Merge branch 'for-linus-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
[muen/linux.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/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43 #include "sysfs.h"
44
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46         [BTRFS_RAID_RAID10] = {
47                 .sub_stripes    = 2,
48                 .dev_stripes    = 1,
49                 .devs_max       = 0,    /* 0 == as many as possible */
50                 .devs_min       = 4,
51                 .tolerated_failures = 1,
52                 .devs_increment = 2,
53                 .ncopies        = 2,
54         },
55         [BTRFS_RAID_RAID1] = {
56                 .sub_stripes    = 1,
57                 .dev_stripes    = 1,
58                 .devs_max       = 2,
59                 .devs_min       = 2,
60                 .tolerated_failures = 1,
61                 .devs_increment = 2,
62                 .ncopies        = 2,
63         },
64         [BTRFS_RAID_DUP] = {
65                 .sub_stripes    = 1,
66                 .dev_stripes    = 2,
67                 .devs_max       = 1,
68                 .devs_min       = 1,
69                 .tolerated_failures = 0,
70                 .devs_increment = 1,
71                 .ncopies        = 2,
72         },
73         [BTRFS_RAID_RAID0] = {
74                 .sub_stripes    = 1,
75                 .dev_stripes    = 1,
76                 .devs_max       = 0,
77                 .devs_min       = 2,
78                 .tolerated_failures = 0,
79                 .devs_increment = 1,
80                 .ncopies        = 1,
81         },
82         [BTRFS_RAID_SINGLE] = {
83                 .sub_stripes    = 1,
84                 .dev_stripes    = 1,
85                 .devs_max       = 1,
86                 .devs_min       = 1,
87                 .tolerated_failures = 0,
88                 .devs_increment = 1,
89                 .ncopies        = 1,
90         },
91         [BTRFS_RAID_RAID5] = {
92                 .sub_stripes    = 1,
93                 .dev_stripes    = 1,
94                 .devs_max       = 0,
95                 .devs_min       = 2,
96                 .tolerated_failures = 1,
97                 .devs_increment = 1,
98                 .ncopies        = 2,
99         },
100         [BTRFS_RAID_RAID6] = {
101                 .sub_stripes    = 1,
102                 .dev_stripes    = 1,
103                 .devs_max       = 0,
104                 .devs_min       = 3,
105                 .tolerated_failures = 2,
106                 .devs_increment = 1,
107                 .ncopies        = 3,
108         },
109 };
110
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
116         [BTRFS_RAID_SINGLE] = 0,
117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
119 };
120
121 /*
122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123  * condition is not met. Zero means there's no corresponding
124  * BTRFS_ERROR_DEV_*_NOT_MET value.
125  */
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129         [BTRFS_RAID_DUP]    = 0,
130         [BTRFS_RAID_RAID0]  = 0,
131         [BTRFS_RAID_SINGLE] = 0,
132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134 };
135
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137                                 struct btrfs_fs_info *fs_info);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
142
143 DEFINE_MUTEX(uuid_mutex);
144 static LIST_HEAD(fs_uuids);
145 struct list_head *btrfs_get_fs_uuids(void)
146 {
147         return &fs_uuids;
148 }
149
150 static struct btrfs_fs_devices *__alloc_fs_devices(void)
151 {
152         struct btrfs_fs_devices *fs_devs;
153
154         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
155         if (!fs_devs)
156                 return ERR_PTR(-ENOMEM);
157
158         mutex_init(&fs_devs->device_list_mutex);
159
160         INIT_LIST_HEAD(&fs_devs->devices);
161         INIT_LIST_HEAD(&fs_devs->resized_devices);
162         INIT_LIST_HEAD(&fs_devs->alloc_list);
163         INIT_LIST_HEAD(&fs_devs->list);
164
165         return fs_devs;
166 }
167
168 /**
169  * alloc_fs_devices - allocate struct btrfs_fs_devices
170  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
171  *              generated.
172  *
173  * Return: a pointer to a new &struct btrfs_fs_devices on success;
174  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
175  * can be destroyed with kfree() right away.
176  */
177 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
178 {
179         struct btrfs_fs_devices *fs_devs;
180
181         fs_devs = __alloc_fs_devices();
182         if (IS_ERR(fs_devs))
183                 return fs_devs;
184
185         if (fsid)
186                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
187         else
188                 generate_random_uuid(fs_devs->fsid);
189
190         return fs_devs;
191 }
192
193 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
194 {
195         struct btrfs_device *device;
196         WARN_ON(fs_devices->opened);
197         while (!list_empty(&fs_devices->devices)) {
198                 device = list_entry(fs_devices->devices.next,
199                                     struct btrfs_device, dev_list);
200                 list_del(&device->dev_list);
201                 rcu_string_free(device->name);
202                 kfree(device);
203         }
204         kfree(fs_devices);
205 }
206
207 static void btrfs_kobject_uevent(struct block_device *bdev,
208                                  enum kobject_action action)
209 {
210         int ret;
211
212         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
213         if (ret)
214                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
215                         action,
216                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
217                         &disk_to_dev(bdev->bd_disk)->kobj);
218 }
219
220 void btrfs_cleanup_fs_uuids(void)
221 {
222         struct btrfs_fs_devices *fs_devices;
223
224         while (!list_empty(&fs_uuids)) {
225                 fs_devices = list_entry(fs_uuids.next,
226                                         struct btrfs_fs_devices, list);
227                 list_del(&fs_devices->list);
228                 free_fs_devices(fs_devices);
229         }
230 }
231
232 static struct btrfs_device *__alloc_device(void)
233 {
234         struct btrfs_device *dev;
235
236         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
237         if (!dev)
238                 return ERR_PTR(-ENOMEM);
239
240         INIT_LIST_HEAD(&dev->dev_list);
241         INIT_LIST_HEAD(&dev->dev_alloc_list);
242         INIT_LIST_HEAD(&dev->resized_list);
243
244         spin_lock_init(&dev->io_lock);
245
246         spin_lock_init(&dev->reada_lock);
247         atomic_set(&dev->reada_in_flight, 0);
248         atomic_set(&dev->dev_stats_ccnt, 0);
249         btrfs_device_data_ordered_init(dev);
250         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
251         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252
253         return dev;
254 }
255
256 static noinline struct btrfs_device *__find_device(struct list_head *head,
257                                                    u64 devid, u8 *uuid)
258 {
259         struct btrfs_device *dev;
260
261         list_for_each_entry(dev, head, dev_list) {
262                 if (dev->devid == devid &&
263                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
264                         return dev;
265                 }
266         }
267         return NULL;
268 }
269
270 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
271 {
272         struct btrfs_fs_devices *fs_devices;
273
274         list_for_each_entry(fs_devices, &fs_uuids, list) {
275                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
276                         return fs_devices;
277         }
278         return NULL;
279 }
280
281 static int
282 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
283                       int flush, struct block_device **bdev,
284                       struct buffer_head **bh)
285 {
286         int ret;
287
288         *bdev = blkdev_get_by_path(device_path, flags, holder);
289
290         if (IS_ERR(*bdev)) {
291                 ret = PTR_ERR(*bdev);
292                 goto error;
293         }
294
295         if (flush)
296                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
297         ret = set_blocksize(*bdev, 4096);
298         if (ret) {
299                 blkdev_put(*bdev, flags);
300                 goto error;
301         }
302         invalidate_bdev(*bdev);
303         *bh = btrfs_read_dev_super(*bdev);
304         if (IS_ERR(*bh)) {
305                 ret = PTR_ERR(*bh);
306                 blkdev_put(*bdev, flags);
307                 goto error;
308         }
309
310         return 0;
311
312 error:
313         *bdev = NULL;
314         *bh = NULL;
315         return ret;
316 }
317
318 static void requeue_list(struct btrfs_pending_bios *pending_bios,
319                         struct bio *head, struct bio *tail)
320 {
321
322         struct bio *old_head;
323
324         old_head = pending_bios->head;
325         pending_bios->head = head;
326         if (pending_bios->tail)
327                 tail->bi_next = old_head;
328         else
329                 pending_bios->tail = tail;
330 }
331
332 /*
333  * we try to collect pending bios for a device so we don't get a large
334  * number of procs sending bios down to the same device.  This greatly
335  * improves the schedulers ability to collect and merge the bios.
336  *
337  * But, it also turns into a long list of bios to process and that is sure
338  * to eventually make the worker thread block.  The solution here is to
339  * make some progress and then put this work struct back at the end of
340  * the list if the block device is congested.  This way, multiple devices
341  * can make progress from a single worker thread.
342  */
343 static noinline void run_scheduled_bios(struct btrfs_device *device)
344 {
345         struct btrfs_fs_info *fs_info = device->fs_info;
346         struct bio *pending;
347         struct backing_dev_info *bdi;
348         struct btrfs_pending_bios *pending_bios;
349         struct bio *tail;
350         struct bio *cur;
351         int again = 0;
352         unsigned long num_run;
353         unsigned long batch_run = 0;
354         unsigned long limit;
355         unsigned long last_waited = 0;
356         int force_reg = 0;
357         int sync_pending = 0;
358         struct blk_plug plug;
359
360         /*
361          * this function runs all the bios we've collected for
362          * a particular device.  We don't want to wander off to
363          * another device without first sending all of these down.
364          * So, setup a plug here and finish it off before we return
365          */
366         blk_start_plug(&plug);
367
368         bdi = device->bdev->bd_bdi;
369         limit = btrfs_async_submit_limit(fs_info);
370         limit = limit * 2 / 3;
371
372 loop:
373         spin_lock(&device->io_lock);
374
375 loop_lock:
376         num_run = 0;
377
378         /* take all the bios off the list at once and process them
379          * later on (without the lock held).  But, remember the
380          * tail and other pointers so the bios can be properly reinserted
381          * into the list if we hit congestion
382          */
383         if (!force_reg && device->pending_sync_bios.head) {
384                 pending_bios = &device->pending_sync_bios;
385                 force_reg = 1;
386         } else {
387                 pending_bios = &device->pending_bios;
388                 force_reg = 0;
389         }
390
391         pending = pending_bios->head;
392         tail = pending_bios->tail;
393         WARN_ON(pending && !tail);
394
395         /*
396          * if pending was null this time around, no bios need processing
397          * at all and we can stop.  Otherwise it'll loop back up again
398          * and do an additional check so no bios are missed.
399          *
400          * device->running_pending is used to synchronize with the
401          * schedule_bio code.
402          */
403         if (device->pending_sync_bios.head == NULL &&
404             device->pending_bios.head == NULL) {
405                 again = 0;
406                 device->running_pending = 0;
407         } else {
408                 again = 1;
409                 device->running_pending = 1;
410         }
411
412         pending_bios->head = NULL;
413         pending_bios->tail = NULL;
414
415         spin_unlock(&device->io_lock);
416
417         while (pending) {
418
419                 rmb();
420                 /* we want to work on both lists, but do more bios on the
421                  * sync list than the regular list
422                  */
423                 if ((num_run > 32 &&
424                     pending_bios != &device->pending_sync_bios &&
425                     device->pending_sync_bios.head) ||
426                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
427                     device->pending_bios.head)) {
428                         spin_lock(&device->io_lock);
429                         requeue_list(pending_bios, pending, tail);
430                         goto loop_lock;
431                 }
432
433                 cur = pending;
434                 pending = pending->bi_next;
435                 cur->bi_next = NULL;
436
437                 /*
438                  * atomic_dec_return implies a barrier for waitqueue_active
439                  */
440                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
441                     waitqueue_active(&fs_info->async_submit_wait))
442                         wake_up(&fs_info->async_submit_wait);
443
444                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
445
446                 /*
447                  * if we're doing the sync list, record that our
448                  * plug has some sync requests on it
449                  *
450                  * If we're doing the regular list and there are
451                  * sync requests sitting around, unplug before
452                  * we add more
453                  */
454                 if (pending_bios == &device->pending_sync_bios) {
455                         sync_pending = 1;
456                 } else if (sync_pending) {
457                         blk_finish_plug(&plug);
458                         blk_start_plug(&plug);
459                         sync_pending = 0;
460                 }
461
462                 btrfsic_submit_bio(cur);
463                 num_run++;
464                 batch_run++;
465
466                 cond_resched();
467
468                 /*
469                  * we made progress, there is more work to do and the bdi
470                  * is now congested.  Back off and let other work structs
471                  * run instead
472                  */
473                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
474                     fs_info->fs_devices->open_devices > 1) {
475                         struct io_context *ioc;
476
477                         ioc = current->io_context;
478
479                         /*
480                          * the main goal here is that we don't want to
481                          * block if we're going to be able to submit
482                          * more requests without blocking.
483                          *
484                          * This code does two great things, it pokes into
485                          * the elevator code from a filesystem _and_
486                          * it makes assumptions about how batching works.
487                          */
488                         if (ioc && ioc->nr_batch_requests > 0 &&
489                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
490                             (last_waited == 0 ||
491                              ioc->last_waited == last_waited)) {
492                                 /*
493                                  * we want to go through our batch of
494                                  * requests and stop.  So, we copy out
495                                  * the ioc->last_waited time and test
496                                  * against it before looping
497                                  */
498                                 last_waited = ioc->last_waited;
499                                 cond_resched();
500                                 continue;
501                         }
502                         spin_lock(&device->io_lock);
503                         requeue_list(pending_bios, pending, tail);
504                         device->running_pending = 1;
505
506                         spin_unlock(&device->io_lock);
507                         btrfs_queue_work(fs_info->submit_workers,
508                                          &device->work);
509                         goto done;
510                 }
511                 /* unplug every 64 requests just for good measure */
512                 if (batch_run % 64 == 0) {
513                         blk_finish_plug(&plug);
514                         blk_start_plug(&plug);
515                         sync_pending = 0;
516                 }
517         }
518
519         cond_resched();
520         if (again)
521                 goto loop;
522
523         spin_lock(&device->io_lock);
524         if (device->pending_bios.head || device->pending_sync_bios.head)
525                 goto loop_lock;
526         spin_unlock(&device->io_lock);
527
528 done:
529         blk_finish_plug(&plug);
530 }
531
532 static void pending_bios_fn(struct btrfs_work *work)
533 {
534         struct btrfs_device *device;
535
536         device = container_of(work, struct btrfs_device, work);
537         run_scheduled_bios(device);
538 }
539
540
541 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
542 {
543         struct btrfs_fs_devices *fs_devs;
544         struct btrfs_device *dev;
545
546         if (!cur_dev->name)
547                 return;
548
549         list_for_each_entry(fs_devs, &fs_uuids, list) {
550                 int del = 1;
551
552                 if (fs_devs->opened)
553                         continue;
554                 if (fs_devs->seeding)
555                         continue;
556
557                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
558
559                         if (dev == cur_dev)
560                                 continue;
561                         if (!dev->name)
562                                 continue;
563
564                         /*
565                          * Todo: This won't be enough. What if the same device
566                          * comes back (with new uuid and) with its mapper path?
567                          * But for now, this does help as mostly an admin will
568                          * either use mapper or non mapper path throughout.
569                          */
570                         rcu_read_lock();
571                         del = strcmp(rcu_str_deref(dev->name),
572                                                 rcu_str_deref(cur_dev->name));
573                         rcu_read_unlock();
574                         if (!del)
575                                 break;
576                 }
577
578                 if (!del) {
579                         /* delete the stale device */
580                         if (fs_devs->num_devices == 1) {
581                                 btrfs_sysfs_remove_fsid(fs_devs);
582                                 list_del(&fs_devs->list);
583                                 free_fs_devices(fs_devs);
584                         } else {
585                                 fs_devs->num_devices--;
586                                 list_del(&dev->dev_list);
587                                 rcu_string_free(dev->name);
588                                 kfree(dev);
589                         }
590                         break;
591                 }
592         }
593 }
594
595 /*
596  * Add new device to list of registered devices
597  *
598  * Returns:
599  * 1   - first time device is seen
600  * 0   - device already known
601  * < 0 - error
602  */
603 static noinline int device_list_add(const char *path,
604                            struct btrfs_super_block *disk_super,
605                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
606 {
607         struct btrfs_device *device;
608         struct btrfs_fs_devices *fs_devices;
609         struct rcu_string *name;
610         int ret = 0;
611         u64 found_transid = btrfs_super_generation(disk_super);
612
613         fs_devices = find_fsid(disk_super->fsid);
614         if (!fs_devices) {
615                 fs_devices = alloc_fs_devices(disk_super->fsid);
616                 if (IS_ERR(fs_devices))
617                         return PTR_ERR(fs_devices);
618
619                 list_add(&fs_devices->list, &fs_uuids);
620
621                 device = NULL;
622         } else {
623                 device = __find_device(&fs_devices->devices, devid,
624                                        disk_super->dev_item.uuid);
625         }
626
627         if (!device) {
628                 if (fs_devices->opened)
629                         return -EBUSY;
630
631                 device = btrfs_alloc_device(NULL, &devid,
632                                             disk_super->dev_item.uuid);
633                 if (IS_ERR(device)) {
634                         /* we can safely leave the fs_devices entry around */
635                         return PTR_ERR(device);
636                 }
637
638                 name = rcu_string_strdup(path, GFP_NOFS);
639                 if (!name) {
640                         kfree(device);
641                         return -ENOMEM;
642                 }
643                 rcu_assign_pointer(device->name, name);
644
645                 mutex_lock(&fs_devices->device_list_mutex);
646                 list_add_rcu(&device->dev_list, &fs_devices->devices);
647                 fs_devices->num_devices++;
648                 mutex_unlock(&fs_devices->device_list_mutex);
649
650                 ret = 1;
651                 device->fs_devices = fs_devices;
652         } else if (!device->name || strcmp(device->name->str, path)) {
653                 /*
654                  * When FS is already mounted.
655                  * 1. If you are here and if the device->name is NULL that
656                  *    means this device was missing at time of FS mount.
657                  * 2. If you are here and if the device->name is different
658                  *    from 'path' that means either
659                  *      a. The same device disappeared and reappeared with
660                  *         different name. or
661                  *      b. The missing-disk-which-was-replaced, has
662                  *         reappeared now.
663                  *
664                  * We must allow 1 and 2a above. But 2b would be a spurious
665                  * and unintentional.
666                  *
667                  * Further in case of 1 and 2a above, the disk at 'path'
668                  * would have missed some transaction when it was away and
669                  * in case of 2a the stale bdev has to be updated as well.
670                  * 2b must not be allowed at all time.
671                  */
672
673                 /*
674                  * For now, we do allow update to btrfs_fs_device through the
675                  * btrfs dev scan cli after FS has been mounted.  We're still
676                  * tracking a problem where systems fail mount by subvolume id
677                  * when we reject replacement on a mounted FS.
678                  */
679                 if (!fs_devices->opened && found_transid < device->generation) {
680                         /*
681                          * That is if the FS is _not_ mounted and if you
682                          * are here, that means there is more than one
683                          * disk with same uuid and devid.We keep the one
684                          * with larger generation number or the last-in if
685                          * generation are equal.
686                          */
687                         return -EEXIST;
688                 }
689
690                 name = rcu_string_strdup(path, GFP_NOFS);
691                 if (!name)
692                         return -ENOMEM;
693                 rcu_string_free(device->name);
694                 rcu_assign_pointer(device->name, name);
695                 if (device->missing) {
696                         fs_devices->missing_devices--;
697                         device->missing = 0;
698                 }
699         }
700
701         /*
702          * Unmount does not free the btrfs_device struct but would zero
703          * generation along with most of the other members. So just update
704          * it back. We need it to pick the disk with largest generation
705          * (as above).
706          */
707         if (!fs_devices->opened)
708                 device->generation = found_transid;
709
710         /*
711          * if there is new btrfs on an already registered device,
712          * then remove the stale device entry.
713          */
714         if (ret > 0)
715                 btrfs_free_stale_device(device);
716
717         *fs_devices_ret = fs_devices;
718
719         return ret;
720 }
721
722 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
723 {
724         struct btrfs_fs_devices *fs_devices;
725         struct btrfs_device *device;
726         struct btrfs_device *orig_dev;
727
728         fs_devices = alloc_fs_devices(orig->fsid);
729         if (IS_ERR(fs_devices))
730                 return fs_devices;
731
732         mutex_lock(&orig->device_list_mutex);
733         fs_devices->total_devices = orig->total_devices;
734
735         /* We have held the volume lock, it is safe to get the devices. */
736         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
737                 struct rcu_string *name;
738
739                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
740                                             orig_dev->uuid);
741                 if (IS_ERR(device))
742                         goto error;
743
744                 /*
745                  * This is ok to do without rcu read locked because we hold the
746                  * uuid mutex so nothing we touch in here is going to disappear.
747                  */
748                 if (orig_dev->name) {
749                         name = rcu_string_strdup(orig_dev->name->str,
750                                         GFP_KERNEL);
751                         if (!name) {
752                                 kfree(device);
753                                 goto error;
754                         }
755                         rcu_assign_pointer(device->name, name);
756                 }
757
758                 list_add(&device->dev_list, &fs_devices->devices);
759                 device->fs_devices = fs_devices;
760                 fs_devices->num_devices++;
761         }
762         mutex_unlock(&orig->device_list_mutex);
763         return fs_devices;
764 error:
765         mutex_unlock(&orig->device_list_mutex);
766         free_fs_devices(fs_devices);
767         return ERR_PTR(-ENOMEM);
768 }
769
770 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
771 {
772         struct btrfs_device *device, *next;
773         struct btrfs_device *latest_dev = NULL;
774
775         mutex_lock(&uuid_mutex);
776 again:
777         /* This is the initialized path, it is safe to release the devices. */
778         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
779                 if (device->in_fs_metadata) {
780                         if (!device->is_tgtdev_for_dev_replace &&
781                             (!latest_dev ||
782                              device->generation > latest_dev->generation)) {
783                                 latest_dev = device;
784                         }
785                         continue;
786                 }
787
788                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
789                         /*
790                          * In the first step, keep the device which has
791                          * the correct fsid and the devid that is used
792                          * for the dev_replace procedure.
793                          * In the second step, the dev_replace state is
794                          * read from the device tree and it is known
795                          * whether the procedure is really active or
796                          * not, which means whether this device is
797                          * used or whether it should be removed.
798                          */
799                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
800                                 continue;
801                         }
802                 }
803                 if (device->bdev) {
804                         blkdev_put(device->bdev, device->mode);
805                         device->bdev = NULL;
806                         fs_devices->open_devices--;
807                 }
808                 if (device->writeable) {
809                         list_del_init(&device->dev_alloc_list);
810                         device->writeable = 0;
811                         if (!device->is_tgtdev_for_dev_replace)
812                                 fs_devices->rw_devices--;
813                 }
814                 list_del_init(&device->dev_list);
815                 fs_devices->num_devices--;
816                 rcu_string_free(device->name);
817                 kfree(device);
818         }
819
820         if (fs_devices->seed) {
821                 fs_devices = fs_devices->seed;
822                 goto again;
823         }
824
825         fs_devices->latest_bdev = latest_dev->bdev;
826
827         mutex_unlock(&uuid_mutex);
828 }
829
830 static void __free_device(struct work_struct *work)
831 {
832         struct btrfs_device *device;
833
834         device = container_of(work, struct btrfs_device, rcu_work);
835         rcu_string_free(device->name);
836         kfree(device);
837 }
838
839 static void free_device(struct rcu_head *head)
840 {
841         struct btrfs_device *device;
842
843         device = container_of(head, struct btrfs_device, rcu);
844
845         INIT_WORK(&device->rcu_work, __free_device);
846         schedule_work(&device->rcu_work);
847 }
848
849 static void btrfs_close_bdev(struct btrfs_device *device)
850 {
851         if (device->bdev && device->writeable) {
852                 sync_blockdev(device->bdev);
853                 invalidate_bdev(device->bdev);
854         }
855
856         if (device->bdev)
857                 blkdev_put(device->bdev, device->mode);
858 }
859
860 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
861 {
862         struct btrfs_fs_devices *fs_devices = device->fs_devices;
863         struct btrfs_device *new_device;
864         struct rcu_string *name;
865
866         if (device->bdev)
867                 fs_devices->open_devices--;
868
869         if (device->writeable &&
870             device->devid != BTRFS_DEV_REPLACE_DEVID) {
871                 list_del_init(&device->dev_alloc_list);
872                 fs_devices->rw_devices--;
873         }
874
875         if (device->missing)
876                 fs_devices->missing_devices--;
877
878         new_device = btrfs_alloc_device(NULL, &device->devid,
879                                         device->uuid);
880         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
881
882         /* Safe because we are under uuid_mutex */
883         if (device->name) {
884                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
885                 BUG_ON(!name); /* -ENOMEM */
886                 rcu_assign_pointer(new_device->name, name);
887         }
888
889         list_replace_rcu(&device->dev_list, &new_device->dev_list);
890         new_device->fs_devices = device->fs_devices;
891 }
892
893 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
894 {
895         struct btrfs_device *device, *tmp;
896         struct list_head pending_put;
897
898         INIT_LIST_HEAD(&pending_put);
899
900         if (--fs_devices->opened > 0)
901                 return 0;
902
903         mutex_lock(&fs_devices->device_list_mutex);
904         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
905                 btrfs_prepare_close_one_device(device);
906                 list_add(&device->dev_list, &pending_put);
907         }
908         mutex_unlock(&fs_devices->device_list_mutex);
909
910         /*
911          * btrfs_show_devname() is using the device_list_mutex,
912          * sometimes call to blkdev_put() leads vfs calling
913          * into this func. So do put outside of device_list_mutex,
914          * as of now.
915          */
916         while (!list_empty(&pending_put)) {
917                 device = list_first_entry(&pending_put,
918                                 struct btrfs_device, dev_list);
919                 list_del(&device->dev_list);
920                 btrfs_close_bdev(device);
921                 call_rcu(&device->rcu, free_device);
922         }
923
924         WARN_ON(fs_devices->open_devices);
925         WARN_ON(fs_devices->rw_devices);
926         fs_devices->opened = 0;
927         fs_devices->seeding = 0;
928
929         return 0;
930 }
931
932 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
933 {
934         struct btrfs_fs_devices *seed_devices = NULL;
935         int ret;
936
937         mutex_lock(&uuid_mutex);
938         ret = __btrfs_close_devices(fs_devices);
939         if (!fs_devices->opened) {
940                 seed_devices = fs_devices->seed;
941                 fs_devices->seed = NULL;
942         }
943         mutex_unlock(&uuid_mutex);
944
945         while (seed_devices) {
946                 fs_devices = seed_devices;
947                 seed_devices = fs_devices->seed;
948                 __btrfs_close_devices(fs_devices);
949                 free_fs_devices(fs_devices);
950         }
951         /*
952          * Wait for rcu kworkers under __btrfs_close_devices
953          * to finish all blkdev_puts so device is really
954          * free when umount is done.
955          */
956         rcu_barrier();
957         return ret;
958 }
959
960 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
961                                 fmode_t flags, void *holder)
962 {
963         struct request_queue *q;
964         struct block_device *bdev;
965         struct list_head *head = &fs_devices->devices;
966         struct btrfs_device *device;
967         struct btrfs_device *latest_dev = NULL;
968         struct buffer_head *bh;
969         struct btrfs_super_block *disk_super;
970         u64 devid;
971         int seeding = 1;
972         int ret = 0;
973
974         flags |= FMODE_EXCL;
975
976         list_for_each_entry(device, head, dev_list) {
977                 if (device->bdev)
978                         continue;
979                 if (!device->name)
980                         continue;
981
982                 /* Just open everything we can; ignore failures here */
983                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
984                                             &bdev, &bh))
985                         continue;
986
987                 disk_super = (struct btrfs_super_block *)bh->b_data;
988                 devid = btrfs_stack_device_id(&disk_super->dev_item);
989                 if (devid != device->devid)
990                         goto error_brelse;
991
992                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
993                            BTRFS_UUID_SIZE))
994                         goto error_brelse;
995
996                 device->generation = btrfs_super_generation(disk_super);
997                 if (!latest_dev ||
998                     device->generation > latest_dev->generation)
999                         latest_dev = device;
1000
1001                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1002                         device->writeable = 0;
1003                 } else {
1004                         device->writeable = !bdev_read_only(bdev);
1005                         seeding = 0;
1006                 }
1007
1008                 q = bdev_get_queue(bdev);
1009                 if (blk_queue_discard(q))
1010                         device->can_discard = 1;
1011
1012                 device->bdev = bdev;
1013                 device->in_fs_metadata = 0;
1014                 device->mode = flags;
1015
1016                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1017                         fs_devices->rotating = 1;
1018
1019                 fs_devices->open_devices++;
1020                 if (device->writeable &&
1021                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1022                         fs_devices->rw_devices++;
1023                         list_add(&device->dev_alloc_list,
1024                                  &fs_devices->alloc_list);
1025                 }
1026                 brelse(bh);
1027                 continue;
1028
1029 error_brelse:
1030                 brelse(bh);
1031                 blkdev_put(bdev, flags);
1032                 continue;
1033         }
1034         if (fs_devices->open_devices == 0) {
1035                 ret = -EINVAL;
1036                 goto out;
1037         }
1038         fs_devices->seeding = seeding;
1039         fs_devices->opened = 1;
1040         fs_devices->latest_bdev = latest_dev->bdev;
1041         fs_devices->total_rw_bytes = 0;
1042 out:
1043         return ret;
1044 }
1045
1046 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1047                        fmode_t flags, void *holder)
1048 {
1049         int ret;
1050
1051         mutex_lock(&uuid_mutex);
1052         if (fs_devices->opened) {
1053                 fs_devices->opened++;
1054                 ret = 0;
1055         } else {
1056                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1057         }
1058         mutex_unlock(&uuid_mutex);
1059         return ret;
1060 }
1061
1062 void btrfs_release_disk_super(struct page *page)
1063 {
1064         kunmap(page);
1065         put_page(page);
1066 }
1067
1068 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1069                 struct page **page, struct btrfs_super_block **disk_super)
1070 {
1071         void *p;
1072         pgoff_t index;
1073
1074         /* make sure our super fits in the device */
1075         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1076                 return 1;
1077
1078         /* make sure our super fits in the page */
1079         if (sizeof(**disk_super) > PAGE_SIZE)
1080                 return 1;
1081
1082         /* make sure our super doesn't straddle pages on disk */
1083         index = bytenr >> PAGE_SHIFT;
1084         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1085                 return 1;
1086
1087         /* pull in the page with our super */
1088         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1089                                    index, GFP_KERNEL);
1090
1091         if (IS_ERR_OR_NULL(*page))
1092                 return 1;
1093
1094         p = kmap(*page);
1095
1096         /* align our pointer to the offset of the super block */
1097         *disk_super = p + (bytenr & ~PAGE_MASK);
1098
1099         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1100             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1101                 btrfs_release_disk_super(*page);
1102                 return 1;
1103         }
1104
1105         if ((*disk_super)->label[0] &&
1106                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1107                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1108
1109         return 0;
1110 }
1111
1112 /*
1113  * Look for a btrfs signature on a device. This may be called out of the mount path
1114  * and we are not allowed to call set_blocksize during the scan. The superblock
1115  * is read via pagecache
1116  */
1117 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1118                           struct btrfs_fs_devices **fs_devices_ret)
1119 {
1120         struct btrfs_super_block *disk_super;
1121         struct block_device *bdev;
1122         struct page *page;
1123         int ret = -EINVAL;
1124         u64 devid;
1125         u64 transid;
1126         u64 total_devices;
1127         u64 bytenr;
1128
1129         /*
1130          * we would like to check all the supers, but that would make
1131          * a btrfs mount succeed after a mkfs from a different FS.
1132          * So, we need to add a special mount option to scan for
1133          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1134          */
1135         bytenr = btrfs_sb_offset(0);
1136         flags |= FMODE_EXCL;
1137         mutex_lock(&uuid_mutex);
1138
1139         bdev = blkdev_get_by_path(path, flags, holder);
1140         if (IS_ERR(bdev)) {
1141                 ret = PTR_ERR(bdev);
1142                 goto error;
1143         }
1144
1145         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1146                 goto error_bdev_put;
1147
1148         devid = btrfs_stack_device_id(&disk_super->dev_item);
1149         transid = btrfs_super_generation(disk_super);
1150         total_devices = btrfs_super_num_devices(disk_super);
1151
1152         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1153         if (ret > 0) {
1154                 if (disk_super->label[0]) {
1155                         pr_info("BTRFS: device label %s ", disk_super->label);
1156                 } else {
1157                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1158                 }
1159
1160                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1161                 ret = 0;
1162         }
1163         if (!ret && fs_devices_ret)
1164                 (*fs_devices_ret)->total_devices = total_devices;
1165
1166         btrfs_release_disk_super(page);
1167
1168 error_bdev_put:
1169         blkdev_put(bdev, flags);
1170 error:
1171         mutex_unlock(&uuid_mutex);
1172         return ret;
1173 }
1174
1175 /* helper to account the used device space in the range */
1176 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1177                                    u64 end, u64 *length)
1178 {
1179         struct btrfs_key key;
1180         struct btrfs_root *root = device->fs_info->dev_root;
1181         struct btrfs_dev_extent *dev_extent;
1182         struct btrfs_path *path;
1183         u64 extent_end;
1184         int ret;
1185         int slot;
1186         struct extent_buffer *l;
1187
1188         *length = 0;
1189
1190         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1191                 return 0;
1192
1193         path = btrfs_alloc_path();
1194         if (!path)
1195                 return -ENOMEM;
1196         path->reada = READA_FORWARD;
1197
1198         key.objectid = device->devid;
1199         key.offset = start;
1200         key.type = BTRFS_DEV_EXTENT_KEY;
1201
1202         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1203         if (ret < 0)
1204                 goto out;
1205         if (ret > 0) {
1206                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1207                 if (ret < 0)
1208                         goto out;
1209         }
1210
1211         while (1) {
1212                 l = path->nodes[0];
1213                 slot = path->slots[0];
1214                 if (slot >= btrfs_header_nritems(l)) {
1215                         ret = btrfs_next_leaf(root, path);
1216                         if (ret == 0)
1217                                 continue;
1218                         if (ret < 0)
1219                                 goto out;
1220
1221                         break;
1222                 }
1223                 btrfs_item_key_to_cpu(l, &key, slot);
1224
1225                 if (key.objectid < device->devid)
1226                         goto next;
1227
1228                 if (key.objectid > device->devid)
1229                         break;
1230
1231                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1232                         goto next;
1233
1234                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1235                 extent_end = key.offset + btrfs_dev_extent_length(l,
1236                                                                   dev_extent);
1237                 if (key.offset <= start && extent_end > end) {
1238                         *length = end - start + 1;
1239                         break;
1240                 } else if (key.offset <= start && extent_end > start)
1241                         *length += extent_end - start;
1242                 else if (key.offset > start && extent_end <= end)
1243                         *length += extent_end - key.offset;
1244                 else if (key.offset > start && key.offset <= end) {
1245                         *length += end - key.offset + 1;
1246                         break;
1247                 } else if (key.offset > end)
1248                         break;
1249
1250 next:
1251                 path->slots[0]++;
1252         }
1253         ret = 0;
1254 out:
1255         btrfs_free_path(path);
1256         return ret;
1257 }
1258
1259 static int contains_pending_extent(struct btrfs_transaction *transaction,
1260                                    struct btrfs_device *device,
1261                                    u64 *start, u64 len)
1262 {
1263         struct btrfs_fs_info *fs_info = device->fs_info;
1264         struct extent_map *em;
1265         struct list_head *search_list = &fs_info->pinned_chunks;
1266         int ret = 0;
1267         u64 physical_start = *start;
1268
1269         if (transaction)
1270                 search_list = &transaction->pending_chunks;
1271 again:
1272         list_for_each_entry(em, search_list, list) {
1273                 struct map_lookup *map;
1274                 int i;
1275
1276                 map = em->map_lookup;
1277                 for (i = 0; i < map->num_stripes; i++) {
1278                         u64 end;
1279
1280                         if (map->stripes[i].dev != device)
1281                                 continue;
1282                         if (map->stripes[i].physical >= physical_start + len ||
1283                             map->stripes[i].physical + em->orig_block_len <=
1284                             physical_start)
1285                                 continue;
1286                         /*
1287                          * Make sure that while processing the pinned list we do
1288                          * not override our *start with a lower value, because
1289                          * we can have pinned chunks that fall within this
1290                          * device hole and that have lower physical addresses
1291                          * than the pending chunks we processed before. If we
1292                          * do not take this special care we can end up getting
1293                          * 2 pending chunks that start at the same physical
1294                          * device offsets because the end offset of a pinned
1295                          * chunk can be equal to the start offset of some
1296                          * pending chunk.
1297                          */
1298                         end = map->stripes[i].physical + em->orig_block_len;
1299                         if (end > *start) {
1300                                 *start = end;
1301                                 ret = 1;
1302                         }
1303                 }
1304         }
1305         if (search_list != &fs_info->pinned_chunks) {
1306                 search_list = &fs_info->pinned_chunks;
1307                 goto again;
1308         }
1309
1310         return ret;
1311 }
1312
1313
1314 /*
1315  * find_free_dev_extent_start - find free space in the specified device
1316  * @device:       the device which we search the free space in
1317  * @num_bytes:    the size of the free space that we need
1318  * @search_start: the position from which to begin the search
1319  * @start:        store the start of the free space.
1320  * @len:          the size of the free space. that we find, or the size
1321  *                of the max free space if we don't find suitable free space
1322  *
1323  * this uses a pretty simple search, the expectation is that it is
1324  * called very infrequently and that a given device has a small number
1325  * of extents
1326  *
1327  * @start is used to store the start of the free space if we find. But if we
1328  * don't find suitable free space, it will be used to store the start position
1329  * of the max free space.
1330  *
1331  * @len is used to store the size of the free space that we find.
1332  * But if we don't find suitable free space, it is used to store the size of
1333  * the max free space.
1334  */
1335 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1336                                struct btrfs_device *device, u64 num_bytes,
1337                                u64 search_start, u64 *start, u64 *len)
1338 {
1339         struct btrfs_fs_info *fs_info = device->fs_info;
1340         struct btrfs_root *root = fs_info->dev_root;
1341         struct btrfs_key key;
1342         struct btrfs_dev_extent *dev_extent;
1343         struct btrfs_path *path;
1344         u64 hole_size;
1345         u64 max_hole_start;
1346         u64 max_hole_size;
1347         u64 extent_end;
1348         u64 search_end = device->total_bytes;
1349         int ret;
1350         int slot;
1351         struct extent_buffer *l;
1352         u64 min_search_start;
1353
1354         /*
1355          * We don't want to overwrite the superblock on the drive nor any area
1356          * used by the boot loader (grub for example), so we make sure to start
1357          * at an offset of at least 1MB.
1358          */
1359         min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1360         search_start = max(search_start, min_search_start);
1361
1362         path = btrfs_alloc_path();
1363         if (!path)
1364                 return -ENOMEM;
1365
1366         max_hole_start = search_start;
1367         max_hole_size = 0;
1368
1369 again:
1370         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1371                 ret = -ENOSPC;
1372                 goto out;
1373         }
1374
1375         path->reada = READA_FORWARD;
1376         path->search_commit_root = 1;
1377         path->skip_locking = 1;
1378
1379         key.objectid = device->devid;
1380         key.offset = search_start;
1381         key.type = BTRFS_DEV_EXTENT_KEY;
1382
1383         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1384         if (ret < 0)
1385                 goto out;
1386         if (ret > 0) {
1387                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1388                 if (ret < 0)
1389                         goto out;
1390         }
1391
1392         while (1) {
1393                 l = path->nodes[0];
1394                 slot = path->slots[0];
1395                 if (slot >= btrfs_header_nritems(l)) {
1396                         ret = btrfs_next_leaf(root, path);
1397                         if (ret == 0)
1398                                 continue;
1399                         if (ret < 0)
1400                                 goto out;
1401
1402                         break;
1403                 }
1404                 btrfs_item_key_to_cpu(l, &key, slot);
1405
1406                 if (key.objectid < device->devid)
1407                         goto next;
1408
1409                 if (key.objectid > device->devid)
1410                         break;
1411
1412                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1413                         goto next;
1414
1415                 if (key.offset > search_start) {
1416                         hole_size = key.offset - search_start;
1417
1418                         /*
1419                          * Have to check before we set max_hole_start, otherwise
1420                          * we could end up sending back this offset anyway.
1421                          */
1422                         if (contains_pending_extent(transaction, device,
1423                                                     &search_start,
1424                                                     hole_size)) {
1425                                 if (key.offset >= search_start) {
1426                                         hole_size = key.offset - search_start;
1427                                 } else {
1428                                         WARN_ON_ONCE(1);
1429                                         hole_size = 0;
1430                                 }
1431                         }
1432
1433                         if (hole_size > max_hole_size) {
1434                                 max_hole_start = search_start;
1435                                 max_hole_size = hole_size;
1436                         }
1437
1438                         /*
1439                          * If this free space is greater than which we need,
1440                          * it must be the max free space that we have found
1441                          * until now, so max_hole_start must point to the start
1442                          * of this free space and the length of this free space
1443                          * is stored in max_hole_size. Thus, we return
1444                          * max_hole_start and max_hole_size and go back to the
1445                          * caller.
1446                          */
1447                         if (hole_size >= num_bytes) {
1448                                 ret = 0;
1449                                 goto out;
1450                         }
1451                 }
1452
1453                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1454                 extent_end = key.offset + btrfs_dev_extent_length(l,
1455                                                                   dev_extent);
1456                 if (extent_end > search_start)
1457                         search_start = extent_end;
1458 next:
1459                 path->slots[0]++;
1460                 cond_resched();
1461         }
1462
1463         /*
1464          * At this point, search_start should be the end of
1465          * allocated dev extents, and when shrinking the device,
1466          * search_end may be smaller than search_start.
1467          */
1468         if (search_end > search_start) {
1469                 hole_size = search_end - search_start;
1470
1471                 if (contains_pending_extent(transaction, device, &search_start,
1472                                             hole_size)) {
1473                         btrfs_release_path(path);
1474                         goto again;
1475                 }
1476
1477                 if (hole_size > max_hole_size) {
1478                         max_hole_start = search_start;
1479                         max_hole_size = hole_size;
1480                 }
1481         }
1482
1483         /* See above. */
1484         if (max_hole_size < num_bytes)
1485                 ret = -ENOSPC;
1486         else
1487                 ret = 0;
1488
1489 out:
1490         btrfs_free_path(path);
1491         *start = max_hole_start;
1492         if (len)
1493                 *len = max_hole_size;
1494         return ret;
1495 }
1496
1497 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1498                          struct btrfs_device *device, u64 num_bytes,
1499                          u64 *start, u64 *len)
1500 {
1501         /* FIXME use last free of some kind */
1502         return find_free_dev_extent_start(trans->transaction, device,
1503                                           num_bytes, 0, start, len);
1504 }
1505
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1507                           struct btrfs_device *device,
1508                           u64 start, u64 *dev_extent_len)
1509 {
1510         struct btrfs_fs_info *fs_info = device->fs_info;
1511         struct btrfs_root *root = fs_info->dev_root;
1512         int ret;
1513         struct btrfs_path *path;
1514         struct btrfs_key key;
1515         struct btrfs_key found_key;
1516         struct extent_buffer *leaf = NULL;
1517         struct btrfs_dev_extent *extent = NULL;
1518
1519         path = btrfs_alloc_path();
1520         if (!path)
1521                 return -ENOMEM;
1522
1523         key.objectid = device->devid;
1524         key.offset = start;
1525         key.type = BTRFS_DEV_EXTENT_KEY;
1526 again:
1527         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1528         if (ret > 0) {
1529                 ret = btrfs_previous_item(root, path, key.objectid,
1530                                           BTRFS_DEV_EXTENT_KEY);
1531                 if (ret)
1532                         goto out;
1533                 leaf = path->nodes[0];
1534                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535                 extent = btrfs_item_ptr(leaf, path->slots[0],
1536                                         struct btrfs_dev_extent);
1537                 BUG_ON(found_key.offset > start || found_key.offset +
1538                        btrfs_dev_extent_length(leaf, extent) < start);
1539                 key = found_key;
1540                 btrfs_release_path(path);
1541                 goto again;
1542         } else if (ret == 0) {
1543                 leaf = path->nodes[0];
1544                 extent = btrfs_item_ptr(leaf, path->slots[0],
1545                                         struct btrfs_dev_extent);
1546         } else {
1547                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1548                 goto out;
1549         }
1550
1551         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1552
1553         ret = btrfs_del_item(trans, root, path);
1554         if (ret) {
1555                 btrfs_handle_fs_error(fs_info, ret,
1556                                       "Failed to remove dev extent item");
1557         } else {
1558                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1559         }
1560 out:
1561         btrfs_free_path(path);
1562         return ret;
1563 }
1564
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566                                   struct btrfs_device *device,
1567                                   u64 chunk_tree, u64 chunk_objectid,
1568                                   u64 chunk_offset, u64 start, u64 num_bytes)
1569 {
1570         int ret;
1571         struct btrfs_path *path;
1572         struct btrfs_fs_info *fs_info = device->fs_info;
1573         struct btrfs_root *root = fs_info->dev_root;
1574         struct btrfs_dev_extent *extent;
1575         struct extent_buffer *leaf;
1576         struct btrfs_key key;
1577
1578         WARN_ON(!device->in_fs_metadata);
1579         WARN_ON(device->is_tgtdev_for_dev_replace);
1580         path = btrfs_alloc_path();
1581         if (!path)
1582                 return -ENOMEM;
1583
1584         key.objectid = device->devid;
1585         key.offset = start;
1586         key.type = BTRFS_DEV_EXTENT_KEY;
1587         ret = btrfs_insert_empty_item(trans, root, path, &key,
1588                                       sizeof(*extent));
1589         if (ret)
1590                 goto out;
1591
1592         leaf = path->nodes[0];
1593         extent = btrfs_item_ptr(leaf, path->slots[0],
1594                                 struct btrfs_dev_extent);
1595         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1596         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1597         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1598
1599         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1600
1601         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1602         btrfs_mark_buffer_dirty(leaf);
1603 out:
1604         btrfs_free_path(path);
1605         return ret;
1606 }
1607
1608 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1609 {
1610         struct extent_map_tree *em_tree;
1611         struct extent_map *em;
1612         struct rb_node *n;
1613         u64 ret = 0;
1614
1615         em_tree = &fs_info->mapping_tree.map_tree;
1616         read_lock(&em_tree->lock);
1617         n = rb_last(&em_tree->map);
1618         if (n) {
1619                 em = rb_entry(n, struct extent_map, rb_node);
1620                 ret = em->start + em->len;
1621         }
1622         read_unlock(&em_tree->lock);
1623
1624         return ret;
1625 }
1626
1627 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1628                                     u64 *devid_ret)
1629 {
1630         int ret;
1631         struct btrfs_key key;
1632         struct btrfs_key found_key;
1633         struct btrfs_path *path;
1634
1635         path = btrfs_alloc_path();
1636         if (!path)
1637                 return -ENOMEM;
1638
1639         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1640         key.type = BTRFS_DEV_ITEM_KEY;
1641         key.offset = (u64)-1;
1642
1643         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1644         if (ret < 0)
1645                 goto error;
1646
1647         BUG_ON(ret == 0); /* Corruption */
1648
1649         ret = btrfs_previous_item(fs_info->chunk_root, path,
1650                                   BTRFS_DEV_ITEMS_OBJECTID,
1651                                   BTRFS_DEV_ITEM_KEY);
1652         if (ret) {
1653                 *devid_ret = 1;
1654         } else {
1655                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656                                       path->slots[0]);
1657                 *devid_ret = found_key.offset + 1;
1658         }
1659         ret = 0;
1660 error:
1661         btrfs_free_path(path);
1662         return ret;
1663 }
1664
1665 /*
1666  * the device information is stored in the chunk root
1667  * the btrfs_device struct should be fully filled in
1668  */
1669 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1670                             struct btrfs_fs_info *fs_info,
1671                             struct btrfs_device *device)
1672 {
1673         struct btrfs_root *root = fs_info->chunk_root;
1674         int ret;
1675         struct btrfs_path *path;
1676         struct btrfs_dev_item *dev_item;
1677         struct extent_buffer *leaf;
1678         struct btrfs_key key;
1679         unsigned long ptr;
1680
1681         path = btrfs_alloc_path();
1682         if (!path)
1683                 return -ENOMEM;
1684
1685         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1686         key.type = BTRFS_DEV_ITEM_KEY;
1687         key.offset = device->devid;
1688
1689         ret = btrfs_insert_empty_item(trans, root, path, &key,
1690                                       sizeof(*dev_item));
1691         if (ret)
1692                 goto out;
1693
1694         leaf = path->nodes[0];
1695         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1696
1697         btrfs_set_device_id(leaf, dev_item, device->devid);
1698         btrfs_set_device_generation(leaf, dev_item, 0);
1699         btrfs_set_device_type(leaf, dev_item, device->type);
1700         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1701         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1702         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1703         btrfs_set_device_total_bytes(leaf, dev_item,
1704                                      btrfs_device_get_disk_total_bytes(device));
1705         btrfs_set_device_bytes_used(leaf, dev_item,
1706                                     btrfs_device_get_bytes_used(device));
1707         btrfs_set_device_group(leaf, dev_item, 0);
1708         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1709         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1710         btrfs_set_device_start_offset(leaf, dev_item, 0);
1711
1712         ptr = btrfs_device_uuid(dev_item);
1713         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1714         ptr = btrfs_device_fsid(dev_item);
1715         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1716         btrfs_mark_buffer_dirty(leaf);
1717
1718         ret = 0;
1719 out:
1720         btrfs_free_path(path);
1721         return ret;
1722 }
1723
1724 /*
1725  * Function to update ctime/mtime for a given device path.
1726  * Mainly used for ctime/mtime based probe like libblkid.
1727  */
1728 static void update_dev_time(const char *path_name)
1729 {
1730         struct file *filp;
1731
1732         filp = filp_open(path_name, O_RDWR, 0);
1733         if (IS_ERR(filp))
1734                 return;
1735         file_update_time(filp);
1736         filp_close(filp, NULL);
1737 }
1738
1739 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1740                              struct btrfs_device *device)
1741 {
1742         struct btrfs_root *root = fs_info->chunk_root;
1743         int ret;
1744         struct btrfs_path *path;
1745         struct btrfs_key key;
1746         struct btrfs_trans_handle *trans;
1747
1748         path = btrfs_alloc_path();
1749         if (!path)
1750                 return -ENOMEM;
1751
1752         trans = btrfs_start_transaction(root, 0);
1753         if (IS_ERR(trans)) {
1754                 btrfs_free_path(path);
1755                 return PTR_ERR(trans);
1756         }
1757         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1758         key.type = BTRFS_DEV_ITEM_KEY;
1759         key.offset = device->devid;
1760
1761         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762         if (ret < 0)
1763                 goto out;
1764
1765         if (ret > 0) {
1766                 ret = -ENOENT;
1767                 goto out;
1768         }
1769
1770         ret = btrfs_del_item(trans, root, path);
1771         if (ret)
1772                 goto out;
1773 out:
1774         btrfs_free_path(path);
1775         btrfs_commit_transaction(trans);
1776         return ret;
1777 }
1778
1779 /*
1780  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781  * filesystem. It's up to the caller to adjust that number regarding eg. device
1782  * replace.
1783  */
1784 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1785                 u64 num_devices)
1786 {
1787         u64 all_avail;
1788         unsigned seq;
1789         int i;
1790
1791         do {
1792                 seq = read_seqbegin(&fs_info->profiles_lock);
1793
1794                 all_avail = fs_info->avail_data_alloc_bits |
1795                             fs_info->avail_system_alloc_bits |
1796                             fs_info->avail_metadata_alloc_bits;
1797         } while (read_seqretry(&fs_info->profiles_lock, seq));
1798
1799         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1800                 if (!(all_avail & btrfs_raid_group[i]))
1801                         continue;
1802
1803                 if (num_devices < btrfs_raid_array[i].devs_min) {
1804                         int ret = btrfs_raid_mindev_error[i];
1805
1806                         if (ret)
1807                                 return ret;
1808                 }
1809         }
1810
1811         return 0;
1812 }
1813
1814 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1815                                         struct btrfs_device *device)
1816 {
1817         struct btrfs_device *next_device;
1818
1819         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1820                 if (next_device != device &&
1821                         !next_device->missing && next_device->bdev)
1822                         return next_device;
1823         }
1824
1825         return NULL;
1826 }
1827
1828 /*
1829  * Helper function to check if the given device is part of s_bdev / latest_bdev
1830  * and replace it with the provided or the next active device, in the context
1831  * where this function called, there should be always be another device (or
1832  * this_dev) which is active.
1833  */
1834 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1835                 struct btrfs_device *device, struct btrfs_device *this_dev)
1836 {
1837         struct btrfs_device *next_device;
1838
1839         if (this_dev)
1840                 next_device = this_dev;
1841         else
1842                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1843                                                                 device);
1844         ASSERT(next_device);
1845
1846         if (fs_info->sb->s_bdev &&
1847                         (fs_info->sb->s_bdev == device->bdev))
1848                 fs_info->sb->s_bdev = next_device->bdev;
1849
1850         if (fs_info->fs_devices->latest_bdev == device->bdev)
1851                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1852 }
1853
1854 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1855                 u64 devid)
1856 {
1857         struct btrfs_device *device;
1858         struct btrfs_fs_devices *cur_devices;
1859         u64 num_devices;
1860         int ret = 0;
1861         bool clear_super = false;
1862
1863         mutex_lock(&uuid_mutex);
1864
1865         num_devices = fs_info->fs_devices->num_devices;
1866         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1867         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1868                 WARN_ON(num_devices < 1);
1869                 num_devices--;
1870         }
1871         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1872
1873         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1874         if (ret)
1875                 goto out;
1876
1877         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1878                                            &device);
1879         if (ret)
1880                 goto out;
1881
1882         if (device->is_tgtdev_for_dev_replace) {
1883                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1884                 goto out;
1885         }
1886
1887         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1888                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1889                 goto out;
1890         }
1891
1892         if (device->writeable) {
1893                 mutex_lock(&fs_info->chunk_mutex);
1894                 list_del_init(&device->dev_alloc_list);
1895                 device->fs_devices->rw_devices--;
1896                 mutex_unlock(&fs_info->chunk_mutex);
1897                 clear_super = true;
1898         }
1899
1900         mutex_unlock(&uuid_mutex);
1901         ret = btrfs_shrink_device(device, 0);
1902         mutex_lock(&uuid_mutex);
1903         if (ret)
1904                 goto error_undo;
1905
1906         /*
1907          * TODO: the superblock still includes this device in its num_devices
1908          * counter although write_all_supers() is not locked out. This
1909          * could give a filesystem state which requires a degraded mount.
1910          */
1911         ret = btrfs_rm_dev_item(fs_info, device);
1912         if (ret)
1913                 goto error_undo;
1914
1915         device->in_fs_metadata = 0;
1916         btrfs_scrub_cancel_dev(fs_info, device);
1917
1918         /*
1919          * the device list mutex makes sure that we don't change
1920          * the device list while someone else is writing out all
1921          * the device supers. Whoever is writing all supers, should
1922          * lock the device list mutex before getting the number of
1923          * devices in the super block (super_copy). Conversely,
1924          * whoever updates the number of devices in the super block
1925          * (super_copy) should hold the device list mutex.
1926          */
1927
1928         cur_devices = device->fs_devices;
1929         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1930         list_del_rcu(&device->dev_list);
1931
1932         device->fs_devices->num_devices--;
1933         device->fs_devices->total_devices--;
1934
1935         if (device->missing)
1936                 device->fs_devices->missing_devices--;
1937
1938         btrfs_assign_next_active_device(fs_info, device, NULL);
1939
1940         if (device->bdev) {
1941                 device->fs_devices->open_devices--;
1942                 /* remove sysfs entry */
1943                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1944         }
1945
1946         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1947         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1948         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1949
1950         /*
1951          * at this point, the device is zero sized and detached from
1952          * the devices list.  All that's left is to zero out the old
1953          * supers and free the device.
1954          */
1955         if (device->writeable)
1956                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1957
1958         btrfs_close_bdev(device);
1959         call_rcu(&device->rcu, free_device);
1960
1961         if (cur_devices->open_devices == 0) {
1962                 struct btrfs_fs_devices *fs_devices;
1963                 fs_devices = fs_info->fs_devices;
1964                 while (fs_devices) {
1965                         if (fs_devices->seed == cur_devices) {
1966                                 fs_devices->seed = cur_devices->seed;
1967                                 break;
1968                         }
1969                         fs_devices = fs_devices->seed;
1970                 }
1971                 cur_devices->seed = NULL;
1972                 __btrfs_close_devices(cur_devices);
1973                 free_fs_devices(cur_devices);
1974         }
1975
1976         fs_info->num_tolerated_disk_barrier_failures =
1977                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1978
1979 out:
1980         mutex_unlock(&uuid_mutex);
1981         return ret;
1982
1983 error_undo:
1984         if (device->writeable) {
1985                 mutex_lock(&fs_info->chunk_mutex);
1986                 list_add(&device->dev_alloc_list,
1987                          &fs_info->fs_devices->alloc_list);
1988                 device->fs_devices->rw_devices++;
1989                 mutex_unlock(&fs_info->chunk_mutex);
1990         }
1991         goto out;
1992 }
1993
1994 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1995                                         struct btrfs_device *srcdev)
1996 {
1997         struct btrfs_fs_devices *fs_devices;
1998
1999         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2000
2001         /*
2002          * in case of fs with no seed, srcdev->fs_devices will point
2003          * to fs_devices of fs_info. However when the dev being replaced is
2004          * a seed dev it will point to the seed's local fs_devices. In short
2005          * srcdev will have its correct fs_devices in both the cases.
2006          */
2007         fs_devices = srcdev->fs_devices;
2008
2009         list_del_rcu(&srcdev->dev_list);
2010         list_del_rcu(&srcdev->dev_alloc_list);
2011         fs_devices->num_devices--;
2012         if (srcdev->missing)
2013                 fs_devices->missing_devices--;
2014
2015         if (srcdev->writeable)
2016                 fs_devices->rw_devices--;
2017
2018         if (srcdev->bdev)
2019                 fs_devices->open_devices--;
2020 }
2021
2022 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2023                                       struct btrfs_device *srcdev)
2024 {
2025         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2026
2027         if (srcdev->writeable) {
2028                 /* zero out the old super if it is writable */
2029                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2030         }
2031
2032         btrfs_close_bdev(srcdev);
2033
2034         call_rcu(&srcdev->rcu, free_device);
2035
2036         /*
2037          * unless fs_devices is seed fs, num_devices shouldn't go
2038          * zero
2039          */
2040         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2041
2042         /* if this is no devs we rather delete the fs_devices */
2043         if (!fs_devices->num_devices) {
2044                 struct btrfs_fs_devices *tmp_fs_devices;
2045
2046                 tmp_fs_devices = fs_info->fs_devices;
2047                 while (tmp_fs_devices) {
2048                         if (tmp_fs_devices->seed == fs_devices) {
2049                                 tmp_fs_devices->seed = fs_devices->seed;
2050                                 break;
2051                         }
2052                         tmp_fs_devices = tmp_fs_devices->seed;
2053                 }
2054                 fs_devices->seed = NULL;
2055                 __btrfs_close_devices(fs_devices);
2056                 free_fs_devices(fs_devices);
2057         }
2058 }
2059
2060 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2061                                       struct btrfs_device *tgtdev)
2062 {
2063         mutex_lock(&uuid_mutex);
2064         WARN_ON(!tgtdev);
2065         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2066
2067         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2068
2069         if (tgtdev->bdev)
2070                 fs_info->fs_devices->open_devices--;
2071
2072         fs_info->fs_devices->num_devices--;
2073
2074         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2075
2076         list_del_rcu(&tgtdev->dev_list);
2077
2078         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2079         mutex_unlock(&uuid_mutex);
2080
2081         /*
2082          * The update_dev_time() with in btrfs_scratch_superblocks()
2083          * may lead to a call to btrfs_show_devname() which will try
2084          * to hold device_list_mutex. And here this device
2085          * is already out of device list, so we don't have to hold
2086          * the device_list_mutex lock.
2087          */
2088         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2089
2090         btrfs_close_bdev(tgtdev);
2091         call_rcu(&tgtdev->rcu, free_device);
2092 }
2093
2094 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2095                                      const char *device_path,
2096                                      struct btrfs_device **device)
2097 {
2098         int ret = 0;
2099         struct btrfs_super_block *disk_super;
2100         u64 devid;
2101         u8 *dev_uuid;
2102         struct block_device *bdev;
2103         struct buffer_head *bh;
2104
2105         *device = NULL;
2106         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107                                     fs_info->bdev_holder, 0, &bdev, &bh);
2108         if (ret)
2109                 return ret;
2110         disk_super = (struct btrfs_super_block *)bh->b_data;
2111         devid = btrfs_stack_device_id(&disk_super->dev_item);
2112         dev_uuid = disk_super->dev_item.uuid;
2113         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2114         brelse(bh);
2115         if (!*device)
2116                 ret = -ENOENT;
2117         blkdev_put(bdev, FMODE_READ);
2118         return ret;
2119 }
2120
2121 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2122                                          const char *device_path,
2123                                          struct btrfs_device **device)
2124 {
2125         *device = NULL;
2126         if (strcmp(device_path, "missing") == 0) {
2127                 struct list_head *devices;
2128                 struct btrfs_device *tmp;
2129
2130                 devices = &fs_info->fs_devices->devices;
2131                 /*
2132                  * It is safe to read the devices since the volume_mutex
2133                  * is held by the caller.
2134                  */
2135                 list_for_each_entry(tmp, devices, dev_list) {
2136                         if (tmp->in_fs_metadata && !tmp->bdev) {
2137                                 *device = tmp;
2138                                 break;
2139                         }
2140                 }
2141
2142                 if (!*device)
2143                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2144
2145                 return 0;
2146         } else {
2147                 return btrfs_find_device_by_path(fs_info, device_path, device);
2148         }
2149 }
2150
2151 /*
2152  * Lookup a device given by device id, or the path if the id is 0.
2153  */
2154 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2155                                  const char *devpath,
2156                                  struct btrfs_device **device)
2157 {
2158         int ret;
2159
2160         if (devid) {
2161                 ret = 0;
2162                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2163                 if (!*device)
2164                         ret = -ENOENT;
2165         } else {
2166                 if (!devpath || !devpath[0])
2167                         return -EINVAL;
2168
2169                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2170                                                            device);
2171         }
2172         return ret;
2173 }
2174
2175 /*
2176  * does all the dirty work required for changing file system's UUID.
2177  */
2178 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2179 {
2180         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2181         struct btrfs_fs_devices *old_devices;
2182         struct btrfs_fs_devices *seed_devices;
2183         struct btrfs_super_block *disk_super = fs_info->super_copy;
2184         struct btrfs_device *device;
2185         u64 super_flags;
2186
2187         BUG_ON(!mutex_is_locked(&uuid_mutex));
2188         if (!fs_devices->seeding)
2189                 return -EINVAL;
2190
2191         seed_devices = __alloc_fs_devices();
2192         if (IS_ERR(seed_devices))
2193                 return PTR_ERR(seed_devices);
2194
2195         old_devices = clone_fs_devices(fs_devices);
2196         if (IS_ERR(old_devices)) {
2197                 kfree(seed_devices);
2198                 return PTR_ERR(old_devices);
2199         }
2200
2201         list_add(&old_devices->list, &fs_uuids);
2202
2203         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2204         seed_devices->opened = 1;
2205         INIT_LIST_HEAD(&seed_devices->devices);
2206         INIT_LIST_HEAD(&seed_devices->alloc_list);
2207         mutex_init(&seed_devices->device_list_mutex);
2208
2209         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2210         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2211                               synchronize_rcu);
2212         list_for_each_entry(device, &seed_devices->devices, dev_list)
2213                 device->fs_devices = seed_devices;
2214
2215         mutex_lock(&fs_info->chunk_mutex);
2216         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2217         mutex_unlock(&fs_info->chunk_mutex);
2218
2219         fs_devices->seeding = 0;
2220         fs_devices->num_devices = 0;
2221         fs_devices->open_devices = 0;
2222         fs_devices->missing_devices = 0;
2223         fs_devices->rotating = 0;
2224         fs_devices->seed = seed_devices;
2225
2226         generate_random_uuid(fs_devices->fsid);
2227         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2228         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2229         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2230
2231         super_flags = btrfs_super_flags(disk_super) &
2232                       ~BTRFS_SUPER_FLAG_SEEDING;
2233         btrfs_set_super_flags(disk_super, super_flags);
2234
2235         return 0;
2236 }
2237
2238 /*
2239  * Store the expected generation for seed devices in device items.
2240  */
2241 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2242                                struct btrfs_fs_info *fs_info)
2243 {
2244         struct btrfs_root *root = fs_info->chunk_root;
2245         struct btrfs_path *path;
2246         struct extent_buffer *leaf;
2247         struct btrfs_dev_item *dev_item;
2248         struct btrfs_device *device;
2249         struct btrfs_key key;
2250         u8 fs_uuid[BTRFS_UUID_SIZE];
2251         u8 dev_uuid[BTRFS_UUID_SIZE];
2252         u64 devid;
2253         int ret;
2254
2255         path = btrfs_alloc_path();
2256         if (!path)
2257                 return -ENOMEM;
2258
2259         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2260         key.offset = 0;
2261         key.type = BTRFS_DEV_ITEM_KEY;
2262
2263         while (1) {
2264                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2265                 if (ret < 0)
2266                         goto error;
2267
2268                 leaf = path->nodes[0];
2269 next_slot:
2270                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2271                         ret = btrfs_next_leaf(root, path);
2272                         if (ret > 0)
2273                                 break;
2274                         if (ret < 0)
2275                                 goto error;
2276                         leaf = path->nodes[0];
2277                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2278                         btrfs_release_path(path);
2279                         continue;
2280                 }
2281
2282                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2284                     key.type != BTRFS_DEV_ITEM_KEY)
2285                         break;
2286
2287                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2288                                           struct btrfs_dev_item);
2289                 devid = btrfs_device_id(leaf, dev_item);
2290                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2291                                    BTRFS_UUID_SIZE);
2292                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2293                                    BTRFS_UUID_SIZE);
2294                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2295                 BUG_ON(!device); /* Logic error */
2296
2297                 if (device->fs_devices->seeding) {
2298                         btrfs_set_device_generation(leaf, dev_item,
2299                                                     device->generation);
2300                         btrfs_mark_buffer_dirty(leaf);
2301                 }
2302
2303                 path->slots[0]++;
2304                 goto next_slot;
2305         }
2306         ret = 0;
2307 error:
2308         btrfs_free_path(path);
2309         return ret;
2310 }
2311
2312 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2313 {
2314         struct btrfs_root *root = fs_info->dev_root;
2315         struct request_queue *q;
2316         struct btrfs_trans_handle *trans;
2317         struct btrfs_device *device;
2318         struct block_device *bdev;
2319         struct list_head *devices;
2320         struct super_block *sb = fs_info->sb;
2321         struct rcu_string *name;
2322         u64 tmp;
2323         int seeding_dev = 0;
2324         int ret = 0;
2325
2326         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2327                 return -EROFS;
2328
2329         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2330                                   fs_info->bdev_holder);
2331         if (IS_ERR(bdev))
2332                 return PTR_ERR(bdev);
2333
2334         if (fs_info->fs_devices->seeding) {
2335                 seeding_dev = 1;
2336                 down_write(&sb->s_umount);
2337                 mutex_lock(&uuid_mutex);
2338         }
2339
2340         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2341
2342         devices = &fs_info->fs_devices->devices;
2343
2344         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2345         list_for_each_entry(device, devices, dev_list) {
2346                 if (device->bdev == bdev) {
2347                         ret = -EEXIST;
2348                         mutex_unlock(
2349                                 &fs_info->fs_devices->device_list_mutex);
2350                         goto error;
2351                 }
2352         }
2353         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2354
2355         device = btrfs_alloc_device(fs_info, NULL, NULL);
2356         if (IS_ERR(device)) {
2357                 /* we can safely leave the fs_devices entry around */
2358                 ret = PTR_ERR(device);
2359                 goto error;
2360         }
2361
2362         name = rcu_string_strdup(device_path, GFP_KERNEL);
2363         if (!name) {
2364                 kfree(device);
2365                 ret = -ENOMEM;
2366                 goto error;
2367         }
2368         rcu_assign_pointer(device->name, name);
2369
2370         trans = btrfs_start_transaction(root, 0);
2371         if (IS_ERR(trans)) {
2372                 rcu_string_free(device->name);
2373                 kfree(device);
2374                 ret = PTR_ERR(trans);
2375                 goto error;
2376         }
2377
2378         q = bdev_get_queue(bdev);
2379         if (blk_queue_discard(q))
2380                 device->can_discard = 1;
2381         device->writeable = 1;
2382         device->generation = trans->transid;
2383         device->io_width = fs_info->sectorsize;
2384         device->io_align = fs_info->sectorsize;
2385         device->sector_size = fs_info->sectorsize;
2386         device->total_bytes = i_size_read(bdev->bd_inode);
2387         device->disk_total_bytes = device->total_bytes;
2388         device->commit_total_bytes = device->total_bytes;
2389         device->fs_info = fs_info;
2390         device->bdev = bdev;
2391         device->in_fs_metadata = 1;
2392         device->is_tgtdev_for_dev_replace = 0;
2393         device->mode = FMODE_EXCL;
2394         device->dev_stats_valid = 1;
2395         set_blocksize(device->bdev, 4096);
2396
2397         if (seeding_dev) {
2398                 sb->s_flags &= ~MS_RDONLY;
2399                 ret = btrfs_prepare_sprout(fs_info);
2400                 BUG_ON(ret); /* -ENOMEM */
2401         }
2402
2403         device->fs_devices = fs_info->fs_devices;
2404
2405         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2406         mutex_lock(&fs_info->chunk_mutex);
2407         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2408         list_add(&device->dev_alloc_list,
2409                  &fs_info->fs_devices->alloc_list);
2410         fs_info->fs_devices->num_devices++;
2411         fs_info->fs_devices->open_devices++;
2412         fs_info->fs_devices->rw_devices++;
2413         fs_info->fs_devices->total_devices++;
2414         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2415
2416         spin_lock(&fs_info->free_chunk_lock);
2417         fs_info->free_chunk_space += device->total_bytes;
2418         spin_unlock(&fs_info->free_chunk_lock);
2419
2420         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2421                 fs_info->fs_devices->rotating = 1;
2422
2423         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2424         btrfs_set_super_total_bytes(fs_info->super_copy,
2425                                     tmp + device->total_bytes);
2426
2427         tmp = btrfs_super_num_devices(fs_info->super_copy);
2428         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2429
2430         /* add sysfs device entry */
2431         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2432
2433         /*
2434          * we've got more storage, clear any full flags on the space
2435          * infos
2436          */
2437         btrfs_clear_space_info_full(fs_info);
2438
2439         mutex_unlock(&fs_info->chunk_mutex);
2440         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2441
2442         if (seeding_dev) {
2443                 mutex_lock(&fs_info->chunk_mutex);
2444                 ret = init_first_rw_device(trans, fs_info);
2445                 mutex_unlock(&fs_info->chunk_mutex);
2446                 if (ret) {
2447                         btrfs_abort_transaction(trans, ret);
2448                         goto error_trans;
2449                 }
2450         }
2451
2452         ret = btrfs_add_device(trans, fs_info, device);
2453         if (ret) {
2454                 btrfs_abort_transaction(trans, ret);
2455                 goto error_trans;
2456         }
2457
2458         if (seeding_dev) {
2459                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2460
2461                 ret = btrfs_finish_sprout(trans, fs_info);
2462                 if (ret) {
2463                         btrfs_abort_transaction(trans, ret);
2464                         goto error_trans;
2465                 }
2466
2467                 /* Sprouting would change fsid of the mounted root,
2468                  * so rename the fsid on the sysfs
2469                  */
2470                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2471                                                 fs_info->fsid);
2472                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2473                         btrfs_warn(fs_info,
2474                                    "sysfs: failed to create fsid for sprout");
2475         }
2476
2477         fs_info->num_tolerated_disk_barrier_failures =
2478                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2479         ret = btrfs_commit_transaction(trans);
2480
2481         if (seeding_dev) {
2482                 mutex_unlock(&uuid_mutex);
2483                 up_write(&sb->s_umount);
2484
2485                 if (ret) /* transaction commit */
2486                         return ret;
2487
2488                 ret = btrfs_relocate_sys_chunks(fs_info);
2489                 if (ret < 0)
2490                         btrfs_handle_fs_error(fs_info, ret,
2491                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2492                 trans = btrfs_attach_transaction(root);
2493                 if (IS_ERR(trans)) {
2494                         if (PTR_ERR(trans) == -ENOENT)
2495                                 return 0;
2496                         return PTR_ERR(trans);
2497                 }
2498                 ret = btrfs_commit_transaction(trans);
2499         }
2500
2501         /* Update ctime/mtime for libblkid */
2502         update_dev_time(device_path);
2503         return ret;
2504
2505 error_trans:
2506         btrfs_end_transaction(trans);
2507         rcu_string_free(device->name);
2508         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2509         kfree(device);
2510 error:
2511         blkdev_put(bdev, FMODE_EXCL);
2512         if (seeding_dev) {
2513                 mutex_unlock(&uuid_mutex);
2514                 up_write(&sb->s_umount);
2515         }
2516         return ret;
2517 }
2518
2519 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2520                                   const char *device_path,
2521                                   struct btrfs_device *srcdev,
2522                                   struct btrfs_device **device_out)
2523 {
2524         struct request_queue *q;
2525         struct btrfs_device *device;
2526         struct block_device *bdev;
2527         struct list_head *devices;
2528         struct rcu_string *name;
2529         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2530         int ret = 0;
2531
2532         *device_out = NULL;
2533         if (fs_info->fs_devices->seeding) {
2534                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2535                 return -EINVAL;
2536         }
2537
2538         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2539                                   fs_info->bdev_holder);
2540         if (IS_ERR(bdev)) {
2541                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2542                 return PTR_ERR(bdev);
2543         }
2544
2545         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2546
2547         devices = &fs_info->fs_devices->devices;
2548         list_for_each_entry(device, devices, dev_list) {
2549                 if (device->bdev == bdev) {
2550                         btrfs_err(fs_info,
2551                                   "target device is in the filesystem!");
2552                         ret = -EEXIST;
2553                         goto error;
2554                 }
2555         }
2556
2557
2558         if (i_size_read(bdev->bd_inode) <
2559             btrfs_device_get_total_bytes(srcdev)) {
2560                 btrfs_err(fs_info,
2561                           "target device is smaller than source device!");
2562                 ret = -EINVAL;
2563                 goto error;
2564         }
2565
2566
2567         device = btrfs_alloc_device(NULL, &devid, NULL);
2568         if (IS_ERR(device)) {
2569                 ret = PTR_ERR(device);
2570                 goto error;
2571         }
2572
2573         name = rcu_string_strdup(device_path, GFP_NOFS);
2574         if (!name) {
2575                 kfree(device);
2576                 ret = -ENOMEM;
2577                 goto error;
2578         }
2579         rcu_assign_pointer(device->name, name);
2580
2581         q = bdev_get_queue(bdev);
2582         if (blk_queue_discard(q))
2583                 device->can_discard = 1;
2584         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2585         device->writeable = 1;
2586         device->generation = 0;
2587         device->io_width = fs_info->sectorsize;
2588         device->io_align = fs_info->sectorsize;
2589         device->sector_size = fs_info->sectorsize;
2590         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2591         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2592         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2593         ASSERT(list_empty(&srcdev->resized_list));
2594         device->commit_total_bytes = srcdev->commit_total_bytes;
2595         device->commit_bytes_used = device->bytes_used;
2596         device->fs_info = fs_info;
2597         device->bdev = bdev;
2598         device->in_fs_metadata = 1;
2599         device->is_tgtdev_for_dev_replace = 1;
2600         device->mode = FMODE_EXCL;
2601         device->dev_stats_valid = 1;
2602         set_blocksize(device->bdev, 4096);
2603         device->fs_devices = fs_info->fs_devices;
2604         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2605         fs_info->fs_devices->num_devices++;
2606         fs_info->fs_devices->open_devices++;
2607         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2608
2609         *device_out = device;
2610         return ret;
2611
2612 error:
2613         blkdev_put(bdev, FMODE_EXCL);
2614         return ret;
2615 }
2616
2617 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2618                                               struct btrfs_device *tgtdev)
2619 {
2620         u32 sectorsize = fs_info->sectorsize;
2621
2622         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2623         tgtdev->io_width = sectorsize;
2624         tgtdev->io_align = sectorsize;
2625         tgtdev->sector_size = sectorsize;
2626         tgtdev->fs_info = fs_info;
2627         tgtdev->in_fs_metadata = 1;
2628 }
2629
2630 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2631                                         struct btrfs_device *device)
2632 {
2633         int ret;
2634         struct btrfs_path *path;
2635         struct btrfs_root *root = device->fs_info->chunk_root;
2636         struct btrfs_dev_item *dev_item;
2637         struct extent_buffer *leaf;
2638         struct btrfs_key key;
2639
2640         path = btrfs_alloc_path();
2641         if (!path)
2642                 return -ENOMEM;
2643
2644         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2645         key.type = BTRFS_DEV_ITEM_KEY;
2646         key.offset = device->devid;
2647
2648         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2649         if (ret < 0)
2650                 goto out;
2651
2652         if (ret > 0) {
2653                 ret = -ENOENT;
2654                 goto out;
2655         }
2656
2657         leaf = path->nodes[0];
2658         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2659
2660         btrfs_set_device_id(leaf, dev_item, device->devid);
2661         btrfs_set_device_type(leaf, dev_item, device->type);
2662         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2663         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2664         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2665         btrfs_set_device_total_bytes(leaf, dev_item,
2666                                      btrfs_device_get_disk_total_bytes(device));
2667         btrfs_set_device_bytes_used(leaf, dev_item,
2668                                     btrfs_device_get_bytes_used(device));
2669         btrfs_mark_buffer_dirty(leaf);
2670
2671 out:
2672         btrfs_free_path(path);
2673         return ret;
2674 }
2675
2676 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2677                       struct btrfs_device *device, u64 new_size)
2678 {
2679         struct btrfs_fs_info *fs_info = device->fs_info;
2680         struct btrfs_super_block *super_copy = fs_info->super_copy;
2681         struct btrfs_fs_devices *fs_devices;
2682         u64 old_total;
2683         u64 diff;
2684
2685         if (!device->writeable)
2686                 return -EACCES;
2687
2688         mutex_lock(&fs_info->chunk_mutex);
2689         old_total = btrfs_super_total_bytes(super_copy);
2690         diff = new_size - device->total_bytes;
2691
2692         if (new_size <= device->total_bytes ||
2693             device->is_tgtdev_for_dev_replace) {
2694                 mutex_unlock(&fs_info->chunk_mutex);
2695                 return -EINVAL;
2696         }
2697
2698         fs_devices = fs_info->fs_devices;
2699
2700         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2701         device->fs_devices->total_rw_bytes += diff;
2702
2703         btrfs_device_set_total_bytes(device, new_size);
2704         btrfs_device_set_disk_total_bytes(device, new_size);
2705         btrfs_clear_space_info_full(device->fs_info);
2706         if (list_empty(&device->resized_list))
2707                 list_add_tail(&device->resized_list,
2708                               &fs_devices->resized_devices);
2709         mutex_unlock(&fs_info->chunk_mutex);
2710
2711         return btrfs_update_device(trans, device);
2712 }
2713
2714 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2715                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2716                             u64 chunk_offset)
2717 {
2718         struct btrfs_root *root = fs_info->chunk_root;
2719         int ret;
2720         struct btrfs_path *path;
2721         struct btrfs_key key;
2722
2723         path = btrfs_alloc_path();
2724         if (!path)
2725                 return -ENOMEM;
2726
2727         key.objectid = chunk_objectid;
2728         key.offset = chunk_offset;
2729         key.type = BTRFS_CHUNK_ITEM_KEY;
2730
2731         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2732         if (ret < 0)
2733                 goto out;
2734         else if (ret > 0) { /* Logic error or corruption */
2735                 btrfs_handle_fs_error(fs_info, -ENOENT,
2736                                       "Failed lookup while freeing chunk.");
2737                 ret = -ENOENT;
2738                 goto out;
2739         }
2740
2741         ret = btrfs_del_item(trans, root, path);
2742         if (ret < 0)
2743                 btrfs_handle_fs_error(fs_info, ret,
2744                                       "Failed to delete chunk item.");
2745 out:
2746         btrfs_free_path(path);
2747         return ret;
2748 }
2749
2750 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2751                                u64 chunk_objectid, u64 chunk_offset)
2752 {
2753         struct btrfs_super_block *super_copy = fs_info->super_copy;
2754         struct btrfs_disk_key *disk_key;
2755         struct btrfs_chunk *chunk;
2756         u8 *ptr;
2757         int ret = 0;
2758         u32 num_stripes;
2759         u32 array_size;
2760         u32 len = 0;
2761         u32 cur;
2762         struct btrfs_key key;
2763
2764         mutex_lock(&fs_info->chunk_mutex);
2765         array_size = btrfs_super_sys_array_size(super_copy);
2766
2767         ptr = super_copy->sys_chunk_array;
2768         cur = 0;
2769
2770         while (cur < array_size) {
2771                 disk_key = (struct btrfs_disk_key *)ptr;
2772                 btrfs_disk_key_to_cpu(&key, disk_key);
2773
2774                 len = sizeof(*disk_key);
2775
2776                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2777                         chunk = (struct btrfs_chunk *)(ptr + len);
2778                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2779                         len += btrfs_chunk_item_size(num_stripes);
2780                 } else {
2781                         ret = -EIO;
2782                         break;
2783                 }
2784                 if (key.objectid == chunk_objectid &&
2785                     key.offset == chunk_offset) {
2786                         memmove(ptr, ptr + len, array_size - (cur + len));
2787                         array_size -= len;
2788                         btrfs_set_super_sys_array_size(super_copy, array_size);
2789                 } else {
2790                         ptr += len;
2791                         cur += len;
2792                 }
2793         }
2794         mutex_unlock(&fs_info->chunk_mutex);
2795         return ret;
2796 }
2797
2798 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2799                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2800 {
2801         struct extent_map_tree *em_tree;
2802         struct extent_map *em;
2803         struct map_lookup *map;
2804         u64 dev_extent_len = 0;
2805         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2806         int i, ret = 0;
2807         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2808
2809         em_tree = &fs_info->mapping_tree.map_tree;
2810
2811         read_lock(&em_tree->lock);
2812         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2813         read_unlock(&em_tree->lock);
2814
2815         if (!em || em->start > chunk_offset ||
2816             em->start + em->len < chunk_offset) {
2817                 /*
2818                  * This is a logic error, but we don't want to just rely on the
2819                  * user having built with ASSERT enabled, so if ASSERT doesn't
2820                  * do anything we still error out.
2821                  */
2822                 ASSERT(0);
2823                 if (em)
2824                         free_extent_map(em);
2825                 return -EINVAL;
2826         }
2827         map = em->map_lookup;
2828         mutex_lock(&fs_info->chunk_mutex);
2829         check_system_chunk(trans, fs_info, map->type);
2830         mutex_unlock(&fs_info->chunk_mutex);
2831
2832         /*
2833          * Take the device list mutex to prevent races with the final phase of
2834          * a device replace operation that replaces the device object associated
2835          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2836          */
2837         mutex_lock(&fs_devices->device_list_mutex);
2838         for (i = 0; i < map->num_stripes; i++) {
2839                 struct btrfs_device *device = map->stripes[i].dev;
2840                 ret = btrfs_free_dev_extent(trans, device,
2841                                             map->stripes[i].physical,
2842                                             &dev_extent_len);
2843                 if (ret) {
2844                         mutex_unlock(&fs_devices->device_list_mutex);
2845                         btrfs_abort_transaction(trans, ret);
2846                         goto out;
2847                 }
2848
2849                 if (device->bytes_used > 0) {
2850                         mutex_lock(&fs_info->chunk_mutex);
2851                         btrfs_device_set_bytes_used(device,
2852                                         device->bytes_used - dev_extent_len);
2853                         spin_lock(&fs_info->free_chunk_lock);
2854                         fs_info->free_chunk_space += dev_extent_len;
2855                         spin_unlock(&fs_info->free_chunk_lock);
2856                         btrfs_clear_space_info_full(fs_info);
2857                         mutex_unlock(&fs_info->chunk_mutex);
2858                 }
2859
2860                 if (map->stripes[i].dev) {
2861                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2862                         if (ret) {
2863                                 mutex_unlock(&fs_devices->device_list_mutex);
2864                                 btrfs_abort_transaction(trans, ret);
2865                                 goto out;
2866                         }
2867                 }
2868         }
2869         mutex_unlock(&fs_devices->device_list_mutex);
2870
2871         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2872         if (ret) {
2873                 btrfs_abort_transaction(trans, ret);
2874                 goto out;
2875         }
2876
2877         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2878
2879         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2880                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2881                                           chunk_offset);
2882                 if (ret) {
2883                         btrfs_abort_transaction(trans, ret);
2884                         goto out;
2885                 }
2886         }
2887
2888         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2889         if (ret) {
2890                 btrfs_abort_transaction(trans, ret);
2891                 goto out;
2892         }
2893
2894 out:
2895         /* once for us */
2896         free_extent_map(em);
2897         return ret;
2898 }
2899
2900 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2901 {
2902         struct btrfs_root *root = fs_info->chunk_root;
2903         struct btrfs_trans_handle *trans;
2904         int ret;
2905
2906         /*
2907          * Prevent races with automatic removal of unused block groups.
2908          * After we relocate and before we remove the chunk with offset
2909          * chunk_offset, automatic removal of the block group can kick in,
2910          * resulting in a failure when calling btrfs_remove_chunk() below.
2911          *
2912          * Make sure to acquire this mutex before doing a tree search (dev
2913          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915          * we release the path used to search the chunk/dev tree and before
2916          * the current task acquires this mutex and calls us.
2917          */
2918         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2919
2920         ret = btrfs_can_relocate(fs_info, chunk_offset);
2921         if (ret)
2922                 return -ENOSPC;
2923
2924         /* step one, relocate all the extents inside this chunk */
2925         btrfs_scrub_pause(fs_info);
2926         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2927         btrfs_scrub_continue(fs_info);
2928         if (ret)
2929                 return ret;
2930
2931         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2932                                                      chunk_offset);
2933         if (IS_ERR(trans)) {
2934                 ret = PTR_ERR(trans);
2935                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2936                 return ret;
2937         }
2938
2939         /*
2940          * step two, delete the device extents and the
2941          * chunk tree entries
2942          */
2943         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2944         btrfs_end_transaction(trans);
2945         return ret;
2946 }
2947
2948 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2949 {
2950         struct btrfs_root *chunk_root = fs_info->chunk_root;
2951         struct btrfs_path *path;
2952         struct extent_buffer *leaf;
2953         struct btrfs_chunk *chunk;
2954         struct btrfs_key key;
2955         struct btrfs_key found_key;
2956         u64 chunk_type;
2957         bool retried = false;
2958         int failed = 0;
2959         int ret;
2960
2961         path = btrfs_alloc_path();
2962         if (!path)
2963                 return -ENOMEM;
2964
2965 again:
2966         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967         key.offset = (u64)-1;
2968         key.type = BTRFS_CHUNK_ITEM_KEY;
2969
2970         while (1) {
2971                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2972                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2973                 if (ret < 0) {
2974                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2975                         goto error;
2976                 }
2977                 BUG_ON(ret == 0); /* Corruption */
2978
2979                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2980                                           key.type);
2981                 if (ret)
2982                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2983                 if (ret < 0)
2984                         goto error;
2985                 if (ret > 0)
2986                         break;
2987
2988                 leaf = path->nodes[0];
2989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2990
2991                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992                                        struct btrfs_chunk);
2993                 chunk_type = btrfs_chunk_type(leaf, chunk);
2994                 btrfs_release_path(path);
2995
2996                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2998                         if (ret == -ENOSPC)
2999                                 failed++;
3000                         else
3001                                 BUG_ON(ret);
3002                 }
3003                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3004
3005                 if (found_key.offset == 0)
3006                         break;
3007                 key.offset = found_key.offset - 1;
3008         }
3009         ret = 0;
3010         if (failed && !retried) {
3011                 failed = 0;
3012                 retried = true;
3013                 goto again;
3014         } else if (WARN_ON(failed && retried)) {
3015                 ret = -ENOSPC;
3016         }
3017 error:
3018         btrfs_free_path(path);
3019         return ret;
3020 }
3021
3022 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3023                                struct btrfs_balance_control *bctl)
3024 {
3025         struct btrfs_root *root = fs_info->tree_root;
3026         struct btrfs_trans_handle *trans;
3027         struct btrfs_balance_item *item;
3028         struct btrfs_disk_balance_args disk_bargs;
3029         struct btrfs_path *path;
3030         struct extent_buffer *leaf;
3031         struct btrfs_key key;
3032         int ret, err;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038         trans = btrfs_start_transaction(root, 0);
3039         if (IS_ERR(trans)) {
3040                 btrfs_free_path(path);
3041                 return PTR_ERR(trans);
3042         }
3043
3044         key.objectid = BTRFS_BALANCE_OBJECTID;
3045         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3046         key.offset = 0;
3047
3048         ret = btrfs_insert_empty_item(trans, root, path, &key,
3049                                       sizeof(*item));
3050         if (ret)
3051                 goto out;
3052
3053         leaf = path->nodes[0];
3054         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3055
3056         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3057
3058         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3059         btrfs_set_balance_data(leaf, item, &disk_bargs);
3060         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3061         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3062         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3063         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3064
3065         btrfs_set_balance_flags(leaf, item, bctl->flags);
3066
3067         btrfs_mark_buffer_dirty(leaf);
3068 out:
3069         btrfs_free_path(path);
3070         err = btrfs_commit_transaction(trans);
3071         if (err && !ret)
3072                 ret = err;
3073         return ret;
3074 }
3075
3076 static int del_balance_item(struct btrfs_fs_info *fs_info)
3077 {
3078         struct btrfs_root *root = fs_info->tree_root;
3079         struct btrfs_trans_handle *trans;
3080         struct btrfs_path *path;
3081         struct btrfs_key key;
3082         int ret, err;
3083
3084         path = btrfs_alloc_path();
3085         if (!path)
3086                 return -ENOMEM;
3087
3088         trans = btrfs_start_transaction(root, 0);
3089         if (IS_ERR(trans)) {
3090                 btrfs_free_path(path);
3091                 return PTR_ERR(trans);
3092         }
3093
3094         key.objectid = BTRFS_BALANCE_OBJECTID;
3095         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3096         key.offset = 0;
3097
3098         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3099         if (ret < 0)
3100                 goto out;
3101         if (ret > 0) {
3102                 ret = -ENOENT;
3103                 goto out;
3104         }
3105
3106         ret = btrfs_del_item(trans, root, path);
3107 out:
3108         btrfs_free_path(path);
3109         err = btrfs_commit_transaction(trans);
3110         if (err && !ret)
3111                 ret = err;
3112         return ret;
3113 }
3114
3115 /*
3116  * This is a heuristic used to reduce the number of chunks balanced on
3117  * resume after balance was interrupted.
3118  */
3119 static void update_balance_args(struct btrfs_balance_control *bctl)
3120 {
3121         /*
3122          * Turn on soft mode for chunk types that were being converted.
3123          */
3124         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3125                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3126         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3127                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3128         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3129                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3130
3131         /*
3132          * Turn on usage filter if is not already used.  The idea is
3133          * that chunks that we have already balanced should be
3134          * reasonably full.  Don't do it for chunks that are being
3135          * converted - that will keep us from relocating unconverted
3136          * (albeit full) chunks.
3137          */
3138         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3139             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3140             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3141                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3142                 bctl->data.usage = 90;
3143         }
3144         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3145             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3146             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3147                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3148                 bctl->sys.usage = 90;
3149         }
3150         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3151             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3152             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3153                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3154                 bctl->meta.usage = 90;
3155         }
3156 }
3157
3158 /*
3159  * Should be called with both balance and volume mutexes held to
3160  * serialize other volume operations (add_dev/rm_dev/resize) with
3161  * restriper.  Same goes for unset_balance_control.
3162  */
3163 static void set_balance_control(struct btrfs_balance_control *bctl)
3164 {
3165         struct btrfs_fs_info *fs_info = bctl->fs_info;
3166
3167         BUG_ON(fs_info->balance_ctl);
3168
3169         spin_lock(&fs_info->balance_lock);
3170         fs_info->balance_ctl = bctl;
3171         spin_unlock(&fs_info->balance_lock);
3172 }
3173
3174 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3175 {
3176         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3177
3178         BUG_ON(!fs_info->balance_ctl);
3179
3180         spin_lock(&fs_info->balance_lock);
3181         fs_info->balance_ctl = NULL;
3182         spin_unlock(&fs_info->balance_lock);
3183
3184         kfree(bctl);
3185 }
3186
3187 /*
3188  * Balance filters.  Return 1 if chunk should be filtered out
3189  * (should not be balanced).
3190  */
3191 static int chunk_profiles_filter(u64 chunk_type,
3192                                  struct btrfs_balance_args *bargs)
3193 {
3194         chunk_type = chunk_to_extended(chunk_type) &
3195                                 BTRFS_EXTENDED_PROFILE_MASK;
3196
3197         if (bargs->profiles & chunk_type)
3198                 return 0;
3199
3200         return 1;
3201 }
3202
3203 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3204                               struct btrfs_balance_args *bargs)
3205 {
3206         struct btrfs_block_group_cache *cache;
3207         u64 chunk_used;
3208         u64 user_thresh_min;
3209         u64 user_thresh_max;
3210         int ret = 1;
3211
3212         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3213         chunk_used = btrfs_block_group_used(&cache->item);
3214
3215         if (bargs->usage_min == 0)
3216                 user_thresh_min = 0;
3217         else
3218                 user_thresh_min = div_factor_fine(cache->key.offset,
3219                                         bargs->usage_min);
3220
3221         if (bargs->usage_max == 0)
3222                 user_thresh_max = 1;
3223         else if (bargs->usage_max > 100)
3224                 user_thresh_max = cache->key.offset;
3225         else
3226                 user_thresh_max = div_factor_fine(cache->key.offset,
3227                                         bargs->usage_max);
3228
3229         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3230                 ret = 0;
3231
3232         btrfs_put_block_group(cache);
3233         return ret;
3234 }
3235
3236 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3237                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3238 {
3239         struct btrfs_block_group_cache *cache;
3240         u64 chunk_used, user_thresh;
3241         int ret = 1;
3242
3243         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3244         chunk_used = btrfs_block_group_used(&cache->item);
3245
3246         if (bargs->usage_min == 0)
3247                 user_thresh = 1;
3248         else if (bargs->usage > 100)
3249                 user_thresh = cache->key.offset;
3250         else
3251                 user_thresh = div_factor_fine(cache->key.offset,
3252                                               bargs->usage);
3253
3254         if (chunk_used < user_thresh)
3255                 ret = 0;
3256
3257         btrfs_put_block_group(cache);
3258         return ret;
3259 }
3260
3261 static int chunk_devid_filter(struct extent_buffer *leaf,
3262                               struct btrfs_chunk *chunk,
3263                               struct btrfs_balance_args *bargs)
3264 {
3265         struct btrfs_stripe *stripe;
3266         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3267         int i;
3268
3269         for (i = 0; i < num_stripes; i++) {
3270                 stripe = btrfs_stripe_nr(chunk, i);
3271                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3272                         return 0;
3273         }
3274
3275         return 1;
3276 }
3277
3278 /* [pstart, pend) */
3279 static int chunk_drange_filter(struct extent_buffer *leaf,
3280                                struct btrfs_chunk *chunk,
3281                                u64 chunk_offset,
3282                                struct btrfs_balance_args *bargs)
3283 {
3284         struct btrfs_stripe *stripe;
3285         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3286         u64 stripe_offset;
3287         u64 stripe_length;
3288         int factor;
3289         int i;
3290
3291         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3292                 return 0;
3293
3294         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3295              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3296                 factor = num_stripes / 2;
3297         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3298                 factor = num_stripes - 1;
3299         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3300                 factor = num_stripes - 2;
3301         } else {
3302                 factor = num_stripes;
3303         }
3304
3305         for (i = 0; i < num_stripes; i++) {
3306                 stripe = btrfs_stripe_nr(chunk, i);
3307                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3308                         continue;
3309
3310                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3311                 stripe_length = btrfs_chunk_length(leaf, chunk);
3312                 stripe_length = div_u64(stripe_length, factor);
3313
3314                 if (stripe_offset < bargs->pend &&
3315                     stripe_offset + stripe_length > bargs->pstart)
3316                         return 0;
3317         }
3318
3319         return 1;
3320 }
3321
3322 /* [vstart, vend) */
3323 static int chunk_vrange_filter(struct extent_buffer *leaf,
3324                                struct btrfs_chunk *chunk,
3325                                u64 chunk_offset,
3326                                struct btrfs_balance_args *bargs)
3327 {
3328         if (chunk_offset < bargs->vend &&
3329             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3330                 /* at least part of the chunk is inside this vrange */
3331                 return 0;
3332
3333         return 1;
3334 }
3335
3336 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3337                                struct btrfs_chunk *chunk,
3338                                struct btrfs_balance_args *bargs)
3339 {
3340         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3341
3342         if (bargs->stripes_min <= num_stripes
3343                         && num_stripes <= bargs->stripes_max)
3344                 return 0;
3345
3346         return 1;
3347 }
3348
3349 static int chunk_soft_convert_filter(u64 chunk_type,
3350                                      struct btrfs_balance_args *bargs)
3351 {
3352         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3353                 return 0;
3354
3355         chunk_type = chunk_to_extended(chunk_type) &
3356                                 BTRFS_EXTENDED_PROFILE_MASK;
3357
3358         if (bargs->target == chunk_type)
3359                 return 1;
3360
3361         return 0;
3362 }
3363
3364 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3365                                 struct extent_buffer *leaf,
3366                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3367 {
3368         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3369         struct btrfs_balance_args *bargs = NULL;
3370         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3371
3372         /* type filter */
3373         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3374               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3375                 return 0;
3376         }
3377
3378         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3379                 bargs = &bctl->data;
3380         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3381                 bargs = &bctl->sys;
3382         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3383                 bargs = &bctl->meta;
3384
3385         /* profiles filter */
3386         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3387             chunk_profiles_filter(chunk_type, bargs)) {
3388                 return 0;
3389         }
3390
3391         /* usage filter */
3392         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3393             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3394                 return 0;
3395         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3396             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3397                 return 0;
3398         }
3399
3400         /* devid filter */
3401         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3402             chunk_devid_filter(leaf, chunk, bargs)) {
3403                 return 0;
3404         }
3405
3406         /* drange filter, makes sense only with devid filter */
3407         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3408             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3409                 return 0;
3410         }
3411
3412         /* vrange filter */
3413         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3414             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3415                 return 0;
3416         }