Btrfs: do not mark the chunk as readonly if in degraded mode
[linux-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 <linux/iocontext.h>
24 #include <asm/div64.h>
25 #include "compat.h"
26 #include "ctree.h"
27 #include "extent_map.h"
28 #include "disk-io.h"
29 #include "transaction.h"
30 #include "print-tree.h"
31 #include "volumes.h"
32 #include "async-thread.h"
33
34 struct map_lookup {
35         u64 type;
36         int io_align;
37         int io_width;
38         int stripe_len;
39         int sector_size;
40         int num_stripes;
41         int sub_stripes;
42         struct btrfs_bio_stripe stripes[];
43 };
44
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46                                 struct btrfs_root *root,
47                                 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51                             (sizeof(struct btrfs_bio_stripe) * (n)))
52
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
55
56 void btrfs_lock_volumes(void)
57 {
58         mutex_lock(&uuid_mutex);
59 }
60
61 void btrfs_unlock_volumes(void)
62 {
63         mutex_unlock(&uuid_mutex);
64 }
65
66 static void lock_chunks(struct btrfs_root *root)
67 {
68         mutex_lock(&root->fs_info->chunk_mutex);
69 }
70
71 static void unlock_chunks(struct btrfs_root *root)
72 {
73         mutex_unlock(&root->fs_info->chunk_mutex);
74 }
75
76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
77 {
78         struct btrfs_device *device;
79         WARN_ON(fs_devices->opened);
80         while (!list_empty(&fs_devices->devices)) {
81                 device = list_entry(fs_devices->devices.next,
82                                     struct btrfs_device, dev_list);
83                 list_del(&device->dev_list);
84                 kfree(device->name);
85                 kfree(device);
86         }
87         kfree(fs_devices);
88 }
89
90 int btrfs_cleanup_fs_uuids(void)
91 {
92         struct btrfs_fs_devices *fs_devices;
93
94         while (!list_empty(&fs_uuids)) {
95                 fs_devices = list_entry(fs_uuids.next,
96                                         struct btrfs_fs_devices, list);
97                 list_del(&fs_devices->list);
98                 free_fs_devices(fs_devices);
99         }
100         return 0;
101 }
102
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
104                                                    u64 devid, u8 *uuid)
105 {
106         struct btrfs_device *dev;
107
108         list_for_each_entry(dev, head, dev_list) {
109                 if (dev->devid == devid &&
110                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
111                         return dev;
112                 }
113         }
114         return NULL;
115 }
116
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
118 {
119         struct btrfs_fs_devices *fs_devices;
120
121         list_for_each_entry(fs_devices, &fs_uuids, list) {
122                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
123                         return fs_devices;
124         }
125         return NULL;
126 }
127
128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
129                         struct bio *head, struct bio *tail)
130 {
131
132         struct bio *old_head;
133
134         old_head = pending_bios->head;
135         pending_bios->head = head;
136         if (pending_bios->tail)
137                 tail->bi_next = old_head;
138         else
139                 pending_bios->tail = tail;
140 }
141
142 /*
143  * we try to collect pending bios for a device so we don't get a large
144  * number of procs sending bios down to the same device.  This greatly
145  * improves the schedulers ability to collect and merge the bios.
146  *
147  * But, it also turns into a long list of bios to process and that is sure
148  * to eventually make the worker thread block.  The solution here is to
149  * make some progress and then put this work struct back at the end of
150  * the list if the block device is congested.  This way, multiple devices
151  * can make progress from a single worker thread.
152  */
153 static noinline int run_scheduled_bios(struct btrfs_device *device)
154 {
155         struct bio *pending;
156         struct backing_dev_info *bdi;
157         struct btrfs_fs_info *fs_info;
158         struct btrfs_pending_bios *pending_bios;
159         struct bio *tail;
160         struct bio *cur;
161         int again = 0;
162         unsigned long num_run;
163         unsigned long num_sync_run;
164         unsigned long batch_run = 0;
165         unsigned long limit;
166         unsigned long last_waited = 0;
167         int force_reg = 0;
168
169         bdi = blk_get_backing_dev_info(device->bdev);
170         fs_info = device->dev_root->fs_info;
171         limit = btrfs_async_submit_limit(fs_info);
172         limit = limit * 2 / 3;
173
174         /* we want to make sure that every time we switch from the sync
175          * list to the normal list, we unplug
176          */
177         num_sync_run = 0;
178
179 loop:
180         spin_lock(&device->io_lock);
181
182 loop_lock:
183         num_run = 0;
184
185         /* take all the bios off the list at once and process them
186          * later on (without the lock held).  But, remember the
187          * tail and other pointers so the bios can be properly reinserted
188          * into the list if we hit congestion
189          */
190         if (!force_reg && device->pending_sync_bios.head) {
191                 pending_bios = &device->pending_sync_bios;
192                 force_reg = 1;
193         } else {
194                 pending_bios = &device->pending_bios;
195                 force_reg = 0;
196         }
197
198         pending = pending_bios->head;
199         tail = pending_bios->tail;
200         WARN_ON(pending && !tail);
201
202         /*
203          * if pending was null this time around, no bios need processing
204          * at all and we can stop.  Otherwise it'll loop back up again
205          * and do an additional check so no bios are missed.
206          *
207          * device->running_pending is used to synchronize with the
208          * schedule_bio code.
209          */
210         if (device->pending_sync_bios.head == NULL &&
211             device->pending_bios.head == NULL) {
212                 again = 0;
213                 device->running_pending = 0;
214         } else {
215                 again = 1;
216                 device->running_pending = 1;
217         }
218
219         pending_bios->head = NULL;
220         pending_bios->tail = NULL;
221
222         spin_unlock(&device->io_lock);
223
224         /*
225          * if we're doing the regular priority list, make sure we unplug
226          * for any high prio bios we've sent down
227          */
228         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
229                 num_sync_run = 0;
230                 blk_run_backing_dev(bdi, NULL);
231         }
232
233         while (pending) {
234
235                 rmb();
236                 /* we want to work on both lists, but do more bios on the
237                  * sync list than the regular list
238                  */
239                 if ((num_run > 32 &&
240                     pending_bios != &device->pending_sync_bios &&
241                     device->pending_sync_bios.head) ||
242                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
243                     device->pending_bios.head)) {
244                         spin_lock(&device->io_lock);
245                         requeue_list(pending_bios, pending, tail);
246                         goto loop_lock;
247                 }
248
249                 cur = pending;
250                 pending = pending->bi_next;
251                 cur->bi_next = NULL;
252                 atomic_dec(&fs_info->nr_async_bios);
253
254                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
255                     waitqueue_active(&fs_info->async_submit_wait))
256                         wake_up(&fs_info->async_submit_wait);
257
258                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
259                 submit_bio(cur->bi_rw, cur);
260                 num_run++;
261                 batch_run++;
262
263                 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
264                         num_sync_run++;
265
266                 if (need_resched()) {
267                         if (num_sync_run) {
268                                 blk_run_backing_dev(bdi, NULL);
269                                 num_sync_run = 0;
270                         }
271                         cond_resched();
272                 }
273
274                 /*
275                  * we made progress, there is more work to do and the bdi
276                  * is now congested.  Back off and let other work structs
277                  * run instead
278                  */
279                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
280                     fs_info->fs_devices->open_devices > 1) {
281                         struct io_context *ioc;
282
283                         ioc = current->io_context;
284
285                         /*
286                          * the main goal here is that we don't want to
287                          * block if we're going to be able to submit
288                          * more requests without blocking.
289                          *
290                          * This code does two great things, it pokes into
291                          * the elevator code from a filesystem _and_
292                          * it makes assumptions about how batching works.
293                          */
294                         if (ioc && ioc->nr_batch_requests > 0 &&
295                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
296                             (last_waited == 0 ||
297                              ioc->last_waited == last_waited)) {
298                                 /*
299                                  * we want to go through our batch of
300                                  * requests and stop.  So, we copy out
301                                  * the ioc->last_waited time and test
302                                  * against it before looping
303                                  */
304                                 last_waited = ioc->last_waited;
305                                 if (need_resched()) {
306                                         if (num_sync_run) {
307                                                 blk_run_backing_dev(bdi, NULL);
308                                                 num_sync_run = 0;
309                                         }
310                                         cond_resched();
311                                 }
312                                 continue;
313                         }
314                         spin_lock(&device->io_lock);
315                         requeue_list(pending_bios, pending, tail);
316                         device->running_pending = 1;
317
318                         spin_unlock(&device->io_lock);
319                         btrfs_requeue_work(&device->work);
320                         goto done;
321                 }
322         }
323
324         if (num_sync_run) {
325                 num_sync_run = 0;
326                 blk_run_backing_dev(bdi, NULL);
327         }
328
329         cond_resched();
330         if (again)
331                 goto loop;
332
333         spin_lock(&device->io_lock);
334         if (device->pending_bios.head || device->pending_sync_bios.head)
335                 goto loop_lock;
336         spin_unlock(&device->io_lock);
337
338         /*
339          * IO has already been through a long path to get here.  Checksumming,
340          * async helper threads, perhaps compression.  We've done a pretty
341          * good job of collecting a batch of IO and should just unplug
342          * the device right away.
343          *
344          * This will help anyone who is waiting on the IO, they might have
345          * already unplugged, but managed to do so before the bio they
346          * cared about found its way down here.
347          */
348         blk_run_backing_dev(bdi, NULL);
349 done:
350         return 0;
351 }
352
353 static void pending_bios_fn(struct btrfs_work *work)
354 {
355         struct btrfs_device *device;
356
357         device = container_of(work, struct btrfs_device, work);
358         run_scheduled_bios(device);
359 }
360
361 static noinline int device_list_add(const char *path,
362                            struct btrfs_super_block *disk_super,
363                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
364 {
365         struct btrfs_device *device;
366         struct btrfs_fs_devices *fs_devices;
367         u64 found_transid = btrfs_super_generation(disk_super);
368
369         fs_devices = find_fsid(disk_super->fsid);
370         if (!fs_devices) {
371                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
372                 if (!fs_devices)
373                         return -ENOMEM;
374                 INIT_LIST_HEAD(&fs_devices->devices);
375                 INIT_LIST_HEAD(&fs_devices->alloc_list);
376                 list_add(&fs_devices->list, &fs_uuids);
377                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
378                 fs_devices->latest_devid = devid;
379                 fs_devices->latest_trans = found_transid;
380                 mutex_init(&fs_devices->device_list_mutex);
381                 device = NULL;
382         } else {
383                 device = __find_device(&fs_devices->devices, devid,
384                                        disk_super->dev_item.uuid);
385         }
386         if (!device) {
387                 if (fs_devices->opened)
388                         return -EBUSY;
389
390                 device = kzalloc(sizeof(*device), GFP_NOFS);
391                 if (!device) {
392                         /* we can safely leave the fs_devices entry around */
393                         return -ENOMEM;
394                 }
395                 device->devid = devid;
396                 device->work.func = pending_bios_fn;
397                 memcpy(device->uuid, disk_super->dev_item.uuid,
398                        BTRFS_UUID_SIZE);
399                 device->barriers = 1;
400                 spin_lock_init(&device->io_lock);
401                 device->name = kstrdup(path, GFP_NOFS);
402                 if (!device->name) {
403                         kfree(device);
404                         return -ENOMEM;
405                 }
406                 INIT_LIST_HEAD(&device->dev_alloc_list);
407
408                 mutex_lock(&fs_devices->device_list_mutex);
409                 list_add(&device->dev_list, &fs_devices->devices);
410                 mutex_unlock(&fs_devices->device_list_mutex);
411
412                 device->fs_devices = fs_devices;
413                 fs_devices->num_devices++;
414         }
415
416         if (found_transid > fs_devices->latest_trans) {
417                 fs_devices->latest_devid = devid;
418                 fs_devices->latest_trans = found_transid;
419         }
420         *fs_devices_ret = fs_devices;
421         return 0;
422 }
423
424 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
425 {
426         struct btrfs_fs_devices *fs_devices;
427         struct btrfs_device *device;
428         struct btrfs_device *orig_dev;
429
430         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
431         if (!fs_devices)
432                 return ERR_PTR(-ENOMEM);
433
434         INIT_LIST_HEAD(&fs_devices->devices);
435         INIT_LIST_HEAD(&fs_devices->alloc_list);
436         INIT_LIST_HEAD(&fs_devices->list);
437         mutex_init(&fs_devices->device_list_mutex);
438         fs_devices->latest_devid = orig->latest_devid;
439         fs_devices->latest_trans = orig->latest_trans;
440         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
441
442         mutex_lock(&orig->device_list_mutex);
443         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
444                 device = kzalloc(sizeof(*device), GFP_NOFS);
445                 if (!device)
446                         goto error;
447
448                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
449                 if (!device->name) {
450                         kfree(device);
451                         goto error;
452                 }
453
454                 device->devid = orig_dev->devid;
455                 device->work.func = pending_bios_fn;
456                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
457                 device->barriers = 1;
458                 spin_lock_init(&device->io_lock);
459                 INIT_LIST_HEAD(&device->dev_list);
460                 INIT_LIST_HEAD(&device->dev_alloc_list);
461
462                 list_add(&device->dev_list, &fs_devices->devices);
463                 device->fs_devices = fs_devices;
464                 fs_devices->num_devices++;
465         }
466         mutex_unlock(&orig->device_list_mutex);
467         return fs_devices;
468 error:
469         mutex_unlock(&orig->device_list_mutex);
470         free_fs_devices(fs_devices);
471         return ERR_PTR(-ENOMEM);
472 }
473
474 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
475 {
476         struct btrfs_device *device, *next;
477
478         mutex_lock(&uuid_mutex);
479 again:
480         mutex_lock(&fs_devices->device_list_mutex);
481         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
482                 if (device->in_fs_metadata)
483                         continue;
484
485                 if (device->bdev) {
486                         close_bdev_exclusive(device->bdev, device->mode);
487                         device->bdev = NULL;
488                         fs_devices->open_devices--;
489                 }
490                 if (device->writeable) {
491                         list_del_init(&device->dev_alloc_list);
492                         device->writeable = 0;
493                         fs_devices->rw_devices--;
494                 }
495                 list_del_init(&device->dev_list);
496                 fs_devices->num_devices--;
497                 kfree(device->name);
498                 kfree(device);
499         }
500         mutex_unlock(&fs_devices->device_list_mutex);
501
502         if (fs_devices->seed) {
503                 fs_devices = fs_devices->seed;
504                 goto again;
505         }
506
507         mutex_unlock(&uuid_mutex);
508         return 0;
509 }
510
511 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
512 {
513         struct btrfs_device *device;
514
515         if (--fs_devices->opened > 0)
516                 return 0;
517
518         list_for_each_entry(device, &fs_devices->devices, dev_list) {
519                 if (device->bdev) {
520                         close_bdev_exclusive(device->bdev, device->mode);
521                         fs_devices->open_devices--;
522                 }
523                 if (device->writeable) {
524                         list_del_init(&device->dev_alloc_list);
525                         fs_devices->rw_devices--;
526                 }
527
528                 device->bdev = NULL;
529                 device->writeable = 0;
530                 device->in_fs_metadata = 0;
531         }
532         WARN_ON(fs_devices->open_devices);
533         WARN_ON(fs_devices->rw_devices);
534         fs_devices->opened = 0;
535         fs_devices->seeding = 0;
536
537         return 0;
538 }
539
540 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
541 {
542         struct btrfs_fs_devices *seed_devices = NULL;
543         int ret;
544
545         mutex_lock(&uuid_mutex);
546         ret = __btrfs_close_devices(fs_devices);
547         if (!fs_devices->opened) {
548                 seed_devices = fs_devices->seed;
549                 fs_devices->seed = NULL;
550         }
551         mutex_unlock(&uuid_mutex);
552
553         while (seed_devices) {
554                 fs_devices = seed_devices;
555                 seed_devices = fs_devices->seed;
556                 __btrfs_close_devices(fs_devices);
557                 free_fs_devices(fs_devices);
558         }
559         return ret;
560 }
561
562 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563                                 fmode_t flags, void *holder)
564 {
565         struct block_device *bdev;
566         struct list_head *head = &fs_devices->devices;
567         struct btrfs_device *device;
568         struct block_device *latest_bdev = NULL;
569         struct buffer_head *bh;
570         struct btrfs_super_block *disk_super;
571         u64 latest_devid = 0;
572         u64 latest_transid = 0;
573         u64 devid;
574         int seeding = 1;
575         int ret = 0;
576
577         list_for_each_entry(device, head, dev_list) {
578                 if (device->bdev)
579                         continue;
580                 if (!device->name)
581                         continue;
582
583                 bdev = open_bdev_exclusive(device->name, flags, holder);
584                 if (IS_ERR(bdev)) {
585                         printk(KERN_INFO "open %s failed\n", device->name);
586                         goto error;
587                 }
588                 set_blocksize(bdev, 4096);
589
590                 bh = btrfs_read_dev_super(bdev);
591                 if (!bh)
592                         goto error_close;
593
594                 disk_super = (struct btrfs_super_block *)bh->b_data;
595                 devid = le64_to_cpu(disk_super->dev_item.devid);
596                 if (devid != device->devid)
597                         goto error_brelse;
598
599                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
600                            BTRFS_UUID_SIZE))
601                         goto error_brelse;
602
603                 device->generation = btrfs_super_generation(disk_super);
604                 if (!latest_transid || device->generation > latest_transid) {
605                         latest_devid = devid;
606                         latest_transid = device->generation;
607                         latest_bdev = bdev;
608                 }
609
610                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
611                         device->writeable = 0;
612                 } else {
613                         device->writeable = !bdev_read_only(bdev);
614                         seeding = 0;
615                 }
616
617                 device->bdev = bdev;
618                 device->in_fs_metadata = 0;
619                 device->mode = flags;
620
621                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
622                         fs_devices->rotating = 1;
623
624                 fs_devices->open_devices++;
625                 if (device->writeable) {
626                         fs_devices->rw_devices++;
627                         list_add(&device->dev_alloc_list,
628                                  &fs_devices->alloc_list);
629                 }
630                 continue;
631
632 error_brelse:
633                 brelse(bh);
634 error_close:
635                 close_bdev_exclusive(bdev, FMODE_READ);
636 error:
637                 continue;
638         }
639         if (fs_devices->open_devices == 0) {
640                 ret = -EIO;
641                 goto out;
642         }
643         fs_devices->seeding = seeding;
644         fs_devices->opened = 1;
645         fs_devices->latest_bdev = latest_bdev;
646         fs_devices->latest_devid = latest_devid;
647         fs_devices->latest_trans = latest_transid;
648         fs_devices->total_rw_bytes = 0;
649 out:
650         return ret;
651 }
652
653 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
654                        fmode_t flags, void *holder)
655 {
656         int ret;
657
658         mutex_lock(&uuid_mutex);
659         if (fs_devices->opened) {
660                 fs_devices->opened++;
661                 ret = 0;
662         } else {
663                 ret = __btrfs_open_devices(fs_devices, flags, holder);
664         }
665         mutex_unlock(&uuid_mutex);
666         return ret;
667 }
668
669 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
670                           struct btrfs_fs_devices **fs_devices_ret)
671 {
672         struct btrfs_super_block *disk_super;
673         struct block_device *bdev;
674         struct buffer_head *bh;
675         int ret;
676         u64 devid;
677         u64 transid;
678
679         mutex_lock(&uuid_mutex);
680
681         bdev = open_bdev_exclusive(path, flags, holder);
682
683         if (IS_ERR(bdev)) {
684                 ret = PTR_ERR(bdev);
685                 goto error;
686         }
687
688         ret = set_blocksize(bdev, 4096);
689         if (ret)
690                 goto error_close;
691         bh = btrfs_read_dev_super(bdev);
692         if (!bh) {
693                 ret = -EIO;
694                 goto error_close;
695         }
696         disk_super = (struct btrfs_super_block *)bh->b_data;
697         devid = le64_to_cpu(disk_super->dev_item.devid);
698         transid = btrfs_super_generation(disk_super);
699         if (disk_super->label[0])
700                 printk(KERN_INFO "device label %s ", disk_super->label);
701         else {
702                 /* FIXME, make a readl uuid parser */
703                 printk(KERN_INFO "device fsid %llx-%llx ",
704                        *(unsigned long long *)disk_super->fsid,
705                        *(unsigned long long *)(disk_super->fsid + 8));
706         }
707         printk(KERN_CONT "devid %llu transid %llu %s\n",
708                (unsigned long long)devid, (unsigned long long)transid, path);
709         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
710
711         brelse(bh);
712 error_close:
713         close_bdev_exclusive(bdev, flags);
714 error:
715         mutex_unlock(&uuid_mutex);
716         return ret;
717 }
718
719 /*
720  * this uses a pretty simple search, the expectation is that it is
721  * called very infrequently and that a given device has a small number
722  * of extents
723  */
724 int find_free_dev_extent(struct btrfs_trans_handle *trans,
725                          struct btrfs_device *device, u64 num_bytes,
726                          u64 *start, u64 *max_avail)
727 {
728         struct btrfs_key key;
729         struct btrfs_root *root = device->dev_root;
730         struct btrfs_dev_extent *dev_extent = NULL;
731         struct btrfs_path *path;
732         u64 hole_size = 0;
733         u64 last_byte = 0;
734         u64 search_start = 0;
735         u64 search_end = device->total_bytes;
736         int ret;
737         int slot = 0;
738         int start_found;
739         struct extent_buffer *l;
740
741         path = btrfs_alloc_path();
742         if (!path)
743                 return -ENOMEM;
744         path->reada = 2;
745         start_found = 0;
746
747         /* FIXME use last free of some kind */
748
749         /* we don't want to overwrite the superblock on the drive,
750          * so we make sure to start at an offset of at least 1MB
751          */
752         search_start = max((u64)1024 * 1024, search_start);
753
754         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
755                 search_start = max(root->fs_info->alloc_start, search_start);
756
757         key.objectid = device->devid;
758         key.offset = search_start;
759         key.type = BTRFS_DEV_EXTENT_KEY;
760         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
761         if (ret < 0)
762                 goto error;
763         if (ret > 0) {
764                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
765                 if (ret < 0)
766                         goto error;
767                 if (ret > 0)
768                         start_found = 1;
769         }
770         l = path->nodes[0];
771         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
772         while (1) {
773                 l = path->nodes[0];
774                 slot = path->slots[0];
775                 if (slot >= btrfs_header_nritems(l)) {
776                         ret = btrfs_next_leaf(root, path);
777                         if (ret == 0)
778                                 continue;
779                         if (ret < 0)
780                                 goto error;
781 no_more_items:
782                         if (!start_found) {
783                                 if (search_start >= search_end) {
784                                         ret = -ENOSPC;
785                                         goto error;
786                                 }
787                                 *start = search_start;
788                                 start_found = 1;
789                                 goto check_pending;
790                         }
791                         *start = last_byte > search_start ?
792                                 last_byte : search_start;
793                         if (search_end <= *start) {
794                                 ret = -ENOSPC;
795                                 goto error;
796                         }
797                         goto check_pending;
798                 }
799                 btrfs_item_key_to_cpu(l, &key, slot);
800
801                 if (key.objectid < device->devid)
802                         goto next;
803
804                 if (key.objectid > device->devid)
805                         goto no_more_items;
806
807                 if (key.offset >= search_start && key.offset > last_byte &&
808                     start_found) {
809                         if (last_byte < search_start)
810                                 last_byte = search_start;
811                         hole_size = key.offset - last_byte;
812
813                         if (hole_size > *max_avail)
814                                 *max_avail = hole_size;
815
816                         if (key.offset > last_byte &&
817                             hole_size >= num_bytes) {
818                                 *start = last_byte;
819                                 goto check_pending;
820                         }
821                 }
822                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
823                         goto next;
824
825                 start_found = 1;
826                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
827                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
828 next:
829                 path->slots[0]++;
830                 cond_resched();
831         }
832 check_pending:
833         /* we have to make sure we didn't find an extent that has already
834          * been allocated by the map tree or the original allocation
835          */
836         BUG_ON(*start < search_start);
837
838         if (*start + num_bytes > search_end) {
839                 ret = -ENOSPC;
840                 goto error;
841         }
842         /* check for pending inserts here */
843         ret = 0;
844
845 error:
846         btrfs_free_path(path);
847         return ret;
848 }
849
850 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
851                           struct btrfs_device *device,
852                           u64 start)
853 {
854         int ret;
855         struct btrfs_path *path;
856         struct btrfs_root *root = device->dev_root;
857         struct btrfs_key key;
858         struct btrfs_key found_key;
859         struct extent_buffer *leaf = NULL;
860         struct btrfs_dev_extent *extent = NULL;
861
862         path = btrfs_alloc_path();
863         if (!path)
864                 return -ENOMEM;
865
866         key.objectid = device->devid;
867         key.offset = start;
868         key.type = BTRFS_DEV_EXTENT_KEY;
869
870         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
871         if (ret > 0) {
872                 ret = btrfs_previous_item(root, path, key.objectid,
873                                           BTRFS_DEV_EXTENT_KEY);
874                 BUG_ON(ret);
875                 leaf = path->nodes[0];
876                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
877                 extent = btrfs_item_ptr(leaf, path->slots[0],
878                                         struct btrfs_dev_extent);
879                 BUG_ON(found_key.offset > start || found_key.offset +
880                        btrfs_dev_extent_length(leaf, extent) < start);
881                 ret = 0;
882         } else if (ret == 0) {
883                 leaf = path->nodes[0];
884                 extent = btrfs_item_ptr(leaf, path->slots[0],
885                                         struct btrfs_dev_extent);
886         }
887         BUG_ON(ret);
888
889         if (device->bytes_used > 0)
890                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
891         ret = btrfs_del_item(trans, root, path);
892         BUG_ON(ret);
893
894         btrfs_free_path(path);
895         return ret;
896 }
897
898 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
899                            struct btrfs_device *device,
900                            u64 chunk_tree, u64 chunk_objectid,
901                            u64 chunk_offset, u64 start, u64 num_bytes)
902 {
903         int ret;
904         struct btrfs_path *path;
905         struct btrfs_root *root = device->dev_root;
906         struct btrfs_dev_extent *extent;
907         struct extent_buffer *leaf;
908         struct btrfs_key key;
909
910         WARN_ON(!device->in_fs_metadata);
911         path = btrfs_alloc_path();
912         if (!path)
913                 return -ENOMEM;
914
915         key.objectid = device->devid;
916         key.offset = start;
917         key.type = BTRFS_DEV_EXTENT_KEY;
918         ret = btrfs_insert_empty_item(trans, root, path, &key,
919                                       sizeof(*extent));
920         BUG_ON(ret);
921
922         leaf = path->nodes[0];
923         extent = btrfs_item_ptr(leaf, path->slots[0],
924                                 struct btrfs_dev_extent);
925         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
926         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
927         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
928
929         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
930                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
931                     BTRFS_UUID_SIZE);
932
933         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
934         btrfs_mark_buffer_dirty(leaf);
935         btrfs_free_path(path);
936         return ret;
937 }
938
939 static noinline int find_next_chunk(struct btrfs_root *root,
940                                     u64 objectid, u64 *offset)
941 {
942         struct btrfs_path *path;
943         int ret;
944         struct btrfs_key key;
945         struct btrfs_chunk *chunk;
946         struct btrfs_key found_key;
947
948         path = btrfs_alloc_path();
949         BUG_ON(!path);
950
951         key.objectid = objectid;
952         key.offset = (u64)-1;
953         key.type = BTRFS_CHUNK_ITEM_KEY;
954
955         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
956         if (ret < 0)
957                 goto error;
958
959         BUG_ON(ret == 0);
960
961         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
962         if (ret) {
963                 *offset = 0;
964         } else {
965                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
966                                       path->slots[0]);
967                 if (found_key.objectid != objectid)
968                         *offset = 0;
969                 else {
970                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
971                                                struct btrfs_chunk);
972                         *offset = found_key.offset +
973                                 btrfs_chunk_length(path->nodes[0], chunk);
974                 }
975         }
976         ret = 0;
977 error:
978         btrfs_free_path(path);
979         return ret;
980 }
981
982 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
983 {
984         int ret;
985         struct btrfs_key key;
986         struct btrfs_key found_key;
987         struct btrfs_path *path;
988
989         root = root->fs_info->chunk_root;
990
991         path = btrfs_alloc_path();
992         if (!path)
993                 return -ENOMEM;
994
995         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
996         key.type = BTRFS_DEV_ITEM_KEY;
997         key.offset = (u64)-1;
998
999         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1000         if (ret < 0)
1001                 goto error;
1002
1003         BUG_ON(ret == 0);
1004
1005         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1006                                   BTRFS_DEV_ITEM_KEY);
1007         if (ret) {
1008                 *objectid = 1;
1009         } else {
1010                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1011                                       path->slots[0]);
1012                 *objectid = found_key.offset + 1;
1013         }
1014         ret = 0;
1015 error:
1016         btrfs_free_path(path);
1017         return ret;
1018 }
1019
1020 /*
1021  * the device information is stored in the chunk root
1022  * the btrfs_device struct should be fully filled in
1023  */
1024 int btrfs_add_device(struct btrfs_trans_handle *trans,
1025                      struct btrfs_root *root,
1026                      struct btrfs_device *device)
1027 {
1028         int ret;
1029         struct btrfs_path *path;
1030         struct btrfs_dev_item *dev_item;
1031         struct extent_buffer *leaf;
1032         struct btrfs_key key;
1033         unsigned long ptr;
1034
1035         root = root->fs_info->chunk_root;
1036
1037         path = btrfs_alloc_path();
1038         if (!path)
1039                 return -ENOMEM;
1040
1041         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1042         key.type = BTRFS_DEV_ITEM_KEY;
1043         key.offset = device->devid;
1044
1045         ret = btrfs_insert_empty_item(trans, root, path, &key,
1046                                       sizeof(*dev_item));
1047         if (ret)
1048                 goto out;
1049
1050         leaf = path->nodes[0];
1051         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1052
1053         btrfs_set_device_id(leaf, dev_item, device->devid);
1054         btrfs_set_device_generation(leaf, dev_item, 0);
1055         btrfs_set_device_type(leaf, dev_item, device->type);
1056         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1057         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1058         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1059         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1060         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1061         btrfs_set_device_group(leaf, dev_item, 0);
1062         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1063         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1064         btrfs_set_device_start_offset(leaf, dev_item, 0);
1065
1066         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1067         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1068         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1069         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1070         btrfs_mark_buffer_dirty(leaf);
1071
1072         ret = 0;
1073 out:
1074         btrfs_free_path(path);
1075         return ret;
1076 }
1077
1078 static int btrfs_rm_dev_item(struct btrfs_root *root,
1079                              struct btrfs_device *device)
1080 {
1081         int ret;
1082         struct btrfs_path *path;
1083         struct btrfs_key key;
1084         struct btrfs_trans_handle *trans;
1085
1086         root = root->fs_info->chunk_root;
1087
1088         path = btrfs_alloc_path();
1089         if (!path)
1090                 return -ENOMEM;
1091
1092         trans = btrfs_start_transaction(root, 1);
1093         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1094         key.type = BTRFS_DEV_ITEM_KEY;
1095         key.offset = device->devid;
1096         lock_chunks(root);
1097
1098         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099         if (ret < 0)
1100                 goto out;
1101
1102         if (ret > 0) {
1103                 ret = -ENOENT;
1104                 goto out;
1105         }
1106
1107         ret = btrfs_del_item(trans, root, path);
1108         if (ret)
1109                 goto out;
1110 out:
1111         btrfs_free_path(path);
1112         unlock_chunks(root);
1113         btrfs_commit_transaction(trans, root);
1114         return ret;
1115 }
1116
1117 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1118 {
1119         struct btrfs_device *device;
1120         struct btrfs_device *next_device;
1121         struct block_device *bdev;
1122         struct buffer_head *bh = NULL;
1123         struct btrfs_super_block *disk_super;
1124         u64 all_avail;
1125         u64 devid;
1126         u64 num_devices;
1127         u8 *dev_uuid;
1128         int ret = 0;
1129
1130         mutex_lock(&uuid_mutex);
1131         mutex_lock(&root->fs_info->volume_mutex);
1132
1133         all_avail = root->fs_info->avail_data_alloc_bits |
1134                 root->fs_info->avail_system_alloc_bits |
1135                 root->fs_info->avail_metadata_alloc_bits;
1136
1137         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1138             root->fs_info->fs_devices->rw_devices <= 4) {
1139                 printk(KERN_ERR "btrfs: unable to go below four devices "
1140                        "on raid10\n");
1141                 ret = -EINVAL;
1142                 goto out;
1143         }
1144
1145         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1146             root->fs_info->fs_devices->rw_devices <= 2) {
1147                 printk(KERN_ERR "btrfs: unable to go below two "
1148                        "devices on raid1\n");
1149                 ret = -EINVAL;
1150                 goto out;
1151         }
1152
1153         if (strcmp(device_path, "missing") == 0) {
1154                 struct list_head *devices;
1155                 struct btrfs_device *tmp;
1156
1157                 device = NULL;
1158                 devices = &root->fs_info->fs_devices->devices;
1159                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1160                 list_for_each_entry(tmp, devices, dev_list) {
1161                         if (tmp->in_fs_metadata && !tmp->bdev) {
1162                                 device = tmp;
1163                                 break;
1164                         }
1165                 }
1166                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1167                 bdev = NULL;
1168                 bh = NULL;
1169                 disk_super = NULL;
1170                 if (!device) {
1171                         printk(KERN_ERR "btrfs: no missing devices found to "
1172                                "remove\n");
1173                         goto out;
1174                 }
1175         } else {
1176                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1177                                       root->fs_info->bdev_holder);
1178                 if (IS_ERR(bdev)) {
1179                         ret = PTR_ERR(bdev);
1180                         goto out;
1181                 }
1182
1183                 set_blocksize(bdev, 4096);
1184                 bh = btrfs_read_dev_super(bdev);
1185                 if (!bh) {
1186                         ret = -EIO;
1187                         goto error_close;
1188                 }
1189                 disk_super = (struct btrfs_super_block *)bh->b_data;
1190                 devid = le64_to_cpu(disk_super->dev_item.devid);
1191                 dev_uuid = disk_super->dev_item.uuid;
1192                 device = btrfs_find_device(root, devid, dev_uuid,
1193                                            disk_super->fsid);
1194                 if (!device) {
1195                         ret = -ENOENT;
1196                         goto error_brelse;
1197                 }
1198         }
1199
1200         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1201                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1202                        "device\n");
1203                 ret = -EINVAL;
1204                 goto error_brelse;
1205         }
1206
1207         if (device->writeable) {
1208                 list_del_init(&device->dev_alloc_list);
1209                 root->fs_info->fs_devices->rw_devices--;
1210         }
1211
1212         ret = btrfs_shrink_device(device, 0);
1213         if (ret)
1214                 goto error_brelse;
1215
1216         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1217         if (ret)
1218                 goto error_brelse;
1219
1220         device->in_fs_metadata = 0;
1221
1222         /*
1223          * the device list mutex makes sure that we don't change
1224          * the device list while someone else is writing out all
1225          * the device supers.
1226          */
1227         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1228         list_del_init(&device->dev_list);
1229         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1230
1231         device->fs_devices->num_devices--;
1232
1233         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1234                                  struct btrfs_device, dev_list);
1235         if (device->bdev == root->fs_info->sb->s_bdev)
1236                 root->fs_info->sb->s_bdev = next_device->bdev;
1237         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1238                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1239
1240         if (device->bdev) {
1241                 close_bdev_exclusive(device->bdev, device->mode);
1242                 device->bdev = NULL;
1243                 device->fs_devices->open_devices--;
1244         }
1245
1246         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1247         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1248
1249         if (device->fs_devices->open_devices == 0) {
1250                 struct btrfs_fs_devices *fs_devices;
1251                 fs_devices = root->fs_info->fs_devices;
1252                 while (fs_devices) {
1253                         if (fs_devices->seed == device->fs_devices)
1254                                 break;
1255                         fs_devices = fs_devices->seed;
1256                 }
1257                 fs_devices->seed = device->fs_devices->seed;
1258                 device->fs_devices->seed = NULL;
1259                 __btrfs_close_devices(device->fs_devices);
1260                 free_fs_devices(device->fs_devices);
1261         }
1262
1263         /*
1264          * at this point, the device is zero sized.  We want to
1265          * remove it from the devices list and zero out the old super
1266          */
1267         if (device->writeable) {
1268                 /* make sure this device isn't detected as part of
1269                  * the FS anymore
1270                  */
1271                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1272                 set_buffer_dirty(bh);
1273                 sync_dirty_buffer(bh);
1274         }
1275
1276         kfree(device->name);
1277         kfree(device);
1278         ret = 0;
1279
1280 error_brelse:
1281         brelse(bh);
1282 error_close:
1283         if (bdev)
1284                 close_bdev_exclusive(bdev, FMODE_READ);
1285 out:
1286         mutex_unlock(&root->fs_info->volume_mutex);
1287         mutex_unlock(&uuid_mutex);
1288         return ret;
1289 }
1290
1291 /*
1292  * does all the dirty work required for changing file system's UUID.
1293  */
1294 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1295                                 struct btrfs_root *root)
1296 {
1297         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1298         struct btrfs_fs_devices *old_devices;
1299         struct btrfs_fs_devices *seed_devices;
1300         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1301         struct btrfs_device *device;
1302         u64 super_flags;
1303
1304         BUG_ON(!mutex_is_locked(&uuid_mutex));
1305         if (!fs_devices->seeding)
1306                 return -EINVAL;
1307
1308         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1309         if (!seed_devices)
1310                 return -ENOMEM;
1311
1312         old_devices = clone_fs_devices(fs_devices);
1313         if (IS_ERR(old_devices)) {
1314                 kfree(seed_devices);
1315                 return PTR_ERR(old_devices);
1316         }
1317
1318         list_add(&old_devices->list, &fs_uuids);
1319
1320         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1321         seed_devices->opened = 1;
1322         INIT_LIST_HEAD(&seed_devices->devices);
1323         INIT_LIST_HEAD(&seed_devices->alloc_list);
1324         mutex_init(&seed_devices->device_list_mutex);
1325         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1326         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1327         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1328                 device->fs_devices = seed_devices;
1329         }
1330
1331         fs_devices->seeding = 0;
1332         fs_devices->num_devices = 0;
1333         fs_devices->open_devices = 0;
1334         fs_devices->seed = seed_devices;
1335
1336         generate_random_uuid(fs_devices->fsid);
1337         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1338         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1339         super_flags = btrfs_super_flags(disk_super) &
1340                       ~BTRFS_SUPER_FLAG_SEEDING;
1341         btrfs_set_super_flags(disk_super, super_flags);
1342
1343         return 0;
1344 }
1345
1346 /*
1347  * strore the expected generation for seed devices in device items.
1348  */
1349 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1350                                struct btrfs_root *root)
1351 {
1352         struct btrfs_path *path;
1353         struct extent_buffer *leaf;
1354         struct btrfs_dev_item *dev_item;
1355         struct btrfs_device *device;
1356         struct btrfs_key key;
1357         u8 fs_uuid[BTRFS_UUID_SIZE];
1358         u8 dev_uuid[BTRFS_UUID_SIZE];
1359         u64 devid;
1360         int ret;
1361
1362         path = btrfs_alloc_path();
1363         if (!path)
1364                 return -ENOMEM;
1365
1366         root = root->fs_info->chunk_root;
1367         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1368         key.offset = 0;
1369         key.type = BTRFS_DEV_ITEM_KEY;
1370
1371         while (1) {
1372                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1373                 if (ret < 0)
1374                         goto error;
1375
1376                 leaf = path->nodes[0];
1377 next_slot:
1378                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1379                         ret = btrfs_next_leaf(root, path);
1380                         if (ret > 0)
1381                                 break;
1382                         if (ret < 0)
1383                                 goto error;
1384                         leaf = path->nodes[0];
1385                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1386                         btrfs_release_path(root, path);
1387                         continue;
1388                 }
1389
1390                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1391                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1392                     key.type != BTRFS_DEV_ITEM_KEY)
1393                         break;
1394
1395                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1396                                           struct btrfs_dev_item);
1397                 devid = btrfs_device_id(leaf, dev_item);
1398                 read_extent_buffer(leaf, dev_uuid,
1399                                    (unsigned long)btrfs_device_uuid(dev_item),
1400                                    BTRFS_UUID_SIZE);
1401                 read_extent_buffer(leaf, fs_uuid,
1402                                    (unsigned long)btrfs_device_fsid(dev_item),
1403                                    BTRFS_UUID_SIZE);
1404                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1405                 BUG_ON(!device);
1406
1407                 if (device->fs_devices->seeding) {
1408                         btrfs_set_device_generation(leaf, dev_item,
1409                                                     device->generation);
1410                         btrfs_mark_buffer_dirty(leaf);
1411                 }
1412
1413                 path->slots[0]++;
1414                 goto next_slot;
1415         }
1416         ret = 0;
1417 error:
1418         btrfs_free_path(path);
1419         return ret;
1420 }
1421
1422 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1423 {
1424         struct btrfs_trans_handle *trans;
1425         struct btrfs_device *device;
1426         struct block_device *bdev;
1427         struct list_head *devices;
1428         struct super_block *sb = root->fs_info->sb;
1429         u64 total_bytes;
1430         int seeding_dev = 0;
1431         int ret = 0;
1432
1433         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1434                 return -EINVAL;
1435
1436         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1437         if (!bdev)
1438                 return -EIO;
1439
1440         if (root->fs_info->fs_devices->seeding) {
1441                 seeding_dev = 1;
1442                 down_write(&sb->s_umount);
1443                 mutex_lock(&uuid_mutex);
1444         }
1445
1446         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1447         mutex_lock(&root->fs_info->volume_mutex);
1448
1449         devices = &root->fs_info->fs_devices->devices;
1450         /*
1451          * we have the volume lock, so we don't need the extra
1452          * device list mutex while reading the list here.
1453          */
1454         list_for_each_entry(device, devices, dev_list) {
1455                 if (device->bdev == bdev) {
1456                         ret = -EEXIST;
1457                         goto error;
1458                 }
1459         }
1460
1461         device = kzalloc(sizeof(*device), GFP_NOFS);
1462         if (!device) {
1463                 /* we can safely leave the fs_devices entry around */
1464                 ret = -ENOMEM;
1465                 goto error;
1466         }
1467
1468         device->name = kstrdup(device_path, GFP_NOFS);
1469         if (!device->name) {
1470                 kfree(device);
1471                 ret = -ENOMEM;
1472                 goto error;
1473         }
1474
1475         ret = find_next_devid(root, &device->devid);
1476         if (ret) {
1477                 kfree(device);
1478                 goto error;
1479         }
1480
1481         trans = btrfs_start_transaction(root, 1);
1482         lock_chunks(root);
1483
1484         device->barriers = 1;
1485         device->writeable = 1;
1486         device->work.func = pending_bios_fn;
1487         generate_random_uuid(device->uuid);
1488         spin_lock_init(&device->io_lock);
1489         device->generation = trans->transid;
1490         device->io_width = root->sectorsize;
1491         device->io_align = root->sectorsize;
1492         device->sector_size = root->sectorsize;
1493         device->total_bytes = i_size_read(bdev->bd_inode);
1494         device->disk_total_bytes = device->total_bytes;
1495         device->dev_root = root->fs_info->dev_root;
1496         device->bdev = bdev;
1497         device->in_fs_metadata = 1;
1498         device->mode = 0;
1499         set_blocksize(device->bdev, 4096);
1500
1501         if (seeding_dev) {
1502                 sb->s_flags &= ~MS_RDONLY;
1503                 ret = btrfs_prepare_sprout(trans, root);
1504                 BUG_ON(ret);
1505         }
1506
1507         device->fs_devices = root->fs_info->fs_devices;
1508
1509         /*
1510          * we don't want write_supers to jump in here with our device
1511          * half setup
1512          */
1513         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1514         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1515         list_add(&device->dev_alloc_list,
1516                  &root->fs_info->fs_devices->alloc_list);
1517         root->fs_info->fs_devices->num_devices++;
1518         root->fs_info->fs_devices->open_devices++;
1519         root->fs_info->fs_devices->rw_devices++;
1520         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1521
1522         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1523                 root->fs_info->fs_devices->rotating = 1;
1524
1525         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1526         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1527                                     total_bytes + device->total_bytes);
1528
1529         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1530         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1531                                     total_bytes + 1);
1532         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1533
1534         if (seeding_dev) {
1535                 ret = init_first_rw_device(trans, root, device);
1536                 BUG_ON(ret);
1537                 ret = btrfs_finish_sprout(trans, root);
1538                 BUG_ON(ret);
1539         } else {
1540                 ret = btrfs_add_device(trans, root, device);
1541         }
1542
1543         /*
1544          * we've got more storage, clear any full flags on the space
1545          * infos
1546          */
1547         btrfs_clear_space_info_full(root->fs_info);
1548
1549         unlock_chunks(root);
1550         btrfs_commit_transaction(trans, root);
1551
1552         if (seeding_dev) {
1553                 mutex_unlock(&uuid_mutex);
1554                 up_write(&sb->s_umount);
1555
1556                 ret = btrfs_relocate_sys_chunks(root);
1557                 BUG_ON(ret);
1558         }
1559 out:
1560         mutex_unlock(&root->fs_info->volume_mutex);
1561         return ret;
1562 error:
1563         close_bdev_exclusive(bdev, 0);
1564         if (seeding_dev) {
1565                 mutex_unlock(&uuid_mutex);
1566                 up_write(&sb->s_umount);
1567         }
1568         goto out;
1569 }
1570
1571 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1572                                         struct btrfs_device *device)
1573 {
1574         int ret;
1575         struct btrfs_path *path;
1576         struct btrfs_root *root;
1577         struct btrfs_dev_item *dev_item;
1578         struct extent_buffer *leaf;
1579         struct btrfs_key key;
1580
1581         root = device->dev_root->fs_info->chunk_root;
1582
1583         path = btrfs_alloc_path();
1584         if (!path)
1585                 return -ENOMEM;
1586
1587         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1588         key.type = BTRFS_DEV_ITEM_KEY;
1589         key.offset = device->devid;
1590
1591         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1592         if (ret < 0)
1593                 goto out;
1594
1595         if (ret > 0) {
1596                 ret = -ENOENT;
1597                 goto out;
1598         }
1599
1600         leaf = path->nodes[0];
1601         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1602
1603         btrfs_set_device_id(leaf, dev_item, device->devid);
1604         btrfs_set_device_type(leaf, dev_item, device->type);
1605         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1606         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1607         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1608         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1609         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1610         btrfs_mark_buffer_dirty(leaf);
1611
1612 out:
1613         btrfs_free_path(path);
1614         return ret;
1615 }
1616
1617 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1618                       struct btrfs_device *device, u64 new_size)
1619 {
1620         struct btrfs_super_block *super_copy =
1621                 &device->dev_root->fs_info->super_copy;
1622         u64 old_total = btrfs_super_total_bytes(super_copy);
1623         u64 diff = new_size - device->total_bytes;
1624
1625         if (!device->writeable)
1626                 return -EACCES;
1627         if (new_size <= device->total_bytes)
1628                 return -EINVAL;
1629
1630         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1631         device->fs_devices->total_rw_bytes += diff;
1632
1633         device->total_bytes = new_size;
1634         device->disk_total_bytes = new_size;
1635         btrfs_clear_space_info_full(device->dev_root->fs_info);
1636
1637         return btrfs_update_device(trans, device);
1638 }
1639
1640 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1641                       struct btrfs_device *device, u64 new_size)
1642 {
1643         int ret;
1644         lock_chunks(device->dev_root);
1645         ret = __btrfs_grow_device(trans, device, new_size);
1646         unlock_chunks(device->dev_root);
1647         return ret;
1648 }
1649
1650 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1651                             struct btrfs_root *root,
1652                             u64 chunk_tree, u64 chunk_objectid,
1653                             u64 chunk_offset)
1654 {
1655         int ret;
1656         struct btrfs_path *path;
1657         struct btrfs_key key;
1658
1659         root = root->fs_info->chunk_root;
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663
1664         key.objectid = chunk_objectid;
1665         key.offset = chunk_offset;
1666         key.type = BTRFS_CHUNK_ITEM_KEY;
1667
1668         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1669         BUG_ON(ret);
1670
1671         ret = btrfs_del_item(trans, root, path);
1672         BUG_ON(ret);
1673
1674         btrfs_free_path(path);
1675         return 0;
1676 }
1677
1678 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1679                         chunk_offset)
1680 {
1681         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1682         struct btrfs_disk_key *disk_key;
1683         struct btrfs_chunk *chunk;
1684         u8 *ptr;
1685         int ret = 0;
1686         u32 num_stripes;
1687         u32 array_size;
1688         u32 len = 0;
1689         u32 cur;
1690         struct btrfs_key key;
1691
1692         array_size = btrfs_super_sys_array_size(super_copy);
1693
1694         ptr = super_copy->sys_chunk_array;
1695         cur = 0;
1696
1697         while (cur < array_size) {
1698                 disk_key = (struct btrfs_disk_key *)ptr;
1699                 btrfs_disk_key_to_cpu(&key, disk_key);
1700
1701                 len = sizeof(*disk_key);
1702
1703                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1704                         chunk = (struct btrfs_chunk *)(ptr + len);
1705                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1706                         len += btrfs_chunk_item_size(num_stripes);
1707                 } else {
1708                         ret = -EIO;
1709                         break;
1710                 }
1711                 if (key.objectid == chunk_objectid &&
1712                     key.offset == chunk_offset) {
1713                         memmove(ptr, ptr + len, array_size - (cur + len));
1714                         array_size -= len;
1715                         btrfs_set_super_sys_array_size(super_copy, array_size);
1716                 } else {
1717                         ptr += len;
1718                         cur += len;
1719                 }
1720         }
1721         return ret;
1722 }
1723
1724 static int btrfs_relocate_chunk(struct btrfs_root *root,
1725                          u64 chunk_tree, u64 chunk_objectid,
1726                          u64 chunk_offset)
1727 {
1728         struct extent_map_tree *em_tree;
1729         struct btrfs_root *extent_root;
1730         struct btrfs_trans_handle *trans;
1731         struct extent_map *em;
1732         struct map_lookup *map;
1733         int ret;
1734         int i;
1735
1736         root = root->fs_info->chunk_root;
1737         extent_root = root->fs_info->extent_root;
1738         em_tree = &root->fs_info->mapping_tree.map_tree;
1739
1740         ret = btrfs_can_relocate(extent_root, chunk_offset);
1741         if (ret)
1742                 return -ENOSPC;
1743
1744         /* step one, relocate all the extents inside this chunk */
1745         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1746         BUG_ON(ret);
1747
1748         trans = btrfs_start_transaction(root, 1);
1749         BUG_ON(!trans);
1750
1751         lock_chunks(root);
1752
1753         /*
1754          * step two, delete the device extents and the
1755          * chunk tree entries
1756          */
1757         read_lock(&em_tree->lock);
1758         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1759         read_unlock(&em_tree->lock);
1760
1761         BUG_ON(em->start > chunk_offset ||
1762                em->start + em->len < chunk_offset);
1763         map = (struct map_lookup *)em->bdev;
1764
1765         for (i = 0; i < map->num_stripes; i++) {
1766                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1767                                             map->stripes[i].physical);
1768                 BUG_ON(ret);
1769
1770                 if (map->stripes[i].dev) {
1771                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1772                         BUG_ON(ret);
1773                 }
1774         }
1775         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1776                                chunk_offset);
1777
1778         BUG_ON(ret);
1779
1780         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1781                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1782                 BUG_ON(ret);
1783         }
1784
1785         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1786         BUG_ON(ret);
1787
1788         write_lock(&em_tree->lock);
1789         remove_extent_mapping(em_tree, em);
1790         write_unlock(&em_tree->lock);
1791
1792         kfree(map);
1793         em->bdev = NULL;
1794
1795         /* once for the tree */
1796         free_extent_map(em);
1797         /* once for us */
1798         free_extent_map(em);
1799
1800         unlock_chunks(root);
1801         btrfs_end_transaction(trans, root);
1802         return 0;
1803 }
1804
1805 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1806 {
1807         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1808         struct btrfs_path *path;
1809         struct extent_buffer *leaf;
1810         struct btrfs_chunk *chunk;
1811         struct btrfs_key key;
1812         struct btrfs_key found_key;
1813         u64 chunk_tree = chunk_root->root_key.objectid;
1814         u64 chunk_type;
1815         bool retried = false;
1816         int failed = 0;
1817         int ret;
1818
1819         path = btrfs_alloc_path();
1820         if (!path)
1821                 return -ENOMEM;
1822
1823 again:
1824         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1825         key.offset = (u64)-1;
1826         key.type = BTRFS_CHUNK_ITEM_KEY;
1827
1828         while (1) {
1829                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1830                 if (ret < 0)
1831                         goto error;
1832                 BUG_ON(ret == 0);
1833
1834                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1835                                           key.type);
1836                 if (ret < 0)
1837                         goto error;
1838                 if (ret > 0)
1839                         break;
1840
1841                 leaf = path->nodes[0];
1842                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1843
1844                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1845                                        struct btrfs_chunk);
1846                 chunk_type = btrfs_chunk_type(leaf, chunk);
1847                 btrfs_release_path(chunk_root, path);
1848
1849                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1850                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1851                                                    found_key.objectid,
1852                                                    found_key.offset);
1853                         if (ret == -ENOSPC)
1854                                 failed++;
1855                         else if (ret)
1856                                 BUG();
1857                 }
1858
1859                 if (found_key.offset == 0)
1860                         break;
1861                 key.offset = found_key.offset - 1;
1862         }
1863         ret = 0;
1864         if (failed && !retried) {
1865                 failed = 0;
1866                 retried = true;
1867                 goto again;
1868         } else if (failed && retried) {
1869                 WARN_ON(1);
1870                 ret = -ENOSPC;
1871         }
1872 error:
1873         btrfs_free_path(path);
1874         return ret;
1875 }
1876
1877 static u64 div_factor(u64 num, int factor)
1878 {
1879         if (factor == 10)
1880                 return num;
1881         num *= factor;
1882         do_div(num, 10);
1883         return num;
1884 }
1885
1886 int btrfs_balance(struct btrfs_root *dev_root)
1887 {
1888         int ret;
1889         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1890         struct btrfs_device *device;
1891         u64 old_size;
1892         u64 size_to_free;
1893         struct btrfs_path *path;
1894         struct btrfs_key key;
1895         struct btrfs_chunk *chunk;
1896         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1897         struct btrfs_trans_handle *trans;
1898         struct btrfs_key found_key;
1899
1900         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1901                 return -EROFS;
1902
1903         mutex_lock(&dev_root->fs_info->volume_mutex);
1904         dev_root = dev_root->fs_info->dev_root;
1905
1906         /* step one make some room on all the devices */
1907         list_for_each_entry(device, devices, dev_list) {
1908                 old_size = device->total_bytes;
1909                 size_to_free = div_factor(old_size, 1);
1910                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1911                 if (!device->writeable ||
1912                     device->total_bytes - device->bytes_used > size_to_free)
1913                         continue;
1914
1915                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1916                 if (ret == -ENOSPC)
1917                         break;
1918                 BUG_ON(ret);
1919
1920                 trans = btrfs_start_transaction(dev_root, 1);
1921                 BUG_ON(!trans);
1922
1923                 ret = btrfs_grow_device(trans, device, old_size);
1924                 BUG_ON(ret);
1925
1926                 btrfs_end_transaction(trans, dev_root);
1927         }
1928
1929         /* step two, relocate all the chunks */
1930         path = btrfs_alloc_path();
1931         BUG_ON(!path);
1932
1933         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1934         key.offset = (u64)-1;
1935         key.type = BTRFS_CHUNK_ITEM_KEY;
1936
1937         while (1) {
1938                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1939                 if (ret < 0)
1940                         goto error;
1941
1942                 /*
1943                  * this shouldn't happen, it means the last relocate
1944                  * failed
1945                  */
1946                 if (ret == 0)
1947                         break;
1948
1949                 ret = btrfs_previous_item(chunk_root, path, 0,
1950                                           BTRFS_CHUNK_ITEM_KEY);
1951                 if (ret)
1952                         break;
1953
1954                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1955                                       path->slots[0]);
1956                 if (found_key.objectid != key.objectid)
1957                         break;
1958
1959                 chunk = btrfs_item_ptr(path->nodes[0],
1960                                        path->slots[0],
1961                                        struct btrfs_chunk);
1962                 /* chunk zero is special */
1963                 if (found_key.offset == 0)
1964                         break;
1965
1966                 btrfs_release_path(chunk_root, path);
1967                 ret = btrfs_relocate_chunk(chunk_root,
1968                                            chunk_root->root_key.objectid,
1969                                            found_key.objectid,
1970                                            found_key.offset);
1971                 BUG_ON(ret && ret != -ENOSPC);
1972                 key.offset = found_key.offset - 1;
1973         }
1974         ret = 0;
1975 error:
1976         btrfs_free_path(path);
1977         mutex_unlock(&dev_root->fs_info->volume_mutex);
1978         return ret;
1979 }
1980
1981 /*
1982  * shrinking a device means finding all of the device extents past
1983  * the new size, and then following the back refs to the chunks.
1984  * The chunk relocation code actually frees the device extent
1985  */
1986 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1987 {
1988         struct btrfs_trans_handle *trans;
1989         struct btrfs_root *root = device->dev_root;
1990         struct btrfs_dev_extent *dev_extent = NULL;
1991         struct btrfs_path *path;
1992         u64 length;
1993         u64 chunk_tree;
1994         u64 chunk_objectid;
1995         u64 chunk_offset;
1996         int ret;
1997         int slot;
1998         int failed = 0;
1999         bool retried = false;
2000         struct extent_buffer *l;
2001         struct btrfs_key key;
2002         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2003         u64 old_total = btrfs_super_total_bytes(super_copy);
2004         u64 old_size = device->total_bytes;
2005         u64 diff = device->total_bytes - new_size;
2006
2007         if (new_size >= device->total_bytes)
2008                 return -EINVAL;
2009
2010         path = btrfs_alloc_path();
2011         if (!path)
2012                 return -ENOMEM;
2013
2014         path->reada = 2;
2015
2016         lock_chunks(root);
2017
2018         device->total_bytes = new_size;
2019         if (device->writeable)
2020                 device->fs_devices->total_rw_bytes -= diff;
2021         unlock_chunks(root);
2022
2023 again:
2024         key.objectid = device->devid;
2025         key.offset = (u64)-1;
2026         key.type = BTRFS_DEV_EXTENT_KEY;
2027
2028         while (1) {
2029                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2030                 if (ret < 0)
2031                         goto done;
2032
2033                 ret = btrfs_previous_item(root, path, 0, key.type);
2034                 if (ret < 0)
2035                         goto done;
2036                 if (ret) {
2037                         ret = 0;
2038                         btrfs_release_path(root, path);
2039                         break;
2040                 }
2041
2042                 l = path->nodes[0];
2043                 slot = path->slots[0];
2044                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2045
2046                 if (key.objectid != device->devid) {
2047                         btrfs_release_path(root, path);
2048                         break;
2049                 }
2050
2051                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2052                 length = btrfs_dev_extent_length(l, dev_extent);
2053
2054                 if (key.offset + length <= new_size) {
2055                         btrfs_release_path(root, path);
2056                         break;
2057                 }
2058
2059                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2060                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2061                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2062                 btrfs_release_path(root, path);
2063
2064                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2065                                            chunk_offset);
2066                 if (ret && ret != -ENOSPC)
2067                         goto done;
2068                 if (ret == -ENOSPC)
2069                         failed++;
2070                 key.offset -= 1;
2071         }
2072
2073         if (failed && !retried) {
2074                 failed = 0;
2075                 retried = true;
2076                 goto again;
2077         } else if (failed && retried) {
2078                 ret = -ENOSPC;
2079                 lock_chunks(root);
2080
2081                 device->total_bytes = old_size;
2082                 if (device->writeable)
2083                         device->fs_devices->total_rw_bytes += diff;
2084                 unlock_chunks(root);
2085                 goto done;
2086         }
2087
2088         /* Shrinking succeeded, else we would be at "done". */
2089         trans = btrfs_start_transaction(root, 1);
2090         if (!trans) {
2091                 ret = -ENOMEM;
2092                 goto done;
2093         }
2094         lock_chunks(root);
2095
2096         device->disk_total_bytes = new_size;
2097         /* Now btrfs_update_device() will change the on-disk size. */
2098         ret = btrfs_update_device(trans, device);
2099         if (ret) {
2100                 unlock_chunks(root);
2101                 btrfs_end_transaction(trans, root);
2102                 goto done;
2103         }
2104         WARN_ON(diff > old_total);
2105         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2106         unlock_chunks(root);
2107         btrfs_end_transaction(trans, root);
2108 done:
2109         btrfs_free_path(path);
2110         return ret;
2111 }
2112
2113 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2114                            struct btrfs_root *root,
2115                            struct btrfs_key *key,
2116                            struct btrfs_chunk *chunk, int item_size)
2117 {
2118         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2119         struct btrfs_disk_key disk_key;
2120         u32 array_size;
2121         u8 *ptr;
2122
2123         array_size = btrfs_super_sys_array_size(super_copy);
2124         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2125                 return -EFBIG;
2126
2127         ptr = super_copy->sys_chunk_array + array_size;
2128         btrfs_cpu_key_to_disk(&disk_key, key);
2129         memcpy(ptr, &disk_key, sizeof(disk_key));
2130         ptr += sizeof(disk_key);
2131         memcpy(ptr, chunk, item_size);
2132         item_size += sizeof(disk_key);
2133         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2134         return 0;
2135 }
2136
2137 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2138                                         int num_stripes, int sub_stripes)
2139 {
2140         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2141                 return calc_size;
2142         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2143                 return calc_size * (num_stripes / sub_stripes);
2144         else
2145                 return calc_size * num_stripes;
2146 }
2147
2148 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2149                                struct btrfs_root *extent_root,
2150                                struct map_lookup **map_ret,
2151                                u64 *num_bytes, u64 *stripe_size,
2152                                u64 start, u64 type)
2153 {
2154         struct btrfs_fs_info *info = extent_root->fs_info;
2155         struct btrfs_device *device = NULL;
2156         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2157         struct list_head *cur;
2158         struct map_lookup *map = NULL;
2159         struct extent_map_tree *em_tree;
2160         struct extent_map *em;
2161         struct list_head private_devs;
2162         int min_stripe_size = 1 * 1024 * 1024;
2163         u64 calc_size = 1024 * 1024 * 1024;
2164         u64 max_chunk_size = calc_size;
2165         u64 min_free;
2166         u64 avail;
2167         u64 max_avail = 0;
2168         u64 dev_offset;
2169         int num_stripes = 1;
2170         int min_stripes = 1;
2171         int sub_stripes = 0;
2172         int looped = 0;
2173         int ret;
2174         int index;
2175         int stripe_len = 64 * 1024;
2176
2177         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2178             (type & BTRFS_BLOCK_GROUP_DUP)) {
2179                 WARN_ON(1);
2180                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2181         }
2182         if (list_empty(&fs_devices->alloc_list))
2183                 return -ENOSPC;
2184
2185         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2186                 num_stripes = fs_devices->rw_devices;
2187                 min_stripes = 2;
2188         }
2189         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2190                 num_stripes = 2;
2191                 min_stripes = 2;
2192         }
2193         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2194                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2195                 if (num_stripes < 2)
2196                         return -ENOSPC;
2197                 min_stripes = 2;
2198         }
2199         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2200                 num_stripes = fs_devices->rw_devices;
2201                 if (num_stripes < 4)
2202                         return -ENOSPC;
2203                 num_stripes &= ~(u32)1;
2204                 sub_stripes = 2;
2205                 min_stripes = 4;
2206         }
2207
2208         if (type & BTRFS_BLOCK_GROUP_DATA) {
2209                 max_chunk_size = 10 * calc_size;
2210                 min_stripe_size = 64 * 1024 * 1024;
2211         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2212                 max_chunk_size = 256 * 1024 * 1024;
2213                 min_stripe_size = 32 * 1024 * 1024;
2214         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2215                 calc_size = 8 * 1024 * 1024;
2216                 max_chunk_size = calc_size * 2;
2217                 min_stripe_size = 1 * 1024 * 1024;
2218         }
2219
2220         /* we don't want a chunk larger than 10% of writeable space */
2221         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2222                              max_chunk_size);
2223
2224 again:
2225         max_avail = 0;
2226         if (!map || map->num_stripes != num_stripes) {
2227                 kfree(map);
2228                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2229                 if (!map)
2230                         return -ENOMEM;
2231                 map->num_stripes = num_stripes;
2232         }
2233
2234         if (calc_size * num_stripes > max_chunk_size) {
2235                 calc_size = max_chunk_size;
2236                 do_div(calc_size, num_stripes);
2237                 do_div(calc_size, stripe_len);
2238                 calc_size *= stripe_len;
2239         }
2240         /* we don't want tiny stripes */
2241         calc_size = max_t(u64, min_stripe_size, calc_size);
2242
2243         do_div(calc_size, stripe_len);
2244         calc_size *= stripe_len;
2245
2246         cur = fs_devices->alloc_list.next;
2247         index = 0;
2248
2249         if (type & BTRFS_BLOCK_GROUP_DUP)
2250                 min_free = calc_size * 2;
2251         else
2252                 min_free = calc_size;
2253
2254         /*
2255          * we add 1MB because we never use the first 1MB of the device, unless
2256          * we've looped, then we are likely allocating the maximum amount of
2257          * space left already
2258          */
2259         if (!looped)
2260                 min_free += 1024 * 1024;
2261
2262         INIT_LIST_HEAD(&private_devs);
2263         while (index < num_stripes) {
2264                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2265                 BUG_ON(!device->writeable);
2266                 if (device->total_bytes > device->bytes_used)
2267                         avail = device->total_bytes - device->bytes_used;
2268                 else
2269                         avail = 0;
2270                 cur = cur->next;
2271
2272                 if (device->in_fs_metadata && avail >= min_free) {
2273                         ret = find_free_dev_extent(trans, device,
2274                                                    min_free, &dev_offset,
2275                                                    &max_avail);
2276                         if (ret == 0) {
2277                                 list_move_tail(&device->dev_alloc_list,
2278                                                &private_devs);
2279                                 map->stripes[index].dev = device;
2280                                 map->stripes[index].physical = dev_offset;
2281                                 index++;
2282                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2283                                         map->stripes[index].dev = device;
2284                                         map->stripes[index].physical =
2285                                                 dev_offset + calc_size;
2286                                         index++;
2287                                 }
2288                         }
2289                 } else if (device->in_fs_metadata && avail > max_avail)
2290                         max_avail = avail;
2291                 if (cur == &fs_devices->alloc_list)
2292                         break;
2293         }
2294         list_splice(&private_devs, &fs_devices->alloc_list);
2295         if (index < num_stripes) {
2296                 if (index >= min_stripes) {
2297                         num_stripes = index;
2298                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2299                                 num_stripes /= sub_stripes;
2300                                 num_stripes *= sub_stripes;
2301                         }
2302                         looped = 1;
2303                         goto again;
2304                 }
2305                 if (!looped && max_avail > 0) {
2306                         looped = 1;
2307                         calc_size = max_avail;
2308                         goto again;
2309                 }
2310                 kfree(map);
2311                 return -ENOSPC;
2312         }
2313         map->sector_size = extent_root->sectorsize;
2314         map->stripe_len = stripe_len;
2315         map->io_align = stripe_len;
2316         map->io_width = stripe_len;
2317         map->type = type;
2318         map->num_stripes = num_stripes;
2319         map->sub_stripes = sub_stripes;
2320
2321         *map_ret = map;
2322         *stripe_size = calc_size;
2323         *num_bytes = chunk_bytes_by_type(type, calc_size,
2324                                          num_stripes, sub_stripes);
2325
2326         em = alloc_extent_map(GFP_NOFS);
2327         if (!em) {
2328                 kfree(map);
2329                 return -ENOMEM;
2330         }
2331         em->bdev = (struct block_device *)map;
2332         em->start = start;
2333         em->len = *num_bytes;
2334         em->block_start = 0;
2335         em->block_len = em->len;
2336
2337         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2338         write_lock(&em_tree->lock);
2339         ret = add_extent_mapping(em_tree, em);
2340         write_unlock(&em_tree->lock);
2341         BUG_ON(ret);
2342         free_extent_map(em);
2343
2344         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2345                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2346                                      start, *num_bytes);
2347         BUG_ON(ret);
2348
2349         index = 0;
2350         while (index < map->num_stripes) {
2351                 device = map->stripes[index].dev;
2352                 dev_offset = map->stripes[index].physical;
2353
2354                 ret = btrfs_alloc_dev_extent(trans, device,
2355                                 info->chunk_root->root_key.objectid,
2356                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2357                                 start, dev_offset, calc_size);
2358                 BUG_ON(ret);
2359                 index++;
2360         }
2361
2362         return 0;
2363 }
2364
2365 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2366                                 struct btrfs_root *extent_root,
2367                                 struct map_lookup *map, u64 chunk_offset,
2368                                 u64 chunk_size, u64 stripe_size)
2369 {
2370         u64 dev_offset;
2371         struct btrfs_key key;
2372         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2373         struct btrfs_device *device;
2374         struct btrfs_chunk *chunk;
2375         struct btrfs_stripe *stripe;
2376         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2377         int index = 0;
2378         int ret;
2379
2380         chunk = kzalloc(item_size, GFP_NOFS);
2381         if (!chunk)
2382                 return -ENOMEM;
2383
2384         index = 0;
2385         while (index < map->num_stripes) {
2386                 device = map->stripes[index].dev;
2387                 device->bytes_used += stripe_size;
2388                 ret = btrfs_update_device(trans, device);
2389                 BUG_ON(ret);
2390                 index++;
2391         }
2392
2393         index = 0;
2394         stripe = &chunk->stripe;
2395         while (index < map->num_stripes) {
2396                 device = map->stripes[index].dev;
2397                 dev_offset = map->stripes[index].physical;
2398
2399                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2400                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2401                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2402                 stripe++;
2403                 index++;
2404         }
2405
2406         btrfs_set_stack_chunk_length(chunk, chunk_size);
2407         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2408         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2409         btrfs_set_stack_chunk_type(chunk, map->type);
2410         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2411         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2412         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2413         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2414         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2415
2416         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2417         key.type = BTRFS_CHUNK_ITEM_KEY;
2418         key.offset = chunk_offset;
2419
2420         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2421         BUG_ON(ret);
2422
2423         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2424                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2425                                              item_size);
2426                 BUG_ON(ret);
2427         }
2428         kfree(chunk);
2429         return 0;
2430 }
2431
2432 /*
2433  * Chunk allocation falls into two parts. The first part does works
2434  * that make the new allocated chunk useable, but not do any operation
2435  * that modifies the chunk tree. The second part does the works that
2436  * require modifying the chunk tree. This division is important for the
2437  * bootstrap process of adding storage to a seed btrfs.
2438  */
2439 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2440                       struct btrfs_root *extent_root, u64 type)
2441 {
2442         u64 chunk_offset;
2443         u64 chunk_size;
2444         u64 stripe_size;
2445         struct map_lookup *map;
2446         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2447         int ret;
2448
2449         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2450                               &chunk_offset);
2451         if (ret)
2452                 return ret;
2453
2454         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2455                                   &stripe_size, chunk_offset, type);
2456         if (ret)
2457                 return ret;
2458
2459         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2460                                    chunk_size, stripe_size);
2461         BUG_ON(ret);
2462         return 0;
2463 }
2464
2465 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2466                                          struct btrfs_root *root,
2467                                          struct btrfs_device *device)
2468 {
2469         u64 chunk_offset;
2470         u64 sys_chunk_offset;
2471         u64 chunk_size;
2472         u64 sys_chunk_size;
2473         u64 stripe_size;
2474         u64 sys_stripe_size;
2475         u64 alloc_profile;
2476         struct map_lookup *map;
2477         struct map_lookup *sys_map;
2478         struct btrfs_fs_info *fs_info = root->fs_info;
2479         struct btrfs_root *extent_root = fs_info->extent_root;
2480         int ret;
2481
2482         ret = find_next_chunk(fs_info->chunk_root,
2483                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2484         BUG_ON(ret);
2485
2486         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2487                         (fs_info->metadata_alloc_profile &
2488                          fs_info->avail_metadata_alloc_bits);
2489         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2490
2491         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2492                                   &stripe_size, chunk_offset, alloc_profile);
2493         BUG_ON(ret);
2494
2495         sys_chunk_offset = chunk_offset + chunk_size;
2496
2497         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2498                         (fs_info->system_alloc_profile &
2499                          fs_info->avail_system_alloc_bits);
2500         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2501
2502         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2503                                   &sys_chunk_size, &sys_stripe_size,
2504                                   sys_chunk_offset, alloc_profile);
2505         BUG_ON(ret);
2506
2507         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2508         BUG_ON(ret);
2509
2510         /*
2511          * Modifying chunk tree needs allocating new blocks from both
2512          * system block group and metadata block group. So we only can
2513          * do operations require modifying the chunk tree after both
2514          * block groups were created.
2515          */
2516         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2517                                    chunk_size, stripe_size);
2518         BUG_ON(ret);
2519
2520         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2521                                    sys_chunk_offset, sys_chunk_size,
2522                                    sys_stripe_size);
2523         BUG_ON(ret);
2524         return 0;
2525 }
2526
2527 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2528 {
2529         struct extent_map *em;
2530         struct map_lookup *map;
2531         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2532         int readonly = 0;
2533         int i;
2534
2535         read_lock(&map_tree->map_tree.lock);
2536         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2537         read_unlock(&map_tree->map_tree.lock);
2538         if (!em)
2539                 return 1;
2540
2541         if (btrfs_test_opt(root, DEGRADED)) {
2542                 free_extent_map(em);
2543                 return 0;
2544         }
2545
2546         map = (struct map_lookup *)em->bdev;
2547         for (i = 0; i < map->num_stripes; i++) {
2548                 if (!map->stripes[i].dev->writeable) {
2549                         readonly = 1;
2550                         break;
2551                 }
2552         }
2553         free_extent_map(em);
2554         return readonly;
2555 }
2556
2557 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2558 {
2559         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2560 }
2561
2562 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2563 {
2564         struct extent_map *em;
2565
2566         while (1) {
2567                 write_lock(&tree->map_tree.lock);
2568                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2569                 if (em)
2570                         remove_extent_mapping(&tree->map_tree, em);
2571                 write_unlock(&tree->map_tree.lock);
2572                 if (!em)
2573                         break;
2574                 kfree(em->bdev);
2575                 /* once for us */
2576                 free_extent_map(em);
2577                 /* once for the tree */
2578                 free_extent_map(em);
2579         }
2580 }
2581
2582 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2583 {
2584         struct extent_map *em;
2585         struct map_lookup *map;
2586         struct extent_map_tree *em_tree = &map_tree->map_tree;
2587         int ret;
2588
2589         read_lock(&em_tree->lock);
2590         em = lookup_extent_mapping(em_tree, logical, len);
2591         read_unlock(&em_tree->lock);
2592         BUG_ON(!em);
2593
2594         BUG_ON(em->start > logical || em->start + em->len < logical);
2595         map = (struct map_lookup *)em->bdev;
2596         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2597                 ret = map->num_stripes;
2598         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2599                 ret = map->sub_stripes;
2600         else
2601                 ret = 1;
2602         free_extent_map(em);
2603         return ret;
2604 }
2605
2606 static int find_live_mirror(struct map_lookup *map, int first, int num,
2607                             int optimal)
2608 {
2609         int i;
2610         if (map->stripes[optimal].dev->bdev)
2611                 return optimal;
2612         for (i = first; i < first + num; i++) {
2613                 if (map->stripes[i].dev->bdev)
2614                         return i;
2615         }
2616         /* we couldn't find one that doesn't fail.  Just return something
2617          * and the io error handling code will clean up eventually
2618          */
2619         return optimal;
2620 }
2621
2622 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2623                              u64 logical, u64 *length,
2624                              struct btrfs_multi_bio **multi_ret,
2625                              int mirror_num, struct page *unplug_page)
2626 {
2627         struct extent_map *em;
2628         struct map_lookup *map;
2629         struct extent_map_tree *em_tree = &map_tree->map_tree;
2630         u64 offset;
2631         u64 stripe_offset;
2632         u64 stripe_nr;
2633         int stripes_allocated = 8;
2634         int stripes_required = 1;
2635         int stripe_index;
2636         int i;
2637         int num_stripes;
2638         int max_errors = 0;
2639         struct btrfs_multi_bio *multi = NULL;
2640
2641         if (multi_ret && !(rw & (1 << BIO_RW)))
2642                 stripes_allocated = 1;
2643 again:
2644         if (multi_ret) {
2645                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2646                                 GFP_NOFS);
2647                 if (!multi)
2648                         return -ENOMEM;
2649
2650                 atomic_set(&multi->error, 0);
2651         }
2652
2653         read_lock(&em_tree->lock);
2654         em = lookup_extent_mapping(em_tree, logical, *length);
2655         read_unlock(&em_tree->lock);
2656
2657         if (!em && unplug_page) {
2658                 kfree(multi);
2659                 return 0;
2660         }
2661
2662         if (!em) {
2663                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2664                        (unsigned long long)logical,
2665                        (unsigned long long)*length);
2666                 BUG();
2667         }
2668
2669         BUG_ON(em->start > logical || em->start + em->len < logical);
2670         map = (struct map_lookup *)em->bdev;
2671         offset = logical - em->start;
2672
2673         if (mirror_num > map->num_stripes)
2674                 mirror_num = 0;
2675
2676         /* if our multi bio struct is too small, back off and try again */
2677         if (rw & (1 << BIO_RW)) {
2678                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2679                                  BTRFS_BLOCK_GROUP_DUP)) {
2680                         stripes_required = map->num_stripes;
2681                         max_errors = 1;
2682                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2683                         stripes_required = map->sub_stripes;
2684                         max_errors = 1;
2685                 }
2686         }
2687         if (multi_ret && (rw & (1 << BIO_RW)) &&
2688             stripes_allocated < stripes_required) {
2689                 stripes_allocated = map->num_stripes;
2690                 free_extent_map(em);
2691                 kfree(multi);
2692                 goto again;
2693         }
2694         stripe_nr = offset;
2695         /*
2696          * stripe_nr counts the total number of stripes we have to stride
2697          * to get to this block
2698          */
2699         do_div(stripe_nr, map->stripe_len);
2700
2701         stripe_offset = stripe_nr * map->stripe_len;
2702         BUG_ON(offset < stripe_offset);
2703
2704         /* stripe_offset is the offset of this block in its stripe*/
2705         stripe_offset = offset - stripe_offset;
2706
2707         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2708                          BTRFS_BLOCK_GROUP_RAID10 |
2709                          BTRFS_BLOCK_GROUP_DUP)) {
2710                 /* we limit the length of each bio to what fits in a stripe */
2711                 *length = min_t(u64, em->len - offset,
2712                               map->stripe_len - stripe_offset);
2713         } else {
2714                 *length = em->len - offset;
2715         }
2716
2717         if (!multi_ret && !unplug_page)
2718                 goto out;
2719
2720         num_stripes = 1;
2721         stripe_index = 0;
2722         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2723                 if (unplug_page || (rw & (1 << BIO_RW)))
2724                         num_stripes = map->num_stripes;
2725                 else if (mirror_num)
2726                         stripe_index = mirror_num - 1;
2727                 else {
2728                         stripe_index = find_live_mirror(map, 0,
2729                                             map->num_stripes,
2730                                             current->pid % map->num_stripes);
2731                 }
2732
2733         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2734                 if (rw & (1 << BIO_RW))
2735                         num_stripes = map->num_stripes;
2736                 else if (mirror_num)
2737                         stripe_index = mirror_num - 1;
2738
2739         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2740                 int factor = map->num_stripes / map->sub_stripes;
2741
2742                 stripe_index = do_div(stripe_nr, factor);
2743                 stripe_index *= map->sub_stripes;
2744
2745                 if (unplug_page || (rw & (1 << BIO_RW)))
2746                         num_stripes = map->sub_stripes;
2747                 else if (mirror_num)
2748                         stripe_index += mirror_num - 1;
2749                 else {
2750                         stripe_index = find_live_mirror(map, stripe_index,
2751                                               map->sub_stripes, stripe_index +
2752                                               current->pid % map->sub_stripes);
2753                 }
2754         } else {
2755                 /*
2756                  * after this do_div call, stripe_nr is the number of stripes
2757                  * on this device we have to walk to find the data, and
2758                  * stripe_index is the number of our device in the stripe array
2759                  */
2760                 stripe_index = do_div(stripe_nr, map->num_stripes);
2761         }
2762         BUG_ON(stripe_index >= map->num_stripes);
2763
2764         for (i = 0; i < num_stripes; i++) {
2765                 if (unplug_page) {
2766                         struct btrfs_device *device;
2767                         struct backing_dev_info *bdi;
2768
2769                         device = map->stripes[stripe_index].dev;
2770                         if (device->bdev) {
2771                                 bdi = blk_get_backing_dev_info(device->bdev);
2772                                 if (bdi->unplug_io_fn)
2773                                         bdi->unplug_io_fn(bdi, unplug_page);
2774                         }
2775                 } else {
2776                         multi->stripes[i].physical =
2777                                 map->stripes[stripe_index].physical +
2778                                 stripe_offset + stripe_nr * map->stripe_len;
2779                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2780                 }
2781                 stripe_index++;
2782         }
2783         if (multi_ret) {
2784                 *multi_ret = multi;
2785                 multi->num_stripes = num_stripes;
2786                 multi->max_errors = max_errors;
2787         }
2788 out:
2789         free_extent_map(em);
2790         return 0;
2791 }
2792
2793 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2794                       u64 logical, u64 *length,
2795                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2796 {
2797         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2798                                  mirror_num, NULL);
2799 }
2800
2801 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2802                      u64 chunk_start, u64 physical, u64 devid,
2803                      u64 **logical, int *naddrs, int *stripe_len)
2804 {
2805         struct extent_map_tree *em_tree = &map_tree->map_tree;
2806         struct extent_map *em;
2807         struct map_lookup *map;
2808         u64 *buf;
2809         u64 bytenr;
2810         u64 length;
2811         u64 stripe_nr;
2812         int i, j, nr = 0;
2813
2814         read_lock(&em_tree->lock);
2815         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2816         read_unlock(&em_tree->lock);
2817
2818         BUG_ON(!em || em->start != chunk_start);
2819         map = (struct map_lookup *)em->bdev;
2820
2821         length = em->len;
2822         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2823                 do_div(length, map->num_stripes / map->sub_stripes);
2824         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2825                 do_div(length, map->num_stripes);
2826
2827         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2828         BUG_ON(!buf);
2829
2830         for (i = 0; i < map->num_stripes; i++) {
2831                 if (devid && map->stripes[i].dev->devid != devid)
2832                         continue;
2833                 if (map->stripes[i].physical > physical ||
2834                     map->stripes[i].physical + length <= physical)
2835                         continue;
2836
2837                 stripe_nr = physical - map->stripes[i].physical;
2838                 do_div(stripe_nr, map->stripe_len);
2839
2840                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2841                         stripe_nr = stripe_nr * map->num_stripes + i;
2842                         do_div(stripe_nr, map->sub_stripes);
2843                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2844                         stripe_nr = stripe_nr * map->num_stripes + i;
2845                 }
2846                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2847                 WARN_ON(nr >= map->num_stripes);
2848                 for (j = 0; j < nr; j++) {
2849                         if (buf[j] == bytenr)
2850                                 break;
2851                 }
2852                 if (j == nr) {
2853                         WARN_ON(nr >= map->num_stripes);
2854                         buf[nr++] = bytenr;
2855                 }
2856         }
2857
2858         *logical = buf;
2859         *naddrs = nr;
2860         *stripe_len = map->stripe_len;
2861
2862         free_extent_map(em);
2863         return 0;
2864 }
2865
2866 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2867                       u64 logical, struct page *page)
2868 {
2869         u64 length = PAGE_CACHE_SIZE;
2870         return __btrfs_map_block(map_tree, READ, logical, &length,
2871                                  NULL, 0, page);
2872 }
2873
2874 static void end_bio_multi_stripe(struct bio *bio, int err)
2875 {
2876         struct btrfs_multi_bio *multi = bio->bi_private;
2877         int is_orig_bio = 0;
2878
2879         if (err)
2880                 atomic_inc(&multi->error);
2881
2882         if (bio == multi->orig_bio)
2883                 is_orig_bio = 1;
2884
2885         if (atomic_dec_and_test(&multi->stripes_pending)) {
2886                 if (!is_orig_bio) {
2887                         bio_put(bio);
2888                         bio = multi->orig_bio;
2889                 }
2890                 bio->bi_private = multi->private;
2891                 bio->bi_end_io = multi->end_io;
2892                 /* only send an error to the higher layers if it is
2893                  * beyond the tolerance of the multi-bio
2894                  */
2895                 if (atomic_read(&multi->error) > multi->max_errors) {
2896                         err = -EIO;
2897                 } else if (err) {
2898                         /*
2899                          * this bio is actually up to date, we didn't
2900                          * go over the max number of errors
2901                          */
2902                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2903                         err = 0;
2904                 }
2905                 kfree(multi);
2906
2907                 bio_endio(bio, err);
2908         } else if (!is_orig_bio) {
2909                 bio_put(bio);
2910         }
2911 }
2912
2913 struct async_sched {
2914         struct bio *bio;
2915         int rw;
2916         struct btrfs_fs_info *info;
2917         struct btrfs_work work;
2918 };
2919
2920 /*
2921  * see run_scheduled_bios for a description of why bios are collected for
2922  * async submit.
2923  *
2924  * This will add one bio to the pending list for a device and make sure
2925  * the work struct is scheduled.
2926  */
2927 static noinline int schedule_bio(struct btrfs_root *root,
2928                                  struct btrfs_device *device,
2929                                  int rw, struct bio *bio)
2930 {
2931         int should_queue = 1;
2932         struct btrfs_pending_bios *pending_bios;
2933
2934         /* don't bother with additional async steps for reads, right now */
2935         if (!(rw & (1 << BIO_RW))) {
2936                 bio_get(bio);
2937                 submit_bio(rw, bio);
2938                 bio_put(bio);
2939                 return 0;
2940         }
2941
2942         /*
2943          * nr_async_bios allows us to reliably return congestion to the
2944          * higher layers.  Otherwise, the async bio makes it appear we have
2945          * made progress against dirty pages when we've really just put it
2946          * on a queue for later
2947          */
2948         atomic_inc(&root->fs_info->nr_async_bios);
2949         WARN_ON(bio->bi_next);
2950         bio->bi_next = NULL;
2951         bio->bi_rw |= rw;
2952
2953         spin_lock(&device->io_lock);
2954         if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2955                 pending_bios = &device->pending_sync_bios;
2956         else
2957                 pending_bios = &device->pending_bios;
2958
2959         if (pending_bios->tail)
2960                 pending_bios->tail->bi_next = bio;
2961
2962         pending_bios->tail = bio;
2963         if (!pending_bios->head)
2964                 pending_bios->head = bio;
2965         if (device->running_pending)
2966                 should_queue = 0;
2967
2968         spin_unlock(&device->io_lock);
2969
2970         if (should_queue)
2971                 btrfs_queue_worker(&root->fs_info->submit_workers,
2972                                    &device->work);
2973         return 0;
2974 }
2975
2976 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2977                   int mirror_num, int async_submit)
2978 {
2979         struct btrfs_mapping_tree *map_tree;
2980         struct btrfs_device *dev;
2981         struct bio *first_bio = bio;
2982         u64 logical = (u64)bio->bi_sector << 9;
2983         u64 length = 0;
2984         u64 map_length;
2985         struct btrfs_multi_bio *multi = NULL;
2986         int ret;
2987         int dev_nr = 0;
2988         int total_devs = 1;
2989
2990         length = bio->bi_size;
2991         map_tree = &root->fs_info->mapping_tree;
2992         map_length = length;
2993
2994         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2995                               mirror_num);
2996         BUG_ON(ret);
2997
2998         total_devs = multi->num_stripes;
2999         if (map_length < length) {
3000                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3001                        "len %llu\n", (unsigned long long)logical,
3002                        (unsigned long long)length,
3003                        (unsigned long long)map_length);
3004                 BUG();
3005         }
3006         multi->end_io = first_bio->bi_end_io;
3007         multi->private = first_bio->bi_private;
3008         multi->orig_bio = first_bio;
3009         atomic_set(&multi->stripes_pending, multi->num_stripes);
3010
3011         while (dev_nr < total_devs) {
3012                 if (total_devs > 1) {
3013                         if (dev_nr < total_devs - 1) {
3014                                 bio = bio_clone(first_bio, GFP_NOFS);
3015                                 BUG_ON(!bio);
3016                         } else {
3017                                 bio = first_bio;
3018                         }
3019                         bio->bi_private = multi;
3020                         bio->bi_end_io = end_bio_multi_stripe;
3021                 }
3022                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3023                 dev = multi->stripes[dev_nr].dev;
3024                 BUG_ON(rw == WRITE && !dev->writeable);
3025                 if (dev && dev->bdev) {
3026                         bio->bi_bdev = dev->bdev;
3027                         if (async_submit)
3028                                 schedule_bio(root, dev, rw, bio);
3029                         else
3030                                 submit_bio(rw, bio);
3031                 } else {
3032                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3033                         bio->bi_sector = logical >> 9;
3034                         bio_endio(bio, -EIO);
3035                 }
3036                 dev_nr++;
3037         }
3038         if (total_devs == 1)
3039                 kfree(multi);
3040         return 0;
3041 }
3042
3043 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3044                                        u8 *uuid, u8 *fsid)
3045 {
3046         struct btrfs_device *device;
3047         struct btrfs_fs_devices *cur_devices;
3048
3049         cur_devices = root->fs_info->fs_devices;
3050         while (cur_devices) {
3051                 if (!fsid ||
3052                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3053                         device = __find_device(&cur_devices->devices,
3054                                                devid, uuid);
3055                         if (device)
3056                                 return device;
3057                 }
3058                 cur_devices = cur_devices->seed;
3059         }
3060         return NULL;
3061 }
3062
3063 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3064                                             u64 devid, u8 *dev_uuid)
3065 {
3066         struct btrfs_device *device;
3067         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3068
3069         device = kzalloc(sizeof(*device), GFP_NOFS);
3070         if (!device)
3071                 return NULL;
3072         list_add(&device->dev_list,
3073                  &fs_devices->devices);
3074         device->barriers = 1;
3075         device->dev_root = root->fs_info->dev_root;
3076         device->devid = devid;
3077         device->work.func = pending_bios_fn;
3078         device->fs_devices = fs_devices;
3079         fs_devices->num_devices++;
3080         spin_lock_init(&device->io_lock);
3081         INIT_LIST_HEAD(&device->dev_alloc_list);
3082         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3083         return device;
3084 }
3085
3086 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3087                           struct extent_buffer *leaf,
3088                           struct btrfs_chunk *chunk)
3089 {
3090         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3091         struct map_lookup *map;
3092         struct extent_map *em;
3093         u64 logical;
3094         u64 length;
3095         u64 devid;
3096         u8 uuid[BTRFS_UUID_SIZE];
3097         int num_stripes;
3098         int ret;
3099         int i;
3100
3101         logical = key->offset;
3102         length = btrfs_chunk_length(leaf, chunk);
3103
3104         read_lock(&map_tree->map_tree.lock);
3105         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3106         read_unlock(&map_tree->map_tree.lock);
3107
3108         /* already mapped? */
3109         if (em && em->start <= logical && em->start + em->len > logical) {
3110                 free_extent_map(em);
3111                 return 0;
3112         } else if (em) {
3113                 free_extent_map(em);
3114         }
3115
3116         em = alloc_extent_map(GFP_NOFS);
3117         if (!em)
3118                 return -ENOMEM;
3119         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3120         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3121         if (!map) {
3122                 free_extent_map(em);
3123                 return -ENOMEM;
3124         }
3125
3126         em->bdev = (struct block_device *)map;
3127         em->start = logical;
3128         em->len = length;
3129         em->block_start = 0;
3130         em->block_len = em->len;
3131
3132         map->num_stripes = num_stripes;
3133         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3134         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3135         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3136         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3137         map->type = btrfs_chunk_type(leaf, chunk);
3138         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3139         for (i = 0; i < num_stripes; i++) {
3140                 map->stripes[i].physical =
3141                         btrfs_stripe_offset_nr(leaf, chunk, i);
3142                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3143                 read_extent_buffer(leaf, uuid, (unsigned long)
3144                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3145                                    BTRFS_UUID_SIZE);
3146                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3147                                                         NULL);
3148                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3149                         kfree(map);
3150                         free_extent_map(em);
3151                         return -EIO;
3152                 }
3153                 if (!map->stripes[i].dev) {
3154                         map->stripes[i].dev =
3155                                 add_missing_dev(root, devid, uuid);
3156                         if (!map->stripes[i].dev) {
3157                                 kfree(map);
3158                                 free_extent_map(em);
3159                                 return -EIO;
3160                         }
3161                 }
3162                 map->stripes[i].dev->in_fs_metadata = 1;
3163         }
3164
3165         write_lock(&map_tree->map_tree.lock);
3166         ret = add_extent_mapping(&map_tree->map_tree, em);
3167         write_unlock(&map_tree->map_tree.lock);
3168         BUG_ON(ret);
3169         free_extent_map(em);
3170
3171         return 0;
3172 }
3173
3174 static int fill_device_from_item(struct extent_buffer *leaf,
3175                                  struct btrfs_dev_item *dev_item,
3176                                  struct btrfs_device *device)
3177 {
3178         unsigned long ptr;
3179
3180         device->devid = btrfs_device_id(leaf, dev_item);
3181         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3182         device->total_bytes = device->disk_total_bytes;
3183         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3184         device->type = btrfs_device_type(leaf, dev_item);
3185         device->io_align = btrfs_device_io_align(leaf, dev_item);
3186         device->io_width = btrfs_device_io_width(leaf, dev_item);
3187         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3188
3189         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3190         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3191
3192         return 0;
3193 }
3194
3195 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3196 {
3197         struct btrfs_fs_devices *fs_devices;
3198         int ret;
3199
3200         mutex_lock(&uuid_mutex);
3201
3202         fs_devices = root->fs_info->fs_devices->seed;
3203         while (fs_devices) {
3204                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3205                         ret = 0;
3206                         goto out;
3207                 }
3208                 fs_devices = fs_devices->seed;
3209         }
3210
3211         fs_devices = find_fsid(fsid);
3212         if (!fs_devices) {
3213                 ret = -ENOENT;
3214                 goto out;
3215         }
3216
3217         fs_devices = clone_fs_devices(fs_devices);
3218         if (IS_ERR(fs_devices)) {
3219                 ret = PTR_ERR(fs_devices);
3220                 goto out;
3221         }
3222
3223         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3224                                    root->fs_info->bdev_holder);
3225         if (ret)
3226                 goto out;
3227
3228         if (!fs_devices->seeding) {
3229                 __btrfs_close_devices(fs_devices);
3230                 free_fs_devices(fs_devices);
3231                 ret = -EINVAL;
3232                 goto out;
3233         }
3234
3235         fs_devices->seed = root->fs_info->fs_devices->seed;
3236         root->fs_info->fs_devices->seed = fs_devices;
3237 out:
3238         mutex_unlock(&uuid_mutex);
3239         return ret;
3240 }
3241
3242 static int read_one_dev(struct btrfs_root *root,
3243                         struct extent_buffer *leaf,
3244                         struct btrfs_dev_item *dev_item)
3245 {
3246         struct btrfs_device *device;
3247         u64 devid;
3248         int ret;
3249         u8 fs_uuid[BTRFS_UUID_SIZE];
3250         u8 dev_uuid[BTRFS_UUID_SIZE];
3251
3252         devid = btrfs_device_id(leaf, dev_item);
3253         read_extent_buffer(leaf, dev_uuid,
3254                            (unsigned long)btrfs_device_uuid(dev_item),
3255                            BTRFS_UUID_SIZE);
3256         read_extent_buffer(leaf, fs_uuid,
3257                            (unsigned long)btrfs_device_fsid(dev_item),
3258                            BTRFS_UUID_SIZE);
3259
3260         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3261                 ret = open_seed_devices(root, fs_uuid);
3262                 if (ret && !btrfs_test_opt(root, DEGRADED))
3263                         return ret;
3264         }
3265
3266         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3267         if (!device || !device->bdev) {
3268                 if (!btrfs_test_opt(root, DEGRADED))
3269                         return -EIO;
3270
3271                 if (!device) {
3272                         printk(KERN_WARNING "warning devid %llu missing\n",
3273                                (unsigned long long)devid);
3274                         device = add_missing_dev(root, devid, dev_uuid);
3275                         if (!device)
3276                                 return -ENOMEM;
3277                 }
3278         }
3279
3280         if (device->fs_devices != root->fs_info->fs_devices) {
3281                 BUG_ON(device->writeable);
3282                 if (device->generation !=
3283                     btrfs_device_generation(leaf, dev_item))
3284                         return -EINVAL;
3285         }
3286
3287         fill_device_from_item(leaf, dev_item, device);
3288         device->dev_root = root->fs_info->dev_root;
3289         device->in_fs_metadata = 1;
3290         if (device->writeable)
3291                 device->fs_devices->total_rw_bytes += device->total_bytes;
3292         ret = 0;
3293         return ret;
3294 }
3295
3296 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3297 {
3298         struct btrfs_dev_item *dev_item;
3299
3300         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3301                                                      dev_item);
3302         return read_one_dev(root, buf, dev_item);
3303 }
3304
3305 int btrfs_read_sys_array(struct btrfs_root *root)
3306 {
3307         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3308         struct extent_buffer *sb;
3309         struct btrfs_disk_key *disk_key;
3310         struct btrfs_chunk *chunk;
3311         u8 *ptr;
3312         unsigned long sb_ptr;
3313         int ret = 0;
3314         u32 num_stripes;
3315         u32 array_size;
3316         u32 len = 0;
3317         u32 cur;
3318         struct btrfs_key key;
3319
3320         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3321                                           BTRFS_SUPER_INFO_SIZE);
3322         if (!sb)
3323                 return -ENOMEM;
3324         btrfs_set_buffer_uptodate(sb);
3325         btrfs_set_buffer_lockdep_class(sb, 0);
3326
3327         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3328         array_size = btrfs_super_sys_array_size(super_copy);
3329
3330         ptr = super_copy->sys_chunk_array;
3331         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3332         cur = 0;
3333
3334         while (cur < array_size) {
3335                 disk_key = (struct btrfs_disk_key *)ptr;
3336                 btrfs_disk_key_to_cpu(&key, disk_key);
3337
3338                 len = sizeof(*disk_key); ptr += len;
3339                 sb_ptr += len;
3340                 cur += len;
3341
3342                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3343                         chunk = (struct btrfs_chunk *)sb_ptr;
3344                         ret = read_one_chunk(root, &key, sb, chunk);
3345                         if (ret)
3346                                 break;
3347                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3348                         len = btrfs_chunk_item_size(num_stripes);
3349                 } else {
3350                         ret = -EIO;
3351                         break;
3352                 }
3353                 ptr += len;
3354                 sb_ptr += len;
3355                 cur += len;
3356         }
3357         free_extent_buffer(sb);
3358         return ret;
3359 }
3360
3361 int btrfs_read_chunk_tree(struct btrfs_root *root)
3362 {
3363         struct btrfs_path *path;
3364         struct extent_buffer *leaf;
3365         struct btrfs_key key;
3366         struct btrfs_key found_key;
3367         int ret;
3368         int slot;
3369
3370         root = root->fs_info->chunk_root;
3371
3372         path = btrfs_alloc_path();
3373         if (!path)
3374                 return -ENOMEM;
3375
3376         /* first we search for all of the device items, and then we
3377          * read in all of the chunk items.  This way we can create chunk
3378          * mappings that reference all of the devices that are afound
3379          */
3380         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3381         key.offset = 0;
3382         key.type = 0;
3383 again:
3384         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3385         while (1) {
3386                 leaf = path->nodes[0];
3387                 slot = path->slots[0];
3388                 if (slot >= btrfs_header_nritems(leaf)) {
3389                         ret = btrfs_next_leaf(root, path);
3390                         if (ret == 0)
3391                                 continue;
3392                         if (ret < 0)
3393                                 goto error;
3394                         break;
3395                 }
3396                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3397                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3398                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3399                                 break;
3400                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3401                                 struct btrfs_dev_item *dev_item;
3402                                 dev_item = btrfs_item_ptr(leaf, slot,
3403                                                   struct btrfs_dev_item);
3404                                 ret = read_one_dev(root, leaf, dev_item);
3405                                 if (ret)
3406                                         goto error;
3407                         }
3408                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3409                         struct btrfs_chunk *chunk;
3410                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3411                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3412                         if (ret)
3413                                 goto error;
3414                 }
3415                 path->slots[0]++;
3416         }
3417         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3418                 key.objectid = 0;
3419                 btrfs_release_path(root, path);
3420                 goto again;
3421         }
3422         ret = 0;
3423 error:
3424         btrfs_free_path(path);
3425         return ret;
3426 }