Fix race in create_empty_buffers() vs __set_page_dirty_buffers()
[sfrench/cifs-2.6.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <asm/div64.h>
24 #include "compat.h"
25 #include "ctree.h"
26 #include "extent_map.h"
27 #include "disk-io.h"
28 #include "transaction.h"
29 #include "print-tree.h"
30 #include "volumes.h"
31 #include "async-thread.h"
32
33 struct map_lookup {
34         u64 type;
35         int io_align;
36         int io_width;
37         int stripe_len;
38         int sector_size;
39         int num_stripes;
40         int sub_stripes;
41         struct btrfs_bio_stripe stripes[];
42 };
43
44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
45                                 struct btrfs_root *root,
46                                 struct btrfs_device *device);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
48
49 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
50                             (sizeof(struct btrfs_bio_stripe) * (n)))
51
52 static DEFINE_MUTEX(uuid_mutex);
53 static LIST_HEAD(fs_uuids);
54
55 void btrfs_lock_volumes(void)
56 {
57         mutex_lock(&uuid_mutex);
58 }
59
60 void btrfs_unlock_volumes(void)
61 {
62         mutex_unlock(&uuid_mutex);
63 }
64
65 static void lock_chunks(struct btrfs_root *root)
66 {
67         mutex_lock(&root->fs_info->chunk_mutex);
68 }
69
70 static void unlock_chunks(struct btrfs_root *root)
71 {
72         mutex_unlock(&root->fs_info->chunk_mutex);
73 }
74
75 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
76 {
77         struct btrfs_device *device;
78         WARN_ON(fs_devices->opened);
79         while (!list_empty(&fs_devices->devices)) {
80                 device = list_entry(fs_devices->devices.next,
81                                     struct btrfs_device, dev_list);
82                 list_del(&device->dev_list);
83                 kfree(device->name);
84                 kfree(device);
85         }
86         kfree(fs_devices);
87 }
88
89 int btrfs_cleanup_fs_uuids(void)
90 {
91         struct btrfs_fs_devices *fs_devices;
92
93         while (!list_empty(&fs_uuids)) {
94                 fs_devices = list_entry(fs_uuids.next,
95                                         struct btrfs_fs_devices, list);
96                 list_del(&fs_devices->list);
97                 free_fs_devices(fs_devices);
98         }
99         return 0;
100 }
101
102 static noinline struct btrfs_device *__find_device(struct list_head *head,
103                                                    u64 devid, u8 *uuid)
104 {
105         struct btrfs_device *dev;
106
107         list_for_each_entry(dev, head, dev_list) {
108                 if (dev->devid == devid &&
109                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
110                         return dev;
111                 }
112         }
113         return NULL;
114 }
115
116 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
117 {
118         struct btrfs_fs_devices *fs_devices;
119
120         list_for_each_entry(fs_devices, &fs_uuids, list) {
121                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
122                         return fs_devices;
123         }
124         return NULL;
125 }
126
127 /*
128  * we try to collect pending bios for a device so we don't get a large
129  * number of procs sending bios down to the same device.  This greatly
130  * improves the schedulers ability to collect and merge the bios.
131  *
132  * But, it also turns into a long list of bios to process and that is sure
133  * to eventually make the worker thread block.  The solution here is to
134  * make some progress and then put this work struct back at the end of
135  * the list if the block device is congested.  This way, multiple devices
136  * can make progress from a single worker thread.
137  */
138 static noinline int run_scheduled_bios(struct btrfs_device *device)
139 {
140         struct bio *pending;
141         struct backing_dev_info *bdi;
142         struct btrfs_fs_info *fs_info;
143         struct bio *tail;
144         struct bio *cur;
145         int again = 0;
146         unsigned long num_run = 0;
147         unsigned long limit;
148
149         bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
150         fs_info = device->dev_root->fs_info;
151         limit = btrfs_async_submit_limit(fs_info);
152         limit = limit * 2 / 3;
153
154 loop:
155         spin_lock(&device->io_lock);
156
157 loop_lock:
158         /* take all the bios off the list at once and process them
159          * later on (without the lock held).  But, remember the
160          * tail and other pointers so the bios can be properly reinserted
161          * into the list if we hit congestion
162          */
163         pending = device->pending_bios;
164         tail = device->pending_bio_tail;
165         WARN_ON(pending && !tail);
166         device->pending_bios = NULL;
167         device->pending_bio_tail = NULL;
168
169         /*
170          * if pending was null this time around, no bios need processing
171          * at all and we can stop.  Otherwise it'll loop back up again
172          * and do an additional check so no bios are missed.
173          *
174          * device->running_pending is used to synchronize with the
175          * schedule_bio code.
176          */
177         if (pending) {
178                 again = 1;
179                 device->running_pending = 1;
180         } else {
181                 again = 0;
182                 device->running_pending = 0;
183         }
184         spin_unlock(&device->io_lock);
185
186         while (pending) {
187                 cur = pending;
188                 pending = pending->bi_next;
189                 cur->bi_next = NULL;
190                 atomic_dec(&fs_info->nr_async_bios);
191
192                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
193                     waitqueue_active(&fs_info->async_submit_wait))
194                         wake_up(&fs_info->async_submit_wait);
195
196                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
197                 bio_get(cur);
198                 submit_bio(cur->bi_rw, cur);
199                 bio_put(cur);
200                 num_run++;
201
202                 /*
203                  * we made progress, there is more work to do and the bdi
204                  * is now congested.  Back off and let other work structs
205                  * run instead
206                  */
207                 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
208                     fs_info->fs_devices->open_devices > 1) {
209                         struct bio *old_head;
210
211                         spin_lock(&device->io_lock);
212
213                         old_head = device->pending_bios;
214                         device->pending_bios = pending;
215                         if (device->pending_bio_tail)
216                                 tail->bi_next = old_head;
217                         else
218                                 device->pending_bio_tail = tail;
219
220                         device->running_pending = 1;
221
222                         spin_unlock(&device->io_lock);
223                         btrfs_requeue_work(&device->work);
224                         goto done;
225                 }
226         }
227         if (again)
228                 goto loop;
229
230         spin_lock(&device->io_lock);
231         if (device->pending_bios)
232                 goto loop_lock;
233         spin_unlock(&device->io_lock);
234 done:
235         return 0;
236 }
237
238 static void pending_bios_fn(struct btrfs_work *work)
239 {
240         struct btrfs_device *device;
241
242         device = container_of(work, struct btrfs_device, work);
243         run_scheduled_bios(device);
244 }
245
246 static noinline int device_list_add(const char *path,
247                            struct btrfs_super_block *disk_super,
248                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
249 {
250         struct btrfs_device *device;
251         struct btrfs_fs_devices *fs_devices;
252         u64 found_transid = btrfs_super_generation(disk_super);
253
254         fs_devices = find_fsid(disk_super->fsid);
255         if (!fs_devices) {
256                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
257                 if (!fs_devices)
258                         return -ENOMEM;
259                 INIT_LIST_HEAD(&fs_devices->devices);
260                 INIT_LIST_HEAD(&fs_devices->alloc_list);
261                 list_add(&fs_devices->list, &fs_uuids);
262                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
263                 fs_devices->latest_devid = devid;
264                 fs_devices->latest_trans = found_transid;
265                 device = NULL;
266         } else {
267                 device = __find_device(&fs_devices->devices, devid,
268                                        disk_super->dev_item.uuid);
269         }
270         if (!device) {
271                 if (fs_devices->opened)
272                         return -EBUSY;
273
274                 device = kzalloc(sizeof(*device), GFP_NOFS);
275                 if (!device) {
276                         /* we can safely leave the fs_devices entry around */
277                         return -ENOMEM;
278                 }
279                 device->devid = devid;
280                 device->work.func = pending_bios_fn;
281                 memcpy(device->uuid, disk_super->dev_item.uuid,
282                        BTRFS_UUID_SIZE);
283                 device->barriers = 1;
284                 spin_lock_init(&device->io_lock);
285                 device->name = kstrdup(path, GFP_NOFS);
286                 if (!device->name) {
287                         kfree(device);
288                         return -ENOMEM;
289                 }
290                 INIT_LIST_HEAD(&device->dev_alloc_list);
291                 list_add(&device->dev_list, &fs_devices->devices);
292                 device->fs_devices = fs_devices;
293                 fs_devices->num_devices++;
294         }
295
296         if (found_transid > fs_devices->latest_trans) {
297                 fs_devices->latest_devid = devid;
298                 fs_devices->latest_trans = found_transid;
299         }
300         *fs_devices_ret = fs_devices;
301         return 0;
302 }
303
304 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
305 {
306         struct btrfs_fs_devices *fs_devices;
307         struct btrfs_device *device;
308         struct btrfs_device *orig_dev;
309
310         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
311         if (!fs_devices)
312                 return ERR_PTR(-ENOMEM);
313
314         INIT_LIST_HEAD(&fs_devices->devices);
315         INIT_LIST_HEAD(&fs_devices->alloc_list);
316         INIT_LIST_HEAD(&fs_devices->list);
317         fs_devices->latest_devid = orig->latest_devid;
318         fs_devices->latest_trans = orig->latest_trans;
319         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
320
321         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
322                 device = kzalloc(sizeof(*device), GFP_NOFS);
323                 if (!device)
324                         goto error;
325
326                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
327                 if (!device->name)
328                         goto error;
329
330                 device->devid = orig_dev->devid;
331                 device->work.func = pending_bios_fn;
332                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
333                 device->barriers = 1;
334                 spin_lock_init(&device->io_lock);
335                 INIT_LIST_HEAD(&device->dev_list);
336                 INIT_LIST_HEAD(&device->dev_alloc_list);
337
338                 list_add(&device->dev_list, &fs_devices->devices);
339                 device->fs_devices = fs_devices;
340                 fs_devices->num_devices++;
341         }
342         return fs_devices;
343 error:
344         free_fs_devices(fs_devices);
345         return ERR_PTR(-ENOMEM);
346 }
347
348 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
349 {
350         struct btrfs_device *device, *next;
351
352         mutex_lock(&uuid_mutex);
353 again:
354         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
355                 if (device->in_fs_metadata)
356                         continue;
357
358                 if (device->bdev) {
359                         close_bdev_exclusive(device->bdev, device->mode);
360                         device->bdev = NULL;
361                         fs_devices->open_devices--;
362                 }
363                 if (device->writeable) {
364                         list_del_init(&device->dev_alloc_list);
365                         device->writeable = 0;
366                         fs_devices->rw_devices--;
367                 }
368                 list_del_init(&device->dev_list);
369                 fs_devices->num_devices--;
370                 kfree(device->name);
371                 kfree(device);
372         }
373
374         if (fs_devices->seed) {
375                 fs_devices = fs_devices->seed;
376                 goto again;
377         }
378
379         mutex_unlock(&uuid_mutex);
380         return 0;
381 }
382
383 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
384 {
385         struct btrfs_device *device;
386
387         if (--fs_devices->opened > 0)
388                 return 0;
389
390         list_for_each_entry(device, &fs_devices->devices, dev_list) {
391                 if (device->bdev) {
392                         close_bdev_exclusive(device->bdev, device->mode);
393                         fs_devices->open_devices--;
394                 }
395                 if (device->writeable) {
396                         list_del_init(&device->dev_alloc_list);
397                         fs_devices->rw_devices--;
398                 }
399
400                 device->bdev = NULL;
401                 device->writeable = 0;
402                 device->in_fs_metadata = 0;
403         }
404         WARN_ON(fs_devices->open_devices);
405         WARN_ON(fs_devices->rw_devices);
406         fs_devices->opened = 0;
407         fs_devices->seeding = 0;
408
409         return 0;
410 }
411
412 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
413 {
414         struct btrfs_fs_devices *seed_devices = NULL;
415         int ret;
416
417         mutex_lock(&uuid_mutex);
418         ret = __btrfs_close_devices(fs_devices);
419         if (!fs_devices->opened) {
420                 seed_devices = fs_devices->seed;
421                 fs_devices->seed = NULL;
422         }
423         mutex_unlock(&uuid_mutex);
424
425         while (seed_devices) {
426                 fs_devices = seed_devices;
427                 seed_devices = fs_devices->seed;
428                 __btrfs_close_devices(fs_devices);
429                 free_fs_devices(fs_devices);
430         }
431         return ret;
432 }
433
434 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
435                                 fmode_t flags, void *holder)
436 {
437         struct block_device *bdev;
438         struct list_head *head = &fs_devices->devices;
439         struct btrfs_device *device;
440         struct block_device *latest_bdev = NULL;
441         struct buffer_head *bh;
442         struct btrfs_super_block *disk_super;
443         u64 latest_devid = 0;
444         u64 latest_transid = 0;
445         u64 devid;
446         int seeding = 1;
447         int ret = 0;
448
449         list_for_each_entry(device, head, dev_list) {
450                 if (device->bdev)
451                         continue;
452                 if (!device->name)
453                         continue;
454
455                 bdev = open_bdev_exclusive(device->name, flags, holder);
456                 if (IS_ERR(bdev)) {
457                         printk(KERN_INFO "open %s failed\n", device->name);
458                         goto error;
459                 }
460                 set_blocksize(bdev, 4096);
461
462                 bh = btrfs_read_dev_super(bdev);
463                 if (!bh)
464                         goto error_close;
465
466                 disk_super = (struct btrfs_super_block *)bh->b_data;
467                 devid = le64_to_cpu(disk_super->dev_item.devid);
468                 if (devid != device->devid)
469                         goto error_brelse;
470
471                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
472                            BTRFS_UUID_SIZE))
473                         goto error_brelse;
474
475                 device->generation = btrfs_super_generation(disk_super);
476                 if (!latest_transid || device->generation > latest_transid) {
477                         latest_devid = devid;
478                         latest_transid = device->generation;
479                         latest_bdev = bdev;
480                 }
481
482                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
483                         device->writeable = 0;
484                 } else {
485                         device->writeable = !bdev_read_only(bdev);
486                         seeding = 0;
487                 }
488
489                 device->bdev = bdev;
490                 device->in_fs_metadata = 0;
491                 device->mode = flags;
492
493                 fs_devices->open_devices++;
494                 if (device->writeable) {
495                         fs_devices->rw_devices++;
496                         list_add(&device->dev_alloc_list,
497                                  &fs_devices->alloc_list);
498                 }
499                 continue;
500
501 error_brelse:
502                 brelse(bh);
503 error_close:
504                 close_bdev_exclusive(bdev, FMODE_READ);
505 error:
506                 continue;
507         }
508         if (fs_devices->open_devices == 0) {
509                 ret = -EIO;
510                 goto out;
511         }
512         fs_devices->seeding = seeding;
513         fs_devices->opened = 1;
514         fs_devices->latest_bdev = latest_bdev;
515         fs_devices->latest_devid = latest_devid;
516         fs_devices->latest_trans = latest_transid;
517         fs_devices->total_rw_bytes = 0;
518 out:
519         return ret;
520 }
521
522 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
523                        fmode_t flags, void *holder)
524 {
525         int ret;
526
527         mutex_lock(&uuid_mutex);
528         if (fs_devices->opened) {
529                 fs_devices->opened++;
530                 ret = 0;
531         } else {
532                 ret = __btrfs_open_devices(fs_devices, flags, holder);
533         }
534         mutex_unlock(&uuid_mutex);
535         return ret;
536 }
537
538 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
539                           struct btrfs_fs_devices **fs_devices_ret)
540 {
541         struct btrfs_super_block *disk_super;
542         struct block_device *bdev;
543         struct buffer_head *bh;
544         int ret;
545         u64 devid;
546         u64 transid;
547
548         mutex_lock(&uuid_mutex);
549
550         bdev = open_bdev_exclusive(path, flags, holder);
551
552         if (IS_ERR(bdev)) {
553                 ret = PTR_ERR(bdev);
554                 goto error;
555         }
556
557         ret = set_blocksize(bdev, 4096);
558         if (ret)
559                 goto error_close;
560         bh = btrfs_read_dev_super(bdev);
561         if (!bh) {
562                 ret = -EIO;
563                 goto error_close;
564         }
565         disk_super = (struct btrfs_super_block *)bh->b_data;
566         devid = le64_to_cpu(disk_super->dev_item.devid);
567         transid = btrfs_super_generation(disk_super);
568         if (disk_super->label[0])
569                 printk(KERN_INFO "device label %s ", disk_super->label);
570         else {
571                 /* FIXME, make a readl uuid parser */
572                 printk(KERN_INFO "device fsid %llx-%llx ",
573                        *(unsigned long long *)disk_super->fsid,
574                        *(unsigned long long *)(disk_super->fsid + 8));
575         }
576         printk(KERN_CONT "devid %llu transid %llu %s\n",
577                (unsigned long long)devid, (unsigned long long)transid, path);
578         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
579
580         brelse(bh);
581 error_close:
582         close_bdev_exclusive(bdev, flags);
583 error:
584         mutex_unlock(&uuid_mutex);
585         return ret;
586 }
587
588 /*
589  * this uses a pretty simple search, the expectation is that it is
590  * called very infrequently and that a given device has a small number
591  * of extents
592  */
593 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
594                                          struct btrfs_device *device,
595                                          u64 num_bytes, u64 *start)
596 {
597         struct btrfs_key key;
598         struct btrfs_root *root = device->dev_root;
599         struct btrfs_dev_extent *dev_extent = NULL;
600         struct btrfs_path *path;
601         u64 hole_size = 0;
602         u64 last_byte = 0;
603         u64 search_start = 0;
604         u64 search_end = device->total_bytes;
605         int ret;
606         int slot = 0;
607         int start_found;
608         struct extent_buffer *l;
609
610         path = btrfs_alloc_path();
611         if (!path)
612                 return -ENOMEM;
613         path->reada = 2;
614         start_found = 0;
615
616         /* FIXME use last free of some kind */
617
618         /* we don't want to overwrite the superblock on the drive,
619          * so we make sure to start at an offset of at least 1MB
620          */
621         search_start = max((u64)1024 * 1024, search_start);
622
623         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
624                 search_start = max(root->fs_info->alloc_start, search_start);
625
626         key.objectid = device->devid;
627         key.offset = search_start;
628         key.type = BTRFS_DEV_EXTENT_KEY;
629         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
630         if (ret < 0)
631                 goto error;
632         ret = btrfs_previous_item(root, path, 0, key.type);
633         if (ret < 0)
634                 goto error;
635         l = path->nodes[0];
636         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
637         while (1) {
638                 l = path->nodes[0];
639                 slot = path->slots[0];
640                 if (slot >= btrfs_header_nritems(l)) {
641                         ret = btrfs_next_leaf(root, path);
642                         if (ret == 0)
643                                 continue;
644                         if (ret < 0)
645                                 goto error;
646 no_more_items:
647                         if (!start_found) {
648                                 if (search_start >= search_end) {
649                                         ret = -ENOSPC;
650                                         goto error;
651                                 }
652                                 *start = search_start;
653                                 start_found = 1;
654                                 goto check_pending;
655                         }
656                         *start = last_byte > search_start ?
657                                 last_byte : search_start;
658                         if (search_end <= *start) {
659                                 ret = -ENOSPC;
660                                 goto error;
661                         }
662                         goto check_pending;
663                 }
664                 btrfs_item_key_to_cpu(l, &key, slot);
665
666                 if (key.objectid < device->devid)
667                         goto next;
668
669                 if (key.objectid > device->devid)
670                         goto no_more_items;
671
672                 if (key.offset >= search_start && key.offset > last_byte &&
673                     start_found) {
674                         if (last_byte < search_start)
675                                 last_byte = search_start;
676                         hole_size = key.offset - last_byte;
677                         if (key.offset > last_byte &&
678                             hole_size >= num_bytes) {
679                                 *start = last_byte;
680                                 goto check_pending;
681                         }
682                 }
683                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
684                         goto next;
685
686                 start_found = 1;
687                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
688                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
689 next:
690                 path->slots[0]++;
691                 cond_resched();
692         }
693 check_pending:
694         /* we have to make sure we didn't find an extent that has already
695          * been allocated by the map tree or the original allocation
696          */
697         BUG_ON(*start < search_start);
698
699         if (*start + num_bytes > search_end) {
700                 ret = -ENOSPC;
701                 goto error;
702         }
703         /* check for pending inserts here */
704         ret = 0;
705
706 error:
707         btrfs_free_path(path);
708         return ret;
709 }
710
711 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
712                           struct btrfs_device *device,
713                           u64 start)
714 {
715         int ret;
716         struct btrfs_path *path;
717         struct btrfs_root *root = device->dev_root;
718         struct btrfs_key key;
719         struct btrfs_key found_key;
720         struct extent_buffer *leaf = NULL;
721         struct btrfs_dev_extent *extent = NULL;
722
723         path = btrfs_alloc_path();
724         if (!path)
725                 return -ENOMEM;
726
727         key.objectid = device->devid;
728         key.offset = start;
729         key.type = BTRFS_DEV_EXTENT_KEY;
730
731         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
732         if (ret > 0) {
733                 ret = btrfs_previous_item(root, path, key.objectid,
734                                           BTRFS_DEV_EXTENT_KEY);
735                 BUG_ON(ret);
736                 leaf = path->nodes[0];
737                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
738                 extent = btrfs_item_ptr(leaf, path->slots[0],
739                                         struct btrfs_dev_extent);
740                 BUG_ON(found_key.offset > start || found_key.offset +
741                        btrfs_dev_extent_length(leaf, extent) < start);
742                 ret = 0;
743         } else if (ret == 0) {
744                 leaf = path->nodes[0];
745                 extent = btrfs_item_ptr(leaf, path->slots[0],
746                                         struct btrfs_dev_extent);
747         }
748         BUG_ON(ret);
749
750         if (device->bytes_used > 0)
751                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
752         ret = btrfs_del_item(trans, root, path);
753         BUG_ON(ret);
754
755         btrfs_free_path(path);
756         return ret;
757 }
758
759 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
760                            struct btrfs_device *device,
761                            u64 chunk_tree, u64 chunk_objectid,
762                            u64 chunk_offset, u64 start, u64 num_bytes)
763 {
764         int ret;
765         struct btrfs_path *path;
766         struct btrfs_root *root = device->dev_root;
767         struct btrfs_dev_extent *extent;
768         struct extent_buffer *leaf;
769         struct btrfs_key key;
770
771         WARN_ON(!device->in_fs_metadata);
772         path = btrfs_alloc_path();
773         if (!path)
774                 return -ENOMEM;
775
776         key.objectid = device->devid;
777         key.offset = start;
778         key.type = BTRFS_DEV_EXTENT_KEY;
779         ret = btrfs_insert_empty_item(trans, root, path, &key,
780                                       sizeof(*extent));
781         BUG_ON(ret);
782
783         leaf = path->nodes[0];
784         extent = btrfs_item_ptr(leaf, path->slots[0],
785                                 struct btrfs_dev_extent);
786         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
787         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
788         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
789
790         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
791                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
792                     BTRFS_UUID_SIZE);
793
794         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
795         btrfs_mark_buffer_dirty(leaf);
796         btrfs_free_path(path);
797         return ret;
798 }
799
800 static noinline int find_next_chunk(struct btrfs_root *root,
801                                     u64 objectid, u64 *offset)
802 {
803         struct btrfs_path *path;
804         int ret;
805         struct btrfs_key key;
806         struct btrfs_chunk *chunk;
807         struct btrfs_key found_key;
808
809         path = btrfs_alloc_path();
810         BUG_ON(!path);
811
812         key.objectid = objectid;
813         key.offset = (u64)-1;
814         key.type = BTRFS_CHUNK_ITEM_KEY;
815
816         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
817         if (ret < 0)
818                 goto error;
819
820         BUG_ON(ret == 0);
821
822         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
823         if (ret) {
824                 *offset = 0;
825         } else {
826                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
827                                       path->slots[0]);
828                 if (found_key.objectid != objectid)
829                         *offset = 0;
830                 else {
831                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
832                                                struct btrfs_chunk);
833                         *offset = found_key.offset +
834                                 btrfs_chunk_length(path->nodes[0], chunk);
835                 }
836         }
837         ret = 0;
838 error:
839         btrfs_free_path(path);
840         return ret;
841 }
842
843 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
844 {
845         int ret;
846         struct btrfs_key key;
847         struct btrfs_key found_key;
848         struct btrfs_path *path;
849
850         root = root->fs_info->chunk_root;
851
852         path = btrfs_alloc_path();
853         if (!path)
854                 return -ENOMEM;
855
856         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
857         key.type = BTRFS_DEV_ITEM_KEY;
858         key.offset = (u64)-1;
859
860         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
861         if (ret < 0)
862                 goto error;
863
864         BUG_ON(ret == 0);
865
866         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
867                                   BTRFS_DEV_ITEM_KEY);
868         if (ret) {
869                 *objectid = 1;
870         } else {
871                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
872                                       path->slots[0]);
873                 *objectid = found_key.offset + 1;
874         }
875         ret = 0;
876 error:
877         btrfs_free_path(path);
878         return ret;
879 }
880
881 /*
882  * the device information is stored in the chunk root
883  * the btrfs_device struct should be fully filled in
884  */
885 int btrfs_add_device(struct btrfs_trans_handle *trans,
886                      struct btrfs_root *root,
887                      struct btrfs_device *device)
888 {
889         int ret;
890         struct btrfs_path *path;
891         struct btrfs_dev_item *dev_item;
892         struct extent_buffer *leaf;
893         struct btrfs_key key;
894         unsigned long ptr;
895
896         root = root->fs_info->chunk_root;
897
898         path = btrfs_alloc_path();
899         if (!path)
900                 return -ENOMEM;
901
902         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
903         key.type = BTRFS_DEV_ITEM_KEY;
904         key.offset = device->devid;
905
906         ret = btrfs_insert_empty_item(trans, root, path, &key,
907                                       sizeof(*dev_item));
908         if (ret)
909                 goto out;
910
911         leaf = path->nodes[0];
912         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
913
914         btrfs_set_device_id(leaf, dev_item, device->devid);
915         btrfs_set_device_generation(leaf, dev_item, 0);
916         btrfs_set_device_type(leaf, dev_item, device->type);
917         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
918         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
919         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
920         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
921         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
922         btrfs_set_device_group(leaf, dev_item, 0);
923         btrfs_set_device_seek_speed(leaf, dev_item, 0);
924         btrfs_set_device_bandwidth(leaf, dev_item, 0);
925         btrfs_set_device_start_offset(leaf, dev_item, 0);
926
927         ptr = (unsigned long)btrfs_device_uuid(dev_item);
928         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
929         ptr = (unsigned long)btrfs_device_fsid(dev_item);
930         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
931         btrfs_mark_buffer_dirty(leaf);
932
933         ret = 0;
934 out:
935         btrfs_free_path(path);
936         return ret;
937 }
938
939 static int btrfs_rm_dev_item(struct btrfs_root *root,
940                              struct btrfs_device *device)
941 {
942         int ret;
943         struct btrfs_path *path;
944         struct btrfs_key key;
945         struct btrfs_trans_handle *trans;
946
947         root = root->fs_info->chunk_root;
948
949         path = btrfs_alloc_path();
950         if (!path)
951                 return -ENOMEM;
952
953         trans = btrfs_start_transaction(root, 1);
954         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
955         key.type = BTRFS_DEV_ITEM_KEY;
956         key.offset = device->devid;
957         lock_chunks(root);
958
959         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
960         if (ret < 0)
961                 goto out;
962
963         if (ret > 0) {
964                 ret = -ENOENT;
965                 goto out;
966         }
967
968         ret = btrfs_del_item(trans, root, path);
969         if (ret)
970                 goto out;
971 out:
972         btrfs_free_path(path);
973         unlock_chunks(root);
974         btrfs_commit_transaction(trans, root);
975         return ret;
976 }
977
978 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
979 {
980         struct btrfs_device *device;
981         struct btrfs_device *next_device;
982         struct block_device *bdev;
983         struct buffer_head *bh = NULL;
984         struct btrfs_super_block *disk_super;
985         u64 all_avail;
986         u64 devid;
987         u64 num_devices;
988         u8 *dev_uuid;
989         int ret = 0;
990
991         mutex_lock(&uuid_mutex);
992         mutex_lock(&root->fs_info->volume_mutex);
993
994         all_avail = root->fs_info->avail_data_alloc_bits |
995                 root->fs_info->avail_system_alloc_bits |
996                 root->fs_info->avail_metadata_alloc_bits;
997
998         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
999             root->fs_info->fs_devices->rw_devices <= 4) {
1000                 printk(KERN_ERR "btrfs: unable to go below four devices "
1001                        "on raid10\n");
1002                 ret = -EINVAL;
1003                 goto out;
1004         }
1005
1006         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1007             root->fs_info->fs_devices->rw_devices <= 2) {
1008                 printk(KERN_ERR "btrfs: unable to go below two "
1009                        "devices on raid1\n");
1010                 ret = -EINVAL;
1011                 goto out;
1012         }
1013
1014         if (strcmp(device_path, "missing") == 0) {
1015                 struct list_head *devices;
1016                 struct btrfs_device *tmp;
1017
1018                 device = NULL;
1019                 devices = &root->fs_info->fs_devices->devices;
1020                 list_for_each_entry(tmp, devices, dev_list) {
1021                         if (tmp->in_fs_metadata && !tmp->bdev) {
1022                                 device = tmp;
1023                                 break;
1024                         }
1025                 }
1026                 bdev = NULL;
1027                 bh = NULL;
1028                 disk_super = NULL;
1029                 if (!device) {
1030                         printk(KERN_ERR "btrfs: no missing devices found to "
1031                                "remove\n");
1032                         goto out;
1033                 }
1034         } else {
1035                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1036                                       root->fs_info->bdev_holder);
1037                 if (IS_ERR(bdev)) {
1038                         ret = PTR_ERR(bdev);
1039                         goto out;
1040                 }
1041
1042                 set_blocksize(bdev, 4096);
1043                 bh = btrfs_read_dev_super(bdev);
1044                 if (!bh) {
1045                         ret = -EIO;
1046                         goto error_close;
1047                 }
1048                 disk_super = (struct btrfs_super_block *)bh->b_data;
1049                 devid = le64_to_cpu(disk_super->dev_item.devid);
1050                 dev_uuid = disk_super->dev_item.uuid;
1051                 device = btrfs_find_device(root, devid, dev_uuid,
1052                                            disk_super->fsid);
1053                 if (!device) {
1054                         ret = -ENOENT;
1055                         goto error_brelse;
1056                 }
1057         }
1058
1059         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1060                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1061                        "device\n");
1062                 ret = -EINVAL;
1063                 goto error_brelse;
1064         }
1065
1066         if (device->writeable) {
1067                 list_del_init(&device->dev_alloc_list);
1068                 root->fs_info->fs_devices->rw_devices--;
1069         }
1070
1071         ret = btrfs_shrink_device(device, 0);
1072         if (ret)
1073                 goto error_brelse;
1074
1075         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1076         if (ret)
1077                 goto error_brelse;
1078
1079         device->in_fs_metadata = 0;
1080         list_del_init(&device->dev_list);
1081         device->fs_devices->num_devices--;
1082
1083         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1084                                  struct btrfs_device, dev_list);
1085         if (device->bdev == root->fs_info->sb->s_bdev)
1086                 root->fs_info->sb->s_bdev = next_device->bdev;
1087         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1088                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1089
1090         if (device->bdev) {
1091                 close_bdev_exclusive(device->bdev, device->mode);
1092                 device->bdev = NULL;
1093                 device->fs_devices->open_devices--;
1094         }
1095
1096         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1097         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1098
1099         if (device->fs_devices->open_devices == 0) {
1100                 struct btrfs_fs_devices *fs_devices;
1101                 fs_devices = root->fs_info->fs_devices;
1102                 while (fs_devices) {
1103                         if (fs_devices->seed == device->fs_devices)
1104                                 break;
1105                         fs_devices = fs_devices->seed;
1106                 }
1107                 fs_devices->seed = device->fs_devices->seed;
1108                 device->fs_devices->seed = NULL;
1109                 __btrfs_close_devices(device->fs_devices);
1110                 free_fs_devices(device->fs_devices);
1111         }
1112
1113         /*
1114          * at this point, the device is zero sized.  We want to
1115          * remove it from the devices list and zero out the old super
1116          */
1117         if (device->writeable) {
1118                 /* make sure this device isn't detected as part of
1119                  * the FS anymore
1120                  */
1121                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1122                 set_buffer_dirty(bh);
1123                 sync_dirty_buffer(bh);
1124         }
1125
1126         kfree(device->name);
1127         kfree(device);
1128         ret = 0;
1129
1130 error_brelse:
1131         brelse(bh);
1132 error_close:
1133         if (bdev)
1134                 close_bdev_exclusive(bdev, FMODE_READ);
1135 out:
1136         mutex_unlock(&root->fs_info->volume_mutex);
1137         mutex_unlock(&uuid_mutex);
1138         return ret;
1139 }
1140
1141 /*
1142  * does all the dirty work required for changing file system's UUID.
1143  */
1144 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1145                                 struct btrfs_root *root)
1146 {
1147         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1148         struct btrfs_fs_devices *old_devices;
1149         struct btrfs_fs_devices *seed_devices;
1150         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1151         struct btrfs_device *device;
1152         u64 super_flags;
1153
1154         BUG_ON(!mutex_is_locked(&uuid_mutex));
1155         if (!fs_devices->seeding)
1156                 return -EINVAL;
1157
1158         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1159         if (!seed_devices)
1160                 return -ENOMEM;
1161
1162         old_devices = clone_fs_devices(fs_devices);
1163         if (IS_ERR(old_devices)) {
1164                 kfree(seed_devices);
1165                 return PTR_ERR(old_devices);
1166         }
1167
1168         list_add(&old_devices->list, &fs_uuids);
1169
1170         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1171         seed_devices->opened = 1;
1172         INIT_LIST_HEAD(&seed_devices->devices);
1173         INIT_LIST_HEAD(&seed_devices->alloc_list);
1174         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1175         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1176         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1177                 device->fs_devices = seed_devices;
1178         }
1179
1180         fs_devices->seeding = 0;
1181         fs_devices->num_devices = 0;
1182         fs_devices->open_devices = 0;
1183         fs_devices->seed = seed_devices;
1184
1185         generate_random_uuid(fs_devices->fsid);
1186         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1187         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1188         super_flags = btrfs_super_flags(disk_super) &
1189                       ~BTRFS_SUPER_FLAG_SEEDING;
1190         btrfs_set_super_flags(disk_super, super_flags);
1191
1192         return 0;
1193 }
1194
1195 /*
1196  * strore the expected generation for seed devices in device items.
1197  */
1198 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1199                                struct btrfs_root *root)
1200 {
1201         struct btrfs_path *path;
1202         struct extent_buffer *leaf;
1203         struct btrfs_dev_item *dev_item;
1204         struct btrfs_device *device;
1205         struct btrfs_key key;
1206         u8 fs_uuid[BTRFS_UUID_SIZE];
1207         u8 dev_uuid[BTRFS_UUID_SIZE];
1208         u64 devid;
1209         int ret;
1210
1211         path = btrfs_alloc_path();
1212         if (!path)
1213                 return -ENOMEM;
1214
1215         root = root->fs_info->chunk_root;
1216         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1217         key.offset = 0;
1218         key.type = BTRFS_DEV_ITEM_KEY;
1219
1220         while (1) {
1221                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1222                 if (ret < 0)
1223                         goto error;
1224
1225                 leaf = path->nodes[0];
1226 next_slot:
1227                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1228                         ret = btrfs_next_leaf(root, path);
1229                         if (ret > 0)
1230                                 break;
1231                         if (ret < 0)
1232                                 goto error;
1233                         leaf = path->nodes[0];
1234                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1235                         btrfs_release_path(root, path);
1236                         continue;
1237                 }
1238
1239                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1240                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1241                     key.type != BTRFS_DEV_ITEM_KEY)
1242                         break;
1243
1244                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1245                                           struct btrfs_dev_item);
1246                 devid = btrfs_device_id(leaf, dev_item);
1247                 read_extent_buffer(leaf, dev_uuid,
1248                                    (unsigned long)btrfs_device_uuid(dev_item),
1249                                    BTRFS_UUID_SIZE);
1250                 read_extent_buffer(leaf, fs_uuid,
1251                                    (unsigned long)btrfs_device_fsid(dev_item),
1252                                    BTRFS_UUID_SIZE);
1253                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1254                 BUG_ON(!device);
1255
1256                 if (device->fs_devices->seeding) {
1257                         btrfs_set_device_generation(leaf, dev_item,
1258                                                     device->generation);
1259                         btrfs_mark_buffer_dirty(leaf);
1260                 }
1261
1262                 path->slots[0]++;
1263                 goto next_slot;
1264         }
1265         ret = 0;
1266 error:
1267         btrfs_free_path(path);
1268         return ret;
1269 }
1270
1271 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1272 {
1273         struct btrfs_trans_handle *trans;
1274         struct btrfs_device *device;
1275         struct block_device *bdev;
1276         struct list_head *devices;
1277         struct super_block *sb = root->fs_info->sb;
1278         u64 total_bytes;
1279         int seeding_dev = 0;
1280         int ret = 0;
1281
1282         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1283                 return -EINVAL;
1284
1285         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1286         if (!bdev)
1287                 return -EIO;
1288
1289         if (root->fs_info->fs_devices->seeding) {
1290                 seeding_dev = 1;
1291                 down_write(&sb->s_umount);
1292                 mutex_lock(&uuid_mutex);
1293         }
1294
1295         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1296         mutex_lock(&root->fs_info->volume_mutex);
1297
1298         devices = &root->fs_info->fs_devices->devices;
1299         list_for_each_entry(device, devices, dev_list) {
1300                 if (device->bdev == bdev) {
1301                         ret = -EEXIST;
1302                         goto error;
1303                 }
1304         }
1305
1306         device = kzalloc(sizeof(*device), GFP_NOFS);
1307         if (!device) {
1308                 /* we can safely leave the fs_devices entry around */
1309                 ret = -ENOMEM;
1310                 goto error;
1311         }
1312
1313         device->name = kstrdup(device_path, GFP_NOFS);
1314         if (!device->name) {
1315                 kfree(device);
1316                 ret = -ENOMEM;
1317                 goto error;
1318         }
1319
1320         ret = find_next_devid(root, &device->devid);
1321         if (ret) {
1322                 kfree(device);
1323                 goto error;
1324         }
1325
1326         trans = btrfs_start_transaction(root, 1);
1327         lock_chunks(root);
1328
1329         device->barriers = 1;
1330         device->writeable = 1;
1331         device->work.func = pending_bios_fn;
1332         generate_random_uuid(device->uuid);
1333         spin_lock_init(&device->io_lock);
1334         device->generation = trans->transid;
1335         device->io_width = root->sectorsize;
1336         device->io_align = root->sectorsize;
1337         device->sector_size = root->sectorsize;
1338         device->total_bytes = i_size_read(bdev->bd_inode);
1339         device->dev_root = root->fs_info->dev_root;
1340         device->bdev = bdev;
1341         device->in_fs_metadata = 1;
1342         device->mode = 0;
1343         set_blocksize(device->bdev, 4096);
1344
1345         if (seeding_dev) {
1346                 sb->s_flags &= ~MS_RDONLY;
1347                 ret = btrfs_prepare_sprout(trans, root);
1348                 BUG_ON(ret);
1349         }
1350
1351         device->fs_devices = root->fs_info->fs_devices;
1352         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1353         list_add(&device->dev_alloc_list,
1354                  &root->fs_info->fs_devices->alloc_list);
1355         root->fs_info->fs_devices->num_devices++;
1356         root->fs_info->fs_devices->open_devices++;
1357         root->fs_info->fs_devices->rw_devices++;
1358         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1359
1360         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1361         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1362                                     total_bytes + device->total_bytes);
1363
1364         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1365         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1366                                     total_bytes + 1);
1367
1368         if (seeding_dev) {
1369                 ret = init_first_rw_device(trans, root, device);
1370                 BUG_ON(ret);
1371                 ret = btrfs_finish_sprout(trans, root);
1372                 BUG_ON(ret);
1373         } else {
1374                 ret = btrfs_add_device(trans, root, device);
1375         }
1376
1377         unlock_chunks(root);
1378         btrfs_commit_transaction(trans, root);
1379
1380         if (seeding_dev) {
1381                 mutex_unlock(&uuid_mutex);
1382                 up_write(&sb->s_umount);
1383
1384                 ret = btrfs_relocate_sys_chunks(root);
1385                 BUG_ON(ret);
1386         }
1387 out:
1388         mutex_unlock(&root->fs_info->volume_mutex);
1389         return ret;
1390 error:
1391         close_bdev_exclusive(bdev, 0);
1392         if (seeding_dev) {
1393                 mutex_unlock(&uuid_mutex);
1394                 up_write(&sb->s_umount);
1395         }
1396         goto out;
1397 }
1398
1399 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1400                                         struct btrfs_device *device)
1401 {
1402         int ret;
1403         struct btrfs_path *path;
1404         struct btrfs_root *root;
1405         struct btrfs_dev_item *dev_item;
1406         struct extent_buffer *leaf;
1407         struct btrfs_key key;
1408
1409         root = device->dev_root->fs_info->chunk_root;
1410
1411         path = btrfs_alloc_path();
1412         if (!path)
1413                 return -ENOMEM;
1414
1415         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1416         key.type = BTRFS_DEV_ITEM_KEY;
1417         key.offset = device->devid;
1418
1419         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1420         if (ret < 0)
1421                 goto out;
1422
1423         if (ret > 0) {
1424                 ret = -ENOENT;
1425                 goto out;
1426         }
1427
1428         leaf = path->nodes[0];
1429         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1430
1431         btrfs_set_device_id(leaf, dev_item, device->devid);
1432         btrfs_set_device_type(leaf, dev_item, device->type);
1433         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1434         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1435         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1436         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1437         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1438         btrfs_mark_buffer_dirty(leaf);
1439
1440 out:
1441         btrfs_free_path(path);
1442         return ret;
1443 }
1444
1445 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1446                       struct btrfs_device *device, u64 new_size)
1447 {
1448         struct btrfs_super_block *super_copy =
1449                 &device->dev_root->fs_info->super_copy;
1450         u64 old_total = btrfs_super_total_bytes(super_copy);
1451         u64 diff = new_size - device->total_bytes;
1452
1453         if (!device->writeable)
1454                 return -EACCES;
1455         if (new_size <= device->total_bytes)
1456                 return -EINVAL;
1457
1458         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1459         device->fs_devices->total_rw_bytes += diff;
1460
1461         device->total_bytes = new_size;
1462         return btrfs_update_device(trans, device);
1463 }
1464
1465 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1466                       struct btrfs_device *device, u64 new_size)
1467 {
1468         int ret;
1469         lock_chunks(device->dev_root);
1470         ret = __btrfs_grow_device(trans, device, new_size);
1471         unlock_chunks(device->dev_root);
1472         return ret;
1473 }
1474
1475 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1476                             struct btrfs_root *root,
1477                             u64 chunk_tree, u64 chunk_objectid,
1478                             u64 chunk_offset)
1479 {
1480         int ret;
1481         struct btrfs_path *path;
1482         struct btrfs_key key;
1483
1484         root = root->fs_info->chunk_root;
1485         path = btrfs_alloc_path();
1486         if (!path)
1487                 return -ENOMEM;
1488
1489         key.objectid = chunk_objectid;
1490         key.offset = chunk_offset;
1491         key.type = BTRFS_CHUNK_ITEM_KEY;
1492
1493         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1494         BUG_ON(ret);
1495
1496         ret = btrfs_del_item(trans, root, path);
1497         BUG_ON(ret);
1498
1499         btrfs_free_path(path);
1500         return 0;
1501 }
1502
1503 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1504                         chunk_offset)
1505 {
1506         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1507         struct btrfs_disk_key *disk_key;
1508         struct btrfs_chunk *chunk;
1509         u8 *ptr;
1510         int ret = 0;
1511         u32 num_stripes;
1512         u32 array_size;
1513         u32 len = 0;
1514         u32 cur;
1515         struct btrfs_key key;
1516
1517         array_size = btrfs_super_sys_array_size(super_copy);
1518
1519         ptr = super_copy->sys_chunk_array;
1520         cur = 0;
1521
1522         while (cur < array_size) {
1523                 disk_key = (struct btrfs_disk_key *)ptr;
1524                 btrfs_disk_key_to_cpu(&key, disk_key);
1525
1526                 len = sizeof(*disk_key);
1527
1528                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1529                         chunk = (struct btrfs_chunk *)(ptr + len);
1530                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1531                         len += btrfs_chunk_item_size(num_stripes);
1532                 } else {
1533                         ret = -EIO;
1534                         break;
1535                 }
1536                 if (key.objectid == chunk_objectid &&
1537                     key.offset == chunk_offset) {
1538                         memmove(ptr, ptr + len, array_size - (cur + len));
1539                         array_size -= len;
1540                         btrfs_set_super_sys_array_size(super_copy, array_size);
1541                 } else {
1542                         ptr += len;
1543                         cur += len;
1544                 }
1545         }
1546         return ret;
1547 }
1548
1549 static int btrfs_relocate_chunk(struct btrfs_root *root,
1550                          u64 chunk_tree, u64 chunk_objectid,
1551                          u64 chunk_offset)
1552 {
1553         struct extent_map_tree *em_tree;
1554         struct btrfs_root *extent_root;
1555         struct btrfs_trans_handle *trans;
1556         struct extent_map *em;
1557         struct map_lookup *map;
1558         int ret;
1559         int i;
1560
1561         printk(KERN_INFO "btrfs relocating chunk %llu\n",
1562                (unsigned long long)chunk_offset);
1563         root = root->fs_info->chunk_root;
1564         extent_root = root->fs_info->extent_root;
1565         em_tree = &root->fs_info->mapping_tree.map_tree;
1566
1567         /* step one, relocate all the extents inside this chunk */
1568         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1569         BUG_ON(ret);
1570
1571         trans = btrfs_start_transaction(root, 1);
1572         BUG_ON(!trans);
1573
1574         lock_chunks(root);
1575
1576         /*
1577          * step two, delete the device extents and the
1578          * chunk tree entries
1579          */
1580         spin_lock(&em_tree->lock);
1581         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1582         spin_unlock(&em_tree->lock);
1583
1584         BUG_ON(em->start > chunk_offset ||
1585                em->start + em->len < chunk_offset);
1586         map = (struct map_lookup *)em->bdev;
1587
1588         for (i = 0; i < map->num_stripes; i++) {
1589                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1590                                             map->stripes[i].physical);
1591                 BUG_ON(ret);
1592
1593                 if (map->stripes[i].dev) {
1594                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1595                         BUG_ON(ret);
1596                 }
1597         }
1598         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1599                                chunk_offset);
1600
1601         BUG_ON(ret);
1602
1603         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1604                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1605                 BUG_ON(ret);
1606         }
1607
1608         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1609         BUG_ON(ret);
1610
1611         spin_lock(&em_tree->lock);
1612         remove_extent_mapping(em_tree, em);
1613         spin_unlock(&em_tree->lock);
1614
1615         kfree(map);
1616         em->bdev = NULL;
1617
1618         /* once for the tree */
1619         free_extent_map(em);
1620         /* once for us */
1621         free_extent_map(em);
1622
1623         unlock_chunks(root);
1624         btrfs_end_transaction(trans, root);
1625         return 0;
1626 }
1627
1628 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1629 {
1630         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1631         struct btrfs_path *path;
1632         struct extent_buffer *leaf;
1633         struct btrfs_chunk *chunk;
1634         struct btrfs_key key;
1635         struct btrfs_key found_key;
1636         u64 chunk_tree = chunk_root->root_key.objectid;
1637         u64 chunk_type;
1638         int ret;
1639
1640         path = btrfs_alloc_path();
1641         if (!path)
1642                 return -ENOMEM;
1643
1644         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1645         key.offset = (u64)-1;
1646         key.type = BTRFS_CHUNK_ITEM_KEY;
1647
1648         while (1) {
1649                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1650                 if (ret < 0)
1651                         goto error;
1652                 BUG_ON(ret == 0);
1653
1654                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1655                                           key.type);
1656                 if (ret < 0)
1657                         goto error;
1658                 if (ret > 0)
1659                         break;
1660
1661                 leaf = path->nodes[0];
1662                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1663
1664                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1665                                        struct btrfs_chunk);
1666                 chunk_type = btrfs_chunk_type(leaf, chunk);
1667                 btrfs_release_path(chunk_root, path);
1668
1669                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1670                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1671                                                    found_key.objectid,
1672                                                    found_key.offset);
1673                         BUG_ON(ret);
1674                 }
1675
1676                 if (found_key.offset == 0)
1677                         break;
1678                 key.offset = found_key.offset - 1;
1679         }
1680         ret = 0;
1681 error:
1682         btrfs_free_path(path);
1683         return ret;
1684 }
1685
1686 static u64 div_factor(u64 num, int factor)
1687 {
1688         if (factor == 10)
1689                 return num;
1690         num *= factor;
1691         do_div(num, 10);
1692         return num;
1693 }
1694
1695 int btrfs_balance(struct btrfs_root *dev_root)
1696 {
1697         int ret;
1698         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1699         struct btrfs_device *device;
1700         u64 old_size;
1701         u64 size_to_free;
1702         struct btrfs_path *path;
1703         struct btrfs_key key;
1704         struct btrfs_chunk *chunk;
1705         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1706         struct btrfs_trans_handle *trans;
1707         struct btrfs_key found_key;
1708
1709         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1710                 return -EROFS;
1711
1712         mutex_lock(&dev_root->fs_info->volume_mutex);
1713         dev_root = dev_root->fs_info->dev_root;
1714
1715         /* step one make some room on all the devices */
1716         list_for_each_entry(device, devices, dev_list) {
1717                 old_size = device->total_bytes;
1718                 size_to_free = div_factor(old_size, 1);
1719                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1720                 if (!device->writeable ||
1721                     device->total_bytes - device->bytes_used > size_to_free)
1722                         continue;
1723
1724                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1725                 BUG_ON(ret);
1726
1727                 trans = btrfs_start_transaction(dev_root, 1);
1728                 BUG_ON(!trans);
1729
1730                 ret = btrfs_grow_device(trans, device, old_size);
1731                 BUG_ON(ret);
1732
1733                 btrfs_end_transaction(trans, dev_root);
1734         }
1735
1736         /* step two, relocate all the chunks */
1737         path = btrfs_alloc_path();
1738         BUG_ON(!path);
1739
1740         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1741         key.offset = (u64)-1;
1742         key.type = BTRFS_CHUNK_ITEM_KEY;
1743
1744         while (1) {
1745                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1746                 if (ret < 0)
1747                         goto error;
1748
1749                 /*
1750                  * this shouldn't happen, it means the last relocate
1751                  * failed
1752                  */
1753                 if (ret == 0)
1754                         break;
1755
1756                 ret = btrfs_previous_item(chunk_root, path, 0,
1757                                           BTRFS_CHUNK_ITEM_KEY);
1758                 if (ret)
1759                         break;
1760
1761                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1762                                       path->slots[0]);
1763                 if (found_key.objectid != key.objectid)
1764                         break;
1765
1766                 chunk = btrfs_item_ptr(path->nodes[0],
1767                                        path->slots[0],
1768                                        struct btrfs_chunk);
1769                 key.offset = found_key.offset;
1770                 /* chunk zero is special */
1771                 if (key.offset == 0)
1772                         break;
1773
1774                 btrfs_release_path(chunk_root, path);
1775                 ret = btrfs_relocate_chunk(chunk_root,
1776                                            chunk_root->root_key.objectid,
1777                                            found_key.objectid,
1778                                            found_key.offset);
1779                 BUG_ON(ret);
1780         }
1781         ret = 0;
1782 error:
1783         btrfs_free_path(path);
1784         mutex_unlock(&dev_root->fs_info->volume_mutex);
1785         return ret;
1786 }
1787
1788 /*
1789  * shrinking a device means finding all of the device extents past
1790  * the new size, and then following the back refs to the chunks.
1791  * The chunk relocation code actually frees the device extent
1792  */
1793 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1794 {
1795         struct btrfs_trans_handle *trans;
1796         struct btrfs_root *root = device->dev_root;
1797         struct btrfs_dev_extent *dev_extent = NULL;
1798         struct btrfs_path *path;
1799         u64 length;
1800         u64 chunk_tree;
1801         u64 chunk_objectid;
1802         u64 chunk_offset;
1803         int ret;
1804         int slot;
1805         struct extent_buffer *l;
1806         struct btrfs_key key;
1807         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1808         u64 old_total = btrfs_super_total_bytes(super_copy);
1809         u64 diff = device->total_bytes - new_size;
1810
1811         if (new_size >= device->total_bytes)
1812                 return -EINVAL;
1813
1814         path = btrfs_alloc_path();
1815         if (!path)
1816                 return -ENOMEM;
1817
1818         trans = btrfs_start_transaction(root, 1);
1819         if (!trans) {
1820                 ret = -ENOMEM;
1821                 goto done;
1822         }
1823
1824         path->reada = 2;
1825
1826         lock_chunks(root);
1827
1828         device->total_bytes = new_size;
1829         if (device->writeable)
1830                 device->fs_devices->total_rw_bytes -= diff;
1831         ret = btrfs_update_device(trans, device);
1832         if (ret) {
1833                 unlock_chunks(root);
1834                 btrfs_end_transaction(trans, root);
1835                 goto done;
1836         }
1837         WARN_ON(diff > old_total);
1838         btrfs_set_super_total_bytes(super_copy, old_total - diff);
1839         unlock_chunks(root);
1840         btrfs_end_transaction(trans, root);
1841
1842         key.objectid = device->devid;
1843         key.offset = (u64)-1;
1844         key.type = BTRFS_DEV_EXTENT_KEY;
1845
1846         while (1) {
1847                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1848                 if (ret < 0)
1849                         goto done;
1850
1851                 ret = btrfs_previous_item(root, path, 0, key.type);
1852                 if (ret < 0)
1853                         goto done;
1854                 if (ret) {
1855                         ret = 0;
1856                         goto done;
1857                 }
1858
1859                 l = path->nodes[0];
1860                 slot = path->slots[0];
1861                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1862
1863                 if (key.objectid != device->devid)
1864                         goto done;
1865
1866                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1867                 length = btrfs_dev_extent_length(l, dev_extent);
1868
1869                 if (key.offset + length <= new_size)
1870                         goto done;
1871
1872                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1873                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1874                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1875                 btrfs_release_path(root, path);
1876
1877                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1878                                            chunk_offset);
1879                 if (ret)
1880                         goto done;
1881         }
1882
1883 done:
1884         btrfs_free_path(path);
1885         return ret;
1886 }
1887
1888 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1889                            struct btrfs_root *root,
1890                            struct btrfs_key *key,
1891                            struct btrfs_chunk *chunk, int item_size)
1892 {
1893         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1894         struct btrfs_disk_key disk_key;
1895         u32 array_size;
1896         u8 *ptr;
1897
1898         array_size = btrfs_super_sys_array_size(super_copy);
1899         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1900                 return -EFBIG;
1901
1902         ptr = super_copy->sys_chunk_array + array_size;
1903         btrfs_cpu_key_to_disk(&disk_key, key);
1904         memcpy(ptr, &disk_key, sizeof(disk_key));
1905         ptr += sizeof(disk_key);
1906         memcpy(ptr, chunk, item_size);
1907         item_size += sizeof(disk_key);
1908         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1909         return 0;
1910 }
1911
1912 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
1913                                         int num_stripes, int sub_stripes)
1914 {
1915         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1916                 return calc_size;
1917         else if (type & BTRFS_BLOCK_GROUP_RAID10)
1918                 return calc_size * (num_stripes / sub_stripes);
1919         else
1920                 return calc_size * num_stripes;
1921 }
1922
1923 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1924                                struct btrfs_root *extent_root,
1925                                struct map_lookup **map_ret,
1926                                u64 *num_bytes, u64 *stripe_size,
1927                                u64 start, u64 type)
1928 {
1929         struct btrfs_fs_info *info = extent_root->fs_info;
1930         struct btrfs_device *device = NULL;
1931         struct btrfs_fs_devices *fs_devices = info->fs_devices;
1932         struct list_head *cur;
1933         struct map_lookup *map = NULL;
1934         struct extent_map_tree *em_tree;
1935         struct extent_map *em;
1936         struct list_head private_devs;
1937         int min_stripe_size = 1 * 1024 * 1024;
1938         u64 calc_size = 1024 * 1024 * 1024;
1939         u64 max_chunk_size = calc_size;
1940         u64 min_free;
1941         u64 avail;
1942         u64 max_avail = 0;
1943         u64 dev_offset;
1944         int num_stripes = 1;
1945         int min_stripes = 1;
1946         int sub_stripes = 0;
1947         int looped = 0;
1948         int ret;
1949         int index;
1950         int stripe_len = 64 * 1024;
1951
1952         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1953             (type & BTRFS_BLOCK_GROUP_DUP)) {
1954                 WARN_ON(1);
1955                 type &= ~BTRFS_BLOCK_GROUP_DUP;
1956         }
1957         if (list_empty(&fs_devices->alloc_list))
1958                 return -ENOSPC;
1959
1960         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1961                 num_stripes = fs_devices->rw_devices;
1962                 min_stripes = 2;
1963         }
1964         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1965                 num_stripes = 2;
1966                 min_stripes = 2;
1967         }
1968         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1969                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1970                 if (num_stripes < 2)
1971                         return -ENOSPC;
1972                 min_stripes = 2;
1973         }
1974         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1975                 num_stripes = fs_devices->rw_devices;
1976                 if (num_stripes < 4)
1977                         return -ENOSPC;
1978                 num_stripes &= ~(u32)1;
1979                 sub_stripes = 2;
1980                 min_stripes = 4;
1981         }
1982
1983         if (type & BTRFS_BLOCK_GROUP_DATA) {
1984                 max_chunk_size = 10 * calc_size;
1985                 min_stripe_size = 64 * 1024 * 1024;
1986         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1987                 max_chunk_size = 4 * calc_size;
1988                 min_stripe_size = 32 * 1024 * 1024;
1989         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1990                 calc_size = 8 * 1024 * 1024;
1991                 max_chunk_size = calc_size * 2;
1992                 min_stripe_size = 1 * 1024 * 1024;
1993         }
1994
1995         /* we don't want a chunk larger than 10% of writeable space */
1996         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
1997                              max_chunk_size);
1998
1999 again:
2000         if (!map || map->num_stripes != num_stripes) {
2001                 kfree(map);
2002                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2003                 if (!map)
2004                         return -ENOMEM;
2005                 map->num_stripes = num_stripes;
2006         }
2007
2008         if (calc_size * num_stripes > max_chunk_size) {
2009                 calc_size = max_chunk_size;
2010                 do_div(calc_size, num_stripes);
2011                 do_div(calc_size, stripe_len);
2012                 calc_size *= stripe_len;
2013         }
2014         /* we don't want tiny stripes */
2015         calc_size = max_t(u64, min_stripe_size, calc_size);
2016
2017         do_div(calc_size, stripe_len);
2018         calc_size *= stripe_len;
2019
2020         cur = fs_devices->alloc_list.next;
2021         index = 0;
2022
2023         if (type & BTRFS_BLOCK_GROUP_DUP)
2024                 min_free = calc_size * 2;
2025         else
2026                 min_free = calc_size;
2027
2028         /*
2029          * we add 1MB because we never use the first 1MB of the device, unless
2030          * we've looped, then we are likely allocating the maximum amount of
2031          * space left already
2032          */
2033         if (!looped)
2034                 min_free += 1024 * 1024;
2035
2036         INIT_LIST_HEAD(&private_devs);
2037         while (index < num_stripes) {
2038                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2039                 BUG_ON(!device->writeable);
2040                 if (device->total_bytes > device->bytes_used)
2041                         avail = device->total_bytes - device->bytes_used;
2042                 else
2043                         avail = 0;
2044                 cur = cur->next;
2045
2046                 if (device->in_fs_metadata && avail >= min_free) {
2047                         ret = find_free_dev_extent(trans, device,
2048                                                    min_free, &dev_offset);
2049                         if (ret == 0) {
2050                                 list_move_tail(&device->dev_alloc_list,
2051                                                &private_devs);
2052                                 map->stripes[index].dev = device;
2053                                 map->stripes[index].physical = dev_offset;
2054                                 index++;
2055                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2056                                         map->stripes[index].dev = device;
2057                                         map->stripes[index].physical =
2058                                                 dev_offset + calc_size;
2059                                         index++;
2060                                 }
2061                         }
2062                 } else if (device->in_fs_metadata && avail > max_avail)
2063                         max_avail = avail;
2064                 if (cur == &fs_devices->alloc_list)
2065                         break;
2066         }
2067         list_splice(&private_devs, &fs_devices->alloc_list);
2068         if (index < num_stripes) {
2069                 if (index >= min_stripes) {
2070                         num_stripes = index;
2071                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2072                                 num_stripes /= sub_stripes;
2073                                 num_stripes *= sub_stripes;
2074                         }
2075                         looped = 1;
2076                         goto again;
2077                 }
2078                 if (!looped && max_avail > 0) {
2079                         looped = 1;
2080                         calc_size = max_avail;
2081                         goto again;
2082                 }
2083                 kfree(map);
2084                 return -ENOSPC;
2085         }
2086         map->sector_size = extent_root->sectorsize;
2087         map->stripe_len = stripe_len;
2088         map->io_align = stripe_len;
2089         map->io_width = stripe_len;
2090         map->type = type;
2091         map->num_stripes = num_stripes;
2092         map->sub_stripes = sub_stripes;
2093
2094         *map_ret = map;
2095         *stripe_size = calc_size;
2096         *num_bytes = chunk_bytes_by_type(type, calc_size,
2097                                          num_stripes, sub_stripes);
2098
2099         em = alloc_extent_map(GFP_NOFS);
2100         if (!em) {
2101                 kfree(map);
2102                 return -ENOMEM;
2103         }
2104         em->bdev = (struct block_device *)map;
2105         em->start = start;
2106         em->len = *num_bytes;
2107         em->block_start = 0;
2108         em->block_len = em->len;
2109
2110         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2111         spin_lock(&em_tree->lock);
2112         ret = add_extent_mapping(em_tree, em);
2113         spin_unlock(&em_tree->lock);
2114         BUG_ON(ret);
2115         free_extent_map(em);
2116
2117         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2118                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2119                                      start, *num_bytes);
2120         BUG_ON(ret);
2121
2122         index = 0;
2123         while (index < map->num_stripes) {
2124                 device = map->stripes[index].dev;
2125                 dev_offset = map->stripes[index].physical;
2126
2127                 ret = btrfs_alloc_dev_extent(trans, device,
2128                                 info->chunk_root->root_key.objectid,
2129                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2130                                 start, dev_offset, calc_size);
2131                 BUG_ON(ret);
2132                 index++;
2133         }
2134
2135         return 0;
2136 }
2137
2138 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2139                                 struct btrfs_root *extent_root,
2140                                 struct map_lookup *map, u64 chunk_offset,
2141                                 u64 chunk_size, u64 stripe_size)
2142 {
2143         u64 dev_offset;
2144         struct btrfs_key key;
2145         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2146         struct btrfs_device *device;
2147         struct btrfs_chunk *chunk;
2148         struct btrfs_stripe *stripe;
2149         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2150         int index = 0;
2151         int ret;
2152
2153         chunk = kzalloc(item_size, GFP_NOFS);
2154         if (!chunk)
2155                 return -ENOMEM;
2156
2157         index = 0;
2158         while (index < map->num_stripes) {
2159                 device = map->stripes[index].dev;
2160                 device->bytes_used += stripe_size;
2161                 ret = btrfs_update_device(trans, device);
2162                 BUG_ON(ret);
2163                 index++;
2164         }
2165
2166         index = 0;
2167         stripe = &chunk->stripe;
2168         while (index < map->num_stripes) {
2169                 device = map->stripes[index].dev;
2170                 dev_offset = map->stripes[index].physical;
2171
2172                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2173                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2174                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2175                 stripe++;
2176                 index++;
2177         }
2178
2179         btrfs_set_stack_chunk_length(chunk, chunk_size);
2180         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2181         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2182         btrfs_set_stack_chunk_type(chunk, map->type);
2183         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2184         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2185         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2186         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2187         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2188
2189         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2190         key.type = BTRFS_CHUNK_ITEM_KEY;
2191         key.offset = chunk_offset;
2192
2193         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2194         BUG_ON(ret);
2195
2196         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2197                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2198                                              item_size);
2199                 BUG_ON(ret);
2200         }
2201         kfree(chunk);
2202         return 0;
2203 }
2204
2205 /*
2206  * Chunk allocation falls into two parts. The first part does works
2207  * that make the new allocated chunk useable, but not do any operation
2208  * that modifies the chunk tree. The second part does the works that
2209  * require modifying the chunk tree. This division is important for the
2210  * bootstrap process of adding storage to a seed btrfs.
2211  */
2212 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2213                       struct btrfs_root *extent_root, u64 type)
2214 {
2215         u64 chunk_offset;
2216         u64 chunk_size;
2217         u64 stripe_size;
2218         struct map_lookup *map;
2219         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2220         int ret;
2221
2222         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2223                               &chunk_offset);
2224         if (ret)
2225                 return ret;
2226
2227         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2228                                   &stripe_size, chunk_offset, type);
2229         if (ret)
2230                 return ret;
2231
2232         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2233                                    chunk_size, stripe_size);
2234         BUG_ON(ret);
2235         return 0;
2236 }
2237
2238 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2239                                          struct btrfs_root *root,
2240                                          struct btrfs_device *device)
2241 {
2242         u64 chunk_offset;
2243         u64 sys_chunk_offset;
2244         u64 chunk_size;
2245         u64 sys_chunk_size;
2246         u64 stripe_size;
2247         u64 sys_stripe_size;
2248         u64 alloc_profile;
2249         struct map_lookup *map;
2250         struct map_lookup *sys_map;
2251         struct btrfs_fs_info *fs_info = root->fs_info;
2252         struct btrfs_root *extent_root = fs_info->extent_root;
2253         int ret;
2254
2255         ret = find_next_chunk(fs_info->chunk_root,
2256                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2257         BUG_ON(ret);
2258
2259         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2260                         (fs_info->metadata_alloc_profile &
2261                          fs_info->avail_metadata_alloc_bits);
2262         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2263
2264         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2265                                   &stripe_size, chunk_offset, alloc_profile);
2266         BUG_ON(ret);
2267
2268         sys_chunk_offset = chunk_offset + chunk_size;
2269
2270         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2271                         (fs_info->system_alloc_profile &
2272                          fs_info->avail_system_alloc_bits);
2273         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2274
2275         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2276                                   &sys_chunk_size, &sys_stripe_size,
2277                                   sys_chunk_offset, alloc_profile);
2278         BUG_ON(ret);
2279
2280         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2281         BUG_ON(ret);
2282
2283         /*
2284          * Modifying chunk tree needs allocating new blocks from both
2285          * system block group and metadata block group. So we only can
2286          * do operations require modifying the chunk tree after both
2287          * block groups were created.
2288          */
2289         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2290                                    chunk_size, stripe_size);
2291         BUG_ON(ret);
2292
2293         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2294                                    sys_chunk_offset, sys_chunk_size,
2295                                    sys_stripe_size);
2296         BUG_ON(ret);
2297         return 0;
2298 }
2299
2300 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2301 {
2302         struct extent_map *em;
2303         struct map_lookup *map;
2304         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2305         int readonly = 0;
2306         int i;
2307
2308         spin_lock(&map_tree->map_tree.lock);
2309         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2310         spin_unlock(&map_tree->map_tree.lock);
2311         if (!em)
2312                 return 1;
2313
2314         map = (struct map_lookup *)em->bdev;
2315         for (i = 0; i < map->num_stripes; i++) {
2316                 if (!map->stripes[i].dev->writeable) {
2317                         readonly = 1;
2318                         break;
2319                 }
2320         }
2321         free_extent_map(em);
2322         return readonly;
2323 }
2324
2325 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2326 {
2327         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2328 }
2329
2330 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2331 {
2332         struct extent_map *em;
2333
2334         while (1) {
2335                 spin_lock(&tree->map_tree.lock);
2336                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2337                 if (em)
2338                         remove_extent_mapping(&tree->map_tree, em);
2339                 spin_unlock(&tree->map_tree.lock);
2340                 if (!em)
2341                         break;
2342                 kfree(em->bdev);
2343                 /* once for us */
2344                 free_extent_map(em);
2345                 /* once for the tree */
2346                 free_extent_map(em);
2347         }
2348 }
2349
2350 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2351 {
2352         struct extent_map *em;
2353         struct map_lookup *map;
2354         struct extent_map_tree *em_tree = &map_tree->map_tree;
2355         int ret;
2356
2357         spin_lock(&em_tree->lock);
2358         em = lookup_extent_mapping(em_tree, logical, len);
2359         spin_unlock(&em_tree->lock);
2360         BUG_ON(!em);
2361
2362         BUG_ON(em->start > logical || em->start + em->len < logical);
2363         map = (struct map_lookup *)em->bdev;
2364         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2365                 ret = map->num_stripes;
2366         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2367                 ret = map->sub_stripes;
2368         else
2369                 ret = 1;
2370         free_extent_map(em);
2371         return ret;
2372 }
2373
2374 static int find_live_mirror(struct map_lookup *map, int first, int num,
2375                             int optimal)
2376 {
2377         int i;
2378         if (map->stripes[optimal].dev->bdev)
2379                 return optimal;
2380         for (i = first; i < first + num; i++) {
2381                 if (map->stripes[i].dev->bdev)
2382                         return i;
2383         }
2384         /* we couldn't find one that doesn't fail.  Just return something
2385          * and the io error handling code will clean up eventually
2386          */
2387         return optimal;
2388 }
2389
2390 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2391                              u64 logical, u64 *length,
2392                              struct btrfs_multi_bio **multi_ret,
2393                              int mirror_num, struct page *unplug_page)
2394 {
2395         struct extent_map *em;
2396         struct map_lookup *map;
2397         struct extent_map_tree *em_tree = &map_tree->map_tree;
2398         u64 offset;
2399         u64 stripe_offset;
2400         u64 stripe_nr;
2401         int stripes_allocated = 8;
2402         int stripes_required = 1;
2403         int stripe_index;
2404         int i;
2405         int num_stripes;
2406         int max_errors = 0;
2407         struct btrfs_multi_bio *multi = NULL;
2408
2409         if (multi_ret && !(rw & (1 << BIO_RW)))
2410                 stripes_allocated = 1;
2411 again:
2412         if (multi_ret) {
2413                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2414                                 GFP_NOFS);
2415                 if (!multi)
2416                         return -ENOMEM;
2417
2418                 atomic_set(&multi->error, 0);
2419         }
2420
2421         spin_lock(&em_tree->lock);
2422         em = lookup_extent_mapping(em_tree, logical, *length);
2423         spin_unlock(&em_tree->lock);
2424
2425         if (!em && unplug_page)
2426                 return 0;
2427
2428         if (!em) {
2429                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2430                        (unsigned long long)logical,
2431                        (unsigned long long)*length);
2432                 BUG();
2433         }
2434
2435         BUG_ON(em->start > logical || em->start + em->len < logical);
2436         map = (struct map_lookup *)em->bdev;
2437         offset = logical - em->start;
2438
2439         if (mirror_num > map->num_stripes)
2440                 mirror_num = 0;
2441
2442         /* if our multi bio struct is too small, back off and try again */
2443         if (rw & (1 << BIO_RW)) {
2444                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2445                                  BTRFS_BLOCK_GROUP_DUP)) {
2446                         stripes_required = map->num_stripes;
2447                         max_errors = 1;
2448                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2449                         stripes_required = map->sub_stripes;
2450                         max_errors = 1;
2451                 }
2452         }
2453         if (multi_ret && rw == WRITE &&
2454             stripes_allocated < stripes_required) {
2455                 stripes_allocated = map->num_stripes;
2456                 free_extent_map(em);
2457                 kfree(multi);
2458                 goto again;
2459         }
2460         stripe_nr = offset;
2461         /*
2462          * stripe_nr counts the total number of stripes we have to stride
2463          * to get to this block
2464          */
2465         do_div(stripe_nr, map->stripe_len);
2466
2467         stripe_offset = stripe_nr * map->stripe_len;
2468         BUG_ON(offset < stripe_offset);
2469
2470         /* stripe_offset is the offset of this block in its stripe*/
2471         stripe_offset = offset - stripe_offset;
2472
2473         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2474                          BTRFS_BLOCK_GROUP_RAID10 |
2475                          BTRFS_BLOCK_GROUP_DUP)) {
2476                 /* we limit the length of each bio to what fits in a stripe */
2477                 *length = min_t(u64, em->len - offset,
2478                               map->stripe_len - stripe_offset);
2479         } else {
2480                 *length = em->len - offset;
2481         }
2482
2483         if (!multi_ret && !unplug_page)
2484                 goto out;
2485
2486         num_stripes = 1;
2487         stripe_index = 0;
2488         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2489                 if (unplug_page || (rw & (1 << BIO_RW)))
2490                         num_stripes = map->num_stripes;
2491                 else if (mirror_num)
2492                         stripe_index = mirror_num - 1;
2493                 else {
2494                         stripe_index = find_live_mirror(map, 0,
2495                                             map->num_stripes,
2496                                             current->pid % map->num_stripes);
2497                 }
2498
2499         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2500                 if (rw & (1 << BIO_RW))
2501                         num_stripes = map->num_stripes;
2502                 else if (mirror_num)
2503                         stripe_index = mirror_num - 1;
2504
2505         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2506                 int factor = map->num_stripes / map->sub_stripes;
2507
2508                 stripe_index = do_div(stripe_nr, factor);
2509                 stripe_index *= map->sub_stripes;
2510
2511                 if (unplug_page || (rw & (1 << BIO_RW)))
2512                         num_stripes = map->sub_stripes;
2513                 else if (mirror_num)
2514                         stripe_index += mirror_num - 1;
2515                 else {
2516                         stripe_index = find_live_mirror(map, stripe_index,
2517                                               map->sub_stripes, stripe_index +
2518                                               current->pid % map->sub_stripes);
2519                 }
2520         } else {
2521                 /*
2522                  * after this do_div call, stripe_nr is the number of stripes
2523                  * on this device we have to walk to find the data, and
2524                  * stripe_index is the number of our device in the stripe array
2525                  */
2526                 stripe_index = do_div(stripe_nr, map->num_stripes);
2527         }
2528         BUG_ON(stripe_index >= map->num_stripes);
2529
2530         for (i = 0; i < num_stripes; i++) {
2531                 if (unplug_page) {
2532                         struct btrfs_device *device;
2533                         struct backing_dev_info *bdi;
2534
2535                         device = map->stripes[stripe_index].dev;
2536                         if (device->bdev) {
2537                                 bdi = blk_get_backing_dev_info(device->bdev);
2538                                 if (bdi->unplug_io_fn)
2539                                         bdi->unplug_io_fn(bdi, unplug_page);
2540                         }
2541                 } else {
2542                         multi->stripes[i].physical =
2543                                 map->stripes[stripe_index].physical +
2544                                 stripe_offset + stripe_nr * map->stripe_len;
2545                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2546                 }
2547                 stripe_index++;
2548         }
2549         if (multi_ret) {
2550                 *multi_ret = multi;
2551                 multi->num_stripes = num_stripes;
2552                 multi->max_errors = max_errors;
2553         }
2554 out:
2555         free_extent_map(em);
2556         return 0;
2557 }
2558
2559 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2560                       u64 logical, u64 *length,
2561                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2562 {
2563         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2564                                  mirror_num, NULL);
2565 }
2566
2567 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2568                      u64 chunk_start, u64 physical, u64 devid,
2569                      u64 **logical, int *naddrs, int *stripe_len)
2570 {
2571         struct extent_map_tree *em_tree = &map_tree->map_tree;
2572         struct extent_map *em;
2573         struct map_lookup *map;
2574         u64 *buf;
2575         u64 bytenr;
2576         u64 length;
2577         u64 stripe_nr;
2578         int i, j, nr = 0;
2579
2580         spin_lock(&em_tree->lock);
2581         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2582         spin_unlock(&em_tree->lock);
2583
2584         BUG_ON(!em || em->start != chunk_start);
2585         map = (struct map_lookup *)em->bdev;
2586
2587         length = em->len;
2588         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2589                 do_div(length, map->num_stripes / map->sub_stripes);
2590         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2591                 do_div(length, map->num_stripes);
2592
2593         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2594         BUG_ON(!buf);
2595
2596         for (i = 0; i < map->num_stripes; i++) {
2597                 if (devid && map->stripes[i].dev->devid != devid)
2598                         continue;
2599                 if (map->stripes[i].physical > physical ||
2600                     map->stripes[i].physical + length <= physical)
2601                         continue;
2602
2603                 stripe_nr = physical - map->stripes[i].physical;
2604                 do_div(stripe_nr, map->stripe_len);
2605
2606                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2607                         stripe_nr = stripe_nr * map->num_stripes + i;
2608                         do_div(stripe_nr, map->sub_stripes);
2609                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2610                         stripe_nr = stripe_nr * map->num_stripes + i;
2611                 }
2612                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2613                 WARN_ON(nr >= map->num_stripes);
2614                 for (j = 0; j < nr; j++) {
2615                         if (buf[j] == bytenr)
2616                                 break;
2617                 }
2618                 if (j == nr) {
2619                         WARN_ON(nr >= map->num_stripes);
2620                         buf[nr++] = bytenr;
2621                 }
2622         }
2623
2624         for (i = 0; i > nr; i++) {
2625                 struct btrfs_multi_bio *multi;
2626                 struct btrfs_bio_stripe *stripe;
2627                 int ret;
2628
2629                 length = 1;
2630                 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2631                                       &length, &multi, 0);
2632                 BUG_ON(ret);
2633
2634                 stripe = multi->stripes;
2635                 for (j = 0; j < multi->num_stripes; j++) {
2636                         if (stripe->physical >= physical &&
2637                             physical < stripe->physical + length)
2638                                 break;
2639                 }
2640                 BUG_ON(j >= multi->num_stripes);
2641                 kfree(multi);
2642         }
2643
2644         *logical = buf;
2645         *naddrs = nr;
2646         *stripe_len = map->stripe_len;
2647
2648         free_extent_map(em);
2649         return 0;
2650 }
2651
2652 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2653                       u64 logical, struct page *page)
2654 {
2655         u64 length = PAGE_CACHE_SIZE;
2656         return __btrfs_map_block(map_tree, READ, logical, &length,
2657                                  NULL, 0, page);
2658 }
2659
2660 static void end_bio_multi_stripe(struct bio *bio, int err)
2661 {
2662         struct btrfs_multi_bio *multi = bio->bi_private;
2663         int is_orig_bio = 0;
2664
2665         if (err)
2666                 atomic_inc(&multi->error);
2667
2668         if (bio == multi->orig_bio)
2669                 is_orig_bio = 1;
2670
2671         if (atomic_dec_and_test(&multi->stripes_pending)) {
2672                 if (!is_orig_bio) {
2673                         bio_put(bio);
2674                         bio = multi->orig_bio;
2675                 }
2676                 bio->bi_private = multi->private;
2677                 bio->bi_end_io = multi->end_io;
2678                 /* only send an error to the higher layers if it is
2679                  * beyond the tolerance of the multi-bio
2680                  */
2681                 if (atomic_read(&multi->error) > multi->max_errors) {
2682                         err = -EIO;
2683                 } else if (err) {
2684                         /*
2685                          * this bio is actually up to date, we didn't
2686                          * go over the max number of errors
2687                          */
2688                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2689                         err = 0;
2690                 }
2691                 kfree(multi);
2692
2693                 bio_endio(bio, err);
2694         } else if (!is_orig_bio) {
2695                 bio_put(bio);
2696         }
2697 }
2698
2699 struct async_sched {
2700         struct bio *bio;
2701         int rw;
2702         struct btrfs_fs_info *info;
2703         struct btrfs_work work;
2704 };
2705
2706 /*
2707  * see run_scheduled_bios for a description of why bios are collected for
2708  * async submit.
2709  *
2710  * This will add one bio to the pending list for a device and make sure
2711  * the work struct is scheduled.
2712  */
2713 static noinline int schedule_bio(struct btrfs_root *root,
2714                                  struct btrfs_device *device,
2715                                  int rw, struct bio *bio)
2716 {
2717         int should_queue = 1;
2718
2719         /* don't bother with additional async steps for reads, right now */
2720         if (!(rw & (1 << BIO_RW))) {
2721                 bio_get(bio);
2722                 submit_bio(rw, bio);
2723                 bio_put(bio);
2724                 return 0;
2725         }
2726
2727         /*
2728          * nr_async_bios allows us to reliably return congestion to the
2729          * higher layers.  Otherwise, the async bio makes it appear we have
2730          * made progress against dirty pages when we've really just put it
2731          * on a queue for later
2732          */
2733         atomic_inc(&root->fs_info->nr_async_bios);
2734         WARN_ON(bio->bi_next);
2735         bio->bi_next = NULL;
2736         bio->bi_rw |= rw;
2737
2738         spin_lock(&device->io_lock);
2739
2740         if (device->pending_bio_tail)
2741                 device->pending_bio_tail->bi_next = bio;
2742
2743         device->pending_bio_tail = bio;
2744         if (!device->pending_bios)
2745                 device->pending_bios = bio;
2746         if (device->running_pending)
2747                 should_queue = 0;
2748
2749         spin_unlock(&device->io_lock);
2750
2751         if (should_queue)
2752                 btrfs_queue_worker(&root->fs_info->submit_workers,
2753                                    &device->work);
2754         return 0;
2755 }
2756
2757 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2758                   int mirror_num, int async_submit)
2759 {
2760         struct btrfs_mapping_tree *map_tree;
2761         struct btrfs_device *dev;
2762         struct bio *first_bio = bio;
2763         u64 logical = (u64)bio->bi_sector << 9;
2764         u64 length = 0;
2765         u64 map_length;
2766         struct btrfs_multi_bio *multi = NULL;
2767         int ret;
2768         int dev_nr = 0;
2769         int total_devs = 1;
2770
2771         length = bio->bi_size;
2772         map_tree = &root->fs_info->mapping_tree;
2773         map_length = length;
2774
2775         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2776                               mirror_num);
2777         BUG_ON(ret);
2778
2779         total_devs = multi->num_stripes;
2780         if (map_length < length) {
2781                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2782                        "len %llu\n", (unsigned long long)logical,
2783                        (unsigned long long)length,
2784                        (unsigned long long)map_length);
2785                 BUG();
2786         }
2787         multi->end_io = first_bio->bi_end_io;
2788         multi->private = first_bio->bi_private;
2789         multi->orig_bio = first_bio;
2790         atomic_set(&multi->stripes_pending, multi->num_stripes);
2791
2792         while (dev_nr < total_devs) {
2793                 if (total_devs > 1) {
2794                         if (dev_nr < total_devs - 1) {
2795                                 bio = bio_clone(first_bio, GFP_NOFS);
2796                                 BUG_ON(!bio);
2797                         } else {
2798                                 bio = first_bio;
2799                         }
2800                         bio->bi_private = multi;
2801                         bio->bi_end_io = end_bio_multi_stripe;
2802                 }
2803                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2804                 dev = multi->stripes[dev_nr].dev;
2805                 BUG_ON(rw == WRITE && !dev->writeable);
2806                 if (dev && dev->bdev) {
2807                         bio->bi_bdev = dev->bdev;
2808                         if (async_submit)
2809                                 schedule_bio(root, dev, rw, bio);
2810                         else
2811                                 submit_bio(rw, bio);
2812                 } else {
2813                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2814                         bio->bi_sector = logical >> 9;
2815                         bio_endio(bio, -EIO);
2816                 }
2817                 dev_nr++;
2818         }
2819         if (total_devs == 1)
2820                 kfree(multi);
2821         return 0;
2822 }
2823
2824 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2825                                        u8 *uuid, u8 *fsid)
2826 {
2827         struct btrfs_device *device;
2828         struct btrfs_fs_devices *cur_devices;
2829
2830         cur_devices = root->fs_info->fs_devices;
2831         while (cur_devices) {
2832                 if (!fsid ||
2833                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2834                         device = __find_device(&cur_devices->devices,
2835                                                devid, uuid);
2836                         if (device)
2837                                 return device;
2838                 }
2839                 cur_devices = cur_devices->seed;
2840         }
2841         return NULL;
2842 }
2843
2844 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2845                                             u64 devid, u8 *dev_uuid)
2846 {
2847         struct btrfs_device *device;
2848         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2849
2850         device = kzalloc(sizeof(*device), GFP_NOFS);
2851         if (!device)
2852                 return NULL;
2853         list_add(&device->dev_list,
2854                  &fs_devices->devices);
2855         device->barriers = 1;
2856         device->dev_root = root->fs_info->dev_root;
2857         device->devid = devid;
2858         device->work.func = pending_bios_fn;
2859         device->fs_devices = fs_devices;
2860         fs_devices->num_devices++;
2861         spin_lock_init(&device->io_lock);
2862         INIT_LIST_HEAD(&device->dev_alloc_list);
2863         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2864         return device;
2865 }
2866
2867 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2868                           struct extent_buffer *leaf,
2869                           struct btrfs_chunk *chunk)
2870 {
2871         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2872         struct map_lookup *map;
2873         struct extent_map *em;
2874         u64 logical;
2875         u64 length;
2876         u64 devid;
2877         u8 uuid[BTRFS_UUID_SIZE];
2878         int num_stripes;
2879         int ret;
2880         int i;
2881
2882         logical = key->offset;
2883         length = btrfs_chunk_length(leaf, chunk);
2884
2885         spin_lock(&map_tree->map_tree.lock);
2886         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2887         spin_unlock(&map_tree->map_tree.lock);
2888
2889         /* already mapped? */
2890         if (em && em->start <= logical && em->start + em->len > logical) {
2891                 free_extent_map(em);
2892                 return 0;
2893         } else if (em) {
2894                 free_extent_map(em);
2895         }
2896
2897         em = alloc_extent_map(GFP_NOFS);
2898         if (!em)
2899                 return -ENOMEM;
2900         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2901         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2902         if (!map) {
2903                 free_extent_map(em);
2904                 return -ENOMEM;
2905         }
2906
2907         em->bdev = (struct block_device *)map;
2908         em->start = logical;
2909         em->len = length;
2910         em->block_start = 0;
2911         em->block_len = em->len;
2912
2913         map->num_stripes = num_stripes;
2914         map->io_width = btrfs_chunk_io_width(leaf, chunk);
2915         map->io_align = btrfs_chunk_io_align(leaf, chunk);
2916         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2917         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2918         map->type = btrfs_chunk_type(leaf, chunk);
2919         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2920         for (i = 0; i < num_stripes; i++) {
2921                 map->stripes[i].physical =
2922                         btrfs_stripe_offset_nr(leaf, chunk, i);
2923                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2924                 read_extent_buffer(leaf, uuid, (unsigned long)
2925                                    btrfs_stripe_dev_uuid_nr(chunk, i),
2926                                    BTRFS_UUID_SIZE);
2927                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2928                                                         NULL);
2929                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2930                         kfree(map);
2931                         free_extent_map(em);
2932                         return -EIO;
2933                 }
2934                 if (!map->stripes[i].dev) {
2935                         map->stripes[i].dev =
2936                                 add_missing_dev(root, devid, uuid);
2937                         if (!map->stripes[i].dev) {
2938                                 kfree(map);
2939                                 free_extent_map(em);
2940                                 return -EIO;
2941                         }
2942                 }
2943                 map->stripes[i].dev->in_fs_metadata = 1;
2944         }
2945
2946         spin_lock(&map_tree->map_tree.lock);
2947         ret = add_extent_mapping(&map_tree->map_tree, em);
2948         spin_unlock(&map_tree->map_tree.lock);
2949         BUG_ON(ret);
2950         free_extent_map(em);
2951
2952         return 0;
2953 }
2954
2955 static int fill_device_from_item(struct extent_buffer *leaf,
2956                                  struct btrfs_dev_item *dev_item,
2957                                  struct btrfs_device *device)
2958 {
2959         unsigned long ptr;
2960
2961         device->devid = btrfs_device_id(leaf, dev_item);
2962         device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2963         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2964         device->type = btrfs_device_type(leaf, dev_item);
2965         device->io_align = btrfs_device_io_align(leaf, dev_item);
2966         device->io_width = btrfs_device_io_width(leaf, dev_item);
2967         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2968
2969         ptr = (unsigned long)btrfs_device_uuid(dev_item);
2970         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2971
2972         return 0;
2973 }
2974
2975 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2976 {
2977         struct btrfs_fs_devices *fs_devices;
2978         int ret;
2979
2980         mutex_lock(&uuid_mutex);
2981
2982         fs_devices = root->fs_info->fs_devices->seed;
2983         while (fs_devices) {
2984                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2985                         ret = 0;
2986                         goto out;
2987                 }
2988                 fs_devices = fs_devices->seed;
2989         }
2990
2991         fs_devices = find_fsid(fsid);
2992         if (!fs_devices) {
2993                 ret = -ENOENT;
2994                 goto out;
2995         }
2996
2997         fs_devices = clone_fs_devices(fs_devices);
2998         if (IS_ERR(fs_devices)) {
2999                 ret = PTR_ERR(fs_devices);
3000                 goto out;
3001         }
3002
3003         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3004                                    root->fs_info->bdev_holder);
3005         if (ret)
3006                 goto out;
3007
3008         if (!fs_devices->seeding) {
3009                 __btrfs_close_devices(fs_devices);
3010                 free_fs_devices(fs_devices);
3011                 ret = -EINVAL;
3012                 goto out;
3013         }
3014
3015         fs_devices->seed = root->fs_info->fs_devices->seed;
3016         root->fs_info->fs_devices->seed = fs_devices;
3017 out:
3018         mutex_unlock(&uuid_mutex);
3019         return ret;
3020 }
3021
3022 static int read_one_dev(struct btrfs_root *root,
3023                         struct extent_buffer *leaf,
3024                         struct btrfs_dev_item *dev_item)
3025 {
3026         struct btrfs_device *device;
3027         u64 devid;
3028         int ret;
3029         u8 fs_uuid[BTRFS_UUID_SIZE];
3030         u8 dev_uuid[BTRFS_UUID_SIZE];
3031
3032         devid = btrfs_device_id(leaf, dev_item);
3033         read_extent_buffer(leaf, dev_uuid,
3034                            (unsigned long)btrfs_device_uuid(dev_item),
3035                            BTRFS_UUID_SIZE);
3036         read_extent_buffer(leaf, fs_uuid,
3037                            (unsigned long)btrfs_device_fsid(dev_item),
3038                            BTRFS_UUID_SIZE);
3039
3040         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3041                 ret = open_seed_devices(root, fs_uuid);
3042                 if (ret && !btrfs_test_opt(root, DEGRADED))
3043                         return ret;
3044         }
3045
3046         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3047         if (!device || !device->bdev) {
3048                 if (!btrfs_test_opt(root, DEGRADED))
3049                         return -EIO;
3050
3051                 if (!device) {
3052                         printk(KERN_WARNING "warning devid %llu missing\n",
3053                                (unsigned long long)devid);
3054                         device = add_missing_dev(root, devid, dev_uuid);
3055                         if (!device)
3056                                 return -ENOMEM;
3057                 }
3058         }
3059
3060         if (device->fs_devices != root->fs_info->fs_devices) {
3061                 BUG_ON(device->writeable);
3062                 if (device->generation !=
3063                     btrfs_device_generation(leaf, dev_item))
3064                         return -EINVAL;
3065         }
3066
3067         fill_device_from_item(leaf, dev_item, device);
3068         device->dev_root = root->fs_info->dev_root;
3069         device->in_fs_metadata = 1;
3070         if (device->writeable)
3071                 device->fs_devices->total_rw_bytes += device->total_bytes;
3072         ret = 0;
3073         return ret;
3074 }
3075
3076 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3077 {
3078         struct btrfs_dev_item *dev_item;
3079
3080         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3081                                                      dev_item);
3082         return read_one_dev(root, buf, dev_item);
3083 }
3084
3085 int btrfs_read_sys_array(struct btrfs_root *root)
3086 {
3087         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3088         struct extent_buffer *sb;
3089         struct btrfs_disk_key *disk_key;
3090         struct btrfs_chunk *chunk;
3091         u8 *ptr;
3092         unsigned long sb_ptr;
3093         int ret = 0;
3094         u32 num_stripes;
3095         u32 array_size;
3096         u32 len = 0;
3097         u32 cur;
3098         struct btrfs_key key;
3099
3100         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3101                                           BTRFS_SUPER_INFO_SIZE);
3102         if (!sb)
3103                 return -ENOMEM;
3104         btrfs_set_buffer_uptodate(sb);
3105         btrfs_set_buffer_lockdep_class(sb, 0);
3106
3107         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3108         array_size = btrfs_super_sys_array_size(super_copy);
3109
3110         ptr = super_copy->sys_chunk_array;
3111         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3112         cur = 0;
3113
3114         while (cur < array_size) {
3115                 disk_key = (struct btrfs_disk_key *)ptr;
3116                 btrfs_disk_key_to_cpu(&key, disk_key);
3117
3118                 len = sizeof(*disk_key); ptr += len;
3119                 sb_ptr += len;
3120                 cur += len;
3121
3122                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3123                         chunk = (struct btrfs_chunk *)sb_ptr;
3124                         ret = read_one_chunk(root, &key, sb, chunk);
3125                         if (ret)
3126                                 break;
3127                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3128                         len = btrfs_chunk_item_size(num_stripes);
3129                 } else {
3130                         ret = -EIO;
3131                         break;
3132                 }
3133                 ptr += len;
3134                 sb_ptr += len;
3135                 cur += len;
3136         }
3137         free_extent_buffer(sb);
3138         return ret;
3139 }
3140
3141 int btrfs_read_chunk_tree(struct btrfs_root *root)
3142 {
3143         struct btrfs_path *path;
3144         struct extent_buffer *leaf;
3145         struct btrfs_key key;
3146         struct btrfs_key found_key;
3147         int ret;
3148         int slot;
3149
3150         root = root->fs_info->chunk_root;
3151
3152         path = btrfs_alloc_path();
3153         if (!path)
3154                 return -ENOMEM;
3155
3156         /* first we search for all of the device items, and then we
3157          * read in all of the chunk items.  This way we can create chunk
3158          * mappings that reference all of the devices that are afound
3159          */
3160         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3161         key.offset = 0;
3162         key.type = 0;
3163 again:
3164         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3165         while (1) {
3166                 leaf = path->nodes[0];
3167                 slot = path->slots[0];
3168                 if (slot >= btrfs_header_nritems(leaf)) {
3169                         ret = btrfs_next_leaf(root, path);
3170                         if (ret == 0)
3171                                 continue;
3172                         if (ret < 0)
3173                                 goto error;
3174                         break;
3175                 }
3176                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3177                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3178                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3179                                 break;
3180                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3181                                 struct btrfs_dev_item *dev_item;
3182                                 dev_item = btrfs_item_ptr(leaf, slot,
3183                                                   struct btrfs_dev_item);
3184                                 ret = read_one_dev(root, leaf, dev_item);
3185                                 if (ret)
3186                                         goto error;
3187                         }
3188                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3189                         struct btrfs_chunk *chunk;
3190                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3191                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3192                         if (ret)
3193                                 goto error;
3194                 }
3195                 path->slots[0]++;
3196         }
3197         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3198                 key.objectid = 0;
3199                 btrfs_release_path(root, path);
3200                 goto again;
3201         }
3202         ret = 0;
3203 error:
3204         btrfs_free_path(path);
3205         return ret;
3206 }