btrfs: switch order of unlocks of space_info and bg in do_trimming()
[sfrench/cifs-2.6.git] / fs / btrfs / free-space-cache.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2008 Red Hat.  All rights reserved.
4  */
5
6 #include <linux/pagemap.h>
7 #include <linux/sched.h>
8 #include <linux/sched/signal.h>
9 #include <linux/slab.h>
10 #include <linux/math64.h>
11 #include <linux/ratelimit.h>
12 #include <linux/error-injection.h>
13 #include <linux/sched/mm.h>
14 #include "ctree.h"
15 #include "free-space-cache.h"
16 #include "transaction.h"
17 #include "disk-io.h"
18 #include "extent_io.h"
19 #include "inode-map.h"
20 #include "volumes.h"
21
22 #define BITS_PER_BITMAP         (PAGE_SIZE * 8UL)
23 #define MAX_CACHE_BYTES_PER_GIG SZ_32K
24
25 struct btrfs_trim_range {
26         u64 start;
27         u64 bytes;
28         struct list_head list;
29 };
30
31 static int link_free_space(struct btrfs_free_space_ctl *ctl,
32                            struct btrfs_free_space *info);
33 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
34                               struct btrfs_free_space *info);
35 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
36                              struct btrfs_trans_handle *trans,
37                              struct btrfs_io_ctl *io_ctl,
38                              struct btrfs_path *path);
39
40 static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
41                                                struct btrfs_path *path,
42                                                u64 offset)
43 {
44         struct btrfs_fs_info *fs_info = root->fs_info;
45         struct btrfs_key key;
46         struct btrfs_key location;
47         struct btrfs_disk_key disk_key;
48         struct btrfs_free_space_header *header;
49         struct extent_buffer *leaf;
50         struct inode *inode = NULL;
51         unsigned nofs_flag;
52         int ret;
53
54         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
55         key.offset = offset;
56         key.type = 0;
57
58         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
59         if (ret < 0)
60                 return ERR_PTR(ret);
61         if (ret > 0) {
62                 btrfs_release_path(path);
63                 return ERR_PTR(-ENOENT);
64         }
65
66         leaf = path->nodes[0];
67         header = btrfs_item_ptr(leaf, path->slots[0],
68                                 struct btrfs_free_space_header);
69         btrfs_free_space_key(leaf, header, &disk_key);
70         btrfs_disk_key_to_cpu(&location, &disk_key);
71         btrfs_release_path(path);
72
73         /*
74          * We are often under a trans handle at this point, so we need to make
75          * sure NOFS is set to keep us from deadlocking.
76          */
77         nofs_flag = memalloc_nofs_save();
78         inode = btrfs_iget_path(fs_info->sb, &location, root, NULL, path);
79         btrfs_release_path(path);
80         memalloc_nofs_restore(nofs_flag);
81         if (IS_ERR(inode))
82                 return inode;
83
84         mapping_set_gfp_mask(inode->i_mapping,
85                         mapping_gfp_constraint(inode->i_mapping,
86                         ~(__GFP_FS | __GFP_HIGHMEM)));
87
88         return inode;
89 }
90
91 struct inode *lookup_free_space_inode(
92                 struct btrfs_block_group_cache *block_group,
93                 struct btrfs_path *path)
94 {
95         struct btrfs_fs_info *fs_info = block_group->fs_info;
96         struct inode *inode = NULL;
97         u32 flags = BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
98
99         spin_lock(&block_group->lock);
100         if (block_group->inode)
101                 inode = igrab(block_group->inode);
102         spin_unlock(&block_group->lock);
103         if (inode)
104                 return inode;
105
106         inode = __lookup_free_space_inode(fs_info->tree_root, path,
107                                           block_group->key.objectid);
108         if (IS_ERR(inode))
109                 return inode;
110
111         spin_lock(&block_group->lock);
112         if (!((BTRFS_I(inode)->flags & flags) == flags)) {
113                 btrfs_info(fs_info, "Old style space inode found, converting.");
114                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM |
115                         BTRFS_INODE_NODATACOW;
116                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
117         }
118
119         if (!block_group->iref) {
120                 block_group->inode = igrab(inode);
121                 block_group->iref = 1;
122         }
123         spin_unlock(&block_group->lock);
124
125         return inode;
126 }
127
128 static int __create_free_space_inode(struct btrfs_root *root,
129                                      struct btrfs_trans_handle *trans,
130                                      struct btrfs_path *path,
131                                      u64 ino, u64 offset)
132 {
133         struct btrfs_key key;
134         struct btrfs_disk_key disk_key;
135         struct btrfs_free_space_header *header;
136         struct btrfs_inode_item *inode_item;
137         struct extent_buffer *leaf;
138         u64 flags = BTRFS_INODE_NOCOMPRESS | BTRFS_INODE_PREALLOC;
139         int ret;
140
141         ret = btrfs_insert_empty_inode(trans, root, path, ino);
142         if (ret)
143                 return ret;
144
145         /* We inline crc's for the free disk space cache */
146         if (ino != BTRFS_FREE_INO_OBJECTID)
147                 flags |= BTRFS_INODE_NODATASUM | BTRFS_INODE_NODATACOW;
148
149         leaf = path->nodes[0];
150         inode_item = btrfs_item_ptr(leaf, path->slots[0],
151                                     struct btrfs_inode_item);
152         btrfs_item_key(leaf, &disk_key, path->slots[0]);
153         memzero_extent_buffer(leaf, (unsigned long)inode_item,
154                              sizeof(*inode_item));
155         btrfs_set_inode_generation(leaf, inode_item, trans->transid);
156         btrfs_set_inode_size(leaf, inode_item, 0);
157         btrfs_set_inode_nbytes(leaf, inode_item, 0);
158         btrfs_set_inode_uid(leaf, inode_item, 0);
159         btrfs_set_inode_gid(leaf, inode_item, 0);
160         btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
161         btrfs_set_inode_flags(leaf, inode_item, flags);
162         btrfs_set_inode_nlink(leaf, inode_item, 1);
163         btrfs_set_inode_transid(leaf, inode_item, trans->transid);
164         btrfs_set_inode_block_group(leaf, inode_item, offset);
165         btrfs_mark_buffer_dirty(leaf);
166         btrfs_release_path(path);
167
168         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
169         key.offset = offset;
170         key.type = 0;
171         ret = btrfs_insert_empty_item(trans, root, path, &key,
172                                       sizeof(struct btrfs_free_space_header));
173         if (ret < 0) {
174                 btrfs_release_path(path);
175                 return ret;
176         }
177
178         leaf = path->nodes[0];
179         header = btrfs_item_ptr(leaf, path->slots[0],
180                                 struct btrfs_free_space_header);
181         memzero_extent_buffer(leaf, (unsigned long)header, sizeof(*header));
182         btrfs_set_free_space_key(leaf, header, &disk_key);
183         btrfs_mark_buffer_dirty(leaf);
184         btrfs_release_path(path);
185
186         return 0;
187 }
188
189 int create_free_space_inode(struct btrfs_trans_handle *trans,
190                             struct btrfs_block_group_cache *block_group,
191                             struct btrfs_path *path)
192 {
193         int ret;
194         u64 ino;
195
196         ret = btrfs_find_free_objectid(trans->fs_info->tree_root, &ino);
197         if (ret < 0)
198                 return ret;
199
200         return __create_free_space_inode(trans->fs_info->tree_root, trans, path,
201                                          ino, block_group->key.objectid);
202 }
203
204 int btrfs_check_trunc_cache_free_space(struct btrfs_fs_info *fs_info,
205                                        struct btrfs_block_rsv *rsv)
206 {
207         u64 needed_bytes;
208         int ret;
209
210         /* 1 for slack space, 1 for updating the inode */
211         needed_bytes = btrfs_calc_trunc_metadata_size(fs_info, 1) +
212                 btrfs_calc_trans_metadata_size(fs_info, 1);
213
214         spin_lock(&rsv->lock);
215         if (rsv->reserved < needed_bytes)
216                 ret = -ENOSPC;
217         else
218                 ret = 0;
219         spin_unlock(&rsv->lock);
220         return ret;
221 }
222
223 int btrfs_truncate_free_space_cache(struct btrfs_trans_handle *trans,
224                                     struct btrfs_block_group_cache *block_group,
225                                     struct inode *inode)
226 {
227         struct btrfs_root *root = BTRFS_I(inode)->root;
228         int ret = 0;
229         bool locked = false;
230
231         if (block_group) {
232                 struct btrfs_path *path = btrfs_alloc_path();
233
234                 if (!path) {
235                         ret = -ENOMEM;
236                         goto fail;
237                 }
238                 locked = true;
239                 mutex_lock(&trans->transaction->cache_write_mutex);
240                 if (!list_empty(&block_group->io_list)) {
241                         list_del_init(&block_group->io_list);
242
243                         btrfs_wait_cache_io(trans, block_group, path);
244                         btrfs_put_block_group(block_group);
245                 }
246
247                 /*
248                  * now that we've truncated the cache away, its no longer
249                  * setup or written
250                  */
251                 spin_lock(&block_group->lock);
252                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
253                 spin_unlock(&block_group->lock);
254                 btrfs_free_path(path);
255         }
256
257         btrfs_i_size_write(BTRFS_I(inode), 0);
258         truncate_pagecache(inode, 0);
259
260         /*
261          * We skip the throttling logic for free space cache inodes, so we don't
262          * need to check for -EAGAIN.
263          */
264         ret = btrfs_truncate_inode_items(trans, root, inode,
265                                          0, BTRFS_EXTENT_DATA_KEY);
266         if (ret)
267                 goto fail;
268
269         ret = btrfs_update_inode(trans, root, inode);
270
271 fail:
272         if (locked)
273                 mutex_unlock(&trans->transaction->cache_write_mutex);
274         if (ret)
275                 btrfs_abort_transaction(trans, ret);
276
277         return ret;
278 }
279
280 static void readahead_cache(struct inode *inode)
281 {
282         struct file_ra_state *ra;
283         unsigned long last_index;
284
285         ra = kzalloc(sizeof(*ra), GFP_NOFS);
286         if (!ra)
287                 return;
288
289         file_ra_state_init(ra, inode->i_mapping);
290         last_index = (i_size_read(inode) - 1) >> PAGE_SHIFT;
291
292         page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
293
294         kfree(ra);
295 }
296
297 static int io_ctl_init(struct btrfs_io_ctl *io_ctl, struct inode *inode,
298                        int write)
299 {
300         int num_pages;
301         int check_crcs = 0;
302
303         num_pages = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
304
305         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FREE_INO_OBJECTID)
306                 check_crcs = 1;
307
308         /* Make sure we can fit our crcs and generation into the first page */
309         if (write && check_crcs &&
310             (num_pages * sizeof(u32) + sizeof(u64)) > PAGE_SIZE)
311                 return -ENOSPC;
312
313         memset(io_ctl, 0, sizeof(struct btrfs_io_ctl));
314
315         io_ctl->pages = kcalloc(num_pages, sizeof(struct page *), GFP_NOFS);
316         if (!io_ctl->pages)
317                 return -ENOMEM;
318
319         io_ctl->num_pages = num_pages;
320         io_ctl->fs_info = btrfs_sb(inode->i_sb);
321         io_ctl->check_crcs = check_crcs;
322         io_ctl->inode = inode;
323
324         return 0;
325 }
326 ALLOW_ERROR_INJECTION(io_ctl_init, ERRNO);
327
328 static void io_ctl_free(struct btrfs_io_ctl *io_ctl)
329 {
330         kfree(io_ctl->pages);
331         io_ctl->pages = NULL;
332 }
333
334 static void io_ctl_unmap_page(struct btrfs_io_ctl *io_ctl)
335 {
336         if (io_ctl->cur) {
337                 io_ctl->cur = NULL;
338                 io_ctl->orig = NULL;
339         }
340 }
341
342 static void io_ctl_map_page(struct btrfs_io_ctl *io_ctl, int clear)
343 {
344         ASSERT(io_ctl->index < io_ctl->num_pages);
345         io_ctl->page = io_ctl->pages[io_ctl->index++];
346         io_ctl->cur = page_address(io_ctl->page);
347         io_ctl->orig = io_ctl->cur;
348         io_ctl->size = PAGE_SIZE;
349         if (clear)
350                 clear_page(io_ctl->cur);
351 }
352
353 static void io_ctl_drop_pages(struct btrfs_io_ctl *io_ctl)
354 {
355         int i;
356
357         io_ctl_unmap_page(io_ctl);
358
359         for (i = 0; i < io_ctl->num_pages; i++) {
360                 if (io_ctl->pages[i]) {
361                         ClearPageChecked(io_ctl->pages[i]);
362                         unlock_page(io_ctl->pages[i]);
363                         put_page(io_ctl->pages[i]);
364                 }
365         }
366 }
367
368 static int io_ctl_prepare_pages(struct btrfs_io_ctl *io_ctl, struct inode *inode,
369                                 int uptodate)
370 {
371         struct page *page;
372         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
373         int i;
374
375         for (i = 0; i < io_ctl->num_pages; i++) {
376                 page = find_or_create_page(inode->i_mapping, i, mask);
377                 if (!page) {
378                         io_ctl_drop_pages(io_ctl);
379                         return -ENOMEM;
380                 }
381                 io_ctl->pages[i] = page;
382                 if (uptodate && !PageUptodate(page)) {
383                         btrfs_readpage(NULL, page);
384                         lock_page(page);
385                         if (!PageUptodate(page)) {
386                                 btrfs_err(BTRFS_I(inode)->root->fs_info,
387                                            "error reading free space cache");
388                                 io_ctl_drop_pages(io_ctl);
389                                 return -EIO;
390                         }
391                 }
392         }
393
394         for (i = 0; i < io_ctl->num_pages; i++) {
395                 clear_page_dirty_for_io(io_ctl->pages[i]);
396                 set_page_extent_mapped(io_ctl->pages[i]);
397         }
398
399         return 0;
400 }
401
402 static void io_ctl_set_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
403 {
404         __le64 *val;
405
406         io_ctl_map_page(io_ctl, 1);
407
408         /*
409          * Skip the csum areas.  If we don't check crcs then we just have a
410          * 64bit chunk at the front of the first page.
411          */
412         if (io_ctl->check_crcs) {
413                 io_ctl->cur += (sizeof(u32) * io_ctl->num_pages);
414                 io_ctl->size -= sizeof(u64) + (sizeof(u32) * io_ctl->num_pages);
415         } else {
416                 io_ctl->cur += sizeof(u64);
417                 io_ctl->size -= sizeof(u64) * 2;
418         }
419
420         val = io_ctl->cur;
421         *val = cpu_to_le64(generation);
422         io_ctl->cur += sizeof(u64);
423 }
424
425 static int io_ctl_check_generation(struct btrfs_io_ctl *io_ctl, u64 generation)
426 {
427         __le64 *gen;
428
429         /*
430          * Skip the crc area.  If we don't check crcs then we just have a 64bit
431          * chunk at the front of the first page.
432          */
433         if (io_ctl->check_crcs) {
434                 io_ctl->cur += sizeof(u32) * io_ctl->num_pages;
435                 io_ctl->size -= sizeof(u64) +
436                         (sizeof(u32) * io_ctl->num_pages);
437         } else {
438                 io_ctl->cur += sizeof(u64);
439                 io_ctl->size -= sizeof(u64) * 2;
440         }
441
442         gen = io_ctl->cur;
443         if (le64_to_cpu(*gen) != generation) {
444                 btrfs_err_rl(io_ctl->fs_info,
445                         "space cache generation (%llu) does not match inode (%llu)",
446                                 *gen, generation);
447                 io_ctl_unmap_page(io_ctl);
448                 return -EIO;
449         }
450         io_ctl->cur += sizeof(u64);
451         return 0;
452 }
453
454 static void io_ctl_set_crc(struct btrfs_io_ctl *io_ctl, int index)
455 {
456         u32 *tmp;
457         u32 crc = ~(u32)0;
458         unsigned offset = 0;
459
460         if (!io_ctl->check_crcs) {
461                 io_ctl_unmap_page(io_ctl);
462                 return;
463         }
464
465         if (index == 0)
466                 offset = sizeof(u32) * io_ctl->num_pages;
467
468         crc = btrfs_csum_data(io_ctl->orig + offset, crc,
469                               PAGE_SIZE - offset);
470         btrfs_csum_final(crc, (u8 *)&crc);
471         io_ctl_unmap_page(io_ctl);
472         tmp = page_address(io_ctl->pages[0]);
473         tmp += index;
474         *tmp = crc;
475 }
476
477 static int io_ctl_check_crc(struct btrfs_io_ctl *io_ctl, int index)
478 {
479         u32 *tmp, val;
480         u32 crc = ~(u32)0;
481         unsigned offset = 0;
482
483         if (!io_ctl->check_crcs) {
484                 io_ctl_map_page(io_ctl, 0);
485                 return 0;
486         }
487
488         if (index == 0)
489                 offset = sizeof(u32) * io_ctl->num_pages;
490
491         tmp = page_address(io_ctl->pages[0]);
492         tmp += index;
493         val = *tmp;
494
495         io_ctl_map_page(io_ctl, 0);
496         crc = btrfs_csum_data(io_ctl->orig + offset, crc,
497                               PAGE_SIZE - offset);
498         btrfs_csum_final(crc, (u8 *)&crc);
499         if (val != crc) {
500                 btrfs_err_rl(io_ctl->fs_info,
501                         "csum mismatch on free space cache");
502                 io_ctl_unmap_page(io_ctl);
503                 return -EIO;
504         }
505
506         return 0;
507 }
508
509 static int io_ctl_add_entry(struct btrfs_io_ctl *io_ctl, u64 offset, u64 bytes,
510                             void *bitmap)
511 {
512         struct btrfs_free_space_entry *entry;
513
514         if (!io_ctl->cur)
515                 return -ENOSPC;
516
517         entry = io_ctl->cur;
518         entry->offset = cpu_to_le64(offset);
519         entry->bytes = cpu_to_le64(bytes);
520         entry->type = (bitmap) ? BTRFS_FREE_SPACE_BITMAP :
521                 BTRFS_FREE_SPACE_EXTENT;
522         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
523         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
524
525         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
526                 return 0;
527
528         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
529
530         /* No more pages to map */
531         if (io_ctl->index >= io_ctl->num_pages)
532                 return 0;
533
534         /* map the next page */
535         io_ctl_map_page(io_ctl, 1);
536         return 0;
537 }
538
539 static int io_ctl_add_bitmap(struct btrfs_io_ctl *io_ctl, void *bitmap)
540 {
541         if (!io_ctl->cur)
542                 return -ENOSPC;
543
544         /*
545          * If we aren't at the start of the current page, unmap this one and
546          * map the next one if there is any left.
547          */
548         if (io_ctl->cur != io_ctl->orig) {
549                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
550                 if (io_ctl->index >= io_ctl->num_pages)
551                         return -ENOSPC;
552                 io_ctl_map_page(io_ctl, 0);
553         }
554
555         copy_page(io_ctl->cur, bitmap);
556         io_ctl_set_crc(io_ctl, io_ctl->index - 1);
557         if (io_ctl->index < io_ctl->num_pages)
558                 io_ctl_map_page(io_ctl, 0);
559         return 0;
560 }
561
562 static void io_ctl_zero_remaining_pages(struct btrfs_io_ctl *io_ctl)
563 {
564         /*
565          * If we're not on the boundary we know we've modified the page and we
566          * need to crc the page.
567          */
568         if (io_ctl->cur != io_ctl->orig)
569                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
570         else
571                 io_ctl_unmap_page(io_ctl);
572
573         while (io_ctl->index < io_ctl->num_pages) {
574                 io_ctl_map_page(io_ctl, 1);
575                 io_ctl_set_crc(io_ctl, io_ctl->index - 1);
576         }
577 }
578
579 static int io_ctl_read_entry(struct btrfs_io_ctl *io_ctl,
580                             struct btrfs_free_space *entry, u8 *type)
581 {
582         struct btrfs_free_space_entry *e;
583         int ret;
584
585         if (!io_ctl->cur) {
586                 ret = io_ctl_check_crc(io_ctl, io_ctl->index);
587                 if (ret)
588                         return ret;
589         }
590
591         e = io_ctl->cur;
592         entry->offset = le64_to_cpu(e->offset);
593         entry->bytes = le64_to_cpu(e->bytes);
594         *type = e->type;
595         io_ctl->cur += sizeof(struct btrfs_free_space_entry);
596         io_ctl->size -= sizeof(struct btrfs_free_space_entry);
597
598         if (io_ctl->size >= sizeof(struct btrfs_free_space_entry))
599                 return 0;
600
601         io_ctl_unmap_page(io_ctl);
602
603         return 0;
604 }
605
606 static int io_ctl_read_bitmap(struct btrfs_io_ctl *io_ctl,
607                               struct btrfs_free_space *entry)
608 {
609         int ret;
610
611         ret = io_ctl_check_crc(io_ctl, io_ctl->index);
612         if (ret)
613                 return ret;
614
615         copy_page(entry->bitmap, io_ctl->cur);
616         io_ctl_unmap_page(io_ctl);
617
618         return 0;
619 }
620
621 /*
622  * Since we attach pinned extents after the fact we can have contiguous sections
623  * of free space that are split up in entries.  This poses a problem with the
624  * tree logging stuff since it could have allocated across what appears to be 2
625  * entries since we would have merged the entries when adding the pinned extents
626  * back to the free space cache.  So run through the space cache that we just
627  * loaded and merge contiguous entries.  This will make the log replay stuff not
628  * blow up and it will make for nicer allocator behavior.
629  */
630 static void merge_space_tree(struct btrfs_free_space_ctl *ctl)
631 {
632         struct btrfs_free_space *e, *prev = NULL;
633         struct rb_node *n;
634
635 again:
636         spin_lock(&ctl->tree_lock);
637         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
638                 e = rb_entry(n, struct btrfs_free_space, offset_index);
639                 if (!prev)
640                         goto next;
641                 if (e->bitmap || prev->bitmap)
642                         goto next;
643                 if (prev->offset + prev->bytes == e->offset) {
644                         unlink_free_space(ctl, prev);
645                         unlink_free_space(ctl, e);
646                         prev->bytes += e->bytes;
647                         kmem_cache_free(btrfs_free_space_cachep, e);
648                         link_free_space(ctl, prev);
649                         prev = NULL;
650                         spin_unlock(&ctl->tree_lock);
651                         goto again;
652                 }
653 next:
654                 prev = e;
655         }
656         spin_unlock(&ctl->tree_lock);
657 }
658
659 static int __load_free_space_cache(struct btrfs_root *root, struct inode *inode,
660                                    struct btrfs_free_space_ctl *ctl,
661                                    struct btrfs_path *path, u64 offset)
662 {
663         struct btrfs_fs_info *fs_info = root->fs_info;
664         struct btrfs_free_space_header *header;
665         struct extent_buffer *leaf;
666         struct btrfs_io_ctl io_ctl;
667         struct btrfs_key key;
668         struct btrfs_free_space *e, *n;
669         LIST_HEAD(bitmaps);
670         u64 num_entries;
671         u64 num_bitmaps;
672         u64 generation;
673         u8 type;
674         int ret = 0;
675
676         /* Nothing in the space cache, goodbye */
677         if (!i_size_read(inode))
678                 return 0;
679
680         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
681         key.offset = offset;
682         key.type = 0;
683
684         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
685         if (ret < 0)
686                 return 0;
687         else if (ret > 0) {
688                 btrfs_release_path(path);
689                 return 0;
690         }
691
692         ret = -1;
693
694         leaf = path->nodes[0];
695         header = btrfs_item_ptr(leaf, path->slots[0],
696                                 struct btrfs_free_space_header);
697         num_entries = btrfs_free_space_entries(leaf, header);
698         num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
699         generation = btrfs_free_space_generation(leaf, header);
700         btrfs_release_path(path);
701
702         if (!BTRFS_I(inode)->generation) {
703                 btrfs_info(fs_info,
704                            "the free space cache file (%llu) is invalid, skip it",
705                            offset);
706                 return 0;
707         }
708
709         if (BTRFS_I(inode)->generation != generation) {
710                 btrfs_err(fs_info,
711                           "free space inode generation (%llu) did not match free space cache generation (%llu)",
712                           BTRFS_I(inode)->generation, generation);
713                 return 0;
714         }
715
716         if (!num_entries)
717                 return 0;
718
719         ret = io_ctl_init(&io_ctl, inode, 0);
720         if (ret)
721                 return ret;
722
723         readahead_cache(inode);
724
725         ret = io_ctl_prepare_pages(&io_ctl, inode, 1);
726         if (ret)
727                 goto out;
728
729         ret = io_ctl_check_crc(&io_ctl, 0);
730         if (ret)
731                 goto free_cache;
732
733         ret = io_ctl_check_generation(&io_ctl, generation);
734         if (ret)
735                 goto free_cache;
736
737         while (num_entries) {
738                 e = kmem_cache_zalloc(btrfs_free_space_cachep,
739                                       GFP_NOFS);
740                 if (!e)
741                         goto free_cache;
742
743                 ret = io_ctl_read_entry(&io_ctl, e, &type);
744                 if (ret) {
745                         kmem_cache_free(btrfs_free_space_cachep, e);
746                         goto free_cache;
747                 }
748
749                 if (!e->bytes) {
750                         kmem_cache_free(btrfs_free_space_cachep, e);
751                         goto free_cache;
752                 }
753
754                 if (type == BTRFS_FREE_SPACE_EXTENT) {
755                         spin_lock(&ctl->tree_lock);
756                         ret = link_free_space(ctl, e);
757                         spin_unlock(&ctl->tree_lock);
758                         if (ret) {
759                                 btrfs_err(fs_info,
760                                         "Duplicate entries in free space cache, dumping");
761                                 kmem_cache_free(btrfs_free_space_cachep, e);
762                                 goto free_cache;
763                         }
764                 } else {
765                         ASSERT(num_bitmaps);
766                         num_bitmaps--;
767                         e->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
768                         if (!e->bitmap) {
769                                 kmem_cache_free(
770                                         btrfs_free_space_cachep, e);
771                                 goto free_cache;
772                         }
773                         spin_lock(&ctl->tree_lock);
774                         ret = link_free_space(ctl, e);
775                         ctl->total_bitmaps++;
776                         ctl->op->recalc_thresholds(ctl);
777                         spin_unlock(&ctl->tree_lock);
778                         if (ret) {
779                                 btrfs_err(fs_info,
780                                         "Duplicate entries in free space cache, dumping");
781                                 kmem_cache_free(btrfs_free_space_cachep, e);
782                                 goto free_cache;
783                         }
784                         list_add_tail(&e->list, &bitmaps);
785                 }
786
787                 num_entries--;
788         }
789
790         io_ctl_unmap_page(&io_ctl);
791
792         /*
793          * We add the bitmaps at the end of the entries in order that
794          * the bitmap entries are added to the cache.
795          */
796         list_for_each_entry_safe(e, n, &bitmaps, list) {
797                 list_del_init(&e->list);
798                 ret = io_ctl_read_bitmap(&io_ctl, e);
799                 if (ret)
800                         goto free_cache;
801         }
802
803         io_ctl_drop_pages(&io_ctl);
804         merge_space_tree(ctl);
805         ret = 1;
806 out:
807         io_ctl_free(&io_ctl);
808         return ret;
809 free_cache:
810         io_ctl_drop_pages(&io_ctl);
811         __btrfs_remove_free_space_cache(ctl);
812         goto out;
813 }
814
815 int load_free_space_cache(struct btrfs_block_group_cache *block_group)
816 {
817         struct btrfs_fs_info *fs_info = block_group->fs_info;
818         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
819         struct inode *inode;
820         struct btrfs_path *path;
821         int ret = 0;
822         bool matched;
823         u64 used = btrfs_block_group_used(&block_group->item);
824
825         /*
826          * If this block group has been marked to be cleared for one reason or
827          * another then we can't trust the on disk cache, so just return.
828          */
829         spin_lock(&block_group->lock);
830         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
831                 spin_unlock(&block_group->lock);
832                 return 0;
833         }
834         spin_unlock(&block_group->lock);
835
836         path = btrfs_alloc_path();
837         if (!path)
838                 return 0;
839         path->search_commit_root = 1;
840         path->skip_locking = 1;
841
842         /*
843          * We must pass a path with search_commit_root set to btrfs_iget in
844          * order to avoid a deadlock when allocating extents for the tree root.
845          *
846          * When we are COWing an extent buffer from the tree root, when looking
847          * for a free extent, at extent-tree.c:find_free_extent(), we can find
848          * block group without its free space cache loaded. When we find one
849          * we must load its space cache which requires reading its free space
850          * cache's inode item from the root tree. If this inode item is located
851          * in the same leaf that we started COWing before, then we end up in
852          * deadlock on the extent buffer (trying to read lock it when we
853          * previously write locked it).
854          *
855          * It's safe to read the inode item using the commit root because
856          * block groups, once loaded, stay in memory forever (until they are
857          * removed) as well as their space caches once loaded. New block groups
858          * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
859          * we will never try to read their inode item while the fs is mounted.
860          */
861         inode = lookup_free_space_inode(block_group, path);
862         if (IS_ERR(inode)) {
863                 btrfs_free_path(path);
864                 return 0;
865         }
866
867         /* We may have converted the inode and made the cache invalid. */
868         spin_lock(&block_group->lock);
869         if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
870                 spin_unlock(&block_group->lock);
871                 btrfs_free_path(path);
872                 goto out;
873         }
874         spin_unlock(&block_group->lock);
875
876         ret = __load_free_space_cache(fs_info->tree_root, inode, ctl,
877                                       path, block_group->key.objectid);
878         btrfs_free_path(path);
879         if (ret <= 0)
880                 goto out;
881
882         spin_lock(&ctl->tree_lock);
883         matched = (ctl->free_space == (block_group->key.offset - used -
884                                        block_group->bytes_super));
885         spin_unlock(&ctl->tree_lock);
886
887         if (!matched) {
888                 __btrfs_remove_free_space_cache(ctl);
889                 btrfs_warn(fs_info,
890                            "block group %llu has wrong amount of free space",
891                            block_group->key.objectid);
892                 ret = -1;
893         }
894 out:
895         if (ret < 0) {
896                 /* This cache is bogus, make sure it gets cleared */
897                 spin_lock(&block_group->lock);
898                 block_group->disk_cache_state = BTRFS_DC_CLEAR;
899                 spin_unlock(&block_group->lock);
900                 ret = 0;
901
902                 btrfs_warn(fs_info,
903                            "failed to load free space cache for block group %llu, rebuilding it now",
904                            block_group->key.objectid);
905         }
906
907         iput(inode);
908         return ret;
909 }
910
911 static noinline_for_stack
912 int write_cache_extent_entries(struct btrfs_io_ctl *io_ctl,
913                               struct btrfs_free_space_ctl *ctl,
914                               struct btrfs_block_group_cache *block_group,
915                               int *entries, int *bitmaps,
916                               struct list_head *bitmap_list)
917 {
918         int ret;
919         struct btrfs_free_cluster *cluster = NULL;
920         struct btrfs_free_cluster *cluster_locked = NULL;
921         struct rb_node *node = rb_first(&ctl->free_space_offset);
922         struct btrfs_trim_range *trim_entry;
923
924         /* Get the cluster for this block_group if it exists */
925         if (block_group && !list_empty(&block_group->cluster_list)) {
926                 cluster = list_entry(block_group->cluster_list.next,
927                                      struct btrfs_free_cluster,
928                                      block_group_list);
929         }
930
931         if (!node && cluster) {
932                 cluster_locked = cluster;
933                 spin_lock(&cluster_locked->lock);
934                 node = rb_first(&cluster->root);
935                 cluster = NULL;
936         }
937
938         /* Write out the extent entries */
939         while (node) {
940                 struct btrfs_free_space *e;
941
942                 e = rb_entry(node, struct btrfs_free_space, offset_index);
943                 *entries += 1;
944
945                 ret = io_ctl_add_entry(io_ctl, e->offset, e->bytes,
946                                        e->bitmap);
947                 if (ret)
948                         goto fail;
949
950                 if (e->bitmap) {
951                         list_add_tail(&e->list, bitmap_list);
952                         *bitmaps += 1;
953                 }
954                 node = rb_next(node);
955                 if (!node && cluster) {
956                         node = rb_first(&cluster->root);
957                         cluster_locked = cluster;
958                         spin_lock(&cluster_locked->lock);
959                         cluster = NULL;
960                 }
961         }
962         if (cluster_locked) {
963                 spin_unlock(&cluster_locked->lock);
964                 cluster_locked = NULL;
965         }
966
967         /*
968          * Make sure we don't miss any range that was removed from our rbtree
969          * because trimming is running. Otherwise after a umount+mount (or crash
970          * after committing the transaction) we would leak free space and get
971          * an inconsistent free space cache report from fsck.
972          */
973         list_for_each_entry(trim_entry, &ctl->trimming_ranges, list) {
974                 ret = io_ctl_add_entry(io_ctl, trim_entry->start,
975                                        trim_entry->bytes, NULL);
976                 if (ret)
977                         goto fail;
978                 *entries += 1;
979         }
980
981         return 0;
982 fail:
983         if (cluster_locked)
984                 spin_unlock(&cluster_locked->lock);
985         return -ENOSPC;
986 }
987
988 static noinline_for_stack int
989 update_cache_item(struct btrfs_trans_handle *trans,
990                   struct btrfs_root *root,
991                   struct inode *inode,
992                   struct btrfs_path *path, u64 offset,
993                   int entries, int bitmaps)
994 {
995         struct btrfs_key key;
996         struct btrfs_free_space_header *header;
997         struct extent_buffer *leaf;
998         int ret;
999
1000         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
1001         key.offset = offset;
1002         key.type = 0;
1003
1004         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1005         if (ret < 0) {
1006                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1007                                  EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL);
1008                 goto fail;
1009         }
1010         leaf = path->nodes[0];
1011         if (ret > 0) {
1012                 struct btrfs_key found_key;
1013                 ASSERT(path->slots[0]);
1014                 path->slots[0]--;
1015                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1016                 if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
1017                     found_key.offset != offset) {
1018                         clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
1019                                          inode->i_size - 1,
1020                                          EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
1021                                          NULL);
1022                         btrfs_release_path(path);
1023                         goto fail;
1024                 }
1025         }
1026
1027         BTRFS_I(inode)->generation = trans->transid;
1028         header = btrfs_item_ptr(leaf, path->slots[0],
1029                                 struct btrfs_free_space_header);
1030         btrfs_set_free_space_entries(leaf, header, entries);
1031         btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
1032         btrfs_set_free_space_generation(leaf, header, trans->transid);
1033         btrfs_mark_buffer_dirty(leaf);
1034         btrfs_release_path(path);
1035
1036         return 0;
1037
1038 fail:
1039         return -1;
1040 }
1041
1042 static noinline_for_stack int write_pinned_extent_entries(
1043                             struct btrfs_block_group_cache *block_group,
1044                             struct btrfs_io_ctl *io_ctl,
1045                             int *entries)
1046 {
1047         u64 start, extent_start, extent_end, len;
1048         struct extent_io_tree *unpin = NULL;
1049         int ret;
1050
1051         if (!block_group)
1052                 return 0;
1053
1054         /*
1055          * We want to add any pinned extents to our free space cache
1056          * so we don't leak the space
1057          *
1058          * We shouldn't have switched the pinned extents yet so this is the
1059          * right one
1060          */
1061         unpin = block_group->fs_info->pinned_extents;
1062
1063         start = block_group->key.objectid;
1064
1065         while (start < block_group->key.objectid + block_group->key.offset) {
1066                 ret = find_first_extent_bit(unpin, start,
1067                                             &extent_start, &extent_end,
1068                                             EXTENT_DIRTY, NULL);
1069                 if (ret)
1070                         return 0;
1071
1072                 /* This pinned extent is out of our range */
1073                 if (extent_start >= block_group->key.objectid +
1074                     block_group->key.offset)
1075                         return 0;
1076
1077                 extent_start = max(extent_start, start);
1078                 extent_end = min(block_group->key.objectid +
1079                                  block_group->key.offset, extent_end + 1);
1080                 len = extent_end - extent_start;
1081
1082                 *entries += 1;
1083                 ret = io_ctl_add_entry(io_ctl, extent_start, len, NULL);
1084                 if (ret)
1085                         return -ENOSPC;
1086
1087                 start = extent_end;
1088         }
1089
1090         return 0;
1091 }
1092
1093 static noinline_for_stack int
1094 write_bitmap_entries(struct btrfs_io_ctl *io_ctl, struct list_head *bitmap_list)
1095 {
1096         struct btrfs_free_space *entry, *next;
1097         int ret;
1098
1099         /* Write out the bitmaps */
1100         list_for_each_entry_safe(entry, next, bitmap_list, list) {
1101                 ret = io_ctl_add_bitmap(io_ctl, entry->bitmap);
1102                 if (ret)
1103                         return -ENOSPC;
1104                 list_del_init(&entry->list);
1105         }
1106
1107         return 0;
1108 }
1109
1110 static int flush_dirty_cache(struct inode *inode)
1111 {
1112         int ret;
1113
1114         ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
1115         if (ret)
1116                 clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
1117                                  EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL);
1118
1119         return ret;
1120 }
1121
1122 static void noinline_for_stack
1123 cleanup_bitmap_list(struct list_head *bitmap_list)
1124 {
1125         struct btrfs_free_space *entry, *next;
1126
1127         list_for_each_entry_safe(entry, next, bitmap_list, list)
1128                 list_del_init(&entry->list);
1129 }
1130
1131 static void noinline_for_stack
1132 cleanup_write_cache_enospc(struct inode *inode,
1133                            struct btrfs_io_ctl *io_ctl,
1134                            struct extent_state **cached_state)
1135 {
1136         io_ctl_drop_pages(io_ctl);
1137         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1138                              i_size_read(inode) - 1, cached_state);
1139 }
1140
1141 static int __btrfs_wait_cache_io(struct btrfs_root *root,
1142                                  struct btrfs_trans_handle *trans,
1143                                  struct btrfs_block_group_cache *block_group,
1144                                  struct btrfs_io_ctl *io_ctl,
1145                                  struct btrfs_path *path, u64 offset)
1146 {
1147         int ret;
1148         struct inode *inode = io_ctl->inode;
1149
1150         if (!inode)
1151                 return 0;
1152
1153         /* Flush the dirty pages in the cache file. */
1154         ret = flush_dirty_cache(inode);
1155         if (ret)
1156                 goto out;
1157
1158         /* Update the cache item to tell everyone this cache file is valid. */
1159         ret = update_cache_item(trans, root, inode, path, offset,
1160                                 io_ctl->entries, io_ctl->bitmaps);
1161 out:
1162         io_ctl_free(io_ctl);
1163         if (ret) {
1164                 invalidate_inode_pages2(inode->i_mapping);
1165                 BTRFS_I(inode)->generation = 0;
1166                 if (block_group) {
1167 #ifdef DEBUG
1168                         btrfs_err(root->fs_info,
1169                                   "failed to write free space cache for block group %llu",
1170                                   block_group->key.objectid);
1171 #endif
1172                 }
1173         }
1174         btrfs_update_inode(trans, root, inode);
1175
1176         if (block_group) {
1177                 /* the dirty list is protected by the dirty_bgs_lock */
1178                 spin_lock(&trans->transaction->dirty_bgs_lock);
1179
1180                 /* the disk_cache_state is protected by the block group lock */
1181                 spin_lock(&block_group->lock);
1182
1183                 /*
1184                  * only mark this as written if we didn't get put back on
1185                  * the dirty list while waiting for IO.   Otherwise our
1186                  * cache state won't be right, and we won't get written again
1187                  */
1188                 if (!ret && list_empty(&block_group->dirty_list))
1189                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1190                 else if (ret)
1191                         block_group->disk_cache_state = BTRFS_DC_ERROR;
1192
1193                 spin_unlock(&block_group->lock);
1194                 spin_unlock(&trans->transaction->dirty_bgs_lock);
1195                 io_ctl->inode = NULL;
1196                 iput(inode);
1197         }
1198
1199         return ret;
1200
1201 }
1202
1203 static int btrfs_wait_cache_io_root(struct btrfs_root *root,
1204                                     struct btrfs_trans_handle *trans,
1205                                     struct btrfs_io_ctl *io_ctl,
1206                                     struct btrfs_path *path)
1207 {
1208         return __btrfs_wait_cache_io(root, trans, NULL, io_ctl, path, 0);
1209 }
1210
1211 int btrfs_wait_cache_io(struct btrfs_trans_handle *trans,
1212                         struct btrfs_block_group_cache *block_group,
1213                         struct btrfs_path *path)
1214 {
1215         return __btrfs_wait_cache_io(block_group->fs_info->tree_root, trans,
1216                                      block_group, &block_group->io_ctl,
1217                                      path, block_group->key.objectid);
1218 }
1219
1220 /**
1221  * __btrfs_write_out_cache - write out cached info to an inode
1222  * @root - the root the inode belongs to
1223  * @ctl - the free space cache we are going to write out
1224  * @block_group - the block_group for this cache if it belongs to a block_group
1225  * @trans - the trans handle
1226  *
1227  * This function writes out a free space cache struct to disk for quick recovery
1228  * on mount.  This will return 0 if it was successful in writing the cache out,
1229  * or an errno if it was not.
1230  */
1231 static int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode,
1232                                    struct btrfs_free_space_ctl *ctl,
1233                                    struct btrfs_block_group_cache *block_group,
1234                                    struct btrfs_io_ctl *io_ctl,
1235                                    struct btrfs_trans_handle *trans)
1236 {
1237         struct extent_state *cached_state = NULL;
1238         LIST_HEAD(bitmap_list);
1239         int entries = 0;
1240         int bitmaps = 0;
1241         int ret;
1242         int must_iput = 0;
1243
1244         if (!i_size_read(inode))
1245                 return -EIO;
1246
1247         WARN_ON(io_ctl->pages);
1248         ret = io_ctl_init(io_ctl, inode, 1);
1249         if (ret)
1250                 return ret;
1251
1252         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)) {
1253                 down_write(&block_group->data_rwsem);
1254                 spin_lock(&block_group->lock);
1255                 if (block_group->delalloc_bytes) {
1256                         block_group->disk_cache_state = BTRFS_DC_WRITTEN;
1257                         spin_unlock(&block_group->lock);
1258                         up_write(&block_group->data_rwsem);
1259                         BTRFS_I(inode)->generation = 0;
1260                         ret = 0;
1261                         must_iput = 1;
1262                         goto out;
1263                 }
1264                 spin_unlock(&block_group->lock);
1265         }
1266
1267         /* Lock all pages first so we can lock the extent safely. */
1268         ret = io_ctl_prepare_pages(io_ctl, inode, 0);
1269         if (ret)
1270                 goto out_unlock;
1271
1272         lock_extent_bits(&BTRFS_I(inode)->io_tree, 0, i_size_read(inode) - 1,
1273                          &cached_state);
1274
1275         io_ctl_set_generation(io_ctl, trans->transid);
1276
1277         mutex_lock(&ctl->cache_writeout_mutex);
1278         /* Write out the extent entries in the free space cache */
1279         spin_lock(&ctl->tree_lock);
1280         ret = write_cache_extent_entries(io_ctl, ctl,
1281                                          block_group, &entries, &bitmaps,
1282                                          &bitmap_list);
1283         if (ret)
1284                 goto out_nospc_locked;
1285
1286         /*
1287          * Some spaces that are freed in the current transaction are pinned,
1288          * they will be added into free space cache after the transaction is
1289          * committed, we shouldn't lose them.
1290          *
1291          * If this changes while we are working we'll get added back to
1292          * the dirty list and redo it.  No locking needed
1293          */
1294         ret = write_pinned_extent_entries(block_group, io_ctl, &entries);
1295         if (ret)
1296                 goto out_nospc_locked;
1297
1298         /*
1299          * At last, we write out all the bitmaps and keep cache_writeout_mutex
1300          * locked while doing it because a concurrent trim can be manipulating
1301          * or freeing the bitmap.
1302          */
1303         ret = write_bitmap_entries(io_ctl, &bitmap_list);
1304         spin_unlock(&ctl->tree_lock);
1305         mutex_unlock(&ctl->cache_writeout_mutex);
1306         if (ret)
1307                 goto out_nospc;
1308
1309         /* Zero out the rest of the pages just to make sure */
1310         io_ctl_zero_remaining_pages(io_ctl);
1311
1312         /* Everything is written out, now we dirty the pages in the file. */
1313         ret = btrfs_dirty_pages(inode, io_ctl->pages, io_ctl->num_pages, 0,
1314                                 i_size_read(inode), &cached_state);
1315         if (ret)
1316                 goto out_nospc;
1317
1318         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1319                 up_write(&block_group->data_rwsem);
1320         /*
1321          * Release the pages and unlock the extent, we will flush
1322          * them out later
1323          */
1324         io_ctl_drop_pages(io_ctl);
1325
1326         unlock_extent_cached(&BTRFS_I(inode)->io_tree, 0,
1327                              i_size_read(inode) - 1, &cached_state);
1328
1329         /*
1330          * at this point the pages are under IO and we're happy,
1331          * The caller is responsible for waiting on them and updating the
1332          * the cache and the inode
1333          */
1334         io_ctl->entries = entries;
1335         io_ctl->bitmaps = bitmaps;
1336
1337         ret = btrfs_fdatawrite_range(inode, 0, (u64)-1);
1338         if (ret)
1339                 goto out;
1340
1341         return 0;
1342
1343 out:
1344         io_ctl->inode = NULL;
1345         io_ctl_free(io_ctl);
1346         if (ret) {
1347                 invalidate_inode_pages2(inode->i_mapping);
1348                 BTRFS_I(inode)->generation = 0;
1349         }
1350         btrfs_update_inode(trans, root, inode);
1351         if (must_iput)
1352                 iput(inode);
1353         return ret;
1354
1355 out_nospc_locked:
1356         cleanup_bitmap_list(&bitmap_list);
1357         spin_unlock(&ctl->tree_lock);
1358         mutex_unlock(&ctl->cache_writeout_mutex);
1359
1360 out_nospc:
1361         cleanup_write_cache_enospc(inode, io_ctl, &cached_state);
1362
1363 out_unlock:
1364         if (block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA))
1365                 up_write(&block_group->data_rwsem);
1366
1367         goto out;
1368 }
1369
1370 int btrfs_write_out_cache(struct btrfs_trans_handle *trans,
1371                           struct btrfs_block_group_cache *block_group,
1372                           struct btrfs_path *path)
1373 {
1374         struct btrfs_fs_info *fs_info = trans->fs_info;
1375         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
1376         struct inode *inode;
1377         int ret = 0;
1378
1379         spin_lock(&block_group->lock);
1380         if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
1381                 spin_unlock(&block_group->lock);
1382                 return 0;
1383         }
1384         spin_unlock(&block_group->lock);
1385
1386         inode = lookup_free_space_inode(block_group, path);
1387         if (IS_ERR(inode))
1388                 return 0;
1389
1390         ret = __btrfs_write_out_cache(fs_info->tree_root, inode, ctl,
1391                                 block_group, &block_group->io_ctl, trans);
1392         if (ret) {
1393 #ifdef DEBUG
1394                 btrfs_err(fs_info,
1395                           "failed to write free space cache for block group %llu",
1396                           block_group->key.objectid);
1397 #endif
1398                 spin_lock(&block_group->lock);
1399                 block_group->disk_cache_state = BTRFS_DC_ERROR;
1400                 spin_unlock(&block_group->lock);
1401
1402                 block_group->io_ctl.inode = NULL;
1403                 iput(inode);
1404         }
1405
1406         /*
1407          * if ret == 0 the caller is expected to call btrfs_wait_cache_io
1408          * to wait for IO and put the inode
1409          */
1410
1411         return ret;
1412 }
1413
1414 static inline unsigned long offset_to_bit(u64 bitmap_start, u32 unit,
1415                                           u64 offset)
1416 {
1417         ASSERT(offset >= bitmap_start);
1418         offset -= bitmap_start;
1419         return (unsigned long)(div_u64(offset, unit));
1420 }
1421
1422 static inline unsigned long bytes_to_bits(u64 bytes, u32 unit)
1423 {
1424         return (unsigned long)(div_u64(bytes, unit));
1425 }
1426
1427 static inline u64 offset_to_bitmap(struct btrfs_free_space_ctl *ctl,
1428                                    u64 offset)
1429 {
1430         u64 bitmap_start;
1431         u64 bytes_per_bitmap;
1432
1433         bytes_per_bitmap = BITS_PER_BITMAP * ctl->unit;
1434         bitmap_start = offset - ctl->start;
1435         bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
1436         bitmap_start *= bytes_per_bitmap;
1437         bitmap_start += ctl->start;
1438
1439         return bitmap_start;
1440 }
1441
1442 static int tree_insert_offset(struct rb_root *root, u64 offset,
1443                               struct rb_node *node, int bitmap)
1444 {
1445         struct rb_node **p = &root->rb_node;
1446         struct rb_node *parent = NULL;
1447         struct btrfs_free_space *info;
1448
1449         while (*p) {
1450                 parent = *p;
1451                 info = rb_entry(parent, struct btrfs_free_space, offset_index);
1452
1453                 if (offset < info->offset) {
1454                         p = &(*p)->rb_left;
1455                 } else if (offset > info->offset) {
1456                         p = &(*p)->rb_right;
1457                 } else {
1458                         /*
1459                          * we could have a bitmap entry and an extent entry
1460                          * share the same offset.  If this is the case, we want
1461                          * the extent entry to always be found first if we do a
1462                          * linear search through the tree, since we want to have
1463                          * the quickest allocation time, and allocating from an
1464                          * extent is faster than allocating from a bitmap.  So
1465                          * if we're inserting a bitmap and we find an entry at
1466                          * this offset, we want to go right, or after this entry
1467                          * logically.  If we are inserting an extent and we've
1468                          * found a bitmap, we want to go left, or before
1469                          * logically.
1470                          */
1471                         if (bitmap) {
1472                                 if (info->bitmap) {
1473                                         WARN_ON_ONCE(1);
1474                                         return -EEXIST;
1475                                 }
1476                                 p = &(*p)->rb_right;
1477                         } else {
1478                                 if (!info->bitmap) {
1479                                         WARN_ON_ONCE(1);
1480                                         return -EEXIST;
1481                                 }
1482                                 p = &(*p)->rb_left;
1483                         }
1484                 }
1485         }
1486
1487         rb_link_node(node, parent, p);
1488         rb_insert_color(node, root);
1489
1490         return 0;
1491 }
1492
1493 /*
1494  * searches the tree for the given offset.
1495  *
1496  * fuzzy - If this is set, then we are trying to make an allocation, and we just
1497  * want a section that has at least bytes size and comes at or after the given
1498  * offset.
1499  */
1500 static struct btrfs_free_space *
1501 tree_search_offset(struct btrfs_free_space_ctl *ctl,
1502                    u64 offset, int bitmap_only, int fuzzy)
1503 {
1504         struct rb_node *n = ctl->free_space_offset.rb_node;
1505         struct btrfs_free_space *entry, *prev = NULL;
1506
1507         /* find entry that is closest to the 'offset' */
1508         while (1) {
1509                 if (!n) {
1510                         entry = NULL;
1511                         break;
1512                 }
1513
1514                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1515                 prev = entry;
1516
1517                 if (offset < entry->offset)
1518                         n = n->rb_left;
1519                 else if (offset > entry->offset)
1520                         n = n->rb_right;
1521                 else
1522                         break;
1523         }
1524
1525         if (bitmap_only) {
1526                 if (!entry)
1527                         return NULL;
1528                 if (entry->bitmap)
1529                         return entry;
1530
1531                 /*
1532                  * bitmap entry and extent entry may share same offset,
1533                  * in that case, bitmap entry comes after extent entry.
1534                  */
1535                 n = rb_next(n);
1536                 if (!n)
1537                         return NULL;
1538                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1539                 if (entry->offset != offset)
1540                         return NULL;
1541
1542                 WARN_ON(!entry->bitmap);
1543                 return entry;
1544         } else if (entry) {
1545                 if (entry->bitmap) {
1546                         /*
1547                          * if previous extent entry covers the offset,
1548                          * we should return it instead of the bitmap entry
1549                          */
1550                         n = rb_prev(&entry->offset_index);
1551                         if (n) {
1552                                 prev = rb_entry(n, struct btrfs_free_space,
1553                                                 offset_index);
1554                                 if (!prev->bitmap &&
1555                                     prev->offset + prev->bytes > offset)
1556                                         entry = prev;
1557                         }
1558                 }
1559                 return entry;
1560         }
1561
1562         if (!prev)
1563                 return NULL;
1564
1565         /* find last entry before the 'offset' */
1566         entry = prev;
1567         if (entry->offset > offset) {
1568                 n = rb_prev(&entry->offset_index);
1569                 if (n) {
1570                         entry = rb_entry(n, struct btrfs_free_space,
1571                                         offset_index);
1572                         ASSERT(entry->offset <= offset);
1573                 } else {
1574                         if (fuzzy)
1575                                 return entry;
1576                         else
1577                                 return NULL;
1578                 }
1579         }
1580
1581         if (entry->bitmap) {
1582                 n = rb_prev(&entry->offset_index);
1583                 if (n) {
1584                         prev = rb_entry(n, struct btrfs_free_space,
1585                                         offset_index);
1586                         if (!prev->bitmap &&
1587                             prev->offset + prev->bytes > offset)
1588                                 return prev;
1589                 }
1590                 if (entry->offset + BITS_PER_BITMAP * ctl->unit > offset)
1591                         return entry;
1592         } else if (entry->offset + entry->bytes > offset)
1593                 return entry;
1594
1595         if (!fuzzy)
1596                 return NULL;
1597
1598         while (1) {
1599                 if (entry->bitmap) {
1600                         if (entry->offset + BITS_PER_BITMAP *
1601                             ctl->unit > offset)
1602                                 break;
1603                 } else {
1604                         if (entry->offset + entry->bytes > offset)
1605                                 break;
1606                 }
1607
1608                 n = rb_next(&entry->offset_index);
1609                 if (!n)
1610                         return NULL;
1611                 entry = rb_entry(n, struct btrfs_free_space, offset_index);
1612         }
1613         return entry;
1614 }
1615
1616 static inline void
1617 __unlink_free_space(struct btrfs_free_space_ctl *ctl,
1618                     struct btrfs_free_space *info)
1619 {
1620         rb_erase(&info->offset_index, &ctl->free_space_offset);
1621         ctl->free_extents--;
1622 }
1623
1624 static void unlink_free_space(struct btrfs_free_space_ctl *ctl,
1625                               struct btrfs_free_space *info)
1626 {
1627         __unlink_free_space(ctl, info);
1628         ctl->free_space -= info->bytes;
1629 }
1630
1631 static int link_free_space(struct btrfs_free_space_ctl *ctl,
1632                            struct btrfs_free_space *info)
1633 {
1634         int ret = 0;
1635
1636         ASSERT(info->bytes || info->bitmap);
1637         ret = tree_insert_offset(&ctl->free_space_offset, info->offset,
1638                                  &info->offset_index, (info->bitmap != NULL));
1639         if (ret)
1640                 return ret;
1641
1642         ctl->free_space += info->bytes;
1643         ctl->free_extents++;
1644         return ret;
1645 }
1646
1647 static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
1648 {
1649         struct btrfs_block_group_cache *block_group = ctl->private;
1650         u64 max_bytes;
1651         u64 bitmap_bytes;
1652         u64 extent_bytes;
1653         u64 size = block_group->key.offset;
1654         u64 bytes_per_bg = BITS_PER_BITMAP * ctl->unit;
1655         u64 max_bitmaps = div64_u64(size + bytes_per_bg - 1, bytes_per_bg);
1656
1657         max_bitmaps = max_t(u64, max_bitmaps, 1);
1658
1659         ASSERT(ctl->total_bitmaps <= max_bitmaps);
1660
1661         /*
1662          * The goal is to keep the total amount of memory used per 1gb of space
1663          * at or below 32k, so we need to adjust how much memory we allow to be
1664          * used by extent based free space tracking
1665          */
1666         if (size < SZ_1G)
1667                 max_bytes = MAX_CACHE_BYTES_PER_GIG;
1668         else
1669                 max_bytes = MAX_CACHE_BYTES_PER_GIG * div_u64(size, SZ_1G);
1670
1671         /*
1672          * we want to account for 1 more bitmap than what we have so we can make
1673          * sure we don't go over our overall goal of MAX_CACHE_BYTES_PER_GIG as
1674          * we add more bitmaps.
1675          */
1676         bitmap_bytes = (ctl->total_bitmaps + 1) * ctl->unit;
1677
1678         if (bitmap_bytes >= max_bytes) {
1679                 ctl->extents_thresh = 0;
1680                 return;
1681         }
1682
1683         /*
1684          * we want the extent entry threshold to always be at most 1/2 the max
1685          * bytes we can have, or whatever is less than that.
1686          */
1687         extent_bytes = max_bytes - bitmap_bytes;
1688         extent_bytes = min_t(u64, extent_bytes, max_bytes >> 1);
1689
1690         ctl->extents_thresh =
1691                 div_u64(extent_bytes, sizeof(struct btrfs_free_space));
1692 }
1693
1694 static inline void __bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1695                                        struct btrfs_free_space *info,
1696                                        u64 offset, u64 bytes)
1697 {
1698         unsigned long start, count;
1699
1700         start = offset_to_bit(info->offset, ctl->unit, offset);
1701         count = bytes_to_bits(bytes, ctl->unit);
1702         ASSERT(start + count <= BITS_PER_BITMAP);
1703
1704         bitmap_clear(info->bitmap, start, count);
1705
1706         info->bytes -= bytes;
1707         if (info->max_extent_size > ctl->unit)
1708                 info->max_extent_size = 0;
1709 }
1710
1711 static void bitmap_clear_bits(struct btrfs_free_space_ctl *ctl,
1712                               struct btrfs_free_space *info, u64 offset,
1713                               u64 bytes)
1714 {
1715         __bitmap_clear_bits(ctl, info, offset, bytes);
1716         ctl->free_space -= bytes;
1717 }
1718
1719 static void bitmap_set_bits(struct btrfs_free_space_ctl *ctl,
1720                             struct btrfs_free_space *info, u64 offset,
1721                             u64 bytes)
1722 {
1723         unsigned long start, count;
1724
1725         start = offset_to_bit(info->offset, ctl->unit, offset);
1726         count = bytes_to_bits(bytes, ctl->unit);
1727         ASSERT(start + count <= BITS_PER_BITMAP);
1728
1729         bitmap_set(info->bitmap, start, count);
1730
1731         info->bytes += bytes;
1732         ctl->free_space += bytes;
1733 }
1734
1735 /*
1736  * If we can not find suitable extent, we will use bytes to record
1737  * the size of the max extent.
1738  */
1739 static int search_bitmap(struct btrfs_free_space_ctl *ctl,
1740                          struct btrfs_free_space *bitmap_info, u64 *offset,
1741                          u64 *bytes, bool for_alloc)
1742 {
1743         unsigned long found_bits = 0;
1744         unsigned long max_bits = 0;
1745         unsigned long bits, i;
1746         unsigned long next_zero;
1747         unsigned long extent_bits;
1748
1749         /*
1750          * Skip searching the bitmap if we don't have a contiguous section that
1751          * is large enough for this allocation.
1752          */
1753         if (for_alloc &&
1754             bitmap_info->max_extent_size &&
1755             bitmap_info->max_extent_size < *bytes) {
1756                 *bytes = bitmap_info->max_extent_size;
1757                 return -1;
1758         }
1759
1760         i = offset_to_bit(bitmap_info->offset, ctl->unit,
1761                           max_t(u64, *offset, bitmap_info->offset));
1762         bits = bytes_to_bits(*bytes, ctl->unit);
1763
1764         for_each_set_bit_from(i, bitmap_info->bitmap, BITS_PER_BITMAP) {
1765                 if (for_alloc && bits == 1) {
1766                         found_bits = 1;
1767                         break;
1768                 }
1769                 next_zero = find_next_zero_bit(bitmap_info->bitmap,
1770                                                BITS_PER_BITMAP, i);
1771                 extent_bits = next_zero - i;
1772                 if (extent_bits >= bits) {
1773                         found_bits = extent_bits;
1774                         break;
1775                 } else if (extent_bits > max_bits) {
1776                         max_bits = extent_bits;
1777                 }
1778                 i = next_zero;
1779         }
1780
1781         if (found_bits) {
1782                 *offset = (u64)(i * ctl->unit) + bitmap_info->offset;
1783                 *bytes = (u64)(found_bits) * ctl->unit;
1784                 return 0;
1785         }
1786
1787         *bytes = (u64)(max_bits) * ctl->unit;
1788         bitmap_info->max_extent_size = *bytes;
1789         return -1;
1790 }
1791
1792 static inline u64 get_max_extent_size(struct btrfs_free_space *entry)
1793 {
1794         if (entry->bitmap)
1795                 return entry->max_extent_size;
1796         return entry->bytes;
1797 }
1798
1799 /* Cache the size of the max extent in bytes */
1800 static struct btrfs_free_space *
1801 find_free_space(struct btrfs_free_space_ctl *ctl, u64 *offset, u64 *bytes,
1802                 unsigned long align, u64 *max_extent_size)
1803 {
1804         struct btrfs_free_space *entry;
1805         struct rb_node *node;
1806         u64 tmp;
1807         u64 align_off;
1808         int ret;
1809
1810         if (!ctl->free_space_offset.rb_node)
1811                 goto out;
1812
1813         entry = tree_search_offset(ctl, offset_to_bitmap(ctl, *offset), 0, 1);
1814         if (!entry)
1815                 goto out;
1816
1817         for (node = &entry->offset_index; node; node = rb_next(node)) {
1818                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
1819                 if (entry->bytes < *bytes) {
1820                         *max_extent_size = max(get_max_extent_size(entry),
1821                                                *max_extent_size);
1822                         continue;
1823                 }
1824
1825                 /* make sure the space returned is big enough
1826                  * to match our requested alignment
1827                  */
1828                 if (*bytes >= align) {
1829                         tmp = entry->offset - ctl->start + align - 1;
1830                         tmp = div64_u64(tmp, align);
1831                         tmp = tmp * align + ctl->start;
1832                         align_off = tmp - entry->offset;
1833                 } else {
1834                         align_off = 0;
1835                         tmp = entry->offset;
1836                 }
1837
1838                 if (entry->bytes < *bytes + align_off) {
1839                         *max_extent_size = max(get_max_extent_size(entry),
1840                                                *max_extent_size);
1841                         continue;
1842                 }
1843
1844                 if (entry->bitmap) {
1845                         u64 size = *bytes;
1846
1847                         ret = search_bitmap(ctl, entry, &tmp, &size, true);
1848                         if (!ret) {
1849                                 *offset = tmp;
1850                                 *bytes = size;
1851                                 return entry;
1852                         } else {
1853                                 *max_extent_size =
1854                                         max(get_max_extent_size(entry),
1855                                             *max_extent_size);
1856                         }
1857                         continue;
1858                 }
1859
1860                 *offset = tmp;
1861                 *bytes = entry->bytes - align_off;
1862                 return entry;
1863         }
1864 out:
1865         return NULL;
1866 }
1867
1868 static void add_new_bitmap(struct btrfs_free_space_ctl *ctl,
1869                            struct btrfs_free_space *info, u64 offset)
1870 {
1871         info->offset = offset_to_bitmap(ctl, offset);
1872         info->bytes = 0;
1873         INIT_LIST_HEAD(&info->list);
1874         link_free_space(ctl, info);
1875         ctl->total_bitmaps++;
1876
1877         ctl->op->recalc_thresholds(ctl);
1878 }
1879
1880 static void free_bitmap(struct btrfs_free_space_ctl *ctl,
1881                         struct btrfs_free_space *bitmap_info)
1882 {
1883         unlink_free_space(ctl, bitmap_info);
1884         kfree(bitmap_info->bitmap);
1885         kmem_cache_free(btrfs_free_space_cachep, bitmap_info);
1886         ctl->total_bitmaps--;
1887         ctl->op->recalc_thresholds(ctl);
1888 }
1889
1890 static noinline int remove_from_bitmap(struct btrfs_free_space_ctl *ctl,
1891                               struct btrfs_free_space *bitmap_info,
1892                               u64 *offset, u64 *bytes)
1893 {
1894         u64 end;
1895         u64 search_start, search_bytes;
1896         int ret;
1897
1898 again:
1899         end = bitmap_info->offset + (u64)(BITS_PER_BITMAP * ctl->unit) - 1;
1900
1901         /*
1902          * We need to search for bits in this bitmap.  We could only cover some
1903          * of the extent in this bitmap thanks to how we add space, so we need
1904          * to search for as much as it as we can and clear that amount, and then
1905          * go searching for the next bit.
1906          */
1907         search_start = *offset;
1908         search_bytes = ctl->unit;
1909         search_bytes = min(search_bytes, end - search_start + 1);
1910         ret = search_bitmap(ctl, bitmap_info, &search_start, &search_bytes,
1911                             false);
1912         if (ret < 0 || search_start != *offset)
1913                 return -EINVAL;
1914
1915         /* We may have found more bits than what we need */
1916         search_bytes = min(search_bytes, *bytes);
1917
1918         /* Cannot clear past the end of the bitmap */
1919         search_bytes = min(search_bytes, end - search_start + 1);
1920
1921         bitmap_clear_bits(ctl, bitmap_info, search_start, search_bytes);
1922         *offset += search_bytes;
1923         *bytes -= search_bytes;
1924
1925         if (*bytes) {
1926                 struct rb_node *next = rb_next(&bitmap_info->offset_index);
1927                 if (!bitmap_info->bytes)
1928                         free_bitmap(ctl, bitmap_info);
1929
1930                 /*
1931                  * no entry after this bitmap, but we still have bytes to
1932                  * remove, so something has gone wrong.
1933                  */
1934                 if (!next)
1935                         return -EINVAL;
1936
1937                 bitmap_info = rb_entry(next, struct btrfs_free_space,
1938                                        offset_index);
1939
1940                 /*
1941                  * if the next entry isn't a bitmap we need to return to let the
1942                  * extent stuff do its work.
1943                  */
1944                 if (!bitmap_info->bitmap)
1945                         return -EAGAIN;
1946
1947                 /*
1948                  * Ok the next item is a bitmap, but it may not actually hold
1949                  * the information for the rest of this free space stuff, so
1950                  * look for it, and if we don't find it return so we can try
1951                  * everything over again.
1952                  */
1953                 search_start = *offset;
1954                 search_bytes = ctl->unit;
1955                 ret = search_bitmap(ctl, bitmap_info, &search_start,
1956                                     &search_bytes, false);
1957                 if (ret < 0 || search_start != *offset)
1958                         return -EAGAIN;
1959
1960                 goto again;
1961         } else if (!bitmap_info->bytes)
1962                 free_bitmap(ctl, bitmap_info);
1963
1964         return 0;
1965 }
1966
1967 static u64 add_bytes_to_bitmap(struct btrfs_free_space_ctl *ctl,
1968                                struct btrfs_free_space *info, u64 offset,
1969                                u64 bytes)
1970 {
1971         u64 bytes_to_set = 0;
1972         u64 end;
1973
1974         end = info->offset + (u64)(BITS_PER_BITMAP * ctl->unit);
1975
1976         bytes_to_set = min(end - offset, bytes);
1977
1978         bitmap_set_bits(ctl, info, offset, bytes_to_set);
1979
1980         /*
1981          * We set some bytes, we have no idea what the max extent size is
1982          * anymore.
1983          */
1984         info->max_extent_size = 0;
1985
1986         return bytes_to_set;
1987
1988 }
1989
1990 static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
1991                       struct btrfs_free_space *info)
1992 {
1993         struct btrfs_block_group_cache *block_group = ctl->private;
1994         struct btrfs_fs_info *fs_info = block_group->fs_info;
1995         bool forced = false;
1996
1997 #ifdef CONFIG_BTRFS_DEBUG
1998         if (btrfs_should_fragment_free_space(block_group))
1999                 forced = true;
2000 #endif
2001
2002         /*
2003          * If we are below the extents threshold then we can add this as an
2004          * extent, and don't have to deal with the bitmap
2005          */
2006         if (!forced && ctl->free_extents < ctl->extents_thresh) {
2007                 /*
2008                  * If this block group has some small extents we don't want to
2009                  * use up all of our free slots in the cache with them, we want
2010                  * to reserve them to larger extents, however if we have plenty
2011                  * of cache left then go ahead an dadd them, no sense in adding
2012                  * the overhead of a bitmap if we don't have to.
2013                  */
2014                 if (info->bytes <= fs_info->sectorsize * 4) {
2015                         if (ctl->free_extents * 2 <= ctl->extents_thresh)
2016                                 return false;
2017                 } else {
2018                         return false;
2019                 }
2020         }
2021
2022         /*
2023          * The original block groups from mkfs can be really small, like 8
2024          * megabytes, so don't bother with a bitmap for those entries.  However
2025          * some block groups can be smaller than what a bitmap would cover but
2026          * are still large enough that they could overflow the 32k memory limit,
2027          * so allow those block groups to still be allowed to have a bitmap
2028          * entry.
2029          */
2030         if (((BITS_PER_BITMAP * ctl->unit) >> 1) > block_group->key.offset)
2031                 return false;
2032
2033         return true;
2034 }
2035
2036 static const struct btrfs_free_space_op free_space_op = {
2037         .recalc_thresholds      = recalculate_thresholds,
2038         .use_bitmap             = use_bitmap,
2039 };
2040
2041 static int insert_into_bitmap(struct btrfs_free_space_ctl *ctl,
2042                               struct btrfs_free_space *info)
2043 {
2044         struct btrfs_free_space *bitmap_info;
2045         struct btrfs_block_group_cache *block_group = NULL;
2046         int added = 0;
2047         u64 bytes, offset, bytes_added;
2048         int ret;
2049
2050         bytes = info->bytes;
2051         offset = info->offset;
2052
2053         if (!ctl->op->use_bitmap(ctl, info))
2054                 return 0;
2055
2056         if (ctl->op == &free_space_op)
2057                 block_group = ctl->private;
2058 again:
2059         /*
2060          * Since we link bitmaps right into the cluster we need to see if we
2061          * have a cluster here, and if so and it has our bitmap we need to add
2062          * the free space to that bitmap.
2063          */
2064         if (block_group && !list_empty(&block_group->cluster_list)) {
2065                 struct btrfs_free_cluster *cluster;
2066                 struct rb_node *node;
2067                 struct btrfs_free_space *entry;
2068
2069                 cluster = list_entry(block_group->cluster_list.next,
2070                                      struct btrfs_free_cluster,
2071                                      block_group_list);
2072                 spin_lock(&cluster->lock);
2073                 node = rb_first(&cluster->root);
2074                 if (!node) {
2075                         spin_unlock(&cluster->lock);
2076                         goto no_cluster_bitmap;
2077                 }
2078
2079                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2080                 if (!entry->bitmap) {
2081                         spin_unlock(&cluster->lock);
2082                         goto no_cluster_bitmap;
2083                 }
2084
2085                 if (entry->offset == offset_to_bitmap(ctl, offset)) {
2086                         bytes_added = add_bytes_to_bitmap(ctl, entry,
2087                                                           offset, bytes);
2088                         bytes -= bytes_added;
2089                         offset += bytes_added;
2090                 }
2091                 spin_unlock(&cluster->lock);
2092                 if (!bytes) {
2093                         ret = 1;
2094                         goto out;
2095                 }
2096         }
2097
2098 no_cluster_bitmap:
2099         bitmap_info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2100                                          1, 0);
2101         if (!bitmap_info) {
2102                 ASSERT(added == 0);
2103                 goto new_bitmap;
2104         }
2105
2106         bytes_added = add_bytes_to_bitmap(ctl, bitmap_info, offset, bytes);
2107         bytes -= bytes_added;
2108         offset += bytes_added;
2109         added = 0;
2110
2111         if (!bytes) {
2112                 ret = 1;
2113                 goto out;
2114         } else
2115                 goto again;
2116
2117 new_bitmap:
2118         if (info && info->bitmap) {
2119                 add_new_bitmap(ctl, info, offset);
2120                 added = 1;
2121                 info = NULL;
2122                 goto again;
2123         } else {
2124                 spin_unlock(&ctl->tree_lock);
2125
2126                 /* no pre-allocated info, allocate a new one */
2127                 if (!info) {
2128                         info = kmem_cache_zalloc(btrfs_free_space_cachep,
2129                                                  GFP_NOFS);
2130                         if (!info) {
2131                                 spin_lock(&ctl->tree_lock);
2132                                 ret = -ENOMEM;
2133                                 goto out;
2134                         }
2135                 }
2136
2137                 /* allocate the bitmap */
2138                 info->bitmap = kzalloc(PAGE_SIZE, GFP_NOFS);
2139                 spin_lock(&ctl->tree_lock);
2140                 if (!info->bitmap) {
2141                         ret = -ENOMEM;
2142                         goto out;
2143                 }
2144                 goto again;
2145         }
2146
2147 out:
2148         if (info) {
2149                 kfree(info->bitmap);
2150                 kmem_cache_free(btrfs_free_space_cachep, info);
2151         }
2152
2153         return ret;
2154 }
2155
2156 static bool try_merge_free_space(struct btrfs_free_space_ctl *ctl,
2157                           struct btrfs_free_space *info, bool update_stat)
2158 {
2159         struct btrfs_free_space *left_info;
2160         struct btrfs_free_space *right_info;
2161         bool merged = false;
2162         u64 offset = info->offset;
2163         u64 bytes = info->bytes;
2164
2165         /*
2166          * first we want to see if there is free space adjacent to the range we
2167          * are adding, if there is remove that struct and add a new one to
2168          * cover the entire range
2169          */
2170         right_info = tree_search_offset(ctl, offset + bytes, 0, 0);
2171         if (right_info && rb_prev(&right_info->offset_index))
2172                 left_info = rb_entry(rb_prev(&right_info->offset_index),
2173                                      struct btrfs_free_space, offset_index);
2174         else
2175                 left_info = tree_search_offset(ctl, offset - 1, 0, 0);
2176
2177         if (right_info && !right_info->bitmap) {
2178                 if (update_stat)
2179                         unlink_free_space(ctl, right_info);
2180                 else
2181                         __unlink_free_space(ctl, right_info);
2182                 info->bytes += right_info->bytes;
2183                 kmem_cache_free(btrfs_free_space_cachep, right_info);
2184                 merged = true;
2185         }
2186
2187         if (left_info && !left_info->bitmap &&
2188             left_info->offset + left_info->bytes == offset) {
2189                 if (update_stat)
2190                         unlink_free_space(ctl, left_info);
2191                 else
2192                         __unlink_free_space(ctl, left_info);
2193                 info->offset = left_info->offset;
2194                 info->bytes += left_info->bytes;
2195                 kmem_cache_free(btrfs_free_space_cachep, left_info);
2196                 merged = true;
2197         }
2198
2199         return merged;
2200 }
2201
2202 static bool steal_from_bitmap_to_end(struct btrfs_free_space_ctl *ctl,
2203                                      struct btrfs_free_space *info,
2204                                      bool update_stat)
2205 {
2206         struct btrfs_free_space *bitmap;
2207         unsigned long i;
2208         unsigned long j;
2209         const u64 end = info->offset + info->bytes;
2210         const u64 bitmap_offset = offset_to_bitmap(ctl, end);
2211         u64 bytes;
2212
2213         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2214         if (!bitmap)
2215                 return false;
2216
2217         i = offset_to_bit(bitmap->offset, ctl->unit, end);
2218         j = find_next_zero_bit(bitmap->bitmap, BITS_PER_BITMAP, i);
2219         if (j == i)
2220                 return false;
2221         bytes = (j - i) * ctl->unit;
2222         info->bytes += bytes;
2223
2224         if (update_stat)
2225                 bitmap_clear_bits(ctl, bitmap, end, bytes);
2226         else
2227                 __bitmap_clear_bits(ctl, bitmap, end, bytes);
2228
2229         if (!bitmap->bytes)
2230                 free_bitmap(ctl, bitmap);
2231
2232         return true;
2233 }
2234
2235 static bool steal_from_bitmap_to_front(struct btrfs_free_space_ctl *ctl,
2236                                        struct btrfs_free_space *info,
2237                                        bool update_stat)
2238 {
2239         struct btrfs_free_space *bitmap;
2240         u64 bitmap_offset;
2241         unsigned long i;
2242         unsigned long j;
2243         unsigned long prev_j;
2244         u64 bytes;
2245
2246         bitmap_offset = offset_to_bitmap(ctl, info->offset);
2247         /* If we're on a boundary, try the previous logical bitmap. */
2248         if (bitmap_offset == info->offset) {
2249                 if (info->offset == 0)
2250                         return false;
2251                 bitmap_offset = offset_to_bitmap(ctl, info->offset - 1);
2252         }
2253
2254         bitmap = tree_search_offset(ctl, bitmap_offset, 1, 0);
2255         if (!bitmap)
2256                 return false;
2257
2258         i = offset_to_bit(bitmap->offset, ctl->unit, info->offset) - 1;
2259         j = 0;
2260         prev_j = (unsigned long)-1;
2261         for_each_clear_bit_from(j, bitmap->bitmap, BITS_PER_BITMAP) {
2262                 if (j > i)
2263                         break;
2264                 prev_j = j;
2265         }
2266         if (prev_j == i)
2267                 return false;
2268
2269         if (prev_j == (unsigned long)-1)
2270                 bytes = (i + 1) * ctl->unit;
2271         else
2272                 bytes = (i - prev_j) * ctl->unit;
2273
2274         info->offset -= bytes;
2275         info->bytes += bytes;
2276
2277         if (update_stat)
2278                 bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2279         else
2280                 __bitmap_clear_bits(ctl, bitmap, info->offset, bytes);
2281
2282         if (!bitmap->bytes)
2283                 free_bitmap(ctl, bitmap);
2284
2285         return true;
2286 }
2287
2288 /*
2289  * We prefer always to allocate from extent entries, both for clustered and
2290  * non-clustered allocation requests. So when attempting to add a new extent
2291  * entry, try to see if there's adjacent free space in bitmap entries, and if
2292  * there is, migrate that space from the bitmaps to the extent.
2293  * Like this we get better chances of satisfying space allocation requests
2294  * because we attempt to satisfy them based on a single cache entry, and never
2295  * on 2 or more entries - even if the entries represent a contiguous free space
2296  * region (e.g. 1 extent entry + 1 bitmap entry starting where the extent entry
2297  * ends).
2298  */
2299 static void steal_from_bitmap(struct btrfs_free_space_ctl *ctl,
2300                               struct btrfs_free_space *info,
2301                               bool update_stat)
2302 {
2303         /*
2304          * Only work with disconnected entries, as we can change their offset,
2305          * and must be extent entries.
2306          */
2307         ASSERT(!info->bitmap);
2308         ASSERT(RB_EMPTY_NODE(&info->offset_index));
2309
2310         if (ctl->total_bitmaps > 0) {
2311                 bool stole_end;
2312                 bool stole_front = false;
2313
2314                 stole_end = steal_from_bitmap_to_end(ctl, info, update_stat);
2315                 if (ctl->total_bitmaps > 0)
2316                         stole_front = steal_from_bitmap_to_front(ctl, info,
2317                                                                  update_stat);
2318
2319                 if (stole_end || stole_front)
2320                         try_merge_free_space(ctl, info, update_stat);
2321         }
2322 }
2323
2324 int __btrfs_add_free_space(struct btrfs_fs_info *fs_info,
2325                            struct btrfs_free_space_ctl *ctl,
2326                            u64 offset, u64 bytes)
2327 {
2328         struct btrfs_free_space *info;
2329         int ret = 0;
2330
2331         info = kmem_cache_zalloc(btrfs_free_space_cachep, GFP_NOFS);
2332         if (!info)
2333                 return -ENOMEM;
2334
2335         info->offset = offset;
2336         info->bytes = bytes;
2337         RB_CLEAR_NODE(&info->offset_index);
2338
2339         spin_lock(&ctl->tree_lock);
2340
2341         if (try_merge_free_space(ctl, info, true))
2342                 goto link;
2343
2344         /*
2345          * There was no extent directly to the left or right of this new
2346          * extent then we know we're going to have to allocate a new extent, so
2347          * before we do that see if we need to drop this into a bitmap
2348          */
2349         ret = insert_into_bitmap(ctl, info);
2350         if (ret < 0) {
2351                 goto out;
2352         } else if (ret) {
2353                 ret = 0;
2354                 goto out;
2355         }
2356 link:
2357         /*
2358          * Only steal free space from adjacent bitmaps if we're sure we're not
2359          * going to add the new free space to existing bitmap entries - because
2360          * that would mean unnecessary work that would be reverted. Therefore
2361          * attempt to steal space from bitmaps if we're adding an extent entry.
2362          */
2363         steal_from_bitmap(ctl, info, true);
2364
2365         ret = link_free_space(ctl, info);
2366         if (ret)
2367                 kmem_cache_free(btrfs_free_space_cachep, info);
2368 out:
2369         spin_unlock(&ctl->tree_lock);
2370
2371         if (ret) {
2372                 btrfs_crit(fs_info, "unable to add free space :%d", ret);
2373                 ASSERT(ret != -EEXIST);
2374         }
2375
2376         return ret;
2377 }
2378
2379 int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
2380                             u64 offset, u64 bytes)
2381 {
2382         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2383         struct btrfs_free_space *info;
2384         int ret;
2385         bool re_search = false;
2386
2387         spin_lock(&ctl->tree_lock);
2388
2389 again:
2390         ret = 0;
2391         if (!bytes)
2392                 goto out_lock;
2393
2394         info = tree_search_offset(ctl, offset, 0, 0);
2395         if (!info) {
2396                 /*
2397                  * oops didn't find an extent that matched the space we wanted
2398                  * to remove, look for a bitmap instead
2399                  */
2400                 info = tree_search_offset(ctl, offset_to_bitmap(ctl, offset),
2401                                           1, 0);
2402                 if (!info) {
2403                         /*
2404                          * If we found a partial bit of our free space in a
2405                          * bitmap but then couldn't find the other part this may
2406                          * be a problem, so WARN about it.
2407                          */
2408                         WARN_ON(re_search);
2409                         goto out_lock;
2410                 }
2411         }
2412
2413         re_search = false;
2414         if (!info->bitmap) {
2415                 unlink_free_space(ctl, info);
2416                 if (offset == info->offset) {
2417                         u64 to_free = min(bytes, info->bytes);
2418
2419                         info->bytes -= to_free;
2420                         info->offset += to_free;
2421                         if (info->bytes) {
2422                                 ret = link_free_space(ctl, info);
2423                                 WARN_ON(ret);
2424                         } else {
2425                                 kmem_cache_free(btrfs_free_space_cachep, info);
2426                         }
2427
2428                         offset += to_free;
2429                         bytes -= to_free;
2430                         goto again;
2431                 } else {
2432                         u64 old_end = info->bytes + info->offset;
2433
2434                         info->bytes = offset - info->offset;
2435                         ret = link_free_space(ctl, info);
2436                         WARN_ON(ret);
2437                         if (ret)
2438                                 goto out_lock;
2439
2440                         /* Not enough bytes in this entry to satisfy us */
2441                         if (old_end < offset + bytes) {
2442                                 bytes -= old_end - offset;
2443                                 offset = old_end;
2444                                 goto again;
2445                         } else if (old_end == offset + bytes) {
2446                                 /* all done */
2447                                 goto out_lock;
2448                         }
2449                         spin_unlock(&ctl->tree_lock);
2450
2451                         ret = btrfs_add_free_space(block_group, offset + bytes,
2452                                                    old_end - (offset + bytes));
2453                         WARN_ON(ret);
2454                         goto out;
2455                 }
2456         }
2457
2458         ret = remove_from_bitmap(ctl, info, &offset, &bytes);
2459         if (ret == -EAGAIN) {
2460                 re_search = true;
2461                 goto again;
2462         }
2463 out_lock:
2464         spin_unlock(&ctl->tree_lock);
2465 out:
2466         return ret;
2467 }
2468
2469 void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
2470                            u64 bytes)
2471 {
2472         struct btrfs_fs_info *fs_info = block_group->fs_info;
2473         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2474         struct btrfs_free_space *info;
2475         struct rb_node *n;
2476         int count = 0;
2477
2478         spin_lock(&ctl->tree_lock);
2479         for (n = rb_first(&ctl->free_space_offset); n; n = rb_next(n)) {
2480                 info = rb_entry(n, struct btrfs_free_space, offset_index);
2481                 if (info->bytes >= bytes && !block_group->ro)
2482                         count++;
2483                 btrfs_crit(fs_info, "entry offset %llu, bytes %llu, bitmap %s",
2484                            info->offset, info->bytes,
2485                        (info->bitmap) ? "yes" : "no");
2486         }
2487         spin_unlock(&ctl->tree_lock);
2488         btrfs_info(fs_info, "block group has cluster?: %s",
2489                list_empty(&block_group->cluster_list) ? "no" : "yes");
2490         btrfs_info(fs_info,
2491                    "%d blocks of free space at or bigger than bytes is", count);
2492 }
2493
2494 void btrfs_init_free_space_ctl(struct btrfs_block_group_cache *block_group)
2495 {
2496         struct btrfs_fs_info *fs_info = block_group->fs_info;
2497         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2498
2499         spin_lock_init(&ctl->tree_lock);
2500         ctl->unit = fs_info->sectorsize;
2501         ctl->start = block_group->key.objectid;
2502         ctl->private = block_group;
2503         ctl->op = &free_space_op;
2504         INIT_LIST_HEAD(&ctl->trimming_ranges);
2505         mutex_init(&ctl->cache_writeout_mutex);
2506
2507         /*
2508          * we only want to have 32k of ram per block group for keeping
2509          * track of free space, and if we pass 1/2 of that we want to
2510          * start converting things over to using bitmaps
2511          */
2512         ctl->extents_thresh = (SZ_32K / 2) / sizeof(struct btrfs_free_space);
2513 }
2514
2515 /*
2516  * for a given cluster, put all of its extents back into the free
2517  * space cache.  If the block group passed doesn't match the block group
2518  * pointed to by the cluster, someone else raced in and freed the
2519  * cluster already.  In that case, we just return without changing anything
2520  */
2521 static int
2522 __btrfs_return_cluster_to_free_space(
2523                              struct btrfs_block_group_cache *block_group,
2524                              struct btrfs_free_cluster *cluster)
2525 {
2526         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2527         struct btrfs_free_space *entry;
2528         struct rb_node *node;
2529
2530         spin_lock(&cluster->lock);
2531         if (cluster->block_group != block_group)
2532                 goto out;
2533
2534         cluster->block_group = NULL;
2535         cluster->window_start = 0;
2536         list_del_init(&cluster->block_group_list);
2537
2538         node = rb_first(&cluster->root);
2539         while (node) {
2540                 bool bitmap;
2541
2542                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2543                 node = rb_next(&entry->offset_index);
2544                 rb_erase(&entry->offset_index, &cluster->root);
2545                 RB_CLEAR_NODE(&entry->offset_index);
2546
2547                 bitmap = (entry->bitmap != NULL);
2548                 if (!bitmap) {
2549                         try_merge_free_space(ctl, entry, false);
2550                         steal_from_bitmap(ctl, entry, false);
2551                 }
2552                 tree_insert_offset(&ctl->free_space_offset,
2553                                    entry->offset, &entry->offset_index, bitmap);
2554         }
2555         cluster->root = RB_ROOT;
2556
2557 out:
2558         spin_unlock(&cluster->lock);
2559         btrfs_put_block_group(block_group);
2560         return 0;
2561 }
2562
2563 static void __btrfs_remove_free_space_cache_locked(
2564                                 struct btrfs_free_space_ctl *ctl)
2565 {
2566         struct btrfs_free_space *info;
2567         struct rb_node *node;
2568
2569         while ((node = rb_last(&ctl->free_space_offset)) != NULL) {
2570                 info = rb_entry(node, struct btrfs_free_space, offset_index);
2571                 if (!info->bitmap) {
2572                         unlink_free_space(ctl, info);
2573                         kmem_cache_free(btrfs_free_space_cachep, info);
2574                 } else {
2575                         free_bitmap(ctl, info);
2576                 }
2577
2578                 cond_resched_lock(&ctl->tree_lock);
2579         }
2580 }
2581
2582 void __btrfs_remove_free_space_cache(struct btrfs_free_space_ctl *ctl)
2583 {
2584         spin_lock(&ctl->tree_lock);
2585         __btrfs_remove_free_space_cache_locked(ctl);
2586         spin_unlock(&ctl->tree_lock);
2587 }
2588
2589 void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
2590 {
2591         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2592         struct btrfs_free_cluster *cluster;
2593         struct list_head *head;
2594
2595         spin_lock(&ctl->tree_lock);
2596         while ((head = block_group->cluster_list.next) !=
2597                &block_group->cluster_list) {
2598                 cluster = list_entry(head, struct btrfs_free_cluster,
2599                                      block_group_list);
2600
2601                 WARN_ON(cluster->block_group != block_group);
2602                 __btrfs_return_cluster_to_free_space(block_group, cluster);
2603
2604                 cond_resched_lock(&ctl->tree_lock);
2605         }
2606         __btrfs_remove_free_space_cache_locked(ctl);
2607         spin_unlock(&ctl->tree_lock);
2608
2609 }
2610
2611 u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
2612                                u64 offset, u64 bytes, u64 empty_size,
2613                                u64 *max_extent_size)
2614 {
2615         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2616         struct btrfs_free_space *entry = NULL;
2617         u64 bytes_search = bytes + empty_size;
2618         u64 ret = 0;
2619         u64 align_gap = 0;
2620         u64 align_gap_len = 0;
2621
2622         spin_lock(&ctl->tree_lock);
2623         entry = find_free_space(ctl, &offset, &bytes_search,
2624                                 block_group->full_stripe_len, max_extent_size);
2625         if (!entry)
2626                 goto out;
2627
2628         ret = offset;
2629         if (entry->bitmap) {
2630                 bitmap_clear_bits(ctl, entry, offset, bytes);
2631                 if (!entry->bytes)
2632                         free_bitmap(ctl, entry);
2633         } else {
2634                 unlink_free_space(ctl, entry);
2635                 align_gap_len = offset - entry->offset;
2636                 align_gap = entry->offset;
2637
2638                 entry->offset = offset + bytes;
2639                 WARN_ON(entry->bytes < bytes + align_gap_len);
2640
2641                 entry->bytes -= bytes + align_gap_len;
2642                 if (!entry->bytes)
2643                         kmem_cache_free(btrfs_free_space_cachep, entry);
2644                 else
2645                         link_free_space(ctl, entry);
2646         }
2647 out:
2648         spin_unlock(&ctl->tree_lock);
2649
2650         if (align_gap_len)
2651                 __btrfs_add_free_space(block_group->fs_info, ctl,
2652                                        align_gap, align_gap_len);
2653         return ret;
2654 }
2655
2656 /*
2657  * given a cluster, put all of its extents back into the free space
2658  * cache.  If a block group is passed, this function will only free
2659  * a cluster that belongs to the passed block group.
2660  *
2661  * Otherwise, it'll get a reference on the block group pointed to by the
2662  * cluster and remove the cluster from it.
2663  */
2664 int btrfs_return_cluster_to_free_space(
2665                                struct btrfs_block_group_cache *block_group,
2666                                struct btrfs_free_cluster *cluster)
2667 {
2668         struct btrfs_free_space_ctl *ctl;
2669         int ret;
2670
2671         /* first, get a safe pointer to the block group */
2672         spin_lock(&cluster->lock);
2673         if (!block_group) {
2674                 block_group = cluster->block_group;
2675                 if (!block_group) {
2676                         spin_unlock(&cluster->lock);
2677                         return 0;
2678                 }
2679         } else if (cluster->block_group != block_group) {
2680                 /* someone else has already freed it don't redo their work */
2681                 spin_unlock(&cluster->lock);
2682                 return 0;
2683         }
2684         atomic_inc(&block_group->count);
2685         spin_unlock(&cluster->lock);
2686
2687         ctl = block_group->free_space_ctl;
2688
2689         /* now return any extents the cluster had on it */
2690         spin_lock(&ctl->tree_lock);
2691         ret = __btrfs_return_cluster_to_free_space(block_group, cluster);
2692         spin_unlock(&ctl->tree_lock);
2693
2694         /* finally drop our ref */
2695         btrfs_put_block_group(block_group);
2696         return ret;
2697 }
2698
2699 static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
2700                                    struct btrfs_free_cluster *cluster,
2701                                    struct btrfs_free_space *entry,
2702                                    u64 bytes, u64 min_start,
2703                                    u64 *max_extent_size)
2704 {
2705         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2706         int err;
2707         u64 search_start = cluster->window_start;
2708         u64 search_bytes = bytes;
2709         u64 ret = 0;
2710
2711         search_start = min_start;
2712         search_bytes = bytes;
2713
2714         err = search_bitmap(ctl, entry, &search_start, &search_bytes, true);
2715         if (err) {
2716                 *max_extent_size = max(get_max_extent_size(entry),
2717                                        *max_extent_size);
2718                 return 0;
2719         }
2720
2721         ret = search_start;
2722         __bitmap_clear_bits(ctl, entry, ret, bytes);
2723
2724         return ret;
2725 }
2726
2727 /*
2728  * given a cluster, try to allocate 'bytes' from it, returns 0
2729  * if it couldn't find anything suitably large, or a logical disk offset
2730  * if things worked out
2731  */
2732 u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
2733                              struct btrfs_free_cluster *cluster, u64 bytes,
2734                              u64 min_start, u64 *max_extent_size)
2735 {
2736         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2737         struct btrfs_free_space *entry = NULL;
2738         struct rb_node *node;
2739         u64 ret = 0;
2740
2741         spin_lock(&cluster->lock);
2742         if (bytes > cluster->max_size)
2743                 goto out;
2744
2745         if (cluster->block_group != block_group)
2746                 goto out;
2747
2748         node = rb_first(&cluster->root);
2749         if (!node)
2750                 goto out;
2751
2752         entry = rb_entry(node, struct btrfs_free_space, offset_index);
2753         while (1) {
2754                 if (entry->bytes < bytes)
2755                         *max_extent_size = max(get_max_extent_size(entry),
2756                                                *max_extent_size);
2757
2758                 if (entry->bytes < bytes ||
2759                     (!entry->bitmap && entry->offset < min_start)) {
2760                         node = rb_next(&entry->offset_index);
2761                         if (!node)
2762                                 break;
2763                         entry = rb_entry(node, struct btrfs_free_space,
2764                                          offset_index);
2765                         continue;
2766                 }
2767
2768                 if (entry->bitmap) {
2769                         ret = btrfs_alloc_from_bitmap(block_group,
2770                                                       cluster, entry, bytes,
2771                                                       cluster->window_start,
2772                                                       max_extent_size);
2773                         if (ret == 0) {
2774                                 node = rb_next(&entry->offset_index);
2775                                 if (!node)
2776                                         break;
2777                                 entry = rb_entry(node, struct btrfs_free_space,
2778                                                  offset_index);
2779                                 continue;
2780                         }
2781                         cluster->window_start += bytes;
2782                 } else {
2783                         ret = entry->offset;
2784
2785                         entry->offset += bytes;
2786                         entry->bytes -= bytes;
2787                 }
2788
2789                 if (entry->bytes == 0)
2790                         rb_erase(&entry->offset_index, &cluster->root);
2791                 break;
2792         }
2793 out:
2794         spin_unlock(&cluster->lock);
2795
2796         if (!ret)
2797                 return 0;
2798
2799         spin_lock(&ctl->tree_lock);
2800
2801         ctl->free_space -= bytes;
2802         if (entry->bytes == 0) {
2803                 ctl->free_extents--;
2804                 if (entry->bitmap) {
2805                         kfree(entry->bitmap);
2806                         ctl->total_bitmaps--;
2807                         ctl->op->recalc_thresholds(ctl);
2808                 }
2809                 kmem_cache_free(btrfs_free_space_cachep, entry);
2810         }
2811
2812         spin_unlock(&ctl->tree_lock);
2813
2814         return ret;
2815 }
2816
2817 static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
2818                                 struct btrfs_free_space *entry,
2819                                 struct btrfs_free_cluster *cluster,
2820                                 u64 offset, u64 bytes,
2821                                 u64 cont1_bytes, u64 min_bytes)
2822 {
2823         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2824         unsigned long next_zero;
2825         unsigned long i;
2826         unsigned long want_bits;
2827         unsigned long min_bits;
2828         unsigned long found_bits;
2829         unsigned long max_bits = 0;
2830         unsigned long start = 0;
2831         unsigned long total_found = 0;
2832         int ret;
2833
2834         i = offset_to_bit(entry->offset, ctl->unit,
2835                           max_t(u64, offset, entry->offset));
2836         want_bits = bytes_to_bits(bytes, ctl->unit);
2837         min_bits = bytes_to_bits(min_bytes, ctl->unit);
2838
2839         /*
2840          * Don't bother looking for a cluster in this bitmap if it's heavily
2841          * fragmented.
2842          */
2843         if (entry->max_extent_size &&
2844             entry->max_extent_size < cont1_bytes)
2845                 return -ENOSPC;
2846 again:
2847         found_bits = 0;
2848         for_each_set_bit_from(i, entry->bitmap, BITS_PER_BITMAP) {
2849                 next_zero = find_next_zero_bit(entry->bitmap,
2850                                                BITS_PER_BITMAP, i);
2851                 if (next_zero - i >= min_bits) {
2852                         found_bits = next_zero - i;
2853                         if (found_bits > max_bits)
2854                                 max_bits = found_bits;
2855                         break;
2856                 }
2857                 if (next_zero - i > max_bits)
2858                         max_bits = next_zero - i;
2859                 i = next_zero;
2860         }
2861
2862         if (!found_bits) {
2863                 entry->max_extent_size = (u64)max_bits * ctl->unit;
2864                 return -ENOSPC;
2865         }
2866
2867         if (!total_found) {
2868                 start = i;
2869                 cluster->max_size = 0;
2870         }
2871
2872         total_found += found_bits;
2873
2874         if (cluster->max_size < found_bits * ctl->unit)
2875                 cluster->max_size = found_bits * ctl->unit;
2876
2877         if (total_found < want_bits || cluster->max_size < cont1_bytes) {
2878                 i = next_zero + 1;
2879                 goto again;
2880         }
2881
2882         cluster->window_start = start * ctl->unit + entry->offset;
2883         rb_erase(&entry->offset_index, &ctl->free_space_offset);
2884         ret = tree_insert_offset(&cluster->root, entry->offset,
2885                                  &entry->offset_index, 1);
2886         ASSERT(!ret); /* -EEXIST; Logic error */
2887
2888         trace_btrfs_setup_cluster(block_group, cluster,
2889                                   total_found * ctl->unit, 1);
2890         return 0;
2891 }
2892
2893 /*
2894  * This searches the block group for just extents to fill the cluster with.
2895  * Try to find a cluster with at least bytes total bytes, at least one
2896  * extent of cont1_bytes, and other clusters of at least min_bytes.
2897  */
2898 static noinline int
2899 setup_cluster_no_bitmap(struct btrfs_block_group_cache *block_group,
2900                         struct btrfs_free_cluster *cluster,
2901                         struct list_head *bitmaps, u64 offset, u64 bytes,
2902                         u64 cont1_bytes, u64 min_bytes)
2903 {
2904         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2905         struct btrfs_free_space *first = NULL;
2906         struct btrfs_free_space *entry = NULL;
2907         struct btrfs_free_space *last;
2908         struct rb_node *node;
2909         u64 window_free;
2910         u64 max_extent;
2911         u64 total_size = 0;
2912
2913         entry = tree_search_offset(ctl, offset, 0, 1);
2914         if (!entry)
2915                 return -ENOSPC;
2916
2917         /*
2918          * We don't want bitmaps, so just move along until we find a normal
2919          * extent entry.
2920          */
2921         while (entry->bitmap || entry->bytes < min_bytes) {
2922                 if (entry->bitmap && list_empty(&entry->list))
2923                         list_add_tail(&entry->list, bitmaps);
2924                 node = rb_next(&entry->offset_index);
2925                 if (!node)
2926                         return -ENOSPC;
2927                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2928         }
2929
2930         window_free = entry->bytes;
2931         max_extent = entry->bytes;
2932         first = entry;
2933         last = entry;
2934
2935         for (node = rb_next(&entry->offset_index); node;
2936              node = rb_next(&entry->offset_index)) {
2937                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2938
2939                 if (entry->bitmap) {
2940                         if (list_empty(&entry->list))
2941                                 list_add_tail(&entry->list, bitmaps);
2942                         continue;
2943                 }
2944
2945                 if (entry->bytes < min_bytes)
2946                         continue;
2947
2948                 last = entry;
2949                 window_free += entry->bytes;
2950                 if (entry->bytes > max_extent)
2951                         max_extent = entry->bytes;
2952         }
2953
2954         if (window_free < bytes || max_extent < cont1_bytes)
2955                 return -ENOSPC;
2956
2957         cluster->window_start = first->offset;
2958
2959         node = &first->offset_index;
2960
2961         /*
2962          * now we've found our entries, pull them out of the free space
2963          * cache and put them into the cluster rbtree
2964          */
2965         do {
2966                 int ret;
2967
2968                 entry = rb_entry(node, struct btrfs_free_space, offset_index);
2969                 node = rb_next(&entry->offset_index);
2970                 if (entry->bitmap || entry->bytes < min_bytes)
2971                         continue;
2972
2973                 rb_erase(&entry->offset_index, &ctl->free_space_offset);
2974                 ret = tree_insert_offset(&cluster->root, entry->offset,
2975                                          &entry->offset_index, 0);
2976                 total_size += entry->bytes;
2977                 ASSERT(!ret); /* -EEXIST; Logic error */
2978         } while (node && entry != last);
2979
2980         cluster->max_size = max_extent;
2981         trace_btrfs_setup_cluster(block_group, cluster, total_size, 0);
2982         return 0;
2983 }
2984
2985 /*
2986  * This specifically looks for bitmaps that may work in the cluster, we assume
2987  * that we have already failed to find extents that will work.
2988  */
2989 static noinline int
2990 setup_cluster_bitmap(struct btrfs_block_group_cache *block_group,
2991                      struct btrfs_free_cluster *cluster,
2992                      struct list_head *bitmaps, u64 offset, u64 bytes,
2993                      u64 cont1_bytes, u64 min_bytes)
2994 {
2995         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
2996         struct btrfs_free_space *entry = NULL;
2997         int ret = -ENOSPC;
2998         u64 bitmap_offset = offset_to_bitmap(ctl, offset);
2999
3000         if (ctl->total_bitmaps == 0)
3001                 return -ENOSPC;
3002
3003         /*
3004          * The bitmap that covers offset won't be in the list unless offset
3005          * is just its start offset.
3006          */
3007         if (!list_empty(bitmaps))
3008                 entry = list_first_entry(bitmaps, struct btrfs_free_space, list);
3009
3010         if (!entry || entry->offset != bitmap_offset) {
3011                 entry = tree_search_offset(ctl, bitmap_offset, 1, 0);
3012                 if (entry && list_empty(&entry->list))
3013                         list_add(&entry->list, bitmaps);
3014         }
3015
3016         list_for_each_entry(entry, bitmaps, list) {
3017                 if (entry->bytes < bytes)
3018                         continue;
3019                 ret = btrfs_bitmap_cluster(block_group, entry, cluster, offset,
3020                                            bytes, cont1_bytes, min_bytes);
3021                 if (!ret)
3022                         return 0;
3023         }
3024
3025         /*
3026          * The bitmaps list has all the bitmaps that record free space
3027          * starting after offset, so no more search is required.
3028          */
3029         return -ENOSPC;
3030 }
3031
3032 /*
3033  * here we try to find a cluster of blocks in a block group.  The goal
3034  * is to find at least bytes+empty_size.
3035  * We might not find them all in one contiguous area.
3036  *
3037  * returns zero and sets up cluster if things worked out, otherwise
3038  * it returns -enospc
3039  */
3040 int btrfs_find_space_cluster(struct btrfs_block_group_cache *block_group,
3041                              struct btrfs_free_cluster *cluster,
3042                              u64 offset, u64 bytes, u64 empty_size)
3043 {
3044         struct btrfs_fs_info *fs_info = block_group->fs_info;
3045         struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3046         struct btrfs_free_space *entry, *tmp;
3047         LIST_HEAD(bitmaps);
3048         u64 min_bytes;
3049         u64 cont1_bytes;
3050         int ret;
3051
3052         /*
3053          * Choose the minimum extent size we'll require for this
3054          * cluster.  For SSD_SPREAD, don't allow any fragmentation.
3055          * For metadata, allow allocates with smaller extents.  For
3056          * data, keep it dense.
3057          */
3058         if (btrfs_test_opt(fs_info, SSD_SPREAD)) {
3059                 cont1_bytes = min_bytes = bytes + empty_size;
3060         } else if (block_group->flags & BTRFS_BLOCK_GROUP_METADATA) {
3061                 cont1_bytes = bytes;
3062                 min_bytes = fs_info->sectorsize;
3063         } else {
3064                 cont1_bytes = max(bytes, (bytes + empty_size) >> 2);
3065                 min_bytes = fs_info->sectorsize;
3066         }
3067
3068         spin_lock(&ctl->tree_lock);
3069
3070         /*
3071          * If we know we don't have enough space to make a cluster don't even
3072          * bother doing all the work to try and find one.
3073          */
3074         if (ctl->free_space < bytes) {
3075                 spin_unlock(&ctl->tree_lock);
3076                 return -ENOSPC;
3077         }
3078
3079         spin_lock(&cluster->lock);
3080
3081         /* someone already found a cluster, hooray */
3082         if (cluster->block_group) {
3083                 ret = 0;
3084                 goto out;
3085         }
3086
3087         trace_btrfs_find_cluster(block_group, offset, bytes, empty_size,
3088                                  min_bytes);
3089
3090         ret = setup_cluster_no_bitmap(block_group, cluster, &bitmaps, offset,
3091                                       bytes + empty_size,
3092                                       cont1_bytes, min_bytes);
3093         if (ret)
3094                 ret = setup_cluster_bitmap(block_group, cluster, &bitmaps,
3095                                            offset, bytes + empty_size,
3096                                            cont1_bytes, min_bytes);
3097
3098         /* Clear our temporary list */
3099         list_for_each_entry_safe(entry, tmp, &bitmaps, list)
3100                 list_del_init(&entry->list);
3101
3102         if (!ret) {
3103                 atomic_inc(&block_group->count);
3104                 list_add_tail(&cluster->block_group_list,
3105                               &block_group->cluster_list);
3106                 cluster->block_group = block_group;
3107         } else {
3108                 trace_btrfs_failed_cluster_setup(block_group);
3109         }
3110 out:
3111         spin_unlock(&cluster->lock);
3112         spin_unlock(&ctl->tree_lock);
3113
3114         return ret;
3115 }
3116
3117 /*
3118  * simple code to zero out a cluster
3119  */
3120 void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
3121 {
3122         spin_lock_init(&cluster->lock);
3123         spin_lock_init(&cluster->refill_lock);
3124         cluster->root = RB_ROOT;
3125         cluster->max_size = 0;
3126         cluster->fragmented = false;
3127         INIT_LIST_HEAD(&cluster->block_group_list);
3128         cluster->block_group = NULL;
3129 }