block: make blkdev_get/put() handle exclusive access
[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/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <asm/div64.h>
26 #include "compat.h"
27 #include "ctree.h"
28 #include "extent_map.h"
29 #include "disk-io.h"
30 #include "transaction.h"
31 #include "print-tree.h"
32 #include "volumes.h"
33 #include "async-thread.h"
34
35 struct map_lookup {
36         u64 type;
37         int io_align;
38         int io_width;
39         int stripe_len;
40         int sector_size;
41         int num_stripes;
42         int sub_stripes;
43         struct btrfs_bio_stripe stripes[];
44 };
45
46 static int init_first_rw_device(struct btrfs_trans_handle *trans,
47                                 struct btrfs_root *root,
48                                 struct btrfs_device *device);
49 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50
51 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
52                             (sizeof(struct btrfs_bio_stripe) * (n)))
53
54 static DEFINE_MUTEX(uuid_mutex);
55 static LIST_HEAD(fs_uuids);
56
57 void btrfs_lock_volumes(void)
58 {
59         mutex_lock(&uuid_mutex);
60 }
61
62 void btrfs_unlock_volumes(void)
63 {
64         mutex_unlock(&uuid_mutex);
65 }
66
67 static void lock_chunks(struct btrfs_root *root)
68 {
69         mutex_lock(&root->fs_info->chunk_mutex);
70 }
71
72 static void unlock_chunks(struct btrfs_root *root)
73 {
74         mutex_unlock(&root->fs_info->chunk_mutex);
75 }
76
77 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78 {
79         struct btrfs_device *device;
80         WARN_ON(fs_devices->opened);
81         while (!list_empty(&fs_devices->devices)) {
82                 device = list_entry(fs_devices->devices.next,
83                                     struct btrfs_device, dev_list);
84                 list_del(&device->dev_list);
85                 kfree(device->name);
86                 kfree(device);
87         }
88         kfree(fs_devices);
89 }
90
91 int btrfs_cleanup_fs_uuids(void)
92 {
93         struct btrfs_fs_devices *fs_devices;
94
95         while (!list_empty(&fs_uuids)) {
96                 fs_devices = list_entry(fs_uuids.next,
97                                         struct btrfs_fs_devices, list);
98                 list_del(&fs_devices->list);
99                 free_fs_devices(fs_devices);
100         }
101         return 0;
102 }
103
104 static noinline struct btrfs_device *__find_device(struct list_head *head,
105                                                    u64 devid, u8 *uuid)
106 {
107         struct btrfs_device *dev;
108
109         list_for_each_entry(dev, head, dev_list) {
110                 if (dev->devid == devid &&
111                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
112                         return dev;
113                 }
114         }
115         return NULL;
116 }
117
118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 {
120         struct btrfs_fs_devices *fs_devices;
121
122         list_for_each_entry(fs_devices, &fs_uuids, list) {
123                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
124                         return fs_devices;
125         }
126         return NULL;
127 }
128
129 static void requeue_list(struct btrfs_pending_bios *pending_bios,
130                         struct bio *head, struct bio *tail)
131 {
132
133         struct bio *old_head;
134
135         old_head = pending_bios->head;
136         pending_bios->head = head;
137         if (pending_bios->tail)
138                 tail->bi_next = old_head;
139         else
140                 pending_bios->tail = tail;
141 }
142
143 /*
144  * we try to collect pending bios for a device so we don't get a large
145  * number of procs sending bios down to the same device.  This greatly
146  * improves the schedulers ability to collect and merge the bios.
147  *
148  * But, it also turns into a long list of bios to process and that is sure
149  * to eventually make the worker thread block.  The solution here is to
150  * make some progress and then put this work struct back at the end of
151  * the list if the block device is congested.  This way, multiple devices
152  * can make progress from a single worker thread.
153  */
154 static noinline int run_scheduled_bios(struct btrfs_device *device)
155 {
156         struct bio *pending;
157         struct backing_dev_info *bdi;
158         struct btrfs_fs_info *fs_info;
159         struct btrfs_pending_bios *pending_bios;
160         struct bio *tail;
161         struct bio *cur;
162         int again = 0;
163         unsigned long num_run;
164         unsigned long num_sync_run;
165         unsigned long batch_run = 0;
166         unsigned long limit;
167         unsigned long last_waited = 0;
168         int force_reg = 0;
169
170         bdi = blk_get_backing_dev_info(device->bdev);
171         fs_info = device->dev_root->fs_info;
172         limit = btrfs_async_submit_limit(fs_info);
173         limit = limit * 2 / 3;
174
175         /* we want to make sure that every time we switch from the sync
176          * list to the normal list, we unplug
177          */
178         num_sync_run = 0;
179
180 loop:
181         spin_lock(&device->io_lock);
182
183 loop_lock:
184         num_run = 0;
185
186         /* take all the bios off the list at once and process them
187          * later on (without the lock held).  But, remember the
188          * tail and other pointers so the bios can be properly reinserted
189          * into the list if we hit congestion
190          */
191         if (!force_reg && device->pending_sync_bios.head) {
192                 pending_bios = &device->pending_sync_bios;
193                 force_reg = 1;
194         } else {
195                 pending_bios = &device->pending_bios;
196                 force_reg = 0;
197         }
198
199         pending = pending_bios->head;
200         tail = pending_bios->tail;
201         WARN_ON(pending && !tail);
202
203         /*
204          * if pending was null this time around, no bios need processing
205          * at all and we can stop.  Otherwise it'll loop back up again
206          * and do an additional check so no bios are missed.
207          *
208          * device->running_pending is used to synchronize with the
209          * schedule_bio code.
210          */
211         if (device->pending_sync_bios.head == NULL &&
212             device->pending_bios.head == NULL) {
213                 again = 0;
214                 device->running_pending = 0;
215         } else {
216                 again = 1;
217                 device->running_pending = 1;
218         }
219
220         pending_bios->head = NULL;
221         pending_bios->tail = NULL;
222
223         spin_unlock(&device->io_lock);
224
225         /*
226          * if we're doing the regular priority list, make sure we unplug
227          * for any high prio bios we've sent down
228          */
229         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
230                 num_sync_run = 0;
231                 blk_run_backing_dev(bdi, NULL);
232         }
233
234         while (pending) {
235
236                 rmb();
237                 /* we want to work on both lists, but do more bios on the
238                  * sync list than the regular list
239                  */
240                 if ((num_run > 32 &&
241                     pending_bios != &device->pending_sync_bios &&
242                     device->pending_sync_bios.head) ||
243                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
244                     device->pending_bios.head)) {
245                         spin_lock(&device->io_lock);
246                         requeue_list(pending_bios, pending, tail);
247                         goto loop_lock;
248                 }
249
250                 cur = pending;
251                 pending = pending->bi_next;
252                 cur->bi_next = NULL;
253                 atomic_dec(&fs_info->nr_async_bios);
254
255                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
256                     waitqueue_active(&fs_info->async_submit_wait))
257                         wake_up(&fs_info->async_submit_wait);
258
259                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
260
261                 if (cur->bi_rw & REQ_SYNC)
262                         num_sync_run++;
263
264                 submit_bio(cur->bi_rw, cur);
265                 num_run++;
266                 batch_run++;
267                 if (need_resched()) {
268                         if (num_sync_run) {
269                                 blk_run_backing_dev(bdi, NULL);
270                                 num_sync_run = 0;
271                         }
272                         cond_resched();
273                 }
274
275                 /*
276                  * we made progress, there is more work to do and the bdi
277                  * is now congested.  Back off and let other work structs
278                  * run instead
279                  */
280                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
281                     fs_info->fs_devices->open_devices > 1) {
282                         struct io_context *ioc;
283
284                         ioc = current->io_context;
285
286                         /*
287                          * the main goal here is that we don't want to
288                          * block if we're going to be able to submit
289                          * more requests without blocking.
290                          *
291                          * This code does two great things, it pokes into
292                          * the elevator code from a filesystem _and_
293                          * it makes assumptions about how batching works.
294                          */
295                         if (ioc && ioc->nr_batch_requests > 0 &&
296                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
297                             (last_waited == 0 ||
298                              ioc->last_waited == last_waited)) {
299                                 /*
300                                  * we want to go through our batch of
301                                  * requests and stop.  So, we copy out
302                                  * the ioc->last_waited time and test
303                                  * against it before looping
304                                  */
305                                 last_waited = ioc->last_waited;
306                                 if (need_resched()) {
307                                         if (num_sync_run) {
308                                                 blk_run_backing_dev(bdi, NULL);
309                                                 num_sync_run = 0;
310                                         }
311                                         cond_resched();
312                                 }
313                                 continue;
314                         }
315                         spin_lock(&device->io_lock);
316                         requeue_list(pending_bios, pending, tail);
317                         device->running_pending = 1;
318
319                         spin_unlock(&device->io_lock);
320                         btrfs_requeue_work(&device->work);
321                         goto done;
322                 }
323         }
324
325         if (num_sync_run) {
326                 num_sync_run = 0;
327                 blk_run_backing_dev(bdi, NULL);
328         }
329         /*
330          * IO has already been through a long path to get here.  Checksumming,
331          * async helper threads, perhaps compression.  We've done a pretty
332          * good job of collecting a batch of IO and should just unplug
333          * the device right away.
334          *
335          * This will help anyone who is waiting on the IO, they might have
336          * already unplugged, but managed to do so before the bio they
337          * cared about found its way down here.
338          */
339         blk_run_backing_dev(bdi, NULL);
340
341         cond_resched();
342         if (again)
343                 goto loop;
344
345         spin_lock(&device->io_lock);
346         if (device->pending_bios.head || device->pending_sync_bios.head)
347                 goto loop_lock;
348         spin_unlock(&device->io_lock);
349
350 done:
351         return 0;
352 }
353
354 static void pending_bios_fn(struct btrfs_work *work)
355 {
356         struct btrfs_device *device;
357
358         device = container_of(work, struct btrfs_device, work);
359         run_scheduled_bios(device);
360 }
361
362 static noinline int device_list_add(const char *path,
363                            struct btrfs_super_block *disk_super,
364                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
365 {
366         struct btrfs_device *device;
367         struct btrfs_fs_devices *fs_devices;
368         u64 found_transid = btrfs_super_generation(disk_super);
369         char *name;
370
371         fs_devices = find_fsid(disk_super->fsid);
372         if (!fs_devices) {
373                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
374                 if (!fs_devices)
375                         return -ENOMEM;
376                 INIT_LIST_HEAD(&fs_devices->devices);
377                 INIT_LIST_HEAD(&fs_devices->alloc_list);
378                 list_add(&fs_devices->list, &fs_uuids);
379                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
380                 fs_devices->latest_devid = devid;
381                 fs_devices->latest_trans = found_transid;
382                 mutex_init(&fs_devices->device_list_mutex);
383                 device = NULL;
384         } else {
385                 device = __find_device(&fs_devices->devices, devid,
386                                        disk_super->dev_item.uuid);
387         }
388         if (!device) {
389                 if (fs_devices->opened)
390                         return -EBUSY;
391
392                 device = kzalloc(sizeof(*device), GFP_NOFS);
393                 if (!device) {
394                         /* we can safely leave the fs_devices entry around */
395                         return -ENOMEM;
396                 }
397                 device->devid = devid;
398                 device->work.func = pending_bios_fn;
399                 memcpy(device->uuid, disk_super->dev_item.uuid,
400                        BTRFS_UUID_SIZE);
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                 spin_lock_init(&device->io_lock);
465                 INIT_LIST_HEAD(&device->dev_list);
466                 INIT_LIST_HEAD(&device->dev_alloc_list);
467
468                 list_add(&device->dev_list, &fs_devices->devices);
469                 device->fs_devices = fs_devices;
470                 fs_devices->num_devices++;
471         }
472         mutex_unlock(&orig->device_list_mutex);
473         return fs_devices;
474 error:
475         mutex_unlock(&orig->device_list_mutex);
476         free_fs_devices(fs_devices);
477         return ERR_PTR(-ENOMEM);
478 }
479
480 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
481 {
482         struct btrfs_device *device, *next;
483
484         mutex_lock(&uuid_mutex);
485 again:
486         mutex_lock(&fs_devices->device_list_mutex);
487         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
488                 if (device->in_fs_metadata)
489                         continue;
490
491                 if (device->bdev) {
492                         blkdev_put(device->bdev, device->mode | FMODE_EXCL);
493                         device->bdev = NULL;
494                         fs_devices->open_devices--;
495                 }
496                 if (device->writeable) {
497                         list_del_init(&device->dev_alloc_list);
498                         device->writeable = 0;
499                         fs_devices->rw_devices--;
500                 }
501                 list_del_init(&device->dev_list);
502                 fs_devices->num_devices--;
503                 kfree(device->name);
504                 kfree(device);
505         }
506         mutex_unlock(&fs_devices->device_list_mutex);
507
508         if (fs_devices->seed) {
509                 fs_devices = fs_devices->seed;
510                 goto again;
511         }
512
513         mutex_unlock(&uuid_mutex);
514         return 0;
515 }
516
517 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
518 {
519         struct btrfs_device *device;
520
521         if (--fs_devices->opened > 0)
522                 return 0;
523
524         list_for_each_entry(device, &fs_devices->devices, dev_list) {
525                 if (device->bdev) {
526                         blkdev_put(device->bdev, device->mode | FMODE_EXCL);
527                         fs_devices->open_devices--;
528                 }
529                 if (device->writeable) {
530                         list_del_init(&device->dev_alloc_list);
531                         fs_devices->rw_devices--;
532                 }
533
534                 device->bdev = NULL;
535                 device->writeable = 0;
536                 device->in_fs_metadata = 0;
537         }
538         WARN_ON(fs_devices->open_devices);
539         WARN_ON(fs_devices->rw_devices);
540         fs_devices->opened = 0;
541         fs_devices->seeding = 0;
542
543         return 0;
544 }
545
546 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
547 {
548         struct btrfs_fs_devices *seed_devices = NULL;
549         int ret;
550
551         mutex_lock(&uuid_mutex);
552         ret = __btrfs_close_devices(fs_devices);
553         if (!fs_devices->opened) {
554                 seed_devices = fs_devices->seed;
555                 fs_devices->seed = NULL;
556         }
557         mutex_unlock(&uuid_mutex);
558
559         while (seed_devices) {
560                 fs_devices = seed_devices;
561                 seed_devices = fs_devices->seed;
562                 __btrfs_close_devices(fs_devices);
563                 free_fs_devices(fs_devices);
564         }
565         return ret;
566 }
567
568 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
569                                 fmode_t flags, void *holder)
570 {
571         struct block_device *bdev;
572         struct list_head *head = &fs_devices->devices;
573         struct btrfs_device *device;
574         struct block_device *latest_bdev = NULL;
575         struct buffer_head *bh;
576         struct btrfs_super_block *disk_super;
577         u64 latest_devid = 0;
578         u64 latest_transid = 0;
579         u64 devid;
580         int seeding = 1;
581         int ret = 0;
582
583         list_for_each_entry(device, head, dev_list) {
584                 if (device->bdev)
585                         continue;
586                 if (!device->name)
587                         continue;
588
589                 bdev = open_bdev_exclusive(device->name, flags, holder);
590                 if (IS_ERR(bdev)) {
591                         printk(KERN_INFO "open %s failed\n", device->name);
592                         goto error;
593                 }
594                 set_blocksize(bdev, 4096);
595
596                 bh = btrfs_read_dev_super(bdev);
597                 if (!bh)
598                         goto error_close;
599
600                 disk_super = (struct btrfs_super_block *)bh->b_data;
601                 devid = btrfs_stack_device_id(&disk_super->dev_item);
602                 if (devid != device->devid)
603                         goto error_brelse;
604
605                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
606                            BTRFS_UUID_SIZE))
607                         goto error_brelse;
608
609                 device->generation = btrfs_super_generation(disk_super);
610                 if (!latest_transid || device->generation > latest_transid) {
611                         latest_devid = devid;
612                         latest_transid = device->generation;
613                         latest_bdev = bdev;
614                 }
615
616                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
617                         device->writeable = 0;
618                 } else {
619                         device->writeable = !bdev_read_only(bdev);
620                         seeding = 0;
621                 }
622
623                 device->bdev = bdev;
624                 device->in_fs_metadata = 0;
625                 device->mode = flags;
626
627                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
628                         fs_devices->rotating = 1;
629
630                 fs_devices->open_devices++;
631                 if (device->writeable) {
632                         fs_devices->rw_devices++;
633                         list_add(&device->dev_alloc_list,
634                                  &fs_devices->alloc_list);
635                 }
636                 continue;
637
638 error_brelse:
639                 brelse(bh);
640 error_close:
641                 blkdev_put(bdev, flags | FMODE_EXCL);
642 error:
643                 continue;
644         }
645         if (fs_devices->open_devices == 0) {
646                 ret = -EIO;
647                 goto out;
648         }
649         fs_devices->seeding = seeding;
650         fs_devices->opened = 1;
651         fs_devices->latest_bdev = latest_bdev;
652         fs_devices->latest_devid = latest_devid;
653         fs_devices->latest_trans = latest_transid;
654         fs_devices->total_rw_bytes = 0;
655 out:
656         return ret;
657 }
658
659 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
660                        fmode_t flags, void *holder)
661 {
662         int ret;
663
664         mutex_lock(&uuid_mutex);
665         if (fs_devices->opened) {
666                 fs_devices->opened++;
667                 ret = 0;
668         } else {
669                 ret = __btrfs_open_devices(fs_devices, flags, holder);
670         }
671         mutex_unlock(&uuid_mutex);
672         return ret;
673 }
674
675 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
676                           struct btrfs_fs_devices **fs_devices_ret)
677 {
678         struct btrfs_super_block *disk_super;
679         struct block_device *bdev;
680         struct buffer_head *bh;
681         int ret;
682         u64 devid;
683         u64 transid;
684
685         mutex_lock(&uuid_mutex);
686
687         bdev = open_bdev_exclusive(path, flags, holder);
688
689         if (IS_ERR(bdev)) {
690                 ret = PTR_ERR(bdev);
691                 goto error;
692         }
693
694         ret = set_blocksize(bdev, 4096);
695         if (ret)
696                 goto error_close;
697         bh = btrfs_read_dev_super(bdev);
698         if (!bh) {
699                 ret = -EIO;
700                 goto error_close;
701         }
702         disk_super = (struct btrfs_super_block *)bh->b_data;
703         devid = btrfs_stack_device_id(&disk_super->dev_item);
704         transid = btrfs_super_generation(disk_super);
705         if (disk_super->label[0])
706                 printk(KERN_INFO "device label %s ", disk_super->label);
707         else {
708                 /* FIXME, make a readl uuid parser */
709                 printk(KERN_INFO "device fsid %llx-%llx ",
710                        *(unsigned long long *)disk_super->fsid,
711                        *(unsigned long long *)(disk_super->fsid + 8));
712         }
713         printk(KERN_CONT "devid %llu transid %llu %s\n",
714                (unsigned long long)devid, (unsigned long long)transid, path);
715         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
716
717         brelse(bh);
718 error_close:
719         blkdev_put(bdev, flags | FMODE_EXCL);
720 error:
721         mutex_unlock(&uuid_mutex);
722         return ret;
723 }
724
725 /*
726  * this uses a pretty simple search, the expectation is that it is
727  * called very infrequently and that a given device has a small number
728  * of extents
729  */
730 int find_free_dev_extent(struct btrfs_trans_handle *trans,
731                          struct btrfs_device *device, u64 num_bytes,
732                          u64 *start, u64 *max_avail)
733 {
734         struct btrfs_key key;
735         struct btrfs_root *root = device->dev_root;
736         struct btrfs_dev_extent *dev_extent = NULL;
737         struct btrfs_path *path;
738         u64 hole_size = 0;
739         u64 last_byte = 0;
740         u64 search_start = 0;
741         u64 search_end = device->total_bytes;
742         int ret;
743         int slot = 0;
744         int start_found;
745         struct extent_buffer *l;
746
747         path = btrfs_alloc_path();
748         if (!path)
749                 return -ENOMEM;
750         path->reada = 2;
751         start_found = 0;
752
753         /* FIXME use last free of some kind */
754
755         /* we don't want to overwrite the superblock on the drive,
756          * so we make sure to start at an offset of at least 1MB
757          */
758         search_start = max((u64)1024 * 1024, search_start);
759
760         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
761                 search_start = max(root->fs_info->alloc_start, search_start);
762
763         key.objectid = device->devid;
764         key.offset = search_start;
765         key.type = BTRFS_DEV_EXTENT_KEY;
766         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
767         if (ret < 0)
768                 goto error;
769         if (ret > 0) {
770                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
771                 if (ret < 0)
772                         goto error;
773                 if (ret > 0)
774                         start_found = 1;
775         }
776         l = path->nodes[0];
777         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
778         while (1) {
779                 l = path->nodes[0];
780                 slot = path->slots[0];
781                 if (slot >= btrfs_header_nritems(l)) {
782                         ret = btrfs_next_leaf(root, path);
783                         if (ret == 0)
784                                 continue;
785                         if (ret < 0)
786                                 goto error;
787 no_more_items:
788                         if (!start_found) {
789                                 if (search_start >= search_end) {
790                                         ret = -ENOSPC;
791                                         goto error;
792                                 }
793                                 *start = search_start;
794                                 start_found = 1;
795                                 goto check_pending;
796                         }
797                         *start = last_byte > search_start ?
798                                 last_byte : search_start;
799                         if (search_end <= *start) {
800                                 ret = -ENOSPC;
801                                 goto error;
802                         }
803                         goto check_pending;
804                 }
805                 btrfs_item_key_to_cpu(l, &key, slot);
806
807                 if (key.objectid < device->devid)
808                         goto next;
809
810                 if (key.objectid > device->devid)
811                         goto no_more_items;
812
813                 if (key.offset >= search_start && key.offset > last_byte &&
814                     start_found) {
815                         if (last_byte < search_start)
816                                 last_byte = search_start;
817                         hole_size = key.offset - last_byte;
818
819                         if (hole_size > *max_avail)
820                                 *max_avail = hole_size;
821
822                         if (key.offset > last_byte &&
823                             hole_size >= num_bytes) {
824                                 *start = last_byte;
825                                 goto check_pending;
826                         }
827                 }
828                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
829                         goto next;
830
831                 start_found = 1;
832                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
833                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
834 next:
835                 path->slots[0]++;
836                 cond_resched();
837         }
838 check_pending:
839         /* we have to make sure we didn't find an extent that has already
840          * been allocated by the map tree or the original allocation
841          */
842         BUG_ON(*start < search_start);
843
844         if (*start + num_bytes > search_end) {
845                 ret = -ENOSPC;
846                 goto error;
847         }
848         /* check for pending inserts here */
849         ret = 0;
850
851 error:
852         btrfs_free_path(path);
853         return ret;
854 }
855
856 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
857                           struct btrfs_device *device,
858                           u64 start)
859 {
860         int ret;
861         struct btrfs_path *path;
862         struct btrfs_root *root = device->dev_root;
863         struct btrfs_key key;
864         struct btrfs_key found_key;
865         struct extent_buffer *leaf = NULL;
866         struct btrfs_dev_extent *extent = NULL;
867
868         path = btrfs_alloc_path();
869         if (!path)
870                 return -ENOMEM;
871
872         key.objectid = device->devid;
873         key.offset = start;
874         key.type = BTRFS_DEV_EXTENT_KEY;
875
876         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
877         if (ret > 0) {
878                 ret = btrfs_previous_item(root, path, key.objectid,
879                                           BTRFS_DEV_EXTENT_KEY);
880                 BUG_ON(ret);
881                 leaf = path->nodes[0];
882                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
883                 extent = btrfs_item_ptr(leaf, path->slots[0],
884                                         struct btrfs_dev_extent);
885                 BUG_ON(found_key.offset > start || found_key.offset +
886                        btrfs_dev_extent_length(leaf, extent) < start);
887                 ret = 0;
888         } else if (ret == 0) {
889                 leaf = path->nodes[0];
890                 extent = btrfs_item_ptr(leaf, path->slots[0],
891                                         struct btrfs_dev_extent);
892         }
893         BUG_ON(ret);
894
895         if (device->bytes_used > 0)
896                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
897         ret = btrfs_del_item(trans, root, path);
898         BUG_ON(ret);
899
900         btrfs_free_path(path);
901         return ret;
902 }
903
904 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
905                            struct btrfs_device *device,
906                            u64 chunk_tree, u64 chunk_objectid,
907                            u64 chunk_offset, u64 start, u64 num_bytes)
908 {
909         int ret;
910         struct btrfs_path *path;
911         struct btrfs_root *root = device->dev_root;
912         struct btrfs_dev_extent *extent;
913         struct extent_buffer *leaf;
914         struct btrfs_key key;
915
916         WARN_ON(!device->in_fs_metadata);
917         path = btrfs_alloc_path();
918         if (!path)
919                 return -ENOMEM;
920
921         key.objectid = device->devid;
922         key.offset = start;
923         key.type = BTRFS_DEV_EXTENT_KEY;
924         ret = btrfs_insert_empty_item(trans, root, path, &key,
925                                       sizeof(*extent));
926         BUG_ON(ret);
927
928         leaf = path->nodes[0];
929         extent = btrfs_item_ptr(leaf, path->slots[0],
930                                 struct btrfs_dev_extent);
931         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
932         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
933         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
934
935         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
936                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
937                     BTRFS_UUID_SIZE);
938
939         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
940         btrfs_mark_buffer_dirty(leaf);
941         btrfs_free_path(path);
942         return ret;
943 }
944
945 static noinline int find_next_chunk(struct btrfs_root *root,
946                                     u64 objectid, u64 *offset)
947 {
948         struct btrfs_path *path;
949         int ret;
950         struct btrfs_key key;
951         struct btrfs_chunk *chunk;
952         struct btrfs_key found_key;
953
954         path = btrfs_alloc_path();
955         BUG_ON(!path);
956
957         key.objectid = objectid;
958         key.offset = (u64)-1;
959         key.type = BTRFS_CHUNK_ITEM_KEY;
960
961         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
962         if (ret < 0)
963                 goto error;
964
965         BUG_ON(ret == 0);
966
967         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
968         if (ret) {
969                 *offset = 0;
970         } else {
971                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
972                                       path->slots[0]);
973                 if (found_key.objectid != objectid)
974                         *offset = 0;
975                 else {
976                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
977                                                struct btrfs_chunk);
978                         *offset = found_key.offset +
979                                 btrfs_chunk_length(path->nodes[0], chunk);
980                 }
981         }
982         ret = 0;
983 error:
984         btrfs_free_path(path);
985         return ret;
986 }
987
988 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
989 {
990         int ret;
991         struct btrfs_key key;
992         struct btrfs_key found_key;
993         struct btrfs_path *path;
994
995         root = root->fs_info->chunk_root;
996
997         path = btrfs_alloc_path();
998         if (!path)
999                 return -ENOMEM;
1000
1001         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1002         key.type = BTRFS_DEV_ITEM_KEY;
1003         key.offset = (u64)-1;
1004
1005         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1006         if (ret < 0)
1007                 goto error;
1008
1009         BUG_ON(ret == 0);
1010
1011         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1012                                   BTRFS_DEV_ITEM_KEY);
1013         if (ret) {
1014                 *objectid = 1;
1015         } else {
1016                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1017                                       path->slots[0]);
1018                 *objectid = found_key.offset + 1;
1019         }
1020         ret = 0;
1021 error:
1022         btrfs_free_path(path);
1023         return ret;
1024 }
1025
1026 /*
1027  * the device information is stored in the chunk root
1028  * the btrfs_device struct should be fully filled in
1029  */
1030 int btrfs_add_device(struct btrfs_trans_handle *trans,
1031                      struct btrfs_root *root,
1032                      struct btrfs_device *device)
1033 {
1034         int ret;
1035         struct btrfs_path *path;
1036         struct btrfs_dev_item *dev_item;
1037         struct extent_buffer *leaf;
1038         struct btrfs_key key;
1039         unsigned long ptr;
1040
1041         root = root->fs_info->chunk_root;
1042
1043         path = btrfs_alloc_path();
1044         if (!path)
1045                 return -ENOMEM;
1046
1047         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1048         key.type = BTRFS_DEV_ITEM_KEY;
1049         key.offset = device->devid;
1050
1051         ret = btrfs_insert_empty_item(trans, root, path, &key,
1052                                       sizeof(*dev_item));
1053         if (ret)
1054                 goto out;
1055
1056         leaf = path->nodes[0];
1057         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1058
1059         btrfs_set_device_id(leaf, dev_item, device->devid);
1060         btrfs_set_device_generation(leaf, dev_item, 0);
1061         btrfs_set_device_type(leaf, dev_item, device->type);
1062         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1063         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1064         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1065         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1066         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1067         btrfs_set_device_group(leaf, dev_item, 0);
1068         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1069         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1070         btrfs_set_device_start_offset(leaf, dev_item, 0);
1071
1072         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1073         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1074         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1075         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1076         btrfs_mark_buffer_dirty(leaf);
1077
1078         ret = 0;
1079 out:
1080         btrfs_free_path(path);
1081         return ret;
1082 }
1083
1084 static int btrfs_rm_dev_item(struct btrfs_root *root,
1085                              struct btrfs_device *device)
1086 {
1087         int ret;
1088         struct btrfs_path *path;
1089         struct btrfs_key key;
1090         struct btrfs_trans_handle *trans;
1091
1092         root = root->fs_info->chunk_root;
1093
1094         path = btrfs_alloc_path();
1095         if (!path)
1096                 return -ENOMEM;
1097
1098         trans = btrfs_start_transaction(root, 0);
1099         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1100         key.type = BTRFS_DEV_ITEM_KEY;
1101         key.offset = device->devid;
1102         lock_chunks(root);
1103
1104         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105         if (ret < 0)
1106                 goto out;
1107
1108         if (ret > 0) {
1109                 ret = -ENOENT;
1110                 goto out;
1111         }
1112
1113         ret = btrfs_del_item(trans, root, path);
1114         if (ret)
1115                 goto out;
1116 out:
1117         btrfs_free_path(path);
1118         unlock_chunks(root);
1119         btrfs_commit_transaction(trans, root);
1120         return ret;
1121 }
1122
1123 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1124 {
1125         struct btrfs_device *device;
1126         struct btrfs_device *next_device;
1127         struct block_device *bdev;
1128         struct buffer_head *bh = NULL;
1129         struct btrfs_super_block *disk_super;
1130         u64 all_avail;
1131         u64 devid;
1132         u64 num_devices;
1133         u8 *dev_uuid;
1134         int ret = 0;
1135
1136         mutex_lock(&uuid_mutex);
1137         mutex_lock(&root->fs_info->volume_mutex);
1138
1139         all_avail = root->fs_info->avail_data_alloc_bits |
1140                 root->fs_info->avail_system_alloc_bits |
1141                 root->fs_info->avail_metadata_alloc_bits;
1142
1143         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1144             root->fs_info->fs_devices->num_devices <= 4) {
1145                 printk(KERN_ERR "btrfs: unable to go below four devices "
1146                        "on raid10\n");
1147                 ret = -EINVAL;
1148                 goto out;
1149         }
1150
1151         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1152             root->fs_info->fs_devices->num_devices <= 2) {
1153                 printk(KERN_ERR "btrfs: unable to go below two "
1154                        "devices on raid1\n");
1155                 ret = -EINVAL;
1156                 goto out;
1157         }
1158
1159         if (strcmp(device_path, "missing") == 0) {
1160                 struct list_head *devices;
1161                 struct btrfs_device *tmp;
1162
1163                 device = NULL;
1164                 devices = &root->fs_info->fs_devices->devices;
1165                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1166                 list_for_each_entry(tmp, devices, dev_list) {
1167                         if (tmp->in_fs_metadata && !tmp->bdev) {
1168                                 device = tmp;
1169                                 break;
1170                         }
1171                 }
1172                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1173                 bdev = NULL;
1174                 bh = NULL;
1175                 disk_super = NULL;
1176                 if (!device) {
1177                         printk(KERN_ERR "btrfs: no missing devices found to "
1178                                "remove\n");
1179                         goto out;
1180                 }
1181         } else {
1182                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1183                                       root->fs_info->bdev_holder);
1184                 if (IS_ERR(bdev)) {
1185                         ret = PTR_ERR(bdev);
1186                         goto out;
1187                 }
1188
1189                 set_blocksize(bdev, 4096);
1190                 bh = btrfs_read_dev_super(bdev);
1191                 if (!bh) {
1192                         ret = -EIO;
1193                         goto error_close;
1194                 }
1195                 disk_super = (struct btrfs_super_block *)bh->b_data;
1196                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1197                 dev_uuid = disk_super->dev_item.uuid;
1198                 device = btrfs_find_device(root, devid, dev_uuid,
1199                                            disk_super->fsid);
1200                 if (!device) {
1201                         ret = -ENOENT;
1202                         goto error_brelse;
1203                 }
1204         }
1205
1206         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1207                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1208                        "device\n");
1209                 ret = -EINVAL;
1210                 goto error_brelse;
1211         }
1212
1213         if (device->writeable) {
1214                 list_del_init(&device->dev_alloc_list);
1215                 root->fs_info->fs_devices->rw_devices--;
1216         }
1217
1218         ret = btrfs_shrink_device(device, 0);
1219         if (ret)
1220                 goto error_brelse;
1221
1222         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1223         if (ret)
1224                 goto error_brelse;
1225
1226         device->in_fs_metadata = 0;
1227
1228         /*
1229          * the device list mutex makes sure that we don't change
1230          * the device list while someone else is writing out all
1231          * the device supers.
1232          */
1233         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1234         list_del_init(&device->dev_list);
1235         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1236
1237         device->fs_devices->num_devices--;
1238
1239         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1240                                  struct btrfs_device, dev_list);
1241         if (device->bdev == root->fs_info->sb->s_bdev)
1242                 root->fs_info->sb->s_bdev = next_device->bdev;
1243         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1244                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1245
1246         if (device->bdev) {
1247                 blkdev_put(device->bdev, device->mode | FMODE_EXCL);
1248                 device->bdev = NULL;
1249                 device->fs_devices->open_devices--;
1250         }
1251
1252         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1253         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1254
1255         if (device->fs_devices->open_devices == 0) {
1256                 struct btrfs_fs_devices *fs_devices;
1257                 fs_devices = root->fs_info->fs_devices;
1258                 while (fs_devices) {
1259                         if (fs_devices->seed == device->fs_devices)
1260                                 break;
1261                         fs_devices = fs_devices->seed;
1262                 }
1263                 fs_devices->seed = device->fs_devices->seed;
1264                 device->fs_devices->seed = NULL;
1265                 __btrfs_close_devices(device->fs_devices);
1266                 free_fs_devices(device->fs_devices);
1267         }
1268
1269         /*
1270          * at this point, the device is zero sized.  We want to
1271          * remove it from the devices list and zero out the old super
1272          */
1273         if (device->writeable) {
1274                 /* make sure this device isn't detected as part of
1275                  * the FS anymore
1276                  */
1277                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1278                 set_buffer_dirty(bh);
1279                 sync_dirty_buffer(bh);
1280         }
1281
1282         kfree(device->name);
1283         kfree(device);
1284         ret = 0;
1285
1286 error_brelse:
1287         brelse(bh);
1288 error_close:
1289         if (bdev)
1290                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1291 out:
1292         mutex_unlock(&root->fs_info->volume_mutex);
1293         mutex_unlock(&uuid_mutex);
1294         return ret;
1295 }
1296
1297 /*
1298  * does all the dirty work required for changing file system's UUID.
1299  */
1300 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1301                                 struct btrfs_root *root)
1302 {
1303         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1304         struct btrfs_fs_devices *old_devices;
1305         struct btrfs_fs_devices *seed_devices;
1306         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1307         struct btrfs_device *device;
1308         u64 super_flags;
1309
1310         BUG_ON(!mutex_is_locked(&uuid_mutex));
1311         if (!fs_devices->seeding)
1312                 return -EINVAL;
1313
1314         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1315         if (!seed_devices)
1316                 return -ENOMEM;
1317
1318         old_devices = clone_fs_devices(fs_devices);
1319         if (IS_ERR(old_devices)) {
1320                 kfree(seed_devices);
1321                 return PTR_ERR(old_devices);
1322         }
1323
1324         list_add(&old_devices->list, &fs_uuids);
1325
1326         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1327         seed_devices->opened = 1;
1328         INIT_LIST_HEAD(&seed_devices->devices);
1329         INIT_LIST_HEAD(&seed_devices->alloc_list);
1330         mutex_init(&seed_devices->device_list_mutex);
1331         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1332         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1333         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1334                 device->fs_devices = seed_devices;
1335         }
1336
1337         fs_devices->seeding = 0;
1338         fs_devices->num_devices = 0;
1339         fs_devices->open_devices = 0;
1340         fs_devices->seed = seed_devices;
1341
1342         generate_random_uuid(fs_devices->fsid);
1343         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1344         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1345         super_flags = btrfs_super_flags(disk_super) &
1346                       ~BTRFS_SUPER_FLAG_SEEDING;
1347         btrfs_set_super_flags(disk_super, super_flags);
1348
1349         return 0;
1350 }
1351
1352 /*
1353  * strore the expected generation for seed devices in device items.
1354  */
1355 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1356                                struct btrfs_root *root)
1357 {
1358         struct btrfs_path *path;
1359         struct extent_buffer *leaf;
1360         struct btrfs_dev_item *dev_item;
1361         struct btrfs_device *device;
1362         struct btrfs_key key;
1363         u8 fs_uuid[BTRFS_UUID_SIZE];
1364         u8 dev_uuid[BTRFS_UUID_SIZE];
1365         u64 devid;
1366         int ret;
1367
1368         path = btrfs_alloc_path();
1369         if (!path)
1370                 return -ENOMEM;
1371
1372         root = root->fs_info->chunk_root;
1373         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1374         key.offset = 0;
1375         key.type = BTRFS_DEV_ITEM_KEY;
1376
1377         while (1) {
1378                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1379                 if (ret < 0)
1380                         goto error;
1381
1382                 leaf = path->nodes[0];
1383 next_slot:
1384                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1385                         ret = btrfs_next_leaf(root, path);
1386                         if (ret > 0)
1387                                 break;
1388                         if (ret < 0)
1389                                 goto error;
1390                         leaf = path->nodes[0];
1391                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1392                         btrfs_release_path(root, path);
1393                         continue;
1394                 }
1395
1396                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1397                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1398                     key.type != BTRFS_DEV_ITEM_KEY)
1399                         break;
1400
1401                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1402                                           struct btrfs_dev_item);
1403                 devid = btrfs_device_id(leaf, dev_item);
1404                 read_extent_buffer(leaf, dev_uuid,
1405                                    (unsigned long)btrfs_device_uuid(dev_item),
1406                                    BTRFS_UUID_SIZE);
1407                 read_extent_buffer(leaf, fs_uuid,
1408                                    (unsigned long)btrfs_device_fsid(dev_item),
1409                                    BTRFS_UUID_SIZE);
1410                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1411                 BUG_ON(!device);
1412
1413                 if (device->fs_devices->seeding) {
1414                         btrfs_set_device_generation(leaf, dev_item,
1415                                                     device->generation);
1416                         btrfs_mark_buffer_dirty(leaf);
1417                 }
1418
1419                 path->slots[0]++;
1420                 goto next_slot;
1421         }
1422         ret = 0;
1423 error:
1424         btrfs_free_path(path);
1425         return ret;
1426 }
1427
1428 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1429 {
1430         struct btrfs_trans_handle *trans;
1431         struct btrfs_device *device;
1432         struct block_device *bdev;
1433         struct list_head *devices;
1434         struct super_block *sb = root->fs_info->sb;
1435         u64 total_bytes;
1436         int seeding_dev = 0;
1437         int ret = 0;
1438
1439         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1440                 return -EINVAL;
1441
1442         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1443         if (IS_ERR(bdev))
1444                 return PTR_ERR(bdev);
1445
1446         if (root->fs_info->fs_devices->seeding) {
1447                 seeding_dev = 1;
1448                 down_write(&sb->s_umount);
1449                 mutex_lock(&uuid_mutex);
1450         }
1451
1452         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1453         mutex_lock(&root->fs_info->volume_mutex);
1454
1455         devices = &root->fs_info->fs_devices->devices;
1456         /*
1457          * we have the volume lock, so we don't need the extra
1458          * device list mutex while reading the list here.
1459          */
1460         list_for_each_entry(device, devices, dev_list) {
1461                 if (device->bdev == bdev) {
1462                         ret = -EEXIST;
1463                         goto error;
1464                 }
1465         }
1466
1467         device = kzalloc(sizeof(*device), GFP_NOFS);
1468         if (!device) {
1469                 /* we can safely leave the fs_devices entry around */
1470                 ret = -ENOMEM;
1471                 goto error;
1472         }
1473
1474         device->name = kstrdup(device_path, GFP_NOFS);
1475         if (!device->name) {
1476                 kfree(device);
1477                 ret = -ENOMEM;
1478                 goto error;
1479         }
1480
1481         ret = find_next_devid(root, &device->devid);
1482         if (ret) {
1483                 kfree(device);
1484                 goto error;
1485         }
1486
1487         trans = btrfs_start_transaction(root, 0);
1488         lock_chunks(root);
1489
1490         device->writeable = 1;
1491         device->work.func = pending_bios_fn;
1492         generate_random_uuid(device->uuid);
1493         spin_lock_init(&device->io_lock);
1494         device->generation = trans->transid;
1495         device->io_width = root->sectorsize;
1496         device->io_align = root->sectorsize;
1497         device->sector_size = root->sectorsize;
1498         device->total_bytes = i_size_read(bdev->bd_inode);
1499         device->disk_total_bytes = device->total_bytes;
1500         device->dev_root = root->fs_info->dev_root;
1501         device->bdev = bdev;
1502         device->in_fs_metadata = 1;
1503         device->mode = 0;
1504         set_blocksize(device->bdev, 4096);
1505
1506         if (seeding_dev) {
1507                 sb->s_flags &= ~MS_RDONLY;
1508                 ret = btrfs_prepare_sprout(trans, root);
1509                 BUG_ON(ret);
1510         }
1511
1512         device->fs_devices = root->fs_info->fs_devices;
1513
1514         /*
1515          * we don't want write_supers to jump in here with our device
1516          * half setup
1517          */
1518         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1519         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1520         list_add(&device->dev_alloc_list,
1521                  &root->fs_info->fs_devices->alloc_list);
1522         root->fs_info->fs_devices->num_devices++;
1523         root->fs_info->fs_devices->open_devices++;
1524         root->fs_info->fs_devices->rw_devices++;
1525         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1526
1527         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1528                 root->fs_info->fs_devices->rotating = 1;
1529
1530         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1531         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1532                                     total_bytes + device->total_bytes);
1533
1534         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1535         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1536                                     total_bytes + 1);
1537         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1538
1539         if (seeding_dev) {
1540                 ret = init_first_rw_device(trans, root, device);
1541                 BUG_ON(ret);
1542                 ret = btrfs_finish_sprout(trans, root);
1543                 BUG_ON(ret);
1544         } else {
1545                 ret = btrfs_add_device(trans, root, device);
1546         }
1547
1548         /*
1549          * we've got more storage, clear any full flags on the space
1550          * infos
1551          */
1552         btrfs_clear_space_info_full(root->fs_info);
1553
1554         unlock_chunks(root);
1555         btrfs_commit_transaction(trans, root);
1556
1557         if (seeding_dev) {
1558                 mutex_unlock(&uuid_mutex);
1559                 up_write(&sb->s_umount);
1560
1561                 ret = btrfs_relocate_sys_chunks(root);
1562                 BUG_ON(ret);
1563         }
1564 out:
1565         mutex_unlock(&root->fs_info->volume_mutex);
1566         return ret;
1567 error:
1568         blkdev_put(bdev, FMODE_EXCL);
1569         if (seeding_dev) {
1570                 mutex_unlock(&uuid_mutex);
1571                 up_write(&sb->s_umount);
1572         }
1573         goto out;
1574 }
1575
1576 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1577                                         struct btrfs_device *device)
1578 {
1579         int ret;
1580         struct btrfs_path *path;
1581         struct btrfs_root *root;
1582         struct btrfs_dev_item *dev_item;
1583         struct extent_buffer *leaf;
1584         struct btrfs_key key;
1585
1586         root = device->dev_root->fs_info->chunk_root;
1587
1588         path = btrfs_alloc_path();
1589         if (!path)
1590                 return -ENOMEM;
1591
1592         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1593         key.type = BTRFS_DEV_ITEM_KEY;
1594         key.offset = device->devid;
1595
1596         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1597         if (ret < 0)
1598                 goto out;
1599
1600         if (ret > 0) {
1601                 ret = -ENOENT;
1602                 goto out;
1603         }
1604
1605         leaf = path->nodes[0];
1606         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1607
1608         btrfs_set_device_id(leaf, dev_item, device->devid);
1609         btrfs_set_device_type(leaf, dev_item, device->type);
1610         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1611         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1612         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1613         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1614         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1615         btrfs_mark_buffer_dirty(leaf);
1616
1617 out:
1618         btrfs_free_path(path);
1619         return ret;
1620 }
1621
1622 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1623                       struct btrfs_device *device, u64 new_size)
1624 {
1625         struct btrfs_super_block *super_copy =
1626                 &device->dev_root->fs_info->super_copy;
1627         u64 old_total = btrfs_super_total_bytes(super_copy);
1628         u64 diff = new_size - device->total_bytes;
1629
1630         if (!device->writeable)
1631                 return -EACCES;
1632         if (new_size <= device->total_bytes)
1633                 return -EINVAL;
1634
1635         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1636         device->fs_devices->total_rw_bytes += diff;
1637
1638         device->total_bytes = new_size;
1639         device->disk_total_bytes = new_size;
1640         btrfs_clear_space_info_full(device->dev_root->fs_info);
1641
1642         return btrfs_update_device(trans, device);
1643 }
1644
1645 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1646                       struct btrfs_device *device, u64 new_size)
1647 {
1648         int ret;
1649         lock_chunks(device->dev_root);
1650         ret = __btrfs_grow_device(trans, device, new_size);
1651         unlock_chunks(device->dev_root);
1652         return ret;
1653 }
1654
1655 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1656                             struct btrfs_root *root,
1657                             u64 chunk_tree, u64 chunk_objectid,
1658                             u64 chunk_offset)
1659 {
1660         int ret;
1661         struct btrfs_path *path;
1662         struct btrfs_key key;
1663
1664         root = root->fs_info->chunk_root;
1665         path = btrfs_alloc_path();
1666         if (!path)
1667                 return -ENOMEM;
1668
1669         key.objectid = chunk_objectid;
1670         key.offset = chunk_offset;
1671         key.type = BTRFS_CHUNK_ITEM_KEY;
1672
1673         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1674         BUG_ON(ret);
1675
1676         ret = btrfs_del_item(trans, root, path);
1677         BUG_ON(ret);
1678
1679         btrfs_free_path(path);
1680         return 0;
1681 }
1682
1683 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1684                         chunk_offset)
1685 {
1686         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1687         struct btrfs_disk_key *disk_key;
1688         struct btrfs_chunk *chunk;
1689         u8 *ptr;
1690         int ret = 0;
1691         u32 num_stripes;
1692         u32 array_size;
1693         u32 len = 0;
1694         u32 cur;
1695         struct btrfs_key key;
1696
1697         array_size = btrfs_super_sys_array_size(super_copy);
1698
1699         ptr = super_copy->sys_chunk_array;
1700         cur = 0;
1701
1702         while (cur < array_size) {
1703                 disk_key = (struct btrfs_disk_key *)ptr;
1704                 btrfs_disk_key_to_cpu(&key, disk_key);
1705
1706                 len = sizeof(*disk_key);
1707
1708                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1709                         chunk = (struct btrfs_chunk *)(ptr + len);
1710                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1711                         len += btrfs_chunk_item_size(num_stripes);
1712                 } else {
1713                         ret = -EIO;
1714                         break;
1715                 }
1716                 if (key.objectid == chunk_objectid &&
1717                     key.offset == chunk_offset) {
1718                         memmove(ptr, ptr + len, array_size - (cur + len));
1719                         array_size -= len;
1720                         btrfs_set_super_sys_array_size(super_copy, array_size);
1721                 } else {
1722                         ptr += len;
1723                         cur += len;
1724                 }
1725         }
1726         return ret;
1727 }
1728
1729 static int btrfs_relocate_chunk(struct btrfs_root *root,
1730                          u64 chunk_tree, u64 chunk_objectid,
1731                          u64 chunk_offset)
1732 {
1733         struct extent_map_tree *em_tree;
1734         struct btrfs_root *extent_root;
1735         struct btrfs_trans_handle *trans;
1736         struct extent_map *em;
1737         struct map_lookup *map;
1738         int ret;
1739         int i;
1740
1741         root = root->fs_info->chunk_root;
1742         extent_root = root->fs_info->extent_root;
1743         em_tree = &root->fs_info->mapping_tree.map_tree;
1744
1745         ret = btrfs_can_relocate(extent_root, chunk_offset);
1746         if (ret)
1747                 return -ENOSPC;
1748
1749         /* step one, relocate all the extents inside this chunk */
1750         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1751         if (ret)
1752                 return ret;
1753
1754         trans = btrfs_start_transaction(root, 0);
1755         BUG_ON(!trans);
1756
1757         lock_chunks(root);
1758
1759         /*
1760          * step two, delete the device extents and the
1761          * chunk tree entries
1762          */
1763         read_lock(&em_tree->lock);
1764         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1765         read_unlock(&em_tree->lock);
1766
1767         BUG_ON(em->start > chunk_offset ||
1768                em->start + em->len < chunk_offset);
1769         map = (struct map_lookup *)em->bdev;
1770
1771         for (i = 0; i < map->num_stripes; i++) {
1772                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1773                                             map->stripes[i].physical);
1774                 BUG_ON(ret);
1775
1776                 if (map->stripes[i].dev) {
1777                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1778                         BUG_ON(ret);
1779                 }
1780         }
1781         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1782                                chunk_offset);
1783
1784         BUG_ON(ret);
1785
1786         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1787                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1788                 BUG_ON(ret);
1789         }
1790
1791         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1792         BUG_ON(ret);
1793
1794         write_lock(&em_tree->lock);
1795         remove_extent_mapping(em_tree, em);
1796         write_unlock(&em_tree->lock);
1797
1798         kfree(map);
1799         em->bdev = NULL;
1800
1801         /* once for the tree */
1802         free_extent_map(em);
1803         /* once for us */
1804         free_extent_map(em);
1805
1806         unlock_chunks(root);
1807         btrfs_end_transaction(trans, root);
1808         return 0;
1809 }
1810
1811 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1812 {
1813         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1814         struct btrfs_path *path;
1815         struct extent_buffer *leaf;
1816         struct btrfs_chunk *chunk;
1817         struct btrfs_key key;
1818         struct btrfs_key found_key;
1819         u64 chunk_tree = chunk_root->root_key.objectid;
1820         u64 chunk_type;
1821         bool retried = false;
1822         int failed = 0;
1823         int ret;
1824
1825         path = btrfs_alloc_path();
1826         if (!path)
1827                 return -ENOMEM;
1828
1829 again:
1830         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1831         key.offset = (u64)-1;
1832         key.type = BTRFS_CHUNK_ITEM_KEY;
1833
1834         while (1) {
1835                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1836                 if (ret < 0)
1837                         goto error;
1838                 BUG_ON(ret == 0);
1839
1840                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1841                                           key.type);
1842                 if (ret < 0)
1843                         goto error;
1844                 if (ret > 0)
1845                         break;
1846
1847                 leaf = path->nodes[0];
1848                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1849
1850                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1851                                        struct btrfs_chunk);
1852                 chunk_type = btrfs_chunk_type(leaf, chunk);
1853                 btrfs_release_path(chunk_root, path);
1854
1855                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1856                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1857                                                    found_key.objectid,
1858                                                    found_key.offset);
1859                         if (ret == -ENOSPC)
1860                                 failed++;
1861                         else if (ret)
1862                                 BUG();
1863                 }
1864
1865                 if (found_key.offset == 0)
1866                         break;
1867                 key.offset = found_key.offset - 1;
1868         }
1869         ret = 0;
1870         if (failed && !retried) {
1871                 failed = 0;
1872                 retried = true;
1873                 goto again;
1874         } else if (failed && retried) {
1875                 WARN_ON(1);
1876                 ret = -ENOSPC;
1877         }
1878 error:
1879         btrfs_free_path(path);
1880         return ret;
1881 }
1882
1883 static u64 div_factor(u64 num, int factor)
1884 {
1885         if (factor == 10)
1886                 return num;
1887         num *= factor;
1888         do_div(num, 10);
1889         return num;
1890 }
1891
1892 int btrfs_balance(struct btrfs_root *dev_root)
1893 {
1894         int ret;
1895         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1896         struct btrfs_device *device;
1897         u64 old_size;
1898         u64 size_to_free;
1899         struct btrfs_path *path;
1900         struct btrfs_key key;
1901         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1902         struct btrfs_trans_handle *trans;
1903         struct btrfs_key found_key;
1904
1905         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1906                 return -EROFS;
1907
1908         mutex_lock(&dev_root->fs_info->volume_mutex);
1909         dev_root = dev_root->fs_info->dev_root;
1910
1911         /* step one make some room on all the devices */
1912         list_for_each_entry(device, devices, dev_list) {
1913                 old_size = device->total_bytes;
1914                 size_to_free = div_factor(old_size, 1);
1915                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1916                 if (!device->writeable ||
1917                     device->total_bytes - device->bytes_used > size_to_free)
1918                         continue;
1919
1920                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1921                 if (ret == -ENOSPC)
1922                         break;
1923                 BUG_ON(ret);
1924
1925                 trans = btrfs_start_transaction(dev_root, 0);
1926                 BUG_ON(!trans);
1927
1928                 ret = btrfs_grow_device(trans, device, old_size);
1929                 BUG_ON(ret);
1930
1931                 btrfs_end_transaction(trans, dev_root);
1932         }
1933
1934         /* step two, relocate all the chunks */
1935         path = btrfs_alloc_path();
1936         BUG_ON(!path);
1937
1938         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1939         key.offset = (u64)-1;
1940         key.type = BTRFS_CHUNK_ITEM_KEY;
1941
1942         while (1) {
1943                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1944                 if (ret < 0)
1945                         goto error;
1946
1947                 /*
1948                  * this shouldn't happen, it means the last relocate
1949                  * failed
1950                  */
1951                 if (ret == 0)
1952                         break;
1953
1954                 ret = btrfs_previous_item(chunk_root, path, 0,
1955                                           BTRFS_CHUNK_ITEM_KEY);
1956                 if (ret)
1957                         break;
1958
1959                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1960                                       path->slots[0]);
1961                 if (found_key.objectid != key.objectid)
1962                         break;
1963
1964                 /* chunk zero is special */
1965                 if (found_key.offset == 0)
1966                         break;
1967
1968                 btrfs_release_path(chunk_root, path);
1969                 ret = btrfs_relocate_chunk(chunk_root,
1970                                            chunk_root->root_key.objectid,
1971                                            found_key.objectid,
1972                                            found_key.offset);
1973                 BUG_ON(ret && ret != -ENOSPC);
1974                 key.offset = found_key.offset - 1;
1975         }
1976         ret = 0;
1977 error:
1978         btrfs_free_path(path);
1979         mutex_unlock(&dev_root->fs_info->volume_mutex);
1980         return ret;
1981 }
1982
1983 /*
1984  * shrinking a device means finding all of the device extents past
1985  * the new size, and then following the back refs to the chunks.
1986  * The chunk relocation code actually frees the device extent
1987  */
1988 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1989 {
1990         struct btrfs_trans_handle *trans;
1991         struct btrfs_root *root = device->dev_root;
1992         struct btrfs_dev_extent *dev_extent = NULL;
1993         struct btrfs_path *path;
1994         u64 length;
1995         u64 chunk_tree;
1996         u64 chunk_objectid;
1997         u64 chunk_offset;
1998         int ret;
1999         int slot;
2000         int failed = 0;
2001         bool retried = false;
2002         struct extent_buffer *l;
2003         struct btrfs_key key;
2004         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2005         u64 old_total = btrfs_super_total_bytes(super_copy);
2006         u64 old_size = device->total_bytes;
2007         u64 diff = device->total_bytes - new_size;
2008
2009         if (new_size >= device->total_bytes)
2010                 return -EINVAL;
2011
2012         path = btrfs_alloc_path();
2013         if (!path)
2014                 return -ENOMEM;
2015
2016         path->reada = 2;
2017
2018         lock_chunks(root);
2019
2020         device->total_bytes = new_size;
2021         if (device->writeable)
2022                 device->fs_devices->total_rw_bytes -= diff;
2023         unlock_chunks(root);
2024
2025 again:
2026         key.objectid = device->devid;
2027         key.offset = (u64)-1;
2028         key.type = BTRFS_DEV_EXTENT_KEY;
2029
2030         while (1) {
2031                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2032                 if (ret < 0)
2033                         goto done;
2034
2035                 ret = btrfs_previous_item(root, path, 0, key.type);
2036                 if (ret < 0)
2037                         goto done;
2038                 if (ret) {
2039                         ret = 0;
2040                         btrfs_release_path(root, path);
2041                         break;
2042                 }
2043
2044                 l = path->nodes[0];
2045                 slot = path->slots[0];
2046                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2047
2048                 if (key.objectid != device->devid) {
2049                         btrfs_release_path(root, path);
2050                         break;
2051                 }
2052
2053                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2054                 length = btrfs_dev_extent_length(l, dev_extent);
2055
2056                 if (key.offset + length <= new_size) {
2057                         btrfs_release_path(root, path);
2058                         break;
2059                 }
2060
2061                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2062                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2063                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2064                 btrfs_release_path(root, path);
2065
2066                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2067                                            chunk_offset);
2068                 if (ret && ret != -ENOSPC)
2069                         goto done;
2070                 if (ret == -ENOSPC)
2071                         failed++;
2072                 key.offset -= 1;
2073         }
2074
2075         if (failed && !retried) {
2076                 failed = 0;
2077                 retried = true;
2078                 goto again;
2079         } else if (failed && retried) {
2080                 ret = -ENOSPC;
2081                 lock_chunks(root);
2082
2083                 device->total_bytes = old_size;
2084                 if (device->writeable)
2085                         device->fs_devices->total_rw_bytes += diff;
2086                 unlock_chunks(root);
2087                 goto done;
2088         }
2089
2090         /* Shrinking succeeded, else we would be at "done". */
2091         trans = btrfs_start_transaction(root, 0);
2092         lock_chunks(root);
2093
2094         device->disk_total_bytes = new_size;
2095         /* Now btrfs_update_device() will change the on-disk size. */
2096         ret = btrfs_update_device(trans, device);
2097         if (ret) {
2098                 unlock_chunks(root);
2099                 btrfs_end_transaction(trans, root);
2100                 goto done;
2101         }
2102         WARN_ON(diff > old_total);
2103         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2104         unlock_chunks(root);
2105         btrfs_end_transaction(trans, root);
2106 done:
2107         btrfs_free_path(path);
2108         return ret;
2109 }
2110
2111 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2112                            struct btrfs_root *root,
2113                            struct btrfs_key *key,
2114                            struct btrfs_chunk *chunk, int item_size)
2115 {
2116         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2117         struct btrfs_disk_key disk_key;
2118         u32 array_size;
2119         u8 *ptr;
2120
2121         array_size = btrfs_super_sys_array_size(super_copy);
2122         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2123                 return -EFBIG;
2124
2125         ptr = super_copy->sys_chunk_array + array_size;
2126         btrfs_cpu_key_to_disk(&disk_key, key);
2127         memcpy(ptr, &disk_key, sizeof(disk_key));
2128         ptr += sizeof(disk_key);
2129         memcpy(ptr, chunk, item_size);
2130         item_size += sizeof(disk_key);
2131         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2132         return 0;
2133 }
2134
2135 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2136                                         int num_stripes, int sub_stripes)
2137 {
2138         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2139                 return calc_size;
2140         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2141                 return calc_size * (num_stripes / sub_stripes);
2142         else
2143                 return calc_size * num_stripes;
2144 }
2145
2146 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2147                                struct btrfs_root *extent_root,
2148                                struct map_lookup **map_ret,
2149                                u64 *num_bytes, u64 *stripe_size,
2150                                u64 start, u64 type)
2151 {
2152         struct btrfs_fs_info *info = extent_root->fs_info;
2153         struct btrfs_device *device = NULL;
2154         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2155         struct list_head *cur;
2156         struct map_lookup *map = NULL;
2157         struct extent_map_tree *em_tree;
2158         struct extent_map *em;
2159         struct list_head private_devs;
2160         int min_stripe_size = 1 * 1024 * 1024;
2161         u64 calc_size = 1024 * 1024 * 1024;
2162         u64 max_chunk_size = calc_size;
2163         u64 min_free;
2164         u64 avail;
2165         u64 max_avail = 0;
2166         u64 dev_offset;
2167         int num_stripes = 1;
2168         int min_stripes = 1;
2169         int sub_stripes = 0;
2170         int looped = 0;
2171         int ret;
2172         int index;
2173         int stripe_len = 64 * 1024;
2174
2175         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2176             (type & BTRFS_BLOCK_GROUP_DUP)) {
2177                 WARN_ON(1);
2178                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2179         }
2180         if (list_empty(&fs_devices->alloc_list))
2181                 return -ENOSPC;
2182
2183         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2184                 num_stripes = fs_devices->rw_devices;
2185                 min_stripes = 2;
2186         }
2187         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2188                 num_stripes = 2;
2189                 min_stripes = 2;
2190         }
2191         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2192                 if (fs_devices->rw_devices < 2)
2193                         return -ENOSPC;
2194                 num_stripes = 2;
2195                 min_stripes = 2;
2196         }
2197         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2198                 num_stripes = fs_devices->rw_devices;
2199                 if (num_stripes < 4)
2200                         return -ENOSPC;
2201                 num_stripes &= ~(u32)1;
2202                 sub_stripes = 2;
2203                 min_stripes = 4;
2204         }
2205
2206         if (type & BTRFS_BLOCK_GROUP_DATA) {
2207                 max_chunk_size = 10 * calc_size;
2208                 min_stripe_size = 64 * 1024 * 1024;
2209         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2210                 max_chunk_size = 256 * 1024 * 1024;
2211                 min_stripe_size = 32 * 1024 * 1024;
2212         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2213                 calc_size = 8 * 1024 * 1024;
2214                 max_chunk_size = calc_size * 2;
2215                 min_stripe_size = 1 * 1024 * 1024;
2216         }
2217
2218         /* we don't want a chunk larger than 10% of writeable space */
2219         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2220                              max_chunk_size);
2221
2222 again:
2223         max_avail = 0;
2224         if (!map || map->num_stripes != num_stripes) {
2225                 kfree(map);
2226                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2227                 if (!map)
2228                         return -ENOMEM;
2229                 map->num_stripes = num_stripes;
2230         }
2231
2232         if (calc_size * num_stripes > max_chunk_size) {
2233                 calc_size = max_chunk_size;
2234                 do_div(calc_size, num_stripes);
2235                 do_div(calc_size, stripe_len);
2236                 calc_size *= stripe_len;
2237         }
2238
2239         /* we don't want tiny stripes */
2240         if (!looped)
2241                 calc_size = max_t(u64, min_stripe_size, calc_size);
2242
2243         /*
2244          * we're about to do_div by the stripe_len so lets make sure
2245          * we end up with something bigger than a stripe
2246          */
2247         calc_size = max_t(u64, calc_size, stripe_len * 4);
2248
2249         do_div(calc_size, stripe_len);
2250         calc_size *= stripe_len;
2251
2252         cur = fs_devices->alloc_list.next;
2253         index = 0;
2254
2255         if (type & BTRFS_BLOCK_GROUP_DUP)
2256                 min_free = calc_size * 2;
2257         else
2258                 min_free = calc_size;
2259
2260         /*
2261          * we add 1MB because we never use the first 1MB of the device, unless
2262          * we've looped, then we are likely allocating the maximum amount of
2263          * space left already
2264          */
2265         if (!looped)
2266                 min_free += 1024 * 1024;
2267
2268         INIT_LIST_HEAD(&private_devs);
2269         while (index < num_stripes) {
2270                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2271                 BUG_ON(!device->writeable);
2272                 if (device->total_bytes > device->bytes_used)
2273                         avail = device->total_bytes - device->bytes_used;
2274                 else
2275                         avail = 0;
2276                 cur = cur->next;
2277
2278                 if (device->in_fs_metadata && avail >= min_free) {
2279                         ret = find_free_dev_extent(trans, device,
2280                                                    min_free, &dev_offset,
2281                                                    &max_avail);
2282                         if (ret == 0) {
2283                                 list_move_tail(&device->dev_alloc_list,
2284                                                &private_devs);
2285                                 map->stripes[index].dev = device;
2286                                 map->stripes[index].physical = dev_offset;
2287                                 index++;
2288                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2289                                         map->stripes[index].dev = device;
2290                                         map->stripes[index].physical =
2291                                                 dev_offset + calc_size;
2292                                         index++;
2293                                 }
2294                         }
2295                 } else if (device->in_fs_metadata && avail > max_avail)
2296                         max_avail = avail;
2297                 if (cur == &fs_devices->alloc_list)
2298                         break;
2299         }
2300         list_splice(&private_devs, &fs_devices->alloc_list);
2301         if (index < num_stripes) {
2302                 if (index >= min_stripes) {
2303                         num_stripes = index;
2304                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2305                                 num_stripes /= sub_stripes;
2306                                 num_stripes *= sub_stripes;
2307                         }
2308                         looped = 1;
2309                         goto again;
2310                 }
2311                 if (!looped && max_avail > 0) {
2312                         looped = 1;
2313                         calc_size = max_avail;
2314                         goto again;
2315                 }
2316                 kfree(map);
2317                 return -ENOSPC;
2318         }
2319         map->sector_size = extent_root->sectorsize;
2320         map->stripe_len = stripe_len;
2321         map->io_align = stripe_len;
2322         map->io_width = stripe_len;
2323         map->type = type;
2324         map->num_stripes = num_stripes;
2325         map->sub_stripes = sub_stripes;
2326
2327         *map_ret = map;
2328         *stripe_size = calc_size;
2329         *num_bytes = chunk_bytes_by_type(type, calc_size,
2330                                          num_stripes, sub_stripes);
2331
2332         em = alloc_extent_map(GFP_NOFS);
2333         if (!em) {
2334                 kfree(map);
2335                 return -ENOMEM;
2336         }
2337         em->bdev = (struct block_device *)map;
2338         em->start = start;
2339         em->len = *num_bytes;
2340         em->block_start = 0;
2341         em->block_len = em->len;
2342
2343         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2344         write_lock(&em_tree->lock);
2345         ret = add_extent_mapping(em_tree, em);
2346         write_unlock(&em_tree->lock);
2347         BUG_ON(ret);
2348         free_extent_map(em);
2349
2350         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2351                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2352                                      start, *num_bytes);
2353         BUG_ON(ret);
2354
2355         index = 0;
2356         while (index < map->num_stripes) {
2357                 device = map->stripes[index].dev;
2358                 dev_offset = map->stripes[index].physical;
2359
2360                 ret = btrfs_alloc_dev_extent(trans, device,
2361                                 info->chunk_root->root_key.objectid,
2362                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2363                                 start, dev_offset, calc_size);
2364                 BUG_ON(ret);
2365                 index++;
2366         }
2367
2368         return 0;
2369 }
2370
2371 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2372                                 struct btrfs_root *extent_root,
2373                                 struct map_lookup *map, u64 chunk_offset,
2374                                 u64 chunk_size, u64 stripe_size)
2375 {
2376         u64 dev_offset;
2377         struct btrfs_key key;
2378         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2379         struct btrfs_device *device;
2380         struct btrfs_chunk *chunk;
2381         struct btrfs_stripe *stripe;
2382         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2383         int index = 0;
2384         int ret;
2385
2386         chunk = kzalloc(item_size, GFP_NOFS);
2387         if (!chunk)
2388                 return -ENOMEM;
2389
2390         index = 0;
2391         while (index < map->num_stripes) {
2392                 device = map->stripes[index].dev;
2393                 device->bytes_used += stripe_size;
2394                 ret = btrfs_update_device(trans, device);
2395                 BUG_ON(ret);
2396                 index++;
2397         }
2398
2399         index = 0;
2400         stripe = &chunk->stripe;
2401         while (index < map->num_stripes) {
2402                 device = map->stripes[index].dev;
2403                 dev_offset = map->stripes[index].physical;
2404
2405                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2406                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2407                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2408                 stripe++;
2409                 index++;
2410         }
2411
2412         btrfs_set_stack_chunk_length(chunk, chunk_size);
2413         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2414         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2415         btrfs_set_stack_chunk_type(chunk, map->type);
2416         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2417         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2418         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2419         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2420         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2421
2422         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2423         key.type = BTRFS_CHUNK_ITEM_KEY;
2424         key.offset = chunk_offset;
2425
2426         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2427         BUG_ON(ret);
2428
2429         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2430                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2431                                              item_size);
2432                 BUG_ON(ret);
2433         }
2434         kfree(chunk);
2435         return 0;
2436 }
2437
2438 /*
2439  * Chunk allocation falls into two parts. The first part does works
2440  * that make the new allocated chunk useable, but not do any operation
2441  * that modifies the chunk tree. The second part does the works that
2442  * require modifying the chunk tree. This division is important for the
2443  * bootstrap process of adding storage to a seed btrfs.
2444  */
2445 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2446                       struct btrfs_root *extent_root, u64 type)
2447 {
2448         u64 chunk_offset;
2449         u64 chunk_size;
2450         u64 stripe_size;
2451         struct map_lookup *map;
2452         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2453         int ret;
2454
2455         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2456                               &chunk_offset);
2457         if (ret)
2458                 return ret;
2459
2460         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2461                                   &stripe_size, chunk_offset, type);
2462         if (ret)
2463                 return ret;
2464
2465         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2466                                    chunk_size, stripe_size);
2467         BUG_ON(ret);
2468         return 0;
2469 }
2470
2471 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2472                                          struct btrfs_root *root,
2473                                          struct btrfs_device *device)
2474 {
2475         u64 chunk_offset;
2476         u64 sys_chunk_offset;
2477         u64 chunk_size;
2478         u64 sys_chunk_size;
2479         u64 stripe_size;
2480         u64 sys_stripe_size;
2481         u64 alloc_profile;
2482         struct map_lookup *map;
2483         struct map_lookup *sys_map;
2484         struct btrfs_fs_info *fs_info = root->fs_info;
2485         struct btrfs_root *extent_root = fs_info->extent_root;
2486         int ret;
2487
2488         ret = find_next_chunk(fs_info->chunk_root,
2489                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2490         BUG_ON(ret);
2491
2492         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2493                         (fs_info->metadata_alloc_profile &
2494                          fs_info->avail_metadata_alloc_bits);
2495         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2496
2497         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2498                                   &stripe_size, chunk_offset, alloc_profile);
2499         BUG_ON(ret);
2500
2501         sys_chunk_offset = chunk_offset + chunk_size;
2502
2503         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2504                         (fs_info->system_alloc_profile &
2505                          fs_info->avail_system_alloc_bits);
2506         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2507
2508         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2509                                   &sys_chunk_size, &sys_stripe_size,
2510                                   sys_chunk_offset, alloc_profile);
2511         BUG_ON(ret);
2512
2513         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2514         BUG_ON(ret);
2515
2516         /*
2517          * Modifying chunk tree needs allocating new blocks from both
2518          * system block group and metadata block group. So we only can
2519          * do operations require modifying the chunk tree after both
2520          * block groups were created.
2521          */
2522         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2523                                    chunk_size, stripe_size);
2524         BUG_ON(ret);
2525
2526         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2527                                    sys_chunk_offset, sys_chunk_size,
2528                                    sys_stripe_size);
2529         BUG_ON(ret);
2530         return 0;
2531 }
2532
2533 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2534 {
2535         struct extent_map *em;
2536         struct map_lookup *map;
2537         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2538         int readonly = 0;
2539         int i;
2540
2541         read_lock(&map_tree->map_tree.lock);
2542         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2543         read_unlock(&map_tree->map_tree.lock);
2544         if (!em)
2545                 return 1;
2546
2547         if (btrfs_test_opt(root, DEGRADED)) {
2548                 free_extent_map(em);
2549                 return 0;
2550         }
2551
2552         map = (struct map_lookup *)em->bdev;
2553         for (i = 0; i < map->num_stripes; i++) {
2554                 if (!map->stripes[i].dev->writeable) {
2555                         readonly = 1;
2556                         break;
2557                 }
2558         }
2559         free_extent_map(em);
2560         return readonly;
2561 }
2562
2563 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2564 {
2565         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2566 }
2567
2568 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2569 {
2570         struct extent_map *em;
2571
2572         while (1) {
2573                 write_lock(&tree->map_tree.lock);
2574                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2575                 if (em)
2576                         remove_extent_mapping(&tree->map_tree, em);
2577                 write_unlock(&tree->map_tree.lock);
2578                 if (!em)
2579                         break;
2580                 kfree(em->bdev);
2581                 /* once for us */
2582                 free_extent_map(em);
2583                 /* once for the tree */
2584                 free_extent_map(em);
2585         }
2586 }
2587
2588 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2589 {
2590         struct extent_map *em;
2591         struct map_lookup *map;
2592         struct extent_map_tree *em_tree = &map_tree->map_tree;
2593         int ret;
2594
2595         read_lock(&em_tree->lock);
2596         em = lookup_extent_mapping(em_tree, logical, len);
2597         read_unlock(&em_tree->lock);
2598         BUG_ON(!em);
2599
2600         BUG_ON(em->start > logical || em->start + em->len < logical);
2601         map = (struct map_lookup *)em->bdev;
2602         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2603                 ret = map->num_stripes;
2604         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2605                 ret = map->sub_stripes;
2606         else
2607                 ret = 1;
2608         free_extent_map(em);
2609         return ret;
2610 }
2611
2612 static int find_live_mirror(struct map_lookup *map, int first, int num,
2613                             int optimal)
2614 {
2615         int i;
2616         if (map->stripes[optimal].dev->bdev)
2617                 return optimal;
2618         for (i = first; i < first + num; i++) {
2619                 if (map->stripes[i].dev->bdev)
2620                         return i;
2621         }
2622         /* we couldn't find one that doesn't fail.  Just return something
2623          * and the io error handling code will clean up eventually
2624          */
2625         return optimal;
2626 }
2627
2628 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2629                              u64 logical, u64 *length,
2630                              struct btrfs_multi_bio **multi_ret,
2631                              int mirror_num, struct page *unplug_page)
2632 {
2633         struct extent_map *em;
2634         struct map_lookup *map;
2635         struct extent_map_tree *em_tree = &map_tree->map_tree;
2636         u64 offset;
2637         u64 stripe_offset;
2638         u64 stripe_nr;
2639         int stripes_allocated = 8;
2640         int stripes_required = 1;
2641         int stripe_index;
2642         int i;
2643         int num_stripes;
2644         int max_errors = 0;
2645         struct btrfs_multi_bio *multi = NULL;
2646
2647         if (multi_ret && !(rw & REQ_WRITE))
2648                 stripes_allocated = 1;
2649 again:
2650         if (multi_ret) {
2651                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2652                                 GFP_NOFS);
2653                 if (!multi)
2654                         return -ENOMEM;
2655
2656                 atomic_set(&multi->error, 0);
2657         }
2658
2659         read_lock(&em_tree->lock);
2660         em = lookup_extent_mapping(em_tree, logical, *length);
2661         read_unlock(&em_tree->lock);
2662
2663         if (!em && unplug_page) {
2664                 kfree(multi);
2665                 return 0;
2666         }
2667
2668         if (!em) {
2669                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2670                        (unsigned long long)logical,
2671                        (unsigned long long)*length);
2672                 BUG();
2673         }
2674
2675         BUG_ON(em->start > logical || em->start + em->len < logical);
2676         map = (struct map_lookup *)em->bdev;
2677         offset = logical - em->start;
2678
2679         if (mirror_num > map->num_stripes)
2680                 mirror_num = 0;
2681
2682         /* if our multi bio struct is too small, back off and try again */
2683         if (rw & REQ_WRITE) {
2684                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2685                                  BTRFS_BLOCK_GROUP_DUP)) {
2686                         stripes_required = map->num_stripes;
2687                         max_errors = 1;
2688                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2689                         stripes_required = map->sub_stripes;
2690                         max_errors = 1;
2691                 }
2692         }
2693         if (multi_ret && (rw & REQ_WRITE) &&
2694             stripes_allocated < stripes_required) {
2695                 stripes_allocated = map->num_stripes;
2696                 free_extent_map(em);
2697                 kfree(multi);
2698                 goto again;
2699         }
2700         stripe_nr = offset;
2701         /*
2702          * stripe_nr counts the total number of stripes we have to stride
2703          * to get to this block
2704          */
2705         do_div(stripe_nr, map->stripe_len);
2706
2707         stripe_offset = stripe_nr * map->stripe_len;
2708         BUG_ON(offset < stripe_offset);
2709
2710         /* stripe_offset is the offset of this block in its stripe*/
2711         stripe_offset = offset - stripe_offset;
2712
2713         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2714                          BTRFS_BLOCK_GROUP_RAID10 |
2715                          BTRFS_BLOCK_GROUP_DUP)) {
2716                 /* we limit the length of each bio to what fits in a stripe */
2717                 *length = min_t(u64, em->len - offset,
2718                               map->stripe_len - stripe_offset);
2719         } else {
2720                 *length = em->len - offset;
2721         }
2722
2723         if (!multi_ret && !unplug_page)
2724                 goto out;
2725
2726         num_stripes = 1;
2727         stripe_index = 0;
2728         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2729                 if (unplug_page || (rw & REQ_WRITE))
2730                         num_stripes = map->num_stripes;
2731                 else if (mirror_num)
2732                         stripe_index = mirror_num - 1;
2733                 else {
2734                         stripe_index = find_live_mirror(map, 0,
2735                                             map->num_stripes,
2736                                             current->pid % map->num_stripes);
2737                 }
2738
2739         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2740                 if (rw & REQ_WRITE)
2741                         num_stripes = map->num_stripes;
2742                 else if (mirror_num)
2743                         stripe_index = mirror_num - 1;
2744
2745         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2746                 int factor = map->num_stripes / map->sub_stripes;
2747
2748                 stripe_index = do_div(stripe_nr, factor);
2749                 stripe_index *= map->sub_stripes;
2750
2751                 if (unplug_page || (rw & REQ_WRITE))
2752                         num_stripes = map->sub_stripes;
2753                 else if (mirror_num)
2754                         stripe_index += mirror_num - 1;
2755                 else {
2756                         stripe_index = find_live_mirror(map, stripe_index,
2757                                               map->sub_stripes, stripe_index +
2758                                               current->pid % map->sub_stripes);
2759                 }
2760         } else {
2761                 /*
2762                  * after this do_div call, stripe_nr is the number of stripes
2763                  * on this device we have to walk to find the data, and
2764                  * stripe_index is the number of our device in the stripe array
2765                  */
2766                 stripe_index = do_div(stripe_nr, map->num_stripes);
2767         }
2768         BUG_ON(stripe_index >= map->num_stripes);
2769
2770         for (i = 0; i < num_stripes; i++) {
2771                 if (unplug_page) {
2772                         struct btrfs_device *device;
2773                         struct backing_dev_info *bdi;
2774
2775                         device = map->stripes[stripe_index].dev;
2776                         if (device->bdev) {
2777                                 bdi = blk_get_backing_dev_info(device->bdev);
2778                                 if (bdi->unplug_io_fn)
2779                                         bdi->unplug_io_fn(bdi, unplug_page);
2780                         }
2781                 } else {
2782                         multi->stripes[i].physical =
2783                                 map->stripes[stripe_index].physical +
2784                                 stripe_offset + stripe_nr * map->stripe_len;
2785                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2786                 }
2787                 stripe_index++;
2788         }
2789         if (multi_ret) {
2790                 *multi_ret = multi;
2791                 multi->num_stripes = num_stripes;
2792                 multi->max_errors = max_errors;
2793         }
2794 out:
2795         free_extent_map(em);
2796         return 0;
2797 }
2798
2799 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2800                       u64 logical, u64 *length,
2801                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2802 {
2803         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2804                                  mirror_num, NULL);
2805 }
2806
2807 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2808                      u64 chunk_start, u64 physical, u64 devid,
2809                      u64 **logical, int *naddrs, int *stripe_len)
2810 {
2811         struct extent_map_tree *em_tree = &map_tree->map_tree;
2812         struct extent_map *em;
2813         struct map_lookup *map;
2814         u64 *buf;
2815         u64 bytenr;
2816         u64 length;
2817         u64 stripe_nr;
2818         int i, j, nr = 0;
2819
2820         read_lock(&em_tree->lock);
2821         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2822         read_unlock(&em_tree->lock);
2823
2824         BUG_ON(!em || em->start != chunk_start);
2825         map = (struct map_lookup *)em->bdev;
2826
2827         length = em->len;
2828         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2829                 do_div(length, map->num_stripes / map->sub_stripes);
2830         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2831                 do_div(length, map->num_stripes);
2832
2833         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2834         BUG_ON(!buf);
2835
2836         for (i = 0; i < map->num_stripes; i++) {
2837                 if (devid && map->stripes[i].dev->devid != devid)
2838                         continue;
2839                 if (map->stripes[i].physical > physical ||
2840                     map->stripes[i].physical + length <= physical)
2841                         continue;
2842
2843                 stripe_nr = physical - map->stripes[i].physical;
2844                 do_div(stripe_nr, map->stripe_len);
2845
2846                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2847                         stripe_nr = stripe_nr * map->num_stripes + i;
2848                         do_div(stripe_nr, map->sub_stripes);
2849                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2850                         stripe_nr = stripe_nr * map->num_stripes + i;
2851                 }
2852                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2853                 WARN_ON(nr >= map->num_stripes);
2854                 for (j = 0; j < nr; j++) {
2855                         if (buf[j] == bytenr)
2856                                 break;
2857                 }
2858                 if (j == nr) {
2859                         WARN_ON(nr >= map->num_stripes);
2860                         buf[nr++] = bytenr;
2861                 }
2862         }
2863
2864         *logical = buf;
2865         *naddrs = nr;
2866         *stripe_len = map->stripe_len;
2867
2868         free_extent_map(em);
2869         return 0;
2870 }
2871
2872 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2873                       u64 logical, struct page *page)
2874 {
2875         u64 length = PAGE_CACHE_SIZE;
2876         return __btrfs_map_block(map_tree, READ, logical, &length,
2877                                  NULL, 0, page);
2878 }
2879
2880 static void end_bio_multi_stripe(struct bio *bio, int err)
2881 {
2882         struct btrfs_multi_bio *multi = bio->bi_private;
2883         int is_orig_bio = 0;
2884
2885         if (err)
2886                 atomic_inc(&multi->error);
2887
2888         if (bio == multi->orig_bio)
2889                 is_orig_bio = 1;
2890
2891         if (atomic_dec_and_test(&multi->stripes_pending)) {
2892                 if (!is_orig_bio) {
2893                         bio_put(bio);
2894                         bio = multi->orig_bio;
2895                 }
2896                 bio->bi_private = multi->private;
2897                 bio->bi_end_io = multi->end_io;
2898                 /* only send an error to the higher layers if it is
2899                  * beyond the tolerance of the multi-bio
2900                  */
2901                 if (atomic_read(&multi->error) > multi->max_errors) {
2902                         err = -EIO;
2903                 } else if (err) {
2904                         /*
2905                          * this bio is actually up to date, we didn't
2906                          * go over the max number of errors
2907                          */
2908                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2909                         err = 0;
2910                 }
2911                 kfree(multi);
2912
2913                 bio_endio(bio, err);
2914         } else if (!is_orig_bio) {
2915                 bio_put(bio);
2916         }
2917 }
2918
2919 struct async_sched {
2920         struct bio *bio;
2921         int rw;
2922         struct btrfs_fs_info *info;
2923         struct btrfs_work work;
2924 };
2925
2926 /*
2927  * see run_scheduled_bios for a description of why bios are collected for
2928  * async submit.
2929  *
2930  * This will add one bio to the pending list for a device and make sure
2931  * the work struct is scheduled.
2932  */
2933 static noinline int schedule_bio(struct btrfs_root *root,
2934                                  struct btrfs_device *device,
2935                                  int rw, struct bio *bio)
2936 {
2937         int should_queue = 1;
2938         struct btrfs_pending_bios *pending_bios;
2939
2940         /* don't bother with additional async steps for reads, right now */
2941         if (!(rw & REQ_WRITE)) {
2942                 bio_get(bio);
2943                 submit_bio(rw, bio);
2944                 bio_put(bio);
2945                 return 0;
2946         }
2947
2948         /*
2949          * nr_async_bios allows us to reliably return congestion to the
2950          * higher layers.  Otherwise, the async bio makes it appear we have
2951          * made progress against dirty pages when we've really just put it
2952          * on a queue for later
2953          */
2954         atomic_inc(&root->fs_info->nr_async_bios);
2955         WARN_ON(bio->bi_next);
2956         bio->bi_next = NULL;
2957         bio->bi_rw |= rw;
2958
2959         spin_lock(&device->io_lock);
2960         if (bio->bi_rw & REQ_SYNC)
2961                 pending_bios = &device->pending_sync_bios;
2962         else
2963                 pending_bios = &device->pending_bios;
2964
2965         if (pending_bios->tail)
2966                 pending_bios->tail->bi_next = bio;
2967
2968         pending_bios->tail = bio;
2969         if (!pending_bios->head)
2970                 pending_bios->head = bio;
2971         if (device->running_pending)
2972                 should_queue = 0;
2973
2974         spin_unlock(&device->io_lock);
2975
2976         if (should_queue)
2977                 btrfs_queue_worker(&root->fs_info->submit_workers,
2978                                    &device->work);
2979         return 0;
2980 }
2981
2982 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2983                   int mirror_num, int async_submit)
2984 {
2985         struct btrfs_mapping_tree *map_tree;
2986         struct btrfs_device *dev;
2987         struct bio *first_bio = bio;
2988         u64 logical = (u64)bio->bi_sector << 9;
2989         u64 length = 0;
2990         u64 map_length;
2991         struct btrfs_multi_bio *multi = NULL;
2992         int ret;
2993         int dev_nr = 0;
2994         int total_devs = 1;
2995
2996         length = bio->bi_size;
2997         map_tree = &root->fs_info->mapping_tree;
2998         map_length = length;
2999
3000         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3001                               mirror_num);
3002         BUG_ON(ret);
3003
3004         total_devs = multi->num_stripes;
3005         if (map_length < length) {
3006                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3007                        "len %llu\n", (unsigned long long)logical,
3008                        (unsigned long long)length,
3009                        (unsigned long long)map_length);
3010                 BUG();
3011         }
3012         multi->end_io = first_bio->bi_end_io;
3013         multi->private = first_bio->bi_private;
3014         multi->orig_bio = first_bio;
3015         atomic_set(&multi->stripes_pending, multi->num_stripes);
3016
3017         while (dev_nr < total_devs) {
3018                 if (total_devs > 1) {
3019                         if (dev_nr < total_devs - 1) {
3020                                 bio = bio_clone(first_bio, GFP_NOFS);
3021                                 BUG_ON(!bio);
3022                         } else {
3023                                 bio = first_bio;
3024                         }
3025                         bio->bi_private = multi;
3026                         bio->bi_end_io = end_bio_multi_stripe;
3027                 }
3028                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3029                 dev = multi->stripes[dev_nr].dev;
3030                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3031                         bio->bi_bdev = dev->bdev;
3032                         if (async_submit)
3033                                 schedule_bio(root, dev, rw, bio);
3034                         else
3035                                 submit_bio(rw, bio);
3036                 } else {
3037                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3038                         bio->bi_sector = logical >> 9;
3039                         bio_endio(bio, -EIO);
3040                 }
3041                 dev_nr++;
3042         }
3043         if (total_devs == 1)
3044                 kfree(multi);
3045         return 0;
3046 }
3047
3048 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3049                                        u8 *uuid, u8 *fsid)
3050 {
3051         struct btrfs_device *device;
3052         struct btrfs_fs_devices *cur_devices;
3053
3054         cur_devices = root->fs_info->fs_devices;
3055         while (cur_devices) {
3056                 if (!fsid ||
3057                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3058                         device = __find_device(&cur_devices->devices,
3059                                                devid, uuid);
3060                         if (device)
3061                                 return device;
3062                 }
3063                 cur_devices = cur_devices->seed;
3064         }
3065         return NULL;
3066 }
3067
3068 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3069                                             u64 devid, u8 *dev_uuid)
3070 {
3071         struct btrfs_device *device;
3072         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3073
3074         device = kzalloc(sizeof(*device), GFP_NOFS);
3075         if (!device)
3076                 return NULL;
3077         list_add(&device->dev_list,
3078                  &fs_devices->devices);
3079         device->dev_root = root->fs_info->dev_root;
3080         device->devid = devid;
3081         device->work.func = pending_bios_fn;
3082         device->fs_devices = fs_devices;
3083         fs_devices->num_devices++;
3084         spin_lock_init(&device->io_lock);
3085         INIT_LIST_HEAD(&device->dev_alloc_list);
3086         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3087         return device;
3088 }
3089
3090 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3091                           struct extent_buffer *leaf,
3092                           struct btrfs_chunk *chunk)
3093 {
3094         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3095         struct map_lookup *map;
3096         struct extent_map *em;
3097         u64 logical;
3098         u64 length;
3099         u64 devid;
3100         u8 uuid[BTRFS_UUID_SIZE];
3101         int num_stripes;
3102         int ret;
3103         int i;
3104
3105         logical = key->offset;
3106         length = btrfs_chunk_length(leaf, chunk);
3107
3108         read_lock(&map_tree->map_tree.lock);
3109         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3110         read_unlock(&map_tree->map_tree.lock);
3111
3112         /* already mapped? */
3113         if (em && em->start <= logical && em->start + em->len > logical) {
3114                 free_extent_map(em);
3115                 return 0;
3116         } else if (em) {
3117                 free_extent_map(em);
3118         }
3119
3120         em = alloc_extent_map(GFP_NOFS);
3121         if (!em)
3122                 return -ENOMEM;
3123         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3124         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3125         if (!map) {
3126                 free_extent_map(em);
3127                 return -ENOMEM;
3128         }
3129
3130         em->bdev = (struct block_device *)map;
3131         em->start = logical;
3132         em->len = length;
3133         em->block_start = 0;
3134         em->block_len = em->len;
3135
3136         map->num_stripes = num_stripes;
3137         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3138         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3139         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3140         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3141         map->type = btrfs_chunk_type(leaf, chunk);
3142         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3143         for (i = 0; i < num_stripes; i++) {
3144                 map->stripes[i].physical =
3145                         btrfs_stripe_offset_nr(leaf, chunk, i);
3146                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3147                 read_extent_buffer(leaf, uuid, (unsigned long)
3148                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3149                                    BTRFS_UUID_SIZE);
3150                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3151                                                         NULL);
3152                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3153                         kfree(map);
3154                         free_extent_map(em);
3155                         return -EIO;
3156                 }
3157                 if (!map->stripes[i].dev) {
3158                         map->stripes[i].dev =
3159                                 add_missing_dev(root, devid, uuid);
3160                         if (!map->stripes[i].dev) {
3161                                 kfree(map);
3162                                 free_extent_map(em);
3163                                 return -EIO;
3164                         }
3165                 }
3166                 map->stripes[i].dev->in_fs_metadata = 1;
3167         }
3168
3169         write_lock(&map_tree->map_tree.lock);
3170         ret = add_extent_mapping(&map_tree->map_tree, em);
3171         write_unlock(&map_tree->map_tree.lock);
3172         BUG_ON(ret);
3173         free_extent_map(em);
3174
3175         return 0;
3176 }
3177
3178 static int fill_device_from_item(struct extent_buffer *leaf,
3179                                  struct btrfs_dev_item *dev_item,
3180                                  struct btrfs_device *device)
3181 {
3182         unsigned long ptr;
3183
3184         device->devid = btrfs_device_id(leaf, dev_item);
3185         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3186         device->total_bytes = device->disk_total_bytes;
3187         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3188         device->type = btrfs_device_type(leaf, dev_item);
3189         device->io_align = btrfs_device_io_align(leaf, dev_item);
3190         device->io_width = btrfs_device_io_width(leaf, dev_item);
3191         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3192
3193         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3194         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3195
3196         return 0;
3197 }
3198
3199 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3200 {
3201         struct btrfs_fs_devices *fs_devices;
3202         int ret;
3203
3204         mutex_lock(&uuid_mutex);
3205
3206         fs_devices = root->fs_info->fs_devices->seed;
3207         while (fs_devices) {
3208                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3209                         ret = 0;
3210                         goto out;
3211                 }
3212                 fs_devices = fs_devices->seed;
3213         }
3214
3215         fs_devices = find_fsid(fsid);
3216         if (!fs_devices) {
3217                 ret = -ENOENT;
3218                 goto out;
3219         }
3220
3221         fs_devices = clone_fs_devices(fs_devices);
3222         if (IS_ERR(fs_devices)) {
3223                 ret = PTR_ERR(fs_devices);
3224                 goto out;
3225         }
3226
3227         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3228                                    root->fs_info->bdev_holder);
3229         if (ret)
3230                 goto out;
3231
3232         if (!fs_devices->seeding) {
3233                 __btrfs_close_devices(fs_devices);
3234                 free_fs_devices(fs_devices);
3235                 ret = -EINVAL;
3236                 goto out;
3237         }
3238
3239         fs_devices->seed = root->fs_info->fs_devices->seed;
3240         root->fs_info->fs_devices->seed = fs_devices;
3241 out:
3242         mutex_unlock(&uuid_mutex);
3243         return ret;
3244 }
3245
3246 static int read_one_dev(struct btrfs_root *root,
3247                         struct extent_buffer *leaf,
3248                         struct btrfs_dev_item *dev_item)
3249 {
3250         struct btrfs_device *device;
3251         u64 devid;
3252         int ret;
3253         u8 fs_uuid[BTRFS_UUID_SIZE];
3254         u8 dev_uuid[BTRFS_UUID_SIZE];
3255
3256         devid = btrfs_device_id(leaf, dev_item);
3257         read_extent_buffer(leaf, dev_uuid,
3258                            (unsigned long)btrfs_device_uuid(dev_item),
3259                            BTRFS_UUID_SIZE);
3260         read_extent_buffer(leaf, fs_uuid,
3261                            (unsigned long)btrfs_device_fsid(dev_item),
3262                            BTRFS_UUID_SIZE);
3263
3264         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3265                 ret = open_seed_devices(root, fs_uuid);
3266                 if (ret && !btrfs_test_opt(root, DEGRADED))
3267                         return ret;
3268         }
3269
3270         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3271         if (!device || !device->bdev) {
3272                 if (!btrfs_test_opt(root, DEGRADED))
3273                         return -EIO;
3274
3275                 if (!device) {
3276                         printk(KERN_WARNING "warning devid %llu missing\n",
3277                                (unsigned long long)devid);
3278                         device = add_missing_dev(root, devid, dev_uuid);
3279                         if (!device)
3280                                 return -ENOMEM;
3281                 }
3282         }
3283
3284         if (device->fs_devices != root->fs_info->fs_devices) {
3285                 BUG_ON(device->writeable);
3286                 if (device->generation !=
3287                     btrfs_device_generation(leaf, dev_item))
3288                         return -EINVAL;
3289         }
3290
3291         fill_device_from_item(leaf, dev_item, device);
3292         device->dev_root = root->fs_info->dev_root;
3293         device->in_fs_metadata = 1;
3294         if (device->writeable)
3295                 device->fs_devices->total_rw_bytes += device->total_bytes;
3296         ret = 0;
3297         return ret;
3298 }
3299
3300 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3301 {
3302         struct btrfs_dev_item *dev_item;
3303
3304         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3305                                                      dev_item);
3306         return read_one_dev(root, buf, dev_item);
3307 }
3308
3309 int btrfs_read_sys_array(struct btrfs_root *root)
3310 {
3311         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3312         struct extent_buffer *sb;
3313         struct btrfs_disk_key *disk_key;
3314         struct btrfs_chunk *chunk;
3315         u8 *ptr;
3316         unsigned long sb_ptr;
3317         int ret = 0;
3318         u32 num_stripes;
3319         u32 array_size;
3320         u32 len = 0;
3321         u32 cur;
3322         struct btrfs_key key;
3323
3324         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3325                                           BTRFS_SUPER_INFO_SIZE);
3326         if (!sb)
3327                 return -ENOMEM;
3328         btrfs_set_buffer_uptodate(sb);
3329         btrfs_set_buffer_lockdep_class(sb, 0);
3330
3331         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3332         array_size = btrfs_super_sys_array_size(super_copy);
3333
3334         ptr = super_copy->sys_chunk_array;
3335         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3336         cur = 0;
3337
3338         while (cur < array_size) {
3339                 disk_key = (struct btrfs_disk_key *)ptr;
3340                 btrfs_disk_key_to_cpu(&key, disk_key);
3341
3342                 len = sizeof(*disk_key); ptr += len;
3343                 sb_ptr += len;
3344                 cur += len;
3345
3346                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3347                         chunk = (struct btrfs_chunk *)sb_ptr;
3348                         ret = read_one_chunk(root, &key, sb, chunk);
3349                         if (ret)
3350                                 break;
3351                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3352                         len = btrfs_chunk_item_size(num_stripes);
3353                 } else {
3354                         ret = -EIO;
3355                         break;
3356                 }
3357                 ptr += len;
3358                 sb_ptr += len;
3359                 cur += len;
3360         }
3361         free_extent_buffer(sb);
3362         return ret;
3363 }
3364
3365 int btrfs_read_chunk_tree(struct btrfs_root *root)
3366 {
3367         struct btrfs_path *path;
3368         struct extent_buffer *leaf;
3369         struct btrfs_key key;
3370         struct btrfs_key found_key;
3371         int ret;
3372         int slot;
3373
3374         root = root->fs_info->chunk_root;
3375
3376         path = btrfs_alloc_path();
3377         if (!path)
3378                 return -ENOMEM;
3379
3380         /* first we search for all of the device items, and then we
3381          * read in all of the chunk items.  This way we can create chunk
3382          * mappings that reference all of the devices that are afound
3383          */
3384         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3385         key.offset = 0;
3386         key.type = 0;
3387 again:
3388         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3389         if (ret < 0)
3390                 goto error;
3391         while (1) {
3392                 leaf = path->nodes[0];
3393                 slot = path->slots[0];
3394                 if (slot >= btrfs_header_nritems(leaf)) {
3395                         ret = btrfs_next_leaf(root, path);
3396                         if (ret == 0)
3397                                 continue;
3398                         if (ret < 0)
3399                                 goto error;
3400                         break;
3401                 }
3402                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3403                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3404                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3405                                 break;
3406                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3407                                 struct btrfs_dev_item *dev_item;
3408                                 dev_item = btrfs_item_ptr(leaf, slot,
3409                                                   struct btrfs_dev_item);
3410                                 ret = read_one_dev(root, leaf, dev_item);
3411                                 if (ret)
3412                                         goto error;
3413                         }
3414                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3415                         struct btrfs_chunk *chunk;
3416                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3417                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3418                         if (ret)
3419                                 goto error;
3420                 }
3421                 path->slots[0]++;
3422         }
3423         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3424                 key.objectid = 0;
3425                 btrfs_release_path(root, path);
3426                 goto again;
3427         }
3428         ret = 0;
3429 error:
3430         btrfs_free_path(path);
3431         return ret;
3432 }