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