_offset(ctl, offset, 0, 1);
2360         if (!entry)
2361                 return -ENOSPC;
2362
2363         /*
2364          * We don't want bitmaps, so just move along until we find a normal
2365          * extent entry.
2366          */
2367         while (entry->bitmap || entry->bytes < min_bytes) {
2368                 if (entry->bitmap && list_empty(&entry->list))
2369                         list_add_tail(&entry->list, bitmaps);
2370                 node = rb_next(&entry->offset_index);
2371                 if (!node)
2372                         return -ENOSPC;
2373                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2374         }
2375
2376         window_start = entry->offset;
2377         window_free = entry->bytes;
2378         max_extent = entry->bytes;
2379         first = entry;
2380         last = entry;
2381
2382         for (node = rb_next(&entry->offset_index); node;
2383              node = rb_next(&entry->offset_index)) {
2384                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2385
2386                 if (entry->bitmap) {
2387                         if (list_empty(&entry->list))
2388                                 list_add_tail(&entry->list, bitmaps);
2389                         continue;
2390                 }
2391
2392                 if (entry->bytes < min_bytes)
2393                         continue;
2394
2395                 last = entry;
2396                 window_free += entry->bytes;
2397                 if (entry->bytes > max_extent)
2398                         max_extent = entry->bytes;
2399         }
2400
2401         if (window_free < bytes || max_extent < cont1_bytes)
2402                 return -ENOSPC;
2403
2404         cluster->window_start = first->offset;
2405
2406         node = &first->offset_index;
2407
2408         /*
2409          * now we've found our entries, pull them out of the free space
2410          * cache and put them into the cluster rbtree
2411          */
2412         do {
2413                 int ret;
2414
2415                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2416                 node = rb_next(&entry->offset_index);
2417                 if (entry->bitmap || entry->bytes < min_bytes)
2418                         continue;
2419
2420                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2421                 ret = tree_insert_offset(&cluster->root, entry->offset,
2422                                          &entry->offset_index, 0);
2423                 total_size += entry->bytes;
2424                 BUG_ON(ret); /* -EEXIST; Logic error */
2425         } while (node && entry != last);
2426
2427         cluster->max_size = max_extent;
2428         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2429         return 0;
2430 }
2431
2432 /*
2433  * This specifically looks for bitmaps that may work in the cluster, we assume
2434  * that we have already failed to find extents that will work.
2435  */
2436 static noinline int
2437 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2438                      struct btrfs_free_cluster *cluster,
2439                      struct list_head *bitmaps, u64 offset, u64 bytes,
2440                      u64 cont1_bytes, u64 min_bytes)
2441 {
2442         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2443         struct btrfs_free_space *entry;
2444         int ret = -ENOSPC;
2445         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2446
2447         if (ctl->total_bitmaps == 0)
2448                 return -ENOSPC;
2449
2450         /*
2451          * The bitmap that covers offset won't be in the list unless offset
2452          * is just its start offset.
2453          */
2454         entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2455         if (entry->offset != bitmap_offset) {
2456                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2457                 if (entry && list_empty(&entry->list))
2458                         list_add(&entry->list, bitmaps);
2459         }
2460
2461         list_for_each_entry(entry, bitmaps, list) {
2462                 if (entry->bytes < bytes)
2463                         continue;
2464                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2465                                            bytes, cont1_bytes, min_bytes);
2466                 if (!ret)
2467                         return 0;
2468         }
2469
2470         /*
2471          * The bitmaps list has all the bitmaps that record free space
2472          * starting after offset, so no more search is required.
2473          */
2474         return -ENOSPC;
2475 }
2476
2477 /*
2478  * here we try to find a cluster of blocks in a block group.  The goal
2479  * is to find at least bytes+empty_size.
2480  * We might not find them all in one contiguous area.
2481  *
2482  * returns zero and sets up cluster if things worked out, otherwise
2483  * it returns -enospc
2484  */
2485 int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
2486                              struct btrfs_root *root,
2487                              struct btrfs_block_group_cache *block_group,
2488                              struct btrfs_free_cluster *cluster,
2489                              u64 offset, u64 bytes, u64 empty_size)
2490 {
2491         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2492         struct btrfs_free_space *entry, *tmp;
2493         LIST_HEAD(bitmaps);
2494         u64 min_bytes;
2495         u64 cont1_bytes;
2496         int ret;
2497
2498         /*
2499          * Choose the minimum extent size we'll require for this
2500          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
2501          * For metadata, allow allocates with smaller extents.  For
2502          * data, keep it dense.
2503          */
2504         if (btrfs_test_opt(root, SSD_SPREAD)) {
2505                 cont1_bytes = min_bytes = bytes + empty_size;
2506         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2507                 cont1_bytes = bytes;
2508                 min_bytes = block_group->sectorsize;
2509         } else {
2510                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2511                 min_bytes = block_group->sectorsize;
2512         }
2513
2514         spin_lock(&ctl->tree_lock);
2515
2516         /*
2517          * If we know we don't have enough space to make a cluster don't even
2518          * bother doing all the work to try and find one.
2519          */
2520         if (ctl->free_space < bytes) {
2521                 spin_unlock(&ctl->tree_lock);
2522                 return -ENOSPC;
2523         }
2524
2525         spin_lock(&cluster->lock);
2526
2527         /* someone already found a cluster, hooray */
2528         if (cluster->block_group) {
2529                 ret = 0;
2530                 goto out;
2531         }
2532
2533         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2534                                  min_bytes);
2535
2536         INIT_LIST_HEAD(&bitmaps);
2537         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2538                                       bytes + empty_size,
2539                                       cont1_bytes, min_bytes);
2540         if (ret)
2541                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2542                                            offset, bytes + empty_size,
2543                                            cont1_bytes, min_bytes);
2544
2545         /* Clear our temporary list */
2546         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2547                 list_del_init(&entry->list);
2548
2549         if (!ret) {
2550                 atomic_inc(&block_group->count);
2551                 list_add_tail(&cluster->block_group_list,
2552                               &block_group->cluster_list);
2553                 cluster->block_group = block_group;
2554         } else {
2555                 trace_btrfs_failed_cluster_setup(block_group);
2556         }
2557 out:
2558         spin_unlock(&cluster->lock);
2559         spin_unlock(&ctl->tree_lock);
2560
2561         return ret;
2562 }
2563
2564 /*
2565  * simple code to zero out a cluster
2566  */
2567 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2568 {
2569         spin_lock_init(&cluster->lock);
2570         spin_lock_init(&cluster->refill_lock);
2571         cluster->root = RB_ROOT;
2572         cluster->max_size = 0;
2573         INIT_LIST_HEAD(&cluster->block_group_list);
2574         cluster->block_group = NULL;
2575 }
2576
2577 static int do_trimming(struct btrfs_block_group_cache *block_group,
2578                        u64 *total_trimmed, u64 start, u64 bytes,
2579                        u64 reserved_start, u64 reserved_bytes)
2580 {
2581         struct btrfs_space_info *space_info = block_group->space_info;
2582         struct btrfs_fs_info *fs_info = block_group->fs_info;
2583         int ret;
2584         int update = 0;
2585         u64 trimmed = 0;
2586
2587         spin_lock(&space_info->lock);
2588         spin_lock(&block_group->lock);
2589         if (!block_group->ro) {
2590                 block_group->reserved += reserved_bytes;
2591                 space_info->bytes_reserved += reserved_bytes;
2592                 update = 1;
2593         }
2594         spin_unlock(&block_group->lock);
2595         spin_unlock(&space_info->lock);
2596
2597         ret = btrfs_error_discard_extent(fs_info->extent_root,
2598                                          start, bytes, &trimmed);
2599         if (!ret)
2600                 *total_trimmed += trimmed;
2601
2602         btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2603
2604         if (update) {
2605                 spin_lock(&space_info->lock);
2606                 spin_lock(&block_group->lock);
2607                 if (block_group->ro)
2608                         space_info->bytes_readonly += reserved_bytes;
2609                 block_group->reserved -= reserved_bytes;
2610                 space_info->bytes_reserved -= reserved_bytes;
2611                 spin_unlock(&space_info->lock);
2612                 spin_unlock(&block_group->lock);
2613         }
2614
2615         return ret;
2616 }
2617
2618 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2619                           u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2620 {
2621         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2622         struct btrfs_free_space *entry;
2623         struct rb_node *node;
2624         int ret = 0;
2625         u64 extent_start;
2626         u64 extent_bytes;
2627         u64 bytes;
2628
2629         while (start < end) {
2630                 spin_lock(&ctl->tree_lock);
2631
2632                 if (ctl->free_space < minlen) {
2633                         spin_unlock(&ctl->tree_lock);
2634                         break;
2635                 }
2636
2637                 entry = tree_search_offset(ctl, start, 0, 1);
2638                 if (!entry) {
2639                         spin_unlock(&ctl->tree_lock);
2640                         break;
2641                 }
2642
2643                 /* skip bitmaps */
2644                 while (entry->bitmap) {
2645                         node = rb_next(&entry->offset_index);
2646                         if (!node) {
2647                                 spin_unlock(&ctl->tree_lock);
2648                                 goto out;
2649                         }
2650                         entry = rb_entry(node, struct btrfs_free_space,
2651                                          offset_index);
2652                 }
2653
2654                 if (entry->offset >= end) {
2655                         spin_unlock(&ctl->tree_lock);
2656                         break;
2657                 }
2658
2659                 extent_start = entry->offset;
2660                 extent_bytes = entry->bytes;
2661                 start = max(start, extent_start);
2662                 bytes = min(extent_start + extent_bytes, end) - start;
2663                 if (bytes < minlen) {
2664                         spin_unlock(&ctl->tree_lock);
2665                         goto next;
2666                 }
2667
2668                 unlink_free_space(ctl, entry);
2669                 kmem_cache_free(btrfs_free_space_cachep, entry);
2670
2671                 spin_unlock(&ctl->tree_lock);
2672
2673                 ret = do_trimming(block_group, total_trimmed, start, bytes,
2674                                   extent_start, extent_bytes);
2675                 if (ret)
2676                         break;
2677 next:
2678                 start += bytes;
2679
2680                 if (fatal_signal_pending(current)) {
2681                         ret = -ERESTARTSYS;
2682                         break;
2683                 }
2684
2685                 cond_resched();
2686         }
2687 out:
2688         return ret;
2689 }
2690
2691 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2692                         u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2693 {
2694         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2695         struct btrfs_free_space *entry;
2696         int ret = 0;
2697         int ret2;
2698         u64 bytes;
2699         u64 offset = offset_to_bitmap(ctl, start);
2700
2701         while (offset < end) {
2702                 bool next_bitmap = false;
2703
2704                 spin_lock(&ctl->tree_lock);
2705
2706                 if (ctl->free_space < minlen) {
2707                         spin_unlock(&ctl->tree_lock);
2708                         break;
2709                 }
2710
2711                 entry = tree_search_offset(ctl, offset, 1, 0);
2712                 if (!entry) {
2713                         spin_unlock(&ctl->tree_lock);
2714                         next_bitmap = true;
2715                         goto next;
2716                 }
2717
2718                 bytes = minlen;
2719                 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2720                 if (ret2 || start >= end) {
2721                         spin_unlock(&ctl->tree_lock);
2722                         next_bitmap = true;
2723                         goto next;
2724                 }
2725
2726                 bytes = min(bytes, end - start);
2727                 if (bytes < minlen) {
2728                         spin_unlock(&ctl->tree_lock);
2729                         goto next;
2730                 }
2731
2732                 bitmap_clear_bits(ctl, entry, start, bytes);
2733                 if (entry->bytes == 0)
2734                         free_bitmap(ctl, entry);
2735
2736                 spin_unlock(&ctl->tree_lock);
2737
2738                 ret = do_trimming(block_group, total_trimmed, start, bytes,
2739                                   start, bytes);
2740                 if (ret)
2741                         break;
2742 next:
2743                 if (next_bitmap) {
2744                         offset += BITS_PER_BITMAP * ctl->unit;
2745                 } else {
2746                         start += bytes;
2747                         if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2748                                 offset += BITS_PER_BITMAP * ctl->unit;
2749                 }
2750
2751                 if (fatal_signal_pending(current)) {
2752                         ret = -ERESTARTSYS;
2753                         break;
2754                 }
2755
2756                 cond_resched();
2757         }
2758
2759         return ret;
2760 }
2761
2762 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2763                            u64 *trimmed, u64 start, u64 end, u64 minlen)
2764 {
2765         int ret;
2766
2767         *trimmed = 0;
2768
2769         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2770         if (ret)
2771                 return ret;
2772
2773         ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2774
2775         return ret;
2776 }
2777
2778 /*
2779  * Find the left-most item in the cache tree, and then return the
2780  * smallest inode number in the item.
2781  *
2782  * Note: the returned inode number may not be the smallest one in
2783  * the tree, if the left-most item is a bitmap.
2784  */
2785 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2786 {
2787         struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2788         struct btrfs_free_space *entry = NULL;
2789         u64 ino = 0;
2790
2791         spin_lock(&ctl->tree_lock);
2792
2793         if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2794                 goto out;
2795
2796         entry = rb_entry(rb_first(&ctl->free_space_offset),
2797                          struct btrfs_free_space, offset_index);
2798
2799         if (!entry->bitmap) {
2800                 ino = entry->offset;
2801
2802                 unlink_free_space(ctl, entry);
2803                 entry->offset++;
2804                 entry->bytes--;
2805                 if (!entry->bytes)
2806                         kmem_cache_free(btrfs_free_space_cachep, entry);
2807                 else
2808                         link_free_space(ctl, entry);
2809         } else {
2810                 u64 offset = 0;
2811                 u64 count = 1;
2812                 int ret;
2813
2814                 ret = search_bitmap(ctl, entry, &offset, &count);
2815                 /* Logic error; Should be empty if it can't find anything */
2816                 BUG_ON(ret);
2817
2818                 ino = offset;
2819                 bitmap_clear_bits(ctl, entry, offset, 1);
2820                 if (entry->bytes == 0)
2821                         free_bitmap(ctl, entry);
2822         }
2823 out:
2824         spin_unlock(&ctl->tree_lock);
2825
2826         return ino;
2827 }
2828
2829 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2830                                     struct btrfs_path *path)
2831 {
2832         struct inode *inode = NULL;
2833
2834         spin_lock(&root->cache_lock);
2835         if (root->cache_inode)
2836                 inode = igrab(root->cache_inode);
2837         spin_unlock(&root->cache_lock);
2838         if (inode)
2839                 return inode;
2840
2841         inode = __lookup_free_space_inode(root, path, 0);
2842         if (IS_ERR(inode))
2843                 return inode;
2844
2845         spin_lock(&root->cache_lock);
2846         if (!btrfs_fs_closing(root->fs_info))
2847                 root->cache_inode = igrab(inode);
2848         spin_unlock(&root->cache_lock);
2849
2850         return inode;
2851 }
2852
2853 int create_free_ino_inode(struct btrfs_root *root,
2854                           struct btrfs_trans_handle *trans,
2855                           struct btrfs_path *path)
2856 {
2857         return __create_free_space_inode(root, trans, path,
2858                                          BTRFS_FREE_INO_OBJECTID, 0);
2859 }
2860
2861 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2862 {
2863         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2864         struct btrfs_path *path;
2865         struct inode *inode;
2866         int ret = 0;
2867         u64 root_gen = btrfs_root_generation(&root->root_item);
2868
2869         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2870                 return 0;
2871
2872         /*
2873          * If we're unmounting then just return, since this does a search on the
2874          * normal root and not the commit root and we could deadlock.
2875          */
2876         if (btrfs_fs_closing(fs_info))
2877                 return 0;
2878
2879         path = btrfs_alloc_path();
2880         if (!path)
2881                 return 0;
2882
2883         inode = lookup_free_ino_inode(root, path);
2884         if (IS_ERR(inode))
2885                 goto out;
2886
2887         if (root_gen != BTRFS_I(inode)->generation)
2888                 goto out_put;
2889
2890         ret = __load_free_space_cache(root, inode, ctl, path, 0);
2891
2892         if (ret < 0)
2893                 printk(KERN_ERR "btrfs: failed to load free ino cache for "
2894                        "root %llu\n", root->root_key.objectid);
2895 out_put:
2896         iput(inode);
2897 out:
2898         btrfs_free_path(path);
2899         return ret;
2900 }
2901
2902 int btrfs_write_out_ino_cache(struct btrfs_root *root,
2903                               struct btrfs_trans_handle *trans,
2904                               struct btrfs_path *path)
2905 {
2906         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
2907         struct inode *inode;
2908         int ret;
2909
2910         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
2911                 return 0;
2912
2913         inode = lookup_free_ino_inode(root, path);
2914         if (IS_ERR(inode))
2915                 return 0;
2916
2917         ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
2918         if (ret) {
2919                 btrfs_delalloc_release_metadata(inode, inode->i_size);
2920 #ifdef DEBUG
2921                 printk(KERN_ERR "btrfs: failed to write free ino cache "
2922                        "for root %llu\n", root->root_key.objectid);
2923 #endif
2924         }
2925
2926         iput(inode);
2927         return ret;
2928 }