btrfs: false BUG_ON when degraded
[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                 /* FIXME, make a readl uuid parser */
694                 printk(KERN_INFO "device fsid %llx-%llx ",
695                        *(unsigned long long *)disk_super->fsid,
696                        *(unsigned long long *)(disk_super->fsid + 8));
697         }
698         printk(KERN_CONT "devid %llu transid %llu %s\n",
699                (unsigned long long)devid, (unsigned long long)transid, path);
700         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
701
702         brelse(bh);
703 error_close:
704         blkdev_put(bdev, flags);
705 error:
706         mutex_unlock(&uuid_mutex);
707         return ret;
708 }
709
710 /* helper to account the used device space in the range */
711 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
712                                    u64 end, u64 *length)
713 {
714         struct btrfs_key key;
715         struct btrfs_root *root = device->dev_root;
716         struct btrfs_dev_extent *dev_extent;
717         struct btrfs_path *path;
718         u64 extent_end;
719         int ret;
720         int slot;
721         struct extent_buffer *l;
722
723         *length = 0;
724
725         if (start >= device->total_bytes)
726                 return 0;
727
728         path = btrfs_alloc_path();
729         if (!path)
730                 return -ENOMEM;
731         path->reada = 2;
732
733         key.objectid = device->devid;
734         key.offset = start;
735         key.type = BTRFS_DEV_EXTENT_KEY;
736
737         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
738         if (ret < 0)
739                 goto out;
740         if (ret > 0) {
741                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
742                 if (ret < 0)
743                         goto out;
744         }
745
746         while (1) {
747                 l = path->nodes[0];
748                 slot = path->slots[0];
749                 if (slot >= btrfs_header_nritems(l)) {
750                         ret = btrfs_next_leaf(root, path);
751                         if (ret == 0)
752                                 continue;
753                         if (ret < 0)
754                                 goto out;
755
756                         break;
757                 }
758                 btrfs_item_key_to_cpu(l, &key, slot);
759
760                 if (key.objectid < device->devid)
761                         goto next;
762
763                 if (key.objectid > device->devid)
764                         break;
765
766                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
767                         goto next;
768
769                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
770                 extent_end = key.offset + btrfs_dev_extent_length(l,
771                                                                   dev_extent);
772                 if (key.offset <= start && extent_end > end) {
773                         *length = end - start + 1;
774                         break;
775                 } else if (key.offset <= start && extent_end > start)
776                         *length += extent_end - start;
777                 else if (key.offset > start && extent_end <= end)
778                         *length += extent_end - key.offset;
779                 else if (key.offset > start && key.offset <= end) {
780                         *length += end - key.offset + 1;
781                         break;
782                 } else if (key.offset > end)
783                         break;
784
785 next:
786                 path->slots[0]++;
787         }
788         ret = 0;
789 out:
790         btrfs_free_path(path);
791         return ret;
792 }
793
794 /*
795  * find_free_dev_extent - find free space in the specified device
796  * @trans:      transaction handler
797  * @device:     the device which we search the free space in
798  * @num_bytes:  the size of the free space that we need
799  * @start:      store the start of the free space.
800  * @len:        the size of the free space. that we find, or the size of the max
801  *              free space if we don't find suitable free space
802  *
803  * this uses a pretty simple search, the expectation is that it is
804  * called very infrequently and that a given device has a small number
805  * of extents
806  *
807  * @start is used to store the start of the free space if we find. But if we
808  * don't find suitable free space, it will be used to store the start position
809  * of the max free space.
810  *
811  * @len is used to store the size of the free space that we find.
812  * But if we don't find suitable free space, it is used to store the size of
813  * the max free space.
814  */
815 int find_free_dev_extent(struct btrfs_trans_handle *trans,
816                          struct btrfs_device *device, u64 num_bytes,
817                          u64 *start, u64 *len)
818 {
819         struct btrfs_key key;
820         struct btrfs_root *root = device->dev_root;
821         struct btrfs_dev_extent *dev_extent;
822         struct btrfs_path *path;
823         u64 hole_size;
824         u64 max_hole_start;
825         u64 max_hole_size;
826         u64 extent_end;
827         u64 search_start;
828         u64 search_end = device->total_bytes;
829         int ret;
830         int slot;
831         struct extent_buffer *l;
832
833         /* FIXME use last free of some kind */
834
835         /* we don't want to overwrite the superblock on the drive,
836          * so we make sure to start at an offset of at least 1MB
837          */
838         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
839
840         max_hole_start = search_start;
841         max_hole_size = 0;
842
843         if (search_start >= search_end) {
844                 ret = -ENOSPC;
845                 goto error;
846         }
847
848         path = btrfs_alloc_path();
849         if (!path) {
850                 ret = -ENOMEM;
851                 goto error;
852         }
853         path->reada = 2;
854
855         key.objectid = device->devid;
856         key.offset = search_start;
857         key.type = BTRFS_DEV_EXTENT_KEY;
858
859         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
860         if (ret < 0)
861                 goto out;
862         if (ret > 0) {
863                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
864                 if (ret < 0)
865                         goto out;
866         }
867
868         while (1) {
869                 l = path->nodes[0];
870                 slot = path->slots[0];
871                 if (slot >= btrfs_header_nritems(l)) {
872                         ret = btrfs_next_leaf(root, path);
873                         if (ret == 0)
874                                 continue;
875                         if (ret < 0)
876                                 goto out;
877
878                         break;
879                 }
880                 btrfs_item_key_to_cpu(l, &key, slot);
881
882                 if (key.objectid < device->devid)
883                         goto next;
884
885                 if (key.objectid > device->devid)
886                         break;
887
888                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
889                         goto next;
890
891                 if (key.offset > search_start) {
892                         hole_size = key.offset - search_start;
893
894                         if (hole_size > max_hole_size) {
895                                 max_hole_start = search_start;
896                                 max_hole_size = hole_size;
897                         }
898
899                         /*
900                          * If this free space is greater than which we need,
901                          * it must be the max free space that we have found
902                          * until now, so max_hole_start must point to the start
903                          * of this free space and the length of this free space
904                          * is stored in max_hole_size. Thus, we return
905                          * max_hole_start and max_hole_size and go back to the
906                          * caller.
907                          */
908                         if (hole_size >= num_bytes) {
909                                 ret = 0;
910                                 goto out;
911                         }
912                 }
913
914                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
915                 extent_end = key.offset + btrfs_dev_extent_length(l,
916                                                                   dev_extent);
917                 if (extent_end > search_start)
918                         search_start = extent_end;
919 next:
920                 path->slots[0]++;
921                 cond_resched();
922         }
923
924         hole_size = search_end- search_start;
925         if (hole_size > max_hole_size) {
926                 max_hole_start = search_start;
927                 max_hole_size = hole_size;
928         }
929
930         /* See above. */
931         if (hole_size < num_bytes)
932                 ret = -ENOSPC;
933         else
934                 ret = 0;
935
936 out:
937         btrfs_free_path(path);
938 error:
939         *start = max_hole_start;
940         if (len)
941                 *len = max_hole_size;
942         return ret;
943 }
944
945 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
946                           struct btrfs_device *device,
947                           u64 start)
948 {
949         int ret;
950         struct btrfs_path *path;
951         struct btrfs_root *root = device->dev_root;
952         struct btrfs_key key;
953         struct btrfs_key found_key;
954         struct extent_buffer *leaf = NULL;
955         struct btrfs_dev_extent *extent = NULL;
956
957         path = btrfs_alloc_path();
958         if (!path)
959                 return -ENOMEM;
960
961         key.objectid = device->devid;
962         key.offset = start;
963         key.type = BTRFS_DEV_EXTENT_KEY;
964
965         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
966         if (ret > 0) {
967                 ret = btrfs_previous_item(root, path, key.objectid,
968                                           BTRFS_DEV_EXTENT_KEY);
969                 if (ret)
970                         goto out;
971                 leaf = path->nodes[0];
972                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
973                 extent = btrfs_item_ptr(leaf, path->slots[0],
974                                         struct btrfs_dev_extent);
975                 BUG_ON(found_key.offset > start || found_key.offset +
976                        btrfs_dev_extent_length(leaf, extent) < start);
977         } else if (ret == 0) {
978                 leaf = path->nodes[0];
979                 extent = btrfs_item_ptr(leaf, path->slots[0],
980                                         struct btrfs_dev_extent);
981         }
982         BUG_ON(ret);
983
984         if (device->bytes_used > 0)
985                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
986         ret = btrfs_del_item(trans, root, path);
987
988 out:
989         btrfs_free_path(path);
990         return ret;
991 }
992
993 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
994                            struct btrfs_device *device,
995                            u64 chunk_tree, u64 chunk_objectid,
996                            u64 chunk_offset, u64 start, u64 num_bytes)
997 {
998         int ret;
999         struct btrfs_path *path;
1000         struct btrfs_root *root = device->dev_root;
1001         struct btrfs_dev_extent *extent;
1002         struct extent_buffer *leaf;
1003         struct btrfs_key key;
1004
1005         WARN_ON(!device->in_fs_metadata);
1006         path = btrfs_alloc_path();
1007         if (!path)
1008                 return -ENOMEM;
1009
1010         key.objectid = device->devid;
1011         key.offset = start;
1012         key.type = BTRFS_DEV_EXTENT_KEY;
1013         ret = btrfs_insert_empty_item(trans, root, path, &key,
1014                                       sizeof(*extent));
1015         BUG_ON(ret);
1016
1017         leaf = path->nodes[0];
1018         extent = btrfs_item_ptr(leaf, path->slots[0],
1019                                 struct btrfs_dev_extent);
1020         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1021         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1022         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1023
1024         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1025                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1026                     BTRFS_UUID_SIZE);
1027
1028         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1029         btrfs_mark_buffer_dirty(leaf);
1030         btrfs_free_path(path);
1031         return ret;
1032 }
1033
1034 static noinline int find_next_chunk(struct btrfs_root *root,
1035                                     u64 objectid, u64 *offset)
1036 {
1037         struct btrfs_path *path;
1038         int ret;
1039         struct btrfs_key key;
1040         struct btrfs_chunk *chunk;
1041         struct btrfs_key found_key;
1042
1043         path = btrfs_alloc_path();
1044         BUG_ON(!path);
1045
1046         key.objectid = objectid;
1047         key.offset = (u64)-1;
1048         key.type = BTRFS_CHUNK_ITEM_KEY;
1049
1050         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1051         if (ret < 0)
1052                 goto error;
1053
1054         BUG_ON(ret == 0);
1055
1056         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1057         if (ret) {
1058                 *offset = 0;
1059         } else {
1060                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1061                                       path->slots[0]);
1062                 if (found_key.objectid != objectid)
1063                         *offset = 0;
1064                 else {
1065                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1066                                                struct btrfs_chunk);
1067                         *offset = found_key.offset +
1068                                 btrfs_chunk_length(path->nodes[0], chunk);
1069                 }
1070         }
1071         ret = 0;
1072 error:
1073         btrfs_free_path(path);
1074         return ret;
1075 }
1076
1077 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1078 {
1079         int ret;
1080         struct btrfs_key key;
1081         struct btrfs_key found_key;
1082         struct btrfs_path *path;
1083
1084         root = root->fs_info->chunk_root;
1085
1086         path = btrfs_alloc_path();
1087         if (!path)
1088                 return -ENOMEM;
1089
1090         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1091         key.type = BTRFS_DEV_ITEM_KEY;
1092         key.offset = (u64)-1;
1093
1094         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1095         if (ret < 0)
1096                 goto error;
1097
1098         BUG_ON(ret == 0);
1099
1100         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1101                                   BTRFS_DEV_ITEM_KEY);
1102         if (ret) {
1103                 *objectid = 1;
1104         } else {
1105                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1106                                       path->slots[0]);
1107                 *objectid = found_key.offset + 1;
1108         }
1109         ret = 0;
1110 error:
1111         btrfs_free_path(path);
1112         return ret;
1113 }
1114
1115 /*
1116  * the device information is stored in the chunk root
1117  * the btrfs_device struct should be fully filled in
1118  */
1119 int btrfs_add_device(struct btrfs_trans_handle *trans,
1120                      struct btrfs_root *root,
1121                      struct btrfs_device *device)
1122 {
1123         int ret;
1124         struct btrfs_path *path;
1125         struct btrfs_dev_item *dev_item;
1126         struct extent_buffer *leaf;
1127         struct btrfs_key key;
1128         unsigned long ptr;
1129
1130         root = root->fs_info->chunk_root;
1131
1132         path = btrfs_alloc_path();
1133         if (!path)
1134                 return -ENOMEM;
1135
1136         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1137         key.type = BTRFS_DEV_ITEM_KEY;
1138         key.offset = device->devid;
1139
1140         ret = btrfs_insert_empty_item(trans, root, path, &key,
1141                                       sizeof(*dev_item));
1142         if (ret)
1143                 goto out;
1144
1145         leaf = path->nodes[0];
1146         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1147
1148         btrfs_set_device_id(leaf, dev_item, device->devid);
1149         btrfs_set_device_generation(leaf, dev_item, 0);
1150         btrfs_set_device_type(leaf, dev_item, device->type);
1151         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1152         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1153         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1154         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1155         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1156         btrfs_set_device_group(leaf, dev_item, 0);
1157         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1158         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1159         btrfs_set_device_start_offset(leaf, dev_item, 0);
1160
1161         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1162         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1163         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1164         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1165         btrfs_mark_buffer_dirty(leaf);
1166
1167         ret = 0;
1168 out:
1169         btrfs_free_path(path);
1170         return ret;
1171 }
1172
1173 static int btrfs_rm_dev_item(struct btrfs_root *root,
1174                              struct btrfs_device *device)
1175 {
1176         int ret;
1177         struct btrfs_path *path;
1178         struct btrfs_key key;
1179         struct btrfs_trans_handle *trans;
1180
1181         root = root->fs_info->chunk_root;
1182
1183         path = btrfs_alloc_path();
1184         if (!path)
1185                 return -ENOMEM;
1186
1187         trans = btrfs_start_transaction(root, 0);
1188         if (IS_ERR(trans)) {
1189                 btrfs_free_path(path);
1190                 return PTR_ERR(trans);
1191         }
1192         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1193         key.type = BTRFS_DEV_ITEM_KEY;
1194         key.offset = device->devid;
1195         lock_chunks(root);
1196
1197         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1198         if (ret < 0)
1199                 goto out;
1200
1201         if (ret > 0) {
1202                 ret = -ENOENT;
1203                 goto out;
1204         }
1205
1206         ret = btrfs_del_item(trans, root, path);
1207         if (ret)
1208                 goto out;
1209 out:
1210         btrfs_free_path(path);
1211         unlock_chunks(root);
1212         btrfs_commit_transaction(trans, root);
1213         return ret;
1214 }
1215
1216 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1217 {
1218         struct btrfs_device *device;
1219         struct btrfs_device *next_device;
1220         struct block_device *bdev;
1221         struct buffer_head *bh = NULL;
1222         struct btrfs_super_block *disk_super;
1223         struct btrfs_fs_devices *cur_devices;
1224         u64 all_avail;
1225         u64 devid;
1226         u64 num_devices;
1227         u8 *dev_uuid;
1228         int ret = 0;
1229         bool clear_super = false;
1230
1231         mutex_lock(&uuid_mutex);
1232         mutex_lock(&root->fs_info->volume_mutex);
1233
1234         all_avail = root->fs_info->avail_data_alloc_bits |
1235                 root->fs_info->avail_system_alloc_bits |
1236                 root->fs_info->avail_metadata_alloc_bits;
1237
1238         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1239             root->fs_info->fs_devices->num_devices <= 4) {
1240                 printk(KERN_ERR "btrfs: unable to go below four devices "
1241                        "on raid10\n");
1242                 ret = -EINVAL;
1243                 goto out;
1244         }
1245
1246         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1247             root->fs_info->fs_devices->num_devices <= 2) {
1248                 printk(KERN_ERR "btrfs: unable to go below two "
1249                        "devices on raid1\n");
1250                 ret = -EINVAL;
1251                 goto out;
1252         }
1253
1254         if (strcmp(device_path, "missing") == 0) {
1255                 struct list_head *devices;
1256                 struct btrfs_device *tmp;
1257
1258                 device = NULL;
1259                 devices = &root->fs_info->fs_devices->devices;
1260                 /*
1261                  * It is safe to read the devices since the volume_mutex
1262                  * is held.
1263                  */
1264                 list_for_each_entry(tmp, devices, dev_list) {
1265                         if (tmp->in_fs_metadata && !tmp->bdev) {
1266                                 device = tmp;
1267                                 break;
1268                         }
1269                 }
1270                 bdev = NULL;
1271                 bh = NULL;
1272                 disk_super = NULL;
1273                 if (!device) {
1274                         printk(KERN_ERR "btrfs: no missing devices found to "
1275                                "remove\n");
1276                         goto out;
1277                 }
1278         } else {
1279                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1280                                           root->fs_info->bdev_holder);
1281                 if (IS_ERR(bdev)) {
1282                         ret = PTR_ERR(bdev);
1283                         goto out;
1284                 }
1285
1286                 set_blocksize(bdev, 4096);
1287                 bh = btrfs_read_dev_super(bdev);
1288                 if (!bh) {
1289                         ret = -EINVAL;
1290                         goto error_close;
1291                 }
1292                 disk_super = (struct btrfs_super_block *)bh->b_data;
1293                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1294                 dev_uuid = disk_super->dev_item.uuid;
1295                 device = btrfs_find_device(root, devid, dev_uuid,
1296                                            disk_super->fsid);
1297                 if (!device) {
1298                         ret = -ENOENT;
1299                         goto error_brelse;
1300                 }
1301         }
1302
1303         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1304                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1305                        "device\n");
1306                 ret = -EINVAL;
1307                 goto error_brelse;
1308         }
1309
1310         if (device->writeable) {
1311                 lock_chunks(root);
1312                 list_del_init(&device->dev_alloc_list);
1313                 unlock_chunks(root);
1314                 root->fs_info->fs_devices->rw_devices--;
1315                 clear_super = true;
1316         }
1317
1318         ret = btrfs_shrink_device(device, 0);
1319         if (ret)
1320                 goto error_undo;
1321
1322         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1323         if (ret)
1324                 goto error_undo;
1325
1326         device->in_fs_metadata = 0;
1327         btrfs_scrub_cancel_dev(root, device);
1328
1329         /*
1330          * the device list mutex makes sure that we don't change
1331          * the device list while someone else is writing out all
1332          * the device supers.
1333          */
1334
1335         cur_devices = device->fs_devices;
1336         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1337         list_del_rcu(&device->dev_list);
1338
1339         device->fs_devices->num_devices--;
1340
1341         if (device->missing)
1342                 root->fs_info->fs_devices->missing_devices--;
1343
1344         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1345                                  struct btrfs_device, dev_list);
1346         if (device->bdev == root->fs_info->sb->s_bdev)
1347                 root->fs_info->sb->s_bdev = next_device->bdev;
1348         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1349                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1350
1351         if (device->bdev)
1352                 device->fs_devices->open_devices--;
1353
1354         call_rcu(&device->rcu, free_device);
1355         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1356
1357         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1358         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1359
1360         if (cur_devices->open_devices == 0) {
1361                 struct btrfs_fs_devices *fs_devices;
1362                 fs_devices = root->fs_info->fs_devices;
1363                 while (fs_devices) {
1364                         if (fs_devices->seed == cur_devices)
1365                                 break;
1366                         fs_devices = fs_devices->seed;
1367                 }
1368                 fs_devices->seed = cur_devices->seed;
1369                 cur_devices->seed = NULL;
1370                 lock_chunks(root);
1371                 __btrfs_close_devices(cur_devices);
1372                 unlock_chunks(root);
1373                 free_fs_devices(cur_devices);
1374         }
1375
1376         /*
1377          * at this point, the device is zero sized.  We want to
1378          * remove it from the devices list and zero out the old super
1379          */
1380         if (clear_super) {
1381                 /* make sure this device isn't detected as part of
1382                  * the FS anymore
1383                  */
1384                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1385                 set_buffer_dirty(bh);
1386                 sync_dirty_buffer(bh);
1387         }
1388
1389         ret = 0;
1390
1391 error_brelse:
1392         brelse(bh);
1393 error_close:
1394         if (bdev)
1395                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1396 out:
1397         mutex_unlock(&root->fs_info->volume_mutex);
1398         mutex_unlock(&uuid_mutex);
1399         return ret;
1400 error_undo:
1401         if (device->writeable) {
1402                 lock_chunks(root);
1403                 list_add(&device->dev_alloc_list,
1404                          &root->fs_info->fs_devices->alloc_list);
1405                 unlock_chunks(root);
1406                 root->fs_info->fs_devices->rw_devices++;
1407         }
1408         goto error_brelse;
1409 }
1410
1411 /*
1412  * does all the dirty work required for changing file system's UUID.
1413  */
1414 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1415                                 struct btrfs_root *root)
1416 {
1417         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1418         struct btrfs_fs_devices *old_devices;
1419         struct btrfs_fs_devices *seed_devices;
1420         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1421         struct btrfs_device *device;
1422         u64 super_flags;
1423
1424         BUG_ON(!mutex_is_locked(&uuid_mutex));
1425         if (!fs_devices->seeding)
1426                 return -EINVAL;
1427
1428         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1429         if (!seed_devices)
1430                 return -ENOMEM;
1431
1432         old_devices = clone_fs_devices(fs_devices);
1433         if (IS_ERR(old_devices)) {
1434                 kfree(seed_devices);
1435                 return PTR_ERR(old_devices);
1436         }
1437
1438         list_add(&old_devices->list, &fs_uuids);
1439
1440         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1441         seed_devices->opened = 1;
1442         INIT_LIST_HEAD(&seed_devices->devices);
1443         INIT_LIST_HEAD(&seed_devices->alloc_list);
1444         mutex_init(&seed_devices->device_list_mutex);
1445
1446         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1447         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1448                               synchronize_rcu);
1449         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1450
1451         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1452         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1453                 device->fs_devices = seed_devices;
1454         }
1455
1456         fs_devices->seeding = 0;
1457         fs_devices->num_devices = 0;
1458         fs_devices->open_devices = 0;
1459         fs_devices->seed = seed_devices;
1460
1461         generate_random_uuid(fs_devices->fsid);
1462         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1463         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1464         super_flags = btrfs_super_flags(disk_super) &
1465                       ~BTRFS_SUPER_FLAG_SEEDING;
1466         btrfs_set_super_flags(disk_super, super_flags);
1467
1468         return 0;
1469 }
1470
1471 /*
1472  * strore the expected generation for seed devices in device items.
1473  */
1474 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1475                                struct btrfs_root *root)
1476 {
1477         struct btrfs_path *path;
1478         struct extent_buffer *leaf;
1479         struct btrfs_dev_item *dev_item;
1480         struct btrfs_device *device;
1481         struct btrfs_key key;
1482         u8 fs_uuid[BTRFS_UUID_SIZE];
1483         u8 dev_uuid[BTRFS_UUID_SIZE];
1484         u64 devid;
1485         int ret;
1486
1487         path = btrfs_alloc_path();
1488         if (!path)
1489                 return -ENOMEM;
1490
1491         root = root->fs_info->chunk_root;
1492         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1493         key.offset = 0;
1494         key.type = BTRFS_DEV_ITEM_KEY;
1495
1496         while (1) {
1497                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1498                 if (ret < 0)
1499                         goto error;
1500
1501                 leaf = path->nodes[0];
1502 next_slot:
1503                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1504                         ret = btrfs_next_leaf(root, path);
1505                         if (ret > 0)
1506                                 break;
1507                         if (ret < 0)
1508                                 goto error;
1509                         leaf = path->nodes[0];
1510                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1511                         btrfs_release_path(path);
1512                         continue;
1513                 }
1514
1515                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1516                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1517                     key.type != BTRFS_DEV_ITEM_KEY)
1518                         break;
1519
1520                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1521                                           struct btrfs_dev_item);
1522                 devid = btrfs_device_id(leaf, dev_item);
1523                 read_extent_buffer(leaf, dev_uuid,
1524                                    (unsigned long)btrfs_device_uuid(dev_item),
1525                                    BTRFS_UUID_SIZE);
1526                 read_extent_buffer(leaf, fs_uuid,
1527                                    (unsigned long)btrfs_device_fsid(dev_item),
1528                                    BTRFS_UUID_SIZE);
1529                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1530                 BUG_ON(!device);
1531
1532                 if (device->fs_devices->seeding) {
1533                         btrfs_set_device_generation(leaf, dev_item,
1534                                                     device->generation);
1535                         btrfs_mark_buffer_dirty(leaf);
1536                 }
1537
1538                 path->slots[0]++;
1539                 goto next_slot;
1540         }
1541         ret = 0;
1542 error:
1543         btrfs_free_path(path);
1544         return ret;
1545 }
1546
1547 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1548 {
1549         struct btrfs_trans_handle *trans;
1550         struct btrfs_device *device;
1551         struct block_device *bdev;
1552         struct list_head *devices;
1553         struct super_block *sb = root->fs_info->sb;
1554         u64 total_bytes;
1555         int seeding_dev = 0;
1556         int ret = 0;
1557
1558         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1559                 return -EINVAL;
1560
1561         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1562                                   root->fs_info->bdev_holder);
1563         if (IS_ERR(bdev))
1564                 return PTR_ERR(bdev);
1565
1566         if (root->fs_info->fs_devices->seeding) {
1567                 seeding_dev = 1;
1568                 down_write(&sb->s_umount);
1569                 mutex_lock(&uuid_mutex);
1570         }
1571
1572         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1573         mutex_lock(&root->fs_info->volume_mutex);
1574
1575         devices = &root->fs_info->fs_devices->devices;
1576         /*
1577          * we have the volume lock, so we don't need the extra
1578          * device list mutex while reading the list here.
1579          */
1580         list_for_each_entry(device, devices, dev_list) {
1581                 if (device->bdev == bdev) {
1582                         ret = -EEXIST;
1583                         goto error;
1584                 }
1585         }
1586
1587         device = kzalloc(sizeof(*device), GFP_NOFS);
1588         if (!device) {
1589                 /* we can safely leave the fs_devices entry around */
1590                 ret = -ENOMEM;
1591                 goto error;
1592         }
1593
1594         device->name = kstrdup(device_path, GFP_NOFS);
1595         if (!device->name) {
1596                 kfree(device);
1597                 ret = -ENOMEM;
1598                 goto error;
1599         }
1600
1601         ret = find_next_devid(root, &device->devid);
1602         if (ret) {
1603                 kfree(device->name);
1604                 kfree(device);
1605                 goto error;
1606         }
1607
1608         trans = btrfs_start_transaction(root, 0);
1609         if (IS_ERR(trans)) {
1610                 kfree(device->name);
1611                 kfree(device);
1612                 ret = PTR_ERR(trans);
1613                 goto error;
1614         }
1615
1616         lock_chunks(root);
1617
1618         device->writeable = 1;
1619         device->work.func = pending_bios_fn;
1620         generate_random_uuid(device->uuid);
1621         spin_lock_init(&device->io_lock);
1622         device->generation = trans->transid;
1623         device->io_width = root->sectorsize;
1624         device->io_align = root->sectorsize;
1625         device->sector_size = root->sectorsize;
1626         device->total_bytes = i_size_read(bdev->bd_inode);
1627         device->disk_total_bytes = device->total_bytes;
1628         device->dev_root = root->fs_info->dev_root;
1629         device->bdev = bdev;
1630         device->in_fs_metadata = 1;
1631         device->mode = FMODE_EXCL;
1632         set_blocksize(device->bdev, 4096);
1633
1634         if (seeding_dev) {
1635                 sb->s_flags &= ~MS_RDONLY;
1636                 ret = btrfs_prepare_sprout(trans, root);
1637                 BUG_ON(ret);
1638         }
1639
1640         device->fs_devices = root->fs_info->fs_devices;
1641
1642         /*
1643          * we don't want write_supers to jump in here with our device
1644          * half setup
1645          */
1646         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1647         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1648         list_add(&device->dev_alloc_list,
1649                  &root->fs_info->fs_devices->alloc_list);
1650         root->fs_info->fs_devices->num_devices++;
1651         root->fs_info->fs_devices->open_devices++;
1652         root->fs_info->fs_devices->rw_devices++;
1653         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1654
1655         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1656                 root->fs_info->fs_devices->rotating = 1;
1657
1658         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1659         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1660                                     total_bytes + device->total_bytes);
1661
1662         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1663         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1664                                     total_bytes + 1);
1665         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1666
1667         if (seeding_dev) {
1668                 ret = init_first_rw_device(trans, root, device);
1669                 BUG_ON(ret);
1670                 ret = btrfs_finish_sprout(trans, root);
1671                 BUG_ON(ret);
1672         } else {
1673                 ret = btrfs_add_device(trans, root, device);
1674         }
1675
1676         /*
1677          * we've got more storage, clear any full flags on the space
1678          * infos
1679          */
1680         btrfs_clear_space_info_full(root->fs_info);
1681
1682         unlock_chunks(root);
1683         btrfs_commit_transaction(trans, root);
1684
1685         if (seeding_dev) {
1686                 mutex_unlock(&uuid_mutex);
1687                 up_write(&sb->s_umount);
1688
1689                 ret = btrfs_relocate_sys_chunks(root);
1690                 BUG_ON(ret);
1691         }
1692 out:
1693         mutex_unlock(&root->fs_info->volume_mutex);
1694         return ret;
1695 error:
1696         blkdev_put(bdev, FMODE_EXCL);
1697         if (seeding_dev) {
1698                 mutex_unlock(&uuid_mutex);
1699                 up_write(&sb->s_umount);
1700         }
1701         goto out;
1702 }
1703
1704 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1705                                         struct btrfs_device *device)
1706 {
1707         int ret;
1708         struct btrfs_path *path;
1709         struct btrfs_root *root;
1710         struct btrfs_dev_item *dev_item;
1711         struct extent_buffer *leaf;
1712         struct btrfs_key key;
1713
1714         root = device->dev_root->fs_info->chunk_root;
1715
1716         path = btrfs_alloc_path();
1717         if (!path)
1718                 return -ENOMEM;
1719
1720         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1721         key.type = BTRFS_DEV_ITEM_KEY;
1722         key.offset = device->devid;
1723
1724         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1725         if (ret < 0)
1726                 goto out;
1727
1728         if (ret > 0) {
1729                 ret = -ENOENT;
1730                 goto out;
1731         }
1732
1733         leaf = path->nodes[0];
1734         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1735
1736         btrfs_set_device_id(leaf, dev_item, device->devid);
1737         btrfs_set_device_type(leaf, dev_item, device->type);
1738         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1739         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1740         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1741         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1742         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1743         btrfs_mark_buffer_dirty(leaf);
1744
1745 out:
1746         btrfs_free_path(path);
1747         return ret;
1748 }
1749
1750 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1751                       struct btrfs_device *device, u64 new_size)
1752 {
1753         struct btrfs_super_block *super_copy =
1754                 &device->dev_root->fs_info->super_copy;
1755         u64 old_total = btrfs_super_total_bytes(super_copy);
1756         u64 diff = new_size - device->total_bytes;
1757
1758         if (!device->writeable)
1759                 return -EACCES;
1760         if (new_size <= device->total_bytes)
1761                 return -EINVAL;
1762
1763         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1764         device->fs_devices->total_rw_bytes += diff;
1765
1766         device->total_bytes = new_size;
1767         device->disk_total_bytes = new_size;
1768         btrfs_clear_space_info_full(device->dev_root->fs_info);
1769
1770         return btrfs_update_device(trans, device);
1771 }
1772
1773 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1774                       struct btrfs_device *device, u64 new_size)
1775 {
1776         int ret;
1777         lock_chunks(device->dev_root);
1778         ret = __btrfs_grow_device(trans, device, new_size);
1779         unlock_chunks(device->dev_root);
1780         return ret;
1781 }
1782
1783 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1784                             struct btrfs_root *root,
1785                             u64 chunk_tree, u64 chunk_objectid,
1786                             u64 chunk_offset)
1787 {
1788         int ret;
1789         struct btrfs_path *path;
1790         struct btrfs_key key;
1791
1792         root = root->fs_info->chunk_root;
1793         path = btrfs_alloc_path();
1794         if (!path)
1795                 return -ENOMEM;
1796
1797         key.objectid = chunk_objectid;
1798         key.offset = chunk_offset;
1799         key.type = BTRFS_CHUNK_ITEM_KEY;
1800
1801         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1802         BUG_ON(ret);
1803
1804         ret = btrfs_del_item(trans, root, path);
1805
1806         btrfs_free_path(path);
1807         return ret;
1808 }
1809
1810 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1811                         chunk_offset)
1812 {
1813         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1814         struct btrfs_disk_key *disk_key;
1815         struct btrfs_chunk *chunk;
1816         u8 *ptr;
1817         int ret = 0;
1818         u32 num_stripes;
1819         u32 array_size;
1820         u32 len = 0;
1821         u32 cur;
1822         struct btrfs_key key;
1823
1824         array_size = btrfs_super_sys_array_size(super_copy);
1825
1826         ptr = super_copy->sys_chunk_array;
1827         cur = 0;
1828
1829         while (cur < array_size) {
1830                 disk_key = (struct btrfs_disk_key *)ptr;
1831                 btrfs_disk_key_to_cpu(&key, disk_key);
1832
1833                 len = sizeof(*disk_key);
1834
1835                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1836                         chunk = (struct btrfs_chunk *)(ptr + len);
1837                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1838                         len += btrfs_chunk_item_size(num_stripes);
1839                 } else {
1840                         ret = -EIO;
1841                         break;
1842                 }
1843                 if (key.objectid == chunk_objectid &&
1844                     key.offset == chunk_offset) {
1845                         memmove(ptr, ptr + len, array_size - (cur + len));
1846                         array_size -= len;
1847                         btrfs_set_super_sys_array_size(super_copy, array_size);
1848                 } else {
1849                         ptr += len;
1850                         cur += len;
1851                 }
1852         }
1853         return ret;
1854 }
1855
1856 static int btrfs_relocate_chunk(struct btrfs_root *root,
1857                          u64 chunk_tree, u64 chunk_objectid,
1858                          u64 chunk_offset)
1859 {
1860         struct extent_map_tree *em_tree;
1861         struct btrfs_root *extent_root;
1862         struct btrfs_trans_handle *trans;
1863         struct extent_map *em;
1864         struct map_lookup *map;
1865         int ret;
1866         int i;
1867
1868         root = root->fs_info->chunk_root;
1869         extent_root = root->fs_info->extent_root;
1870         em_tree = &root->fs_info->mapping_tree.map_tree;
1871
1872         ret = btrfs_can_relocate(extent_root, chunk_offset);
1873         if (ret)
1874                 return -ENOSPC;
1875
1876         /* step one, relocate all the extents inside this chunk */
1877         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1878         if (ret)
1879                 return ret;
1880
1881         trans = btrfs_start_transaction(root, 0);
1882         BUG_ON(IS_ERR(trans));
1883
1884         lock_chunks(root);
1885
1886         /*
1887          * step two, delete the device extents and the
1888          * chunk tree entries
1889          */
1890         read_lock(&em_tree->lock);
1891         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1892         read_unlock(&em_tree->lock);
1893
1894         BUG_ON(em->start > chunk_offset ||
1895                em->start + em->len < chunk_offset);
1896         map = (struct map_lookup *)em->bdev;
1897
1898         for (i = 0; i < map->num_stripes; i++) {
1899                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1900                                             map->stripes[i].physical);
1901                 BUG_ON(ret);
1902
1903                 if (map->stripes[i].dev) {
1904                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1905                         BUG_ON(ret);
1906                 }
1907         }
1908         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1909                                chunk_offset);
1910
1911         BUG_ON(ret);
1912
1913         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1914
1915         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1916                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1917                 BUG_ON(ret);
1918         }
1919
1920         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1921         BUG_ON(ret);
1922
1923         write_lock(&em_tree->lock);
1924         remove_extent_mapping(em_tree, em);
1925         write_unlock(&em_tree->lock);
1926
1927         kfree(map);
1928         em->bdev = NULL;
1929
1930         /* once for the tree */
1931         free_extent_map(em);
1932         /* once for us */
1933         free_extent_map(em);
1934
1935         unlock_chunks(root);
1936         btrfs_end_transaction(trans, root);
1937         return 0;
1938 }
1939
1940 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1941 {
1942         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1943         struct btrfs_path *path;
1944         struct extent_buffer *leaf;
1945         struct btrfs_chunk *chunk;
1946         struct btrfs_key key;
1947         struct btrfs_key found_key;
1948         u64 chunk_tree = chunk_root->root_key.objectid;
1949         u64 chunk_type;
1950         bool retried = false;
1951         int failed = 0;
1952         int ret;
1953
1954         path = btrfs_alloc_path();
1955         if (!path)
1956                 return -ENOMEM;
1957
1958 again:
1959         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1960         key.offset = (u64)-1;
1961         key.type = BTRFS_CHUNK_ITEM_KEY;
1962
1963         while (1) {
1964                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1965                 if (ret < 0)
1966                         goto error;
1967                 BUG_ON(ret == 0);
1968
1969                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1970                                           key.type);
1971                 if (ret < 0)
1972                         goto error;
1973                 if (ret > 0)
1974                         break;
1975
1976                 leaf = path->nodes[0];
1977                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1978
1979                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1980                                        struct btrfs_chunk);
1981                 chunk_type = btrfs_chunk_type(leaf, chunk);
1982                 btrfs_release_path(path);
1983
1984                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1985                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1986                                                    found_key.objectid,
1987                                                    found_key.offset);
1988                         if (ret == -ENOSPC)
1989                                 failed++;
1990                         else if (ret)
1991                                 BUG();
1992                 }
1993
1994                 if (found_key.offset == 0)
1995                         break;
1996                 key.offset = found_key.offset - 1;
1997         }
1998         ret = 0;
1999         if (failed && !retried) {
2000                 failed = 0;
2001                 retried = true;
2002                 goto again;
2003         } else if (failed && retried) {
2004                 WARN_ON(1);
2005                 ret = -ENOSPC;
2006         }
2007 error:
2008         btrfs_free_path(path);
2009         return ret;
2010 }
2011
2012 static u64 div_factor(u64 num, int factor)
2013 {
2014         if (factor == 10)
2015                 return num;
2016         num *= factor;
2017         do_div(num, 10);
2018         return num;
2019 }
2020
2021 int btrfs_balance(struct btrfs_root *dev_root)
2022 {
2023         int ret;
2024         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2025         struct btrfs_device *device;
2026         u64 old_size;
2027         u64 size_to_free;
2028         struct btrfs_path *path;
2029         struct btrfs_key key;
2030         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2031         struct btrfs_trans_handle *trans;
2032         struct btrfs_key found_key;
2033
2034         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2035                 return -EROFS;
2036
2037         if (!capable(CAP_SYS_ADMIN))
2038                 return -EPERM;
2039
2040         mutex_lock(&dev_root->fs_info->volume_mutex);
2041         dev_root = dev_root->fs_info->dev_root;
2042
2043         /* step one make some room on all the devices */
2044         list_for_each_entry(device, devices, dev_list) {
2045                 old_size = device->total_bytes;
2046                 size_to_free = div_factor(old_size, 1);
2047                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2048                 if (!device->writeable ||
2049                     device->total_bytes - device->bytes_used > size_to_free)
2050                         continue;
2051
2052                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2053                 if (ret == -ENOSPC)
2054                         break;
2055                 BUG_ON(ret);
2056
2057                 trans = btrfs_start_transaction(dev_root, 0);
2058                 BUG_ON(IS_ERR(trans));
2059
2060                 ret = btrfs_grow_device(trans, device, old_size);
2061                 BUG_ON(ret);
2062
2063                 btrfs_end_transaction(trans, dev_root);
2064         }
2065
2066         /* step two, relocate all the chunks */
2067         path = btrfs_alloc_path();
2068         BUG_ON(!path);
2069
2070         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2071         key.offset = (u64)-1;
2072         key.type = BTRFS_CHUNK_ITEM_KEY;
2073
2074         while (1) {
2075                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2076                 if (ret < 0)
2077                         goto error;
2078
2079                 /*
2080                  * this shouldn't happen, it means the last relocate
2081                  * failed
2082                  */
2083                 if (ret == 0)
2084                         break;
2085
2086                 ret = btrfs_previous_item(chunk_root, path, 0,
2087                                           BTRFS_CHUNK_ITEM_KEY);
2088                 if (ret)
2089                         break;
2090
2091                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2092                                       path->slots[0]);
2093                 if (found_key.objectid != key.objectid)
2094                         break;
2095
2096                 /* chunk zero is special */
2097                 if (found_key.offset == 0)
2098                         break;
2099
2100                 btrfs_release_path(path);
2101                 ret = btrfs_relocate_chunk(chunk_root,
2102                                            chunk_root->root_key.objectid,
2103                                            found_key.objectid,
2104                                            found_key.offset);
2105                 BUG_ON(ret && ret != -ENOSPC);
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         BUG_ON(ret);
2668
2669         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2670                         (fs_info->metadata_alloc_profile &
2671                          fs_info->avail_metadata_alloc_bits);
2672         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2673
2674         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2675                                   &stripe_size, chunk_offset, alloc_profile);
2676         BUG_ON(ret);
2677
2678         sys_chunk_offset = chunk_offset + chunk_size;
2679
2680         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2681                         (fs_info->system_alloc_profile &
2682                          fs_info->avail_system_alloc_bits);
2683         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2684
2685         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2686                                   &sys_chunk_size, &sys_stripe_size,
2687                                   sys_chunk_offset, alloc_profile);
2688         BUG_ON(ret);
2689
2690         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2691         BUG_ON(ret);
2692
2693         /*
2694          * Modifying chunk tree needs allocating new blocks from both
2695          * system block group and metadata block group. So we only can
2696          * do operations require modifying the chunk tree after both
2697          * block groups were created.
2698          */
2699         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2700                                    chunk_size, stripe_size);
2701         BUG_ON(ret);
2702
2703         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2704                                    sys_chunk_offset, sys_chunk_size,
2705                                    sys_stripe_size);
2706         BUG_ON(ret);
2707         return 0;
2708 }
2709
2710 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2711 {
2712         struct extent_map *em;
2713         struct map_lookup *map;
2714         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2715         int readonly = 0;
2716         int i;
2717
2718         read_lock(&map_tree->map_tree.lock);
2719         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2720         read_unlock(&map_tree->map_tree.lock);
2721         if (!em)
2722                 return 1;
2723
2724         if (btrfs_test_opt(root, DEGRADED)) {
2725                 free_extent_map(em);
2726                 return 0;
2727         }
2728
2729         map = (struct map_lookup *)em->bdev;
2730         for (i = 0; i < map->num_stripes; i++) {
2731                 if (!map->stripes[i].dev->writeable) {
2732                         readonly = 1;
2733                         break;
2734                 }
2735         }
2736         free_extent_map(em);
2737         return readonly;
2738 }
2739
2740 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2741 {
2742         extent_map_tree_init(&tree->map_tree);
2743 }
2744
2745 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2746 {
2747         struct extent_map *em;
2748
2749         while (1) {
2750                 write_lock(&tree->map_tree.lock);
2751                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2752                 if (em)
2753                         remove_extent_mapping(&tree->map_tree, em);
2754                 write_unlock(&tree->map_tree.lock);
2755                 if (!em)
2756                         break;
2757                 kfree(em->bdev);
2758                 /* once for us */
2759                 free_extent_map(em);
2760                 /* once for the tree */
2761                 free_extent_map(em);
2762         }
2763 }
2764
2765 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2766 {
2767         struct extent_map *em;
2768         struct map_lookup *map;
2769         struct extent_map_tree *em_tree = &map_tree->map_tree;
2770         int ret;
2771
2772         read_lock(&em_tree->lock);
2773         em = lookup_extent_mapping(em_tree, logical, len);
2774         read_unlock(&em_tree->lock);
2775         BUG_ON(!em);
2776
2777         BUG_ON(em->start > logical || em->start + em->len < logical);
2778         map = (struct map_lookup *)em->bdev;
2779         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2780                 ret = map->num_stripes;
2781         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2782                 ret = map->sub_stripes;
2783         else
2784                 ret = 1;
2785         free_extent_map(em);
2786         return ret;
2787 }
2788
2789 static int find_live_mirror(struct map_lookup *map, int first, int num,
2790                             int optimal)
2791 {
2792         int i;
2793         if (map->stripes[optimal].dev->bdev)
2794                 return optimal;
2795         for (i = first; i < first + num; i++) {
2796                 if (map->stripes[i].dev->bdev)
2797                         return i;
2798         }
2799         /* we couldn't find one that doesn't fail.  Just return something
2800          * and the io error handling code will clean up eventually
2801          */
2802         return optimal;
2803 }
2804
2805 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2806                              u64 logical, u64 *length,
2807                              struct btrfs_multi_bio **multi_ret,
2808                              int mirror_num)
2809 {
2810         struct extent_map *em;
2811         struct map_lookup *map;
2812         struct extent_map_tree *em_tree = &map_tree->map_tree;
2813         u64 offset;
2814         u64 stripe_offset;
2815         u64 stripe_end_offset;
2816         u64 stripe_nr;
2817         u64 stripe_nr_orig;
2818         u64 stripe_nr_end;
2819         int stripes_allocated = 8;
2820         int stripes_required = 1;
2821         int stripe_index;
2822         int i;
2823         int num_stripes;
2824         int max_errors = 0;
2825         struct btrfs_multi_bio *multi = NULL;
2826
2827         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2828                 stripes_allocated = 1;
2829 again:
2830         if (multi_ret) {
2831                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2832                                 GFP_NOFS);
2833                 if (!multi)
2834                         return -ENOMEM;
2835
2836                 atomic_set(&multi->error, 0);
2837         }
2838
2839         read_lock(&em_tree->lock);
2840         em = lookup_extent_mapping(em_tree, logical, *length);
2841         read_unlock(&em_tree->lock);
2842
2843         if (!em) {
2844                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2845                        (unsigned long long)logical,
2846                        (unsigned long long)*length);
2847                 BUG();
2848         }
2849
2850         BUG_ON(em->start > logical || em->start + em->len < logical);
2851         map = (struct map_lookup *)em->bdev;
2852         offset = logical - em->start;
2853
2854         if (mirror_num > map->num_stripes)
2855                 mirror_num = 0;
2856
2857         /* if our multi bio struct is too small, back off and try again */
2858         if (rw & REQ_WRITE) {
2859                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2860                                  BTRFS_BLOCK_GROUP_DUP)) {
2861                         stripes_required = map->num_stripes;
2862                         max_errors = 1;
2863                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2864                         stripes_required = map->sub_stripes;
2865                         max_errors = 1;
2866                 }
2867         }
2868         if (rw & REQ_DISCARD) {
2869                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2870                                  BTRFS_BLOCK_GROUP_RAID1 |
2871                                  BTRFS_BLOCK_GROUP_DUP |
2872                                  BTRFS_BLOCK_GROUP_RAID10)) {
2873                         stripes_required = map->num_stripes;
2874                 }
2875         }
2876         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2877             stripes_allocated < stripes_required) {
2878                 stripes_allocated = map->num_stripes;
2879                 free_extent_map(em);
2880                 kfree(multi);
2881                 goto again;
2882         }
2883         stripe_nr = offset;
2884         /*
2885          * stripe_nr counts the total number of stripes we have to stride
2886          * to get to this block
2887          */
2888         do_div(stripe_nr, map->stripe_len);
2889
2890         stripe_offset = stripe_nr * map->stripe_len;
2891         BUG_ON(offset < stripe_offset);
2892
2893         /* stripe_offset is the offset of this block in its stripe*/
2894         stripe_offset = offset - stripe_offset;
2895
2896         if (rw & REQ_DISCARD)
2897                 *length = min_t(u64, em->len - offset, *length);
2898         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2899                               BTRFS_BLOCK_GROUP_RAID1 |
2900                               BTRFS_BLOCK_GROUP_RAID10 |
2901                               BTRFS_BLOCK_GROUP_DUP)) {
2902                 /* we limit the length of each bio to what fits in a stripe */
2903                 *length = min_t(u64, em->len - offset,
2904                                 map->stripe_len - stripe_offset);
2905         } else {
2906                 *length = em->len - offset;
2907         }
2908
2909         if (!multi_ret)
2910                 goto out;
2911
2912         num_stripes = 1;
2913         stripe_index = 0;
2914         stripe_nr_orig = stripe_nr;
2915         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2916                         (~(map->stripe_len - 1));
2917         do_div(stripe_nr_end, map->stripe_len);
2918         stripe_end_offset = stripe_nr_end * map->stripe_len -
2919                             (offset + *length);
2920         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2921                 if (rw & REQ_DISCARD)
2922                         num_stripes = min_t(u64, map->num_stripes,
2923                                             stripe_nr_end - stripe_nr_orig);
2924                 stripe_index = do_div(stripe_nr, map->num_stripes);
2925         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2926                 if (rw & (REQ_WRITE | REQ_DISCARD))
2927                         num_stripes = map->num_stripes;
2928                 else if (mirror_num)
2929                         stripe_index = mirror_num - 1;
2930                 else {
2931                         stripe_index = find_live_mirror(map, 0,
2932                                             map->num_stripes,
2933                                             current->pid % map->num_stripes);
2934                 }
2935
2936         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2937                 if (rw & (REQ_WRITE | REQ_DISCARD))
2938                         num_stripes = map->num_stripes;
2939                 else if (mirror_num)
2940                         stripe_index = mirror_num - 1;
2941
2942         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2943                 int factor = map->num_stripes / map->sub_stripes;
2944
2945                 stripe_index = do_div(stripe_nr, factor);
2946                 stripe_index *= map->sub_stripes;
2947
2948                 if (rw & REQ_WRITE)
2949                         num_stripes = map->sub_stripes;
2950                 else if (rw & REQ_DISCARD)
2951                         num_stripes = min_t(u64, map->sub_stripes *
2952                                             (stripe_nr_end - stripe_nr_orig),
2953                                             map->num_stripes);
2954                 else if (mirror_num)
2955                         stripe_index += mirror_num - 1;
2956                 else {
2957                         stripe_index = find_live_mirror(map, stripe_index,
2958                                               map->sub_stripes, stripe_index +
2959                                               current->pid % map->sub_stripes);
2960                 }
2961         } else {
2962                 /*
2963                  * after this do_div call, stripe_nr is the number of stripes
2964                  * on this device we have to walk to find the data, and
2965                  * stripe_index is the number of our device in the stripe array
2966                  */
2967                 stripe_index = do_div(stripe_nr, map->num_stripes);
2968         }
2969         BUG_ON(stripe_index >= map->num_stripes);
2970
2971         if (rw & REQ_DISCARD) {
2972                 for (i = 0; i < num_stripes; i++) {
2973                         multi->stripes[i].physical =
2974                                 map->stripes[stripe_index].physical +
2975                                 stripe_offset + stripe_nr * map->stripe_len;
2976                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2977
2978                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2979                                 u64 stripes;
2980                                 u32 last_stripe = 0;
2981                                 int j;
2982
2983                                 div_u64_rem(stripe_nr_end - 1,
2984                                             map->num_stripes,
2985                                             &last_stripe);
2986
2987                                 for (j = 0; j < map->num_stripes; j++) {
2988                                         u32 test;
2989
2990                                         div_u64_rem(stripe_nr_end - 1 - j,
2991                                                     map->num_stripes, &test);
2992                                         if (test == stripe_index)
2993                                                 break;
2994                                 }
2995                                 stripes = stripe_nr_end - 1 - j;
2996                                 do_div(stripes, map->num_stripes);
2997                                 multi->stripes[i].length = map->stripe_len *
2998                                         (stripes - stripe_nr + 1);
2999
3000                                 if (i == 0) {
3001                                         multi->stripes[i].length -=
3002                                                 stripe_offset;
3003                                         stripe_offset = 0;
3004                                 }
3005                                 if (stripe_index == last_stripe)
3006                                         multi->stripes[i].length -=
3007                                                 stripe_end_offset;
3008                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3009                                 u64 stripes;
3010                                 int j;
3011                                 int factor = map->num_stripes /
3012                                              map->sub_stripes;
3013                                 u32 last_stripe = 0;
3014
3015                                 div_u64_rem(stripe_nr_end - 1,
3016                                             factor, &last_stripe);
3017                                 last_stripe *= map->sub_stripes;
3018
3019                                 for (j = 0; j < factor; j++) {
3020                                         u32 test;
3021
3022                                         div_u64_rem(stripe_nr_end - 1 - j,
3023                                                     factor, &test);
3024
3025                                         if (test ==
3026                                             stripe_index / map->sub_stripes)
3027                                                 break;
3028                                 }
3029                                 stripes = stripe_nr_end - 1 - j;
3030                                 do_div(stripes, factor);
3031                                 multi->stripes[i].length = map->stripe_len *
3032                                         (stripes - stripe_nr + 1);
3033
3034                                 if (i < map->sub_stripes) {
3035                                         multi->stripes[i].length -=
3036                                                 stripe_offset;
3037                                         if (i == map->sub_stripes - 1)
3038                                                 stripe_offset = 0;
3039                                 }
3040                                 if (stripe_index >= last_stripe &&
3041                                     stripe_index <= (last_stripe +
3042                                                      map->sub_stripes - 1)) {
3043                                         multi->stripes[i].length -=
3044                                                 stripe_end_offset;
3045                                 }
3046                         } else
3047                                 multi->stripes[i].length = *length;
3048
3049                         stripe_index++;
3050                         if (stripe_index == map->num_stripes) {
3051                                 /* This could only happen for RAID0/10 */
3052                                 stripe_index = 0;
3053                                 stripe_nr++;
3054                         }
3055                 }
3056         } else {
3057                 for (i = 0; i < num_stripes; i++) {
3058                         multi->stripes[i].physical =
3059                                 map->stripes[stripe_index].physical +
3060                                 stripe_offset +
3061                                 stripe_nr * map->stripe_len;
3062                         multi->stripes[i].dev =
3063                                 map->stripes[stripe_index].dev;
3064                         stripe_index++;
3065                 }
3066         }
3067         if (multi_ret) {
3068                 *multi_ret = multi;
3069                 multi->num_stripes = num_stripes;
3070                 multi->max_errors = max_errors;
3071         }
3072 out:
3073         free_extent_map(em);
3074         return 0;
3075 }
3076
3077 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3078                       u64 logical, u64 *length,
3079                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3080 {
3081         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3082                                  mirror_num);
3083 }
3084
3085 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3086                      u64 chunk_start, u64 physical, u64 devid,
3087                      u64 **logical, int *naddrs, int *stripe_len)
3088 {
3089         struct extent_map_tree *em_tree = &map_tree->map_tree;
3090         struct extent_map *em;
3091         struct map_lookup *map;
3092         u64 *buf;
3093         u64 bytenr;
3094         u64 length;
3095         u64 stripe_nr;
3096         int i, j, nr = 0;
3097
3098         read_lock(&em_tree->lock);
3099         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3100         read_unlock(&em_tree->lock);
3101
3102         BUG_ON(!em || em->start != chunk_start);
3103         map = (struct map_lookup *)em->bdev;
3104
3105         length = em->len;
3106         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3107                 do_div(length, map->num_stripes / map->sub_stripes);
3108         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3109                 do_div(length, map->num_stripes);
3110
3111         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3112         BUG_ON(!buf);
3113
3114         for (i = 0; i < map->num_stripes; i++) {
3115                 if (devid && map->stripes[i].dev->devid != devid)
3116                         continue;
3117                 if (map->stripes[i].physical > physical ||
3118                     map->stripes[i].physical + length <= physical)
3119                         continue;
3120
3121                 stripe_nr = physical - map->stripes[i].physical;
3122                 do_div(stripe_nr, map->stripe_len);
3123
3124                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3125                         stripe_nr = stripe_nr * map->num_stripes + i;
3126                         do_div(stripe_nr, map->sub_stripes);
3127                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3128                         stripe_nr = stripe_nr * map->num_stripes + i;
3129                 }
3130                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3131                 WARN_ON(nr >= map->num_stripes);
3132                 for (j = 0; j < nr; j++) {
3133                         if (buf[j] == bytenr)
3134                                 break;
3135                 }
3136                 if (j == nr) {
3137                         WARN_ON(nr >= map->num_stripes);
3138                         buf[nr++] = bytenr;
3139                 }
3140         }
3141
3142         *logical = buf;
3143         *naddrs = nr;
3144         *stripe_len = map->stripe_len;
3145
3146         free_extent_map(em);
3147         return 0;
3148 }
3149
3150 static void end_bio_multi_stripe(struct bio *bio, int err)
3151 {
3152         struct btrfs_multi_bio *multi = bio->bi_private;
3153         int is_orig_bio = 0;
3154
3155         if (err)
3156                 atomic_inc(&multi->error);
3157
3158         if (bio == multi->orig_bio)
3159                 is_orig_bio = 1;
3160
3161         if (atomic_dec_and_test(&multi->stripes_pending)) {
3162                 if (!is_orig_bio) {
3163                         bio_put(bio);
3164                         bio = multi->orig_bio;
3165                 }
3166                 bio->bi_private = multi->private;
3167                 bio->bi_end_io = multi->end_io;
3168                 /* only send an error to the higher layers if it is
3169                  * beyond the tolerance of the multi-bio
3170                  */
3171                 if (atomic_read(&multi->error) > multi->max_errors) {
3172                         err = -EIO;
3173                 } else if (err) {
3174                         /*
3175                          * this bio is actually up to date, we didn't
3176                          * go over the max number of errors
3177                          */
3178                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3179                         err = 0;
3180                 }
3181                 kfree(multi);
3182
3183                 bio_endio(bio, err);
3184         } else if (!is_orig_bio) {
3185                 bio_put(bio);
3186         }
3187 }
3188
3189 struct async_sched {
3190         struct bio *bio;
3191         int rw;
3192         struct btrfs_fs_info *info;
3193         struct btrfs_work work;
3194 };
3195
3196 /*
3197  * see run_scheduled_bios for a description of why bios are collected for
3198  * async submit.
3199  *
3200  * This will add one bio to the pending list for a device and make sure
3201  * the work struct is scheduled.
3202  */
3203 static noinline int schedule_bio(struct btrfs_root *root,
3204                                  struct btrfs_device *device,
3205                                  int rw, struct bio *bio)
3206 {
3207         int should_queue = 1;
3208         struct btrfs_pending_bios *pending_bios;
3209
3210         /* don't bother with additional async steps for reads, right now */
3211         if (!(rw & REQ_WRITE)) {
3212                 bio_get(bio);
3213                 submit_bio(rw, bio);
3214                 bio_put(bio);
3215                 return 0;
3216         }
3217
3218         /*
3219          * nr_async_bios allows us to reliably return congestion to the
3220          * higher layers.  Otherwise, the async bio makes it appear we have
3221          * made progress against dirty pages when we've really just put it
3222          * on a queue for later
3223          */
3224         atomic_inc(&root->fs_info->nr_async_bios);
3225         WARN_ON(bio->bi_next);
3226         bio->bi_next = NULL;
3227         bio->bi_rw |= rw;
3228
3229         spin_lock(&device->io_lock);
3230         if (bio->bi_rw & REQ_SYNC)
3231                 pending_bios = &device->pending_sync_bios;
3232         else
3233                 pending_bios = &device->pending_bios;
3234
3235         if (pending_bios->tail)
3236                 pending_bios->tail->bi_next = bio;
3237
3238         pending_bios->tail = bio;
3239         if (!pending_bios->head)
3240                 pending_bios->head = bio;
3241         if (device->running_pending)
3242                 should_queue = 0;
3243
3244         spin_unlock(&device->io_lock);
3245
3246         if (should_queue)
3247                 btrfs_queue_worker(&root->fs_info->submit_workers,
3248                                    &device->work);
3249         return 0;
3250 }
3251
3252 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3253                   int mirror_num, int async_submit)
3254 {
3255         struct btrfs_mapping_tree *map_tree;
3256         struct btrfs_device *dev;
3257         struct bio *first_bio = bio;
3258         u64 logical = (u64)bio->bi_sector << 9;
3259         u64 length = 0;
3260         u64 map_length;
3261         struct btrfs_multi_bio *multi = NULL;
3262         int ret;
3263         int dev_nr = 0;
3264         int total_devs = 1;
3265
3266         length = bio->bi_size;
3267         map_tree = &root->fs_info->mapping_tree;
3268         map_length = length;
3269
3270         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3271                               mirror_num);
3272         BUG_ON(ret);
3273
3274         total_devs = multi->num_stripes;
3275         if (map_length < length) {
3276                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3277                        "len %llu\n", (unsigned long long)logical,
3278                        (unsigned long long)length,
3279                        (unsigned long long)map_length);
3280                 BUG();
3281         }
3282         multi->end_io = first_bio->bi_end_io;
3283         multi->private = first_bio->bi_private;
3284         multi->orig_bio = first_bio;
3285         atomic_set(&multi->stripes_pending, multi->num_stripes);
3286
3287         while (dev_nr < total_devs) {
3288                 if (total_devs > 1) {
3289                         if (dev_nr < total_devs - 1) {
3290                                 bio = bio_clone(first_bio, GFP_NOFS);
3291                                 BUG_ON(!bio);
3292                         } else {
3293                                 bio = first_bio;
3294                         }
3295                         bio->bi_private = multi;
3296                         bio->bi_end_io = end_bio_multi_stripe;
3297                 }
3298                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3299                 dev = multi->stripes[dev_nr].dev;
3300                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3301                         bio->bi_bdev = dev->bdev;
3302                         if (async_submit)
3303                                 schedule_bio(root, dev, rw, bio);
3304                         else
3305                                 submit_bio(rw, bio);
3306                 } else {
3307                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3308                         bio->bi_sector = logical >> 9;
3309                         bio_endio(bio, -EIO);
3310                 }
3311                 dev_nr++;
3312         }
3313         if (total_devs == 1)
3314                 kfree(multi);
3315         return 0;
3316 }
3317
3318 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3319                                        u8 *uuid, u8 *fsid)
3320 {
3321         struct btrfs_device *device;
3322         struct btrfs_fs_devices *cur_devices;
3323
3324         cur_devices = root->fs_info->fs_devices;
3325         while (cur_devices) {
3326                 if (!fsid ||
3327                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3328                         device = __find_device(&cur_devices->devices,
3329                                                devid, uuid);
3330                         if (device)
3331                                 return device;
3332                 }
3333                 cur_devices = cur_devices->seed;
3334         }
3335         return NULL;
3336 }
3337
3338 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3339                                             u64 devid, u8 *dev_uuid)
3340 {
3341         struct btrfs_device *device;
3342         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3343
3344         device = kzalloc(sizeof(*device), GFP_NOFS);
3345         if (!device)
3346                 return NULL;
3347         list_add(&device->dev_list,
3348                  &fs_devices->devices);
3349         device->dev_root = root->fs_info->dev_root;
3350         device->devid = devid;
3351         device->work.func = pending_bios_fn;
3352         device->fs_devices = fs_devices;
3353         device->missing = 1;
3354         fs_devices->num_devices++;
3355         fs_devices->missing_devices++;
3356         spin_lock_init(&device->io_lock);
3357         INIT_LIST_HEAD(&device->dev_alloc_list);
3358         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3359         return device;
3360 }
3361
3362 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3363                           struct extent_buffer *leaf,
3364                           struct btrfs_chunk *chunk)
3365 {
3366         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3367         struct map_lookup *map;
3368         struct extent_map *em;
3369         u64 logical;
3370         u64 length;
3371         u64 devid;
3372         u8 uuid[BTRFS_UUID_SIZE];
3373         int num_stripes;
3374         int ret;
3375         int i;
3376
3377         logical = key->offset;
3378         length = btrfs_chunk_length(leaf, chunk);
3379
3380         read_lock(&map_tree->map_tree.lock);
3381         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3382         read_unlock(&map_tree->map_tree.lock);
3383
3384         /* already mapped? */
3385         if (em && em->start <= logical && em->start + em->len > logical) {
3386                 free_extent_map(em);
3387                 return 0;
3388         } else if (em) {
3389                 free_extent_map(em);
3390         }
3391
3392         em = alloc_extent_map();
3393         if (!em)
3394                 return -ENOMEM;
3395         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3396         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3397         if (!map) {
3398                 free_extent_map(em);
3399                 return -ENOMEM;
3400         }
3401
3402         em->bdev = (struct block_device *)map;
3403         em->start = logical;
3404         em->len = length;
3405         em->block_start = 0;
3406         em->block_len = em->len;
3407
3408         map->num_stripes = num_stripes;
3409         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3410         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3411         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3412         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3413         map->type = btrfs_chunk_type(leaf, chunk);
3414         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3415         for (i = 0; i < num_stripes; i++) {
3416                 map->stripes[i].physical =
3417                         btrfs_stripe_offset_nr(leaf, chunk, i);
3418                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3419                 read_extent_buffer(leaf, uuid, (unsigned long)
3420                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3421                                    BTRFS_UUID_SIZE);
3422                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3423                                                         NULL);
3424                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3425                         kfree(map);
3426                         free_extent_map(em);
3427                         return -EIO;
3428                 }
3429                 if (!map->stripes[i].dev) {
3430                         map->stripes[i].dev =
3431                                 add_missing_dev(root, devid, uuid);
3432                         if (!map->stripes[i].dev) {
3433                                 kfree(map);
3434                                 free_extent_map(em);
3435                                 return -EIO;
3436                         }
3437                 }
3438                 map->stripes[i].dev->in_fs_metadata = 1;
3439         }
3440
3441         write_lock(&map_tree->map_tree.lock);
3442         ret = add_extent_mapping(&map_tree->map_tree, em);
3443         write_unlock(&map_tree->map_tree.lock);
3444         BUG_ON(ret);
3445         free_extent_map(em);
3446
3447         return 0;
3448 }
3449
3450 static int fill_device_from_item(struct extent_buffer *leaf,
3451                                  struct btrfs_dev_item *dev_item,
3452                                  struct btrfs_device *device)
3453 {
3454         unsigned long ptr;
3455
3456         device->devid = btrfs_device_id(leaf, dev_item);
3457         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3458         device->total_bytes = device->disk_total_bytes;
3459         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3460         device->type = btrfs_device_type(leaf, dev_item);
3461         device->io_align = btrfs_device_io_align(leaf, dev_item);
3462         device->io_width = btrfs_device_io_width(leaf, dev_item);
3463         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3464
3465         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3466         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3467
3468         return 0;
3469 }
3470
3471 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3472 {
3473         struct btrfs_fs_devices *fs_devices;
3474         int ret;
3475
3476         mutex_lock(&uuid_mutex);
3477
3478         fs_devices = root->fs_info->fs_devices->seed;
3479         while (fs_devices) {
3480                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3481                         ret = 0;
3482                         goto out;
3483                 }
3484                 fs_devices = fs_devices->seed;
3485         }
3486
3487         fs_devices = find_fsid(fsid);
3488         if (!fs_devices) {
3489                 ret = -ENOENT;
3490                 goto out;
3491         }
3492
3493         fs_devices = clone_fs_devices(fs_devices);
3494         if (IS_ERR(fs_devices)) {
3495                 ret = PTR_ERR(fs_devices);
3496                 goto out;
3497         }
3498
3499         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3500                                    root->fs_info->bdev_holder);
3501         if (ret)
3502                 goto out;
3503
3504         if (!fs_devices->seeding) {
3505                 __btrfs_close_devices(fs_devices);
3506                 free_fs_devices(fs_devices);
3507                 ret = -EINVAL;
3508                 goto out;
3509         }
3510
3511         fs_devices->seed = root->fs_info->fs_devices->seed;
3512         root->fs_info->fs_devices->seed = fs_devices;
3513 out:
3514         mutex_unlock(&uuid_mutex);
3515         return ret;
3516 }
3517
3518 static int read_one_dev(struct btrfs_root *root,
3519                         struct extent_buffer *leaf,
3520                         struct btrfs_dev_item *dev_item)
3521 {
3522         struct btrfs_device *device;
3523         u64 devid;
3524         int ret;
3525         u8 fs_uuid[BTRFS_UUID_SIZE];
3526         u8 dev_uuid[BTRFS_UUID_SIZE];
3527
3528         devid = btrfs_device_id(leaf, dev_item);
3529         read_extent_buffer(leaf, dev_uuid,
3530                            (unsigned long)btrfs_device_uuid(dev_item),
3531                            BTRFS_UUID_SIZE);
3532         read_extent_buffer(leaf, fs_uuid,
3533                            (unsigned long)btrfs_device_fsid(dev_item),
3534                            BTRFS_UUID_SIZE);
3535
3536         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3537                 ret = open_seed_devices(root, fs_uuid);
3538                 if (ret && !btrfs_test_opt(root, DEGRADED))
3539                         return ret;
3540         }
3541
3542         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3543         if (!device || !device->bdev) {
3544                 if (!btrfs_test_opt(root, DEGRADED))
3545                         return -EIO;
3546
3547                 if (!device) {
3548                         printk(KERN_WARNING "warning devid %llu missing\n",
3549                                (unsigned long long)devid);
3550                         device = add_missing_dev(root, devid, dev_uuid);
3551                         if (!device)
3552                                 return -ENOMEM;
3553                 } else if (!device->missing) {
3554                         /*
3555                          * this happens when a device that was properly setup
3556                          * in the device info lists suddenly goes bad.
3557                          * device->bdev is NULL, and so we have to set
3558                          * device->missing to one here
3559                          */
3560                         root->fs_info->fs_devices->missing_devices++;
3561                         device->missing = 1;
3562                 }
3563         }
3564
3565         if (device->fs_devices != root->fs_info->fs_devices) {
3566                 BUG_ON(device->writeable);
3567                 if (device->generation !=
3568                     btrfs_device_generation(leaf, dev_item))
3569                         return -EINVAL;
3570         }
3571
3572         fill_device_from_item(leaf, dev_item, device);
3573         device->dev_root = root->fs_info->dev_root;
3574         device->in_fs_metadata = 1;
3575         if (device->writeable)
3576                 device->fs_devices->total_rw_bytes += device->total_bytes;
3577         ret = 0;
3578         return ret;
3579 }
3580
3581 int btrfs_read_sys_array(struct btrfs_root *root)
3582 {
3583         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3584         struct extent_buffer *sb;
3585         struct btrfs_disk_key *disk_key;
3586         struct btrfs_chunk *chunk;
3587         u8 *ptr;
3588         unsigned long sb_ptr;
3589         int ret = 0;
3590         u32 num_stripes;
3591         u32 array_size;
3592         u32 len = 0;
3593         u32 cur;
3594         struct btrfs_key key;
3595
3596         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3597                                           BTRFS_SUPER_INFO_SIZE);
3598         if (!sb)
3599                 return -ENOMEM;
3600         btrfs_set_buffer_uptodate(sb);
3601         btrfs_set_buffer_lockdep_class(sb, 0);
3602
3603         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3604         array_size = btrfs_super_sys_array_size(super_copy);
3605
3606         ptr = super_copy->sys_chunk_array;
3607         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3608         cur = 0;
3609
3610         while (cur < array_size) {
3611                 disk_key = (struct btrfs_disk_key *)ptr;
3612                 btrfs_disk_key_to_cpu(&key, disk_key);
3613
3614                 len = sizeof(*disk_key); ptr += len;
3615                 sb_ptr += len;
3616                 cur += len;
3617
3618                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3619                         chunk = (struct btrfs_chunk *)sb_ptr;
3620                         ret = read_one_chunk(root, &key, sb, chunk);
3621                         if (ret)
3622                                 break;
3623                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3624                         len = btrfs_chunk_item_size(num_stripes);
3625                 } else {
3626                         ret = -EIO;
3627                         break;
3628                 }
3629                 ptr += len;
3630                 sb_ptr += len;
3631                 cur += len;
3632         }
3633         free_extent_buffer(sb);
3634         return ret;
3635 }
3636
3637 int btrfs_read_chunk_tree(struct btrfs_root *root)
3638 {
3639         struct btrfs_path *path;
3640         struct extent_buffer *leaf;
3641         struct btrfs_key key;
3642         struct btrfs_key found_key;
3643         int ret;
3644         int slot;
3645
3646         root = root->fs_info->chunk_root;
3647
3648         path = btrfs_alloc_path();
3649         if (!path)
3650                 return -ENOMEM;
3651
3652         /* first we search for all of the device items, and then we
3653          * read in all of the chunk items.  This way we can create chunk
3654          * mappings that reference all of the devices that are afound
3655          */
3656         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3657         key.offset = 0;
3658         key.type = 0;
3659 again:
3660         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3661         if (ret < 0)
3662                 goto error;
3663         while (1) {
3664                 leaf = path->nodes[0];
3665                 slot = path->slots[0];
3666                 if (slot >= btrfs_header_nritems(leaf)) {
3667                         ret = btrfs_next_leaf(root, path);
3668                         if (ret == 0)
3669                                 continue;
3670                         if (ret < 0)
3671                                 goto error;
3672                         break;
3673                 }
3674                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3675                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3676                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3677                                 break;
3678                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3679                                 struct btrfs_dev_item *dev_item;
3680                                 dev_item = btrfs_item_ptr(leaf, slot,
3681                                                   struct btrfs_dev_item);
3682                                 ret = read_one_dev(root, leaf, dev_item);
3683                                 if (ret)
3684                                         goto error;
3685                         }
3686                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3687                         struct btrfs_chunk *chunk;
3688                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3689                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3690                         if (ret)
3691                                 goto error;
3692                 }
3693                 path->slots[0]++;
3694         }
3695         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3696                 key.objectid = 0;
3697                 btrfs_release_path(path);
3698                 goto again;
3699         }
3700         ret = 0;
3701 error:
3702         btrfs_free_path(path);
3703         return ret;
3704 }