btrfs: Don't BUG_ON alloc_path errors in find_next_chunk
[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 <linux/capability.h>
26 #include <asm/div64.h>
27 #include "compat.h"
28 #include "ctree.h"
29 #include "extent_map.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "print-tree.h"
33 #include "volumes.h"
34 #include "async-thread.h"
35
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37                                 struct btrfs_root *root,
38                                 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
41 static DEFINE_MUTEX(uuid_mutex);
42 static LIST_HEAD(fs_uuids);
43
44 static void lock_chunks(struct btrfs_root *root)
45 {
46         mutex_lock(&root->fs_info->chunk_mutex);
47 }
48
49 static void unlock_chunks(struct btrfs_root *root)
50 {
51         mutex_unlock(&root->fs_info->chunk_mutex);
52 }
53
54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
55 {
56         struct btrfs_device *device;
57         WARN_ON(fs_devices->opened);
58         while (!list_empty(&fs_devices->devices)) {
59                 device = list_entry(fs_devices->devices.next,
60                                     struct btrfs_device, dev_list);
61                 list_del(&device->dev_list);
62                 kfree(device->name);
63                 kfree(device);
64         }
65         kfree(fs_devices);
66 }
67
68 int btrfs_cleanup_fs_uuids(void)
69 {
70         struct btrfs_fs_devices *fs_devices;
71
72         while (!list_empty(&fs_uuids)) {
73                 fs_devices = list_entry(fs_uuids.next,
74                                         struct btrfs_fs_devices, list);
75                 list_del(&fs_devices->list);
76                 free_fs_devices(fs_devices);
77         }
78         return 0;
79 }
80
81 static noinline struct btrfs_device *__find_device(struct list_head *head,
82                                                    u64 devid, u8 *uuid)
83 {
84         struct btrfs_device *dev;
85
86         list_for_each_entry(dev, head, dev_list) {
87                 if (dev->devid == devid &&
88                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
89                         return dev;
90                 }
91         }
92         return NULL;
93 }
94
95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
96 {
97         struct btrfs_fs_devices *fs_devices;
98
99         list_for_each_entry(fs_devices, &fs_uuids, list) {
100                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
101                         return fs_devices;
102         }
103         return NULL;
104 }
105
106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
107                         struct bio *head, struct bio *tail)
108 {
109
110         struct bio *old_head;
111
112         old_head = pending_bios->head;
113         pending_bios->head = head;
114         if (pending_bios->tail)
115                 tail->bi_next = old_head;
116         else
117                 pending_bios->tail = tail;
118 }
119
120 /*
121  * we try to collect pending bios for a device so we don't get a large
122  * number of procs sending bios down to the same device.  This greatly
123  * improves the schedulers ability to collect and merge the bios.
124  *
125  * But, it also turns into a long list of bios to process and that is sure
126  * to eventually make the worker thread block.  The solution here is to
127  * make some progress and then put this work struct back at the end of
128  * the list if the block device is congested.  This way, multiple devices
129  * can make progress from a single worker thread.
130  */
131 static noinline int run_scheduled_bios(struct btrfs_device *device)
132 {
133         struct bio *pending;
134         struct backing_dev_info *bdi;
135         struct btrfs_fs_info *fs_info;
136         struct btrfs_pending_bios *pending_bios;
137         struct bio *tail;
138         struct bio *cur;
139         int again = 0;
140         unsigned long num_run;
141         unsigned long batch_run = 0;
142         unsigned long limit;
143         unsigned long last_waited = 0;
144         int force_reg = 0;
145         struct blk_plug plug;
146
147         /*
148          * this function runs all the bios we've collected for
149          * a particular device.  We don't want to wander off to
150          * another device without first sending all of these down.
151          * So, setup a plug here and finish it off before we return
152          */
153         blk_start_plug(&plug);
154
155         bdi = blk_get_backing_dev_info(device->bdev);
156         fs_info = device->dev_root->fs_info;
157         limit = btrfs_async_submit_limit(fs_info);
158         limit = limit * 2 / 3;
159
160 loop:
161         spin_lock(&device->io_lock);
162
163 loop_lock:
164         num_run = 0;
165
166         /* take all the bios off the list at once and process them
167          * later on (without the lock held).  But, remember the
168          * tail and other pointers so the bios can be properly reinserted
169          * into the list if we hit congestion
170          */
171         if (!force_reg && device->pending_sync_bios.head) {
172                 pending_bios = &device->pending_sync_bios;
173                 force_reg = 1;
174         } else {
175                 pending_bios = &device->pending_bios;
176                 force_reg = 0;
177         }
178
179         pending = pending_bios->head;
180         tail = pending_bios->tail;
181         WARN_ON(pending && !tail);
182
183         /*
184          * if pending was null this time around, no bios need processing
185          * at all and we can stop.  Otherwise it'll loop back up again
186          * and do an additional check so no bios are missed.
187          *
188          * device->running_pending is used to synchronize with the
189          * schedule_bio code.
190          */
191         if (device->pending_sync_bios.head == NULL &&
192             device->pending_bios.head == NULL) {
193                 again = 0;
194                 device->running_pending = 0;
195         } else {
196                 again = 1;
197                 device->running_pending = 1;
198         }
199
200         pending_bios->head = NULL;
201         pending_bios->tail = NULL;
202
203         spin_unlock(&device->io_lock);
204
205         while (pending) {
206
207                 rmb();
208                 /* we want to work on both lists, but do more bios on the
209                  * sync list than the regular list
210                  */
211                 if ((num_run > 32 &&
212                     pending_bios != &device->pending_sync_bios &&
213                     device->pending_sync_bios.head) ||
214                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
215                     device->pending_bios.head)) {
216                         spin_lock(&device->io_lock);
217                         requeue_list(pending_bios, pending, tail);
218                         goto loop_lock;
219                 }
220
221                 cur = pending;
222                 pending = pending->bi_next;
223                 cur->bi_next = NULL;
224                 atomic_dec(&fs_info->nr_async_bios);
225
226                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
227                     waitqueue_active(&fs_info->async_submit_wait))
228                         wake_up(&fs_info->async_submit_wait);
229
230                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
231
232                 submit_bio(cur->bi_rw, cur);
233                 num_run++;
234                 batch_run++;
235                 if (need_resched())
236                         cond_resched();
237
238                 /*
239                  * we made progress, there is more work to do and the bdi
240                  * is now congested.  Back off and let other work structs
241                  * run instead
242                  */
243                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
244                     fs_info->fs_devices->open_devices > 1) {
245                         struct io_context *ioc;
246
247                         ioc = current->io_context;
248
249                         /*
250                          * the main goal here is that we don't want to
251                          * block if we're going to be able to submit
252                          * more requests without blocking.
253                          *
254                          * This code does two great things, it pokes into
255                          * the elevator code from a filesystem _and_
256                          * it makes assumptions about how batching works.
257                          */
258                         if (ioc && ioc->nr_batch_requests > 0 &&
259                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
260                             (last_waited == 0 ||
261                              ioc->last_waited == last_waited)) {
262                                 /*
263                                  * we want to go through our batch of
264                                  * requests and stop.  So, we copy out
265                                  * the ioc->last_waited time and test
266                                  * against it before looping
267                                  */
268                                 last_waited = ioc->last_waited;
269                                 if (need_resched())
270                                         cond_resched();
271                                 continue;
272                         }
273                         spin_lock(&device->io_lock);
274                         requeue_list(pending_bios, pending, tail);
275                         device->running_pending = 1;
276
277                         spin_unlock(&device->io_lock);
278                         btrfs_requeue_work(&device->work);
279                         goto done;
280                 }
281         }
282
283         cond_resched();
284         if (again)
285                 goto loop;
286
287         spin_lock(&device->io_lock);
288         if (device->pending_bios.head || device->pending_sync_bios.head)
289                 goto loop_lock;
290         spin_unlock(&device->io_lock);
291
292 done:
293         blk_finish_plug(&plug);
294         return 0;
295 }
296
297 static void pending_bios_fn(struct btrfs_work *work)
298 {
299         struct btrfs_device *device;
300
301         device = container_of(work, struct btrfs_device, work);
302         run_scheduled_bios(device);
303 }
304
305 static noinline int device_list_add(const char *path,
306                            struct btrfs_super_block *disk_super,
307                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
308 {
309         struct btrfs_device *device;
310         struct btrfs_fs_devices *fs_devices;
311         u64 found_transid = btrfs_super_generation(disk_super);
312         char *name;
313
314         fs_devices = find_fsid(disk_super->fsid);
315         if (!fs_devices) {
316                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
317                 if (!fs_devices)
318                         return -ENOMEM;
319                 INIT_LIST_HEAD(&fs_devices->devices);
320                 INIT_LIST_HEAD(&fs_devices->alloc_list);
321                 list_add(&fs_devices->list, &fs_uuids);
322                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
323                 fs_devices->latest_devid = devid;
324                 fs_devices->latest_trans = found_transid;
325                 mutex_init(&fs_devices->device_list_mutex);
326                 device = NULL;
327         } else {
328                 device = __find_device(&fs_devices->devices, devid,
329                                        disk_super->dev_item.uuid);
330         }
331         if (!device) {
332                 if (fs_devices->opened)
333                         return -EBUSY;
334
335                 device = kzalloc(sizeof(*device), GFP_NOFS);
336                 if (!device) {
337                         /* we can safely leave the fs_devices entry around */
338                         return -ENOMEM;
339                 }
340                 device->devid = devid;
341                 device->work.func = pending_bios_fn;
342                 memcpy(device->uuid, disk_super->dev_item.uuid,
343                        BTRFS_UUID_SIZE);
344                 spin_lock_init(&device->io_lock);
345                 device->name = kstrdup(path, GFP_NOFS);
346                 if (!device->name) {
347                         kfree(device);
348                         return -ENOMEM;
349                 }
350                 INIT_LIST_HEAD(&device->dev_alloc_list);
351
352                 mutex_lock(&fs_devices->device_list_mutex);
353                 list_add_rcu(&device->dev_list, &fs_devices->devices);
354                 mutex_unlock(&fs_devices->device_list_mutex);
355
356                 device->fs_devices = fs_devices;
357                 fs_devices->num_devices++;
358         } else if (!device->name || strcmp(device->name, path)) {
359                 name = kstrdup(path, GFP_NOFS);
360                 if (!name)
361                         return -ENOMEM;
362                 kfree(device->name);
363                 device->name = name;
364                 if (device->missing) {
365                         fs_devices->missing_devices--;
366                         device->missing = 0;
367                 }
368         }
369
370         if (found_transid > fs_devices->latest_trans) {
371                 fs_devices->latest_devid = devid;
372                 fs_devices->latest_trans = found_transid;
373         }
374         *fs_devices_ret = fs_devices;
375         return 0;
376 }
377
378 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
379 {
380         struct btrfs_fs_devices *fs_devices;
381         struct btrfs_device *device;
382         struct btrfs_device *orig_dev;
383
384         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
385         if (!fs_devices)
386                 return ERR_PTR(-ENOMEM);
387
388         INIT_LIST_HEAD(&fs_devices->devices);
389         INIT_LIST_HEAD(&fs_devices->alloc_list);
390         INIT_LIST_HEAD(&fs_devices->list);
391         mutex_init(&fs_devices->device_list_mutex);
392         fs_devices->latest_devid = orig->latest_devid;
393         fs_devices->latest_trans = orig->latest_trans;
394         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
395
396         /* We have held the volume lock, it is safe to get the devices. */
397         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
398                 device = kzalloc(sizeof(*device), GFP_NOFS);
399                 if (!device)
400                         goto error;
401
402                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
403                 if (!device->name) {
404                         kfree(device);
405                         goto error;
406                 }
407
408                 device->devid = orig_dev->devid;
409                 device->work.func = pending_bios_fn;
410                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
411                 spin_lock_init(&device->io_lock);
412                 INIT_LIST_HEAD(&device->dev_list);
413                 INIT_LIST_HEAD(&device->dev_alloc_list);
414
415                 list_add(&device->dev_list, &fs_devices->devices);
416                 device->fs_devices = fs_devices;
417                 fs_devices->num_devices++;
418         }
419         return fs_devices;
420 error:
421         free_fs_devices(fs_devices);
422         return ERR_PTR(-ENOMEM);
423 }
424
425 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
426 {
427         struct btrfs_device *device, *next;
428
429         mutex_lock(&uuid_mutex);
430 again:
431         /* This is the initialized path, it is safe to release the devices. */
432         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
433                 if (device->in_fs_metadata)
434                         continue;
435
436                 if (device->bdev) {
437                         blkdev_put(device->bdev, device->mode);
438                         device->bdev = NULL;
439                         fs_devices->open_devices--;
440                 }
441                 if (device->writeable) {
442                         list_del_init(&device->dev_alloc_list);
443                         device->writeable = 0;
444                         fs_devices->rw_devices--;
445                 }
446                 list_del_init(&device->dev_list);
447                 fs_devices->num_devices--;
448                 kfree(device->name);
449                 kfree(device);
450         }
451
452         if (fs_devices->seed) {
453                 fs_devices = fs_devices->seed;
454                 goto again;
455         }
456
457         mutex_unlock(&uuid_mutex);
458         return 0;
459 }
460
461 static void __free_device(struct work_struct *work)
462 {
463         struct btrfs_device *device;
464
465         device = container_of(work, struct btrfs_device, rcu_work);
466
467         if (device->bdev)
468                 blkdev_put(device->bdev, device->mode);
469
470         kfree(device->name);
471         kfree(device);
472 }
473
474 static void free_device(struct rcu_head *head)
475 {
476         struct btrfs_device *device;
477
478         device = container_of(head, struct btrfs_device, rcu);
479
480         INIT_WORK(&device->rcu_work, __free_device);
481         schedule_work(&device->rcu_work);
482 }
483
484 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
485 {
486         struct btrfs_device *device;
487
488         if (--fs_devices->opened > 0)
489                 return 0;
490
491         mutex_lock(&fs_devices->device_list_mutex);
492         list_for_each_entry(device, &fs_devices->devices, dev_list) {
493                 struct btrfs_device *new_device;
494
495                 if (device->bdev)
496                         fs_devices->open_devices--;
497
498                 if (device->writeable) {
499                         list_del_init(&device->dev_alloc_list);
500                         fs_devices->rw_devices--;
501                 }
502
503                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
504                 BUG_ON(!new_device);
505                 memcpy(new_device, device, sizeof(*new_device));
506                 new_device->name = kstrdup(device->name, GFP_NOFS);
507                 BUG_ON(device->name && !new_device->name);
508                 new_device->bdev = NULL;
509                 new_device->writeable = 0;
510                 new_device->in_fs_metadata = 0;
511                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
512
513                 call_rcu(&device->rcu, free_device);
514         }
515         mutex_unlock(&fs_devices->device_list_mutex);
516
517         WARN_ON(fs_devices->open_devices);
518         WARN_ON(fs_devices->rw_devices);
519         fs_devices->opened = 0;
520         fs_devices->seeding = 0;
521
522         return 0;
523 }
524
525 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
526 {
527         struct btrfs_fs_devices *seed_devices = NULL;
528         int ret;
529
530         mutex_lock(&uuid_mutex);
531         ret = __btrfs_close_devices(fs_devices);
532         if (!fs_devices->opened) {
533                 seed_devices = fs_devices->seed;
534                 fs_devices->seed = NULL;
535         }
536         mutex_unlock(&uuid_mutex);
537
538         while (seed_devices) {
539                 fs_devices = seed_devices;
540                 seed_devices = fs_devices->seed;
541                 __btrfs_close_devices(fs_devices);
542                 free_fs_devices(fs_devices);
543         }
544         return ret;
545 }
546
547 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
548                                 fmode_t flags, void *holder)
549 {
550         struct block_device *bdev;
551         struct list_head *head = &fs_devices->devices;
552         struct btrfs_device *device;
553         struct block_device *latest_bdev = NULL;
554         struct buffer_head *bh;
555         struct btrfs_super_block *disk_super;
556         u64 latest_devid = 0;
557         u64 latest_transid = 0;
558         u64 devid;
559         int seeding = 1;
560         int ret = 0;
561
562         flags |= FMODE_EXCL;
563
564         list_for_each_entry(device, head, dev_list) {
565                 if (device->bdev)
566                         continue;
567                 if (!device->name)
568                         continue;
569
570                 bdev = blkdev_get_by_path(device->name, flags, holder);
571                 if (IS_ERR(bdev)) {
572                         printk(KERN_INFO "open %s failed\n", device->name);
573                         goto error;
574                 }
575                 set_blocksize(bdev, 4096);
576
577                 bh = btrfs_read_dev_super(bdev);
578                 if (!bh) {
579                         ret = -EINVAL;
580                         goto error_close;
581                 }
582
583                 disk_super = (struct btrfs_super_block *)bh->b_data;
584                 devid = btrfs_stack_device_id(&disk_super->dev_item);
585                 if (devid != device->devid)
586                         goto error_brelse;
587
588                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
589                            BTRFS_UUID_SIZE))
590                         goto error_brelse;
591
592                 device->generation = btrfs_super_generation(disk_super);
593                 if (!latest_transid || device->generation > latest_transid) {
594                         latest_devid = devid;
595                         latest_transid = device->generation;
596                         latest_bdev = bdev;
597                 }
598
599                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
600                         device->writeable = 0;
601                 } else {
602                         device->writeable = !bdev_read_only(bdev);
603                         seeding = 0;
604                 }
605
606                 device->bdev = bdev;
607                 device->in_fs_metadata = 0;
608                 device->mode = flags;
609
610                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
611                         fs_devices->rotating = 1;
612
613                 fs_devices->open_devices++;
614                 if (device->writeable) {
615                         fs_devices->rw_devices++;
616                         list_add(&device->dev_alloc_list,
617                                  &fs_devices->alloc_list);
618                 }
619                 brelse(bh);
620                 continue;
621
622 error_brelse:
623                 brelse(bh);
624 error_close:
625                 blkdev_put(bdev, flags);
626 error:
627                 continue;
628         }
629         if (fs_devices->open_devices == 0) {
630                 ret = -EIO;
631                 goto out;
632         }
633         fs_devices->seeding = seeding;
634         fs_devices->opened = 1;
635         fs_devices->latest_bdev = latest_bdev;
636         fs_devices->latest_devid = latest_devid;
637         fs_devices->latest_trans = latest_transid;
638         fs_devices->total_rw_bytes = 0;
639 out:
640         return ret;
641 }
642
643 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
644                        fmode_t flags, void *holder)
645 {
646         int ret;
647
648         mutex_lock(&uuid_mutex);
649         if (fs_devices->opened) {
650                 fs_devices->opened++;
651                 ret = 0;
652         } else {
653                 ret = __btrfs_open_devices(fs_devices, flags, holder);
654         }
655         mutex_unlock(&uuid_mutex);
656         return ret;
657 }
658
659 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
660                           struct btrfs_fs_devices **fs_devices_ret)
661 {
662         struct btrfs_super_block *disk_super;
663         struct block_device *bdev;
664         struct buffer_head *bh;
665         int ret;
666         u64 devid;
667         u64 transid;
668
669         mutex_lock(&uuid_mutex);
670
671         flags |= FMODE_EXCL;
672         bdev = blkdev_get_by_path(path, flags, holder);
673
674         if (IS_ERR(bdev)) {
675                 ret = PTR_ERR(bdev);
676                 goto error;
677         }
678
679         ret = set_blocksize(bdev, 4096);
680         if (ret)
681                 goto error_close;
682         bh = btrfs_read_dev_super(bdev);
683         if (!bh) {
684                 ret = -EINVAL;
685                 goto error_close;
686         }
687         disk_super = (struct btrfs_super_block *)bh->b_data;
688         devid = btrfs_stack_device_id(&disk_super->dev_item);
689         transid = btrfs_super_generation(disk_super);
690         if (disk_super->label[0])
691                 printk(KERN_INFO "device label %s ", disk_super->label);
692         else
693                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
694         printk(KERN_CONT "devid %llu transid %llu %s\n",
695                (unsigned long long)devid, (unsigned long long)transid, path);
696         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
697
698         brelse(bh);
699 error_close:
700         blkdev_put(bdev, flags);
701 error:
702         mutex_unlock(&uuid_mutex);
703         return ret;
704 }
705
706 /* helper to account the used device space in the range */
707 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
708                                    u64 end, u64 *length)
709 {
710         struct btrfs_key key;
711         struct btrfs_root *root = device->dev_root;
712         struct btrfs_dev_extent *dev_extent;
713         struct btrfs_path *path;
714         u64 extent_end;
715         int ret;
716         int slot;
717         struct extent_buffer *l;
718
719         *length = 0;
720
721         if (start >= device->total_bytes)
722                 return 0;
723
724         path = btrfs_alloc_path();
725         if (!path)
726                 return -ENOMEM;
727         path->reada = 2;
728
729         key.objectid = device->devid;
730         key.offset = start;
731         key.type = BTRFS_DEV_EXTENT_KEY;
732
733         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
734         if (ret < 0)
735                 goto out;
736         if (ret > 0) {
737                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
738                 if (ret < 0)
739                         goto out;
740         }
741
742         while (1) {
743                 l = path->nodes[0];
744                 slot = path->slots[0];
745                 if (slot >= btrfs_header_nritems(l)) {
746                         ret = btrfs_next_leaf(root, path);
747                         if (ret == 0)
748                                 continue;
749                         if (ret < 0)
750                                 goto out;
751
752                         break;
753                 }
754                 btrfs_item_key_to_cpu(l, &key, slot);
755
756                 if (key.objectid < device->devid)
757                         goto next;
758
759                 if (key.objectid > device->devid)
760                         break;
761
762                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
763                         goto next;
764
765                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
766                 extent_end = key.offset + btrfs_dev_extent_length(l,
767                                                                   dev_extent);
768                 if (key.offset <= start && extent_end > end) {
769                         *length = end - start + 1;
770                         break;
771                 } else if (key.offset <= start && extent_end > start)
772                         *length += extent_end - start;
773                 else if (key.offset > start && extent_end <= end)
774                         *length += extent_end - key.offset;
775                 else if (key.offset > start && key.offset <= end) {
776                         *length += end - key.offset + 1;
777                         break;
778                 } else if (key.offset > end)
779                         break;
780
781 next:
782                 path->slots[0]++;
783         }
784         ret = 0;
785 out:
786         btrfs_free_path(path);
787         return ret;
788 }
789
790 /*
791  * find_free_dev_extent - find free space in the specified device
792  * @trans:      transaction handler
793  * @device:     the device which we search the free space in
794  * @num_bytes:  the size of the free space that we need
795  * @start:      store the start of the free space.
796  * @len:        the size of the free space. that we find, or the size of the max
797  *              free space if we don't find suitable free space
798  *
799  * this uses a pretty simple search, the expectation is that it is
800  * called very infrequently and that a given device has a small number
801  * of extents
802  *
803  * @start is used to store the start of the free space if we find. But if we
804  * don't find suitable free space, it will be used to store the start position
805  * of the max free space.
806  *
807  * @len is used to store the size of the free space that we find.
808  * But if we don't find suitable free space, it is used to store the size of
809  * the max free space.
810  */
811 int find_free_dev_extent(struct btrfs_trans_handle *trans,
812                          struct btrfs_device *device, u64 num_bytes,
813                          u64 *start, u64 *len)
814 {
815         struct btrfs_key key;
816         struct btrfs_root *root = device->dev_root;
817         struct btrfs_dev_extent *dev_extent;
818         struct btrfs_path *path;
819         u64 hole_size;
820         u64 max_hole_start;
821         u64 max_hole_size;
822         u64 extent_end;
823         u64 search_start;
824         u64 search_end = device->total_bytes;
825         int ret;
826         int slot;
827         struct extent_buffer *l;
828
829         /* FIXME use last free of some kind */
830
831         /* we don't want to overwrite the superblock on the drive,
832          * so we make sure to start at an offset of at least 1MB
833          */
834         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
835
836         max_hole_start = search_start;
837         max_hole_size = 0;
838
839         if (search_start >= search_end) {
840                 ret = -ENOSPC;
841                 goto error;
842         }
843
844         path = btrfs_alloc_path();
845         if (!path) {
846                 ret = -ENOMEM;
847                 goto error;
848         }
849         path->reada = 2;
850
851         key.objectid = device->devid;
852         key.offset = search_start;
853         key.type = BTRFS_DEV_EXTENT_KEY;
854
855         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
856         if (ret < 0)
857                 goto out;
858         if (ret > 0) {
859                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
860                 if (ret < 0)
861                         goto out;
862         }
863
864         while (1) {
865                 l = path->nodes[0];
866                 slot = path->slots[0];
867                 if (slot >= btrfs_header_nritems(l)) {
868                         ret = btrfs_next_leaf(root, path);
869                         if (ret == 0)
870                                 continue;
871                         if (ret < 0)
872                                 goto out;
873
874                         break;
875                 }
876                 btrfs_item_key_to_cpu(l, &key, slot);
877
878                 if (key.objectid < device->devid)
879                         goto next;
880
881                 if (key.objectid > device->devid)
882                         break;
883
884                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
885                         goto next;
886
887                 if (key.offset > search_start) {
888                         hole_size = key.offset - search_start;
889
890                         if (hole_size > max_hole_size) {
891                                 max_hole_start = search_start;
892                                 max_hole_size = hole_size;
893                         }
894
895                         /*
896                          * If this free space is greater than which we need,
897                          * it must be the max free space that we have found
898                          * until now, so max_hole_start must point to the start
899                          * of this free space and the length of this free space
900                          * is stored in max_hole_size. Thus, we return
901                          * max_hole_start and max_hole_size and go back to the
902                          * caller.
903                          */
904                         if (hole_size >= num_bytes) {
905                                 ret = 0;
906                                 goto out;
907                         }
908                 }
909
910                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
911                 extent_end = key.offset + btrfs_dev_extent_length(l,
912                                                                   dev_extent);
913                 if (extent_end > search_start)
914                         search_start = extent_end;
915 next:
916                 path->slots[0]++;
917                 cond_resched();
918         }
919
920         hole_size = search_end- search_start;
921         if (hole_size > max_hole_size) {
922                 max_hole_start = search_start;
923                 max_hole_size = hole_size;
924         }
925
926         /* See above. */
927         if (hole_size < num_bytes)
928                 ret = -ENOSPC;
929         else
930                 ret = 0;
931
932 out:
933         btrfs_free_path(path);
934 error:
935         *start = max_hole_start;
936         if (len)
937                 *len = max_hole_size;
938         return ret;
939 }
940
941 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
942                           struct btrfs_device *device,
943                           u64 start)
944 {
945         int ret;
946         struct btrfs_path *path;
947         struct btrfs_root *root = device->dev_root;
948         struct btrfs_key key;
949         struct btrfs_key found_key;
950         struct extent_buffer *leaf = NULL;
951         struct btrfs_dev_extent *extent = NULL;
952
953         path = btrfs_alloc_path();
954         if (!path)
955                 return -ENOMEM;
956
957         key.objectid = device->devid;
958         key.offset = start;
959         key.type = BTRFS_DEV_EXTENT_KEY;
960
961         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
962         if (ret > 0) {
963                 ret = btrfs_previous_item(root, path, key.objectid,
964                                           BTRFS_DEV_EXTENT_KEY);
965                 if (ret)
966                         goto out;
967                 leaf = path->nodes[0];
968                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
969                 extent = btrfs_item_ptr(leaf, path->slots[0],
970                                         struct btrfs_dev_extent);
971                 BUG_ON(found_key.offset > start || found_key.offset +
972                        btrfs_dev_extent_length(leaf, extent) < start);
973         } else if (ret == 0) {
974                 leaf = path->nodes[0];
975                 extent = btrfs_item_ptr(leaf, path->slots[0],
976                                         struct btrfs_dev_extent);
977         }
978         BUG_ON(ret);
979
980         if (device->bytes_used > 0)
981                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
982         ret = btrfs_del_item(trans, root, path);
983
984 out:
985         btrfs_free_path(path);
986         return ret;
987 }
988
989 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
990                            struct btrfs_device *device,
991                            u64 chunk_tree, u64 chunk_objectid,
992                            u64 chunk_offset, u64 start, u64 num_bytes)
993 {
994         int ret;
995         struct btrfs_path *path;
996         struct btrfs_root *root = device->dev_root;
997         struct btrfs_dev_extent *extent;
998         struct extent_buffer *leaf;
999         struct btrfs_key key;
1000
1001         WARN_ON(!device->in_fs_metadata);
1002         path = btrfs_alloc_path();
1003         if (!path)
1004                 return -ENOMEM;
1005
1006         key.objectid = device->devid;
1007         key.offset = start;
1008         key.type = BTRFS_DEV_EXTENT_KEY;
1009         ret = btrfs_insert_empty_item(trans, root, path, &key,
1010                                       sizeof(*extent));
1011         BUG_ON(ret);
1012
1013         leaf = path->nodes[0];
1014         extent = btrfs_item_ptr(leaf, path->slots[0],
1015                                 struct btrfs_dev_extent);
1016         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1017         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1018         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1019
1020         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1021                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1022                     BTRFS_UUID_SIZE);
1023
1024         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1025         btrfs_mark_buffer_dirty(leaf);
1026         btrfs_free_path(path);
1027         return ret;
1028 }
1029
1030 static noinline int find_next_chunk(struct btrfs_root *root,
1031                                     u64 objectid, u64 *offset)
1032 {
1033         struct btrfs_path *path;
1034         int ret;
1035         struct btrfs_key key;
1036         struct btrfs_chunk *chunk;
1037         struct btrfs_key found_key;
1038
1039         path = btrfs_alloc_path();
1040         if (!path)
1041                 return -ENOMEM;
1042
1043         key.objectid = objectid;
1044         key.offset = (u64)-1;
1045         key.type = BTRFS_CHUNK_ITEM_KEY;
1046
1047         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1048         if (ret < 0)
1049                 goto error;
1050
1051         BUG_ON(ret == 0);
1052
1053         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1054         if (ret) {
1055                 *offset = 0;
1056         } else {
1057                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1058                                       path->slots[0]);
1059                 if (found_key.objectid != objectid)
1060                         *offset = 0;
1061                 else {
1062                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1063                                                struct btrfs_chunk);
1064                         *offset = found_key.offset +
1065                                 btrfs_chunk_length(path->nodes[0], chunk);
1066                 }
1067         }
1068         ret = 0;
1069 error:
1070         btrfs_free_path(path);
1071         return ret;
1072 }
1073
1074 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1075 {
1076         int ret;
1077         struct btrfs_key key;
1078         struct btrfs_key found_key;
1079         struct btrfs_path *path;
1080
1081         root = root->fs_info->chunk_root;
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086
1087         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1088         key.type = BTRFS_DEV_ITEM_KEY;
1089         key.offset = (u64)-1;
1090
1091         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1092         if (ret < 0)
1093                 goto error;
1094
1095         BUG_ON(ret == 0);
1096
1097         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1098                                   BTRFS_DEV_ITEM_KEY);
1099         if (ret) {
1100                 *objectid = 1;
1101         } else {
1102                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1103                                       path->slots[0]);
1104                 *objectid = found_key.offset + 1;
1105         }
1106         ret = 0;
1107 error:
1108         btrfs_free_path(path);
1109         return ret;
1110 }
1111
1112 /*
1113  * the device information is stored in the chunk root
1114  * the btrfs_device struct should be fully filled in
1115  */
1116 int btrfs_add_device(struct btrfs_trans_handle *trans,
1117                      struct btrfs_root *root,
1118                      struct btrfs_device *device)
1119 {
1120         int ret;
1121         struct btrfs_path *path;
1122         struct btrfs_dev_item *dev_item;
1123         struct extent_buffer *leaf;
1124         struct btrfs_key key;
1125         unsigned long ptr;
1126
1127         root = root->fs_info->chunk_root;
1128
1129         path = btrfs_alloc_path();
1130         if (!path)
1131                 return -ENOMEM;
1132
1133         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1134         key.type = BTRFS_DEV_ITEM_KEY;
1135         key.offset = device->devid;
1136
1137         ret = btrfs_insert_empty_item(trans, root, path, &key,
1138                                       sizeof(*dev_item));
1139         if (ret)
1140                 goto out;
1141
1142         leaf = path->nodes[0];
1143         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1144
1145         btrfs_set_device_id(leaf, dev_item, device->devid);
1146         btrfs_set_device_generation(leaf, dev_item, 0);
1147         btrfs_set_device_type(leaf, dev_item, device->type);
1148         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1149         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1150         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1151         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1152         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1153         btrfs_set_device_group(leaf, dev_item, 0);
1154         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1155         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1156         btrfs_set_device_start_offset(leaf, dev_item, 0);
1157
1158         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1159         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1160         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1161         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1162         btrfs_mark_buffer_dirty(leaf);
1163
1164         ret = 0;
1165 out:
1166         btrfs_free_path(path);
1167         return ret;
1168 }
1169
1170 static int btrfs_rm_dev_item(struct btrfs_root *root,
1171                              struct btrfs_device *device)
1172 {
1173         int ret;
1174         struct btrfs_path *path;
1175         struct btrfs_key key;
1176         struct btrfs_trans_handle *trans;
1177
1178         root = root->fs_info->chunk_root;
1179
1180         path = btrfs_alloc_path();
1181         if (!path)
1182                 return -ENOMEM;
1183
1184         trans = btrfs_start_transaction(root, 0);
1185         if (IS_ERR(trans)) {
1186                 btrfs_free_path(path);
1187                 return PTR_ERR(trans);
1188         }
1189         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1190         key.type = BTRFS_DEV_ITEM_KEY;
1191         key.offset = device->devid;
1192         lock_chunks(root);
1193
1194         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1195         if (ret < 0)
1196                 goto out;
1197
1198         if (ret > 0) {
1199                 ret = -ENOENT;
1200                 goto out;
1201         }
1202
1203         ret = btrfs_del_item(trans, root, path);
1204         if (ret)
1205                 goto out;
1206 out:
1207         btrfs_free_path(path);
1208         unlock_chunks(root);
1209         btrfs_commit_transaction(trans, root);
1210         return ret;
1211 }
1212
1213 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1214 {
1215         struct btrfs_device *device;
1216         struct btrfs_device *next_device;
1217         struct block_device *bdev;
1218         struct buffer_head *bh = NULL;
1219         struct btrfs_super_block *disk_super;
1220         struct btrfs_fs_devices *cur_devices;
1221         u64 all_avail;
1222         u64 devid;
1223         u64 num_devices;
1224         u8 *dev_uuid;
1225         int ret = 0;
1226         bool clear_super = false;
1227
1228         mutex_lock(&uuid_mutex);
1229         mutex_lock(&root->fs_info->volume_mutex);
1230
1231         all_avail = root->fs_info->avail_data_alloc_bits |
1232                 root->fs_info->avail_system_alloc_bits |
1233                 root->fs_info->avail_metadata_alloc_bits;
1234
1235         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1236             root->fs_info->fs_devices->num_devices <= 4) {
1237                 printk(KERN_ERR "btrfs: unable to go below four devices "
1238                        "on raid10\n");
1239                 ret = -EINVAL;
1240                 goto out;
1241         }
1242
1243         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1244             root->fs_info->fs_devices->num_devices <= 2) {
1245                 printk(KERN_ERR "btrfs: unable to go below two "
1246                        "devices on raid1\n");
1247                 ret = -EINVAL;
1248                 goto out;
1249         }
1250
1251         if (strcmp(device_path, "missing") == 0) {
1252                 struct list_head *devices;
1253                 struct btrfs_device *tmp;
1254
1255                 device = NULL;
1256                 devices = &root->fs_info->fs_devices->devices;
1257                 /*
1258                  * It is safe to read the devices since the volume_mutex
1259                  * is held.
1260                  */
1261                 list_for_each_entry(tmp, devices, dev_list) {
1262                         if (tmp->in_fs_metadata && !tmp->bdev) {
1263                                 device = tmp;
1264                                 break;
1265                         }
1266                 }
1267                 bdev = NULL;
1268                 bh = NULL;
1269                 disk_super = NULL;
1270                 if (!device) {
1271                         printk(KERN_ERR "btrfs: no missing devices found to "
1272                                "remove\n");
1273                         goto out;
1274                 }
1275         } else {
1276                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1277                                           root->fs_info->bdev_holder);
1278                 if (IS_ERR(bdev)) {
1279                         ret = PTR_ERR(bdev);
1280                         goto out;
1281                 }
1282
1283                 set_blocksize(bdev, 4096);
1284                 bh = btrfs_read_dev_super(bdev);
1285                 if (!bh) {
1286                         ret = -EINVAL;
1287                         goto error_close;
1288                 }
1289                 disk_super = (struct btrfs_super_block *)bh->b_data;
1290                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1291                 dev_uuid = disk_super->dev_item.uuid;
1292                 device = btrfs_find_device(root, devid, dev_uuid,
1293                                            disk_super->fsid);
1294                 if (!device) {
1295                         ret = -ENOENT;
1296                         goto error_brelse;
1297                 }
1298         }
1299
1300         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1301                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1302                        "device\n");
1303                 ret = -EINVAL;
1304                 goto error_brelse;
1305         }
1306
1307         if (device->writeable) {
1308                 lock_chunks(root);
1309                 list_del_init(&device->dev_alloc_list);
1310                 unlock_chunks(root);
1311                 root->fs_info->fs_devices->rw_devices--;
1312                 clear_super = true;
1313         }
1314
1315         ret = btrfs_shrink_device(device, 0);
1316         if (ret)
1317                 goto error_undo;
1318
1319         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1320         if (ret)
1321                 goto error_undo;
1322
1323         device->in_fs_metadata = 0;
1324         btrfs_scrub_cancel_dev(root, device);
1325
1326         /*
1327          * the device list mutex makes sure that we don't change
1328          * the device list while someone else is writing out all
1329          * the device supers.
1330          */
1331
1332         cur_devices = device->fs_devices;
1333         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1334         list_del_rcu(&device->dev_list);
1335
1336         device->fs_devices->num_devices--;
1337
1338         if (device->missing)
1339                 root->fs_info->fs_devices->missing_devices--;
1340
1341         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1342                                  struct btrfs_device, dev_list);
1343         if (device->bdev == root->fs_info->sb->s_bdev)
1344                 root->fs_info->sb->s_bdev = next_device->bdev;
1345         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1346                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1347
1348         if (device->bdev)
1349                 device->fs_devices->open_devices--;
1350
1351         call_rcu(&device->rcu, free_device);
1352         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1353
1354         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1355         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1356
1357         if (cur_devices->open_devices == 0) {
1358                 struct btrfs_fs_devices *fs_devices;
1359                 fs_devices = root->fs_info->fs_devices;
1360                 while (fs_devices) {
1361                         if (fs_devices->seed == cur_devices)
1362                                 break;
1363                         fs_devices = fs_devices->seed;
1364                 }
1365                 fs_devices->seed = cur_devices->seed;
1366                 cur_devices->seed = NULL;
1367                 lock_chunks(root);
1368                 __btrfs_close_devices(cur_devices);
1369                 unlock_chunks(root);
1370                 free_fs_devices(cur_devices);
1371         }
1372
1373         /*
1374          * at this point, the device is zero sized.  We want to
1375          * remove it from the devices list and zero out the old super
1376          */
1377         if (clear_super) {
1378                 /* make sure this device isn't detected as part of
1379                  * the FS anymore
1380                  */
1381                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1382                 set_buffer_dirty(bh);
1383                 sync_dirty_buffer(bh);
1384         }
1385
1386         ret = 0;
1387
1388 error_brelse:
1389         brelse(bh);
1390 error_close:
1391         if (bdev)
1392                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1393 out:
1394         mutex_unlock(&root->fs_info->volume_mutex);
1395         mutex_unlock(&uuid_mutex);
1396         return ret;
1397 error_undo:
1398         if (device->writeable) {
1399                 lock_chunks(root);
1400                 list_add(&device->dev_alloc_list,
1401                          &root->fs_info->fs_devices->alloc_list);
1402                 unlock_chunks(root);
1403                 root->fs_info->fs_devices->rw_devices++;
1404         }
1405         goto error_brelse;
1406 }
1407
1408 /*
1409  * does all the dirty work required for changing file system's UUID.
1410  */
1411 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1412                                 struct btrfs_root *root)
1413 {
1414         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1415         struct btrfs_fs_devices *old_devices;
1416         struct btrfs_fs_devices *seed_devices;
1417         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1418         struct btrfs_device *device;
1419         u64 super_flags;
1420
1421         BUG_ON(!mutex_is_locked(&uuid_mutex));
1422         if (!fs_devices->seeding)
1423                 return -EINVAL;
1424
1425         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1426         if (!seed_devices)
1427                 return -ENOMEM;
1428
1429         old_devices = clone_fs_devices(fs_devices);
1430         if (IS_ERR(old_devices)) {
1431                 kfree(seed_devices);
1432                 return PTR_ERR(old_devices);
1433         }
1434
1435         list_add(&old_devices->list, &fs_uuids);
1436
1437         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1438         seed_devices->opened = 1;
1439         INIT_LIST_HEAD(&seed_devices->devices);
1440         INIT_LIST_HEAD(&seed_devices->alloc_list);
1441         mutex_init(&seed_devices->device_list_mutex);
1442
1443         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1444         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1445                               synchronize_rcu);
1446         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1447
1448         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1449         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1450                 device->fs_devices = seed_devices;
1451         }
1452
1453         fs_devices->seeding = 0;
1454         fs_devices->num_devices = 0;
1455         fs_devices->open_devices = 0;
1456         fs_devices->seed = seed_devices;
1457
1458         generate_random_uuid(fs_devices->fsid);
1459         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1460         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1461         super_flags = btrfs_super_flags(disk_super) &
1462                       ~BTRFS_SUPER_FLAG_SEEDING;
1463         btrfs_set_super_flags(disk_super, super_flags);
1464
1465         return 0;
1466 }
1467
1468 /*
1469  * strore the expected generation for seed devices in device items.
1470  */
1471 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1472                                struct btrfs_root *root)
1473 {
1474         struct btrfs_path *path;
1475         struct extent_buffer *leaf;
1476         struct btrfs_dev_item *dev_item;
1477         struct btrfs_device *device;
1478         struct btrfs_key key;
1479         u8 fs_uuid[BTRFS_UUID_SIZE];
1480         u8 dev_uuid[BTRFS_UUID_SIZE];
1481         u64 devid;
1482         int ret;
1483
1484         path = btrfs_alloc_path();
1485         if (!path)
1486                 return -ENOMEM;
1487
1488         root = root->fs_info->chunk_root;
1489         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1490         key.offset = 0;
1491         key.type = BTRFS_DEV_ITEM_KEY;
1492
1493         while (1) {
1494                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1495                 if (ret < 0)
1496                         goto error;
1497
1498                 leaf = path->nodes[0];
1499 next_slot:
1500                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1501                         ret = btrfs_next_leaf(root, path);
1502                         if (ret > 0)
1503                                 break;
1504                         if (ret < 0)
1505                                 goto error;
1506                         leaf = path->nodes[0];
1507                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1508                         btrfs_release_path(path);
1509                         continue;
1510                 }
1511
1512                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1513                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1514                     key.type != BTRFS_DEV_ITEM_KEY)
1515                         break;
1516
1517                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1518                                           struct btrfs_dev_item);
1519                 devid = btrfs_device_id(leaf, dev_item);
1520                 read_extent_buffer(leaf, dev_uuid,
1521                                    (unsigned long)btrfs_device_uuid(dev_item),
1522                                    BTRFS_UUID_SIZE);
1523                 read_extent_buffer(leaf, fs_uuid,
1524                                    (unsigned long)btrfs_device_fsid(dev_item),
1525                                    BTRFS_UUID_SIZE);
1526                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1527                 BUG_ON(!device);
1528
1529                 if (device->fs_devices->seeding) {
1530                         btrfs_set_device_generation(leaf, dev_item,
1531                                                     device->generation);
1532                         btrfs_mark_buffer_dirty(leaf);
1533                 }
1534
1535                 path->slots[0]++;
1536                 goto next_slot;
1537         }
1538         ret = 0;
1539 error:
1540         btrfs_free_path(path);
1541         return ret;
1542 }
1543
1544 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1545 {
1546         struct btrfs_trans_handle *trans;
1547         struct btrfs_device *device;
1548         struct block_device *bdev;
1549         struct list_head *devices;
1550         struct super_block *sb = root->fs_info->sb;
1551         u64 total_bytes;
1552         int seeding_dev = 0;
1553         int ret = 0;
1554
1555         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1556                 return -EINVAL;
1557
1558         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1559                                   root->fs_info->bdev_holder);
1560         if (IS_ERR(bdev))
1561                 return PTR_ERR(bdev);
1562
1563         if (root->fs_info->fs_devices->seeding) {
1564                 seeding_dev = 1;
1565                 down_write(&sb->s_umount);
1566                 mutex_lock(&uuid_mutex);
1567         }
1568
1569         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1570         mutex_lock(&root->fs_info->volume_mutex);
1571
1572         devices = &root->fs_info->fs_devices->devices;
1573         /*
1574          * we have the volume lock, so we don't need the extra
1575          * device list mutex while reading the list here.
1576          */
1577         list_for_each_entry(device, devices, dev_list) {
1578                 if (device->bdev == bdev) {
1579                         ret = -EEXIST;
1580                         goto error;
1581                 }
1582         }
1583
1584         device = kzalloc(sizeof(*device), GFP_NOFS);
1585         if (!device) {
1586                 /* we can safely leave the fs_devices entry around */
1587                 ret = -ENOMEM;
1588                 goto error;
1589         }
1590
1591         device->name = kstrdup(device_path, GFP_NOFS);
1592         if (!device->name) {
1593                 kfree(device);
1594                 ret = -ENOMEM;
1595                 goto error;
1596         }
1597
1598         ret = find_next_devid(root, &device->devid);
1599         if (ret) {
1600                 kfree(device->name);
1601                 kfree(device);
1602                 goto error;
1603         }
1604
1605         trans = btrfs_start_transaction(root, 0);
1606         if (IS_ERR(trans)) {
1607                 kfree(device->name);
1608                 kfree(device);
1609                 ret = PTR_ERR(trans);
1610                 goto error;
1611         }
1612
1613         lock_chunks(root);
1614
1615         device->writeable = 1;
1616         device->work.func = pending_bios_fn;
1617         generate_random_uuid(device->uuid);
1618         spin_lock_init(&device->io_lock);
1619         device->generation = trans->transid;
1620         device->io_width = root->sectorsize;
1621         device->io_align = root->sectorsize;
1622         device->sector_size = root->sectorsize;
1623         device->total_bytes = i_size_read(bdev->bd_inode);
1624         device->disk_total_bytes = device->total_bytes;
1625         device->dev_root = root->fs_info->dev_root;
1626         device->bdev = bdev;
1627         device->in_fs_metadata = 1;
1628         device->mode = FMODE_EXCL;
1629         set_blocksize(device->bdev, 4096);
1630
1631         if (seeding_dev) {
1632                 sb->s_flags &= ~MS_RDONLY;
1633                 ret = btrfs_prepare_sprout(trans, root);
1634                 BUG_ON(ret);
1635         }
1636
1637         device->fs_devices = root->fs_info->fs_devices;
1638
1639         /*
1640          * we don't want write_supers to jump in here with our device
1641          * half setup
1642          */
1643         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1644         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1645         list_add(&device->dev_alloc_list,
1646                  &root->fs_info->fs_devices->alloc_list);
1647         root->fs_info->fs_devices->num_devices++;
1648         root->fs_info->fs_devices->open_devices++;
1649         root->fs_info->fs_devices->rw_devices++;
1650         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1651
1652         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1653                 root->fs_info->fs_devices->rotating = 1;
1654
1655         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1656         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1657                                     total_bytes + device->total_bytes);
1658
1659         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1660         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1661                                     total_bytes + 1);
1662         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1663
1664         if (seeding_dev) {
1665                 ret = init_first_rw_device(trans, root, device);
1666                 BUG_ON(ret);
1667                 ret = btrfs_finish_sprout(trans, root);
1668                 BUG_ON(ret);
1669         } else {
1670                 ret = btrfs_add_device(trans, root, device);
1671         }
1672
1673         /*
1674          * we've got more storage, clear any full flags on the space
1675          * infos
1676          */
1677         btrfs_clear_space_info_full(root->fs_info);
1678
1679         unlock_chunks(root);
1680         btrfs_commit_transaction(trans, root);
1681
1682         if (seeding_dev) {
1683                 mutex_unlock(&uuid_mutex);
1684                 up_write(&sb->s_umount);
1685
1686                 ret = btrfs_relocate_sys_chunks(root);
1687                 BUG_ON(ret);
1688         }
1689 out:
1690         mutex_unlock(&root->fs_info->volume_mutex);
1691         return ret;
1692 error:
1693         blkdev_put(bdev, FMODE_EXCL);
1694         if (seeding_dev) {
1695                 mutex_unlock(&uuid_mutex);
1696                 up_write(&sb->s_umount);
1697         }
1698         goto out;
1699 }
1700
1701 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1702                                         struct btrfs_device *device)
1703 {
1704         int ret;
1705         struct btrfs_path *path;
1706         struct btrfs_root *root;
1707         struct btrfs_dev_item *dev_item;
1708         struct extent_buffer *leaf;
1709         struct btrfs_key key;
1710
1711         root = device->dev_root->fs_info->chunk_root;
1712
1713         path = btrfs_alloc_path();
1714         if (!path)
1715                 return -ENOMEM;
1716
1717         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1718         key.type = BTRFS_DEV_ITEM_KEY;
1719         key.offset = device->devid;
1720
1721         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1722         if (ret < 0)
1723                 goto out;
1724
1725         if (ret > 0) {
1726                 ret = -ENOENT;
1727                 goto out;
1728         }
1729
1730         leaf = path->nodes[0];
1731         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1732
1733         btrfs_set_device_id(leaf, dev_item, device->devid);
1734         btrfs_set_device_type(leaf, dev_item, device->type);
1735         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1736         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1737         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1738         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1739         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1740         btrfs_mark_buffer_dirty(leaf);
1741
1742 out:
1743         btrfs_free_path(path);
1744         return ret;
1745 }
1746
1747 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1748                       struct btrfs_device *device, u64 new_size)
1749 {
1750         struct btrfs_super_block *super_copy =
1751                 &device->dev_root->fs_info->super_copy;
1752         u64 old_total = btrfs_super_total_bytes(super_copy);
1753         u64 diff = new_size - device->total_bytes;
1754
1755         if (!device->writeable)
1756                 return -EACCES;
1757         if (new_size <= device->total_bytes)
1758                 return -EINVAL;
1759
1760         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1761         device->fs_devices->total_rw_bytes += diff;
1762
1763         device->total_bytes = new_size;
1764         device->disk_total_bytes = new_size;
1765         btrfs_clear_space_info_full(device->dev_root->fs_info);
1766
1767         return btrfs_update_device(trans, device);
1768 }
1769
1770 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1771                       struct btrfs_device *device, u64 new_size)
1772 {
1773         int ret;
1774         lock_chunks(device->dev_root);
1775         ret = __btrfs_grow_device(trans, device, new_size);
1776         unlock_chunks(device->dev_root);
1777         return ret;
1778 }
1779
1780 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1781                             struct btrfs_root *root,
1782                             u64 chunk_tree, u64 chunk_objectid,
1783                             u64 chunk_offset)
1784 {
1785         int ret;
1786         struct btrfs_path *path;
1787         struct btrfs_key key;
1788
1789         root = root->fs_info->chunk_root;
1790         path = btrfs_alloc_path();
1791         if (!path)
1792                 return -ENOMEM;
1793
1794         key.objectid = chunk_objectid;
1795         key.offset = chunk_offset;
1796         key.type = BTRFS_CHUNK_ITEM_KEY;
1797
1798         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1799         BUG_ON(ret);
1800
1801         ret = btrfs_del_item(trans, root, path);
1802
1803         btrfs_free_path(path);
1804         return ret;
1805 }
1806
1807 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1808                         chunk_offset)
1809 {
1810         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1811         struct btrfs_disk_key *disk_key;
1812         struct btrfs_chunk *chunk;
1813         u8 *ptr;
1814         int ret = 0;
1815         u32 num_stripes;
1816         u32 array_size;
1817         u32 len = 0;
1818         u32 cur;
1819         struct btrfs_key key;
1820
1821         array_size = btrfs_super_sys_array_size(super_copy);
1822
1823         ptr = super_copy->sys_chunk_array;
1824         cur = 0;
1825
1826         while (cur < array_size) {
1827                 disk_key = (struct btrfs_disk_key *)ptr;
1828                 btrfs_disk_key_to_cpu(&key, disk_key);
1829
1830                 len = sizeof(*disk_key);
1831
1832                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1833                         chunk = (struct btrfs_chunk *)(ptr + len);
1834                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1835                         len += btrfs_chunk_item_size(num_stripes);
1836                 } else {
1837                         ret = -EIO;
1838                         break;
1839                 }
1840                 if (key.objectid == chunk_objectid &&
1841                     key.offset == chunk_offset) {
1842                         memmove(ptr, ptr + len, array_size - (cur + len));
1843                         array_size -= len;
1844                         btrfs_set_super_sys_array_size(super_copy, array_size);
1845                 } else {
1846                         ptr += len;
1847                         cur += len;
1848                 }
1849         }
1850         return ret;
1851 }
1852
1853 static int btrfs_relocate_chunk(struct btrfs_root *root,
1854                          u64 chunk_tree, u64 chunk_objectid,
1855                          u64 chunk_offset)
1856 {
1857         struct extent_map_tree *em_tree;
1858         struct btrfs_root *extent_root;
1859         struct btrfs_trans_handle *trans;
1860         struct extent_map *em;
1861         struct map_lookup *map;
1862         int ret;
1863         int i;
1864
1865         root = root->fs_info->chunk_root;
1866         extent_root = root->fs_info->extent_root;
1867         em_tree = &root->fs_info->mapping_tree.map_tree;
1868
1869         ret = btrfs_can_relocate(extent_root, chunk_offset);
1870         if (ret)
1871                 return -ENOSPC;
1872
1873         /* step one, relocate all the extents inside this chunk */
1874         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1875         if (ret)
1876                 return ret;
1877
1878         trans = btrfs_start_transaction(root, 0);
1879         BUG_ON(IS_ERR(trans));
1880
1881         lock_chunks(root);
1882
1883         /*
1884          * step two, delete the device extents and the
1885          * chunk tree entries
1886          */
1887         read_lock(&em_tree->lock);
1888         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1889         read_unlock(&em_tree->lock);
1890
1891         BUG_ON(em->start > chunk_offset ||
1892                em->start + em->len < chunk_offset);
1893         map = (struct map_lookup *)em->bdev;
1894
1895         for (i = 0; i < map->num_stripes; i++) {
1896                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1897                                             map->stripes[i].physical);
1898                 BUG_ON(ret);
1899
1900                 if (map->stripes[i].dev) {
1901                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1902                         BUG_ON(ret);
1903                 }
1904         }
1905         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1906                                chunk_offset);
1907
1908         BUG_ON(ret);
1909
1910         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1911
1912         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1913                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1914                 BUG_ON(ret);
1915         }
1916
1917         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1918         BUG_ON(ret);
1919
1920         write_lock(&em_tree->lock);
1921         remove_extent_mapping(em_tree, em);
1922         write_unlock(&em_tree->lock);
1923
1924         kfree(map);
1925         em->bdev = NULL;
1926
1927         /* once for the tree */
1928         free_extent_map(em);
1929         /* once for us */
1930         free_extent_map(em);
1931
1932         unlock_chunks(root);
1933         btrfs_end_transaction(trans, root);
1934         return 0;
1935 }
1936
1937 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1938 {
1939         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1940         struct btrfs_path *path;
1941         struct extent_buffer *leaf;
1942         struct btrfs_chunk *chunk;
1943         struct btrfs_key key;
1944         struct btrfs_key found_key;
1945         u64 chunk_tree = chunk_root->root_key.objectid;
1946         u64 chunk_type;
1947         bool retried = false;
1948         int failed = 0;
1949         int ret;
1950
1951         path = btrfs_alloc_path();
1952         if (!path)
1953                 return -ENOMEM;
1954
1955 again:
1956         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1957         key.offset = (u64)-1;
1958         key.type = BTRFS_CHUNK_ITEM_KEY;
1959
1960         while (1) {
1961                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1962                 if (ret < 0)
1963                         goto error;
1964                 BUG_ON(ret == 0);
1965
1966                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1967                                           key.type);
1968                 if (ret < 0)
1969                         goto error;
1970                 if (ret > 0)
1971                         break;
1972
1973                 leaf = path->nodes[0];
1974                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1975
1976                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1977                                        struct btrfs_chunk);
1978                 chunk_type = btrfs_chunk_type(leaf, chunk);
1979                 btrfs_release_path(path);
1980
1981                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1982                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1983                                                    found_key.objectid,
1984                                                    found_key.offset);
1985                         if (ret == -ENOSPC)
1986                                 failed++;
1987                         else if (ret)
1988                                 BUG();
1989                 }
1990
1991                 if (found_key.offset == 0)
1992                         break;
1993                 key.offset = found_key.offset - 1;
1994         }
1995         ret = 0;
1996         if (failed && !retried) {
1997                 failed = 0;
1998                 retried = true;
1999                 goto again;
2000         } else if (failed && retried) {
2001                 WARN_ON(1);
2002                 ret = -ENOSPC;
2003         }
2004 error:
2005         btrfs_free_path(path);
2006         return ret;
2007 }
2008
2009 static u64 div_factor(u64 num, int factor)
2010 {
2011         if (factor == 10)
2012                 return num;
2013         num *= factor;
2014         do_div(num, 10);
2015         return num;
2016 }
2017
2018 int btrfs_balance(struct btrfs_root *dev_root)
2019 {
2020         int ret;
2021         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2022         struct btrfs_device *device;
2023         u64 old_size;
2024         u64 size_to_free;
2025         struct btrfs_path *path;
2026         struct btrfs_key key;
2027         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2028         struct btrfs_trans_handle *trans;
2029         struct btrfs_key found_key;
2030
2031         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2032                 return -EROFS;
2033
2034         if (!capable(CAP_SYS_ADMIN))
2035                 return -EPERM;
2036
2037         mutex_lock(&dev_root->fs_info->volume_mutex);
2038         dev_root = dev_root->fs_info->dev_root;
2039
2040         /* step one make some room on all the devices */
2041         list_for_each_entry(device, devices, dev_list) {
2042                 old_size = device->total_bytes;
2043                 size_to_free = div_factor(old_size, 1);
2044                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2045                 if (!device->writeable ||
2046                     device->total_bytes - device->bytes_used > size_to_free)
2047                         continue;
2048
2049                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2050                 if (ret == -ENOSPC)
2051                         break;
2052                 BUG_ON(ret);
2053
2054                 trans = btrfs_start_transaction(dev_root, 0);
2055                 BUG_ON(IS_ERR(trans));
2056
2057                 ret = btrfs_grow_device(trans, device, old_size);
2058                 BUG_ON(ret);
2059
2060                 btrfs_end_transaction(trans, dev_root);
2061         }
2062
2063         /* step two, relocate all the chunks */
2064         path = btrfs_alloc_path();
2065         if (!path) {
2066                 ret = -ENOMEM;
2067                 goto error;
2068         }
2069         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2070         key.offset = (u64)-1;
2071         key.type = BTRFS_CHUNK_ITEM_KEY;
2072
2073         while (1) {
2074                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2075                 if (ret < 0)
2076                         goto error;
2077
2078                 /*
2079                  * this shouldn't happen, it means the last relocate
2080                  * failed
2081                  */
2082                 if (ret == 0)
2083                         break;
2084
2085                 ret = btrfs_previous_item(chunk_root, path, 0,
2086                                           BTRFS_CHUNK_ITEM_KEY);
2087                 if (ret)
2088                         break;
2089
2090                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2091                                       path->slots[0]);
2092                 if (found_key.objectid != key.objectid)
2093                         break;
2094
2095                 /* chunk zero is special */
2096                 if (found_key.offset == 0)
2097                         break;
2098
2099                 btrfs_release_path(path);
2100                 ret = btrfs_relocate_chunk(chunk_root,
2101                                            chunk_root->root_key.objectid,
2102                                            found_key.objectid,
2103                                            found_key.offset);
2104                 if (ret && ret != -ENOSPC)
2105                         goto error;
2106                 key.offset = found_key.offset - 1;
2107         }
2108         ret = 0;
2109 error:
2110         btrfs_free_path(path);
2111         mutex_unlock(&dev_root->fs_info->volume_mutex);
2112         return ret;
2113 }
2114
2115 /*
2116  * shrinking a device means finding all of the device extents past
2117  * the new size, and then following the back refs to the chunks.
2118  * The chunk relocation code actually frees the device extent
2119  */
2120 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2121 {
2122         struct btrfs_trans_handle *trans;
2123         struct btrfs_root *root = device->dev_root;
2124         struct btrfs_dev_extent *dev_extent = NULL;
2125         struct btrfs_path *path;
2126         u64 length;
2127         u64 chunk_tree;
2128         u64 chunk_objectid;
2129         u64 chunk_offset;
2130         int ret;
2131         int slot;
2132         int failed = 0;
2133         bool retried = false;
2134         struct extent_buffer *l;
2135         struct btrfs_key key;
2136         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2137         u64 old_total = btrfs_super_total_bytes(super_copy);
2138         u64 old_size = device->total_bytes;
2139         u64 diff = device->total_bytes - new_size;
2140
2141         if (new_size >= device->total_bytes)
2142                 return -EINVAL;
2143
2144         path = btrfs_alloc_path();
2145         if (!path)
2146                 return -ENOMEM;
2147
2148         path->reada = 2;
2149
2150         lock_chunks(root);
2151
2152         device->total_bytes = new_size;
2153         if (device->writeable)
2154                 device->fs_devices->total_rw_bytes -= diff;
2155         unlock_chunks(root);
2156
2157 again:
2158         key.objectid = device->devid;
2159         key.offset = (u64)-1;
2160         key.type = BTRFS_DEV_EXTENT_KEY;
2161
2162         while (1) {
2163                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2164                 if (ret < 0)
2165                         goto done;
2166
2167                 ret = btrfs_previous_item(root, path, 0, key.type);
2168                 if (ret < 0)
2169                         goto done;
2170                 if (ret) {
2171                         ret = 0;
2172                         btrfs_release_path(path);
2173                         break;
2174                 }
2175
2176                 l = path->nodes[0];
2177                 slot = path->slots[0];
2178                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2179
2180                 if (key.objectid != device->devid) {
2181                         btrfs_release_path(path);
2182                         break;
2183                 }
2184
2185                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2186                 length = btrfs_dev_extent_length(l, dev_extent);
2187
2188                 if (key.offset + length <= new_size) {
2189                         btrfs_release_path(path);
2190                         break;
2191                 }
2192
2193                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2194                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2195                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2196                 btrfs_release_path(path);
2197
2198                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2199                                            chunk_offset);
2200                 if (ret && ret != -ENOSPC)
2201                         goto done;
2202                 if (ret == -ENOSPC)
2203                         failed++;
2204                 key.offset -= 1;
2205         }
2206
2207         if (failed && !retried) {
2208                 failed = 0;
2209                 retried = true;
2210                 goto again;
2211         } else if (failed && retried) {
2212                 ret = -ENOSPC;
2213                 lock_chunks(root);
2214
2215                 device->total_bytes = old_size;
2216                 if (device->writeable)
2217                         device->fs_devices->total_rw_bytes += diff;
2218                 unlock_chunks(root);
2219                 goto done;
2220         }
2221
2222         /* Shrinking succeeded, else we would be at "done". */
2223         trans = btrfs_start_transaction(root, 0);
2224         if (IS_ERR(trans)) {
2225                 ret = PTR_ERR(trans);
2226                 goto done;
2227         }
2228
2229         lock_chunks(root);
2230
2231         device->disk_total_bytes = new_size;
2232         /* Now btrfs_update_device() will change the on-disk size. */
2233         ret = btrfs_update_device(trans, device);
2234         if (ret) {
2235                 unlock_chunks(root);
2236                 btrfs_end_transaction(trans, root);
2237                 goto done;
2238         }
2239         WARN_ON(diff > old_total);
2240         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2241         unlock_chunks(root);
2242         btrfs_end_transaction(trans, root);
2243 done:
2244         btrfs_free_path(path);
2245         return ret;
2246 }
2247
2248 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2249                            struct btrfs_root *root,
2250                            struct btrfs_key *key,
2251                            struct btrfs_chunk *chunk, int item_size)
2252 {
2253         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2254         struct btrfs_disk_key disk_key;
2255         u32 array_size;
2256         u8 *ptr;
2257
2258         array_size = btrfs_super_sys_array_size(super_copy);
2259         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2260                 return -EFBIG;
2261
2262         ptr = super_copy->sys_chunk_array + array_size;
2263         btrfs_cpu_key_to_disk(&disk_key, key);
2264         memcpy(ptr, &disk_key, sizeof(disk_key));
2265         ptr += sizeof(disk_key);
2266         memcpy(ptr, chunk, item_size);
2267         item_size += sizeof(disk_key);
2268         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2269         return 0;
2270 }
2271
2272 /*
2273  * sort the devices in descending order by max_avail, total_avail
2274  */
2275 static int btrfs_cmp_device_info(const void *a, const void *b)
2276 {
2277         const struct btrfs_device_info *di_a = a;
2278         const struct btrfs_device_info *di_b = b;
2279
2280         if (di_a->max_avail > di_b->max_avail)
2281                 return -1;
2282         if (di_a->max_avail < di_b->max_avail)
2283                 return 1;
2284         if (di_a->total_avail > di_b->total_avail)
2285                 return -1;
2286         if (di_a->total_avail < di_b->total_avail)
2287                 return 1;
2288         return 0;
2289 }
2290
2291 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2292                                struct btrfs_root *extent_root,
2293                                struct map_lookup **map_ret,
2294                                u64 *num_bytes_out, u64 *stripe_size_out,
2295                                u64 start, u64 type)
2296 {
2297         struct btrfs_fs_info *info = extent_root->fs_info;
2298         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2299         struct list_head *cur;
2300         struct map_lookup *map = NULL;
2301         struct extent_map_tree *em_tree;
2302         struct extent_map *em;
2303         struct btrfs_device_info *devices_info = NULL;
2304         u64 total_avail;
2305         int num_stripes;        /* total number of stripes to allocate */
2306         int sub_stripes;        /* sub_stripes info for map */
2307         int dev_stripes;        /* stripes per dev */
2308         int devs_max;           /* max devs to use */
2309         int devs_min;           /* min devs needed */
2310         int devs_increment;     /* ndevs has to be a multiple of this */
2311         int ncopies;            /* how many copies to data has */
2312         int ret;
2313         u64 max_stripe_size;
2314         u64 max_chunk_size;
2315         u64 stripe_size;
2316         u64 num_bytes;
2317         int ndevs;
2318         int i;
2319         int j;
2320
2321         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2322             (type & BTRFS_BLOCK_GROUP_DUP)) {
2323                 WARN_ON(1);
2324                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2325         }
2326
2327         if (list_empty(&fs_devices->alloc_list))
2328                 return -ENOSPC;
2329
2330         sub_stripes = 1;
2331         dev_stripes = 1;
2332         devs_increment = 1;
2333         ncopies = 1;
2334         devs_max = 0;   /* 0 == as many as possible */
2335         devs_min = 1;
2336
2337         /*
2338          * define the properties of each RAID type.
2339          * FIXME: move this to a global table and use it in all RAID
2340          * calculation code
2341          */
2342         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2343                 dev_stripes = 2;
2344                 ncopies = 2;
2345                 devs_max = 1;
2346         } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2347                 devs_min = 2;
2348         } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2349                 devs_increment = 2;
2350                 ncopies = 2;
2351                 devs_max = 2;
2352                 devs_min = 2;
2353         } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2354                 sub_stripes = 2;
2355                 devs_increment = 2;
2356                 ncopies = 2;
2357                 devs_min = 4;
2358         } else {
2359                 devs_max = 1;
2360         }
2361
2362         if (type & BTRFS_BLOCK_GROUP_DATA) {
2363                 max_stripe_size = 1024 * 1024 * 1024;
2364                 max_chunk_size = 10 * max_stripe_size;
2365         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2366                 max_stripe_size = 256 * 1024 * 1024;
2367                 max_chunk_size = max_stripe_size;
2368         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2369                 max_stripe_size = 8 * 1024 * 1024;
2370                 max_chunk_size = 2 * max_stripe_size;
2371         } else {
2372                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2373                        type);
2374                 BUG_ON(1);
2375         }
2376
2377         /* we don't want a chunk larger than 10% of writeable space */
2378         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2379                              max_chunk_size);
2380
2381         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2382                                GFP_NOFS);
2383         if (!devices_info)
2384                 return -ENOMEM;
2385
2386         cur = fs_devices->alloc_list.next;
2387
2388         /*
2389          * in the first pass through the devices list, we gather information
2390          * about the available holes on each device.
2391          */
2392         ndevs = 0;
2393         while (cur != &fs_devices->alloc_list) {
2394                 struct btrfs_device *device;
2395                 u64 max_avail;
2396                 u64 dev_offset;
2397
2398                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2399
2400                 cur = cur->next;
2401
2402                 if (!device->writeable) {
2403                         printk(KERN_ERR
2404                                "btrfs: read-only device in alloc_list\n");
2405                         WARN_ON(1);
2406                         continue;
2407                 }
2408
2409                 if (!device->in_fs_metadata)
2410                         continue;
2411
2412                 if (device->total_bytes > device->bytes_used)
2413                         total_avail = device->total_bytes - device->bytes_used;
2414                 else
2415                         total_avail = 0;
2416                 /* avail is off by max(alloc_start, 1MB), but that is the same
2417                  * for all devices, so it doesn't hurt the sorting later on
2418                  */
2419
2420                 ret = find_free_dev_extent(trans, device,
2421                                            max_stripe_size * dev_stripes,
2422                                            &dev_offset, &max_avail);
2423                 if (ret && ret != -ENOSPC)
2424                         goto error;
2425
2426                 if (ret == 0)
2427                         max_avail = max_stripe_size * dev_stripes;
2428
2429                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2430                         continue;
2431
2432                 devices_info[ndevs].dev_offset = dev_offset;
2433                 devices_info[ndevs].max_avail = max_avail;
2434                 devices_info[ndevs].total_avail = total_avail;
2435                 devices_info[ndevs].dev = device;
2436                 ++ndevs;
2437         }
2438
2439         /*
2440          * now sort the devices by hole size / available space
2441          */
2442         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2443              btrfs_cmp_device_info, NULL);
2444
2445         /* round down to number of usable stripes */
2446         ndevs -= ndevs % devs_increment;
2447
2448         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2449                 ret = -ENOSPC;
2450                 goto error;
2451         }
2452
2453         if (devs_max && ndevs > devs_max)
2454                 ndevs = devs_max;
2455         /*
2456          * the primary goal is to maximize the number of stripes, so use as many
2457          * devices as possible, even if the stripes are not maximum sized.
2458          */
2459         stripe_size = devices_info[ndevs-1].max_avail;
2460         num_stripes = ndevs * dev_stripes;
2461
2462         if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2463                 stripe_size = max_chunk_size * ncopies;
2464                 do_div(stripe_size, num_stripes);
2465         }
2466
2467         do_div(stripe_size, dev_stripes);
2468         do_div(stripe_size, BTRFS_STRIPE_LEN);
2469         stripe_size *= BTRFS_STRIPE_LEN;
2470
2471         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2472         if (!map) {
2473                 ret = -ENOMEM;
2474                 goto error;
2475         }
2476         map->num_stripes = num_stripes;
2477
2478         for (i = 0; i < ndevs; ++i) {
2479                 for (j = 0; j < dev_stripes; ++j) {
2480                         int s = i * dev_stripes + j;
2481                         map->stripes[s].dev = devices_info[i].dev;
2482                         map->stripes[s].physical = devices_info[i].dev_offset +
2483                                                    j * stripe_size;
2484                 }
2485         }
2486         map->sector_size = extent_root->sectorsize;
2487         map->stripe_len = BTRFS_STRIPE_LEN;
2488         map->io_align = BTRFS_STRIPE_LEN;
2489         map->io_width = BTRFS_STRIPE_LEN;
2490         map->type = type;
2491         map->sub_stripes = sub_stripes;
2492
2493         *map_ret = map;
2494         num_bytes = stripe_size * (num_stripes / ncopies);
2495
2496         *stripe_size_out = stripe_size;
2497         *num_bytes_out = num_bytes;
2498
2499         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2500
2501         em = alloc_extent_map();
2502         if (!em) {
2503                 ret = -ENOMEM;
2504                 goto error;
2505         }
2506         em->bdev = (struct block_device *)map;
2507         em->start = start;
2508         em->len = num_bytes;
2509         em->block_start = 0;
2510         em->block_len = em->len;
2511
2512         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2513         write_lock(&em_tree->lock);
2514         ret = add_extent_mapping(em_tree, em);
2515         write_unlock(&em_tree->lock);
2516         BUG_ON(ret);
2517         free_extent_map(em);
2518
2519         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2520                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2521                                      start, num_bytes);
2522         BUG_ON(ret);
2523
2524         for (i = 0; i < map->num_stripes; ++i) {
2525                 struct btrfs_device *device;
2526                 u64 dev_offset;
2527
2528                 device = map->stripes[i].dev;
2529                 dev_offset = map->stripes[i].physical;
2530
2531                 ret = btrfs_alloc_dev_extent(trans, device,
2532                                 info->chunk_root->root_key.objectid,
2533                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2534                                 start, dev_offset, stripe_size);
2535                 BUG_ON(ret);
2536         }
2537
2538         kfree(devices_info);
2539         return 0;
2540
2541 error:
2542         kfree(map);
2543         kfree(devices_info);
2544         return ret;
2545 }
2546
2547 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2548                                 struct btrfs_root *extent_root,
2549                                 struct map_lookup *map, u64 chunk_offset,
2550                                 u64 chunk_size, u64 stripe_size)
2551 {
2552         u64 dev_offset;
2553         struct btrfs_key key;
2554         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2555         struct btrfs_device *device;
2556         struct btrfs_chunk *chunk;
2557         struct btrfs_stripe *stripe;
2558         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2559         int index = 0;
2560         int ret;
2561
2562         chunk = kzalloc(item_size, GFP_NOFS);
2563         if (!chunk)
2564                 return -ENOMEM;
2565
2566         index = 0;
2567         while (index < map->num_stripes) {
2568                 device = map->stripes[index].dev;
2569                 device->bytes_used += stripe_size;
2570                 ret = btrfs_update_device(trans, device);
2571                 BUG_ON(ret);
2572                 index++;
2573         }
2574
2575         index = 0;
2576         stripe = &chunk->stripe;
2577         while (index < map->num_stripes) {
2578                 device = map->stripes[index].dev;
2579                 dev_offset = map->stripes[index].physical;
2580
2581                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2582                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2583                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2584                 stripe++;
2585                 index++;
2586         }
2587
2588         btrfs_set_stack_chunk_length(chunk, chunk_size);
2589         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2590         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2591         btrfs_set_stack_chunk_type(chunk, map->type);
2592         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2593         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2594         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2595         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2596         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2597
2598         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2599         key.type = BTRFS_CHUNK_ITEM_KEY;
2600         key.offset = chunk_offset;
2601
2602         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2603         BUG_ON(ret);
2604
2605         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2606                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2607                                              item_size);
2608                 BUG_ON(ret);
2609         }
2610
2611         kfree(chunk);
2612         return 0;
2613 }
2614
2615 /*
2616  * Chunk allocation falls into two parts. The first part does works
2617  * that make the new allocated chunk useable, but not do any operation
2618  * that modifies the chunk tree. The second part does the works that
2619  * require modifying the chunk tree. This division is important for the
2620  * bootstrap process of adding storage to a seed btrfs.
2621  */
2622 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2623                       struct btrfs_root *extent_root, u64 type)
2624 {
2625         u64 chunk_offset;
2626         u64 chunk_size;
2627         u64 stripe_size;
2628         struct map_lookup *map;
2629         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2630         int ret;
2631
2632         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2633                               &chunk_offset);
2634         if (ret)
2635                 return ret;
2636
2637         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2638                                   &stripe_size, chunk_offset, type);
2639         if (ret)
2640                 return ret;
2641
2642         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2643                                    chunk_size, stripe_size);
2644         BUG_ON(ret);
2645         return 0;
2646 }
2647
2648 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2649                                          struct btrfs_root *root,
2650                                          struct btrfs_device *device)
2651 {
2652         u64 chunk_offset;
2653         u64 sys_chunk_offset;
2654         u64 chunk_size;
2655         u64 sys_chunk_size;
2656         u64 stripe_size;
2657         u64 sys_stripe_size;
2658         u64 alloc_profile;
2659         struct map_lookup *map;
2660         struct map_lookup *sys_map;
2661         struct btrfs_fs_info *fs_info = root->fs_info;
2662         struct btrfs_root *extent_root = fs_info->extent_root;
2663         int ret;
2664
2665         ret = find_next_chunk(fs_info->chunk_root,
2666                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2667         if (ret)
2668                 return ret;
2669
2670         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2671                         (fs_info->metadata_alloc_profile &
2672                          fs_info->avail_metadata_alloc_bits);
2673         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2674
2675         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2676                                   &stripe_size, chunk_offset, alloc_profile);
2677         BUG_ON(ret);
2678
2679         sys_chunk_offset = chunk_offset + chunk_size;
2680
2681         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2682                         (fs_info->system_alloc_profile &
2683                          fs_info->avail_system_alloc_bits);
2684         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2685
2686         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2687                                   &sys_chunk_size, &sys_stripe_size,
2688                                   sys_chunk_offset, alloc_profile);
2689         BUG_ON(ret);
2690
2691         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2692         BUG_ON(ret);
2693
2694         /*
2695          * Modifying chunk tree needs allocating new blocks from both
2696          * system block group and metadata block group. So we only can
2697          * do operations require modifying the chunk tree after both
2698          * block groups were created.
2699          */
2700         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2701                                    chunk_size, stripe_size);
2702         BUG_ON(ret);
2703
2704         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2705                                    sys_chunk_offset, sys_chunk_size,
2706                                    sys_stripe_size);
2707         BUG_ON(ret);
2708         return 0;
2709 }
2710
2711 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2712 {
2713         struct extent_map *em;
2714         struct map_lookup *map;
2715         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2716         int readonly = 0;
2717         int i;
2718
2719         read_lock(&map_tree->map_tree.lock);
2720         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2721         read_unlock(&map_tree->map_tree.lock);
2722         if (!em)
2723                 return 1;
2724
2725         if (btrfs_test_opt(root, DEGRADED)) {
2726                 free_extent_map(em);
2727                 return 0;
2728         }
2729
2730         map = (struct map_lookup *)em->bdev;
2731         for (i = 0; i < map->num_stripes; i++) {
2732                 if (!map->stripes[i].dev->writeable) {
2733                         readonly = 1;
2734                         break;
2735                 }
2736         }
2737         free_extent_map(em);
2738         return readonly;
2739 }
2740
2741 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2742 {
2743         extent_map_tree_init(&tree->map_tree);
2744 }
2745
2746 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2747 {
2748         struct extent_map *em;
2749
2750         while (1) {
2751                 write_lock(&tree->map_tree.lock);
2752                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2753                 if (em)
2754                         remove_extent_mapping(&tree->map_tree, em);
2755                 write_unlock(&tree->map_tree.lock);
2756                 if (!em)
2757                         break;
2758                 kfree(em->bdev);
2759                 /* once for us */
2760                 free_extent_map(em);
2761                 /* once for the tree */
2762                 free_extent_map(em);
2763         }
2764 }
2765
2766 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2767 {
2768         struct extent_map *em;
2769         struct map_lookup *map;
2770         struct extent_map_tree *em_tree = &map_tree->map_tree;
2771         int ret;
2772
2773         read_lock(&em_tree->lock);
2774         em = lookup_extent_mapping(em_tree, logical, len);
2775         read_unlock(&em_tree->lock);
2776         BUG_ON(!em);
2777
2778         BUG_ON(em->start > logical || em->start + em->len < logical);
2779         map = (struct map_lookup *)em->bdev;
2780         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2781                 ret = map->num_stripes;
2782         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2783                 ret = map->sub_stripes;
2784         else
2785                 ret = 1;
2786         free_extent_map(em);
2787         return ret;
2788 }
2789
2790 static int find_live_mirror(struct map_lookup *map, int first, int num,
2791                             int optimal)
2792 {
2793         int i;
2794         if (map->stripes[optimal].dev->bdev)
2795                 return optimal;
2796         for (i = first; i < first + num; i++) {
2797                 if (map->stripes[i].dev->bdev)
2798                         return i;
2799         }
2800         /* we couldn't find one that doesn't fail.  Just return something
2801          * and the io error handling code will clean up eventually
2802          */
2803         return optimal;
2804 }
2805
2806 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2807                              u64 logical, u64 *length,
2808                              struct btrfs_multi_bio **multi_ret,
2809                              int mirror_num)
2810 {
2811         struct extent_map *em;
2812         struct map_lookup *map;
2813         struct extent_map_tree *em_tree = &map_tree->map_tree;
2814         u64 offset;
2815         u64 stripe_offset;
2816         u64 stripe_end_offset;
2817         u64 stripe_nr;
2818         u64 stripe_nr_orig;
2819         u64 stripe_nr_end;
2820         int stripes_allocated = 8;
2821         int stripes_required = 1;
2822         int stripe_index;
2823         int i;
2824         int num_stripes;
2825         int max_errors = 0;
2826         struct btrfs_multi_bio *multi = NULL;
2827
2828         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2829                 stripes_allocated = 1;
2830 again:
2831         if (multi_ret) {
2832                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2833                                 GFP_NOFS);
2834                 if (!multi)
2835                         return -ENOMEM;
2836
2837                 atomic_set(&multi->error, 0);
2838         }
2839
2840         read_lock(&em_tree->lock);
2841         em = lookup_extent_mapping(em_tree, logical, *length);
2842         read_unlock(&em_tree->lock);
2843
2844         if (!em) {
2845                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2846                        (unsigned long long)logical,
2847                        (unsigned long long)*length);
2848                 BUG();
2849         }
2850
2851         BUG_ON(em->start > logical || em->start + em->len < logical);
2852         map = (struct map_lookup *)em->bdev;
2853         offset = logical - em->start;
2854
2855         if (mirror_num > map->num_stripes)
2856                 mirror_num = 0;
2857
2858         /* if our multi bio struct is too small, back off and try again */
2859         if (rw & REQ_WRITE) {
2860                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2861                                  BTRFS_BLOCK_GROUP_DUP)) {
2862                         stripes_required = map->num_stripes;
2863                         max_errors = 1;
2864                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2865                         stripes_required = map->sub_stripes;
2866                         max_errors = 1;
2867                 }
2868         }
2869         if (rw & REQ_DISCARD) {
2870                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2871                                  BTRFS_BLOCK_GROUP_RAID1 |
2872                                  BTRFS_BLOCK_GROUP_DUP |
2873                                  BTRFS_BLOCK_GROUP_RAID10)) {
2874                         stripes_required = map->num_stripes;
2875                 }
2876         }
2877         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2878             stripes_allocated < stripes_required) {
2879                 stripes_allocated = map->num_stripes;
2880                 free_extent_map(em);
2881                 kfree(multi);
2882                 goto again;
2883         }
2884         stripe_nr = offset;
2885         /*
2886          * stripe_nr counts the total number of stripes we have to stride
2887          * to get to this block
2888          */
2889         do_div(stripe_nr, map->stripe_len);
2890
2891         stripe_offset = stripe_nr * map->stripe_len;
2892         BUG_ON(offset < stripe_offset);
2893
2894         /* stripe_offset is the offset of this block in its stripe*/
2895         stripe_offset = offset - stripe_offset;
2896
2897         if (rw & REQ_DISCARD)
2898                 *length = min_t(u64, em->len - offset, *length);
2899         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2900                               BTRFS_BLOCK_GROUP_RAID1 |
2901                               BTRFS_BLOCK_GROUP_RAID10 |
2902                               BTRFS_BLOCK_GROUP_DUP)) {
2903                 /* we limit the length of each bio to what fits in a stripe */
2904                 *length = min_t(u64, em->len - offset,
2905                                 map->stripe_len - stripe_offset);
2906         } else {
2907                 *length = em->len - offset;
2908         }
2909
2910         if (!multi_ret)
2911                 goto out;
2912
2913         num_stripes = 1;
2914         stripe_index = 0;
2915         stripe_nr_orig = stripe_nr;
2916         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2917                         (~(map->stripe_len - 1));
2918         do_div(stripe_nr_end, map->stripe_len);
2919         stripe_end_offset = stripe_nr_end * map->stripe_len -
2920                             (offset + *length);
2921         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2922                 if (rw & REQ_DISCARD)
2923                         num_stripes = min_t(u64, map->num_stripes,
2924                                             stripe_nr_end - stripe_nr_orig);
2925                 stripe_index = do_div(stripe_nr, map->num_stripes);
2926         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2927                 if (rw & (REQ_WRITE | REQ_DISCARD))
2928                         num_stripes = map->num_stripes;
2929                 else if (mirror_num)
2930                         stripe_index = mirror_num - 1;
2931                 else {
2932                         stripe_index = find_live_mirror(map, 0,
2933                                             map->num_stripes,
2934                                             current->pid % map->num_stripes);
2935                 }
2936
2937         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2938                 if (rw & (REQ_WRITE | REQ_DISCARD))
2939                         num_stripes = map->num_stripes;
2940                 else if (mirror_num)
2941                         stripe_index = mirror_num - 1;
2942
2943         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2944                 int factor = map->num_stripes / map->sub_stripes;
2945
2946                 stripe_index = do_div(stripe_nr, factor);
2947                 stripe_index *= map->sub_stripes;
2948
2949                 if (rw & REQ_WRITE)
2950                         num_stripes = map->sub_stripes;
2951                 else if (rw & REQ_DISCARD)
2952                         num_stripes = min_t(u64, map->sub_stripes *
2953                                             (stripe_nr_end - stripe_nr_orig),
2954                                             map->num_stripes);
2955                 else if (mirror_num)
2956                         stripe_index += mirror_num - 1;
2957                 else {
2958                         stripe_index = find_live_mirror(map, stripe_index,
2959                                               map->sub_stripes, stripe_index +
2960                                               current->pid % map->sub_stripes);
2961                 }
2962         } else {
2963                 /*
2964                  * after this do_div call, stripe_nr is the number of stripes
2965                  * on this device we have to walk to find the data, and
2966                  * stripe_index is the number of our device in the stripe array
2967                  */
2968                 stripe_index = do_div(stripe_nr, map->num_stripes);
2969         }
2970         BUG_ON(stripe_index >= map->num_stripes);
2971
2972         if (rw & REQ_DISCARD) {
2973                 for (i = 0; i < num_stripes; i++) {
2974                         multi->stripes[i].physical =
2975                                 map->stripes[stripe_index].physical +
2976                                 stripe_offset + stripe_nr * map->stripe_len;
2977                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2978
2979                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2980                                 u64 stripes;
2981                                 u32 last_stripe = 0;
2982                                 int j;
2983
2984                                 div_u64_rem(stripe_nr_end - 1,
2985                                             map->num_stripes,
2986                                             &last_stripe);
2987
2988                                 for (j = 0; j < map->num_stripes; j++) {
2989                                         u32 test;
2990
2991                                         div_u64_rem(stripe_nr_end - 1 - j,
2992                                                     map->num_stripes, &test);
2993                                         if (test == stripe_index)
2994                                                 break;
2995                                 }
2996                                 stripes = stripe_nr_end - 1 - j;
2997                                 do_div(stripes, map->num_stripes);
2998                                 multi->stripes[i].length = map->stripe_len *
2999                                         (stripes - stripe_nr + 1);
3000
3001                                 if (i == 0) {
3002                                         multi->stripes[i].length -=
3003                                                 stripe_offset;
3004                                         stripe_offset = 0;
3005                                 }
3006                                 if (stripe_index == last_stripe)
3007                                         multi->stripes[i].length -=
3008                                                 stripe_end_offset;
3009                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3010                                 u64 stripes;
3011                                 int j;
3012                                 int factor = map->num_stripes /
3013                                              map->sub_stripes;
3014                                 u32 last_stripe = 0;
3015
3016                                 div_u64_rem(stripe_nr_end - 1,
3017                                             factor, &last_stripe);
3018                                 last_stripe *= map->sub_stripes;
3019
3020                                 for (j = 0; j < factor; j++) {
3021                                         u32 test;
3022
3023                                         div_u64_rem(stripe_nr_end - 1 - j,
3024                                                     factor, &test);
3025
3026                                         if (test ==
3027                                             stripe_index / map->sub_stripes)
3028                                                 break;
3029                                 }
3030                                 stripes = stripe_nr_end - 1 - j;
3031                                 do_div(stripes, factor);
3032                                 multi->stripes[i].length = map->stripe_len *
3033                                         (stripes - stripe_nr + 1);
3034
3035                                 if (i < map->sub_stripes) {
3036                                         multi->stripes[i].length -=
3037                                                 stripe_offset;
3038                                         if (i == map->sub_stripes - 1)
3039                                                 stripe_offset = 0;
3040                                 }
3041                                 if (stripe_index >= last_stripe &&
3042                                     stripe_index <= (last_stripe +
3043                                                      map->sub_stripes - 1)) {
3044                                         multi->stripes[i].length -=
3045                                                 stripe_end_offset;
3046                                 }
3047                         } else
3048                                 multi->stripes[i].length = *length;
3049
3050                         stripe_index++;
3051                         if (stripe_index == map->num_stripes) {
3052                                 /* This could only happen for RAID0/10 */
3053                                 stripe_index = 0;
3054                                 stripe_nr++;
3055                         }
3056                 }
3057         } else {
3058                 for (i = 0; i < num_stripes; i++) {
3059                         multi->stripes[i].physical =
3060                                 map->stripes[stripe_index].physical +
3061                                 stripe_offset +
3062                                 stripe_nr * map->stripe_len;
3063                         multi->stripes[i].dev =
3064                                 map->stripes[stripe_index].dev;
3065                         stripe_index++;
3066                 }
3067         }
3068         if (multi_ret) {
3069                 *multi_ret = multi;
3070                 multi->num_stripes = num_stripes;
3071                 multi->max_errors = max_errors;
3072         }
3073 out:
3074         free_extent_map(em);
3075         return 0;
3076 }
3077
3078 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3079                       u64 logical, u64 *length,
3080                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3081 {
3082         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3083                                  mirror_num);
3084 }
3085
3086 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3087                      u64 chunk_start, u64 physical, u64 devid,
3088                      u64 **logical, int *naddrs, int *stripe_len)
3089 {
3090         struct extent_map_tree *em_tree = &map_tree->map_tree;
3091         struct extent_map *em;
3092         struct map_lookup *map;
3093         u64 *buf;
3094         u64 bytenr;
3095         u64 length;
3096         u64 stripe_nr;
3097         int i, j, nr = 0;
3098
3099         read_lock(&em_tree->lock);
3100         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3101         read_unlock(&em_tree->lock);
3102
3103         BUG_ON(!em || em->start != chunk_start);
3104         map = (struct map_lookup *)em->bdev;
3105
3106         length = em->len;
3107         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3108                 do_div(length, map->num_stripes / map->sub_stripes);
3109         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3110                 do_div(length, map->num_stripes);
3111
3112         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3113         BUG_ON(!buf);
3114
3115         for (i = 0; i < map->num_stripes; i++) {
3116                 if (devid && map->stripes[i].dev->devid != devid)
3117                         continue;
3118                 if (map->stripes[i].physical > physical ||
3119                     map->stripes[i].physical + length <= physical)
3120                         continue;
3121
3122                 stripe_nr = physical - map->stripes[i].physical;
3123                 do_div(stripe_nr, map->stripe_len);
3124
3125                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3126                         stripe_nr = stripe_nr * map->num_stripes + i;
3127                         do_div(stripe_nr, map->sub_stripes);
3128                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3129                         stripe_nr = stripe_nr * map->num_stripes + i;
3130                 }
3131                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3132                 WARN_ON(nr >= map->num_stripes);
3133                 for (j = 0; j < nr; j++) {
3134                         if (buf[j] == bytenr)
3135                                 break;
3136                 }
3137                 if (j == nr) {
3138                         WARN_ON(nr >= map->num_stripes);
3139                         buf[nr++] = bytenr;
3140                 }
3141         }
3142
3143         *logical = buf;
3144         *naddrs = nr;
3145         *stripe_len = map->stripe_len;
3146
3147         free_extent_map(em);
3148         return 0;
3149 }
3150
3151 static void end_bio_multi_stripe(struct bio *bio, int err)
3152 {
3153         struct btrfs_multi_bio *multi = bio->bi_private;
3154         int is_orig_bio = 0;
3155
3156         if (err)
3157                 atomic_inc(&multi->error);
3158
3159         if (bio == multi->orig_bio)
3160                 is_orig_bio = 1;
3161
3162         if (atomic_dec_and_test(&multi->stripes_pending)) {
3163                 if (!is_orig_bio) {
3164                         bio_put(bio);
3165                         bio = multi->orig_bio;
3166                 }
3167                 bio->bi_private = multi->private;
3168                 bio->bi_end_io = multi->end_io;
3169                 /* only send an error to the higher layers if it is
3170                  * beyond the tolerance of the multi-bio
3171                  */
3172                 if (atomic_read(&multi->error) > multi->max_errors) {
3173                         err = -EIO;
3174                 } else if (err) {
3175                         /*
3176                          * this bio is actually up to date, we didn't
3177                          * go over the max number of errors
3178                          */
3179                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3180                         err = 0;
3181                 }
3182                 kfree(multi);
3183
3184                 bio_endio(bio, err);
3185         } else if (!is_orig_bio) {
3186                 bio_put(bio);
3187         }
3188 }
3189
3190 struct async_sched {
3191         struct bio *bio;
3192         int rw;
3193         struct btrfs_fs_info *info;
3194         struct btrfs_work work;
3195 };
3196
3197 /*
3198  * see run_scheduled_bios for a description of why bios are collected for
3199  * async submit.
3200  *
3201  * This will add one bio to the pending list for a device and make sure
3202  * the work struct is scheduled.
3203  */
3204 static noinline int schedule_bio(struct btrfs_root *root,
3205                                  struct btrfs_device *device,
3206                                  int rw, struct bio *bio)
3207 {
3208         int should_queue = 1;
3209         struct btrfs_pending_bios *pending_bios;
3210
3211         /* don't bother with additional async steps for reads, right now */
3212         if (!(rw & REQ_WRITE)) {
3213                 bio_get(bio);
3214                 submit_bio(rw, bio);
3215                 bio_put(bio);
3216                 return 0;
3217         }
3218
3219         /*
3220          * nr_async_bios allows us to reliably return congestion to the
3221          * higher layers.  Otherwise, the async bio makes it appear we have
3222          * made progress against dirty pages when we've really just put it
3223          * on a queue for later
3224          */
3225         atomic_inc(&root->fs_info->nr_async_bios);
3226         WARN_ON(bio->bi_next);
3227         bio->bi_next = NULL;
3228         bio->bi_rw |= rw;
3229
3230         spin_lock(&device->io_lock);
3231         if (bio->bi_rw & REQ_SYNC)
3232                 pending_bios = &device->pending_sync_bios;
3233         else
3234                 pending_bios = &device->pending_bios;
3235
3236         if (pending_bios->tail)
3237                 pending_bios->tail->bi_next = bio;
3238
3239         pending_bios->tail = bio;
3240         if (!pending_bios->head)
3241                 pending_bios->head = bio;
3242         if (device->running_pending)
3243                 should_queue = 0;
3244
3245         spin_unlock(&device->io_lock);
3246
3247         if (should_queue)
3248                 btrfs_queue_worker(&root->fs_info->submit_workers,
3249                                    &device->work);
3250         return 0;
3251 }
3252
3253 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3254                   int mirror_num, int async_submit)
3255 {
3256         struct btrfs_mapping_tree *map_tree;
3257         struct btrfs_device *dev;
3258         struct bio *first_bio = bio;
3259         u64 logical = (u64)bio->bi_sector << 9;
3260         u64 length = 0;
3261         u64 map_length;
3262         struct btrfs_multi_bio *multi = NULL;
3263         int ret;
3264         int dev_nr = 0;
3265         int total_devs = 1;
3266
3267         length = bio->bi_size;
3268         map_tree = &root->fs_info->mapping_tree;
3269         map_length = length;
3270
3271         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3272                               mirror_num);
3273         BUG_ON(ret);
3274
3275         total_devs = multi->num_stripes;
3276         if (map_length < length) {
3277                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3278                        "len %llu\n", (unsigned long long)logical,
3279                        (unsigned long long)length,
3280                        (unsigned long long)map_length);
3281                 BUG();
3282         }
3283         multi->end_io = first_bio->bi_end_io;
3284         multi->private = first_bio->bi_private;
3285         multi->orig_bio = first_bio;
3286         atomic_set(&multi->stripes_pending, multi->num_stripes);
3287
3288         while (dev_nr < total_devs) {
3289                 if (total_devs > 1) {
3290                         if (dev_nr < total_devs - 1) {
3291                                 bio = bio_clone(first_bio, GFP_NOFS);
3292                                 BUG_ON(!bio);
3293                         } else {
3294                                 bio = first_bio;
3295                         }
3296                         bio->bi_private = multi;
3297                         bio->bi_end_io = end_bio_multi_stripe;
3298                 }
3299                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3300                 dev = multi->stripes[dev_nr].dev;
3301                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3302                         bio->bi_bdev = dev->bdev;
3303                         if (async_submit)
3304                                 schedule_bio(root, dev, rw, bio);
3305                         else
3306                                 submit_bio(rw, bio);
3307                 } else {
3308                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3309                         bio->bi_sector = logical >> 9;
3310                         bio_endio(bio, -EIO);
3311                 }
3312                 dev_nr++;
3313         }
3314         if (total_devs == 1)
3315                 kfree(multi);
3316         return 0;
3317 }
3318
3319 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3320                                        u8 *uuid, u8 *fsid)
3321 {
3322         struct btrfs_device *device;
3323         struct btrfs_fs_devices *cur_devices;
3324
3325         cur_devices = root->fs_info->fs_devices;
3326         while (cur_devices) {
3327                 if (!fsid ||
3328                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3329                         device = __find_device(&cur_devices->devices,
3330                                                devid, uuid);
3331                         if (device)
3332                                 return device;
3333                 }
3334                 cur_devices = cur_devices->seed;
3335         }
3336         return NULL;
3337 }
3338
3339 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3340                                             u64 devid, u8 *dev_uuid)
3341 {
3342         struct btrfs_device *device;
3343         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3344
3345         device = kzalloc(sizeof(*device), GFP_NOFS);
3346         if (!device)
3347                 return NULL;
3348         list_add(&device->dev_list,
3349                  &fs_devices->devices);
3350         device->dev_root = root->fs_info->dev_root;
3351         device->devid = devid;
3352         device->work.func = pending_bios_fn;
3353         device->fs_devices = fs_devices;
3354         device->missing = 1;
3355         fs_devices->num_devices++;
3356         fs_devices->missing_devices++;
3357         spin_lock_init(&device->io_lock);
3358         INIT_LIST_HEAD(&device->dev_alloc_list);
3359         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3360         return device;
3361 }
3362
3363 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3364                           struct extent_buffer *leaf,
3365                           struct btrfs_chunk *chunk)
3366 {
3367         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3368         struct map_lookup *map;
3369         struct extent_map *em;
3370         u64 logical;
3371         u64 length;
3372         u64 devid;
3373         u8 uuid[BTRFS_UUID_SIZE];
3374         int num_stripes;
3375         int ret;
3376         int i;
3377
3378         logical = key->offset;
3379         length = btrfs_chunk_length(leaf, chunk);
3380
3381         read_lock(&map_tree->map_tree.lock);
3382         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3383         read_unlock(&map_tree->map_tree.lock);
3384
3385         /* already mapped? */
3386         if (em && em->start <= logical && em->start + em->len > logical) {
3387                 free_extent_map(em);
3388                 return 0;
3389         } else if (em) {
3390                 free_extent_map(em);
3391         }
3392
3393         em = alloc_extent_map();
3394         if (!em)
3395                 return -ENOMEM;
3396         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3397         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3398         if (!map) {
3399                 free_extent_map(em);
3400                 return -ENOMEM;
3401         }
3402
3403         em->bdev = (struct block_device *)map;
3404         em->start = logical;
3405         em->len = length;
3406         em->block_start = 0;
3407         em->block_len = em->len;
3408
3409         map->num_stripes = num_stripes;
3410         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3411         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3412         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3413         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3414         map->type = btrfs_chunk_type(leaf, chunk);
3415         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3416         for (i = 0; i < num_stripes; i++) {
3417                 map->stripes[i].physical =
3418                         btrfs_stripe_offset_nr(leaf, chunk, i);
3419                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3420                 read_extent_buffer(leaf, uuid, (unsigned long)
3421                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3422                                    BTRFS_UUID_SIZE);
3423                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3424                                                         NULL);
3425                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3426                         kfree(map);
3427                         free_extent_map(em);
3428                         return -EIO;
3429                 }
3430                 if (!map->stripes[i].dev) {
3431                         map->stripes[i].dev =
3432                                 add_missing_dev(root, devid, uuid);
3433                         if (!map->stripes[i].dev) {
3434                                 kfree(map);
3435                                 free_extent_map(em);
3436                                 return -EIO;
3437                         }
3438                 }
3439                 map->stripes[i].dev->in_fs_metadata = 1;
3440         }
3441
3442         write_lock(&map_tree->map_tree.lock);
3443         ret = add_extent_mapping(&map_tree->map_tree, em);
3444         write_unlock(&map_tree->map_tree.lock);
3445         BUG_ON(ret);
3446         free_extent_map(em);
3447
3448         return 0;
3449 }
3450
3451 static int fill_device_from_item(struct extent_buffer *leaf,
3452                                  struct btrfs_dev_item *dev_item,
3453                                  struct btrfs_device *device)
3454 {
3455         unsigned long ptr;
3456
3457         device->devid = btrfs_device_id(leaf, dev_item);
3458         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3459         device->total_bytes = device->disk_total_bytes;
3460         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3461         device->type = btrfs_device_type(leaf, dev_item);
3462         device->io_align = btrfs_device_io_align(leaf, dev_item);
3463         device->io_width = btrfs_device_io_width(leaf, dev_item);
3464         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3465
3466         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3467         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3468
3469         return 0;
3470 }
3471
3472 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3473 {
3474         struct btrfs_fs_devices *fs_devices;
3475         int ret;
3476
3477         mutex_lock(&uuid_mutex);
3478
3479         fs_devices = root->fs_info->fs_devices->seed;
3480         while (fs_devices) {
3481                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3482                         ret = 0;
3483                         goto out;
3484                 }
3485                 fs_devices = fs_devices->seed;
3486         }
3487
3488         fs_devices = find_fsid(fsid);
3489         if (!fs_devices) {
3490                 ret = -ENOENT;
3491                 goto out;
3492         }
3493
3494         fs_devices = clone_fs_devices(fs_devices);
3495         if (IS_ERR(fs_devices)) {
3496                 ret = PTR_ERR(fs_devices);
3497                 goto out;
3498         }
3499
3500         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3501                                    root->fs_info->bdev_holder);
3502         if (ret)
3503                 goto out;
3504
3505         if (!fs_devices->seeding) {
3506                 __btrfs_close_devices(fs_devices);
3507                 free_fs_devices(fs_devices);
3508                 ret = -EINVAL;
3509                 goto out;
3510         }
3511
3512         fs_devices->seed = root->fs_info->fs_devices->seed;
3513         root->fs_info->fs_devices->seed = fs_devices;
3514 out:
3515         mutex_unlock(&uuid_mutex);
3516         return ret;
3517 }
3518
3519 static int read_one_dev(struct btrfs_root *root,
3520                         struct extent_buffer *leaf,
3521                         struct btrfs_dev_item *dev_item)
3522 {
3523         struct btrfs_device *device;
3524         u64 devid;
3525         int ret;
3526         u8 fs_uuid[BTRFS_UUID_SIZE];
3527         u8 dev_uuid[BTRFS_UUID_SIZE];
3528
3529         devid = btrfs_device_id(leaf, dev_item);
3530         read_extent_buffer(leaf, dev_uuid,
3531                            (unsigned long)btrfs_device_uuid(dev_item),
3532                            BTRFS_UUID_SIZE);
3533         read_extent_buffer(leaf, fs_uuid,
3534                            (unsigned long)btrfs_device_fsid(dev_item),
3535                            BTRFS_UUID_SIZE);
3536
3537         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3538                 ret = open_seed_devices(root, fs_uuid);
3539                 if (ret && !btrfs_test_opt(root, DEGRADED))
3540                         return ret;
3541         }
3542
3543         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3544         if (!device || !device->bdev) {
3545                 if (!btrfs_test_opt(root, DEGRADED))
3546                         return -EIO;
3547
3548                 if (!device) {
3549                         printk(KERN_WARNING "warning devid %llu missing\n",
3550                                (unsigned long long)devid);
3551                         device = add_missing_dev(root, devid, dev_uuid);
3552                         if (!device)
3553                                 return -ENOMEM;
3554                 } else if (!device->missing) {
3555                         /*
3556                          * this happens when a device that was properly setup
3557                          * in the device info lists suddenly goes bad.
3558                          * device->bdev is NULL, and so we have to set
3559                          * device->missing to one here
3560                          */
3561                         root->fs_info->fs_devices->missing_devices++;
3562                         device->missing = 1;
3563                 }
3564         }
3565
3566         if (device->fs_devices != root->fs_info->fs_devices) {
3567                 BUG_ON(device->writeable);
3568                 if (device->generation !=
3569                     btrfs_device_generation(leaf, dev_item))
3570                         return -EINVAL;
3571         }
3572
3573         fill_device_from_item(leaf, dev_item, device);
3574         device->dev_root = root->fs_info->dev_root;
3575         device->in_fs_metadata = 1;
3576         if (device->writeable)
3577                 device->fs_devices->total_rw_bytes += device->total_bytes;
3578         ret = 0;
3579         return ret;
3580 }
3581
3582 int btrfs_read_sys_array(struct btrfs_root *root)
3583 {
3584         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3585         struct extent_buffer *sb;
3586         struct btrfs_disk_key *disk_key;
3587         struct btrfs_chunk *chunk;
3588         u8 *ptr;
3589         unsigned long sb_ptr;
3590         int ret = 0;
3591         u32 num_stripes;
3592         u32 array_size;
3593         u32 len = 0;
3594         u32 cur;
3595         struct btrfs_key key;
3596
3597         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3598                                           BTRFS_SUPER_INFO_SIZE);
3599         if (!sb)
3600                 return -ENOMEM;
3601         btrfs_set_buffer_uptodate(sb);
3602         btrfs_set_buffer_lockdep_class(sb, 0);
3603
3604         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3605         array_size = btrfs_super_sys_array_size(super_copy);
3606
3607         ptr = super_copy->sys_chunk_array;
3608         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3609         cur = 0;
3610
3611         while (cur < array_size) {
3612                 disk_key = (struct btrfs_disk_key *)ptr;
3613                 btrfs_disk_key_to_cpu(&key, disk_key);
3614
3615                 len = sizeof(*disk_key); ptr += len;
3616                 sb_ptr += len;
3617                 cur += len;
3618
3619                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3620                         chunk = (struct btrfs_chunk *)sb_ptr;
3621                         ret = read_one_chunk(root, &key, sb, chunk);
3622                         if (ret)
3623                                 break;
3624                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3625                         len = btrfs_chunk_item_size(num_stripes);
3626                 } else {
3627                         ret = -EIO;
3628                         break;
3629                 }
3630                 ptr += len;
3631                 sb_ptr += len;
3632                 cur += len;
3633         }
3634         free_extent_buffer(sb);
3635         return ret;
3636 }
3637
3638 int btrfs_read_chunk_tree(struct btrfs_root *root)
3639 {
3640         struct btrfs_path *path;
3641         struct extent_buffer *leaf;
3642         struct btrfs_key key;
3643         struct btrfs_key found_key;
3644         int ret;
3645         int slot;
3646
3647         root = root->fs_info->chunk_root;
3648
3649         path = btrfs_alloc_path();
3650         if (!path)
3651                 return -ENOMEM;
3652
3653         /* first we search for all of the device items, and then we
3654          * read in all of the chunk items.  This way we can create chunk
3655          * mappings that reference all of the devices that are afound
3656          */
3657         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3658         key.offset = 0;
3659         key.type = 0;
3660 again:
3661         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3662         if (ret < 0)
3663                 goto error;
3664         while (1) {
3665                 leaf = path->nodes[0];
3666                 slot = path->slots[0];
3667                 if (slot >= btrfs_header_nritems(leaf)) {
3668                         ret = btrfs_next_leaf(root, path);
3669                         if (ret == 0)
3670                                 continue;
3671                         if (ret < 0)
3672                                 goto error;
3673                         break;
3674                 }
3675                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3676                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3677                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3678                                 break;
3679                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3680                                 struct btrfs_dev_item *dev_item;
3681                                 dev_item = btrfs_item_ptr(leaf, slot,
3682                                                   struct btrfs_dev_item);
3683                                 ret = read_one_dev(root, leaf, dev_item);
3684                                 if (ret)
3685                                         goto error;
3686                         }
3687                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3688                         struct btrfs_chunk *chunk;
3689                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3690                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3691                         if (ret)
3692                                 goto error;
3693                 }
3694                 path->slots[0]++;
3695         }
3696         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3697                 key.objectid = 0;
3698                 btrfs_release_path(path);
3699                 goto again;
3700         }
3701         ret = 0;
3702 error:
3703         btrfs_free_path(path);
3704         return ret;
3705 }