ARM: mvebu: update Armada XP DT for dynamic frequency scaling
[cascardo/linux.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                 btrfs_info(root->fs_info,
108                         "Old style space inode found, converting.");
109                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
110                         BTRFS_INODE_NODATACOW;
111                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
112         }
113
114         if (!block_group->iref) {
115                 block_group->inode = igrab(inode);
116                 block_group->iref = 1;
117         }
118         spin_unlock(&block_group->lock);
119
120         return inode;
121 }
122
123 static int __create_free_space_inode(struct btrfs_root *root,
124                                      struct btrfs_trans_handle *trans,
125                                      struct btrfs_path *path,
126                                      u64 ino, u64 offset)
127 {
128         struct btrfs_key key;
129         struct btrfs_disk_key disk_key;
130         struct btrfs_free_space_header *header;
131         struct btrfs_inode_item *inode_item;
132         struct extent_buffer *leaf;
133         u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
134         int ret;
135
136         ret = btrfs_insert_empty_inode(trans, root, path, ino);
137         if (ret)
138                 return ret;
139
140         /* We inline crc's for the free disk space cache */
141         if (ino != BTRFS_FREE_INO_OBJECTID)
142                 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
143
144         leaf = path->nodes[0];
145         inode_item = btrfs_item_ptr(leaf, path->slots[0],
146                                     struct btrfs_inode_item);
147         btrfs_item_key(leaf, &disk_key, path->slots[0]);
148         memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
149                              sizeof(*inode_item));
150         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
151         btrfs_set_inode_size(leaf, inode_item, 0);
152         btrfs_set_inode_nbytes(leaf, inode_item, 0);
153         btrfs_set_inode_uid(leaf, inode_item, 0);
154         btrfs_set_inode_gid(leaf, inode_item, 0);
155         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
156         btrfs_set_inode_flags(leaf, inode_item, flags);
157         btrfs_set_inode_nlink(leaf, inode_item, 1);
158         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
159         btrfs_set_inode_block_group(leaf, inode_item, offset);
160         btrfs_mark_buffer_dirty(leaf);
161         btrfs_release_path(path);
162
163         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
164         key.offset = offset;
165         key.type = 0;
166
167         ret = btrfs_insert_empty_item(trans, root, path, &key,
168                                       sizeof(struct btrfs_free_space_header));
169         if (ret < 0) {
170                 btrfs_release_path(path);
171                 return ret;
172         }
173         leaf = path->nodes[0];
174         header = btrfs_item_ptr(leaf, path->slots[0],
175                                 struct btrfs_free_space_header);
176         memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
177         btrfs_set_free_space_key(leaf, header, &disk_key);
178         btrfs_mark_buffer_dirty(leaf);
179         btrfs_release_path(path);
180
181         return 0;
182 }
183
184 int create_free_space_inode(struct btrfs_root *root,
185                             struct btrfs_trans_handle *trans,
186                             struct btrfs_block_group_cache *block_group,
187                             struct btrfs_path *path)
188 {
189         int ret;
190         u64 ino;
191
192         ret = btrfs_find_free_objectid(root, &ino);
193         if (ret < 0)
194                 return ret;
195
196         return __create_free_space_inode(root, trans, path, ino,
197                                          block_group->key.objectid);
198 }
199
200 int btrfs_check_trunc_cache_free_space(struct btrfs_root *root,
201                                        struct btrfs_block_rsv *rsv)
202 {
203         u64 needed_bytes;
204         int ret;
205
206         /* 1 for slack space, 1 for updating the inode */
207         needed_bytes = btrfs_calc_trunc_metadata_size(root, 1) +
208                 btrfs_calc_trans_metadata_size(root, 1);
209
210         spin_lock(&rsv->lock);
211         if (rsv->reserved < needed_bytes)
212                 ret = -ENOSPC;
213         else
214                 ret = 0;
215         spin_unlock(&rsv->lock);
216         return ret;
217 }
218
219 int btrfs_truncate_free_space_cache(struct btrfs_root *root,
220                                     struct btrfs_trans_handle *trans,
221                                     struct inode *inode)
222 {
223         int ret = 0;
224
225         btrfs_i_size_write(inode, 0);
226         truncate_pagecache(inode, 0);
227
228         /*
229          * We don't need an orphan item because truncating the free space cache
230          * will never be split across transactions.
231          */
232         ret = btrfs_truncate_inode_items(trans, root, inode,
233                                          0, BTRFS_EXTENT_DATA_KEY);
234         if (ret) {
235                 btrfs_abort_transaction(trans, root, ret);
236                 return ret;
237         }
238
239         ret = btrfs_update_inode(trans, root, inode);
240         if (ret)
241                 btrfs_abort_transaction(trans, root, ret);
242
243         return ret;
244 }
245
246 static int readahead_cache(struct inode *inode)
247 {
248         struct file_ra_state *ra;
249         unsigned long last_index;
250
251         ra = kzalloc(sizeof(*ra), GFP_NOFS);
252         if (!ra)
253                 return -ENOMEM;
254
255         file_ra_state_init(ra, inode->i_mapping);
256         last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
257
258         page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
259
260         kfree(ra);
261
262         return 0;
263 }
264
265 struct io_ctl {
266         void *cur, *orig;
267         struct page *page;
268         struct page **pages;
269         struct btrfs_root *root;
270         unsigned long size;
271         int index;
272         int num_pages;
273         unsigned check_crcs:1;
274 };
275
276 static int io_ctl_init(struct io_ctl *io_ctl, struct inode *inode,
277                        struct btrfs_root *root)
278 {
279         memset(io_ctl, 0, sizeof(struct io_ctl));
280         io_ctl->num_pages = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >>
281                 PAGE_CACHE_SHIFT;
282         io_ctl->pages = kzalloc(sizeof(struct page *) * io_ctl->num_pages,
283                                 GFP_NOFS);
284         if (!io_ctl->pages)
285                 return -ENOMEM;
286         io_ctl->root = root;
287         if (btrfs_ino(inode) != BTRFS_FREE_INO_OBJECTID)
288                 io_ctl->check_crcs = 1;
289         return 0;
290 }
291
292 static void io_ctl_free(struct io_ctl *io_ctl)
293 {
294         kfree(io_ctl->pages);
295 }
296
297 static void io_ctl_unmap_page(struct io_ctl *io_ctl)
298 {
299         if (io_ctl->cur) {
300                 kunmap(io_ctl->page);
301                 io_ctl->cur = NULL;
302                 io_ctl->orig = NULL;
303         }
304 }
305
306 static void io_ctl_map_page(struct io_ctl *io_ctl, int clear)
307 {
308         ASSERT(io_ctl->index < io_ctl->num_pages);
309         io_ctl->page = io_ctl->pages[io_ctl->index++];
310         io_ctl->cur = kmap(io_ctl->page);
311         io_ctl->orig = io_ctl->cur;
312         io_ctl->size = PAGE_CACHE_SIZE;
313         if (clear)
314                 memset(io_ctl->cur, 0, PAGE_CACHE_SIZE);
315 }
316
317 static void io_ctl_drop_pages(struct io_ctl *io_ctl)
318 {
319         int i;
320
321         io_ctl_unmap_page(io_ctl);
322
323         for (i = 0; i < io_ctl->num_pages; i++) {
324                 if (io_ctl->pages[i]) {
325                         ClearPageChecked(io_ctl->pages[i]);
326                         unlock_page(io_ctl->pages[i]);
327                         page_cache_release(io_ctl->pages[i]);
328                 }
329         }
330 }
331
332 static int io_ctl_prepare_pages(struct io_ctl *io_ctl, struct inode *inode,
333                                 int uptodate)
334 {
335         struct page *page;
336         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
337         int i;
338
339         for (i = 0; i < io_ctl->num_pages; i++) {
340                 page = find_or_create_page(inode->i_mapping, i, mask);
341                 if (!page) {
342                         io_ctl_drop_pages(io_ctl);
343                         return -ENOMEM;
344                 }
345                 io_ctl->pages[i] = page;
346                 if (uptodate && !PageUptodate(page)) {
347                         btrfs_readpage(NULL, page);
348                         lock_page(page);
349                         if (!PageUptodate(page)) {
350                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
351                                            "error reading free space cache");
352                                 io_ctl_drop_pages(io_ctl);
353                                 return -EIO;
354                         }
355                 }
356         }
357
358         for (i = 0; i < io_ctl->num_pages; i++) {
359                 clear_page_dirty_for_io(io_ctl->pages[i]);
360                 set_page_extent_mapped(io_ctl->pages[i]);
361         }
362
363         return 0;
364 }
365
366 static void io_ctl_set_generation(struct io_ctl *io_ctl, u64 generation)
367 {
368         __le64 *val;
369
370         io_ctl_map_page(io_ctl, 1);
371
372         /*
373          * Skip the csum areas.  If we don't check crcs then we just have a
374          * 64bit chunk at the front of the first page.
375          */
376         if (io_ctl->check_crcs) {
377                 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
378                 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
379         } else {
380                 io_ctl->cur += sizeof(u64);
381                 io_ctl->size -= sizeof(u64) * 2;
382         }
383
384         val = io_ctl->cur;
385         *val = cpu_to_le64(generation);
386         io_ctl->cur += sizeof(u64);
387 }
388
389 static int io_ctl_check_generation(struct io_ctl *io_ctl, u64 generation)
390 {
391         __le64 *gen;
392
393         /*
394          * Skip the crc area.  If we don't check crcs then we just have a 64bit
395          * chunk at the front of the first page.
396          */
397         if (io_ctl->check_crcs) {
398                 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
399                 io_ctl->size -= sizeof(u64) +
400                         (sizeof(u32) * io_ctl->num_pages);
401         } else {
402                 io_ctl->cur += sizeof(u64);
403                 io_ctl->size -= sizeof(u64) * 2;
404         }
405
406         gen = io_ctl->cur;
407         if (le64_to_cpu(*gen) != generation) {
408                 printk_ratelimited(KERN_ERR "BTRFS: space cache generation "
409                                    "(%Lu) does not match inode (%Lu)\n", *gen,
410                                    generation);
411                 io_ctl_unmap_page(io_ctl);
412                 return -EIO;
413         }
414         io_ctl->cur += sizeof(u64);
415         return 0;
416 }
417
418 static void io_ctl_set_crc(struct io_ctl *io_ctl, int index)
419 {
420         u32 *tmp;
421         u32 crc = ~(u32)0;
422         unsigned offset = 0;
423
424         if (!io_ctl->check_crcs) {
425                 io_ctl_unmap_page(io_ctl);
426                 return;
427         }
428
429         if (index == 0)
430                 offset = sizeof(u32) * io_ctl->num_pages;
431
432         crc = btrfs_csum_data(io_ctl->orig + offset, crc,
433                               PAGE_CACHE_SIZE - offset);
434         btrfs_csum_final(crc, (char *)&crc);
435         io_ctl_unmap_page(io_ctl);
436         tmp = kmap(io_ctl->pages[0]);
437         tmp += index;
438         *tmp = crc;
439         kunmap(io_ctl->pages[0]);
440 }
441
442 static int io_ctl_check_crc(struct io_ctl *io_ctl, int index)
443 {
444         u32 *tmp, val;
445         u32 crc = ~(u32)0;
446         unsigned offset = 0;
447
448         if (!io_ctl->check_crcs) {
449                 io_ctl_map_page(io_ctl, 0);
450                 return 0;
451         }
452
453         if (index == 0)
454                 offset = sizeof(u32) * io_ctl->num_pages;
455
456         tmp = kmap(io_ctl->pages[0]);
457         tmp += index;
458         val = *tmp;
459         kunmap(io_ctl->pages[0]);
460
461         io_ctl_map_page(io_ctl, 0);
462         crc = btrfs_csum_data(io_ctl->orig + offset, crc,
463                               PAGE_CACHE_SIZE - offset);
464         btrfs_csum_final(crc, (char *)&crc);
465         if (val != crc) {
466                 printk_ratelimited(KERN_ERR "BTRFS: csum mismatch on free "
467                                    "space cache\n");
468                 io_ctl_unmap_page(io_ctl);
469                 return -EIO;
470         }
471
472         return 0;
473 }
474
475 static int io_ctl_add_entry(struct io_ctl *io_ctl, u64 offset, u64 bytes,
476                             void *bitmap)
477 {
478         struct btrfs_free_space_entry *entry;
479
480         if (!io_ctl->cur)
481                 return -ENOSPC;
482
483         entry = io_ctl->cur;
484         entry->offset = cpu_to_le64(offset);
485         entry->bytes = cpu_to_le64(bytes);
486         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
487                 BTRFS_FREE_SPACE_EXTENT;
488         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
489         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
490
491         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
492                 return 0;
493
494         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
495
496         /* No more pages to map */
497         if (io_ctl->index >= io_ctl->num_pages)
498                 return 0;
499
500         /* map the next page */
501         io_ctl_map_page(io_ctl, 1);
502         return 0;
503 }
504
505 static int io_ctl_add_bitmap(struct io_ctl *io_ctl, void *bitmap)
506 {
507         if (!io_ctl->cur)
508                 return -ENOSPC;
509
510         /*
511          * If we aren't at the start of the current page, unmap this one and
512          * map the next one if there is any left.
513          */
514         if (io_ctl->cur != io_ctl->orig) {
515                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
516                 if (io_ctl->index >= io_ctl->num_pages)
517                         return -ENOSPC;
518                 io_ctl_map_page(io_ctl, 0);
519         }
520
521         memcpy(io_ctl->cur, bitmap, PAGE_CACHE_SIZE);
522         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
523         if (io_ctl->index < io_ctl->num_pages)
524                 io_ctl_map_page(io_ctl, 0);
525         return 0;
526 }
527
528 static void io_ctl_zero_remaining_pages(struct io_ctl *io_ctl)
529 {
530         /*
531          * If we're not on the boundary we know we've modified the page and we
532          * need to crc the page.
533          */
534         if (io_ctl->cur != io_ctl->orig)
535                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
536         else
537                 io_ctl_unmap_page(io_ctl);
538
539         while (io_ctl->index < io_ctl->num_pages) {
540                 io_ctl_map_page(io_ctl, 1);
541                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
542         }
543 }
544
545 static int io_ctl_read_entry(struct io_ctl *io_ctl,
546                             struct btrfs_free_space *entry, u8 *type)
547 {
548         struct btrfs_free_space_entry *e;
549         int ret;
550
551         if (!io_ctl->cur) {
552                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
553                 if (ret)
554                         return ret;
555         }
556
557         e = io_ctl->cur;
558         entry->offset = le64_to_cpu(e->offset);
559         entry->bytes = le64_to_cpu(e->bytes);
560         *type = e->type;
561         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
562         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
563
564         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
565                 return 0;
566
567         io_ctl_unmap_page(io_ctl);
568
569         return 0;
570 }
571
572 static int io_ctl_read_bitmap(struct io_ctl *io_ctl,
573                               struct btrfs_free_space *entry)
574 {
575         int ret;
576
577         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
578         if (ret)
579                 return ret;
580
581         memcpy(entry->bitmap, io_ctl->cur, PAGE_CACHE_SIZE);
582         io_ctl_unmap_page(io_ctl);
583
584         return 0;
585 }
586
587 /*
588  * Since we attach pinned extents after the fact we can have contiguous sections
589  * of free space that are split up in entries.  This poses a problem with the
590  * tree logging stuff since it could have allocated across what appears to be 2
591  * entries since we would have merged the entries when adding the pinned extents
592  * back to the free space cache.  So run through the space cache that we just
593  * loaded and merge contiguous entries.  This will make the log replay stuff not
594  * blow up and it will make for nicer allocator behavior.
595  */
596 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
597 {
598         struct btrfs_free_space *e, *prev = NULL;
599         struct rb_node *n;
600
601 again:
602         spin_lock(&ctl->tree_lock);
603         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
604                 e = rb_entry(n, struct btrfs_free_space, offset_index);
605                 if (!prev)
606                         goto next;
607                 if (e->bitmap || prev->bitmap)
608                         goto next;
609                 if (prev->offset + prev->bytes == e->offset) {
610                         unlink_free_space(ctl, prev);
611                         unlink_free_space(ctl, e);
612                         prev->bytes += e->bytes;
613                         kmem_cache_free(btrfs_free_space_cachep, e);
614                         link_free_space(ctl, prev);
615                         prev = NULL;
616                         spin_unlock(&ctl->tree_lock);
617                         goto again;
618                 }
619 next:
620                 prev = e;
621         }
622         spin_unlock(&ctl->tree_lock);
623 }
624
625 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
626                                    struct btrfs_free_space_ctl *ctl,
627                                    struct btrfs_path *path, u64 offset)
628 {
629         struct btrfs_free_space_header *header;
630         struct extent_buffer *leaf;
631         struct io_ctl io_ctl;
632         struct btrfs_key key;
633         struct btrfs_free_space *e, *n;
634         struct list_head bitmaps;
635         u64 num_entries;
636         u64 num_bitmaps;
637         u64 generation;
638         u8 type;
639         int ret = 0;
640
641         INIT_LIST_HEAD(&bitmaps);
642
643         /* Nothing in the space cache, goodbye */
644         if (!i_size_read(inode))
645                 return 0;
646
647         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
648         key.offset = offset;
649         key.type = 0;
650
651         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
652         if (ret < 0)
653                 return 0;
654         else if (ret > 0) {
655                 btrfs_release_path(path);
656                 return 0;
657         }
658
659         ret = -1;
660
661         leaf = path->nodes[0];
662         header = btrfs_item_ptr(leaf, path->slots[0],
663                                 struct btrfs_free_space_header);
664         num_entries = btrfs_free_space_entries(leaf, header);
665         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
666         generation = btrfs_free_space_generation(leaf, header);
667         btrfs_release_path(path);
668
669         if (BTRFS_I(inode)->generation != generation) {
670                 btrfs_err(root->fs_info,
671                         "free space inode generation (%llu) "
672                         "did not match free space cache generation (%llu)",
673                         BTRFS_I(inode)->generation, generation);
674                 return 0;
675         }
676
677         if (!num_entries)
678                 return 0;
679
680         ret = io_ctl_init(&io_ctl, inode, root);
681         if (ret)
682                 return ret;
683
684         ret = readahead_cache(inode);
685         if (ret)
686                 goto out;
687
688         ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
689         if (ret)
690                 goto out;
691
692         ret = io_ctl_check_crc(&io_ctl, 0);
693         if (ret)
694                 goto free_cache;
695
696         ret = io_ctl_check_generation(&io_ctl, generation);
697         if (ret)
698                 goto free_cache;
699
700         while (num_entries) {
701                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
702                                       GFP_NOFS);
703                 if (!e)
704                         goto free_cache;
705
706                 ret = io_ctl_read_entry(&io_ctl, e, &type);
707                 if (ret) {
708                         kmem_cache_free(btrfs_free_space_cachep, e);
709                         goto free_cache;
710                 }
711
712                 if (!e->bytes) {
713                         kmem_cache_free(btrfs_free_space_cachep, e);
714                         goto free_cache;
715                 }
716
717                 if (type == BTRFS_FREE_SPACE_EXTENT) {
718                         spin_lock(&ctl->tree_lock);
719                         ret = link_free_space(ctl, e);
720                         spin_unlock(&ctl->tree_lock);
721                         if (ret) {
722                                 btrfs_err(root->fs_info,
723                                         "Duplicate entries in free space cache, dumping");
724                                 kmem_cache_free(btrfs_free_space_cachep, e);
725                                 goto free_cache;
726                         }
727                 } else {
728                         ASSERT(num_bitmaps);
729                         num_bitmaps--;
730                         e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
731                         if (!e->bitmap) {
732                                 kmem_cache_free(
733                                         btrfs_free_space_cachep, e);
734                                 goto free_cache;
735                         }
736                         spin_lock(&ctl->tree_lock);
737                         ret = link_free_space(ctl, e);
738                         ctl->total_bitmaps++;
739                         ctl->op->recalc_thresholds(ctl);
740                         spin_unlock(&ctl->tree_lock);
741                         if (ret) {
742                                 btrfs_err(root->fs_info,
743                                         "Duplicate entries in free space cache, dumping");
744                                 kmem_cache_free(btrfs_free_space_cachep, e);
745                                 goto free_cache;
746                         }
747                         list_add_tail(&e->list, &bitmaps);
748                 }
749
750                 num_entries--;
751         }
752
753         io_ctl_unmap_page(&io_ctl);
754
755         /*
756          * We add the bitmaps at the end of the entries in order that
757          * the bitmap entries are added to the cache.
758          */
759         list_for_each_entry_safe(e, n, &bitmaps, list) {
760                 list_del_init(&e->list);
761                 ret = io_ctl_read_bitmap(&io_ctl, e);
762                 if (ret)
763                         goto free_cache;
764         }
765
766         io_ctl_drop_pages(&io_ctl);
767         merge_space_tree(ctl);
768         ret = 1;
769 out:
770         io_ctl_free(&io_ctl);
771         return ret;
772 free_cache:
773         io_ctl_drop_pages(&io_ctl);
774         __btrfs_remove_free_space_cache(ctl);
775         goto out;
776 }
777
778 int load_free_space_cache(struct btrfs_fs_info *fs_info,
779                           struct btrfs_block_group_cache *block_group)
780 {
781         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
782         struct btrfs_root *root = fs_info->tree_root;
783         struct inode *inode;
784         struct btrfs_path *path;
785         int ret = 0;
786         bool matched;
787         u64 used = btrfs_block_group_used(&block_group->item);
788
789         /*
790          * If this block group has been marked to be cleared for one reason or
791          * another then we can't trust the on disk cache, so just return.
792          */
793         spin_lock(&block_group->lock);
794         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
795                 spin_unlock(&block_group->lock);
796                 return 0;
797         }
798         spin_unlock(&block_group->lock);
799
800         path = btrfs_alloc_path();
801         if (!path)
802                 return 0;
803         path->search_commit_root = 1;
804         path->skip_locking = 1;
805
806         inode = lookup_free_space_inode(root, block_group, path);
807         if (IS_ERR(inode)) {
808                 btrfs_free_path(path);
809                 return 0;
810         }
811
812         /* We may have converted the inode and made the cache invalid. */
813         spin_lock(&block_group->lock);
814         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
815                 spin_unlock(&block_group->lock);
816                 btrfs_free_path(path);
817                 goto out;
818         }
819         spin_unlock(&block_group->lock);
820
821         ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
822                                       path, block_group->key.objectid);
823         btrfs_free_path(path);
824         if (ret <= 0)
825                 goto out;
826
827         spin_lock(&ctl->tree_lock);
828         matched = (ctl->free_space == (block_group->key.offset - used -
829                                        block_group->bytes_super));
830         spin_unlock(&ctl->tree_lock);
831
832         if (!matched) {
833                 __btrfs_remove_free_space_cache(ctl);
834                 btrfs_warn(fs_info, "block group %llu has wrong amount of free space",
835                         block_group->key.objectid);
836                 ret = -1;
837         }
838 out:
839         if (ret < 0) {
840                 /* This cache is bogus, make sure it gets cleared */
841                 spin_lock(&block_group->lock);
842                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
843                 spin_unlock(&block_group->lock);
844                 ret = 0;
845
846                 btrfs_warn(fs_info, "failed to load free space cache for block group %llu, rebuild it now",
847                         block_group->key.objectid);
848         }
849
850         iput(inode);
851         return ret;
852 }
853
854 static noinline_for_stack
855 int write_cache_extent_entries(struct io_ctl *io_ctl,
856                               struct btrfs_free_space_ctl *ctl,
857                               struct btrfs_block_group_cache *block_group,
858                               int *entries, int *bitmaps,
859                               struct list_head *bitmap_list)
860 {
861         int ret;
862         struct btrfs_free_cluster *cluster = NULL;
863         struct rb_node *node = rb_first(&ctl->free_space_offset);
864
865         /* Get the cluster for this block_group if it exists */
866         if (block_group && !list_empty(&block_group->cluster_list)) {
867                 cluster = list_entry(block_group->cluster_list.next,
868                                      struct btrfs_free_cluster,
869                                      block_group_list);
870         }
871
872         if (!node && cluster) {
873                 node = rb_first(&cluster->root);
874                 cluster = NULL;
875         }
876
877         /* Write out the extent entries */
878         while (node) {
879                 struct btrfs_free_space *e;
880
881                 e = rb_entry(node, struct btrfs_free_space, offset_index);
882                 *entries += 1;
883
884                 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
885                                        e->bitmap);
886                 if (ret)
887                         goto fail;
888
889                 if (e->bitmap) {
890                         list_add_tail(&e->list, bitmap_list);
891                         *bitmaps += 1;
892                 }
893                 node = rb_next(node);
894                 if (!node && cluster) {
895                         node = rb_first(&cluster->root);
896                         cluster = NULL;
897                 }
898         }
899         return 0;
900 fail:
901         return -ENOSPC;
902 }
903
904 static noinline_for_stack int
905 update_cache_item(struct btrfs_trans_handle *trans,
906                   struct btrfs_root *root,
907                   struct inode *inode,
908                   struct btrfs_path *path, u64 offset,
909                   int entries, int bitmaps)
910 {
911         struct btrfs_key key;
912         struct btrfs_free_space_header *header;
913         struct extent_buffer *leaf;
914         int ret;
915
916         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
917         key.offset = offset;
918         key.type = 0;
919
920         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
921         if (ret < 0) {
922                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
923                                  EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
924                                  GFP_NOFS);
925                 goto fail;
926         }
927         leaf = path->nodes[0];
928         if (ret > 0) {
929                 struct btrfs_key found_key;
930                 ASSERT(path->slots[0]);
931                 path->slots[0]--;
932                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
933                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
934                     found_key.offset != offset) {
935                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
936                                          inode->i_size - 1,
937                                          EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
938                                          NULL, GFP_NOFS);
939                         btrfs_release_path(path);
940                         goto fail;
941                 }
942         }
943
944         BTRFS_I(inode)->generation = trans->transid;
945         header = btrfs_item_ptr(leaf, path->slots[0],
946                                 struct btrfs_free_space_header);
947         btrfs_set_free_space_entries(leaf, header, entries);
948         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
949         btrfs_set_free_space_generation(leaf, header, trans->transid);
950         btrfs_mark_buffer_dirty(leaf);
951         btrfs_release_path(path);
952
953         return 0;
954
955 fail:
956         return -1;
957 }
958
959 static noinline_for_stack int
960 add_ioctl_entries(struct btrfs_root *root,
961                   struct inode *inode,
962                   struct btrfs_block_group_cache *block_group,
963                   struct io_ctl *io_ctl,
964                   struct extent_state **cached_state,
965                   struct list_head *bitmap_list,
966                   int *entries)
967 {
968         u64 start, extent_start, extent_end, len;
969         struct list_head *pos, *n;
970         struct extent_io_tree *unpin = NULL;
971         int ret;
972
973         /*
974          * We want to add any pinned extents to our free space cache
975          * so we don't leak the space
976          *
977          * We shouldn't have switched the pinned extents yet so this is the
978          * right one
979          */
980         unpin = root->fs_info->pinned_extents;
981
982         if (block_group)
983                 start = block_group->key.objectid;
984
985         while (block_group && (start < block_group->key.objectid +
986                                block_group->key.offset)) {
987                 ret = find_first_extent_bit(unpin, start,
988                                             &extent_start, &extent_end,
989                                             EXTENT_DIRTY, NULL);
990                 if (ret) {
991                         ret = 0;
992                         break;
993                 }
994
995                 /* This pinned extent is out of our range */
996                 if (extent_start >= block_group->key.objectid +
997                     block_group->key.offset)
998                         break;
999
1000                 extent_start = max(extent_start, start);
1001                 extent_end = min(block_group->key.objectid +
1002                                  block_group->key.offset, extent_end + 1);
1003                 len = extent_end - extent_start;
1004
1005                 *entries += 1;
1006                 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1007                 if (ret)
1008                         goto out_nospc;
1009
1010                 start = extent_end;
1011         }
1012
1013         /* Write out the bitmaps */
1014         list_for_each_safe(pos, n, bitmap_list) {
1015                 struct btrfs_free_space *entry =
1016                         list_entry(pos, struct btrfs_free_space, list);
1017
1018                 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1019                 if (ret)
1020                         goto out_nospc;
1021                 list_del_init(&entry->list);
1022         }
1023
1024         /* Zero out the rest of the pages just to make sure */
1025         io_ctl_zero_remaining_pages(io_ctl);
1026
1027         ret = btrfs_dirty_pages(root, inode, io_ctl->pages, io_ctl->num_pages,
1028                                 0, i_size_read(inode), cached_state);
1029         io_ctl_drop_pages(io_ctl);
1030         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1031                              i_size_read(inode) - 1, cached_state, GFP_NOFS);
1032
1033         if (ret)
1034                 goto fail;
1035
1036         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1037         if (ret) {
1038                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1039                                  EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
1040                                  GFP_NOFS);
1041                 goto fail;
1042         }
1043         return 0;
1044
1045 fail:
1046         return -1;
1047
1048 out_nospc:
1049         return -ENOSPC;
1050 }
1051
1052 static void noinline_for_stack
1053 cleanup_write_cache_enospc(struct inode *inode,
1054                            struct io_ctl *io_ctl,
1055                            struct extent_state **cached_state,
1056                            struct list_head *bitmap_list)
1057 {
1058         struct list_head *pos, *n;
1059         list_for_each_safe(pos, n, bitmap_list) {
1060                 struct btrfs_free_space *entry =
1061                         list_entry(pos, struct btrfs_free_space, list);
1062                 list_del_init(&entry->list);
1063         }
1064         io_ctl_drop_pages(io_ctl);
1065         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1066                              i_size_read(inode) - 1, cached_state,
1067                              GFP_NOFS);
1068 }
1069
1070 /**
1071  * __btrfs_write_out_cache - write out cached info to an inode
1072  * @root - the root the inode belongs to
1073  * @ctl - the free space cache we are going to write out
1074  * @block_group - the block_group for this cache if it belongs to a block_group
1075  * @trans - the trans handle
1076  * @path - the path to use
1077  * @offset - the offset for the key we'll insert
1078  *
1079  * This function writes out a free space cache struct to disk for quick recovery
1080  * on mount.  This will return 0 if it was successfull in writing the cache out,
1081  * and -1 if it was not.
1082  */
1083 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1084                                    struct btrfs_free_space_ctl *ctl,
1085                                    struct btrfs_block_group_cache *block_group,
1086                                    struct btrfs_trans_handle *trans,
1087                                    struct btrfs_path *path, u64 offset)
1088 {
1089         struct extent_state *cached_state = NULL;
1090         struct io_ctl io_ctl;
1091         struct list_head bitmap_list;
1092         int entries = 0;
1093         int bitmaps = 0;
1094         int ret;
1095         int err = -1;
1096
1097         INIT_LIST_HEAD(&bitmap_list);
1098
1099         if (!i_size_read(inode))
1100                 return -1;
1101
1102         ret = io_ctl_init(&io_ctl, inode, root);
1103         if (ret)
1104                 return -1;
1105
1106         /* Lock all pages first so we can lock the extent safely. */
1107         io_ctl_prepare_pages(&io_ctl, inode, 0);
1108
1109         lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1110                          0, &cached_state);
1111
1112
1113         /* Make sure we can fit our crcs into the first page */
1114         if (io_ctl.check_crcs &&
1115             (io_ctl.num_pages * sizeof(u32)) >= PAGE_CACHE_SIZE)
1116                 goto out_nospc;
1117
1118         io_ctl_set_generation(&io_ctl, trans->transid);
1119
1120         ret = write_cache_extent_entries(&io_ctl, ctl,
1121                                          block_group, &entries, &bitmaps,
1122                                          &bitmap_list);
1123         if (ret)
1124                 goto out_nospc;
1125
1126         ret = add_ioctl_entries(root, inode, block_group, &io_ctl,
1127                                 &cached_state, &bitmap_list, &entries);
1128
1129         if (ret == -ENOSPC)
1130                 goto out_nospc;
1131         else if (ret)
1132                 goto out;
1133
1134         err = update_cache_item(trans, root, inode, path, offset,
1135                                 entries, bitmaps);
1136
1137 out:
1138         io_ctl_free(&io_ctl);
1139         if (err) {
1140                 invalidate_inode_pages2(inode->i_mapping);
1141                 BTRFS_I(inode)->generation = 0;
1142         }
1143         btrfs_update_inode(trans, root, inode);
1144         return err;
1145
1146 out_nospc:
1147
1148         cleanup_write_cache_enospc(inode, &io_ctl, &cached_state, &bitmap_list);
1149         goto out;
1150 }
1151
1152 int btrfs_write_out_cache(struct btrfs_root *root,
1153                           struct btrfs_trans_handle *trans,
1154                           struct btrfs_block_group_cache *block_group,
1155                           struct btrfs_path *path)
1156 {
1157         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1158         struct inode *inode;
1159         int ret = 0;
1160
1161         root = root->fs_info->tree_root;
1162
1163         spin_lock(&block_group->lock);
1164         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1165                 spin_unlock(&block_group->lock);
1166                 return 0;
1167         }
1168         spin_unlock(&block_group->lock);
1169
1170         inode = lookup_free_space_inode(root, block_group, path);
1171         if (IS_ERR(inode))
1172                 return 0;
1173
1174         ret = __btrfs_write_out_cache(root, inode, ctl, block_group, trans,
1175                                       path, block_group->key.objectid);
1176         if (ret) {
1177                 spin_lock(&block_group->lock);
1178                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1179                 spin_unlock(&block_group->lock);
1180                 ret = 0;
1181 #ifdef DEBUG
1182                 btrfs_err(root->fs_info,
1183                         "failed to write free space cache for block group %llu",
1184                         block_group->key.objectid);
1185 #endif
1186         }
1187
1188         iput(inode);
1189         return ret;
1190 }
1191
1192 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1193                                           u64 offset)
1194 {
1195         ASSERT(offset >= bitmap_start);
1196         offset -= bitmap_start;
1197         return (unsigned long)(div_u64(offset, unit));
1198 }
1199
1200 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1201 {
1202         return (unsigned long)(div_u64(bytes, unit));
1203 }
1204
1205 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1206                                    u64 offset)
1207 {
1208         u64 bitmap_start;
1209         u64 bytes_per_bitmap;
1210
1211         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1212         bitmap_start = offset - ctl->start;
1213         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1214         bitmap_start *= bytes_per_bitmap;
1215         bitmap_start += ctl->start;
1216
1217         return bitmap_start;
1218 }
1219
1220 static int tree_insert_offset(struct rb_root *root, u64 offset,
1221                               struct rb_node *node, int bitmap)
1222 {
1223         struct rb_node **p = &root->rb_node;
1224         struct rb_node *parent = NULL;
1225         struct btrfs_free_space *info;
1226
1227         while (*p) {
1228                 parent = *p;
1229                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1230
1231                 if (offset < info->offset) {
1232                         p = &(*p)->rb_left;
1233                 } else if (offset > info->offset) {
1234                         p = &(*p)->rb_right;
1235                 } else {
1236                         /*
1237                          * we could have a bitmap entry and an extent entry
1238                          * share the same offset.  If this is the case, we want
1239                          * the extent entry to always be found first if we do a
1240                          * linear search through the tree, since we want to have
1241                          * the quickest allocation time, and allocating from an
1242                          * extent is faster than allocating from a bitmap.  So
1243                          * if we're inserting a bitmap and we find an entry at
1244                          * this offset, we want to go right, or after this entry
1245                          * logically.  If we are inserting an extent and we've
1246                          * found a bitmap, we want to go left, or before
1247                          * logically.
1248                          */
1249                         if (bitmap) {
1250                                 if (info->bitmap) {
1251                                         WARN_ON_ONCE(1);
1252                                         return -EEXIST;
1253                                 }
1254                                 p = &(*p)->rb_right;
1255                         } else {
1256                                 if (!info->bitmap) {
1257                                         WARN_ON_ONCE(1);
1258                                         return -EEXIST;
1259                                 }
1260                                 p = &(*p)->rb_left;
1261                         }
1262                 }
1263         }
1264
1265         rb_link_node(node, parent, p);
1266         rb_insert_color(node, root);
1267
1268         return 0;
1269 }
1270
1271 /*
1272  * searches the tree for the given offset.
1273  *
1274  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1275  * want a section that has at least bytes size and comes at or after the given
1276  * offset.
1277  */
1278 static struct btrfs_free_space *
1279 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1280                    u64 offset, int bitmap_only, int fuzzy)
1281 {
1282         struct rb_node *n = ctl->free_space_offset.rb_node;
1283         struct btrfs_free_space *entry, *prev = NULL;
1284
1285         /* find entry that is closest to the 'offset' */
1286         while (1) {
1287                 if (!n) {
1288                         entry = NULL;
1289                         break;
1290                 }
1291
1292                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1293                 prev = entry;
1294
1295                 if (offset < entry->offset)
1296                         n = n->rb_left;
1297                 else if (offset > entry->offset)
1298                         n = n->rb_right;
1299                 else
1300                         break;
1301         }
1302
1303         if (bitmap_only) {
1304                 if (!entry)
1305                         return NULL;
1306                 if (entry->bitmap)
1307                         return entry;
1308
1309                 /*
1310                  * bitmap entry and extent entry may share same offset,
1311                  * in that case, bitmap entry comes after extent entry.
1312                  */
1313                 n = rb_next(n);
1314                 if (!n)
1315                         return NULL;
1316                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1317                 if (entry->offset != offset)
1318                         return NULL;
1319
1320                 WARN_ON(!entry->bitmap);
1321                 return entry;
1322         } else if (entry) {
1323                 if (entry->bitmap) {
1324                         /*
1325                          * if previous extent entry covers the offset,
1326                          * we should return it instead of the bitmap entry
1327                          */
1328                         n = rb_prev(&entry->offset_index);
1329                         if (n) {
1330                                 prev = rb_entry(n, struct btrfs_free_space,
1331                                                 offset_index);
1332                                 if (!prev->bitmap &&
1333                                     prev->offset + prev->bytes > offset)
1334                                         entry = prev;
1335                         }
1336                 }
1337                 return entry;
1338         }
1339
1340         if (!prev)
1341                 return NULL;
1342
1343         /* find last entry before the 'offset' */
1344         entry = prev;
1345         if (entry->offset > offset) {
1346                 n = rb_prev(&entry->offset_index);
1347                 if (n) {
1348                         entry = rb_entry(n, struct btrfs_free_space,
1349                                         offset_index);
1350                         ASSERT(entry->offset <= offset);
1351                 } else {
1352                         if (fuzzy)
1353                                 return entry;
1354                         else
1355                                 return NULL;
1356                 }
1357         }
1358
1359         if (entry->bitmap) {
1360                 n = rb_prev(&entry->offset_index);
1361                 if (n) {
1362                         prev = rb_entry(n, struct btrfs_free_space,
1363                                         offset_index);
1364                         if (!prev->bitmap &&
1365                             prev->offset + prev->bytes > offset)
1366                                 return prev;
1367                 }
1368                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1369                         return entry;
1370         } else if (entry->offset + entry->bytes > offset)
1371                 return entry;
1372
1373         if (!fuzzy)
1374                 return NULL;
1375
1376         while (1) {
1377                 if (entry->bitmap) {
1378                         if (entry->offset + BITS_PER_BITMAP *
1379                             ctl->unit > offset)
1380                                 break;
1381                 } else {
1382                         if (entry->offset + entry->bytes > offset)
1383                                 break;
1384                 }
1385
1386                 n = rb_next(&entry->offset_index);
1387                 if (!n)
1388                         return NULL;
1389                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1390         }
1391         return entry;
1392 }
1393
1394 static inline void
1395 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1396                     struct btrfs_free_space *info)
1397 {
1398         rb_erase(&info->offset_index, &ctl->free_space_offset);
1399         ctl->free_extents--;
1400 }
1401
1402 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1403                               struct btrfs_free_space *info)
1404 {
1405         __unlink_free_space(ctl, info);
1406         ctl->free_space -= info->bytes;
1407 }
1408
1409 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1410                            struct btrfs_free_space *info)
1411 {
1412         int ret = 0;
1413
1414         ASSERT(info->bytes || info->bitmap);
1415         ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1416                                  &info->offset_index, (info->bitmap != NULL));
1417         if (ret)
1418                 return ret;
1419
1420         ctl->free_space += info->bytes;
1421         ctl->free_extents++;
1422         return ret;
1423 }
1424
1425 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1426 {
1427         struct btrfs_block_group_cache *block_group = ctl->private;
1428         u64 max_bytes;
1429         u64 bitmap_bytes;
1430         u64 extent_bytes;
1431         u64 size = block_group->key.offset;
1432         u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1433         int max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1434
1435         max_bitmaps = max(max_bitmaps, 1);
1436
1437         ASSERT(ctl->total_bitmaps <= max_bitmaps);
1438
1439         /*
1440          * The goal is to keep the total amount of memory used per 1gb of space
1441          * at or below 32k, so we need to adjust how much memory we allow to be
1442          * used by extent based free space tracking
1443          */
1444         if (size < 1024 * 1024 * 1024)
1445                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1446         else
1447                 max_bytes = MAX_CACHE_BYTES_PER_GIG *
1448                         div64_u64(size, 1024 * 1024 * 1024);
1449
1450         /*
1451          * we want to account for 1 more bitmap than what we have so we can make
1452          * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1453          * we add more bitmaps.
1454          */
1455         bitmap_bytes = (ctl->total_bitmaps + 1) * PAGE_CACHE_SIZE;
1456
1457         if (bitmap_bytes >= max_bytes) {
1458                 ctl->extents_thresh = 0;
1459                 return;
1460         }
1461
1462         /*
1463          * we want the extent entry threshold to always be at most 1/2 the maxw
1464          * bytes we can have, or whatever is less than that.
1465          */
1466         extent_bytes = max_bytes - bitmap_bytes;
1467         extent_bytes = min_t(u64, extent_bytes, div64_u64(max_bytes, 2));
1468
1469         ctl->extents_thresh =
1470                 div64_u64(extent_bytes, (sizeof(struct btrfs_free_space)));
1471 }
1472
1473 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1474                                        struct btrfs_free_space *info,
1475                                        u64 offset, u64 bytes)
1476 {
1477         unsigned long start, count;
1478
1479         start = offset_to_bit(info->offset, ctl->unit, offset);
1480         count = bytes_to_bits(bytes, ctl->unit);
1481         ASSERT(start + count <= BITS_PER_BITMAP);
1482
1483         bitmap_clear(info->bitmap, start, count);
1484
1485         info->bytes -= bytes;
1486 }
1487
1488 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1489                               struct btrfs_free_space *info, u64 offset,
1490                               u64 bytes)
1491 {
1492         __bitmap_clear_bits(ctl, info, offset, bytes);
1493         ctl->free_space -= bytes;
1494 }
1495
1496 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1497                             struct btrfs_free_space *info, u64 offset,
1498                             u64 bytes)
1499 {
1500         unsigned long start, count;
1501
1502         start = offset_to_bit(info->offset, ctl->unit, offset);
1503         count = bytes_to_bits(bytes, ctl->unit);
1504         ASSERT(start + count <= BITS_PER_BITMAP);
1505
1506         bitmap_set(info->bitmap, start, count);
1507
1508         info->bytes += bytes;
1509         ctl->free_space += bytes;
1510 }
1511
1512 /*
1513  * If we can not find suitable extent, we will use bytes to record
1514  * the size of the max extent.
1515  */
1516 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1517                          struct btrfs_free_space *bitmap_info, u64 *offset,
1518                          u64 *bytes)
1519 {
1520         unsigned long found_bits = 0;
1521         unsigned long max_bits = 0;
1522         unsigned long bits, i;
1523         unsigned long next_zero;
1524         unsigned long extent_bits;
1525
1526         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1527                           max_t(u64, *offset, bitmap_info->offset));
1528         bits = bytes_to_bits(*bytes, ctl->unit);
1529
1530         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1531                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1532                                                BITS_PER_BITMAP, i);
1533                 extent_bits = next_zero - i;
1534                 if (extent_bits >= bits) {
1535                         found_bits = extent_bits;
1536                         break;
1537                 } else if (extent_bits > max_bits) {
1538                         max_bits = extent_bits;
1539                 }
1540                 i = next_zero;
1541         }
1542
1543         if (found_bits) {
1544                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1545                 *bytes = (u64)(found_bits) * ctl->unit;
1546                 return 0;
1547         }
1548
1549         *bytes = (u64)(max_bits) * ctl->unit;
1550         return -1;
1551 }
1552
1553 /* Cache the size of the max extent in bytes */
1554 static struct btrfs_free_space *
1555 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1556                 unsigned long align, u64 *max_extent_size)
1557 {
1558         struct btrfs_free_space *entry;
1559         struct rb_node *node;
1560         u64 tmp;
1561         u64 align_off;
1562         int ret;
1563
1564         if (!ctl->free_space_offset.rb_node)
1565                 goto out;
1566
1567         entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1568         if (!entry)
1569                 goto out;
1570
1571         for (node = &entry->offset_index; node; node = rb_next(node)) {
1572                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1573                 if (entry->bytes < *bytes) {
1574                         if (entry->bytes > *max_extent_size)
1575                                 *max_extent_size = entry->bytes;
1576                         continue;
1577                 }
1578
1579                 /* make sure the space returned is big enough
1580                  * to match our requested alignment
1581                  */
1582                 if (*bytes >= align) {
1583                         tmp = entry->offset - ctl->start + align - 1;
1584                         do_div(tmp, align);
1585                         tmp = tmp * align + ctl->start;
1586                         align_off = tmp - entry->offset;
1587                 } else {
1588                         align_off = 0;
1589                         tmp = entry->offset;
1590                 }
1591
1592                 if (entry->bytes < *bytes + align_off) {
1593                         if (entry->bytes > *max_extent_size)
1594                                 *max_extent_size = entry->bytes;
1595                         continue;
1596                 }
1597
1598                 if (entry->bitmap) {
1599                         u64 size = *bytes;
1600
1601                         ret = search_bitmap(ctl, entry, &tmp, &size);
1602                         if (!ret) {
1603                                 *offset = tmp;
1604                                 *bytes = size;
1605                                 return entry;
1606                         } else if (size > *max_extent_size) {
1607                                 *max_extent_size = size;
1608                         }
1609                         continue;
1610                 }
1611
1612                 *offset = tmp;
1613                 *bytes = entry->bytes - align_off;
1614                 return entry;
1615         }
1616 out:
1617         return NULL;
1618 }
1619
1620 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1621                            struct btrfs_free_space *info, u64 offset)
1622 {
1623         info->offset = offset_to_bitmap(ctl, offset);
1624         info->bytes = 0;
1625         INIT_LIST_HEAD(&info->list);
1626         link_free_space(ctl, info);
1627         ctl->total_bitmaps++;
1628
1629         ctl->op->recalc_thresholds(ctl);
1630 }
1631
1632 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1633                         struct btrfs_free_space *bitmap_info)
1634 {
1635         unlink_free_space(ctl, bitmap_info);
1636         kfree(bitmap_info->bitmap);
1637         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1638         ctl->total_bitmaps--;
1639         ctl->op->recalc_thresholds(ctl);
1640 }
1641
1642 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1643                               struct btrfs_free_space *bitmap_info,
1644                               u64 *offset, u64 *bytes)
1645 {
1646         u64 end;
1647         u64 search_start, search_bytes;
1648         int ret;
1649
1650 again:
1651         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1652
1653         /*
1654          * We need to search for bits in this bitmap.  We could only cover some
1655          * of the extent in this bitmap thanks to how we add space, so we need
1656          * to search for as much as it as we can and clear that amount, and then
1657          * go searching for the next bit.
1658          */
1659         search_start = *offset;
1660         search_bytes = ctl->unit;
1661         search_bytes = min(search_bytes, end - search_start + 1);
1662         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes);
1663         if (ret < 0 || search_start != *offset)
1664                 return -EINVAL;
1665
1666         /* We may have found more bits than what we need */
1667         search_bytes = min(search_bytes, *bytes);
1668
1669         /* Cannot clear past the end of the bitmap */
1670         search_bytes = min(search_bytes, end - search_start + 1);
1671
1672         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1673         *offset += search_bytes;
1674         *bytes -= search_bytes;
1675
1676         if (*bytes) {
1677                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1678                 if (!bitmap_info->bytes)
1679                         free_bitmap(ctl, bitmap_info);
1680
1681                 /*
1682                  * no entry after this bitmap, but we still have bytes to
1683                  * remove, so something has gone wrong.
1684                  */
1685                 if (!next)
1686                         return -EINVAL;
1687
1688                 bitmap_info = rb_entry(next, struct btrfs_free_space,
1689                                        offset_index);
1690
1691                 /*
1692                  * if the next entry isn't a bitmap we need to return to let the
1693                  * extent stuff do its work.
1694                  */
1695                 if (!bitmap_info->bitmap)
1696                         return -EAGAIN;
1697
1698                 /*
1699                  * Ok the next item is a bitmap, but it may not actually hold
1700                  * the information for the rest of this free space stuff, so
1701                  * look for it, and if we don't find it return so we can try
1702                  * everything over again.
1703                  */
1704                 search_start = *offset;
1705                 search_bytes = ctl->unit;
1706                 ret = search_bitmap(ctl, bitmap_info, &search_start,
1707                                     &search_bytes);
1708                 if (ret < 0 || search_start != *offset)
1709                         return -EAGAIN;
1710
1711                 goto again;
1712         } else if (!bitmap_info->bytes)
1713                 free_bitmap(ctl, bitmap_info);
1714
1715         return 0;
1716 }
1717
1718 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1719                                struct btrfs_free_space *info, u64 offset,
1720                                u64 bytes)
1721 {
1722         u64 bytes_to_set = 0;
1723         u64 end;
1724
1725         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1726
1727         bytes_to_set = min(end - offset, bytes);
1728
1729         bitmap_set_bits(ctl, info, offset, bytes_to_set);
1730
1731         return bytes_to_set;
1732
1733 }
1734
1735 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1736                       struct btrfs_free_space *info)
1737 {
1738         struct btrfs_block_group_cache *block_group = ctl->private;
1739
1740         /*
1741          * If we are below the extents threshold then we can add this as an
1742          * extent, and don't have to deal with the bitmap
1743          */
1744         if (ctl->free_extents < ctl->extents_thresh) {
1745                 /*
1746                  * If this block group has some small extents we don't want to
1747                  * use up all of our free slots in the cache with them, we want
1748                  * to reserve them to larger extents, however if we have plent
1749                  * of cache left then go ahead an dadd them, no sense in adding
1750                  * the overhead of a bitmap if we don't have to.
1751                  */
1752                 if (info->bytes <= block_group->sectorsize * 4) {
1753                         if (ctl->free_extents * 2 <= ctl->extents_thresh)
1754                                 return false;
1755                 } else {
1756                         return false;
1757                 }
1758         }
1759
1760         /*
1761          * The original block groups from mkfs can be really small, like 8
1762          * megabytes, so don't bother with a bitmap for those entries.  However
1763          * some block groups can be smaller than what a bitmap would cover but
1764          * are still large enough that they could overflow the 32k memory limit,
1765          * so allow those block groups to still be allowed to have a bitmap
1766          * entry.
1767          */
1768         if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
1769                 return false;
1770
1771         return true;
1772 }
1773
1774 static struct btrfs_free_space_op free_space_op = {
1775         .recalc_thresholds      = recalculate_thresholds,
1776         .use_bitmap             = use_bitmap,
1777 };
1778
1779 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
1780                               struct btrfs_free_space *info)
1781 {
1782         struct btrfs_free_space *bitmap_info;
1783         struct btrfs_block_group_cache *block_group = NULL;
1784         int added = 0;
1785         u64 bytes, offset, bytes_added;
1786         int ret;
1787
1788         bytes = info->bytes;
1789         offset = info->offset;
1790
1791         if (!ctl->op->use_bitmap(ctl, info))
1792                 return 0;
1793
1794         if (ctl->op == &free_space_op)
1795                 block_group = ctl->private;
1796 again:
1797         /*
1798          * Since we link bitmaps right into the cluster we need to see if we
1799          * have a cluster here, and if so and it has our bitmap we need to add
1800          * the free space to that bitmap.
1801          */
1802         if (block_group && !list_empty(&block_group->cluster_list)) {
1803                 struct btrfs_free_cluster *cluster;
1804                 struct rb_node *node;
1805                 struct btrfs_free_space *entry;
1806
1807                 cluster = list_entry(block_group->cluster_list.next,
1808                                      struct btrfs_free_cluster,
1809                                      block_group_list);
1810                 spin_lock(&cluster->lock);
1811                 node = rb_first(&cluster->root);
1812                 if (!node) {
1813                         spin_unlock(&cluster->lock);
1814                         goto no_cluster_bitmap;
1815                 }
1816
1817                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1818                 if (!entry->bitmap) {
1819                         spin_unlock(&cluster->lock);
1820                         goto no_cluster_bitmap;
1821                 }
1822
1823                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
1824                         bytes_added = add_bytes_to_bitmap(ctl, entry,
1825                                                           offset, bytes);
1826                         bytes -= bytes_added;
1827                         offset += bytes_added;
1828                 }
1829                 spin_unlock(&cluster->lock);
1830                 if (!bytes) {
1831                         ret = 1;
1832                         goto out;
1833                 }
1834         }
1835
1836 no_cluster_bitmap:
1837         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
1838                                          1, 0);
1839         if (!bitmap_info) {
1840                 ASSERT(added == 0);
1841                 goto new_bitmap;
1842         }
1843
1844         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
1845         bytes -= bytes_added;
1846         offset += bytes_added;
1847         added = 0;
1848
1849         if (!bytes) {
1850                 ret = 1;
1851                 goto out;
1852         } else
1853                 goto again;
1854
1855 new_bitmap:
1856         if (info && info->bitmap) {
1857                 add_new_bitmap(ctl, info, offset);
1858                 added = 1;
1859                 info = NULL;
1860                 goto again;
1861         } else {
1862                 spin_unlock(&ctl->tree_lock);
1863
1864                 /* no pre-allocated info, allocate a new one */
1865                 if (!info) {
1866                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
1867                                                  GFP_NOFS);
1868                         if (!info) {
1869                                 spin_lock(&ctl->tree_lock);
1870                                 ret = -ENOMEM;
1871                                 goto out;
1872                         }
1873                 }
1874
1875                 /* allocate the bitmap */
1876                 info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
1877                 spin_lock(&ctl->tree_lock);
1878                 if (!info->bitmap) {
1879                         ret = -ENOMEM;
1880                         goto out;
1881                 }
1882                 goto again;
1883         }
1884
1885 out:
1886         if (info) {
1887                 if (info->bitmap)
1888                         kfree(info->bitmap);
1889                 kmem_cache_free(btrfs_free_space_cachep, info);
1890         }
1891
1892         return ret;
1893 }
1894
1895 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
1896                           struct btrfs_free_space *info, bool update_stat)
1897 {
1898         struct btrfs_free_space *left_info;
1899         struct btrfs_free_space *right_info;
1900         bool merged = false;
1901         u64 offset = info->offset;
1902         u64 bytes = info->bytes;
1903
1904         /*
1905          * first we want to see if there is free space adjacent to the range we
1906          * are adding, if there is remove that struct and add a new one to
1907          * cover the entire range
1908          */
1909         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
1910         if (right_info && rb_prev(&right_info->offset_index))
1911                 left_info = rb_entry(rb_prev(&right_info->offset_index),
1912                                      struct btrfs_free_space, offset_index);
1913         else
1914                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
1915
1916         if (right_info && !right_info->bitmap) {
1917                 if (update_stat)
1918                         unlink_free_space(ctl, right_info);
1919                 else
1920                         __unlink_free_space(ctl, right_info);
1921                 info->bytes += right_info->bytes;
1922                 kmem_cache_free(btrfs_free_space_cachep, right_info);
1923                 merged = true;
1924         }
1925
1926         if (left_info && !left_info->bitmap &&
1927             left_info->offset + left_info->bytes == offset) {
1928                 if (update_stat)
1929                         unlink_free_space(ctl, left_info);
1930                 else
1931                         __unlink_free_space(ctl, left_info);
1932                 info->offset = left_info->offset;
1933                 info->bytes += left_info->bytes;
1934                 kmem_cache_free(btrfs_free_space_cachep, left_info);
1935                 merged = true;
1936         }
1937
1938         return merged;
1939 }
1940
1941 int __btrfs_add_free_space(struct btrfs_free_space_ctl *ctl,
1942                            u64 offset, u64 bytes)
1943 {
1944         struct btrfs_free_space *info;
1945         int ret = 0;
1946
1947         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
1948         if (!info)
1949                 return -ENOMEM;
1950
1951         info->offset = offset;
1952         info->bytes = bytes;
1953
1954         spin_lock(&ctl->tree_lock);
1955
1956         if (try_merge_free_space(ctl, info, true))
1957                 goto link;
1958
1959         /*
1960          * There was no extent directly to the left or right of this new
1961          * extent then we know we're going to have to allocate a new extent, so
1962          * before we do that see if we need to drop this into a bitmap
1963          */
1964         ret = insert_into_bitmap(ctl, info);
1965         if (ret < 0) {
1966                 goto out;
1967         } else if (ret) {
1968                 ret = 0;
1969                 goto out;
1970         }
1971 link:
1972         ret = link_free_space(ctl, info);
1973         if (ret)
1974                 kmem_cache_free(btrfs_free_space_cachep, info);
1975 out:
1976         spin_unlock(&ctl->tree_lock);
1977
1978         if (ret) {
1979                 printk(KERN_CRIT "BTRFS: unable to add free space :%d\n", ret);
1980                 ASSERT(ret != -EEXIST);
1981         }
1982
1983         return ret;
1984 }
1985
1986 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
1987                             u64 offset, u64 bytes)
1988 {
1989         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1990         struct btrfs_free_space *info;
1991         int ret;
1992         bool re_search = false;
1993
1994         spin_lock(&ctl->tree_lock);
1995
1996 again:
1997         ret = 0;
1998         if (!bytes)
1999                 goto out_lock;
2000
2001         info = tree_search_offset(ctl, offset, 0, 0);
2002         if (!info) {
2003                 /*
2004                  * oops didn't find an extent that matched the space we wanted
2005                  * to remove, look for a bitmap instead
2006                  */
2007                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2008                                           1, 0);
2009                 if (!info) {
2010                         /*
2011                          * If we found a partial bit of our free space in a
2012                          * bitmap but then couldn't find the other part this may
2013                          * be a problem, so WARN about it.
2014                          */
2015                         WARN_ON(re_search);
2016                         goto out_lock;
2017                 }
2018         }
2019
2020         re_search = false;
2021         if (!info->bitmap) {
2022                 unlink_free_space(ctl, info);
2023                 if (offset == info->offset) {
2024                         u64 to_free = min(bytes, info->bytes);
2025
2026                         info->bytes -= to_free;
2027                         info->offset += to_free;
2028                         if (info->bytes) {
2029                                 ret = link_free_space(ctl, info);
2030                                 WARN_ON(ret);
2031                         } else {
2032                                 kmem_cache_free(btrfs_free_space_cachep, info);
2033                         }
2034
2035                         offset += to_free;
2036                         bytes -= to_free;
2037                         goto again;
2038                 } else {
2039                         u64 old_end = info->bytes + info->offset;
2040
2041                         info->bytes = offset - info->offset;
2042                         ret = link_free_space(ctl, info);
2043                         WARN_ON(ret);
2044                         if (ret)
2045                                 goto out_lock;
2046
2047                         /* Not enough bytes in this entry to satisfy us */
2048                         if (old_end < offset + bytes) {
2049                                 bytes -= old_end - offset;
2050                                 offset = old_end;
2051                                 goto again;
2052                         } else if (old_end == offset + bytes) {
2053                                 /* all done */
2054                                 goto out_lock;
2055                         }
2056                         spin_unlock(&ctl->tree_lock);
2057
2058                         ret = btrfs_add_free_space(block_group, offset + bytes,
2059                                                    old_end - (offset + bytes));
2060                         WARN_ON(ret);
2061                         goto out;
2062                 }
2063         }
2064
2065         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2066         if (ret == -EAGAIN) {
2067                 re_search = true;
2068                 goto again;
2069         }
2070 out_lock:
2071         spin_unlock(&ctl->tree_lock);
2072 out:
2073         return ret;
2074 }
2075
2076 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2077                            u64 bytes)
2078 {
2079         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2080         struct btrfs_free_space *info;
2081         struct rb_node *n;
2082         int count = 0;
2083
2084         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2085                 info = rb_entry(n, struct btrfs_free_space, offset_index);
2086                 if (info->bytes >= bytes && !block_group->ro)
2087                         count++;
2088                 btrfs_crit(block_group->fs_info,
2089                            "entry offset %llu, bytes %llu, bitmap %s",
2090                            info->offset, info->bytes,
2091                        (info->bitmap) ? "yes" : "no");
2092         }
2093         btrfs_info(block_group->fs_info, "block group has cluster?: %s",
2094                list_empty(&block_group->cluster_list) ? "no" : "yes");
2095         btrfs_info(block_group->fs_info,
2096                    "%d blocks of free space at or bigger than bytes is", count);
2097 }
2098
2099 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2100 {
2101         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2102
2103         spin_lock_init(&ctl->tree_lock);
2104         ctl->unit = block_group->sectorsize;
2105         ctl->start = block_group->key.objectid;
2106         ctl->private = block_group;
2107         ctl->op = &free_space_op;
2108
2109         /*
2110          * we only want to have 32k of ram per block group for keeping
2111          * track of free space, and if we pass 1/2 of that we want to
2112          * start converting things over to using bitmaps
2113          */
2114         ctl->extents_thresh = ((1024 * 32) / 2) /
2115                                 sizeof(struct btrfs_free_space);
2116 }
2117
2118 /*
2119  * for a given cluster, put all of its extents back into the free
2120  * space cache.  If the block group passed doesn't match the block group
2121  * pointed to by the cluster, someone else raced in and freed the
2122  * cluster already.  In that case, we just return without changing anything
2123  */
2124 static int
2125 __btrfs_return_cluster_to_free_space(
2126                              struct btrfs_block_group_cache *block_group,
2127                              struct btrfs_free_cluster *cluster)
2128 {
2129         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2130         struct btrfs_free_space *entry;
2131         struct rb_node *node;
2132
2133         spin_lock(&cluster->lock);
2134         if (cluster->block_group != block_group)
2135                 goto out;
2136
2137         cluster->block_group = NULL;
2138         cluster->window_start = 0;
2139         list_del_init(&cluster->block_group_list);
2140
2141         node = rb_first(&cluster->root);
2142         while (node) {
2143                 bool bitmap;
2144
2145                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2146                 node = rb_next(&entry->offset_index);
2147                 rb_erase(&entry->offset_index, &cluster->root);
2148
2149                 bitmap = (entry->bitmap != NULL);
2150                 if (!bitmap)
2151                         try_merge_free_space(ctl, entry, false);
2152                 tree_insert_offset(&ctl->free_space_offset,
2153                                    entry->offset, &entry->offset_index, bitmap);
2154         }
2155         cluster->root = RB_ROOT;
2156
2157 out:
2158         spin_unlock(&cluster->lock);
2159         btrfs_put_block_group(block_group);
2160         return 0;
2161 }
2162
2163 static void __btrfs_remove_free_space_cache_locked(
2164                                 struct btrfs_free_space_ctl *ctl)
2165 {
2166         struct btrfs_free_space *info;
2167         struct rb_node *node;
2168
2169         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2170                 info = rb_entry(node, struct btrfs_free_space, offset_index);
2171                 if (!info->bitmap) {
2172                         unlink_free_space(ctl, info);
2173                         kmem_cache_free(btrfs_free_space_cachep, info);
2174                 } else {
2175                         free_bitmap(ctl, info);
2176                 }
2177                 if (need_resched()) {
2178                         spin_unlock(&ctl->tree_lock);
2179                         cond_resched();
2180                         spin_lock(&ctl->tree_lock);
2181                 }
2182         }
2183 }
2184
2185 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2186 {
2187         spin_lock(&ctl->tree_lock);
2188         __btrfs_remove_free_space_cache_locked(ctl);
2189         spin_unlock(&ctl->tree_lock);
2190 }
2191
2192 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2193 {
2194         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2195         struct btrfs_free_cluster *cluster;
2196         struct list_head *head;
2197
2198         spin_lock(&ctl->tree_lock);
2199         while ((head = block_group->cluster_list.next) !=
2200                &block_group->cluster_list) {
2201                 cluster = list_entry(head, struct btrfs_free_cluster,
2202                                      block_group_list);
2203
2204                 WARN_ON(cluster->block_group != block_group);
2205                 __btrfs_return_cluster_to_free_space(block_group, cluster);
2206                 if (need_resched()) {
2207                         spin_unlock(&ctl->tree_lock);
2208                         cond_resched();
2209                         spin_lock(&ctl->tree_lock);
2210                 }
2211         }
2212         __btrfs_remove_free_space_cache_locked(ctl);
2213         spin_unlock(&ctl->tree_lock);
2214
2215 }
2216
2217 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2218                                u64 offset, u64 bytes, u64 empty_size,
2219                                u64 *max_extent_size)
2220 {
2221         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2222         struct btrfs_free_space *entry = NULL;
2223         u64 bytes_search = bytes + empty_size;
2224         u64 ret = 0;
2225         u64 align_gap = 0;
2226         u64 align_gap_len = 0;
2227
2228         spin_lock(&ctl->tree_lock);
2229         entry = find_free_space(ctl, &offset, &bytes_search,
2230                                 block_group->full_stripe_len, max_extent_size);
2231         if (!entry)
2232                 goto out;
2233
2234         ret = offset;
2235         if (entry->bitmap) {
2236                 bitmap_clear_bits(ctl, entry, offset, bytes);
2237                 if (!entry->bytes)
2238                         free_bitmap(ctl, entry);
2239         } else {
2240                 unlink_free_space(ctl, entry);
2241                 align_gap_len = offset - entry->offset;
2242                 align_gap = entry->offset;
2243
2244                 entry->offset = offset + bytes;
2245                 WARN_ON(entry->bytes < bytes + align_gap_len);
2246
2247                 entry->bytes -= bytes + align_gap_len;
2248                 if (!entry->bytes)
2249                         kmem_cache_free(btrfs_free_space_cachep, entry);
2250                 else
2251                         link_free_space(ctl, entry);
2252         }
2253 out:
2254         spin_unlock(&ctl->tree_lock);
2255
2256         if (align_gap_len)
2257                 __btrfs_add_free_space(ctl, align_gap, align_gap_len);
2258         return ret;
2259 }
2260
2261 /*
2262  * given a cluster, put all of its extents back into the free space
2263  * cache.  If a block group is passed, this function will only free
2264  * a cluster that belongs to the passed block group.
2265  *
2266  * Otherwise, it'll get a reference on the block group pointed to by the
2267  * cluster and remove the cluster from it.
2268  */
2269 int btrfs_return_cluster_to_free_space(
2270                                struct btrfs_block_group_cache *block_group,
2271                                struct btrfs_free_cluster *cluster)
2272 {
2273         struct btrfs_free_space_ctl *ctl;
2274         int ret;
2275
2276         /* first, get a safe pointer to the block group */
2277         spin_lock(&cluster->lock);
2278         if (!block_group) {
2279                 block_group = cluster->block_group;
2280                 if (!block_group) {
2281                         spin_unlock(&cluster->lock);
2282                         return 0;
2283                 }
2284         } else if (cluster->block_group != block_group) {
2285                 /* someone else has already freed it don't redo their work */
2286                 spin_unlock(&cluster->lock);
2287                 return 0;
2288         }
2289         atomic_inc(&block_group->count);
2290         spin_unlock(&cluster->lock);
2291
2292         ctl = block_group->free_space_ctl;
2293
2294         /* now return any extents the cluster had on it */
2295         spin_lock(&ctl->tree_lock);
2296         ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2297         spin_unlock(&ctl->tree_lock);
2298
2299         /* finally drop our ref */
2300         btrfs_put_block_group(block_group);
2301         return ret;
2302 }
2303
2304 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2305                                    struct btrfs_free_cluster *cluster,
2306                                    struct btrfs_free_space *entry,
2307                                    u64 bytes, u64 min_start,
2308                                    u64 *max_extent_size)
2309 {
2310         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2311         int err;
2312         u64 search_start = cluster->window_start;
2313         u64 search_bytes = bytes;
2314         u64 ret = 0;
2315
2316         search_start = min_start;
2317         search_bytes = bytes;
2318
2319         err = search_bitmap(ctl, entry, &search_start, &search_bytes);
2320         if (err) {
2321                 if (search_bytes > *max_extent_size)
2322                         *max_extent_size = search_bytes;
2323                 return 0;
2324         }
2325
2326         ret = search_start;
2327         __bitmap_clear_bits(ctl, entry, ret, bytes);
2328
2329         return ret;
2330 }
2331
2332 /*
2333  * given a cluster, try to allocate 'bytes' from it, returns 0
2334  * if it couldn't find anything suitably large, or a logical disk offset
2335  * if things worked out
2336  */
2337 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2338                              struct btrfs_free_cluster *cluster, u64 bytes,
2339                              u64 min_start, u64 *max_extent_size)
2340 {
2341         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2342         struct btrfs_free_space *entry = NULL;
2343         struct rb_node *node;
2344         u64 ret = 0;
2345
2346         spin_lock(&cluster->lock);
2347         if (bytes > cluster->max_size)
2348                 goto out;
2349
2350         if (cluster->block_group != block_group)
2351                 goto out;
2352
2353         node = rb_first(&cluster->root);
2354         if (!node)
2355                 goto out;
2356
2357         entry = rb_entry(node, struct btrfs_free_space, offset_index);
2358         while (1) {
2359                 if (entry->bytes < bytes && entry->bytes > *max_extent_size)
2360                         *max_extent_size = entry->bytes;
2361
2362                 if (entry->bytes < bytes ||
2363                     (!entry->bitmap && entry->offset < min_start)) {
2364                         node = rb_next(&entry->offset_index);
2365                         if (!node)
2366                                 break;
2367                         entry = rb_entry(node, struct btrfs_free_space,
2368                                          offset_index);
2369                         continue;
2370                 }
2371
2372                 if (entry->bitmap) {
2373                         ret = btrfs_alloc_from_bitmap(block_group,
2374                                                       cluster, entry, bytes,
2375                                                       cluster->window_start,
2376                                                       max_extent_size);
2377                         if (ret == 0) {
2378                                 node = rb_next(&entry->offset_index);
2379                                 if (!node)
2380                                         break;
2381                                 entry = rb_entry(node, struct btrfs_free_space,
2382                                                  offset_index);
2383                                 continue;
2384                         }
2385                         cluster->window_start += bytes;
2386                 } else {
2387                         ret = entry->offset;
2388
2389                         entry->offset += bytes;
2390                         entry->bytes -= bytes;
2391                 }
2392
2393                 if (entry->bytes == 0)
2394                         rb_erase(&entry->offset_index, &cluster->root);
2395                 break;
2396         }
2397 out:
2398         spin_unlock(&cluster->lock);
2399
2400         if (!ret)
2401                 return 0;
2402
2403         spin_lock(&ctl->tree_lock);
2404
2405         ctl->free_space -= bytes;
2406         if (entry->bytes == 0) {
2407                 ctl->free_extents--;
2408                 if (entry->bitmap) {
2409                         kfree(entry->bitmap);
2410                         ctl->total_bitmaps--;
2411                         ctl->op->recalc_thresholds(ctl);
2412                 }
2413                 kmem_cache_free(btrfs_free_space_cachep, entry);
2414         }
2415
2416         spin_unlock(&ctl->tree_lock);
2417
2418         return ret;
2419 }
2420
2421 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2422                                 struct btrfs_free_space *entry,
2423                                 struct btrfs_free_cluster *cluster,
2424                                 u64 offset, u64 bytes,
2425                                 u64 cont1_bytes, u64 min_bytes)
2426 {
2427         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2428         unsigned long next_zero;
2429         unsigned long i;
2430         unsigned long want_bits;
2431         unsigned long min_bits;
2432         unsigned long found_bits;
2433         unsigned long start = 0;
2434         unsigned long total_found = 0;
2435         int ret;
2436
2437         i = offset_to_bit(entry->offset, ctl->unit,
2438                           max_t(u64, offset, entry->offset));
2439         want_bits = bytes_to_bits(bytes, ctl->unit);
2440         min_bits = bytes_to_bits(min_bytes, ctl->unit);
2441
2442 again:
2443         found_bits = 0;
2444         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2445                 next_zero = find_next_zero_bit(entry->bitmap,
2446                                                BITS_PER_BITMAP, i);
2447                 if (next_zero - i >= min_bits) {
2448                         found_bits = next_zero - i;
2449                         break;
2450                 }
2451                 i = next_zero;
2452         }
2453
2454         if (!found_bits)
2455                 return -ENOSPC;
2456
2457         if (!total_found) {
2458                 start = i;
2459                 cluster->max_size = 0;
2460         }
2461
2462         total_found += found_bits;
2463
2464         if (cluster->max_size < found_bits * ctl->unit)
2465                 cluster->max_size = found_bits * ctl->unit;
2466
2467         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2468                 i = next_zero + 1;
2469                 goto again;
2470         }
2471
2472         cluster->window_start = start * ctl->unit + entry->offset;
2473         rb_erase(&entry->offset_index, &ctl->free_space_offset);
2474         ret = tree_insert_offset(&cluster->root, entry->offset,
2475                                  &entry->offset_index, 1);
2476         ASSERT(!ret); /* -EEXIST; Logic error */
2477
2478         trace_btrfs_setup_cluster(block_group, cluster,
2479                                   total_found * ctl->unit, 1);
2480         return 0;
2481 }
2482
2483 /*
2484  * This searches the block group for just extents to fill the cluster with.
2485  * Try to find a cluster with at least bytes total bytes, at least one
2486  * extent of cont1_bytes, and other clusters of at least min_bytes.
2487  */
2488 static noinline int
2489 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2490                         struct btrfs_free_cluster *cluster,
2491                         struct list_head *bitmaps, u64 offset, u64 bytes,
2492                         u64 cont1_bytes, u64 min_bytes)
2493 {
2494         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2495         struct btrfs_free_space *first = NULL;
2496         struct btrfs_free_space *entry = NULL;
2497         struct btrfs_free_space *last;
2498         struct rb_node *node;
2499         u64 window_free;
2500         u64 max_extent;
2501         u64 total_size = 0;
2502
2503         entry = tree_search_offset(ctl, offset, 0, 1);
2504         if (!entry)
2505                 return -ENOSPC;
2506
2507         /*
2508          * We don't want bitmaps, so just move along until we find a normal
2509          * extent entry.
2510          */
2511         while (entry->bitmap || entry->bytes < min_bytes) {
2512                 if (entry->bitmap && list_empty(&entry->list))
2513                         list_add_tail(&entry->list, bitmaps);
2514                 node = rb_next(&entry->offset_index);
2515                 if (!node)
2516                         return -ENOSPC;
2517                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2518         }
2519
2520         window_free = entry->bytes;
2521         max_extent = entry->bytes;
2522         first = entry;
2523         last = entry;
2524
2525         for (node = rb_next(&entry->offset_index); node;
2526              node = rb_next(&entry->offset_index)) {
2527                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2528
2529                 if (entry->bitmap) {
2530                         if (list_empty(&entry->list))
2531                                 list_add_tail(&entry->list, bitmaps);
2532                         continue;
2533                 }
2534
2535                 if (entry->bytes < min_bytes)
2536                         continue;
2537
2538                 last = entry;
2539                 window_free += entry->bytes;
2540                 if (entry->bytes > max_extent)
2541                         max_extent = entry->bytes;
2542         }
2543
2544         if (window_free < bytes || max_extent < cont1_bytes)
2545                 return -ENOSPC;
2546
2547         cluster->window_start = first->offset;
2548
2549         node = &first->offset_index;
2550
2551         /*
2552          * now we've found our entries, pull them out of the free space
2553          * cache and put them into the cluster rbtree
2554          */
2555         do {
2556                 int ret;
2557
2558                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2559                 node = rb_next(&entry->offset_index);
2560                 if (entry->bitmap || entry->bytes < min_bytes)
2561                         continue;
2562
2563                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2564                 ret = tree_insert_offset(&cluster->root, entry->offset,
2565                                          &entry->offset_index, 0);
2566                 total_size += entry->bytes;
2567                 ASSERT(!ret); /* -EEXIST; Logic error */
2568         } while (node && entry != last);
2569
2570         cluster->max_size = max_extent;
2571         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2572         return 0;
2573 }
2574
2575 /*
2576  * This specifically looks for bitmaps that may work in the cluster, we assume
2577  * that we have already failed to find extents that will work.
2578  */
2579 static noinline int
2580 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2581                      struct btrfs_free_cluster *cluster,
2582                      struct list_head *bitmaps, u64 offset, u64 bytes,
2583                      u64 cont1_bytes, u64 min_bytes)
2584 {
2585         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2586         struct btrfs_free_space *entry;
2587         int ret = -ENOSPC;
2588         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2589
2590         if (ctl->total_bitmaps == 0)
2591                 return -ENOSPC;
2592
2593         /*
2594          * The bitmap that covers offset won't be in the list unless offset
2595          * is just its start offset.
2596          */
2597         entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
2598         if (entry->offset != bitmap_offset) {
2599                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
2600                 if (entry && list_empty(&entry->list))
2601                         list_add(&entry->list, bitmaps);
2602         }
2603
2604         list_for_each_entry(entry, bitmaps, list) {
2605                 if (entry->bytes < bytes)
2606                         continue;
2607                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
2608                                            bytes, cont1_bytes, min_bytes);
2609                 if (!ret)
2610                         return 0;
2611         }
2612
2613         /*
2614          * The bitmaps list has all the bitmaps that record free space
2615          * starting after offset, so no more search is required.
2616          */
2617         return -ENOSPC;
2618 }
2619
2620 /*
2621  * here we try to find a cluster of blocks in a block group.  The goal
2622  * is to find at least bytes+empty_size.
2623  * We might not find them all in one contiguous area.
2624  *
2625  * returns zero and sets up cluster if things worked out, otherwise
2626  * it returns -enospc
2627  */
2628 int btrfs_find_space_cluster(struct btrfs_root *root,
2629                              struct btrfs_block_group_cache *block_group,
2630                              struct btrfs_free_cluster *cluster,
2631                              u64 offset, u64 bytes, u64 empty_size)
2632 {
2633         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2634         struct btrfs_free_space *entry, *tmp;
2635         LIST_HEAD(bitmaps);
2636         u64 min_bytes;
2637         u64 cont1_bytes;
2638         int ret;
2639
2640         /*
2641          * Choose the minimum extent size we'll require for this
2642          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
2643          * For metadata, allow allocates with smaller extents.  For
2644          * data, keep it dense.
2645          */
2646         if (btrfs_test_opt(root, SSD_SPREAD)) {
2647                 cont1_bytes = min_bytes = bytes + empty_size;
2648         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
2649                 cont1_bytes = bytes;
2650                 min_bytes = block_group->sectorsize;
2651         } else {
2652                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
2653                 min_bytes = block_group->sectorsize;
2654         }
2655
2656         spin_lock(&ctl->tree_lock);
2657
2658         /*
2659          * If we know we don't have enough space to make a cluster don't even
2660          * bother doing all the work to try and find one.
2661          */
2662         if (ctl->free_space < bytes) {
2663                 spin_unlock(&ctl->tree_lock);
2664                 return -ENOSPC;
2665         }
2666
2667         spin_lock(&cluster->lock);
2668
2669         /* someone already found a cluster, hooray */
2670         if (cluster->block_group) {
2671                 ret = 0;
2672                 goto out;
2673         }
2674
2675         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
2676                                  min_bytes);
2677
2678         INIT_LIST_HEAD(&bitmaps);
2679         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
2680                                       bytes + empty_size,
2681                                       cont1_bytes, min_bytes);
2682         if (ret)
2683                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
2684                                            offset, bytes + empty_size,
2685                                            cont1_bytes, min_bytes);
2686
2687         /* Clear our temporary list */
2688         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
2689                 list_del_init(&entry->list);
2690
2691         if (!ret) {
2692                 atomic_inc(&block_group->count);
2693                 list_add_tail(&cluster->block_group_list,
2694                               &block_group->cluster_list);
2695                 cluster->block_group = block_group;
2696         } else {
2697                 trace_btrfs_failed_cluster_setup(block_group);
2698         }
2699 out:
2700         spin_unlock(&cluster->lock);
2701         spin_unlock(&ctl->tree_lock);
2702
2703         return ret;
2704 }
2705
2706 /*
2707  * simple code to zero out a cluster
2708  */
2709 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
2710 {
2711         spin_lock_init(&cluster->lock);
2712         spin_lock_init(&cluster->refill_lock);
2713         cluster->root = RB_ROOT;
2714         cluster->max_size = 0;
2715         INIT_LIST_HEAD(&cluster->block_group_list);
2716         cluster->block_group = NULL;
2717 }
2718
2719 static int do_trimming(struct btrfs_block_group_cache *block_group,
2720                        u64 *total_trimmed, u64 start, u64 bytes,
2721                        u64 reserved_start, u64 reserved_bytes)
2722 {
2723         struct btrfs_space_info *space_info = block_group->space_info;
2724         struct btrfs_fs_info *fs_info = block_group->fs_info;
2725         int ret;
2726         int update = 0;
2727         u64 trimmed = 0;
2728
2729         spin_lock(&space_info->lock);
2730         spin_lock(&block_group->lock);
2731         if (!block_group->ro) {
2732                 block_group->reserved += reserved_bytes;
2733                 space_info->bytes_reserved += reserved_bytes;
2734                 update = 1;
2735         }
2736         spin_unlock(&block_group->lock);
2737         spin_unlock(&space_info->lock);
2738
2739         ret = btrfs_error_discard_extent(fs_info->extent_root,
2740                                          start, bytes, &trimmed);
2741         if (!ret)
2742                 *total_trimmed += trimmed;
2743
2744         btrfs_add_free_space(block_group, reserved_start, reserved_bytes);
2745
2746         if (update) {
2747                 spin_lock(&space_info->lock);
2748                 spin_lock(&block_group->lock);
2749                 if (block_group->ro)
2750                         space_info->bytes_readonly += reserved_bytes;
2751                 block_group->reserved -= reserved_bytes;
2752                 space_info->bytes_reserved -= reserved_bytes;
2753                 spin_unlock(&space_info->lock);
2754                 spin_unlock(&block_group->lock);
2755         }
2756
2757         return ret;
2758 }
2759
2760 static int trim_no_bitmap(struct btrfs_block_group_cache *block_group,
2761                           u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2762 {
2763         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2764         struct btrfs_free_space *entry;
2765         struct rb_node *node;
2766         int ret = 0;
2767         u64 extent_start;
2768         u64 extent_bytes;
2769         u64 bytes;
2770
2771         while (start < end) {
2772                 spin_lock(&ctl->tree_lock);
2773
2774                 if (ctl->free_space < minlen) {
2775                         spin_unlock(&ctl->tree_lock);
2776                         break;
2777                 }
2778
2779                 entry = tree_search_offset(ctl, start, 0, 1);
2780                 if (!entry) {
2781                         spin_unlock(&ctl->tree_lock);
2782                         break;
2783                 }
2784
2785                 /* skip bitmaps */
2786                 while (entry->bitmap) {
2787                         node = rb_next(&entry->offset_index);
2788                         if (!node) {
2789                                 spin_unlock(&ctl->tree_lock);
2790                                 goto out;
2791                         }
2792                         entry = rb_entry(node, struct btrfs_free_space,
2793                                          offset_index);
2794                 }
2795
2796                 if (entry->offset >= end) {
2797                         spin_unlock(&ctl->tree_lock);
2798                         break;
2799                 }
2800
2801                 extent_start = entry->offset;
2802                 extent_bytes = entry->bytes;
2803                 start = max(start, extent_start);
2804                 bytes = min(extent_start + extent_bytes, end) - start;
2805                 if (bytes < minlen) {
2806                         spin_unlock(&ctl->tree_lock);
2807                         goto next;
2808                 }
2809
2810                 unlink_free_space(ctl, entry);
2811                 kmem_cache_free(btrfs_free_space_cachep, entry);
2812
2813                 spin_unlock(&ctl->tree_lock);
2814
2815                 ret = do_trimming(block_group, total_trimmed, start, bytes,
2816                                   extent_start, extent_bytes);
2817                 if (ret)
2818                         break;
2819 next:
2820                 start += bytes;
2821
2822                 if (fatal_signal_pending(current)) {
2823                         ret = -ERESTARTSYS;
2824                         break;
2825                 }
2826
2827                 cond_resched();
2828         }
2829 out:
2830         return ret;
2831 }
2832
2833 static int trim_bitmaps(struct btrfs_block_group_cache *block_group,
2834                         u64 *total_trimmed, u64 start, u64 end, u64 minlen)
2835 {
2836         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2837         struct btrfs_free_space *entry;
2838         int ret = 0;
2839         int ret2;
2840         u64 bytes;
2841         u64 offset = offset_to_bitmap(ctl, start);
2842
2843         while (offset < end) {
2844                 bool next_bitmap = false;
2845
2846                 spin_lock(&ctl->tree_lock);
2847
2848                 if (ctl->free_space < minlen) {
2849                         spin_unlock(&ctl->tree_lock);
2850                         break;
2851                 }
2852
2853                 entry = tree_search_offset(ctl, offset, 1, 0);
2854                 if (!entry) {
2855                         spin_unlock(&ctl->tree_lock);
2856                         next_bitmap = true;
2857                         goto next;
2858                 }
2859
2860                 bytes = minlen;
2861                 ret2 = search_bitmap(ctl, entry, &start, &bytes);
2862                 if (ret2 || start >= end) {
2863                         spin_unlock(&ctl->tree_lock);
2864                         next_bitmap = true;
2865                         goto next;
2866                 }
2867
2868                 bytes = min(bytes, end - start);
2869                 if (bytes < minlen) {
2870                         spin_unlock(&ctl->tree_lock);
2871                         goto next;
2872                 }
2873
2874                 bitmap_clear_bits(ctl, entry, start, bytes);
2875                 if (entry->bytes == 0)
2876                         free_bitmap(ctl, entry);
2877
2878                 spin_unlock(&ctl->tree_lock);
2879
2880                 ret = do_trimming(block_group, total_trimmed, start, bytes,
2881                                   start, bytes);
2882                 if (ret)
2883                         break;
2884 next:
2885                 if (next_bitmap) {
2886                         offset += BITS_PER_BITMAP * ctl->unit;
2887                 } else {
2888                         start += bytes;
2889                         if (start >= offset + BITS_PER_BITMAP * ctl->unit)
2890                                 offset += BITS_PER_BITMAP * ctl->unit;
2891                 }
2892
2893                 if (fatal_signal_pending(current)) {
2894                         ret = -ERESTARTSYS;
2895                         break;
2896                 }
2897
2898                 cond_resched();
2899         }
2900
2901         return ret;
2902 }
2903
2904 int btrfs_trim_block_group(struct btrfs_block_group_cache *block_group,
2905                            u64 *trimmed, u64 start, u64 end, u64 minlen)
2906 {
2907         int ret;
2908
2909         *trimmed = 0;
2910
2911         ret = trim_no_bitmap(block_group, trimmed, start, end, minlen);
2912         if (ret)
2913                 return ret;
2914
2915         ret = trim_bitmaps(block_group, trimmed, start, end, minlen);
2916
2917         return ret;
2918 }
2919
2920 /*
2921  * Find the left-most item in the cache tree, and then return the
2922  * smallest inode number in the item.
2923  *
2924  * Note: the returned inode number may not be the smallest one in
2925  * the tree, if the left-most item is a bitmap.
2926  */
2927 u64 btrfs_find_ino_for_alloc(struct btrfs_root *fs_root)
2928 {
2929         struct btrfs_free_space_ctl *ctl = fs_root->free_ino_ctl;
2930         struct btrfs_free_space *entry = NULL;
2931         u64 ino = 0;
2932
2933         spin_lock(&ctl->tree_lock);
2934
2935         if (RB_EMPTY_ROOT(&ctl->free_space_offset))
2936                 goto out;
2937
2938         entry = rb_entry(rb_first(&ctl->free_space_offset),
2939                          struct btrfs_free_space, offset_index);
2940
2941         if (!entry->bitmap) {
2942                 ino = entry->offset;
2943
2944                 unlink_free_space(ctl, entry);
2945                 entry->offset++;
2946                 entry->bytes--;
2947                 if (!entry->bytes)
2948                         kmem_cache_free(btrfs_free_space_cachep, entry);
2949                 else
2950                         link_free_space(ctl, entry);
2951         } else {
2952                 u64 offset = 0;
2953                 u64 count = 1;
2954                 int ret;
2955
2956                 ret = search_bitmap(ctl, entry, &offset, &count);
2957                 /* Logic error; Should be empty if it can't find anything */
2958                 ASSERT(!ret);
2959
2960                 ino = offset;
2961                 bitmap_clear_bits(ctl, entry, offset, 1);
2962                 if (entry->bytes == 0)
2963                         free_bitmap(ctl, entry);
2964         }
2965 out:
2966         spin_unlock(&ctl->tree_lock);
2967
2968         return ino;
2969 }
2970
2971 struct inode *lookup_free_ino_inode(struct btrfs_root *root,
2972                                     struct btrfs_path *path)
2973 {
2974         struct inode *inode = NULL;
2975
2976         spin_lock(&root->cache_lock);
2977         if (root->cache_inode)
2978                 inode = igrab(root->cache_inode);
2979         spin_unlock(&root->cache_lock);
2980         if (inode)
2981                 return inode;
2982
2983         inode = __lookup_free_space_inode(root, path, 0);
2984         if (IS_ERR(inode))
2985                 return inode;
2986
2987         spin_lock(&root->cache_lock);
2988         if (!btrfs_fs_closing(root->fs_info))
2989                 root->cache_inode = igrab(inode);
2990         spin_unlock(&root->cache_lock);
2991
2992         return inode;
2993 }
2994
2995 int create_free_ino_inode(struct btrfs_root *root,
2996                           struct btrfs_trans_handle *trans,
2997                           struct btrfs_path *path)
2998 {
2999         return __create_free_space_inode(root, trans, path,
3000                                          BTRFS_FREE_INO_OBJECTID, 0);
3001 }
3002
3003 int load_free_ino_cache(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3004 {
3005         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3006         struct btrfs_path *path;
3007         struct inode *inode;
3008         int ret = 0;
3009         u64 root_gen = btrfs_root_generation(&root->root_item);
3010
3011         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3012                 return 0;
3013
3014         /*
3015          * If we're unmounting then just return, since this does a search on the
3016          * normal root and not the commit root and we could deadlock.
3017          */
3018         if (btrfs_fs_closing(fs_info))
3019                 return 0;
3020
3021         path = btrfs_alloc_path();
3022         if (!path)
3023                 return 0;
3024
3025         inode = lookup_free_ino_inode(root, path);
3026         if (IS_ERR(inode))
3027                 goto out;
3028
3029         if (root_gen != BTRFS_I(inode)->generation)
3030                 goto out_put;
3031
3032         ret = __load_free_space_cache(root, inode, ctl, path, 0);
3033
3034         if (ret < 0)
3035                 btrfs_err(fs_info,
3036                         "failed to load free ino cache for root %llu",
3037                         root->root_key.objectid);
3038 out_put:
3039         iput(inode);
3040 out:
3041         btrfs_free_path(path);
3042         return ret;
3043 }
3044
3045 int btrfs_write_out_ino_cache(struct btrfs_root *root,
3046                               struct btrfs_trans_handle *trans,
3047                               struct btrfs_path *path,
3048                               struct inode *inode)
3049 {
3050         struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
3051         int ret;
3052
3053         if (!btrfs_test_opt(root, INODE_MAP_CACHE))
3054                 return 0;
3055
3056         ret = __btrfs_write_out_cache(root, inode, ctl, NULL, trans, path, 0);
3057         if (ret) {
3058                 btrfs_delalloc_release_metadata(inode, inode->i_size);
3059 #ifdef DEBUG
3060                 btrfs_err(root->fs_info,
3061                         "failed to write free ino cache for root %llu",
3062                         root->root_key.objectid);
3063 #endif
3064         }
3065
3066         return ret;
3067 }
3068
3069 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3070 /*
3071  * Use this if you need to make a bitmap or extent entry specifically, it
3072  * doesn't do any of the merging that add_free_space does, this acts a lot like
3073  * how the free space cache loading stuff works, so you can get really weird
3074  * configurations.
3075  */
3076 int test_add_free_space_entry(struct btrfs_block_group_cache *cache,
3077                               u64 offset, u64 bytes, bool bitmap)
3078 {
3079         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3080         struct btrfs_free_space *info = NULL, *bitmap_info;
3081         void *map = NULL;
3082         u64 bytes_added;
3083         int ret;
3084
3085 again:
3086         if (!info) {
3087                 info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
3088                 if (!info)
3089                         return -ENOMEM;
3090         }
3091
3092         if (!bitmap) {
3093                 spin_lock(&ctl->tree_lock);
3094                 info->offset = offset;
3095                 info->bytes = bytes;
3096                 ret = link_free_space(ctl, info);
3097                 spin_unlock(&ctl->tree_lock);
3098                 if (ret)
3099                         kmem_cache_free(btrfs_free_space_cachep, info);
3100                 return ret;
3101         }
3102
3103         if (!map) {
3104                 map = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
3105                 if (!map) {
3106                         kmem_cache_free(btrfs_free_space_cachep, info);
3107                         return -ENOMEM;
3108                 }
3109         }
3110
3111         spin_lock(&ctl->tree_lock);
3112         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3113                                          1, 0);
3114         if (!bitmap_info) {
3115                 info->bitmap = map;
3116                 map = NULL;
3117                 add_new_bitmap(ctl, info, offset);
3118                 bitmap_info = info;
3119         }
3120
3121         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
3122         bytes -= bytes_added;
3123         offset += bytes_added;
3124         spin_unlock(&ctl->tree_lock);
3125
3126         if (bytes)
3127                 goto again;
3128
3129         if (map)
3130                 kfree(map);
3131         return 0;
3132 }
3133
3134 /*
3135  * Checks to see if the given range is in the free space cache.  This is really
3136  * just used to check the absence of space, so if there is free space in the
3137  * range at all we will return 1.
3138  */
3139 int test_check_exists(struct btrfs_block_group_cache *cache,
3140                       u64 offset, u64 bytes)
3141 {
3142         struct btrfs_free_space_ctl *ctl = cache->free_space_ctl;
3143         struct btrfs_free_space *info;
3144         int ret = 0;
3145
3146         spin_lock(&ctl->tree_lock);
3147         info = tree_search_offset(ctl, offset, 0, 0);
3148         if (!info) {
3149                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
3150                                           1, 0);
3151                 if (!info)
3152                         goto out;
3153         }
3154
3155 have_info:
3156         if (info->bitmap) {
3157                 u64 bit_off, bit_bytes;
3158                 struct rb_node *n;
3159                 struct btrfs_free_space *tmp;
3160
3161                 bit_off = offset;
3162                 bit_bytes = ctl->unit;
3163                 ret = search_bitmap(ctl, info, &bit_off, &bit_bytes);
3164                 if (!ret) {
3165                         if (bit_off == offset) {
3166                                 ret = 1;
3167                                 goto out;
3168                         } else if (bit_off > offset &&
3169                                    offset + bytes > bit_off) {
3170                                 ret = 1;
3171                                 goto out;
3172                         }
3173                 }
3174
3175                 n = rb_prev(&info->offset_index);
3176                 while (n) {
3177                         tmp = rb_entry(n, struct btrfs_free_space,
3178                                        offset_index);
3179                         if (tmp->offset + tmp->bytes < offset)
3180                                 break;
3181                         if (offset + bytes < tmp->offset) {
3182                                 n = rb_prev(&info->offset_index);
3183                                 continue;
3184                         }
3185                         info = tmp;
3186                         goto have_info;
3187                 }
3188
3189                 n = rb_next(&info->offset_index);
3190                 while (n) {
3191                         tmp = rb_entry(n, struct btrfs_free_space,
3192                                        offset_index);
3193                         if (offset + bytes < tmp->offset)
3194                                 break;
3195                         if (tmp->offset + tmp->bytes < offset) {
3196                                 n = rb_next(&info->offset_index);
3197                                 continue;
3198                         }
3199                         info = tmp;
3200                         goto have_info;
3201                 }
3202
3203                 goto out;
3204         }
3205
3206         if (info->offset == offset) {
3207                 ret = 1;
3208                 goto out;
3209         }
3210
3211         if (offset > info->offset && offset < info->offset + info->bytes)
3212                 ret = 1;
3213 out:
3214         spin_unlock(&ctl->tree_lock);
3215         return ret;
3216 }
3217 #endif /* CONFIG_BTRFS_FS_RUN_SANITY_TESTS */