Merge branch 'block'
[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 (!device->name || 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                 if (device->missing) {
422                         fs_devices->missing_devices--;
423                         device->missing = 0;
424                 }
425         }
426
427         if (found_transid > fs_devices->latest_trans) {
428                 fs_devices->latest_devid = devid;
429                 fs_devices->latest_trans = found_transid;
430         }
431         *fs_devices_ret = fs_devices;
432         return 0;
433 }
434
435 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
436 {
437         struct btrfs_fs_devices *fs_devices;
438         struct btrfs_device *device;
439         struct btrfs_device *orig_dev;
440
441         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
442         if (!fs_devices)
443                 return ERR_PTR(-ENOMEM);
444
445         INIT_LIST_HEAD(&fs_devices->devices);
446         INIT_LIST_HEAD(&fs_devices->alloc_list);
447         INIT_LIST_HEAD(&fs_devices->list);
448         mutex_init(&fs_devices->device_list_mutex);
449         fs_devices->latest_devid = orig->latest_devid;
450         fs_devices->latest_trans = orig->latest_trans;
451         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
452
453         mutex_lock(&orig->device_list_mutex);
454         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
455                 device = kzalloc(sizeof(*device), GFP_NOFS);
456                 if (!device)
457                         goto error;
458
459                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
460                 if (!device->name) {
461                         kfree(device);
462                         goto error;
463                 }
464
465                 device->devid = orig_dev->devid;
466                 device->work.func = pending_bios_fn;
467                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
468                 spin_lock_init(&device->io_lock);
469                 INIT_LIST_HEAD(&device->dev_list);
470                 INIT_LIST_HEAD(&device->dev_alloc_list);
471
472                 list_add(&device->dev_list, &fs_devices->devices);
473                 device->fs_devices = fs_devices;
474                 fs_devices->num_devices++;
475         }
476         mutex_unlock(&orig->device_list_mutex);
477         return fs_devices;
478 error:
479         mutex_unlock(&orig->device_list_mutex);
480         free_fs_devices(fs_devices);
481         return ERR_PTR(-ENOMEM);
482 }
483
484 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
485 {
486         struct btrfs_device *device, *next;
487
488         mutex_lock(&uuid_mutex);
489 again:
490         mutex_lock(&fs_devices->device_list_mutex);
491         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
492                 if (device->in_fs_metadata)
493                         continue;
494
495                 if (device->bdev) {
496                         blkdev_put(device->bdev, device->mode);
497                         device->bdev = NULL;
498                         fs_devices->open_devices--;
499                 }
500                 if (device->writeable) {
501                         list_del_init(&device->dev_alloc_list);
502                         device->writeable = 0;
503                         fs_devices->rw_devices--;
504                 }
505                 list_del_init(&device->dev_list);
506                 fs_devices->num_devices--;
507                 kfree(device->name);
508                 kfree(device);
509         }
510         mutex_unlock(&fs_devices->device_list_mutex);
511
512         if (fs_devices->seed) {
513                 fs_devices = fs_devices->seed;
514                 goto again;
515         }
516
517         mutex_unlock(&uuid_mutex);
518         return 0;
519 }
520
521 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
522 {
523         struct btrfs_device *device;
524
525         if (--fs_devices->opened > 0)
526                 return 0;
527
528         list_for_each_entry(device, &fs_devices->devices, dev_list) {
529                 if (device->bdev) {
530                         blkdev_put(device->bdev, device->mode);
531                         fs_devices->open_devices--;
532                 }
533                 if (device->writeable) {
534                         list_del_init(&device->dev_alloc_list);
535                         fs_devices->rw_devices--;
536                 }
537
538                 device->bdev = NULL;
539                 device->writeable = 0;
540                 device->in_fs_metadata = 0;
541         }
542         WARN_ON(fs_devices->open_devices);
543         WARN_ON(fs_devices->rw_devices);
544         fs_devices->opened = 0;
545         fs_devices->seeding = 0;
546
547         return 0;
548 }
549
550 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
551 {
552         struct btrfs_fs_devices *seed_devices = NULL;
553         int ret;
554
555         mutex_lock(&uuid_mutex);
556         ret = __btrfs_close_devices(fs_devices);
557         if (!fs_devices->opened) {
558                 seed_devices = fs_devices->seed;
559                 fs_devices->seed = NULL;
560         }
561         mutex_unlock(&uuid_mutex);
562
563         while (seed_devices) {
564                 fs_devices = seed_devices;
565                 seed_devices = fs_devices->seed;
566                 __btrfs_close_devices(fs_devices);
567                 free_fs_devices(fs_devices);
568         }
569         return ret;
570 }
571
572 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
573                                 fmode_t flags, void *holder)
574 {
575         struct block_device *bdev;
576         struct list_head *head = &fs_devices->devices;
577         struct btrfs_device *device;
578         struct block_device *latest_bdev = NULL;
579         struct buffer_head *bh;
580         struct btrfs_super_block *disk_super;
581         u64 latest_devid = 0;
582         u64 latest_transid = 0;
583         u64 devid;
584         int seeding = 1;
585         int ret = 0;
586
587         flags |= FMODE_EXCL;
588
589         list_for_each_entry(device, head, dev_list) {
590                 if (device->bdev)
591                         continue;
592                 if (!device->name)
593                         continue;
594
595                 bdev = blkdev_get_by_path(device->name, flags, holder);
596                 if (IS_ERR(bdev)) {
597                         printk(KERN_INFO "open %s failed\n", device->name);
598                         goto error;
599                 }
600                 set_blocksize(bdev, 4096);
601
602                 bh = btrfs_read_dev_super(bdev);
603                 if (!bh)
604                         goto error_close;
605
606                 disk_super = (struct btrfs_super_block *)bh->b_data;
607                 devid = btrfs_stack_device_id(&disk_super->dev_item);
608                 if (devid != device->devid)
609                         goto error_brelse;
610
611                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
612                            BTRFS_UUID_SIZE))
613                         goto error_brelse;
614
615                 device->generation = btrfs_super_generation(disk_super);
616                 if (!latest_transid || device->generation > latest_transid) {
617                         latest_devid = devid;
618                         latest_transid = device->generation;
619                         latest_bdev = bdev;
620                 }
621
622                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
623                         device->writeable = 0;
624                 } else {
625                         device->writeable = !bdev_read_only(bdev);
626                         seeding = 0;
627                 }
628
629                 device->bdev = bdev;
630                 device->in_fs_metadata = 0;
631                 device->mode = flags;
632
633                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
634                         fs_devices->rotating = 1;
635
636                 fs_devices->open_devices++;
637                 if (device->writeable) {
638                         fs_devices->rw_devices++;
639                         list_add(&device->dev_alloc_list,
640                                  &fs_devices->alloc_list);
641                 }
642                 continue;
643
644 error_brelse:
645                 brelse(bh);
646 error_close:
647                 blkdev_put(bdev, flags);
648 error:
649                 continue;
650         }
651         if (fs_devices->open_devices == 0) {
652                 ret = -EIO;
653                 goto out;
654         }
655         fs_devices->seeding = seeding;
656         fs_devices->opened = 1;
657         fs_devices->latest_bdev = latest_bdev;
658         fs_devices->latest_devid = latest_devid;
659         fs_devices->latest_trans = latest_transid;
660         fs_devices->total_rw_bytes = 0;
661 out:
662         return ret;
663 }
664
665 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
666                        fmode_t flags, void *holder)
667 {
668         int ret;
669
670         mutex_lock(&uuid_mutex);
671         if (fs_devices->opened) {
672                 fs_devices->opened++;
673                 ret = 0;
674         } else {
675                 ret = __btrfs_open_devices(fs_devices, flags, holder);
676         }
677         mutex_unlock(&uuid_mutex);
678         return ret;
679 }
680
681 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
682                           struct btrfs_fs_devices **fs_devices_ret)
683 {
684         struct btrfs_super_block *disk_super;
685         struct block_device *bdev;
686         struct buffer_head *bh;
687         int ret;
688         u64 devid;
689         u64 transid;
690
691         mutex_lock(&uuid_mutex);
692
693         flags |= FMODE_EXCL;
694         bdev = blkdev_get_by_path(path, flags, holder);
695
696         if (IS_ERR(bdev)) {
697                 ret = PTR_ERR(bdev);
698                 goto error;
699         }
700
701         ret = set_blocksize(bdev, 4096);
702         if (ret)
703                 goto error_close;
704         bh = btrfs_read_dev_super(bdev);
705         if (!bh) {
706                 ret = -EIO;
707                 goto error_close;
708         }
709         disk_super = (struct btrfs_super_block *)bh->b_data;
710         devid = btrfs_stack_device_id(&disk_super->dev_item);
711         transid = btrfs_super_generation(disk_super);
712         if (disk_super->label[0])
713                 printk(KERN_INFO "device label %s ", disk_super->label);
714         else {
715                 /* FIXME, make a readl uuid parser */
716                 printk(KERN_INFO "device fsid %llx-%llx ",
717                        *(unsigned long long *)disk_super->fsid,
718                        *(unsigned long long *)(disk_super->fsid + 8));
719         }
720         printk(KERN_CONT "devid %llu transid %llu %s\n",
721                (unsigned long long)devid, (unsigned long long)transid, path);
722         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
723
724         brelse(bh);
725 error_close:
726         blkdev_put(bdev, flags);
727 error:
728         mutex_unlock(&uuid_mutex);
729         return ret;
730 }
731
732 /*
733  * this uses a pretty simple search, the expectation is that it is
734  * called very infrequently and that a given device has a small number
735  * of extents
736  */
737 int find_free_dev_extent(struct btrfs_trans_handle *trans,
738                          struct btrfs_device *device, u64 num_bytes,
739                          u64 *start, u64 *max_avail)
740 {
741         struct btrfs_key key;
742         struct btrfs_root *root = device->dev_root;
743         struct btrfs_dev_extent *dev_extent = NULL;
744         struct btrfs_path *path;
745         u64 hole_size = 0;
746         u64 last_byte = 0;
747         u64 search_start = 0;
748         u64 search_end = device->total_bytes;
749         int ret;
750         int slot = 0;
751         int start_found;
752         struct extent_buffer *l;
753
754         path = btrfs_alloc_path();
755         if (!path)
756                 return -ENOMEM;
757         path->reada = 2;
758         start_found = 0;
759
760         /* FIXME use last free of some kind */
761
762         /* we don't want to overwrite the superblock on the drive,
763          * so we make sure to start at an offset of at least 1MB
764          */
765         search_start = max((u64)1024 * 1024, search_start);
766
767         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
768                 search_start = max(root->fs_info->alloc_start, search_start);
769
770         key.objectid = device->devid;
771         key.offset = search_start;
772         key.type = BTRFS_DEV_EXTENT_KEY;
773         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
774         if (ret < 0)
775                 goto error;
776         if (ret > 0) {
777                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
778                 if (ret < 0)
779                         goto error;
780                 if (ret > 0)
781                         start_found = 1;
782         }
783         l = path->nodes[0];
784         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
785         while (1) {
786                 l = path->nodes[0];
787                 slot = path->slots[0];
788                 if (slot >= btrfs_header_nritems(l)) {
789                         ret = btrfs_next_leaf(root, path);
790                         if (ret == 0)
791                                 continue;
792                         if (ret < 0)
793                                 goto error;
794 no_more_items:
795                         if (!start_found) {
796                                 if (search_start >= search_end) {
797                                         ret = -ENOSPC;
798                                         goto error;
799                                 }
800                                 *start = search_start;
801                                 start_found = 1;
802                                 goto check_pending;
803                         }
804                         *start = last_byte > search_start ?
805                                 last_byte : search_start;
806                         if (search_end <= *start) {
807                                 ret = -ENOSPC;
808                                 goto error;
809                         }
810                         goto check_pending;
811                 }
812                 btrfs_item_key_to_cpu(l, &key, slot);
813
814                 if (key.objectid < device->devid)
815                         goto next;
816
817                 if (key.objectid > device->devid)
818                         goto no_more_items;
819
820                 if (key.offset >= search_start && key.offset > last_byte &&
821                     start_found) {
822                         if (last_byte < search_start)
823                                 last_byte = search_start;
824                         hole_size = key.offset - last_byte;
825
826                         if (hole_size > *max_avail)
827                                 *max_avail = hole_size;
828
829                         if (key.offset > last_byte &&
830                             hole_size >= num_bytes) {
831                                 *start = last_byte;
832                                 goto check_pending;
833                         }
834                 }
835                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
836                         goto next;
837
838                 start_found = 1;
839                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
840                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
841 next:
842                 path->slots[0]++;
843                 cond_resched();
844         }
845 check_pending:
846         /* we have to make sure we didn't find an extent that has already
847          * been allocated by the map tree or the original allocation
848          */
849         BUG_ON(*start < search_start);
850
851         if (*start + num_bytes > search_end) {
852                 ret = -ENOSPC;
853                 goto error;
854         }
855         /* check for pending inserts here */
856         ret = 0;
857
858 error:
859         btrfs_free_path(path);
860         return ret;
861 }
862
863 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
864                           struct btrfs_device *device,
865                           u64 start)
866 {
867         int ret;
868         struct btrfs_path *path;
869         struct btrfs_root *root = device->dev_root;
870         struct btrfs_key key;
871         struct btrfs_key found_key;
872         struct extent_buffer *leaf = NULL;
873         struct btrfs_dev_extent *extent = NULL;
874
875         path = btrfs_alloc_path();
876         if (!path)
877                 return -ENOMEM;
878
879         key.objectid = device->devid;
880         key.offset = start;
881         key.type = BTRFS_DEV_EXTENT_KEY;
882
883         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
884         if (ret > 0) {
885                 ret = btrfs_previous_item(root, path, key.objectid,
886                                           BTRFS_DEV_EXTENT_KEY);
887                 BUG_ON(ret);
888                 leaf = path->nodes[0];
889                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
890                 extent = btrfs_item_ptr(leaf, path->slots[0],
891                                         struct btrfs_dev_extent);
892                 BUG_ON(found_key.offset > start || found_key.offset +
893                        btrfs_dev_extent_length(leaf, extent) < start);
894                 ret = 0;
895         } else if (ret == 0) {
896                 leaf = path->nodes[0];
897                 extent = btrfs_item_ptr(leaf, path->slots[0],
898                                         struct btrfs_dev_extent);
899         }
900         BUG_ON(ret);
901
902         if (device->bytes_used > 0)
903                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
904         ret = btrfs_del_item(trans, root, path);
905         BUG_ON(ret);
906
907         btrfs_free_path(path);
908         return ret;
909 }
910
911 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
912                            struct btrfs_device *device,
913                            u64 chunk_tree, u64 chunk_objectid,
914                            u64 chunk_offset, u64 start, u64 num_bytes)
915 {
916         int ret;
917         struct btrfs_path *path;
918         struct btrfs_root *root = device->dev_root;
919         struct btrfs_dev_extent *extent;
920         struct extent_buffer *leaf;
921         struct btrfs_key key;
922
923         WARN_ON(!device->in_fs_metadata);
924         path = btrfs_alloc_path();
925         if (!path)
926                 return -ENOMEM;
927
928         key.objectid = device->devid;
929         key.offset = start;
930         key.type = BTRFS_DEV_EXTENT_KEY;
931         ret = btrfs_insert_empty_item(trans, root, path, &key,
932                                       sizeof(*extent));
933         BUG_ON(ret);
934
935         leaf = path->nodes[0];
936         extent = btrfs_item_ptr(leaf, path->slots[0],
937                                 struct btrfs_dev_extent);
938         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
939         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
940         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
941
942         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
943                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
944                     BTRFS_UUID_SIZE);
945
946         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
947         btrfs_mark_buffer_dirty(leaf);
948         btrfs_free_path(path);
949         return ret;
950 }
951
952 static noinline int find_next_chunk(struct btrfs_root *root,
953                                     u64 objectid, u64 *offset)
954 {
955         struct btrfs_path *path;
956         int ret;
957         struct btrfs_key key;
958         struct btrfs_chunk *chunk;
959         struct btrfs_key found_key;
960
961         path = btrfs_alloc_path();
962         BUG_ON(!path);
963
964         key.objectid = objectid;
965         key.offset = (u64)-1;
966         key.type = BTRFS_CHUNK_ITEM_KEY;
967
968         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
969         if (ret < 0)
970                 goto error;
971
972         BUG_ON(ret == 0);
973
974         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
975         if (ret) {
976                 *offset = 0;
977         } else {
978                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
979                                       path->slots[0]);
980                 if (found_key.objectid != objectid)
981                         *offset = 0;
982                 else {
983                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
984                                                struct btrfs_chunk);
985                         *offset = found_key.offset +
986                                 btrfs_chunk_length(path->nodes[0], chunk);
987                 }
988         }
989         ret = 0;
990 error:
991         btrfs_free_path(path);
992         return ret;
993 }
994
995 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
996 {
997         int ret;
998         struct btrfs_key key;
999         struct btrfs_key found_key;
1000         struct btrfs_path *path;
1001
1002         root = root->fs_info->chunk_root;
1003
1004         path = btrfs_alloc_path();
1005         if (!path)
1006                 return -ENOMEM;
1007
1008         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1009         key.type = BTRFS_DEV_ITEM_KEY;
1010         key.offset = (u64)-1;
1011
1012         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1013         if (ret < 0)
1014                 goto error;
1015
1016         BUG_ON(ret == 0);
1017
1018         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1019                                   BTRFS_DEV_ITEM_KEY);
1020         if (ret) {
1021                 *objectid = 1;
1022         } else {
1023                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1024                                       path->slots[0]);
1025                 *objectid = found_key.offset + 1;
1026         }
1027         ret = 0;
1028 error:
1029         btrfs_free_path(path);
1030         return ret;
1031 }
1032
1033 /*
1034  * the device information is stored in the chunk root
1035  * the btrfs_device struct should be fully filled in
1036  */
1037 int btrfs_add_device(struct btrfs_trans_handle *trans,
1038                      struct btrfs_root *root,
1039                      struct btrfs_device *device)
1040 {
1041         int ret;
1042         struct btrfs_path *path;
1043         struct btrfs_dev_item *dev_item;
1044         struct extent_buffer *leaf;
1045         struct btrfs_key key;
1046         unsigned long ptr;
1047
1048         root = root->fs_info->chunk_root;
1049
1050         path = btrfs_alloc_path();
1051         if (!path)
1052                 return -ENOMEM;
1053
1054         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1055         key.type = BTRFS_DEV_ITEM_KEY;
1056         key.offset = device->devid;
1057
1058         ret = btrfs_insert_empty_item(trans, root, path, &key,
1059                                       sizeof(*dev_item));
1060         if (ret)
1061                 goto out;
1062
1063         leaf = path->nodes[0];
1064         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1065
1066         btrfs_set_device_id(leaf, dev_item, device->devid);
1067         btrfs_set_device_generation(leaf, dev_item, 0);
1068         btrfs_set_device_type(leaf, dev_item, device->type);
1069         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1070         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1071         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1072         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1073         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1074         btrfs_set_device_group(leaf, dev_item, 0);
1075         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1076         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1077         btrfs_set_device_start_offset(leaf, dev_item, 0);
1078
1079         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1080         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1081         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1082         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1083         btrfs_mark_buffer_dirty(leaf);
1084
1085         ret = 0;
1086 out:
1087         btrfs_free_path(path);
1088         return ret;
1089 }
1090
1091 static int btrfs_rm_dev_item(struct btrfs_root *root,
1092                              struct btrfs_device *device)
1093 {
1094         int ret;
1095         struct btrfs_path *path;
1096         struct btrfs_key key;
1097         struct btrfs_trans_handle *trans;
1098
1099         root = root->fs_info->chunk_root;
1100
1101         path = btrfs_alloc_path();
1102         if (!path)
1103                 return -ENOMEM;
1104
1105         trans = btrfs_start_transaction(root, 0);
1106         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1107         key.type = BTRFS_DEV_ITEM_KEY;
1108         key.offset = device->devid;
1109         lock_chunks(root);
1110
1111         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1112         if (ret < 0)
1113                 goto out;
1114
1115         if (ret > 0) {
1116                 ret = -ENOENT;
1117                 goto out;
1118         }
1119
1120         ret = btrfs_del_item(trans, root, path);
1121         if (ret)
1122                 goto out;
1123 out:
1124         btrfs_free_path(path);
1125         unlock_chunks(root);
1126         btrfs_commit_transaction(trans, root);
1127         return ret;
1128 }
1129
1130 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1131 {
1132         struct btrfs_device *device;
1133         struct btrfs_device *next_device;
1134         struct block_device *bdev;
1135         struct buffer_head *bh = NULL;
1136         struct btrfs_super_block *disk_super;
1137         u64 all_avail;
1138         u64 devid;
1139         u64 num_devices;
1140         u8 *dev_uuid;
1141         int ret = 0;
1142
1143         mutex_lock(&uuid_mutex);
1144         mutex_lock(&root->fs_info->volume_mutex);
1145
1146         all_avail = root->fs_info->avail_data_alloc_bits |
1147                 root->fs_info->avail_system_alloc_bits |
1148                 root->fs_info->avail_metadata_alloc_bits;
1149
1150         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1151             root->fs_info->fs_devices->num_devices <= 4) {
1152                 printk(KERN_ERR "btrfs: unable to go below four devices "
1153                        "on raid10\n");
1154                 ret = -EINVAL;
1155                 goto out;
1156         }
1157
1158         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1159             root->fs_info->fs_devices->num_devices <= 2) {
1160                 printk(KERN_ERR "btrfs: unable to go below two "
1161                        "devices on raid1\n");
1162                 ret = -EINVAL;
1163                 goto out;
1164         }
1165
1166         if (strcmp(device_path, "missing") == 0) {
1167                 struct list_head *devices;
1168                 struct btrfs_device *tmp;
1169
1170                 device = NULL;
1171                 devices = &root->fs_info->fs_devices->devices;
1172                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1173                 list_for_each_entry(tmp, devices, dev_list) {
1174                         if (tmp->in_fs_metadata && !tmp->bdev) {
1175                                 device = tmp;
1176                                 break;
1177                         }
1178                 }
1179                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1180                 bdev = NULL;
1181                 bh = NULL;
1182                 disk_super = NULL;
1183                 if (!device) {
1184                         printk(KERN_ERR "btrfs: no missing devices found to "
1185                                "remove\n");
1186                         goto out;
1187                 }
1188         } else {
1189                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1190                                           root->fs_info->bdev_holder);
1191                 if (IS_ERR(bdev)) {
1192                         ret = PTR_ERR(bdev);
1193                         goto out;
1194                 }
1195
1196                 set_blocksize(bdev, 4096);
1197                 bh = btrfs_read_dev_super(bdev);
1198                 if (!bh) {
1199                         ret = -EIO;
1200                         goto error_close;
1201                 }
1202                 disk_super = (struct btrfs_super_block *)bh->b_data;
1203                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1204                 dev_uuid = disk_super->dev_item.uuid;
1205                 device = btrfs_find_device(root, devid, dev_uuid,
1206                                            disk_super->fsid);
1207                 if (!device) {
1208                         ret = -ENOENT;
1209                         goto error_brelse;
1210                 }
1211         }
1212
1213         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1214                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1215                        "device\n");
1216                 ret = -EINVAL;
1217                 goto error_brelse;
1218         }
1219
1220         if (device->writeable) {
1221                 list_del_init(&device->dev_alloc_list);
1222                 root->fs_info->fs_devices->rw_devices--;
1223         }
1224
1225         ret = btrfs_shrink_device(device, 0);
1226         if (ret)
1227                 goto error_brelse;
1228
1229         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1230         if (ret)
1231                 goto error_brelse;
1232
1233         device->in_fs_metadata = 0;
1234
1235         /*
1236          * the device list mutex makes sure that we don't change
1237          * the device list while someone else is writing out all
1238          * the device supers.
1239          */
1240         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1241         list_del_init(&device->dev_list);
1242         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1243
1244         device->fs_devices->num_devices--;
1245
1246         if (device->missing)
1247                 root->fs_info->fs_devices->missing_devices--;
1248
1249         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1250                                  struct btrfs_device, dev_list);
1251         if (device->bdev == root->fs_info->sb->s_bdev)
1252                 root->fs_info->sb->s_bdev = next_device->bdev;
1253         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1254                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1255
1256         if (device->bdev) {
1257                 blkdev_put(device->bdev, device->mode);
1258                 device->bdev = NULL;
1259                 device->fs_devices->open_devices--;
1260         }
1261
1262         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1263         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1264
1265         if (device->fs_devices->open_devices == 0) {
1266                 struct btrfs_fs_devices *fs_devices;
1267                 fs_devices = root->fs_info->fs_devices;
1268                 while (fs_devices) {
1269                         if (fs_devices->seed == device->fs_devices)
1270                                 break;
1271                         fs_devices = fs_devices->seed;
1272                 }
1273                 fs_devices->seed = device->fs_devices->seed;
1274                 device->fs_devices->seed = NULL;
1275                 __btrfs_close_devices(device->fs_devices);
1276                 free_fs_devices(device->fs_devices);
1277         }
1278
1279         /*
1280          * at this point, the device is zero sized.  We want to
1281          * remove it from the devices list and zero out the old super
1282          */
1283         if (device->writeable) {
1284                 /* make sure this device isn't detected as part of
1285                  * the FS anymore
1286                  */
1287                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1288                 set_buffer_dirty(bh);
1289                 sync_dirty_buffer(bh);
1290         }
1291
1292         kfree(device->name);
1293         kfree(device);
1294         ret = 0;
1295
1296 error_brelse:
1297         brelse(bh);
1298 error_close:
1299         if (bdev)
1300                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1301 out:
1302         mutex_unlock(&root->fs_info->volume_mutex);
1303         mutex_unlock(&uuid_mutex);
1304         return ret;
1305 }
1306
1307 /*
1308  * does all the dirty work required for changing file system's UUID.
1309  */
1310 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1311                                 struct btrfs_root *root)
1312 {
1313         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1314         struct btrfs_fs_devices *old_devices;
1315         struct btrfs_fs_devices *seed_devices;
1316         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1317         struct btrfs_device *device;
1318         u64 super_flags;
1319
1320         BUG_ON(!mutex_is_locked(&uuid_mutex));
1321         if (!fs_devices->seeding)
1322                 return -EINVAL;
1323
1324         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1325         if (!seed_devices)
1326                 return -ENOMEM;
1327
1328         old_devices = clone_fs_devices(fs_devices);
1329         if (IS_ERR(old_devices)) {
1330                 kfree(seed_devices);
1331                 return PTR_ERR(old_devices);
1332         }
1333
1334         list_add(&old_devices->list, &fs_uuids);
1335
1336         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1337         seed_devices->opened = 1;
1338         INIT_LIST_HEAD(&seed_devices->devices);
1339         INIT_LIST_HEAD(&seed_devices->alloc_list);
1340         mutex_init(&seed_devices->device_list_mutex);
1341         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1342         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1343         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1344                 device->fs_devices = seed_devices;
1345         }
1346
1347         fs_devices->seeding = 0;
1348         fs_devices->num_devices = 0;
1349         fs_devices->open_devices = 0;
1350         fs_devices->seed = seed_devices;
1351
1352         generate_random_uuid(fs_devices->fsid);
1353         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1354         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1355         super_flags = btrfs_super_flags(disk_super) &
1356                       ~BTRFS_SUPER_FLAG_SEEDING;
1357         btrfs_set_super_flags(disk_super, super_flags);
1358
1359         return 0;
1360 }
1361
1362 /*
1363  * strore the expected generation for seed devices in device items.
1364  */
1365 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1366                                struct btrfs_root *root)
1367 {
1368         struct btrfs_path *path;
1369         struct extent_buffer *leaf;
1370         struct btrfs_dev_item *dev_item;
1371         struct btrfs_device *device;
1372         struct btrfs_key key;
1373         u8 fs_uuid[BTRFS_UUID_SIZE];
1374         u8 dev_uuid[BTRFS_UUID_SIZE];
1375         u64 devid;
1376         int ret;
1377
1378         path = btrfs_alloc_path();
1379         if (!path)
1380                 return -ENOMEM;
1381
1382         root = root->fs_info->chunk_root;
1383         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1384         key.offset = 0;
1385         key.type = BTRFS_DEV_ITEM_KEY;
1386
1387         while (1) {
1388                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1389                 if (ret < 0)
1390                         goto error;
1391
1392                 leaf = path->nodes[0];
1393 next_slot:
1394                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1395                         ret = btrfs_next_leaf(root, path);
1396                         if (ret > 0)
1397                                 break;
1398                         if (ret < 0)
1399                                 goto error;
1400                         leaf = path->nodes[0];
1401                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1402                         btrfs_release_path(root, path);
1403                         continue;
1404                 }
1405
1406                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1407                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1408                     key.type != BTRFS_DEV_ITEM_KEY)
1409                         break;
1410
1411                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1412                                           struct btrfs_dev_item);
1413                 devid = btrfs_device_id(leaf, dev_item);
1414                 read_extent_buffer(leaf, dev_uuid,
1415                                    (unsigned long)btrfs_device_uuid(dev_item),
1416                                    BTRFS_UUID_SIZE);
1417                 read_extent_buffer(leaf, fs_uuid,
1418                                    (unsigned long)btrfs_device_fsid(dev_item),
1419                                    BTRFS_UUID_SIZE);
1420                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1421                 BUG_ON(!device);
1422
1423                 if (device->fs_devices->seeding) {
1424                         btrfs_set_device_generation(leaf, dev_item,
1425                                                     device->generation);
1426                         btrfs_mark_buffer_dirty(leaf);
1427                 }
1428
1429                 path->slots[0]++;
1430                 goto next_slot;
1431         }
1432         ret = 0;
1433 error:
1434         btrfs_free_path(path);
1435         return ret;
1436 }
1437
1438 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1439 {
1440         struct btrfs_trans_handle *trans;
1441         struct btrfs_device *device;
1442         struct block_device *bdev;
1443         struct list_head *devices;
1444         struct super_block *sb = root->fs_info->sb;
1445         u64 total_bytes;
1446         int seeding_dev = 0;
1447         int ret = 0;
1448
1449         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1450                 return -EINVAL;
1451
1452         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1453                                   root->fs_info->bdev_holder);
1454         if (IS_ERR(bdev))
1455                 return PTR_ERR(bdev);
1456
1457         if (root->fs_info->fs_devices->seeding) {
1458                 seeding_dev = 1;
1459                 down_write(&sb->s_umount);
1460                 mutex_lock(&uuid_mutex);
1461         }
1462
1463         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1464         mutex_lock(&root->fs_info->volume_mutex);
1465
1466         devices = &root->fs_info->fs_devices->devices;
1467         /*
1468          * we have the volume lock, so we don't need the extra
1469          * device list mutex while reading the list here.
1470          */
1471         list_for_each_entry(device, devices, dev_list) {
1472                 if (device->bdev == bdev) {
1473                         ret = -EEXIST;
1474                         goto error;
1475                 }
1476         }
1477
1478         device = kzalloc(sizeof(*device), GFP_NOFS);
1479         if (!device) {
1480                 /* we can safely leave the fs_devices entry around */
1481                 ret = -ENOMEM;
1482                 goto error;
1483         }
1484
1485         device->name = kstrdup(device_path, GFP_NOFS);
1486         if (!device->name) {
1487                 kfree(device);
1488                 ret = -ENOMEM;
1489                 goto error;
1490         }
1491
1492         ret = find_next_devid(root, &device->devid);
1493         if (ret) {
1494                 kfree(device);
1495                 goto error;
1496         }
1497
1498         trans = btrfs_start_transaction(root, 0);
1499         lock_chunks(root);
1500
1501         device->writeable = 1;
1502         device->work.func = pending_bios_fn;
1503         generate_random_uuid(device->uuid);
1504         spin_lock_init(&device->io_lock);
1505         device->generation = trans->transid;
1506         device->io_width = root->sectorsize;
1507         device->io_align = root->sectorsize;
1508         device->sector_size = root->sectorsize;
1509         device->total_bytes = i_size_read(bdev->bd_inode);
1510         device->disk_total_bytes = device->total_bytes;
1511         device->dev_root = root->fs_info->dev_root;
1512         device->bdev = bdev;
1513         device->in_fs_metadata = 1;
1514         device->mode = 0;
1515         set_blocksize(device->bdev, 4096);
1516
1517         if (seeding_dev) {
1518                 sb->s_flags &= ~MS_RDONLY;
1519                 ret = btrfs_prepare_sprout(trans, root);
1520                 BUG_ON(ret);
1521         }
1522
1523         device->fs_devices = root->fs_info->fs_devices;
1524
1525         /*
1526          * we don't want write_supers to jump in here with our device
1527          * half setup
1528          */
1529         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1530         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1531         list_add(&device->dev_alloc_list,
1532                  &root->fs_info->fs_devices->alloc_list);
1533         root->fs_info->fs_devices->num_devices++;
1534         root->fs_info->fs_devices->open_devices++;
1535         root->fs_info->fs_devices->rw_devices++;
1536         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1537
1538         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1539                 root->fs_info->fs_devices->rotating = 1;
1540
1541         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1542         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1543                                     total_bytes + device->total_bytes);
1544
1545         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1546         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1547                                     total_bytes + 1);
1548         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1549
1550         if (seeding_dev) {
1551                 ret = init_first_rw_device(trans, root, device);
1552                 BUG_ON(ret);
1553                 ret = btrfs_finish_sprout(trans, root);
1554                 BUG_ON(ret);
1555         } else {
1556                 ret = btrfs_add_device(trans, root, device);
1557         }
1558
1559         /*
1560          * we've got more storage, clear any full flags on the space
1561          * infos
1562          */
1563         btrfs_clear_space_info_full(root->fs_info);
1564
1565         unlock_chunks(root);
1566         btrfs_commit_transaction(trans, root);
1567
1568         if (seeding_dev) {
1569                 mutex_unlock(&uuid_mutex);
1570                 up_write(&sb->s_umount);
1571
1572                 ret = btrfs_relocate_sys_chunks(root);
1573                 BUG_ON(ret);
1574         }
1575 out:
1576         mutex_unlock(&root->fs_info->volume_mutex);
1577         return ret;
1578 error:
1579         blkdev_put(bdev, FMODE_EXCL);
1580         if (seeding_dev) {
1581                 mutex_unlock(&uuid_mutex);
1582                 up_write(&sb->s_umount);
1583         }
1584         goto out;
1585 }
1586
1587 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1588                                         struct btrfs_device *device)
1589 {
1590         int ret;
1591         struct btrfs_path *path;
1592         struct btrfs_root *root;
1593         struct btrfs_dev_item *dev_item;
1594         struct extent_buffer *leaf;
1595         struct btrfs_key key;
1596
1597         root = device->dev_root->fs_info->chunk_root;
1598
1599         path = btrfs_alloc_path();
1600         if (!path)
1601                 return -ENOMEM;
1602
1603         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1604         key.type = BTRFS_DEV_ITEM_KEY;
1605         key.offset = device->devid;
1606
1607         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1608         if (ret < 0)
1609                 goto out;
1610
1611         if (ret > 0) {
1612                 ret = -ENOENT;
1613                 goto out;
1614         }
1615
1616         leaf = path->nodes[0];
1617         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1618
1619         btrfs_set_device_id(leaf, dev_item, device->devid);
1620         btrfs_set_device_type(leaf, dev_item, device->type);
1621         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1622         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1623         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1624         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1625         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1626         btrfs_mark_buffer_dirty(leaf);
1627
1628 out:
1629         btrfs_free_path(path);
1630         return ret;
1631 }
1632
1633 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1634                       struct btrfs_device *device, u64 new_size)
1635 {
1636         struct btrfs_super_block *super_copy =
1637                 &device->dev_root->fs_info->super_copy;
1638         u64 old_total = btrfs_super_total_bytes(super_copy);
1639         u64 diff = new_size - device->total_bytes;
1640
1641         if (!device->writeable)
1642                 return -EACCES;
1643         if (new_size <= device->total_bytes)
1644                 return -EINVAL;
1645
1646         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1647         device->fs_devices->total_rw_bytes += diff;
1648
1649         device->total_bytes = new_size;
1650         device->disk_total_bytes = new_size;
1651         btrfs_clear_space_info_full(device->dev_root->fs_info);
1652
1653         return btrfs_update_device(trans, device);
1654 }
1655
1656 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1657                       struct btrfs_device *device, u64 new_size)
1658 {
1659         int ret;
1660         lock_chunks(device->dev_root);
1661         ret = __btrfs_grow_device(trans, device, new_size);
1662         unlock_chunks(device->dev_root);
1663         return ret;
1664 }
1665
1666 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1667                             struct btrfs_root *root,
1668                             u64 chunk_tree, u64 chunk_objectid,
1669                             u64 chunk_offset)
1670 {
1671         int ret;
1672         struct btrfs_path *path;
1673         struct btrfs_key key;
1674
1675         root = root->fs_info->chunk_root;
1676         path = btrfs_alloc_path();
1677         if (!path)
1678                 return -ENOMEM;
1679
1680         key.objectid = chunk_objectid;
1681         key.offset = chunk_offset;
1682         key.type = BTRFS_CHUNK_ITEM_KEY;
1683
1684         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1685         BUG_ON(ret);
1686
1687         ret = btrfs_del_item(trans, root, path);
1688         BUG_ON(ret);
1689
1690         btrfs_free_path(path);
1691         return 0;
1692 }
1693
1694 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1695                         chunk_offset)
1696 {
1697         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1698         struct btrfs_disk_key *disk_key;
1699         struct btrfs_chunk *chunk;
1700         u8 *ptr;
1701         int ret = 0;
1702         u32 num_stripes;
1703         u32 array_size;
1704         u32 len = 0;
1705         u32 cur;
1706         struct btrfs_key key;
1707
1708         array_size = btrfs_super_sys_array_size(super_copy);
1709
1710         ptr = super_copy->sys_chunk_array;
1711         cur = 0;
1712
1713         while (cur < array_size) {
1714                 disk_key = (struct btrfs_disk_key *)ptr;
1715                 btrfs_disk_key_to_cpu(&key, disk_key);
1716
1717                 len = sizeof(*disk_key);
1718
1719                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1720                         chunk = (struct btrfs_chunk *)(ptr + len);
1721                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1722                         len += btrfs_chunk_item_size(num_stripes);
1723                 } else {
1724                         ret = -EIO;
1725                         break;
1726                 }
1727                 if (key.objectid == chunk_objectid &&
1728                     key.offset == chunk_offset) {
1729                         memmove(ptr, ptr + len, array_size - (cur + len));
1730                         array_size -= len;
1731                         btrfs_set_super_sys_array_size(super_copy, array_size);
1732                 } else {
1733                         ptr += len;
1734                         cur += len;
1735                 }
1736         }
1737         return ret;
1738 }
1739
1740 static int btrfs_relocate_chunk(struct btrfs_root *root,
1741                          u64 chunk_tree, u64 chunk_objectid,
1742                          u64 chunk_offset)
1743 {
1744         struct extent_map_tree *em_tree;
1745         struct btrfs_root *extent_root;
1746         struct btrfs_trans_handle *trans;
1747         struct extent_map *em;
1748         struct map_lookup *map;
1749         int ret;
1750         int i;
1751
1752         root = root->fs_info->chunk_root;
1753         extent_root = root->fs_info->extent_root;
1754         em_tree = &root->fs_info->mapping_tree.map_tree;
1755
1756         ret = btrfs_can_relocate(extent_root, chunk_offset);
1757         if (ret)
1758                 return -ENOSPC;
1759
1760         /* step one, relocate all the extents inside this chunk */
1761         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1762         if (ret)
1763                 return ret;
1764
1765         trans = btrfs_start_transaction(root, 0);
1766         BUG_ON(!trans);
1767
1768         lock_chunks(root);
1769
1770         /*
1771          * step two, delete the device extents and the
1772          * chunk tree entries
1773          */
1774         read_lock(&em_tree->lock);
1775         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1776         read_unlock(&em_tree->lock);
1777
1778         BUG_ON(em->start > chunk_offset ||
1779                em->start + em->len < chunk_offset);
1780         map = (struct map_lookup *)em->bdev;
1781
1782         for (i = 0; i < map->num_stripes; i++) {
1783                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1784                                             map->stripes[i].physical);
1785                 BUG_ON(ret);
1786
1787                 if (map->stripes[i].dev) {
1788                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1789                         BUG_ON(ret);
1790                 }
1791         }
1792         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1793                                chunk_offset);
1794
1795         BUG_ON(ret);
1796
1797         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1798                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1799                 BUG_ON(ret);
1800         }
1801
1802         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1803         BUG_ON(ret);
1804
1805         write_lock(&em_tree->lock);
1806         remove_extent_mapping(em_tree, em);
1807         write_unlock(&em_tree->lock);
1808
1809         kfree(map);
1810         em->bdev = NULL;
1811
1812         /* once for the tree */
1813         free_extent_map(em);
1814         /* once for us */
1815         free_extent_map(em);
1816
1817         unlock_chunks(root);
1818         btrfs_end_transaction(trans, root);
1819         return 0;
1820 }
1821
1822 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1823 {
1824         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1825         struct btrfs_path *path;
1826         struct extent_buffer *leaf;
1827         struct btrfs_chunk *chunk;
1828         struct btrfs_key key;
1829         struct btrfs_key found_key;
1830         u64 chunk_tree = chunk_root->root_key.objectid;
1831         u64 chunk_type;
1832         bool retried = false;
1833         int failed = 0;
1834         int ret;
1835
1836         path = btrfs_alloc_path();
1837         if (!path)
1838                 return -ENOMEM;
1839
1840 again:
1841         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1842         key.offset = (u64)-1;
1843         key.type = BTRFS_CHUNK_ITEM_KEY;
1844
1845         while (1) {
1846                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1847                 if (ret < 0)
1848                         goto error;
1849                 BUG_ON(ret == 0);
1850
1851                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1852                                           key.type);
1853                 if (ret < 0)
1854                         goto error;
1855                 if (ret > 0)
1856                         break;
1857
1858                 leaf = path->nodes[0];
1859                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1860
1861                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1862                                        struct btrfs_chunk);
1863                 chunk_type = btrfs_chunk_type(leaf, chunk);
1864                 btrfs_release_path(chunk_root, path);
1865
1866                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1867                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1868                                                    found_key.objectid,
1869                                                    found_key.offset);
1870                         if (ret == -ENOSPC)
1871                                 failed++;
1872                         else if (ret)
1873                                 BUG();
1874                 }
1875
1876                 if (found_key.offset == 0)
1877                         break;
1878                 key.offset = found_key.offset - 1;
1879         }
1880         ret = 0;
1881         if (failed && !retried) {
1882                 failed = 0;
1883                 retried = true;
1884                 goto again;
1885         } else if (failed && retried) {
1886                 WARN_ON(1);
1887                 ret = -ENOSPC;
1888         }
1889 error:
1890         btrfs_free_path(path);
1891         return ret;
1892 }
1893
1894 static u64 div_factor(u64 num, int factor)
1895 {
1896         if (factor == 10)
1897                 return num;
1898         num *= factor;
1899         do_div(num, 10);
1900         return num;
1901 }
1902
1903 int btrfs_balance(struct btrfs_root *dev_root)
1904 {
1905         int ret;
1906         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1907         struct btrfs_device *device;
1908         u64 old_size;
1909         u64 size_to_free;
1910         struct btrfs_path *path;
1911         struct btrfs_key key;
1912         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1913         struct btrfs_trans_handle *trans;
1914         struct btrfs_key found_key;
1915
1916         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1917                 return -EROFS;
1918
1919         mutex_lock(&dev_root->fs_info->volume_mutex);
1920         dev_root = dev_root->fs_info->dev_root;
1921
1922         /* step one make some room on all the devices */
1923         list_for_each_entry(device, devices, dev_list) {
1924                 old_size = device->total_bytes;
1925                 size_to_free = div_factor(old_size, 1);
1926                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1927                 if (!device->writeable ||
1928                     device->total_bytes - device->bytes_used > size_to_free)
1929                         continue;
1930
1931                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1932                 if (ret == -ENOSPC)
1933                         break;
1934                 BUG_ON(ret);
1935
1936                 trans = btrfs_start_transaction(dev_root, 0);
1937                 BUG_ON(!trans);
1938
1939                 ret = btrfs_grow_device(trans, device, old_size);
1940                 BUG_ON(ret);
1941
1942                 btrfs_end_transaction(trans, dev_root);
1943         }
1944
1945         /* step two, relocate all the chunks */
1946         path = btrfs_alloc_path();
1947         BUG_ON(!path);
1948
1949         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1950         key.offset = (u64)-1;
1951         key.type = BTRFS_CHUNK_ITEM_KEY;
1952
1953         while (1) {
1954                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1955                 if (ret < 0)
1956                         goto error;
1957
1958                 /*
1959                  * this shouldn't happen, it means the last relocate
1960                  * failed
1961                  */
1962                 if (ret == 0)
1963                         break;
1964
1965                 ret = btrfs_previous_item(chunk_root, path, 0,
1966                                           BTRFS_CHUNK_ITEM_KEY);
1967                 if (ret)
1968                         break;
1969
1970                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1971                                       path->slots[0]);
1972                 if (found_key.objectid != key.objectid)
1973                         break;
1974
1975                 /* chunk zero is special */
1976                 if (found_key.offset == 0)
1977                         break;
1978
1979                 btrfs_release_path(chunk_root, path);
1980                 ret = btrfs_relocate_chunk(chunk_root,
1981                                            chunk_root->root_key.objectid,
1982                                            found_key.objectid,
1983                                            found_key.offset);
1984                 BUG_ON(ret && ret != -ENOSPC);
1985                 key.offset = found_key.offset - 1;
1986         }
1987         ret = 0;
1988 error:
1989         btrfs_free_path(path);
1990         mutex_unlock(&dev_root->fs_info->volume_mutex);
1991         return ret;
1992 }
1993
1994 /*
1995  * shrinking a device means finding all of the device extents past
1996  * the new size, and then following the back refs to the chunks.
1997  * The chunk relocation code actually frees the device extent
1998  */
1999 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2000 {
2001         struct btrfs_trans_handle *trans;
2002         struct btrfs_root *root = device->dev_root;
2003         struct btrfs_dev_extent *dev_extent = NULL;
2004         struct btrfs_path *path;
2005         u64 length;
2006         u64 chunk_tree;
2007         u64 chunk_objectid;
2008         u64 chunk_offset;
2009         int ret;
2010         int slot;
2011         int failed = 0;
2012         bool retried = false;
2013         struct extent_buffer *l;
2014         struct btrfs_key key;
2015         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2016         u64 old_total = btrfs_super_total_bytes(super_copy);
2017         u64 old_size = device->total_bytes;
2018         u64 diff = device->total_bytes - new_size;
2019
2020         if (new_size >= device->total_bytes)
2021                 return -EINVAL;
2022
2023         path = btrfs_alloc_path();
2024         if (!path)
2025                 return -ENOMEM;
2026
2027         path->reada = 2;
2028
2029         lock_chunks(root);
2030
2031         device->total_bytes = new_size;
2032         if (device->writeable)
2033                 device->fs_devices->total_rw_bytes -= diff;
2034         unlock_chunks(root);
2035
2036 again:
2037         key.objectid = device->devid;
2038         key.offset = (u64)-1;
2039         key.type = BTRFS_DEV_EXTENT_KEY;
2040
2041         while (1) {
2042                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2043                 if (ret < 0)
2044                         goto done;
2045
2046                 ret = btrfs_previous_item(root, path, 0, key.type);
2047                 if (ret < 0)
2048                         goto done;
2049                 if (ret) {
2050                         ret = 0;
2051                         btrfs_release_path(root, path);
2052                         break;
2053                 }
2054
2055                 l = path->nodes[0];
2056                 slot = path->slots[0];
2057                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2058
2059                 if (key.objectid != device->devid) {
2060                         btrfs_release_path(root, path);
2061                         break;
2062                 }
2063
2064                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2065                 length = btrfs_dev_extent_length(l, dev_extent);
2066
2067                 if (key.offset + length <= new_size) {
2068                         btrfs_release_path(root, path);
2069                         break;
2070                 }
2071
2072                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2073                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2074                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2075                 btrfs_release_path(root, path);
2076
2077                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2078                                            chunk_offset);
2079                 if (ret && ret != -ENOSPC)
2080                         goto done;
2081                 if (ret == -ENOSPC)
2082                         failed++;
2083                 key.offset -= 1;
2084         }
2085
2086         if (failed && !retried) {
2087                 failed = 0;
2088                 retried = true;
2089                 goto again;
2090         } else if (failed && retried) {
2091                 ret = -ENOSPC;
2092                 lock_chunks(root);
2093
2094                 device->total_bytes = old_size;
2095                 if (device->writeable)
2096                         device->fs_devices->total_rw_bytes += diff;
2097                 unlock_chunks(root);
2098                 goto done;
2099         }
2100
2101         /* Shrinking succeeded, else we would be at "done". */
2102         trans = btrfs_start_transaction(root, 0);
2103         lock_chunks(root);
2104
2105         device->disk_total_bytes = new_size;
2106         /* Now btrfs_update_device() will change the on-disk size. */
2107         ret = btrfs_update_device(trans, device);
2108         if (ret) {
2109                 unlock_chunks(root);
2110                 btrfs_end_transaction(trans, root);
2111                 goto done;
2112         }
2113         WARN_ON(diff > old_total);
2114         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2115         unlock_chunks(root);
2116         btrfs_end_transaction(trans, root);
2117 done:
2118         btrfs_free_path(path);
2119         return ret;
2120 }
2121
2122 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2123                            struct btrfs_root *root,
2124                            struct btrfs_key *key,
2125                            struct btrfs_chunk *chunk, int item_size)
2126 {
2127         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2128         struct btrfs_disk_key disk_key;
2129         u32 array_size;
2130         u8 *ptr;
2131
2132         array_size = btrfs_super_sys_array_size(super_copy);
2133         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2134                 return -EFBIG;
2135
2136         ptr = super_copy->sys_chunk_array + array_size;
2137         btrfs_cpu_key_to_disk(&disk_key, key);
2138         memcpy(ptr, &disk_key, sizeof(disk_key));
2139         ptr += sizeof(disk_key);
2140         memcpy(ptr, chunk, item_size);
2141         item_size += sizeof(disk_key);
2142         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2143         return 0;
2144 }
2145
2146 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2147                                         int num_stripes, int sub_stripes)
2148 {
2149         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2150                 return calc_size;
2151         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2152                 return calc_size * (num_stripes / sub_stripes);
2153         else
2154                 return calc_size * num_stripes;
2155 }
2156
2157 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2158                                struct btrfs_root *extent_root,
2159                                struct map_lookup **map_ret,
2160                                u64 *num_bytes, u64 *stripe_size,
2161                                u64 start, u64 type)
2162 {
2163         struct btrfs_fs_info *info = extent_root->fs_info;
2164         struct btrfs_device *device = NULL;
2165         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2166         struct list_head *cur;
2167         struct map_lookup *map = NULL;
2168         struct extent_map_tree *em_tree;
2169         struct extent_map *em;
2170         struct list_head private_devs;
2171         int min_stripe_size = 1 * 1024 * 1024;
2172         u64 calc_size = 1024 * 1024 * 1024;
2173         u64 max_chunk_size = calc_size;
2174         u64 min_free;
2175         u64 avail;
2176         u64 max_avail = 0;
2177         u64 dev_offset;
2178         int num_stripes = 1;
2179         int min_stripes = 1;
2180         int sub_stripes = 0;
2181         int looped = 0;
2182         int ret;
2183         int index;
2184         int stripe_len = 64 * 1024;
2185
2186         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2187             (type & BTRFS_BLOCK_GROUP_DUP)) {
2188                 WARN_ON(1);
2189                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2190         }
2191         if (list_empty(&fs_devices->alloc_list))
2192                 return -ENOSPC;
2193
2194         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2195                 num_stripes = fs_devices->rw_devices;
2196                 min_stripes = 2;
2197         }
2198         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2199                 num_stripes = 2;
2200                 min_stripes = 2;
2201         }
2202         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2203                 if (fs_devices->rw_devices < 2)
2204                         return -ENOSPC;
2205                 num_stripes = 2;
2206                 min_stripes = 2;
2207         }
2208         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2209                 num_stripes = fs_devices->rw_devices;
2210                 if (num_stripes < 4)
2211                         return -ENOSPC;
2212                 num_stripes &= ~(u32)1;
2213                 sub_stripes = 2;
2214                 min_stripes = 4;
2215         }
2216
2217         if (type & BTRFS_BLOCK_GROUP_DATA) {
2218                 max_chunk_size = 10 * calc_size;
2219                 min_stripe_size = 64 * 1024 * 1024;
2220         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2221                 max_chunk_size = 256 * 1024 * 1024;
2222                 min_stripe_size = 32 * 1024 * 1024;
2223         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2224                 calc_size = 8 * 1024 * 1024;
2225                 max_chunk_size = calc_size * 2;
2226                 min_stripe_size = 1 * 1024 * 1024;
2227         }
2228
2229         /* we don't want a chunk larger than 10% of writeable space */
2230         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2231                              max_chunk_size);
2232
2233 again:
2234         max_avail = 0;
2235         if (!map || map->num_stripes != num_stripes) {
2236                 kfree(map);
2237                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2238                 if (!map)
2239                         return -ENOMEM;
2240                 map->num_stripes = num_stripes;
2241         }
2242
2243         if (calc_size * num_stripes > max_chunk_size) {
2244                 calc_size = max_chunk_size;
2245                 do_div(calc_size, num_stripes);
2246                 do_div(calc_size, stripe_len);
2247                 calc_size *= stripe_len;
2248         }
2249
2250         /* we don't want tiny stripes */
2251         if (!looped)
2252                 calc_size = max_t(u64, min_stripe_size, calc_size);
2253
2254         /*
2255          * we're about to do_div by the stripe_len so lets make sure
2256          * we end up with something bigger than a stripe
2257          */
2258         calc_size = max_t(u64, calc_size, stripe_len * 4);
2259
2260         do_div(calc_size, stripe_len);
2261         calc_size *= stripe_len;
2262
2263         cur = fs_devices->alloc_list.next;
2264         index = 0;
2265
2266         if (type & BTRFS_BLOCK_GROUP_DUP)
2267                 min_free = calc_size * 2;
2268         else
2269                 min_free = calc_size;
2270
2271         /*
2272          * we add 1MB because we never use the first 1MB of the device, unless
2273          * we've looped, then we are likely allocating the maximum amount of
2274          * space left already
2275          */
2276         if (!looped)
2277                 min_free += 1024 * 1024;
2278
2279         INIT_LIST_HEAD(&private_devs);
2280         while (index < num_stripes) {
2281                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2282                 BUG_ON(!device->writeable);
2283                 if (device->total_bytes > device->bytes_used)
2284                         avail = device->total_bytes - device->bytes_used;
2285                 else
2286                         avail = 0;
2287                 cur = cur->next;
2288
2289                 if (device->in_fs_metadata && avail >= min_free) {
2290                         ret = find_free_dev_extent(trans, device,
2291                                                    min_free, &dev_offset,
2292                                                    &max_avail);
2293                         if (ret == 0) {
2294                                 list_move_tail(&device->dev_alloc_list,
2295                                                &private_devs);
2296                                 map->stripes[index].dev = device;
2297                                 map->stripes[index].physical = dev_offset;
2298                                 index++;
2299                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2300                                         map->stripes[index].dev = device;
2301                                         map->stripes[index].physical =
2302                                                 dev_offset + calc_size;
2303                                         index++;
2304                                 }
2305                         }
2306                 } else if (device->in_fs_metadata && avail > max_avail)
2307                         max_avail = avail;
2308                 if (cur == &fs_devices->alloc_list)
2309                         break;
2310         }
2311         list_splice(&private_devs, &fs_devices->alloc_list);
2312         if (index < num_stripes) {
2313                 if (index >= min_stripes) {
2314                         num_stripes = index;
2315                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2316                                 num_stripes /= sub_stripes;
2317                                 num_stripes *= sub_stripes;
2318                         }
2319                         looped = 1;
2320                         goto again;
2321                 }
2322                 if (!looped && max_avail > 0) {
2323                         looped = 1;
2324                         calc_size = max_avail;
2325                         goto again;
2326                 }
2327                 kfree(map);
2328                 return -ENOSPC;
2329         }
2330         map->sector_size = extent_root->sectorsize;
2331         map->stripe_len = stripe_len;
2332         map->io_align = stripe_len;
2333         map->io_width = stripe_len;
2334         map->type = type;
2335         map->num_stripes = num_stripes;
2336         map->sub_stripes = sub_stripes;
2337
2338         *map_ret = map;
2339         *stripe_size = calc_size;
2340         *num_bytes = chunk_bytes_by_type(type, calc_size,
2341                                          num_stripes, sub_stripes);
2342
2343         em = alloc_extent_map(GFP_NOFS);
2344         if (!em) {
2345                 kfree(map);
2346                 return -ENOMEM;
2347         }
2348         em->bdev = (struct block_device *)map;
2349         em->start = start;
2350         em->len = *num_bytes;
2351         em->block_start = 0;
2352         em->block_len = em->len;
2353
2354         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2355         write_lock(&em_tree->lock);
2356         ret = add_extent_mapping(em_tree, em);
2357         write_unlock(&em_tree->lock);
2358         BUG_ON(ret);
2359         free_extent_map(em);
2360
2361         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2362                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2363                                      start, *num_bytes);
2364         BUG_ON(ret);
2365
2366         index = 0;
2367         while (index < map->num_stripes) {
2368                 device = map->stripes[index].dev;
2369                 dev_offset = map->stripes[index].physical;
2370
2371                 ret = btrfs_alloc_dev_extent(trans, device,
2372                                 info->chunk_root->root_key.objectid,
2373                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2374                                 start, dev_offset, calc_size);
2375                 BUG_ON(ret);
2376                 index++;
2377         }
2378
2379         return 0;
2380 }
2381
2382 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2383                                 struct btrfs_root *extent_root,
2384                                 struct map_lookup *map, u64 chunk_offset,
2385                                 u64 chunk_size, u64 stripe_size)
2386 {
2387         u64 dev_offset;
2388         struct btrfs_key key;
2389         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2390         struct btrfs_device *device;
2391         struct btrfs_chunk *chunk;
2392         struct btrfs_stripe *stripe;
2393         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2394         int index = 0;
2395         int ret;
2396
2397         chunk = kzalloc(item_size, GFP_NOFS);
2398         if (!chunk)
2399                 return -ENOMEM;
2400
2401         index = 0;
2402         while (index < map->num_stripes) {
2403                 device = map->stripes[index].dev;
2404                 device->bytes_used += stripe_size;
2405                 ret = btrfs_update_device(trans, device);
2406                 BUG_ON(ret);
2407                 index++;
2408         }
2409
2410         index = 0;
2411         stripe = &chunk->stripe;
2412         while (index < map->num_stripes) {
2413                 device = map->stripes[index].dev;
2414                 dev_offset = map->stripes[index].physical;
2415
2416                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2417                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2418                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2419                 stripe++;
2420                 index++;
2421         }
2422
2423         btrfs_set_stack_chunk_length(chunk, chunk_size);
2424         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2425         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2426         btrfs_set_stack_chunk_type(chunk, map->type);
2427         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2428         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2429         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2430         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2431         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2432
2433         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2434         key.type = BTRFS_CHUNK_ITEM_KEY;
2435         key.offset = chunk_offset;
2436
2437         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2438         BUG_ON(ret);
2439
2440         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2441                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2442                                              item_size);
2443                 BUG_ON(ret);
2444         }
2445         kfree(chunk);
2446         return 0;
2447 }
2448
2449 /*
2450  * Chunk allocation falls into two parts. The first part does works
2451  * that make the new allocated chunk useable, but not do any operation
2452  * that modifies the chunk tree. The second part does the works that
2453  * require modifying the chunk tree. This division is important for the
2454  * bootstrap process of adding storage to a seed btrfs.
2455  */
2456 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2457                       struct btrfs_root *extent_root, u64 type)
2458 {
2459         u64 chunk_offset;
2460         u64 chunk_size;
2461         u64 stripe_size;
2462         struct map_lookup *map;
2463         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2464         int ret;
2465
2466         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2467                               &chunk_offset);
2468         if (ret)
2469                 return ret;
2470
2471         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2472                                   &stripe_size, chunk_offset, type);
2473         if (ret)
2474                 return ret;
2475
2476         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2477                                    chunk_size, stripe_size);
2478         BUG_ON(ret);
2479         return 0;
2480 }
2481
2482 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2483                                          struct btrfs_root *root,
2484                                          struct btrfs_device *device)
2485 {
2486         u64 chunk_offset;
2487         u64 sys_chunk_offset;
2488         u64 chunk_size;
2489         u64 sys_chunk_size;
2490         u64 stripe_size;
2491         u64 sys_stripe_size;
2492         u64 alloc_profile;
2493         struct map_lookup *map;
2494         struct map_lookup *sys_map;
2495         struct btrfs_fs_info *fs_info = root->fs_info;
2496         struct btrfs_root *extent_root = fs_info->extent_root;
2497         int ret;
2498
2499         ret = find_next_chunk(fs_info->chunk_root,
2500                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2501         BUG_ON(ret);
2502
2503         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2504                         (fs_info->metadata_alloc_profile &
2505                          fs_info->avail_metadata_alloc_bits);
2506         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2507
2508         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2509                                   &stripe_size, chunk_offset, alloc_profile);
2510         BUG_ON(ret);
2511
2512         sys_chunk_offset = chunk_offset + chunk_size;
2513
2514         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2515                         (fs_info->system_alloc_profile &
2516                          fs_info->avail_system_alloc_bits);
2517         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2518
2519         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2520                                   &sys_chunk_size, &sys_stripe_size,
2521                                   sys_chunk_offset, alloc_profile);
2522         BUG_ON(ret);
2523
2524         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2525         BUG_ON(ret);
2526
2527         /*
2528          * Modifying chunk tree needs allocating new blocks from both
2529          * system block group and metadata block group. So we only can
2530          * do operations require modifying the chunk tree after both
2531          * block groups were created.
2532          */
2533         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2534                                    chunk_size, stripe_size);
2535         BUG_ON(ret);
2536
2537         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2538                                    sys_chunk_offset, sys_chunk_size,
2539                                    sys_stripe_size);
2540         BUG_ON(ret);
2541         return 0;
2542 }
2543
2544 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2545 {
2546         struct extent_map *em;
2547         struct map_lookup *map;
2548         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2549         int readonly = 0;
2550         int i;
2551
2552         read_lock(&map_tree->map_tree.lock);
2553         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2554         read_unlock(&map_tree->map_tree.lock);
2555         if (!em)
2556                 return 1;
2557
2558         if (btrfs_test_opt(root, DEGRADED)) {
2559                 free_extent_map(em);
2560                 return 0;
2561         }
2562
2563         map = (struct map_lookup *)em->bdev;
2564         for (i = 0; i < map->num_stripes; i++) {
2565                 if (!map->stripes[i].dev->writeable) {
2566                         readonly = 1;
2567                         break;
2568                 }
2569         }
2570         free_extent_map(em);
2571         return readonly;
2572 }
2573
2574 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2575 {
2576         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2577 }
2578
2579 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2580 {
2581         struct extent_map *em;
2582
2583         while (1) {
2584                 write_lock(&tree->map_tree.lock);
2585                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2586                 if (em)
2587                         remove_extent_mapping(&tree->map_tree, em);
2588                 write_unlock(&tree->map_tree.lock);
2589                 if (!em)
2590                         break;
2591                 kfree(em->bdev);
2592                 /* once for us */
2593                 free_extent_map(em);
2594                 /* once for the tree */
2595                 free_extent_map(em);
2596         }
2597 }
2598
2599 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2600 {
2601         struct extent_map *em;
2602         struct map_lookup *map;
2603         struct extent_map_tree *em_tree = &map_tree->map_tree;
2604         int ret;
2605
2606         read_lock(&em_tree->lock);
2607         em = lookup_extent_mapping(em_tree, logical, len);
2608         read_unlock(&em_tree->lock);
2609         BUG_ON(!em);
2610
2611         BUG_ON(em->start > logical || em->start + em->len < logical);
2612         map = (struct map_lookup *)em->bdev;
2613         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2614                 ret = map->num_stripes;
2615         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2616                 ret = map->sub_stripes;
2617         else
2618                 ret = 1;
2619         free_extent_map(em);
2620         return ret;
2621 }
2622
2623 static int find_live_mirror(struct map_lookup *map, int first, int num,
2624                             int optimal)
2625 {
2626         int i;
2627         if (map->stripes[optimal].dev->bdev)
2628                 return optimal;
2629         for (i = first; i < first + num; i++) {
2630                 if (map->stripes[i].dev->bdev)
2631                         return i;
2632         }
2633         /* we couldn't find one that doesn't fail.  Just return something
2634          * and the io error handling code will clean up eventually
2635          */
2636         return optimal;
2637 }
2638
2639 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2640                              u64 logical, u64 *length,
2641                              struct btrfs_multi_bio **multi_ret,
2642                              int mirror_num, struct page *unplug_page)
2643 {
2644         struct extent_map *em;
2645         struct map_lookup *map;
2646         struct extent_map_tree *em_tree = &map_tree->map_tree;
2647         u64 offset;
2648         u64 stripe_offset;
2649         u64 stripe_nr;
2650         int stripes_allocated = 8;
2651         int stripes_required = 1;
2652         int stripe_index;
2653         int i;
2654         int num_stripes;
2655         int max_errors = 0;
2656         struct btrfs_multi_bio *multi = NULL;
2657
2658         if (multi_ret && !(rw & REQ_WRITE))
2659                 stripes_allocated = 1;
2660 again:
2661         if (multi_ret) {
2662                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2663                                 GFP_NOFS);
2664                 if (!multi)
2665                         return -ENOMEM;
2666
2667                 atomic_set(&multi->error, 0);
2668         }
2669
2670         read_lock(&em_tree->lock);
2671         em = lookup_extent_mapping(em_tree, logical, *length);
2672         read_unlock(&em_tree->lock);
2673
2674         if (!em && unplug_page) {
2675                 kfree(multi);
2676                 return 0;
2677         }
2678
2679         if (!em) {
2680                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2681                        (unsigned long long)logical,
2682                        (unsigned long long)*length);
2683                 BUG();
2684         }
2685
2686         BUG_ON(em->start > logical || em->start + em->len < logical);
2687         map = (struct map_lookup *)em->bdev;
2688         offset = logical - em->start;
2689
2690         if (mirror_num > map->num_stripes)
2691                 mirror_num = 0;
2692
2693         /* if our multi bio struct is too small, back off and try again */
2694         if (rw & REQ_WRITE) {
2695                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2696                                  BTRFS_BLOCK_GROUP_DUP)) {
2697                         stripes_required = map->num_stripes;
2698                         max_errors = 1;
2699                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2700                         stripes_required = map->sub_stripes;
2701                         max_errors = 1;
2702                 }
2703         }
2704         if (multi_ret && (rw & REQ_WRITE) &&
2705             stripes_allocated < stripes_required) {
2706                 stripes_allocated = map->num_stripes;
2707                 free_extent_map(em);
2708                 kfree(multi);
2709                 goto again;
2710         }
2711         stripe_nr = offset;
2712         /*
2713          * stripe_nr counts the total number of stripes we have to stride
2714          * to get to this block
2715          */
2716         do_div(stripe_nr, map->stripe_len);
2717
2718         stripe_offset = stripe_nr * map->stripe_len;
2719         BUG_ON(offset < stripe_offset);
2720
2721         /* stripe_offset is the offset of this block in its stripe*/
2722         stripe_offset = offset - stripe_offset;
2723
2724         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2725                          BTRFS_BLOCK_GROUP_RAID10 |
2726                          BTRFS_BLOCK_GROUP_DUP)) {
2727                 /* we limit the length of each bio to what fits in a stripe */
2728                 *length = min_t(u64, em->len - offset,
2729                               map->stripe_len - stripe_offset);
2730         } else {
2731                 *length = em->len - offset;
2732         }
2733
2734         if (!multi_ret && !unplug_page)
2735                 goto out;
2736
2737         num_stripes = 1;
2738         stripe_index = 0;
2739         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2740                 if (unplug_page || (rw & REQ_WRITE))
2741                         num_stripes = map->num_stripes;
2742                 else if (mirror_num)
2743                         stripe_index = mirror_num - 1;
2744                 else {
2745                         stripe_index = find_live_mirror(map, 0,
2746                                             map->num_stripes,
2747                                             current->pid % map->num_stripes);
2748                 }
2749
2750         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2751                 if (rw & REQ_WRITE)
2752                         num_stripes = map->num_stripes;
2753                 else if (mirror_num)
2754                         stripe_index = mirror_num - 1;
2755
2756         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2757                 int factor = map->num_stripes / map->sub_stripes;
2758
2759                 stripe_index = do_div(stripe_nr, factor);
2760                 stripe_index *= map->sub_stripes;
2761
2762                 if (unplug_page || (rw & REQ_WRITE))
2763                         num_stripes = map->sub_stripes;
2764                 else if (mirror_num)
2765                         stripe_index += mirror_num - 1;
2766                 else {
2767                         stripe_index = find_live_mirror(map, stripe_index,
2768                                               map->sub_stripes, stripe_index +
2769                                               current->pid % map->sub_stripes);
2770                 }
2771         } else {
2772                 /*
2773                  * after this do_div call, stripe_nr is the number of stripes
2774                  * on this device we have to walk to find the data, and
2775                  * stripe_index is the number of our device in the stripe array
2776                  */
2777                 stripe_index = do_div(stripe_nr, map->num_stripes);
2778         }
2779         BUG_ON(stripe_index >= map->num_stripes);
2780
2781         for (i = 0; i < num_stripes; i++) {
2782                 if (unplug_page) {
2783                         struct btrfs_device *device;
2784                         struct backing_dev_info *bdi;
2785
2786                         device = map->stripes[stripe_index].dev;
2787                         if (device->bdev) {
2788                                 bdi = blk_get_backing_dev_info(device->bdev);
2789                                 if (bdi->unplug_io_fn)
2790                                         bdi->unplug_io_fn(bdi, unplug_page);
2791                         }
2792                 } else {
2793                         multi->stripes[i].physical =
2794                                 map->stripes[stripe_index].physical +
2795                                 stripe_offset + stripe_nr * map->stripe_len;
2796                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2797                 }
2798                 stripe_index++;
2799         }
2800         if (multi_ret) {
2801                 *multi_ret = multi;
2802                 multi->num_stripes = num_stripes;
2803                 multi->max_errors = max_errors;
2804         }
2805 out:
2806         free_extent_map(em);
2807         return 0;
2808 }
2809
2810 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2811                       u64 logical, u64 *length,
2812                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2813 {
2814         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2815                                  mirror_num, NULL);
2816 }
2817
2818 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2819                      u64 chunk_start, u64 physical, u64 devid,
2820                      u64 **logical, int *naddrs, int *stripe_len)
2821 {
2822         struct extent_map_tree *em_tree = &map_tree->map_tree;
2823         struct extent_map *em;
2824         struct map_lookup *map;
2825         u64 *buf;
2826         u64 bytenr;
2827         u64 length;
2828         u64 stripe_nr;
2829         int i, j, nr = 0;
2830
2831         read_lock(&em_tree->lock);
2832         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2833         read_unlock(&em_tree->lock);
2834
2835         BUG_ON(!em || em->start != chunk_start);
2836         map = (struct map_lookup *)em->bdev;
2837
2838         length = em->len;
2839         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2840                 do_div(length, map->num_stripes / map->sub_stripes);
2841         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2842                 do_div(length, map->num_stripes);
2843
2844         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2845         BUG_ON(!buf);
2846
2847         for (i = 0; i < map->num_stripes; i++) {
2848                 if (devid && map->stripes[i].dev->devid != devid)
2849                         continue;
2850                 if (map->stripes[i].physical > physical ||
2851                     map->stripes[i].physical + length <= physical)
2852                         continue;
2853
2854                 stripe_nr = physical - map->stripes[i].physical;
2855                 do_div(stripe_nr, map->stripe_len);
2856
2857                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2858                         stripe_nr = stripe_nr * map->num_stripes + i;
2859                         do_div(stripe_nr, map->sub_stripes);
2860                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2861                         stripe_nr = stripe_nr * map->num_stripes + i;
2862                 }
2863                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2864                 WARN_ON(nr >= map->num_stripes);
2865                 for (j = 0; j < nr; j++) {
2866                         if (buf[j] == bytenr)
2867                                 break;
2868                 }
2869                 if (j == nr) {
2870                         WARN_ON(nr >= map->num_stripes);
2871                         buf[nr++] = bytenr;
2872                 }
2873         }
2874
2875         *logical = buf;
2876         *naddrs = nr;
2877         *stripe_len = map->stripe_len;
2878
2879         free_extent_map(em);
2880         return 0;
2881 }
2882
2883 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2884                       u64 logical, struct page *page)
2885 {
2886         u64 length = PAGE_CACHE_SIZE;
2887         return __btrfs_map_block(map_tree, READ, logical, &length,
2888                                  NULL, 0, page);
2889 }
2890
2891 static void end_bio_multi_stripe(struct bio *bio, int err)
2892 {
2893         struct btrfs_multi_bio *multi = bio->bi_private;
2894         int is_orig_bio = 0;
2895
2896         if (err)
2897                 atomic_inc(&multi->error);
2898
2899         if (bio == multi->orig_bio)
2900                 is_orig_bio = 1;
2901
2902         if (atomic_dec_and_test(&multi->stripes_pending)) {
2903                 if (!is_orig_bio) {
2904                         bio_put(bio);
2905                         bio = multi->orig_bio;
2906                 }
2907                 bio->bi_private = multi->private;
2908                 bio->bi_end_io = multi->end_io;
2909                 /* only send an error to the higher layers if it is
2910                  * beyond the tolerance of the multi-bio
2911                  */
2912                 if (atomic_read(&multi->error) > multi->max_errors) {
2913                         err = -EIO;
2914                 } else if (err) {
2915                         /*
2916                          * this bio is actually up to date, we didn't
2917                          * go over the max number of errors
2918                          */
2919                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2920                         err = 0;
2921                 }
2922                 kfree(multi);
2923
2924                 bio_endio(bio, err);
2925         } else if (!is_orig_bio) {
2926                 bio_put(bio);
2927         }
2928 }
2929
2930 struct async_sched {
2931         struct bio *bio;
2932         int rw;
2933         struct btrfs_fs_info *info;
2934         struct btrfs_work work;
2935 };
2936
2937 /*
2938  * see run_scheduled_bios for a description of why bios are collected for
2939  * async submit.
2940  *
2941  * This will add one bio to the pending list for a device and make sure
2942  * the work struct is scheduled.
2943  */
2944 static noinline int schedule_bio(struct btrfs_root *root,
2945                                  struct btrfs_device *device,
2946                                  int rw, struct bio *bio)
2947 {
2948         int should_queue = 1;
2949         struct btrfs_pending_bios *pending_bios;
2950
2951         /* don't bother with additional async steps for reads, right now */
2952         if (!(rw & REQ_WRITE)) {
2953                 bio_get(bio);
2954                 submit_bio(rw, bio);
2955                 bio_put(bio);
2956                 return 0;
2957         }
2958
2959         /*
2960          * nr_async_bios allows us to reliably return congestion to the
2961          * higher layers.  Otherwise, the async bio makes it appear we have
2962          * made progress against dirty pages when we've really just put it
2963          * on a queue for later
2964          */
2965         atomic_inc(&root->fs_info->nr_async_bios);
2966         WARN_ON(bio->bi_next);
2967         bio->bi_next = NULL;
2968         bio->bi_rw |= rw;
2969
2970         spin_lock(&device->io_lock);
2971         if (bio->bi_rw & REQ_SYNC)
2972                 pending_bios = &device->pending_sync_bios;
2973         else
2974                 pending_bios = &device->pending_bios;
2975
2976         if (pending_bios->tail)
2977                 pending_bios->tail->bi_next = bio;
2978
2979         pending_bios->tail = bio;
2980         if (!pending_bios->head)
2981                 pending_bios->head = bio;
2982         if (device->running_pending)
2983                 should_queue = 0;
2984
2985         spin_unlock(&device->io_lock);
2986
2987         if (should_queue)
2988                 btrfs_queue_worker(&root->fs_info->submit_workers,
2989                                    &device->work);
2990         return 0;
2991 }
2992
2993 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2994                   int mirror_num, int async_submit)
2995 {
2996         struct btrfs_mapping_tree *map_tree;
2997         struct btrfs_device *dev;
2998         struct bio *first_bio = bio;
2999         u64 logical = (u64)bio->bi_sector << 9;
3000         u64 length = 0;
3001         u64 map_length;
3002         struct btrfs_multi_bio *multi = NULL;
3003         int ret;
3004         int dev_nr = 0;
3005         int total_devs = 1;
3006
3007         length = bio->bi_size;
3008         map_tree = &root->fs_info->mapping_tree;
3009         map_length = length;
3010
3011         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3012                               mirror_num);
3013         BUG_ON(ret);
3014
3015         total_devs = multi->num_stripes;
3016         if (map_length < length) {
3017                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3018                        "len %llu\n", (unsigned long long)logical,
3019                        (unsigned long long)length,
3020                        (unsigned long long)map_length);
3021                 BUG();
3022         }
3023         multi->end_io = first_bio->bi_end_io;
3024         multi->private = first_bio->bi_private;
3025         multi->orig_bio = first_bio;
3026         atomic_set(&multi->stripes_pending, multi->num_stripes);
3027
3028         while (dev_nr < total_devs) {
3029                 if (total_devs > 1) {
3030                         if (dev_nr < total_devs - 1) {
3031                                 bio = bio_clone(first_bio, GFP_NOFS);
3032                                 BUG_ON(!bio);
3033                         } else {
3034                                 bio = first_bio;
3035                         }
3036                         bio->bi_private = multi;
3037                         bio->bi_end_io = end_bio_multi_stripe;
3038                 }
3039                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3040                 dev = multi->stripes[dev_nr].dev;
3041                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3042                         bio->bi_bdev = dev->bdev;
3043                         if (async_submit)
3044                                 schedule_bio(root, dev, rw, bio);
3045                         else
3046                                 submit_bio(rw, bio);
3047                 } else {
3048                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3049                         bio->bi_sector = logical >> 9;
3050                         bio_endio(bio, -EIO);
3051                 }
3052                 dev_nr++;
3053         }
3054         if (total_devs == 1)
3055                 kfree(multi);
3056         return 0;
3057 }
3058
3059 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3060                                        u8 *uuid, u8 *fsid)
3061 {
3062         struct btrfs_device *device;
3063         struct btrfs_fs_devices *cur_devices;
3064
3065         cur_devices = root->fs_info->fs_devices;
3066         while (cur_devices) {
3067                 if (!fsid ||
3068                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3069                         device = __find_device(&cur_devices->devices,
3070                                                devid, uuid);
3071                         if (device)
3072                                 return device;
3073                 }
3074                 cur_devices = cur_devices->seed;
3075         }
3076         return NULL;
3077 }
3078
3079 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3080                                             u64 devid, u8 *dev_uuid)
3081 {
3082         struct btrfs_device *device;
3083         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3084
3085         device = kzalloc(sizeof(*device), GFP_NOFS);
3086         if (!device)
3087                 return NULL;
3088         list_add(&device->dev_list,
3089                  &fs_devices->devices);
3090         device->dev_root = root->fs_info->dev_root;
3091         device->devid = devid;
3092         device->work.func = pending_bios_fn;
3093         device->fs_devices = fs_devices;
3094         device->missing = 1;
3095         fs_devices->num_devices++;
3096         fs_devices->missing_devices++;
3097         spin_lock_init(&device->io_lock);
3098         INIT_LIST_HEAD(&device->dev_alloc_list);
3099         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3100         return device;
3101 }
3102
3103 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3104                           struct extent_buffer *leaf,
3105                           struct btrfs_chunk *chunk)
3106 {
3107         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3108         struct map_lookup *map;
3109         struct extent_map *em;
3110         u64 logical;
3111         u64 length;
3112         u64 devid;
3113         u8 uuid[BTRFS_UUID_SIZE];
3114         int num_stripes;
3115         int ret;
3116         int i;
3117
3118         logical = key->offset;
3119         length = btrfs_chunk_length(leaf, chunk);
3120
3121         read_lock(&map_tree->map_tree.lock);
3122         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3123         read_unlock(&map_tree->map_tree.lock);
3124
3125         /* already mapped? */
3126         if (em && em->start <= logical && em->start + em->len > logical) {
3127                 free_extent_map(em);
3128                 return 0;
3129         } else if (em) {
3130                 free_extent_map(em);
3131         }
3132
3133         em = alloc_extent_map(GFP_NOFS);
3134         if (!em)
3135                 return -ENOMEM;
3136         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3137         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3138         if (!map) {
3139                 free_extent_map(em);
3140                 return -ENOMEM;
3141         }
3142
3143         em->bdev = (struct block_device *)map;
3144         em->start = logical;
3145         em->len = length;
3146         em->block_start = 0;
3147         em->block_len = em->len;
3148
3149         map->num_stripes = num_stripes;
3150         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3151         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3152         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3153         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3154         map->type = btrfs_chunk_type(leaf, chunk);
3155         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3156         for (i = 0; i < num_stripes; i++) {
3157                 map->stripes[i].physical =
3158                         btrfs_stripe_offset_nr(leaf, chunk, i);
3159                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3160                 read_extent_buffer(leaf, uuid, (unsigned long)
3161                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3162                                    BTRFS_UUID_SIZE);
3163                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3164                                                         NULL);
3165                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3166                         kfree(map);
3167                         free_extent_map(em);
3168                         return -EIO;
3169                 }
3170                 if (!map->stripes[i].dev) {
3171                         map->stripes[i].dev =
3172                                 add_missing_dev(root, devid, uuid);
3173                         if (!map->stripes[i].dev) {
3174                                 kfree(map);
3175                                 free_extent_map(em);
3176                                 return -EIO;
3177                         }
3178                 }
3179                 map->stripes[i].dev->in_fs_metadata = 1;
3180         }
3181
3182         write_lock(&map_tree->map_tree.lock);
3183         ret = add_extent_mapping(&map_tree->map_tree, em);
3184         write_unlock(&map_tree->map_tree.lock);
3185         BUG_ON(ret);
3186         free_extent_map(em);
3187
3188         return 0;
3189 }
3190
3191 static int fill_device_from_item(struct extent_buffer *leaf,
3192                                  struct btrfs_dev_item *dev_item,
3193                                  struct btrfs_device *device)
3194 {
3195         unsigned long ptr;
3196
3197         device->devid = btrfs_device_id(leaf, dev_item);
3198         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3199         device->total_bytes = device->disk_total_bytes;
3200         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3201         device->type = btrfs_device_type(leaf, dev_item);
3202         device->io_align = btrfs_device_io_align(leaf, dev_item);
3203         device->io_width = btrfs_device_io_width(leaf, dev_item);
3204         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3205
3206         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3207         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3208
3209         return 0;
3210 }
3211
3212 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3213 {
3214         struct btrfs_fs_devices *fs_devices;
3215         int ret;
3216
3217         mutex_lock(&uuid_mutex);
3218
3219         fs_devices = root->fs_info->fs_devices->seed;
3220         while (fs_devices) {
3221                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3222                         ret = 0;
3223                         goto out;
3224                 }
3225                 fs_devices = fs_devices->seed;
3226         }
3227
3228         fs_devices = find_fsid(fsid);
3229         if (!fs_devices) {
3230                 ret = -ENOENT;
3231                 goto out;
3232         }
3233
3234         fs_devices = clone_fs_devices(fs_devices);
3235         if (IS_ERR(fs_devices)) {
3236                 ret = PTR_ERR(fs_devices);
3237                 goto out;
3238         }
3239
3240         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3241                                    root->fs_info->bdev_holder);
3242         if (ret)
3243                 goto out;
3244
3245         if (!fs_devices->seeding) {
3246                 __btrfs_close_devices(fs_devices);
3247                 free_fs_devices(fs_devices);
3248                 ret = -EINVAL;
3249                 goto out;
3250         }
3251
3252         fs_devices->seed = root->fs_info->fs_devices->seed;
3253         root->fs_info->fs_devices->seed = fs_devices;
3254 out:
3255         mutex_unlock(&uuid_mutex);
3256         return ret;
3257 }
3258
3259 static int read_one_dev(struct btrfs_root *root,
3260                         struct extent_buffer *leaf,
3261                         struct btrfs_dev_item *dev_item)
3262 {
3263         struct btrfs_device *device;
3264         u64 devid;
3265         int ret;
3266         u8 fs_uuid[BTRFS_UUID_SIZE];
3267         u8 dev_uuid[BTRFS_UUID_SIZE];
3268
3269         devid = btrfs_device_id(leaf, dev_item);
3270         read_extent_buffer(leaf, dev_uuid,
3271                            (unsigned long)btrfs_device_uuid(dev_item),
3272                            BTRFS_UUID_SIZE);
3273         read_extent_buffer(leaf, fs_uuid,
3274                            (unsigned long)btrfs_device_fsid(dev_item),
3275                            BTRFS_UUID_SIZE);
3276
3277         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3278                 ret = open_seed_devices(root, fs_uuid);
3279                 if (ret && !btrfs_test_opt(root, DEGRADED))
3280                         return ret;
3281         }
3282
3283         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3284         if (!device || !device->bdev) {
3285                 if (!btrfs_test_opt(root, DEGRADED))
3286                         return -EIO;
3287
3288                 if (!device) {
3289                         printk(KERN_WARNING "warning devid %llu missing\n",
3290                                (unsigned long long)devid);
3291                         device = add_missing_dev(root, devid, dev_uuid);
3292                         if (!device)
3293                                 return -ENOMEM;
3294                 } else if (!device->missing) {
3295                         /*
3296                          * this happens when a device that was properly setup
3297                          * in the device info lists suddenly goes bad.
3298                          * device->bdev is NULL, and so we have to set
3299                          * device->missing to one here
3300                          */
3301                         root->fs_info->fs_devices->missing_devices++;
3302                         device->missing = 1;
3303                 }
3304         }
3305
3306         if (device->fs_devices != root->fs_info->fs_devices) {
3307                 BUG_ON(device->writeable);
3308                 if (device->generation !=
3309                     btrfs_device_generation(leaf, dev_item))
3310                         return -EINVAL;
3311         }
3312
3313         fill_device_from_item(leaf, dev_item, device);
3314         device->dev_root = root->fs_info->dev_root;
3315         device->in_fs_metadata = 1;
3316         if (device->writeable)
3317                 device->fs_devices->total_rw_bytes += device->total_bytes;
3318         ret = 0;
3319         return ret;
3320 }
3321
3322 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3323 {
3324         struct btrfs_dev_item *dev_item;
3325
3326         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3327                                                      dev_item);
3328         return read_one_dev(root, buf, dev_item);
3329 }
3330
3331 int btrfs_read_sys_array(struct btrfs_root *root)
3332 {
3333         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3334         struct extent_buffer *sb;
3335         struct btrfs_disk_key *disk_key;
3336         struct btrfs_chunk *chunk;
3337         u8 *ptr;
3338         unsigned long sb_ptr;
3339         int ret = 0;
3340         u32 num_stripes;
3341         u32 array_size;
3342         u32 len = 0;
3343         u32 cur;
3344         struct btrfs_key key;
3345
3346         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3347                                           BTRFS_SUPER_INFO_SIZE);
3348         if (!sb)
3349                 return -ENOMEM;
3350         btrfs_set_buffer_uptodate(sb);
3351         btrfs_set_buffer_lockdep_class(sb, 0);
3352
3353         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3354         array_size = btrfs_super_sys_array_size(super_copy);
3355
3356         ptr = super_copy->sys_chunk_array;
3357         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3358         cur = 0;
3359
3360         while (cur < array_size) {
3361                 disk_key = (struct btrfs_disk_key *)ptr;
3362                 btrfs_disk_key_to_cpu(&key, disk_key);
3363
3364                 len = sizeof(*disk_key); ptr += len;
3365                 sb_ptr += len;
3366                 cur += len;
3367
3368                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3369                         chunk = (struct btrfs_chunk *)sb_ptr;
3370                         ret = read_one_chunk(root, &key, sb, chunk);
3371                         if (ret)
3372                                 break;
3373                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3374                         len = btrfs_chunk_item_size(num_stripes);
3375                 } else {
3376                         ret = -EIO;
3377                         break;
3378                 }
3379                 ptr += len;
3380                 sb_ptr += len;
3381                 cur += len;
3382         }
3383         free_extent_buffer(sb);
3384         return ret;
3385 }
3386
3387 int btrfs_read_chunk_tree(struct btrfs_root *root)
3388 {
3389         struct btrfs_path *path;
3390         struct extent_buffer *leaf;
3391         struct btrfs_key key;
3392         struct btrfs_key found_key;
3393         int ret;
3394         int slot;
3395
3396         root = root->fs_info->chunk_root;
3397
3398         path = btrfs_alloc_path();
3399         if (!path)
3400                 return -ENOMEM;
3401
3402         /* first we search for all of the device items, and then we
3403          * read in all of the chunk items.  This way we can create chunk
3404          * mappings that reference all of the devices that are afound
3405          */
3406         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3407         key.offset = 0;
3408         key.type = 0;
3409 again:
3410         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3411         if (ret < 0)
3412                 goto error;
3413         while (1) {
3414                 leaf = path->nodes[0];
3415                 slot = path->slots[0];
3416                 if (slot >= btrfs_header_nritems(leaf)) {
3417                         ret = btrfs_next_leaf(root, path);
3418                         if (ret == 0)
3419                                 continue;
3420                         if (ret < 0)
3421                                 goto error;
3422                         break;
3423                 }
3424                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3425                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3426                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3427                                 break;
3428                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3429                                 struct btrfs_dev_item *dev_item;
3430                                 dev_item = btrfs_item_ptr(leaf, slot,
3431                                                   struct btrfs_dev_item);
3432                                 ret = read_one_dev(root, leaf, dev_item);
3433                                 if (ret)
3434                                         goto error;
3435                         }
3436                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3437                         struct btrfs_chunk *chunk;
3438                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3439                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3440                         if (ret)
3441                                 goto error;
3442                 }
3443                 path->slots[0]++;
3444         }
3445         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3446                 key.objectid = 0;
3447                 btrfs_release_path(root, path);
3448                 goto again;
3449         }
3450         ret = 0;
3451 error:
3452         btrfs_free_path(path);
3453         return ret;
3454 }