Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[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
260                 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
261                         num_sync_run++;
262
263                 submit_bio(cur->bi_rw, cur);
264                 num_run++;
265                 batch_run++;
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          * IO has already been through a long path to get here.  Checksumming,
330          * async helper threads, perhaps compression.  We've done a pretty
331          * good job of collecting a batch of IO and should just unplug
332          * the device right away.
333          *
334          * This will help anyone who is waiting on the IO, they might have
335          * already unplugged, but managed to do so before the bio they
336          * cared about found its way down here.
337          */
338         blk_run_backing_dev(bdi, NULL);
339
340         cond_resched();
341         if (again)
342                 goto loop;
343
344         spin_lock(&device->io_lock);
345         if (device->pending_bios.head || device->pending_sync_bios.head)
346                 goto loop_lock;
347         spin_unlock(&device->io_lock);
348
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         char *name;
369
370         fs_devices = find_fsid(disk_super->fsid);
371         if (!fs_devices) {
372                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
373                 if (!fs_devices)
374                         return -ENOMEM;
375                 INIT_LIST_HEAD(&fs_devices->devices);
376                 INIT_LIST_HEAD(&fs_devices->alloc_list);
377                 list_add(&fs_devices->list, &fs_uuids);
378                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
379                 fs_devices->latest_devid = devid;
380                 fs_devices->latest_trans = found_transid;
381                 mutex_init(&fs_devices->device_list_mutex);
382                 device = NULL;
383         } else {
384                 device = __find_device(&fs_devices->devices, devid,
385                                        disk_super->dev_item.uuid);
386         }
387         if (!device) {
388                 if (fs_devices->opened)
389                         return -EBUSY;
390
391                 device = kzalloc(sizeof(*device), GFP_NOFS);
392                 if (!device) {
393                         /* we can safely leave the fs_devices entry around */
394                         return -ENOMEM;
395                 }
396                 device->devid = devid;
397                 device->work.func = pending_bios_fn;
398                 memcpy(device->uuid, disk_super->dev_item.uuid,
399                        BTRFS_UUID_SIZE);
400                 device->barriers = 1;
401                 spin_lock_init(&device->io_lock);
402                 device->name = kstrdup(path, GFP_NOFS);
403                 if (!device->name) {
404                         kfree(device);
405                         return -ENOMEM;
406                 }
407                 INIT_LIST_HEAD(&device->dev_alloc_list);
408
409                 mutex_lock(&fs_devices->device_list_mutex);
410                 list_add(&device->dev_list, &fs_devices->devices);
411                 mutex_unlock(&fs_devices->device_list_mutex);
412
413                 device->fs_devices = fs_devices;
414                 fs_devices->num_devices++;
415         } else if (strcmp(device->name, path)) {
416                 name = kstrdup(path, GFP_NOFS);
417                 if (!name)
418                         return -ENOMEM;
419                 kfree(device->name);
420                 device->name = name;
421         }
422
423         if (found_transid > fs_devices->latest_trans) {
424                 fs_devices->latest_devid = devid;
425                 fs_devices->latest_trans = found_transid;
426         }
427         *fs_devices_ret = fs_devices;
428         return 0;
429 }
430
431 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
432 {
433         struct btrfs_fs_devices *fs_devices;
434         struct btrfs_device *device;
435         struct btrfs_device *orig_dev;
436
437         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
438         if (!fs_devices)
439                 return ERR_PTR(-ENOMEM);
440
441         INIT_LIST_HEAD(&fs_devices->devices);
442         INIT_LIST_HEAD(&fs_devices->alloc_list);
443         INIT_LIST_HEAD(&fs_devices->list);
444         mutex_init(&fs_devices->device_list_mutex);
445         fs_devices->latest_devid = orig->latest_devid;
446         fs_devices->latest_trans = orig->latest_trans;
447         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
448
449         mutex_lock(&orig->device_list_mutex);
450         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
451                 device = kzalloc(sizeof(*device), GFP_NOFS);
452                 if (!device)
453                         goto error;
454
455                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
456                 if (!device->name) {
457                         kfree(device);
458                         goto error;
459                 }
460
461                 device->devid = orig_dev->devid;
462                 device->work.func = pending_bios_fn;
463                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
464                 device->barriers = 1;
465                 spin_lock_init(&device->io_lock);
466                 INIT_LIST_HEAD(&device->dev_list);
467                 INIT_LIST_HEAD(&device->dev_alloc_list);
468
469                 list_add(&device->dev_list, &fs_devices->devices);
470                 device->fs_devices = fs_devices;
471                 fs_devices->num_devices++;
472         }
473         mutex_unlock(&orig->device_list_mutex);
474         return fs_devices;
475 error:
476         mutex_unlock(&orig->device_list_mutex);
477         free_fs_devices(fs_devices);
478         return ERR_PTR(-ENOMEM);
479 }
480
481 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
482 {
483         struct btrfs_device *device, *next;
484
485         mutex_lock(&uuid_mutex);
486 again:
487         mutex_lock(&fs_devices->device_list_mutex);
488         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
489                 if (device->in_fs_metadata)
490                         continue;
491
492                 if (device->bdev) {
493                         close_bdev_exclusive(device->bdev, device->mode);
494                         device->bdev = NULL;
495                         fs_devices->open_devices--;
496                 }
497                 if (device->writeable) {
498                         list_del_init(&device->dev_alloc_list);
499                         device->writeable = 0;
500                         fs_devices->rw_devices--;
501                 }
502                 list_del_init(&device->dev_list);
503                 fs_devices->num_devices--;
504                 kfree(device->name);
505                 kfree(device);
506         }
507         mutex_unlock(&fs_devices->device_list_mutex);
508
509         if (fs_devices->seed) {
510                 fs_devices = fs_devices->seed;
511                 goto again;
512         }
513
514         mutex_unlock(&uuid_mutex);
515         return 0;
516 }
517
518 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
519 {
520         struct btrfs_device *device;
521
522         if (--fs_devices->opened > 0)
523                 return 0;
524
525         list_for_each_entry(device, &fs_devices->devices, dev_list) {
526                 if (device->bdev) {
527                         close_bdev_exclusive(device->bdev, device->mode);
528                         fs_devices->open_devices--;
529                 }
530                 if (device->writeable) {
531                         list_del_init(&device->dev_alloc_list);
532                         fs_devices->rw_devices--;
533                 }
534
535                 device->bdev = NULL;
536                 device->writeable = 0;
537                 device->in_fs_metadata = 0;
538         }
539         WARN_ON(fs_devices->open_devices);
540         WARN_ON(fs_devices->rw_devices);
541         fs_devices->opened = 0;
542         fs_devices->seeding = 0;
543
544         return 0;
545 }
546
547 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
548 {
549         struct btrfs_fs_devices *seed_devices = NULL;
550         int ret;
551
552         mutex_lock(&uuid_mutex);
553         ret = __btrfs_close_devices(fs_devices);
554         if (!fs_devices->opened) {
555                 seed_devices = fs_devices->seed;
556                 fs_devices->seed = NULL;
557         }
558         mutex_unlock(&uuid_mutex);
559
560         while (seed_devices) {
561                 fs_devices = seed_devices;
562                 seed_devices = fs_devices->seed;
563                 __btrfs_close_devices(fs_devices);
564                 free_fs_devices(fs_devices);
565         }
566         return ret;
567 }
568
569 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
570                                 fmode_t flags, void *holder)
571 {
572         struct block_device *bdev;
573         struct list_head *head = &fs_devices->devices;
574         struct btrfs_device *device;
575         struct block_device *latest_bdev = NULL;
576         struct buffer_head *bh;
577         struct btrfs_super_block *disk_super;
578         u64 latest_devid = 0;
579         u64 latest_transid = 0;
580         u64 devid;
581         int seeding = 1;
582         int ret = 0;
583
584         list_for_each_entry(device, head, dev_list) {
585                 if (device->bdev)
586                         continue;
587                 if (!device->name)
588                         continue;
589
590                 bdev = open_bdev_exclusive(device->name, flags, holder);
591                 if (IS_ERR(bdev)) {
592                         printk(KERN_INFO "open %s failed\n", device->name);
593                         goto error;
594                 }
595                 set_blocksize(bdev, 4096);
596
597                 bh = btrfs_read_dev_super(bdev);
598                 if (!bh)
599                         goto error_close;
600
601                 disk_super = (struct btrfs_super_block *)bh->b_data;
602                 devid = btrfs_stack_device_id(&disk_super->dev_item);
603                 if (devid != device->devid)
604                         goto error_brelse;
605
606                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
607                            BTRFS_UUID_SIZE))
608                         goto error_brelse;
609
610                 device->generation = btrfs_super_generation(disk_super);
611                 if (!latest_transid || device->generation > latest_transid) {
612                         latest_devid = devid;
613                         latest_transid = device->generation;
614                         latest_bdev = bdev;
615                 }
616
617                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
618                         device->writeable = 0;
619                 } else {
620                         device->writeable = !bdev_read_only(bdev);
621                         seeding = 0;
622                 }
623
624                 device->bdev = bdev;
625                 device->in_fs_metadata = 0;
626                 device->mode = flags;
627
628                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
629                         fs_devices->rotating = 1;
630
631                 fs_devices->open_devices++;
632                 if (device->writeable) {
633                         fs_devices->rw_devices++;
634                         list_add(&device->dev_alloc_list,
635                                  &fs_devices->alloc_list);
636                 }
637                 continue;
638
639 error_brelse:
640                 brelse(bh);
641 error_close:
642                 close_bdev_exclusive(bdev, FMODE_READ);
643 error:
644                 continue;
645         }
646         if (fs_devices->open_devices == 0) {
647                 ret = -EIO;
648                 goto out;
649         }
650         fs_devices->seeding = seeding;
651         fs_devices->opened = 1;
652         fs_devices->latest_bdev = latest_bdev;
653         fs_devices->latest_devid = latest_devid;
654         fs_devices->latest_trans = latest_transid;
655         fs_devices->total_rw_bytes = 0;
656 out:
657         return ret;
658 }
659
660 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
661                        fmode_t flags, void *holder)
662 {
663         int ret;
664
665         mutex_lock(&uuid_mutex);
666         if (fs_devices->opened) {
667                 fs_devices->opened++;
668                 ret = 0;
669         } else {
670                 ret = __btrfs_open_devices(fs_devices, flags, holder);
671         }
672         mutex_unlock(&uuid_mutex);
673         return ret;
674 }
675
676 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
677                           struct btrfs_fs_devices **fs_devices_ret)
678 {
679         struct btrfs_super_block *disk_super;
680         struct block_device *bdev;
681         struct buffer_head *bh;
682         int ret;
683         u64 devid;
684         u64 transid;
685
686         mutex_lock(&uuid_mutex);
687
688         bdev = open_bdev_exclusive(path, flags, holder);
689
690         if (IS_ERR(bdev)) {
691                 ret = PTR_ERR(bdev);
692                 goto error;
693         }
694
695         ret = set_blocksize(bdev, 4096);
696         if (ret)
697                 goto error_close;
698         bh = btrfs_read_dev_super(bdev);
699         if (!bh) {
700                 ret = -EIO;
701                 goto error_close;
702         }
703         disk_super = (struct btrfs_super_block *)bh->b_data;
704         devid = btrfs_stack_device_id(&disk_super->dev_item);
705         transid = btrfs_super_generation(disk_super);
706         if (disk_super->label[0])
707                 printk(KERN_INFO "device label %s ", disk_super->label);
708         else {
709                 /* FIXME, make a readl uuid parser */
710                 printk(KERN_INFO "device fsid %llx-%llx ",
711                        *(unsigned long long *)disk_super->fsid,
712                        *(unsigned long long *)(disk_super->fsid + 8));
713         }
714         printk(KERN_CONT "devid %llu transid %llu %s\n",
715                (unsigned long long)devid, (unsigned long long)transid, path);
716         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
717
718         brelse(bh);
719 error_close:
720         close_bdev_exclusive(bdev, flags);
721 error:
722         mutex_unlock(&uuid_mutex);
723         return ret;
724 }
725
726 /*
727  * this uses a pretty simple search, the expectation is that it is
728  * called very infrequently and that a given device has a small number
729  * of extents
730  */
731 int find_free_dev_extent(struct btrfs_trans_handle *trans,
732                          struct btrfs_device *device, u64 num_bytes,
733                          u64 *start, u64 *max_avail)
734 {
735         struct btrfs_key key;
736         struct btrfs_root *root = device->dev_root;
737         struct btrfs_dev_extent *dev_extent = NULL;
738         struct btrfs_path *path;
739         u64 hole_size = 0;
740         u64 last_byte = 0;
741         u64 search_start = 0;
742         u64 search_end = device->total_bytes;
743         int ret;
744         int slot = 0;
745         int start_found;
746         struct extent_buffer *l;
747
748         path = btrfs_alloc_path();
749         if (!path)
750                 return -ENOMEM;
751         path->reada = 2;
752         start_found = 0;
753
754         /* FIXME use last free of some kind */
755
756         /* we don't want to overwrite the superblock on the drive,
757          * so we make sure to start at an offset of at least 1MB
758          */
759         search_start = max((u64)1024 * 1024, search_start);
760
761         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
762                 search_start = max(root->fs_info->alloc_start, search_start);
763
764         key.objectid = device->devid;
765         key.offset = search_start;
766         key.type = BTRFS_DEV_EXTENT_KEY;
767         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
768         if (ret < 0)
769                 goto error;
770         if (ret > 0) {
771                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
772                 if (ret < 0)
773                         goto error;
774                 if (ret > 0)
775                         start_found = 1;
776         }
777         l = path->nodes[0];
778         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
779         while (1) {
780                 l = path->nodes[0];
781                 slot = path->slots[0];
782                 if (slot >= btrfs_header_nritems(l)) {
783                         ret = btrfs_next_leaf(root, path);
784                         if (ret == 0)
785                                 continue;
786                         if (ret < 0)
787                                 goto error;
788 no_more_items:
789                         if (!start_found) {
790                                 if (search_start >= search_end) {
791                                         ret = -ENOSPC;
792                                         goto error;
793                                 }
794                                 *start = search_start;
795                                 start_found = 1;
796                                 goto check_pending;
797                         }
798                         *start = last_byte > search_start ?
799                                 last_byte : search_start;
800                         if (search_end <= *start) {
801                                 ret = -ENOSPC;
802                                 goto error;
803                         }
804                         goto check_pending;
805                 }
806                 btrfs_item_key_to_cpu(l, &key, slot);
807
808                 if (key.objectid < device->devid)
809                         goto next;
810
811                 if (key.objectid > device->devid)
812                         goto no_more_items;
813
814                 if (key.offset >= search_start && key.offset > last_byte &&
815                     start_found) {
816                         if (last_byte < search_start)
817                                 last_byte = search_start;
818                         hole_size = key.offset - last_byte;
819
820                         if (hole_size > *max_avail)
821                                 *max_avail = hole_size;
822
823                         if (key.offset > last_byte &&
824                             hole_size >= num_bytes) {
825                                 *start = last_byte;
826                                 goto check_pending;
827                         }
828                 }
829                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
830                         goto next;
831
832                 start_found = 1;
833                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
834                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
835 next:
836                 path->slots[0]++;
837                 cond_resched();
838         }
839 check_pending:
840         /* we have to make sure we didn't find an extent that has already
841          * been allocated by the map tree or the original allocation
842          */
843         BUG_ON(*start < search_start);
844
845         if (*start + num_bytes > search_end) {
846                 ret = -ENOSPC;
847                 goto error;
848         }
849         /* check for pending inserts here */
850         ret = 0;
851
852 error:
853         btrfs_free_path(path);
854         return ret;
855 }
856
857 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
858                           struct btrfs_device *device,
859                           u64 start)
860 {
861         int ret;
862         struct btrfs_path *path;
863         struct btrfs_root *root = device->dev_root;
864         struct btrfs_key key;
865         struct btrfs_key found_key;
866         struct extent_buffer *leaf = NULL;
867         struct btrfs_dev_extent *extent = NULL;
868
869         path = btrfs_alloc_path();
870         if (!path)
871                 return -ENOMEM;
872
873         key.objectid = device->devid;
874         key.offset = start;
875         key.type = BTRFS_DEV_EXTENT_KEY;
876
877         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
878         if (ret > 0) {
879                 ret = btrfs_previous_item(root, path, key.objectid,
880                                           BTRFS_DEV_EXTENT_KEY);
881                 BUG_ON(ret);
882                 leaf = path->nodes[0];
883                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
884                 extent = btrfs_item_ptr(leaf, path->slots[0],
885                                         struct btrfs_dev_extent);
886                 BUG_ON(found_key.offset > start || found_key.offset +
887                        btrfs_dev_extent_length(leaf, extent) < start);
888                 ret = 0;
889         } else if (ret == 0) {
890                 leaf = path->nodes[0];
891                 extent = btrfs_item_ptr(leaf, path->slots[0],
892                                         struct btrfs_dev_extent);
893         }
894         BUG_ON(ret);
895
896         if (device->bytes_used > 0)
897                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
898         ret = btrfs_del_item(trans, root, path);
899         BUG_ON(ret);
900
901         btrfs_free_path(path);
902         return ret;
903 }
904
905 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
906                            struct btrfs_device *device,
907                            u64 chunk_tree, u64 chunk_objectid,
908                            u64 chunk_offset, u64 start, u64 num_bytes)
909 {
910         int ret;
911         struct btrfs_path *path;
912         struct btrfs_root *root = device->dev_root;
913         struct btrfs_dev_extent *extent;
914         struct extent_buffer *leaf;
915         struct btrfs_key key;
916
917         WARN_ON(!device->in_fs_metadata);
918         path = btrfs_alloc_path();
919         if (!path)
920                 return -ENOMEM;
921
922         key.objectid = device->devid;
923         key.offset = start;
924         key.type = BTRFS_DEV_EXTENT_KEY;
925         ret = btrfs_insert_empty_item(trans, root, path, &key,
926                                       sizeof(*extent));
927         BUG_ON(ret);
928
929         leaf = path->nodes[0];
930         extent = btrfs_item_ptr(leaf, path->slots[0],
931                                 struct btrfs_dev_extent);
932         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
933         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
934         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
935
936         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
937                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
938                     BTRFS_UUID_SIZE);
939
940         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
941         btrfs_mark_buffer_dirty(leaf);
942         btrfs_free_path(path);
943         return ret;
944 }
945
946 static noinline int find_next_chunk(struct btrfs_root *root,
947                                     u64 objectid, u64 *offset)
948 {
949         struct btrfs_path *path;
950         int ret;
951         struct btrfs_key key;
952         struct btrfs_chunk *chunk;
953         struct btrfs_key found_key;
954
955         path = btrfs_alloc_path();
956         BUG_ON(!path);
957
958         key.objectid = objectid;
959         key.offset = (u64)-1;
960         key.type = BTRFS_CHUNK_ITEM_KEY;
961
962         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
963         if (ret < 0)
964                 goto error;
965
966         BUG_ON(ret == 0);
967
968         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
969         if (ret) {
970                 *offset = 0;
971         } else {
972                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
973                                       path->slots[0]);
974                 if (found_key.objectid != objectid)
975                         *offset = 0;
976                 else {
977                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
978                                                struct btrfs_chunk);
979                         *offset = found_key.offset +
980                                 btrfs_chunk_length(path->nodes[0], chunk);
981                 }
982         }
983         ret = 0;
984 error:
985         btrfs_free_path(path);
986         return ret;
987 }
988
989 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
990 {
991         int ret;
992         struct btrfs_key key;
993         struct btrfs_key found_key;
994         struct btrfs_path *path;
995
996         root = root->fs_info->chunk_root;
997
998         path = btrfs_alloc_path();
999         if (!path)
1000                 return -ENOMEM;
1001
1002         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1003         key.type = BTRFS_DEV_ITEM_KEY;
1004         key.offset = (u64)-1;
1005
1006         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1007         if (ret < 0)
1008                 goto error;
1009
1010         BUG_ON(ret == 0);
1011
1012         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1013                                   BTRFS_DEV_ITEM_KEY);
1014         if (ret) {
1015                 *objectid = 1;
1016         } else {
1017                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1018                                       path->slots[0]);
1019                 *objectid = found_key.offset + 1;
1020         }
1021         ret = 0;
1022 error:
1023         btrfs_free_path(path);
1024         return ret;
1025 }
1026
1027 /*
1028  * the device information is stored in the chunk root
1029  * the btrfs_device struct should be fully filled in
1030  */
1031 int btrfs_add_device(struct btrfs_trans_handle *trans,
1032                      struct btrfs_root *root,
1033                      struct btrfs_device *device)
1034 {
1035         int ret;
1036         struct btrfs_path *path;
1037         struct btrfs_dev_item *dev_item;
1038         struct extent_buffer *leaf;
1039         struct btrfs_key key;
1040         unsigned long ptr;
1041
1042         root = root->fs_info->chunk_root;
1043
1044         path = btrfs_alloc_path();
1045         if (!path)
1046                 return -ENOMEM;
1047
1048         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1049         key.type = BTRFS_DEV_ITEM_KEY;
1050         key.offset = device->devid;
1051
1052         ret = btrfs_insert_empty_item(trans, root, path, &key,
1053                                       sizeof(*dev_item));
1054         if (ret)
1055                 goto out;
1056
1057         leaf = path->nodes[0];
1058         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1059
1060         btrfs_set_device_id(leaf, dev_item, device->devid);
1061         btrfs_set_device_generation(leaf, dev_item, 0);
1062         btrfs_set_device_type(leaf, dev_item, device->type);
1063         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1064         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1065         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1066         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1067         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1068         btrfs_set_device_group(leaf, dev_item, 0);
1069         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1070         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1071         btrfs_set_device_start_offset(leaf, dev_item, 0);
1072
1073         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1074         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1075         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1076         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1077         btrfs_mark_buffer_dirty(leaf);
1078
1079         ret = 0;
1080 out:
1081         btrfs_free_path(path);
1082         return ret;
1083 }
1084
1085 static int btrfs_rm_dev_item(struct btrfs_root *root,
1086                              struct btrfs_device *device)
1087 {
1088         int ret;
1089         struct btrfs_path *path;
1090         struct btrfs_key key;
1091         struct btrfs_trans_handle *trans;
1092
1093         root = root->fs_info->chunk_root;
1094
1095         path = btrfs_alloc_path();
1096         if (!path)
1097                 return -ENOMEM;
1098
1099         trans = btrfs_start_transaction(root, 1);
1100         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1101         key.type = BTRFS_DEV_ITEM_KEY;
1102         key.offset = device->devid;
1103         lock_chunks(root);
1104
1105         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106         if (ret < 0)
1107                 goto out;
1108
1109         if (ret > 0) {
1110                 ret = -ENOENT;
1111                 goto out;
1112         }
1113
1114         ret = btrfs_del_item(trans, root, path);
1115         if (ret)
1116                 goto out;
1117 out:
1118         btrfs_free_path(path);
1119         unlock_chunks(root);
1120         btrfs_commit_transaction(trans, root);
1121         return ret;
1122 }
1123
1124 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1125 {
1126         struct btrfs_device *device;
1127         struct btrfs_device *next_device;
1128         struct block_device *bdev;
1129         struct buffer_head *bh = NULL;
1130         struct btrfs_super_block *disk_super;
1131         u64 all_avail;
1132         u64 devid;
1133         u64 num_devices;
1134         u8 *dev_uuid;
1135         int ret = 0;
1136
1137         mutex_lock(&uuid_mutex);
1138         mutex_lock(&root->fs_info->volume_mutex);
1139
1140         all_avail = root->fs_info->avail_data_alloc_bits |
1141                 root->fs_info->avail_system_alloc_bits |
1142                 root->fs_info->avail_metadata_alloc_bits;
1143
1144         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1145             root->fs_info->fs_devices->num_devices <= 4) {
1146                 printk(KERN_ERR "btrfs: unable to go below four devices "
1147                        "on raid10\n");
1148                 ret = -EINVAL;
1149                 goto out;
1150         }
1151
1152         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1153             root->fs_info->fs_devices->num_devices <= 2) {
1154                 printk(KERN_ERR "btrfs: unable to go below two "
1155                        "devices on raid1\n");
1156                 ret = -EINVAL;
1157                 goto out;
1158         }
1159
1160         if (strcmp(device_path, "missing") == 0) {
1161                 struct list_head *devices;
1162                 struct btrfs_device *tmp;
1163
1164                 device = NULL;
1165                 devices = &root->fs_info->fs_devices->devices;
1166                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1167                 list_for_each_entry(tmp, devices, dev_list) {
1168                         if (tmp->in_fs_metadata && !tmp->bdev) {
1169                                 device = tmp;
1170                                 break;
1171                         }
1172                 }
1173                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1174                 bdev = NULL;
1175                 bh = NULL;
1176                 disk_super = NULL;
1177                 if (!device) {
1178                         printk(KERN_ERR "btrfs: no missing devices found to "
1179                                "remove\n");
1180                         goto out;
1181                 }
1182         } else {
1183                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1184                                       root->fs_info->bdev_holder);
1185                 if (IS_ERR(bdev)) {
1186                         ret = PTR_ERR(bdev);
1187                         goto out;
1188                 }
1189
1190                 set_blocksize(bdev, 4096);
1191                 bh = btrfs_read_dev_super(bdev);
1192                 if (!bh) {
1193                         ret = -EIO;
1194                         goto error_close;
1195                 }
1196                 disk_super = (struct btrfs_super_block *)bh->b_data;
1197                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1198                 dev_uuid = disk_super->dev_item.uuid;
1199                 device = btrfs_find_device(root, devid, dev_uuid,
1200                                            disk_super->fsid);
1201                 if (!device) {
1202                         ret = -ENOENT;
1203                         goto error_brelse;
1204                 }
1205         }
1206
1207         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1208                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1209                        "device\n");
1210                 ret = -EINVAL;
1211                 goto error_brelse;
1212         }
1213
1214         if (device->writeable) {
1215                 list_del_init(&device->dev_alloc_list);
1216                 root->fs_info->fs_devices->rw_devices--;
1217         }
1218
1219         ret = btrfs_shrink_device(device, 0);
1220         if (ret)
1221                 goto error_brelse;
1222
1223         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1224         if (ret)
1225                 goto error_brelse;
1226
1227         device->in_fs_metadata = 0;
1228
1229         /*
1230          * the device list mutex makes sure that we don't change
1231          * the device list while someone else is writing out all
1232          * the device supers.
1233          */
1234         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1235         list_del_init(&device->dev_list);
1236         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1237
1238         device->fs_devices->num_devices--;
1239
1240         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1241                                  struct btrfs_device, dev_list);
1242         if (device->bdev == root->fs_info->sb->s_bdev)
1243                 root->fs_info->sb->s_bdev = next_device->bdev;
1244         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1245                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1246
1247         if (device->bdev) {
1248                 close_bdev_exclusive(device->bdev, device->mode);
1249                 device->bdev = NULL;
1250                 device->fs_devices->open_devices--;
1251         }
1252
1253         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1254         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1255
1256         if (device->fs_devices->open_devices == 0) {
1257                 struct btrfs_fs_devices *fs_devices;
1258                 fs_devices = root->fs_info->fs_devices;
1259                 while (fs_devices) {
1260                         if (fs_devices->seed == device->fs_devices)
1261                                 break;
1262                         fs_devices = fs_devices->seed;
1263                 }
1264                 fs_devices->seed = device->fs_devices->seed;
1265                 device->fs_devices->seed = NULL;
1266                 __btrfs_close_devices(device->fs_devices);
1267                 free_fs_devices(device->fs_devices);
1268         }
1269
1270         /*
1271          * at this point, the device is zero sized.  We want to
1272          * remove it from the devices list and zero out the old super
1273          */
1274         if (device->writeable) {
1275                 /* make sure this device isn't detected as part of
1276                  * the FS anymore
1277                  */
1278                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1279                 set_buffer_dirty(bh);
1280                 sync_dirty_buffer(bh);
1281         }
1282
1283         kfree(device->name);
1284         kfree(device);
1285         ret = 0;
1286
1287 error_brelse:
1288         brelse(bh);
1289 error_close:
1290         if (bdev)
1291                 close_bdev_exclusive(bdev, FMODE_READ);
1292 out:
1293         mutex_unlock(&root->fs_info->volume_mutex);
1294         mutex_unlock(&uuid_mutex);
1295         return ret;
1296 }
1297
1298 /*
1299  * does all the dirty work required for changing file system's UUID.
1300  */
1301 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1302                                 struct btrfs_root *root)
1303 {
1304         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1305         struct btrfs_fs_devices *old_devices;
1306         struct btrfs_fs_devices *seed_devices;
1307         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1308         struct btrfs_device *device;
1309         u64 super_flags;
1310
1311         BUG_ON(!mutex_is_locked(&uuid_mutex));
1312         if (!fs_devices->seeding)
1313                 return -EINVAL;
1314
1315         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1316         if (!seed_devices)
1317                 return -ENOMEM;
1318
1319         old_devices = clone_fs_devices(fs_devices);
1320         if (IS_ERR(old_devices)) {
1321                 kfree(seed_devices);
1322                 return PTR_ERR(old_devices);
1323         }
1324
1325         list_add(&old_devices->list, &fs_uuids);
1326
1327         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1328         seed_devices->opened = 1;
1329         INIT_LIST_HEAD(&seed_devices->devices);
1330         INIT_LIST_HEAD(&seed_devices->alloc_list);
1331         mutex_init(&seed_devices->device_list_mutex);
1332         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1333         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1334         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1335                 device->fs_devices = seed_devices;
1336         }
1337
1338         fs_devices->seeding = 0;
1339         fs_devices->num_devices = 0;
1340         fs_devices->open_devices = 0;
1341         fs_devices->seed = seed_devices;
1342
1343         generate_random_uuid(fs_devices->fsid);
1344         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1345         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1346         super_flags = btrfs_super_flags(disk_super) &
1347                       ~BTRFS_SUPER_FLAG_SEEDING;
1348         btrfs_set_super_flags(disk_super, super_flags);
1349
1350         return 0;
1351 }
1352
1353 /*
1354  * strore the expected generation for seed devices in device items.
1355  */
1356 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1357                                struct btrfs_root *root)
1358 {
1359         struct btrfs_path *path;
1360         struct extent_buffer *leaf;
1361         struct btrfs_dev_item *dev_item;
1362         struct btrfs_device *device;
1363         struct btrfs_key key;
1364         u8 fs_uuid[BTRFS_UUID_SIZE];
1365         u8 dev_uuid[BTRFS_UUID_SIZE];
1366         u64 devid;
1367         int ret;
1368
1369         path = btrfs_alloc_path();
1370         if (!path)
1371                 return -ENOMEM;
1372
1373         root = root->fs_info->chunk_root;
1374         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1375         key.offset = 0;
1376         key.type = BTRFS_DEV_ITEM_KEY;
1377
1378         while (1) {
1379                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1380                 if (ret < 0)
1381                         goto error;
1382
1383                 leaf = path->nodes[0];
1384 next_slot:
1385                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1386                         ret = btrfs_next_leaf(root, path);
1387                         if (ret > 0)
1388                                 break;
1389                         if (ret < 0)
1390                                 goto error;
1391                         leaf = path->nodes[0];
1392                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1393                         btrfs_release_path(root, path);
1394                         continue;
1395                 }
1396
1397                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1398                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1399                     key.type != BTRFS_DEV_ITEM_KEY)
1400                         break;
1401
1402                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1403                                           struct btrfs_dev_item);
1404                 devid = btrfs_device_id(leaf, dev_item);
1405                 read_extent_buffer(leaf, dev_uuid,
1406                                    (unsigned long)btrfs_device_uuid(dev_item),
1407                                    BTRFS_UUID_SIZE);
1408                 read_extent_buffer(leaf, fs_uuid,
1409                                    (unsigned long)btrfs_device_fsid(dev_item),
1410                                    BTRFS_UUID_SIZE);
1411                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1412                 BUG_ON(!device);
1413
1414                 if (device->fs_devices->seeding) {
1415                         btrfs_set_device_generation(leaf, dev_item,
1416                                                     device->generation);
1417                         btrfs_mark_buffer_dirty(leaf);
1418                 }
1419
1420                 path->slots[0]++;
1421                 goto next_slot;
1422         }
1423         ret = 0;
1424 error:
1425         btrfs_free_path(path);
1426         return ret;
1427 }
1428
1429 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1430 {
1431         struct btrfs_trans_handle *trans;
1432         struct btrfs_device *device;
1433         struct block_device *bdev;
1434         struct list_head *devices;
1435         struct super_block *sb = root->fs_info->sb;
1436         u64 total_bytes;
1437         int seeding_dev = 0;
1438         int ret = 0;
1439
1440         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1441                 return -EINVAL;
1442
1443         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1444         if (IS_ERR(bdev))
1445                 return PTR_ERR(bdev);
1446
1447         if (root->fs_info->fs_devices->seeding) {
1448                 seeding_dev = 1;
1449                 down_write(&sb->s_umount);
1450                 mutex_lock(&uuid_mutex);
1451         }
1452
1453         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1454         mutex_lock(&root->fs_info->volume_mutex);
1455
1456         devices = &root->fs_info->fs_devices->devices;
1457         /*
1458          * we have the volume lock, so we don't need the extra
1459          * device list mutex while reading the list here.
1460          */
1461         list_for_each_entry(device, devices, dev_list) {
1462                 if (device->bdev == bdev) {
1463                         ret = -EEXIST;
1464                         goto error;
1465                 }
1466         }
1467
1468         device = kzalloc(sizeof(*device), GFP_NOFS);
1469         if (!device) {
1470                 /* we can safely leave the fs_devices entry around */
1471                 ret = -ENOMEM;
1472                 goto error;
1473         }
1474
1475         device->name = kstrdup(device_path, GFP_NOFS);
1476         if (!device->name) {
1477                 kfree(device);
1478                 ret = -ENOMEM;
1479                 goto error;
1480         }
1481
1482         ret = find_next_devid(root, &device->devid);
1483         if (ret) {
1484                 kfree(device);
1485                 goto error;
1486         }
1487
1488         trans = btrfs_start_transaction(root, 1);
1489         lock_chunks(root);
1490
1491         device->barriers = 1;
1492         device->writeable = 1;
1493         device->work.func = pending_bios_fn;
1494         generate_random_uuid(device->uuid);
1495         spin_lock_init(&device->io_lock);
1496         device->generation = trans->transid;
1497         device->io_width = root->sectorsize;
1498         device->io_align = root->sectorsize;
1499         device->sector_size = root->sectorsize;
1500         device->total_bytes = i_size_read(bdev->bd_inode);
1501         device->disk_total_bytes = device->total_bytes;
1502         device->dev_root = root->fs_info->dev_root;
1503         device->bdev = bdev;
1504         device->in_fs_metadata = 1;
1505         device->mode = 0;
1506         set_blocksize(device->bdev, 4096);
1507
1508         if (seeding_dev) {
1509                 sb->s_flags &= ~MS_RDONLY;
1510                 ret = btrfs_prepare_sprout(trans, root);
1511                 BUG_ON(ret);
1512         }
1513
1514         device->fs_devices = root->fs_info->fs_devices;
1515
1516         /*
1517          * we don't want write_supers to jump in here with our device
1518          * half setup
1519          */
1520         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1521         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1522         list_add(&device->dev_alloc_list,
1523                  &root->fs_info->fs_devices->alloc_list);
1524         root->fs_info->fs_devices->num_devices++;
1525         root->fs_info->fs_devices->open_devices++;
1526         root->fs_info->fs_devices->rw_devices++;
1527         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1528
1529         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1530                 root->fs_info->fs_devices->rotating = 1;
1531
1532         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1533         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1534                                     total_bytes + device->total_bytes);
1535
1536         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1537         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1538                                     total_bytes + 1);
1539         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1540
1541         if (seeding_dev) {
1542                 ret = init_first_rw_device(trans, root, device);
1543                 BUG_ON(ret);
1544                 ret = btrfs_finish_sprout(trans, root);
1545                 BUG_ON(ret);
1546         } else {
1547                 ret = btrfs_add_device(trans, root, device);
1548         }
1549
1550         /*
1551          * we've got more storage, clear any full flags on the space
1552          * infos
1553          */
1554         btrfs_clear_space_info_full(root->fs_info);
1555
1556         unlock_chunks(root);
1557         btrfs_commit_transaction(trans, root);
1558
1559         if (seeding_dev) {
1560                 mutex_unlock(&uuid_mutex);
1561                 up_write(&sb->s_umount);
1562
1563                 ret = btrfs_relocate_sys_chunks(root);
1564                 BUG_ON(ret);
1565         }
1566 out:
1567         mutex_unlock(&root->fs_info->volume_mutex);
1568         return ret;
1569 error:
1570         close_bdev_exclusive(bdev, 0);
1571         if (seeding_dev) {
1572                 mutex_unlock(&uuid_mutex);
1573                 up_write(&sb->s_umount);
1574         }
1575         goto out;
1576 }
1577
1578 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1579                                         struct btrfs_device *device)
1580 {
1581         int ret;
1582         struct btrfs_path *path;
1583         struct btrfs_root *root;
1584         struct btrfs_dev_item *dev_item;
1585         struct extent_buffer *leaf;
1586         struct btrfs_key key;
1587
1588         root = device->dev_root->fs_info->chunk_root;
1589
1590         path = btrfs_alloc_path();
1591         if (!path)
1592                 return -ENOMEM;
1593
1594         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1595         key.type = BTRFS_DEV_ITEM_KEY;
1596         key.offset = device->devid;
1597
1598         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1599         if (ret < 0)
1600                 goto out;
1601
1602         if (ret > 0) {
1603                 ret = -ENOENT;
1604                 goto out;
1605         }
1606
1607         leaf = path->nodes[0];
1608         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1609
1610         btrfs_set_device_id(leaf, dev_item, device->devid);
1611         btrfs_set_device_type(leaf, dev_item, device->type);
1612         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1613         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1614         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1615         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1616         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1617         btrfs_mark_buffer_dirty(leaf);
1618
1619 out:
1620         btrfs_free_path(path);
1621         return ret;
1622 }
1623
1624 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1625                       struct btrfs_device *device, u64 new_size)
1626 {
1627         struct btrfs_super_block *super_copy =
1628                 &device->dev_root->fs_info->super_copy;
1629         u64 old_total = btrfs_super_total_bytes(super_copy);
1630         u64 diff = new_size - device->total_bytes;
1631
1632         if (!device->writeable)
1633                 return -EACCES;
1634         if (new_size <= device->total_bytes)
1635                 return -EINVAL;
1636
1637         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1638         device->fs_devices->total_rw_bytes += diff;
1639
1640         device->total_bytes = new_size;
1641         device->disk_total_bytes = new_size;
1642         btrfs_clear_space_info_full(device->dev_root->fs_info);
1643
1644         return btrfs_update_device(trans, device);
1645 }
1646
1647 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1648                       struct btrfs_device *device, u64 new_size)
1649 {
1650         int ret;
1651         lock_chunks(device->dev_root);
1652         ret = __btrfs_grow_device(trans, device, new_size);
1653         unlock_chunks(device->dev_root);
1654         return ret;
1655 }
1656
1657 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1658                             struct btrfs_root *root,
1659                             u64 chunk_tree, u64 chunk_objectid,
1660                             u64 chunk_offset)
1661 {
1662         int ret;
1663         struct btrfs_path *path;
1664         struct btrfs_key key;
1665
1666         root = root->fs_info->chunk_root;
1667         path = btrfs_alloc_path();
1668         if (!path)
1669                 return -ENOMEM;
1670
1671         key.objectid = chunk_objectid;
1672         key.offset = chunk_offset;
1673         key.type = BTRFS_CHUNK_ITEM_KEY;
1674
1675         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1676         BUG_ON(ret);
1677
1678         ret = btrfs_del_item(trans, root, path);
1679         BUG_ON(ret);
1680
1681         btrfs_free_path(path);
1682         return 0;
1683 }
1684
1685 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1686                         chunk_offset)
1687 {
1688         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1689         struct btrfs_disk_key *disk_key;
1690         struct btrfs_chunk *chunk;
1691         u8 *ptr;
1692         int ret = 0;
1693         u32 num_stripes;
1694         u32 array_size;
1695         u32 len = 0;
1696         u32 cur;
1697         struct btrfs_key key;
1698
1699         array_size = btrfs_super_sys_array_size(super_copy);
1700
1701         ptr = super_copy->sys_chunk_array;
1702         cur = 0;
1703
1704         while (cur < array_size) {
1705                 disk_key = (struct btrfs_disk_key *)ptr;
1706                 btrfs_disk_key_to_cpu(&key, disk_key);
1707
1708                 len = sizeof(*disk_key);
1709
1710                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1711                         chunk = (struct btrfs_chunk *)(ptr + len);
1712                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1713                         len += btrfs_chunk_item_size(num_stripes);
1714                 } else {
1715                         ret = -EIO;
1716                         break;
1717                 }
1718                 if (key.objectid == chunk_objectid &&
1719                     key.offset == chunk_offset) {
1720                         memmove(ptr, ptr + len, array_size - (cur + len));
1721                         array_size -= len;
1722                         btrfs_set_super_sys_array_size(super_copy, array_size);
1723                 } else {
1724                         ptr += len;
1725                         cur += len;
1726                 }
1727         }
1728         return ret;
1729 }
1730
1731 static int btrfs_relocate_chunk(struct btrfs_root *root,
1732                          u64 chunk_tree, u64 chunk_objectid,
1733                          u64 chunk_offset)
1734 {
1735         struct extent_map_tree *em_tree;
1736         struct btrfs_root *extent_root;
1737         struct btrfs_trans_handle *trans;
1738         struct extent_map *em;
1739         struct map_lookup *map;
1740         int ret;
1741         int i;
1742
1743         root = root->fs_info->chunk_root;
1744         extent_root = root->fs_info->extent_root;
1745         em_tree = &root->fs_info->mapping_tree.map_tree;
1746
1747         ret = btrfs_can_relocate(extent_root, chunk_offset);
1748         if (ret)
1749                 return -ENOSPC;
1750
1751         /* step one, relocate all the extents inside this chunk */
1752         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1753         BUG_ON(ret);
1754
1755         trans = btrfs_start_transaction(root, 1);
1756         BUG_ON(!trans);
1757
1758         lock_chunks(root);
1759
1760         /*
1761          * step two, delete the device extents and the
1762          * chunk tree entries
1763          */
1764         read_lock(&em_tree->lock);
1765         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1766         read_unlock(&em_tree->lock);
1767
1768         BUG_ON(em->start > chunk_offset ||
1769                em->start + em->len < chunk_offset);
1770         map = (struct map_lookup *)em->bdev;
1771
1772         for (i = 0; i < map->num_stripes; i++) {
1773                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1774                                             map->stripes[i].physical);
1775                 BUG_ON(ret);
1776
1777                 if (map->stripes[i].dev) {
1778                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1779                         BUG_ON(ret);
1780                 }
1781         }
1782         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1783                                chunk_offset);
1784
1785         BUG_ON(ret);
1786
1787         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1788                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1789                 BUG_ON(ret);
1790         }
1791
1792         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1793         BUG_ON(ret);
1794
1795         write_lock(&em_tree->lock);
1796         remove_extent_mapping(em_tree, em);
1797         write_unlock(&em_tree->lock);
1798
1799         kfree(map);
1800         em->bdev = NULL;
1801
1802         /* once for the tree */
1803         free_extent_map(em);
1804         /* once for us */
1805         free_extent_map(em);
1806
1807         unlock_chunks(root);
1808         btrfs_end_transaction(trans, root);
1809         return 0;
1810 }
1811
1812 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1813 {
1814         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1815         struct btrfs_path *path;
1816         struct extent_buffer *leaf;
1817         struct btrfs_chunk *chunk;
1818         struct btrfs_key key;
1819         struct btrfs_key found_key;
1820         u64 chunk_tree = chunk_root->root_key.objectid;
1821         u64 chunk_type;
1822         bool retried = false;
1823         int failed = 0;
1824         int ret;
1825
1826         path = btrfs_alloc_path();
1827         if (!path)
1828                 return -ENOMEM;
1829
1830 again:
1831         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1832         key.offset = (u64)-1;
1833         key.type = BTRFS_CHUNK_ITEM_KEY;
1834
1835         while (1) {
1836                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1837                 if (ret < 0)
1838                         goto error;
1839                 BUG_ON(ret == 0);
1840
1841                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1842                                           key.type);
1843                 if (ret < 0)
1844                         goto error;
1845                 if (ret > 0)
1846                         break;
1847
1848                 leaf = path->nodes[0];
1849                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1850
1851                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1852                                        struct btrfs_chunk);
1853                 chunk_type = btrfs_chunk_type(leaf, chunk);
1854                 btrfs_release_path(chunk_root, path);
1855
1856                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1857                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1858                                                    found_key.objectid,
1859                                                    found_key.offset);
1860                         if (ret == -ENOSPC)
1861                                 failed++;
1862                         else if (ret)
1863                                 BUG();
1864                 }
1865
1866                 if (found_key.offset == 0)
1867                         break;
1868                 key.offset = found_key.offset - 1;
1869         }
1870         ret = 0;
1871         if (failed && !retried) {
1872                 failed = 0;
1873                 retried = true;
1874                 goto again;
1875         } else if (failed && retried) {
1876                 WARN_ON(1);
1877                 ret = -ENOSPC;
1878         }
1879 error:
1880         btrfs_free_path(path);
1881         return ret;
1882 }
1883
1884 static u64 div_factor(u64 num, int factor)
1885 {
1886         if (factor == 10)
1887                 return num;
1888         num *= factor;
1889         do_div(num, 10);
1890         return num;
1891 }
1892
1893 int btrfs_balance(struct btrfs_root *dev_root)
1894 {
1895         int ret;
1896         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1897         struct btrfs_device *device;
1898         u64 old_size;
1899         u64 size_to_free;
1900         struct btrfs_path *path;
1901         struct btrfs_key key;
1902         struct btrfs_chunk *chunk;
1903         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1904         struct btrfs_trans_handle *trans;
1905         struct btrfs_key found_key;
1906
1907         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1908                 return -EROFS;
1909
1910         mutex_lock(&dev_root->fs_info->volume_mutex);
1911         dev_root = dev_root->fs_info->dev_root;
1912
1913         /* step one make some room on all the devices */
1914         list_for_each_entry(device, devices, dev_list) {
1915                 old_size = device->total_bytes;
1916                 size_to_free = div_factor(old_size, 1);
1917                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1918                 if (!device->writeable ||
1919                     device->total_bytes - device->bytes_used > size_to_free)
1920                         continue;
1921
1922                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1923                 if (ret == -ENOSPC)
1924                         break;
1925                 BUG_ON(ret);
1926
1927                 trans = btrfs_start_transaction(dev_root, 1);
1928                 BUG_ON(!trans);
1929
1930                 ret = btrfs_grow_device(trans, device, old_size);
1931                 BUG_ON(ret);
1932
1933                 btrfs_end_transaction(trans, dev_root);
1934         }
1935
1936         /* step two, relocate all the chunks */
1937         path = btrfs_alloc_path();
1938         BUG_ON(!path);
1939
1940         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1941         key.offset = (u64)-1;
1942         key.type = BTRFS_CHUNK_ITEM_KEY;
1943
1944         while (1) {
1945                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1946                 if (ret < 0)
1947                         goto error;
1948
1949                 /*
1950                  * this shouldn't happen, it means the last relocate
1951                  * failed
1952                  */
1953                 if (ret == 0)
1954                         break;
1955
1956                 ret = btrfs_previous_item(chunk_root, path, 0,
1957                                           BTRFS_CHUNK_ITEM_KEY);
1958                 if (ret)
1959                         break;
1960
1961                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1962                                       path->slots[0]);
1963                 if (found_key.objectid != key.objectid)
1964                         break;
1965
1966                 chunk = btrfs_item_ptr(path->nodes[0],
1967                                        path->slots[0],
1968                                        struct btrfs_chunk);
1969                 /* chunk zero is special */
1970                 if (found_key.offset == 0)
1971                         break;
1972
1973                 btrfs_release_path(chunk_root, path);
1974                 ret = btrfs_relocate_chunk(chunk_root,
1975                                            chunk_root->root_key.objectid,
1976                                            found_key.objectid,
1977                                            found_key.offset);
1978                 BUG_ON(ret && ret != -ENOSPC);
1979                 key.offset = found_key.offset - 1;
1980         }
1981         ret = 0;
1982 error:
1983         btrfs_free_path(path);
1984         mutex_unlock(&dev_root->fs_info->volume_mutex);
1985         return ret;
1986 }
1987
1988 /*
1989  * shrinking a device means finding all of the device extents past
1990  * the new size, and then following the back refs to the chunks.
1991  * The chunk relocation code actually frees the device extent
1992  */
1993 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1994 {
1995         struct btrfs_trans_handle *trans;
1996         struct btrfs_root *root = device->dev_root;
1997         struct btrfs_dev_extent *dev_extent = NULL;
1998         struct btrfs_path *path;
1999         u64 length;
2000         u64 chunk_tree;
2001         u64 chunk_objectid;
2002         u64 chunk_offset;
2003         int ret;
2004         int slot;
2005         int failed = 0;
2006         bool retried = false;
2007         struct extent_buffer *l;
2008         struct btrfs_key key;
2009         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2010         u64 old_total = btrfs_super_total_bytes(super_copy);
2011         u64 old_size = device->total_bytes;
2012         u64 diff = device->total_bytes - new_size;
2013
2014         if (new_size >= device->total_bytes)
2015                 return -EINVAL;
2016
2017         path = btrfs_alloc_path();
2018         if (!path)
2019                 return -ENOMEM;
2020
2021         path->reada = 2;
2022
2023         lock_chunks(root);
2024
2025         device->total_bytes = new_size;
2026         if (device->writeable)
2027                 device->fs_devices->total_rw_bytes -= diff;
2028         unlock_chunks(root);
2029
2030 again:
2031         key.objectid = device->devid;
2032         key.offset = (u64)-1;
2033         key.type = BTRFS_DEV_EXTENT_KEY;
2034
2035         while (1) {
2036                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2037                 if (ret < 0)
2038                         goto done;
2039
2040                 ret = btrfs_previous_item(root, path, 0, key.type);
2041                 if (ret < 0)
2042                         goto done;
2043                 if (ret) {
2044                         ret = 0;
2045                         btrfs_release_path(root, path);
2046                         break;
2047                 }
2048
2049                 l = path->nodes[0];
2050                 slot = path->slots[0];
2051                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2052
2053                 if (key.objectid != device->devid) {
2054                         btrfs_release_path(root, path);
2055                         break;
2056                 }
2057
2058                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2059                 length = btrfs_dev_extent_length(l, dev_extent);
2060
2061                 if (key.offset + length <= new_size) {
2062                         btrfs_release_path(root, path);
2063                         break;
2064                 }
2065
2066                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2067                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2068                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2069                 btrfs_release_path(root, path);
2070
2071                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2072                                            chunk_offset);
2073                 if (ret && ret != -ENOSPC)
2074                         goto done;
2075                 if (ret == -ENOSPC)
2076                         failed++;
2077                 key.offset -= 1;
2078         }
2079
2080         if (failed && !retried) {
2081                 failed = 0;
2082                 retried = true;
2083                 goto again;
2084         } else if (failed && retried) {
2085                 ret = -ENOSPC;
2086                 lock_chunks(root);
2087
2088                 device->total_bytes = old_size;
2089                 if (device->writeable)
2090                         device->fs_devices->total_rw_bytes += diff;
2091                 unlock_chunks(root);
2092                 goto done;
2093         }
2094
2095         /* Shrinking succeeded, else we would be at "done". */
2096         trans = btrfs_start_transaction(root, 1);
2097         if (!trans) {
2098                 ret = -ENOMEM;
2099                 goto done;
2100         }
2101         lock_chunks(root);
2102
2103         device->disk_total_bytes = new_size;
2104         /* Now btrfs_update_device() will change the on-disk size. */
2105         ret = btrfs_update_device(trans, device);
2106         if (ret) {
2107                 unlock_chunks(root);
2108                 btrfs_end_transaction(trans, root);
2109                 goto done;
2110         }
2111         WARN_ON(diff > old_total);
2112         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2113         unlock_chunks(root);
2114         btrfs_end_transaction(trans, root);
2115 done:
2116         btrfs_free_path(path);
2117         return ret;
2118 }
2119
2120 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2121                            struct btrfs_root *root,
2122                            struct btrfs_key *key,
2123                            struct btrfs_chunk *chunk, int item_size)
2124 {
2125         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2126         struct btrfs_disk_key disk_key;
2127         u32 array_size;
2128         u8 *ptr;
2129
2130         array_size = btrfs_super_sys_array_size(super_copy);
2131         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2132                 return -EFBIG;
2133
2134         ptr = super_copy->sys_chunk_array + array_size;
2135         btrfs_cpu_key_to_disk(&disk_key, key);
2136         memcpy(ptr, &disk_key, sizeof(disk_key));
2137         ptr += sizeof(disk_key);
2138         memcpy(ptr, chunk, item_size);
2139         item_size += sizeof(disk_key);
2140         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2141         return 0;
2142 }
2143
2144 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2145                                         int num_stripes, int sub_stripes)
2146 {
2147         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2148                 return calc_size;
2149         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2150                 return calc_size * (num_stripes / sub_stripes);
2151         else
2152                 return calc_size * num_stripes;
2153 }
2154
2155 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2156                                struct btrfs_root *extent_root,
2157                                struct map_lookup **map_ret,
2158                                u64 *num_bytes, u64 *stripe_size,
2159                                u64 start, u64 type)
2160 {
2161         struct btrfs_fs_info *info = extent_root->fs_info;
2162         struct btrfs_device *device = NULL;
2163         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2164         struct list_head *cur;
2165         struct map_lookup *map = NULL;
2166         struct extent_map_tree *em_tree;
2167         struct extent_map *em;
2168         struct list_head private_devs;
2169         int min_stripe_size = 1 * 1024 * 1024;
2170         u64 calc_size = 1024 * 1024 * 1024;
2171         u64 max_chunk_size = calc_size;
2172         u64 min_free;
2173         u64 avail;
2174         u64 max_avail = 0;
2175         u64 dev_offset;
2176         int num_stripes = 1;
2177         int min_stripes = 1;
2178         int sub_stripes = 0;
2179         int looped = 0;
2180         int ret;
2181         int index;
2182         int stripe_len = 64 * 1024;
2183
2184         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2185             (type & BTRFS_BLOCK_GROUP_DUP)) {
2186                 WARN_ON(1);
2187                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2188         }
2189         if (list_empty(&fs_devices->alloc_list))
2190                 return -ENOSPC;
2191
2192         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2193                 num_stripes = fs_devices->rw_devices;
2194                 min_stripes = 2;
2195         }
2196         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2197                 num_stripes = 2;
2198                 min_stripes = 2;
2199         }
2200         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2201                 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2202                 if (num_stripes < 2)
2203                         return -ENOSPC;
2204                 min_stripes = 2;
2205         }
2206         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2207                 num_stripes = fs_devices->rw_devices;
2208                 if (num_stripes < 4)
2209                         return -ENOSPC;
2210                 num_stripes &= ~(u32)1;
2211                 sub_stripes = 2;
2212                 min_stripes = 4;
2213         }
2214
2215         if (type & BTRFS_BLOCK_GROUP_DATA) {
2216                 max_chunk_size = 10 * calc_size;
2217                 min_stripe_size = 64 * 1024 * 1024;
2218         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2219                 max_chunk_size = 256 * 1024 * 1024;
2220                 min_stripe_size = 32 * 1024 * 1024;
2221         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2222                 calc_size = 8 * 1024 * 1024;
2223                 max_chunk_size = calc_size * 2;
2224                 min_stripe_size = 1 * 1024 * 1024;
2225         }
2226
2227         /* we don't want a chunk larger than 10% of writeable space */
2228         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2229                              max_chunk_size);
2230
2231 again:
2232         max_avail = 0;
2233         if (!map || map->num_stripes != num_stripes) {
2234                 kfree(map);
2235                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2236                 if (!map)
2237                         return -ENOMEM;
2238                 map->num_stripes = num_stripes;
2239         }
2240
2241         if (calc_size * num_stripes > max_chunk_size) {
2242                 calc_size = max_chunk_size;
2243                 do_div(calc_size, num_stripes);
2244                 do_div(calc_size, stripe_len);
2245                 calc_size *= stripe_len;
2246         }
2247         /* we don't want tiny stripes */
2248         calc_size = max_t(u64, min_stripe_size, calc_size);
2249
2250         do_div(calc_size, stripe_len);
2251         calc_size *= stripe_len;
2252
2253         cur = fs_devices->alloc_list.next;
2254         index = 0;
2255
2256         if (type & BTRFS_BLOCK_GROUP_DUP)
2257                 min_free = calc_size * 2;
2258         else
2259                 min_free = calc_size;
2260
2261         /*
2262          * we add 1MB because we never use the first 1MB of the device, unless
2263          * we've looped, then we are likely allocating the maximum amount of
2264          * space left already
2265          */
2266         if (!looped)
2267                 min_free += 1024 * 1024;
2268
2269         INIT_LIST_HEAD(&private_devs);
2270         while (index < num_stripes) {
2271                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2272                 BUG_ON(!device->writeable);
2273                 if (device->total_bytes > device->bytes_used)
2274                         avail = device->total_bytes - device->bytes_used;
2275                 else
2276                         avail = 0;
2277                 cur = cur->next;
2278
2279                 if (device->in_fs_metadata && avail >= min_free) {
2280                         ret = find_free_dev_extent(trans, device,
2281                                                    min_free, &dev_offset,
2282                                                    &max_avail);
2283                         if (ret == 0) {
2284                                 list_move_tail(&device->dev_alloc_list,
2285                                                &private_devs);
2286                                 map->stripes[index].dev = device;
2287                                 map->stripes[index].physical = dev_offset;
2288                                 index++;
2289                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2290                                         map->stripes[index].dev = device;
2291                                         map->stripes[index].physical =
2292                                                 dev_offset + calc_size;
2293                                         index++;
2294                                 }
2295                         }
2296                 } else if (device->in_fs_metadata && avail > max_avail)
2297                         max_avail = avail;
2298                 if (cur == &fs_devices->alloc_list)
2299                         break;
2300         }
2301         list_splice(&private_devs, &fs_devices->alloc_list);
2302         if (index < num_stripes) {
2303                 if (index >= min_stripes) {
2304                         num_stripes = index;
2305                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2306                                 num_stripes /= sub_stripes;
2307                                 num_stripes *= sub_stripes;
2308                         }
2309                         looped = 1;
2310                         goto again;
2311                 }
2312                 if (!looped && max_avail > 0) {
2313                         looped = 1;
2314                         calc_size = max_avail;
2315                         goto again;
2316                 }
2317                 kfree(map);
2318                 return -ENOSPC;
2319         }
2320         map->sector_size = extent_root->sectorsize;
2321         map->stripe_len = stripe_len;
2322         map->io_align = stripe_len;
2323         map->io_width = stripe_len;
2324         map->type = type;
2325         map->num_stripes = num_stripes;
2326         map->sub_stripes = sub_stripes;
2327
2328         *map_ret = map;
2329         *stripe_size = calc_size;
2330         *num_bytes = chunk_bytes_by_type(type, calc_size,
2331                                          num_stripes, sub_stripes);
2332
2333         em = alloc_extent_map(GFP_NOFS);
2334         if (!em) {
2335                 kfree(map);
2336                 return -ENOMEM;
2337         }
2338         em->bdev = (struct block_device *)map;
2339         em->start = start;
2340         em->len = *num_bytes;
2341         em->block_start = 0;
2342         em->block_len = em->len;
2343
2344         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2345         write_lock(&em_tree->lock);
2346         ret = add_extent_mapping(em_tree, em);
2347         write_unlock(&em_tree->lock);
2348         BUG_ON(ret);
2349         free_extent_map(em);
2350
2351         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2352                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2353                                      start, *num_bytes);
2354         BUG_ON(ret);
2355
2356         index = 0;
2357         while (index < map->num_stripes) {
2358                 device = map->stripes[index].dev;
2359                 dev_offset = map->stripes[index].physical;
2360
2361                 ret = btrfs_alloc_dev_extent(trans, device,
2362                                 info->chunk_root->root_key.objectid,
2363                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2364                                 start, dev_offset, calc_size);
2365                 BUG_ON(ret);
2366                 index++;
2367         }
2368
2369         return 0;
2370 }
2371
2372 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2373                                 struct btrfs_root *extent_root,
2374                                 struct map_lookup *map, u64 chunk_offset,
2375                                 u64 chunk_size, u64 stripe_size)
2376 {
2377         u64 dev_offset;
2378         struct btrfs_key key;
2379         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2380         struct btrfs_device *device;
2381         struct btrfs_chunk *chunk;
2382         struct btrfs_stripe *stripe;
2383         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2384         int index = 0;
2385         int ret;
2386
2387         chunk = kzalloc(item_size, GFP_NOFS);
2388         if (!chunk)
2389                 return -ENOMEM;
2390
2391         index = 0;
2392         while (index < map->num_stripes) {
2393                 device = map->stripes[index].dev;
2394                 device->bytes_used += stripe_size;
2395                 ret = btrfs_update_device(trans, device);
2396                 BUG_ON(ret);
2397                 index++;
2398         }
2399
2400         index = 0;
2401         stripe = &chunk->stripe;
2402         while (index < map->num_stripes) {
2403                 device = map->stripes[index].dev;
2404                 dev_offset = map->stripes[index].physical;
2405
2406                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2407                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2408                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2409                 stripe++;
2410                 index++;
2411         }
2412
2413         btrfs_set_stack_chunk_length(chunk, chunk_size);
2414         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2415         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2416         btrfs_set_stack_chunk_type(chunk, map->type);
2417         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2418         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2419         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2420         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2421         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2422
2423         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2424         key.type = BTRFS_CHUNK_ITEM_KEY;
2425         key.offset = chunk_offset;
2426
2427         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2428         BUG_ON(ret);
2429
2430         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2431                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2432                                              item_size);
2433                 BUG_ON(ret);
2434         }
2435         kfree(chunk);
2436         return 0;
2437 }
2438
2439 /*
2440  * Chunk allocation falls into two parts. The first part does works
2441  * that make the new allocated chunk useable, but not do any operation
2442  * that modifies the chunk tree. The second part does the works that
2443  * require modifying the chunk tree. This division is important for the
2444  * bootstrap process of adding storage to a seed btrfs.
2445  */
2446 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2447                       struct btrfs_root *extent_root, u64 type)
2448 {
2449         u64 chunk_offset;
2450         u64 chunk_size;
2451         u64 stripe_size;
2452         struct map_lookup *map;
2453         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2454         int ret;
2455
2456         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2457                               &chunk_offset);
2458         if (ret)
2459                 return ret;
2460
2461         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2462                                   &stripe_size, chunk_offset, type);
2463         if (ret)
2464                 return ret;
2465
2466         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2467                                    chunk_size, stripe_size);
2468         BUG_ON(ret);
2469         return 0;
2470 }
2471
2472 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2473                                          struct btrfs_root *root,
2474                                          struct btrfs_device *device)
2475 {
2476         u64 chunk_offset;
2477         u64 sys_chunk_offset;
2478         u64 chunk_size;
2479         u64 sys_chunk_size;
2480         u64 stripe_size;
2481         u64 sys_stripe_size;
2482         u64 alloc_profile;
2483         struct map_lookup *map;
2484         struct map_lookup *sys_map;
2485         struct btrfs_fs_info *fs_info = root->fs_info;
2486         struct btrfs_root *extent_root = fs_info->extent_root;
2487         int ret;
2488
2489         ret = find_next_chunk(fs_info->chunk_root,
2490                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2491         BUG_ON(ret);
2492
2493         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2494                         (fs_info->metadata_alloc_profile &
2495                          fs_info->avail_metadata_alloc_bits);
2496         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2497
2498         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2499                                   &stripe_size, chunk_offset, alloc_profile);
2500         BUG_ON(ret);
2501
2502         sys_chunk_offset = chunk_offset + chunk_size;
2503
2504         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2505                         (fs_info->system_alloc_profile &
2506                          fs_info->avail_system_alloc_bits);
2507         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2508
2509         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2510                                   &sys_chunk_size, &sys_stripe_size,
2511                                   sys_chunk_offset, alloc_profile);
2512         BUG_ON(ret);
2513
2514         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2515         BUG_ON(ret);
2516
2517         /*
2518          * Modifying chunk tree needs allocating new blocks from both
2519          * system block group and metadata block group. So we only can
2520          * do operations require modifying the chunk tree after both
2521          * block groups were created.
2522          */
2523         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2524                                    chunk_size, stripe_size);
2525         BUG_ON(ret);
2526
2527         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2528                                    sys_chunk_offset, sys_chunk_size,
2529                                    sys_stripe_size);
2530         BUG_ON(ret);
2531         return 0;
2532 }
2533
2534 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2535 {
2536         struct extent_map *em;
2537         struct map_lookup *map;
2538         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2539         int readonly = 0;
2540         int i;
2541
2542         read_lock(&map_tree->map_tree.lock);
2543         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2544         read_unlock(&map_tree->map_tree.lock);
2545         if (!em)
2546                 return 1;
2547
2548         if (btrfs_test_opt(root, DEGRADED)) {
2549                 free_extent_map(em);
2550                 return 0;
2551         }
2552
2553         map = (struct map_lookup *)em->bdev;
2554         for (i = 0; i < map->num_stripes; i++) {
2555                 if (!map->stripes[i].dev->writeable) {
2556                         readonly = 1;
2557                         break;
2558                 }
2559         }
2560         free_extent_map(em);
2561         return readonly;
2562 }
2563
2564 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2565 {
2566         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2567 }
2568
2569 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2570 {
2571         struct extent_map *em;
2572
2573         while (1) {
2574                 write_lock(&tree->map_tree.lock);
2575                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2576                 if (em)
2577                         remove_extent_mapping(&tree->map_tree, em);
2578                 write_unlock(&tree->map_tree.lock);
2579                 if (!em)
2580                         break;
2581                 kfree(em->bdev);
2582                 /* once for us */
2583                 free_extent_map(em);
2584                 /* once for the tree */
2585                 free_extent_map(em);
2586         }
2587 }
2588
2589 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2590 {
2591         struct extent_map *em;
2592         struct map_lookup *map;
2593         struct extent_map_tree *em_tree = &map_tree->map_tree;
2594         int ret;
2595
2596         read_lock(&em_tree->lock);
2597         em = lookup_extent_mapping(em_tree, logical, len);
2598         read_unlock(&em_tree->lock);
2599         BUG_ON(!em);
2600
2601         BUG_ON(em->start > logical || em->start + em->len < logical);
2602         map = (struct map_lookup *)em->bdev;
2603         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2604                 ret = map->num_stripes;
2605         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2606                 ret = map->sub_stripes;
2607         else
2608                 ret = 1;
2609         free_extent_map(em);
2610         return ret;
2611 }
2612
2613 static int find_live_mirror(struct map_lookup *map, int first, int num,
2614                             int optimal)
2615 {
2616         int i;
2617         if (map->stripes[optimal].dev->bdev)
2618                 return optimal;
2619         for (i = first; i < first + num; i++) {
2620                 if (map->stripes[i].dev->bdev)
2621                         return i;
2622         }
2623         /* we couldn't find one that doesn't fail.  Just return something
2624          * and the io error handling code will clean up eventually
2625          */
2626         return optimal;
2627 }
2628
2629 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2630                              u64 logical, u64 *length,
2631                              struct btrfs_multi_bio **multi_ret,
2632                              int mirror_num, struct page *unplug_page)
2633 {
2634         struct extent_map *em;
2635         struct map_lookup *map;
2636         struct extent_map_tree *em_tree = &map_tree->map_tree;
2637         u64 offset;
2638         u64 stripe_offset;
2639         u64 stripe_nr;
2640         int stripes_allocated = 8;
2641         int stripes_required = 1;
2642         int stripe_index;
2643         int i;
2644         int num_stripes;
2645         int max_errors = 0;
2646         struct btrfs_multi_bio *multi = NULL;
2647
2648         if (multi_ret && !(rw & (1 << BIO_RW)))
2649                 stripes_allocated = 1;
2650 again:
2651         if (multi_ret) {
2652                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2653                                 GFP_NOFS);
2654                 if (!multi)
2655                         return -ENOMEM;
2656
2657                 atomic_set(&multi->error, 0);
2658         }
2659
2660         read_lock(&em_tree->lock);
2661         em = lookup_extent_mapping(em_tree, logical, *length);
2662         read_unlock(&em_tree->lock);
2663
2664         if (!em && unplug_page) {
2665                 kfree(multi);
2666                 return 0;
2667         }
2668
2669         if (!em) {
2670                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2671                        (unsigned long long)logical,
2672                        (unsigned long long)*length);
2673                 BUG();
2674         }
2675
2676         BUG_ON(em->start > logical || em->start + em->len < logical);
2677         map = (struct map_lookup *)em->bdev;
2678         offset = logical - em->start;
2679
2680         if (mirror_num > map->num_stripes)
2681                 mirror_num = 0;
2682
2683         /* if our multi bio struct is too small, back off and try again */
2684         if (rw & (1 << BIO_RW)) {
2685                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2686                                  BTRFS_BLOCK_GROUP_DUP)) {
2687                         stripes_required = map->num_stripes;
2688                         max_errors = 1;
2689                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2690                         stripes_required = map->sub_stripes;
2691                         max_errors = 1;
2692                 }
2693         }
2694         if (multi_ret && (rw & (1 << BIO_RW)) &&
2695             stripes_allocated < stripes_required) {
2696                 stripes_allocated = map->num_stripes;
2697                 free_extent_map(em);
2698                 kfree(multi);
2699                 goto again;
2700         }
2701         stripe_nr = offset;
2702         /*
2703          * stripe_nr counts the total number of stripes we have to stride
2704          * to get to this block
2705          */
2706         do_div(stripe_nr, map->stripe_len);
2707
2708         stripe_offset = stripe_nr * map->stripe_len;
2709         BUG_ON(offset < stripe_offset);
2710
2711         /* stripe_offset is the offset of this block in its stripe*/
2712         stripe_offset = offset - stripe_offset;
2713
2714         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2715                          BTRFS_BLOCK_GROUP_RAID10 |
2716                          BTRFS_BLOCK_GROUP_DUP)) {
2717                 /* we limit the length of each bio to what fits in a stripe */
2718                 *length = min_t(u64, em->len - offset,
2719                               map->stripe_len - stripe_offset);
2720         } else {
2721                 *length = em->len - offset;
2722         }
2723
2724         if (!multi_ret && !unplug_page)
2725                 goto out;
2726
2727         num_stripes = 1;
2728         stripe_index = 0;
2729         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2730                 if (unplug_page || (rw & (1 << BIO_RW)))
2731                         num_stripes = map->num_stripes;
2732                 else if (mirror_num)
2733                         stripe_index = mirror_num - 1;
2734                 else {
2735                         stripe_index = find_live_mirror(map, 0,
2736                                             map->num_stripes,
2737                                             current->pid % map->num_stripes);
2738                 }
2739
2740         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2741                 if (rw & (1 << BIO_RW))
2742                         num_stripes = map->num_stripes;
2743                 else if (mirror_num)
2744                         stripe_index = mirror_num - 1;
2745
2746         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2747                 int factor = map->num_stripes / map->sub_stripes;
2748
2749                 stripe_index = do_div(stripe_nr, factor);
2750                 stripe_index *= map->sub_stripes;
2751
2752                 if (unplug_page || (rw & (1 << BIO_RW)))
2753                         num_stripes = map->sub_stripes;
2754                 else if (mirror_num)
2755                         stripe_index += mirror_num - 1;
2756                 else {
2757                         stripe_index = find_live_mirror(map, stripe_index,
2758                                               map->sub_stripes, stripe_index +
2759                                               current->pid % map->sub_stripes);
2760                 }
2761         } else {
2762                 /*
2763                  * after this do_div call, stripe_nr is the number of stripes
2764                  * on this device we have to walk to find the data, and
2765                  * stripe_index is the number of our device in the stripe array
2766                  */
2767                 stripe_index = do_div(stripe_nr, map->num_stripes);
2768         }
2769         BUG_ON(stripe_index >= map->num_stripes);
2770
2771         for (i = 0; i < num_stripes; i++) {
2772                 if (unplug_page) {
2773                         struct btrfs_device *device;
2774                         struct backing_dev_info *bdi;
2775
2776                         device = map->stripes[stripe_index].dev;
2777                         if (device->bdev) {
2778                                 bdi = blk_get_backing_dev_info(device->bdev);
2779                                 if (bdi->unplug_io_fn)
2780                                         bdi->unplug_io_fn(bdi, unplug_page);
2781                         }
2782                 } else {
2783                         multi->stripes[i].physical =
2784                                 map->stripes[stripe_index].physical +
2785                                 stripe_offset + stripe_nr * map->stripe_len;
2786                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2787                 }
2788                 stripe_index++;
2789         }
2790         if (multi_ret) {
2791                 *multi_ret = multi;
2792                 multi->num_stripes = num_stripes;
2793                 multi->max_errors = max_errors;
2794         }
2795 out:
2796         free_extent_map(em);
2797         return 0;
2798 }
2799
2800 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2801                       u64 logical, u64 *length,
2802                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2803 {
2804         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2805                                  mirror_num, NULL);
2806 }
2807
2808 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2809                      u64 chunk_start, u64 physical, u64 devid,
2810                      u64 **logical, int *naddrs, int *stripe_len)
2811 {
2812         struct extent_map_tree *em_tree = &map_tree->map_tree;
2813         struct extent_map *em;
2814         struct map_lookup *map;
2815         u64 *buf;
2816         u64 bytenr;
2817         u64 length;
2818         u64 stripe_nr;
2819         int i, j, nr = 0;
2820
2821         read_lock(&em_tree->lock);
2822         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2823         read_unlock(&em_tree->lock);
2824
2825         BUG_ON(!em || em->start != chunk_start);
2826         map = (struct map_lookup *)em->bdev;
2827
2828         length = em->len;
2829         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2830                 do_div(length, map->num_stripes / map->sub_stripes);
2831         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2832                 do_div(length, map->num_stripes);
2833
2834         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2835         BUG_ON(!buf);
2836
2837         for (i = 0; i < map->num_stripes; i++) {
2838                 if (devid && map->stripes[i].dev->devid != devid)
2839                         continue;
2840                 if (map->stripes[i].physical > physical ||
2841                     map->stripes[i].physical + length <= physical)
2842                         continue;
2843
2844                 stripe_nr = physical - map->stripes[i].physical;
2845                 do_div(stripe_nr, map->stripe_len);
2846
2847                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2848                         stripe_nr = stripe_nr * map->num_stripes + i;
2849                         do_div(stripe_nr, map->sub_stripes);
2850                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2851                         stripe_nr = stripe_nr * map->num_stripes + i;
2852                 }
2853                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2854                 WARN_ON(nr >= map->num_stripes);
2855                 for (j = 0; j < nr; j++) {
2856                         if (buf[j] == bytenr)
2857                                 break;
2858                 }
2859                 if (j == nr) {
2860                         WARN_ON(nr >= map->num_stripes);
2861                         buf[nr++] = bytenr;
2862                 }
2863         }
2864
2865         *logical = buf;
2866         *naddrs = nr;
2867         *stripe_len = map->stripe_len;
2868
2869         free_extent_map(em);
2870         return 0;
2871 }
2872
2873 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2874                       u64 logical, struct page *page)
2875 {
2876         u64 length = PAGE_CACHE_SIZE;
2877         return __btrfs_map_block(map_tree, READ, logical, &length,
2878                                  NULL, 0, page);
2879 }
2880
2881 static void end_bio_multi_stripe(struct bio *bio, int err)
2882 {
2883         struct btrfs_multi_bio *multi = bio->bi_private;
2884         int is_orig_bio = 0;
2885
2886         if (err)
2887                 atomic_inc(&multi->error);
2888
2889         if (bio == multi->orig_bio)
2890                 is_orig_bio = 1;
2891
2892         if (atomic_dec_and_test(&multi->stripes_pending)) {
2893                 if (!is_orig_bio) {
2894                         bio_put(bio);
2895                         bio = multi->orig_bio;
2896                 }
2897                 bio->bi_private = multi->private;
2898                 bio->bi_end_io = multi->end_io;
2899                 /* only send an error to the higher layers if it is
2900                  * beyond the tolerance of the multi-bio
2901                  */
2902                 if (atomic_read(&multi->error) > multi->max_errors) {
2903                         err = -EIO;
2904                 } else if (err) {
2905                         /*
2906                          * this bio is actually up to date, we didn't
2907                          * go over the max number of errors
2908                          */
2909                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2910                         err = 0;
2911                 }
2912                 kfree(multi);
2913
2914                 bio_endio(bio, err);
2915         } else if (!is_orig_bio) {
2916                 bio_put(bio);
2917         }
2918 }
2919
2920 struct async_sched {
2921         struct bio *bio;
2922         int rw;
2923         struct btrfs_fs_info *info;
2924         struct btrfs_work work;
2925 };
2926
2927 /*
2928  * see run_scheduled_bios for a description of why bios are collected for
2929  * async submit.
2930  *
2931  * This will add one bio to the pending list for a device and make sure
2932  * the work struct is scheduled.
2933  */
2934 static noinline int schedule_bio(struct btrfs_root *root,
2935                                  struct btrfs_device *device,
2936                                  int rw, struct bio *bio)
2937 {
2938         int should_queue = 1;
2939         struct btrfs_pending_bios *pending_bios;
2940
2941         /* don't bother with additional async steps for reads, right now */
2942         if (!(rw & (1 << BIO_RW))) {
2943                 bio_get(bio);
2944                 submit_bio(rw, bio);
2945                 bio_put(bio);
2946                 return 0;
2947         }
2948
2949         /*
2950          * nr_async_bios allows us to reliably return congestion to the
2951          * higher layers.  Otherwise, the async bio makes it appear we have
2952          * made progress against dirty pages when we've really just put it
2953          * on a queue for later
2954          */
2955         atomic_inc(&root->fs_info->nr_async_bios);
2956         WARN_ON(bio->bi_next);
2957         bio->bi_next = NULL;
2958         bio->bi_rw |= rw;
2959
2960         spin_lock(&device->io_lock);
2961         if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2962                 pending_bios = &device->pending_sync_bios;
2963         else
2964                 pending_bios = &device->pending_bios;
2965
2966         if (pending_bios->tail)
2967                 pending_bios->tail->bi_next = bio;
2968
2969         pending_bios->tail = bio;
2970         if (!pending_bios->head)
2971                 pending_bios->head = bio;
2972         if (device->running_pending)
2973                 should_queue = 0;
2974
2975         spin_unlock(&device->io_lock);
2976
2977         if (should_queue)
2978                 btrfs_queue_worker(&root->fs_info->submit_workers,
2979                                    &device->work);
2980         return 0;
2981 }
2982
2983 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2984                   int mirror_num, int async_submit)
2985 {
2986         struct btrfs_mapping_tree *map_tree;
2987         struct btrfs_device *dev;
2988         struct bio *first_bio = bio;
2989         u64 logical = (u64)bio->bi_sector << 9;
2990         u64 length = 0;
2991         u64 map_length;
2992         struct btrfs_multi_bio *multi = NULL;
2993         int ret;
2994         int dev_nr = 0;
2995         int total_devs = 1;
2996
2997         length = bio->bi_size;
2998         map_tree = &root->fs_info->mapping_tree;
2999         map_length = length;
3000
3001         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3002                               mirror_num);
3003         BUG_ON(ret);
3004
3005         total_devs = multi->num_stripes;
3006         if (map_length < length) {
3007                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3008                        "len %llu\n", (unsigned long long)logical,
3009                        (unsigned long long)length,
3010                        (unsigned long long)map_length);
3011                 BUG();
3012         }
3013         multi->end_io = first_bio->bi_end_io;
3014         multi->private = first_bio->bi_private;
3015         multi->orig_bio = first_bio;
3016         atomic_set(&multi->stripes_pending, multi->num_stripes);
3017
3018         while (dev_nr < total_devs) {
3019                 if (total_devs > 1) {
3020                         if (dev_nr < total_devs - 1) {
3021                                 bio = bio_clone(first_bio, GFP_NOFS);
3022                                 BUG_ON(!bio);
3023                         } else {
3024                                 bio = first_bio;
3025                         }
3026                         bio->bi_private = multi;
3027                         bio->bi_end_io = end_bio_multi_stripe;
3028                 }
3029                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3030                 dev = multi->stripes[dev_nr].dev;
3031                 BUG_ON(rw == WRITE && !dev->writeable);
3032                 if (dev && dev->bdev) {
3033                         bio->bi_bdev = dev->bdev;
3034                         if (async_submit)
3035                                 schedule_bio(root, dev, rw, bio);
3036                         else
3037                                 submit_bio(rw, bio);
3038                 } else {
3039                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3040                         bio->bi_sector = logical >> 9;
3041                         bio_endio(bio, -EIO);
3042                 }
3043                 dev_nr++;
3044         }
3045         if (total_devs == 1)
3046                 kfree(multi);
3047         return 0;
3048 }
3049
3050 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3051                                        u8 *uuid, u8 *fsid)
3052 {
3053         struct btrfs_device *device;
3054         struct btrfs_fs_devices *cur_devices;
3055
3056         cur_devices = root->fs_info->fs_devices;
3057         while (cur_devices) {
3058                 if (!fsid ||
3059                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3060                         device = __find_device(&cur_devices->devices,
3061                                                devid, uuid);
3062                         if (device)
3063                                 return device;
3064                 }
3065                 cur_devices = cur_devices->seed;
3066         }
3067         return NULL;
3068 }
3069
3070 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3071                                             u64 devid, u8 *dev_uuid)
3072 {
3073         struct btrfs_device *device;
3074         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3075
3076         device = kzalloc(sizeof(*device), GFP_NOFS);
3077         if (!device)
3078                 return NULL;
3079         list_add(&device->dev_list,
3080                  &fs_devices->devices);
3081         device->barriers = 1;
3082         device->dev_root = root->fs_info->dev_root;
3083         device->devid = devid;
3084         device->work.func = pending_bios_fn;
3085         device->fs_devices = fs_devices;
3086         fs_devices->num_devices++;
3087         spin_lock_init(&device->io_lock);
3088         INIT_LIST_HEAD(&device->dev_alloc_list);
3089         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3090         return device;
3091 }
3092
3093 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3094                           struct extent_buffer *leaf,
3095                           struct btrfs_chunk *chunk)
3096 {
3097         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3098         struct map_lookup *map;
3099         struct extent_map *em;
3100         u64 logical;
3101         u64 length;
3102         u64 devid;
3103         u8 uuid[BTRFS_UUID_SIZE];
3104         int num_stripes;
3105         int ret;
3106         int i;
3107
3108         logical = key->offset;
3109         length = btrfs_chunk_length(leaf, chunk);
3110
3111         read_lock(&map_tree->map_tree.lock);
3112         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3113         read_unlock(&map_tree->map_tree.lock);
3114
3115         /* already mapped? */
3116         if (em && em->start <= logical && em->start + em->len > logical) {
3117                 free_extent_map(em);
3118                 return 0;
3119         } else if (em) {
3120                 free_extent_map(em);
3121         }
3122
3123         em = alloc_extent_map(GFP_NOFS);
3124         if (!em)
3125                 return -ENOMEM;
3126         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3127         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3128         if (!map) {
3129                 free_extent_map(em);
3130                 return -ENOMEM;
3131         }
3132
3133         em->bdev = (struct block_device *)map;
3134         em->start = logical;
3135         em->len = length;
3136         em->block_start = 0;
3137         em->block_len = em->len;
3138
3139         map->num_stripes = num_stripes;
3140         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3141         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3142         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3143         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3144         map->type = btrfs_chunk_type(leaf, chunk);
3145         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3146         for (i = 0; i < num_stripes; i++) {
3147                 map->stripes[i].physical =
3148                         btrfs_stripe_offset_nr(leaf, chunk, i);
3149                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3150                 read_extent_buffer(leaf, uuid, (unsigned long)
3151                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3152                                    BTRFS_UUID_SIZE);
3153                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3154                                                         NULL);
3155                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3156                         kfree(map);
3157                         free_extent_map(em);
3158                         return -EIO;
3159                 }
3160                 if (!map->stripes[i].dev) {
3161                         map->stripes[i].dev =
3162                                 add_missing_dev(root, devid, uuid);
3163                         if (!map->stripes[i].dev) {
3164                                 kfree(map);
3165                                 free_extent_map(em);
3166                                 return -EIO;
3167                         }
3168                 }
3169                 map->stripes[i].dev->in_fs_metadata = 1;
3170         }
3171
3172         write_lock(&map_tree->map_tree.lock);
3173         ret = add_extent_mapping(&map_tree->map_tree, em);
3174         write_unlock(&map_tree->map_tree.lock);
3175         BUG_ON(ret);
3176         free_extent_map(em);
3177
3178         return 0;
3179 }
3180
3181 static int fill_device_from_item(struct extent_buffer *leaf,
3182                                  struct btrfs_dev_item *dev_item,
3183                                  struct btrfs_device *device)
3184 {
3185         unsigned long ptr;
3186
3187         device->devid = btrfs_device_id(leaf, dev_item);
3188         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3189         device->total_bytes = device->disk_total_bytes;
3190         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3191         device->type = btrfs_device_type(leaf, dev_item);
3192         device->io_align = btrfs_device_io_align(leaf, dev_item);
3193         device->io_width = btrfs_device_io_width(leaf, dev_item);
3194         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3195
3196         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3197         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3198
3199         return 0;
3200 }
3201
3202 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3203 {
3204         struct btrfs_fs_devices *fs_devices;
3205         int ret;
3206
3207         mutex_lock(&uuid_mutex);
3208
3209         fs_devices = root->fs_info->fs_devices->seed;
3210         while (fs_devices) {
3211                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3212                         ret = 0;
3213                         goto out;
3214                 }
3215                 fs_devices = fs_devices->seed;
3216         }
3217
3218         fs_devices = find_fsid(fsid);
3219         if (!fs_devices) {
3220                 ret = -ENOENT;
3221                 goto out;
3222         }
3223
3224         fs_devices = clone_fs_devices(fs_devices);
3225         if (IS_ERR(fs_devices)) {
3226                 ret = PTR_ERR(fs_devices);
3227                 goto out;
3228         }
3229
3230         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3231                                    root->fs_info->bdev_holder);
3232         if (ret)
3233                 goto out;
3234
3235         if (!fs_devices->seeding) {
3236                 __btrfs_close_devices(fs_devices);
3237                 free_fs_devices(fs_devices);
3238                 ret = -EINVAL;
3239                 goto out;
3240         }
3241
3242         fs_devices->seed = root->fs_info->fs_devices->seed;
3243         root->fs_info->fs_devices->seed = fs_devices;
3244 out:
3245         mutex_unlock(&uuid_mutex);
3246         return ret;
3247 }
3248
3249 static int read_one_dev(struct btrfs_root *root,
3250                         struct extent_buffer *leaf,
3251                         struct btrfs_dev_item *dev_item)
3252 {
3253         struct btrfs_device *device;
3254         u64 devid;
3255         int ret;
3256         u8 fs_uuid[BTRFS_UUID_SIZE];
3257         u8 dev_uuid[BTRFS_UUID_SIZE];
3258
3259         devid = btrfs_device_id(leaf, dev_item);
3260         read_extent_buffer(leaf, dev_uuid,
3261                            (unsigned long)btrfs_device_uuid(dev_item),
3262                            BTRFS_UUID_SIZE);
3263         read_extent_buffer(leaf, fs_uuid,
3264                            (unsigned long)btrfs_device_fsid(dev_item),
3265                            BTRFS_UUID_SIZE);
3266
3267         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3268                 ret = open_seed_devices(root, fs_uuid);
3269                 if (ret && !btrfs_test_opt(root, DEGRADED))
3270                         return ret;
3271         }
3272
3273         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3274         if (!device || !device->bdev) {
3275                 if (!btrfs_test_opt(root, DEGRADED))
3276                         return -EIO;
3277
3278                 if (!device) {
3279                         printk(KERN_WARNING "warning devid %llu missing\n",
3280                                (unsigned long long)devid);
3281                         device = add_missing_dev(root, devid, dev_uuid);
3282                         if (!device)
3283                                 return -ENOMEM;
3284                 }
3285         }
3286
3287         if (device->fs_devices != root->fs_info->fs_devices) {
3288                 BUG_ON(device->writeable);
3289                 if (device->generation !=
3290                     btrfs_device_generation(leaf, dev_item))
3291                         return -EINVAL;
3292         }
3293
3294         fill_device_from_item(leaf, dev_item, device);
3295         device->dev_root = root->fs_info->dev_root;
3296         device->in_fs_metadata = 1;
3297         if (device->writeable)
3298                 device->fs_devices->total_rw_bytes += device->total_bytes;
3299         ret = 0;
3300         return ret;
3301 }
3302
3303 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3304 {
3305         struct btrfs_dev_item *dev_item;
3306
3307         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3308                                                      dev_item);
3309         return read_one_dev(root, buf, dev_item);
3310 }
3311
3312 int btrfs_read_sys_array(struct btrfs_root *root)
3313 {
3314         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3315         struct extent_buffer *sb;
3316         struct btrfs_disk_key *disk_key;
3317         struct btrfs_chunk *chunk;
3318         u8 *ptr;
3319         unsigned long sb_ptr;
3320         int ret = 0;
3321         u32 num_stripes;
3322         u32 array_size;
3323         u32 len = 0;
3324         u32 cur;
3325         struct btrfs_key key;
3326
3327         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3328                                           BTRFS_SUPER_INFO_SIZE);
3329         if (!sb)
3330                 return -ENOMEM;
3331         btrfs_set_buffer_uptodate(sb);
3332         btrfs_set_buffer_lockdep_class(sb, 0);
3333
3334         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3335         array_size = btrfs_super_sys_array_size(super_copy);
3336
3337         ptr = super_copy->sys_chunk_array;
3338         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3339         cur = 0;
3340
3341         while (cur < array_size) {
3342                 disk_key = (struct btrfs_disk_key *)ptr;
3343                 btrfs_disk_key_to_cpu(&key, disk_key);
3344
3345                 len = sizeof(*disk_key); ptr += len;
3346                 sb_ptr += len;
3347                 cur += len;
3348
3349                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3350                         chunk = (struct btrfs_chunk *)sb_ptr;
3351                         ret = read_one_chunk(root, &key, sb, chunk);
3352                         if (ret)
3353                                 break;
3354                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3355                         len = btrfs_chunk_item_size(num_stripes);
3356                 } else {
3357                         ret = -EIO;
3358                         break;
3359                 }
3360                 ptr += len;
3361                 sb_ptr += len;
3362                 cur += len;
3363         }
3364         free_extent_buffer(sb);
3365         return ret;
3366 }
3367
3368 int btrfs_read_chunk_tree(struct btrfs_root *root)
3369 {
3370         struct btrfs_path *path;
3371         struct extent_buffer *leaf;
3372         struct btrfs_key key;
3373         struct btrfs_key found_key;
3374         int ret;
3375         int slot;
3376
3377         root = root->fs_info->chunk_root;
3378
3379         path = btrfs_alloc_path();
3380         if (!path)
3381                 return -ENOMEM;
3382
3383         /* first we search for all of the device items, and then we
3384          * read in all of the chunk items.  This way we can create chunk
3385          * mappings that reference all of the devices that are afound
3386          */
3387         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3388         key.offset = 0;
3389         key.type = 0;
3390 again:
3391         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3392         while (1) {
3393                 leaf = path->nodes[0];
3394                 slot = path->slots[0];
3395                 if (slot >= btrfs_header_nritems(leaf)) {
3396                         ret = btrfs_next_leaf(root, path);
3397                         if (ret == 0)
3398                                 continue;
3399                         if (ret < 0)
3400                                 goto error;
3401                         break;
3402                 }
3403                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3404                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3405                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3406                                 break;
3407                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3408                                 struct btrfs_dev_item *dev_item;
3409                                 dev_item = btrfs_item_ptr(leaf, slot,
3410                                                   struct btrfs_dev_item);
3411                                 ret = read_one_dev(root, leaf, dev_item);
3412                                 if (ret)
3413                                         goto error;
3414                         }
3415                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3416                         struct btrfs_chunk *chunk;
3417                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3418                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3419                         if (ret)
3420                                 goto error;
3421                 }
3422                 path->slots[0]++;
3423         }
3424         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3425                 key.objectid = 0;
3426                 btrfs_release_path(root, path);
3427                 goto again;
3428         }
3429         ret = 0;
3430 error:
3431         btrfs_free_path(path);
3432         return ret;
3433 }