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