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