Merge branch 'for-4.14/block-postmerge' of git://git.kernel.dk/linux-block
[sfrench/cifs-2.6.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/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 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
143                              enum btrfs_map_op op,
144                              u64 logical, u64 *length,
145                              struct btrfs_bio **bbio_ret,
146                              int mirror_num, int need_raid_map);
147
148 DEFINE_MUTEX(uuid_mutex);
149 static LIST_HEAD(fs_uuids);
150 struct list_head *btrfs_get_fs_uuids(void)
151 {
152         return &fs_uuids;
153 }
154
155 static struct btrfs_fs_devices *__alloc_fs_devices(void)
156 {
157         struct btrfs_fs_devices *fs_devs;
158
159         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
160         if (!fs_devs)
161                 return ERR_PTR(-ENOMEM);
162
163         mutex_init(&fs_devs->device_list_mutex);
164
165         INIT_LIST_HEAD(&fs_devs->devices);
166         INIT_LIST_HEAD(&fs_devs->resized_devices);
167         INIT_LIST_HEAD(&fs_devs->alloc_list);
168         INIT_LIST_HEAD(&fs_devs->list);
169
170         return fs_devs;
171 }
172
173 /**
174  * alloc_fs_devices - allocate struct btrfs_fs_devices
175  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
176  *              generated.
177  *
178  * Return: a pointer to a new &struct btrfs_fs_devices on success;
179  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
180  * can be destroyed with kfree() right away.
181  */
182 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
183 {
184         struct btrfs_fs_devices *fs_devs;
185
186         fs_devs = __alloc_fs_devices();
187         if (IS_ERR(fs_devs))
188                 return fs_devs;
189
190         if (fsid)
191                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
192         else
193                 generate_random_uuid(fs_devs->fsid);
194
195         return fs_devs;
196 }
197
198 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
199 {
200         struct btrfs_device *device;
201         WARN_ON(fs_devices->opened);
202         while (!list_empty(&fs_devices->devices)) {
203                 device = list_entry(fs_devices->devices.next,
204                                     struct btrfs_device, dev_list);
205                 list_del(&device->dev_list);
206                 rcu_string_free(device->name);
207                 kfree(device);
208         }
209         kfree(fs_devices);
210 }
211
212 static void btrfs_kobject_uevent(struct block_device *bdev,
213                                  enum kobject_action action)
214 {
215         int ret;
216
217         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
218         if (ret)
219                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
220                         action,
221                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
222                         &disk_to_dev(bdev->bd_disk)->kobj);
223 }
224
225 void btrfs_cleanup_fs_uuids(void)
226 {
227         struct btrfs_fs_devices *fs_devices;
228
229         while (!list_empty(&fs_uuids)) {
230                 fs_devices = list_entry(fs_uuids.next,
231                                         struct btrfs_fs_devices, list);
232                 list_del(&fs_devices->list);
233                 free_fs_devices(fs_devices);
234         }
235 }
236
237 static struct btrfs_device *__alloc_device(void)
238 {
239         struct btrfs_device *dev;
240
241         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
242         if (!dev)
243                 return ERR_PTR(-ENOMEM);
244
245         /*
246          * Preallocate a bio that's always going to be used for flushing device
247          * barriers and matches the device lifespan
248          */
249         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
250         if (!dev->flush_bio) {
251                 kfree(dev);
252                 return ERR_PTR(-ENOMEM);
253         }
254         bio_get(dev->flush_bio);
255
256         INIT_LIST_HEAD(&dev->dev_list);
257         INIT_LIST_HEAD(&dev->dev_alloc_list);
258         INIT_LIST_HEAD(&dev->resized_list);
259
260         spin_lock_init(&dev->io_lock);
261
262         spin_lock_init(&dev->reada_lock);
263         atomic_set(&dev->reada_in_flight, 0);
264         atomic_set(&dev->dev_stats_ccnt, 0);
265         btrfs_device_data_ordered_init(dev);
266         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
267         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
268
269         return dev;
270 }
271
272 static noinline struct btrfs_device *__find_device(struct list_head *head,
273                                                    u64 devid, u8 *uuid)
274 {
275         struct btrfs_device *dev;
276
277         list_for_each_entry(dev, head, dev_list) {
278                 if (dev->devid == devid &&
279                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
280                         return dev;
281                 }
282         }
283         return NULL;
284 }
285
286 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
287 {
288         struct btrfs_fs_devices *fs_devices;
289
290         list_for_each_entry(fs_devices, &fs_uuids, list) {
291                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
292                         return fs_devices;
293         }
294         return NULL;
295 }
296
297 static int
298 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
299                       int flush, struct block_device **bdev,
300                       struct buffer_head **bh)
301 {
302         int ret;
303
304         *bdev = blkdev_get_by_path(device_path, flags, holder);
305
306         if (IS_ERR(*bdev)) {
307                 ret = PTR_ERR(*bdev);
308                 goto error;
309         }
310
311         if (flush)
312                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
313         ret = set_blocksize(*bdev, 4096);
314         if (ret) {
315                 blkdev_put(*bdev, flags);
316                 goto error;
317         }
318         invalidate_bdev(*bdev);
319         *bh = btrfs_read_dev_super(*bdev);
320         if (IS_ERR(*bh)) {
321                 ret = PTR_ERR(*bh);
322                 blkdev_put(*bdev, flags);
323                 goto error;
324         }
325
326         return 0;
327
328 error:
329         *bdev = NULL;
330         *bh = NULL;
331         return ret;
332 }
333
334 static void requeue_list(struct btrfs_pending_bios *pending_bios,
335                         struct bio *head, struct bio *tail)
336 {
337
338         struct bio *old_head;
339
340         old_head = pending_bios->head;
341         pending_bios->head = head;
342         if (pending_bios->tail)
343                 tail->bi_next = old_head;
344         else
345                 pending_bios->tail = tail;
346 }
347
348 /*
349  * we try to collect pending bios for a device so we don't get a large
350  * number of procs sending bios down to the same device.  This greatly
351  * improves the schedulers ability to collect and merge the bios.
352  *
353  * But, it also turns into a long list of bios to process and that is sure
354  * to eventually make the worker thread block.  The solution here is to
355  * make some progress and then put this work struct back at the end of
356  * the list if the block device is congested.  This way, multiple devices
357  * can make progress from a single worker thread.
358  */
359 static noinline void run_scheduled_bios(struct btrfs_device *device)
360 {
361         struct btrfs_fs_info *fs_info = device->fs_info;
362         struct bio *pending;
363         struct backing_dev_info *bdi;
364         struct btrfs_pending_bios *pending_bios;
365         struct bio *tail;
366         struct bio *cur;
367         int again = 0;
368         unsigned long num_run;
369         unsigned long batch_run = 0;
370         unsigned long limit;
371         unsigned long last_waited = 0;
372         int force_reg = 0;
373         int sync_pending = 0;
374         struct blk_plug plug;
375
376         /*
377          * this function runs all the bios we've collected for
378          * a particular device.  We don't want to wander off to
379          * another device without first sending all of these down.
380          * So, setup a plug here and finish it off before we return
381          */
382         blk_start_plug(&plug);
383
384         bdi = device->bdev->bd_bdi;
385         limit = btrfs_async_submit_limit(fs_info);
386         limit = limit * 2 / 3;
387
388 loop:
389         spin_lock(&device->io_lock);
390
391 loop_lock:
392         num_run = 0;
393
394         /* take all the bios off the list at once and process them
395          * later on (without the lock held).  But, remember the
396          * tail and other pointers so the bios can be properly reinserted
397          * into the list if we hit congestion
398          */
399         if (!force_reg && device->pending_sync_bios.head) {
400                 pending_bios = &device->pending_sync_bios;
401                 force_reg = 1;
402         } else {
403                 pending_bios = &device->pending_bios;
404                 force_reg = 0;
405         }
406
407         pending = pending_bios->head;
408         tail = pending_bios->tail;
409         WARN_ON(pending && !tail);
410
411         /*
412          * if pending was null this time around, no bios need processing
413          * at all and we can stop.  Otherwise it'll loop back up again
414          * and do an additional check so no bios are missed.
415          *
416          * device->running_pending is used to synchronize with the
417          * schedule_bio code.
418          */
419         if (device->pending_sync_bios.head == NULL &&
420             device->pending_bios.head == NULL) {
421                 again = 0;
422                 device->running_pending = 0;
423         } else {
424                 again = 1;
425                 device->running_pending = 1;
426         }
427
428         pending_bios->head = NULL;
429         pending_bios->tail = NULL;
430
431         spin_unlock(&device->io_lock);
432
433         while (pending) {
434
435                 rmb();
436                 /* we want to work on both lists, but do more bios on the
437                  * sync list than the regular list
438                  */
439                 if ((num_run > 32 &&
440                     pending_bios != &device->pending_sync_bios &&
441                     device->pending_sync_bios.head) ||
442                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
443                     device->pending_bios.head)) {
444                         spin_lock(&device->io_lock);
445                         requeue_list(pending_bios, pending, tail);
446                         goto loop_lock;
447                 }
448
449                 cur = pending;
450                 pending = pending->bi_next;
451                 cur->bi_next = NULL;
452
453                 /*
454                  * atomic_dec_return implies a barrier for waitqueue_active
455                  */
456                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
457                     waitqueue_active(&fs_info->async_submit_wait))
458                         wake_up(&fs_info->async_submit_wait);
459
460                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
461
462                 /*
463                  * if we're doing the sync list, record that our
464                  * plug has some sync requests on it
465                  *
466                  * If we're doing the regular list and there are
467                  * sync requests sitting around, unplug before
468                  * we add more
469                  */
470                 if (pending_bios == &device->pending_sync_bios) {
471                         sync_pending = 1;
472                 } else if (sync_pending) {
473                         blk_finish_plug(&plug);
474                         blk_start_plug(&plug);
475                         sync_pending = 0;
476                 }
477
478                 btrfsic_submit_bio(cur);
479                 num_run++;
480                 batch_run++;
481
482                 cond_resched();
483
484                 /*
485                  * we made progress, there is more work to do and the bdi
486                  * is now congested.  Back off and let other work structs
487                  * run instead
488                  */
489                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
490                     fs_info->fs_devices->open_devices > 1) {
491                         struct io_context *ioc;
492
493                         ioc = current->io_context;
494
495                         /*
496                          * the main goal here is that we don't want to
497                          * block if we're going to be able to submit
498                          * more requests without blocking.
499                          *
500                          * This code does two great things, it pokes into
501                          * the elevator code from a filesystem _and_
502                          * it makes assumptions about how batching works.
503                          */
504                         if (ioc && ioc->nr_batch_requests > 0 &&
505                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
506                             (last_waited == 0 ||
507                              ioc->last_waited == last_waited)) {
508                                 /*
509                                  * we want to go through our batch of
510                                  * requests and stop.  So, we copy out
511                                  * the ioc->last_waited time and test
512                                  * against it before looping
513                                  */
514                                 last_waited = ioc->last_waited;
515                                 cond_resched();
516                                 continue;
517                         }
518                         spin_lock(&device->io_lock);
519                         requeue_list(pending_bios, pending, tail);
520                         device->running_pending = 1;
521
522                         spin_unlock(&device->io_lock);
523                         btrfs_queue_work(fs_info->submit_workers,
524                                          &device->work);
525                         goto done;
526                 }
527                 /* unplug every 64 requests just for good measure */
528                 if (batch_run % 64 == 0) {
529                         blk_finish_plug(&plug);
530                         blk_start_plug(&plug);
531                         sync_pending = 0;
532                 }
533         }
534
535         cond_resched();
536         if (again)
537                 goto loop;
538
539         spin_lock(&device->io_lock);
540         if (device->pending_bios.head || device->pending_sync_bios.head)
541                 goto loop_lock;
542         spin_unlock(&device->io_lock);
543
544 done:
545         blk_finish_plug(&plug);
546 }
547
548 static void pending_bios_fn(struct btrfs_work *work)
549 {
550         struct btrfs_device *device;
551
552         device = container_of(work, struct btrfs_device, work);
553         run_scheduled_bios(device);
554 }
555
556
557 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
558 {
559         struct btrfs_fs_devices *fs_devs;
560         struct btrfs_device *dev;
561
562         if (!cur_dev->name)
563                 return;
564
565         list_for_each_entry(fs_devs, &fs_uuids, list) {
566                 int del = 1;
567
568                 if (fs_devs->opened)
569                         continue;
570                 if (fs_devs->seeding)
571                         continue;
572
573                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
574
575                         if (dev == cur_dev)
576                                 continue;
577                         if (!dev->name)
578                                 continue;
579
580                         /*
581                          * Todo: This won't be enough. What if the same device
582                          * comes back (with new uuid and) with its mapper path?
583                          * But for now, this does help as mostly an admin will
584                          * either use mapper or non mapper path throughout.
585                          */
586                         rcu_read_lock();
587                         del = strcmp(rcu_str_deref(dev->name),
588                                                 rcu_str_deref(cur_dev->name));
589                         rcu_read_unlock();
590                         if (!del)
591                                 break;
592                 }
593
594                 if (!del) {
595                         /* delete the stale device */
596                         if (fs_devs->num_devices == 1) {
597                                 btrfs_sysfs_remove_fsid(fs_devs);
598                                 list_del(&fs_devs->list);
599                                 free_fs_devices(fs_devs);
600                         } else {
601                                 fs_devs->num_devices--;
602                                 list_del(&dev->dev_list);
603                                 rcu_string_free(dev->name);
604                                 kfree(dev);
605                         }
606                         break;
607                 }
608         }
609 }
610
611 /*
612  * Add new device to list of registered devices
613  *
614  * Returns:
615  * 1   - first time device is seen
616  * 0   - device already known
617  * < 0 - error
618  */
619 static noinline int device_list_add(const char *path,
620                            struct btrfs_super_block *disk_super,
621                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
622 {
623         struct btrfs_device *device;
624         struct btrfs_fs_devices *fs_devices;
625         struct rcu_string *name;
626         int ret = 0;
627         u64 found_transid = btrfs_super_generation(disk_super);
628
629         fs_devices = find_fsid(disk_super->fsid);
630         if (!fs_devices) {
631                 fs_devices = alloc_fs_devices(disk_super->fsid);
632                 if (IS_ERR(fs_devices))
633                         return PTR_ERR(fs_devices);
634
635                 list_add(&fs_devices->list, &fs_uuids);
636
637                 device = NULL;
638         } else {
639                 device = __find_device(&fs_devices->devices, devid,
640                                        disk_super->dev_item.uuid);
641         }
642
643         if (!device) {
644                 if (fs_devices->opened)
645                         return -EBUSY;
646
647                 device = btrfs_alloc_device(NULL, &devid,
648                                             disk_super->dev_item.uuid);
649                 if (IS_ERR(device)) {
650                         /* we can safely leave the fs_devices entry around */
651                         return PTR_ERR(device);
652                 }
653
654                 name = rcu_string_strdup(path, GFP_NOFS);
655                 if (!name) {
656                         kfree(device);
657                         return -ENOMEM;
658                 }
659                 rcu_assign_pointer(device->name, name);
660
661                 mutex_lock(&fs_devices->device_list_mutex);
662                 list_add_rcu(&device->dev_list, &fs_devices->devices);
663                 fs_devices->num_devices++;
664                 mutex_unlock(&fs_devices->device_list_mutex);
665
666                 ret = 1;
667                 device->fs_devices = fs_devices;
668         } else if (!device->name || strcmp(device->name->str, path)) {
669                 /*
670                  * When FS is already mounted.
671                  * 1. If you are here and if the device->name is NULL that
672                  *    means this device was missing at time of FS mount.
673                  * 2. If you are here and if the device->name is different
674                  *    from 'path' that means either
675                  *      a. The same device disappeared and reappeared with
676                  *         different name. or
677                  *      b. The missing-disk-which-was-replaced, has
678                  *         reappeared now.
679                  *
680                  * We must allow 1 and 2a above. But 2b would be a spurious
681                  * and unintentional.
682                  *
683                  * Further in case of 1 and 2a above, the disk at 'path'
684                  * would have missed some transaction when it was away and
685                  * in case of 2a the stale bdev has to be updated as well.
686                  * 2b must not be allowed at all time.
687                  */
688
689                 /*
690                  * For now, we do allow update to btrfs_fs_device through the
691                  * btrfs dev scan cli after FS has been mounted.  We're still
692                  * tracking a problem where systems fail mount by subvolume id
693                  * when we reject replacement on a mounted FS.
694                  */
695                 if (!fs_devices->opened && found_transid < device->generation) {
696                         /*
697                          * That is if the FS is _not_ mounted and if you
698                          * are here, that means there is more than one
699                          * disk with same uuid and devid.We keep the one
700                          * with larger generation number or the last-in if
701                          * generation are equal.
702                          */
703                         return -EEXIST;
704                 }
705
706                 name = rcu_string_strdup(path, GFP_NOFS);
707                 if (!name)
708                         return -ENOMEM;
709                 rcu_string_free(device->name);
710                 rcu_assign_pointer(device->name, name);
711                 if (device->missing) {
712                         fs_devices->missing_devices--;
713                         device->missing = 0;
714                 }
715         }
716
717         /*
718          * Unmount does not free the btrfs_device struct but would zero
719          * generation along with most of the other members. So just update
720          * it back. We need it to pick the disk with largest generation
721          * (as above).
722          */
723         if (!fs_devices->opened)
724                 device->generation = found_transid;
725
726         /*
727          * if there is new btrfs on an already registered device,
728          * then remove the stale device entry.
729          */
730         if (ret > 0)
731                 btrfs_free_stale_device(device);
732
733         *fs_devices_ret = fs_devices;
734
735         return ret;
736 }
737
738 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
739 {
740         struct btrfs_fs_devices *fs_devices;
741         struct btrfs_device *device;
742         struct btrfs_device *orig_dev;
743
744         fs_devices = alloc_fs_devices(orig->fsid);
745         if (IS_ERR(fs_devices))
746                 return fs_devices;
747
748         mutex_lock(&orig->device_list_mutex);
749         fs_devices->total_devices = orig->total_devices;
750
751         /* We have held the volume lock, it is safe to get the devices. */
752         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
753                 struct rcu_string *name;
754
755                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
756                                             orig_dev->uuid);
757                 if (IS_ERR(device))
758                         goto error;
759
760                 /*
761                  * This is ok to do without rcu read locked because we hold the
762                  * uuid mutex so nothing we touch in here is going to disappear.
763                  */
764                 if (orig_dev->name) {
765                         name = rcu_string_strdup(orig_dev->name->str,
766                                         GFP_KERNEL);
767                         if (!name) {
768                                 kfree(device);
769                                 goto error;
770                         }
771                         rcu_assign_pointer(device->name, name);
772                 }
773
774                 list_add(&device->dev_list, &fs_devices->devices);
775                 device->fs_devices = fs_devices;
776                 fs_devices->num_devices++;
777         }
778         mutex_unlock(&orig->device_list_mutex);
779         return fs_devices;
780 error:
781         mutex_unlock(&orig->device_list_mutex);
782         free_fs_devices(fs_devices);
783         return ERR_PTR(-ENOMEM);
784 }
785
786 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
787 {
788         struct btrfs_device *device, *next;
789         struct btrfs_device *latest_dev = NULL;
790
791         mutex_lock(&uuid_mutex);
792 again:
793         /* This is the initialized path, it is safe to release the devices. */
794         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
795                 if (device->in_fs_metadata) {
796                         if (!device->is_tgtdev_for_dev_replace &&
797                             (!latest_dev ||
798                              device->generation > latest_dev->generation)) {
799                                 latest_dev = device;
800                         }
801                         continue;
802                 }
803
804                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
805                         /*
806                          * In the first step, keep the device which has
807                          * the correct fsid and the devid that is used
808                          * for the dev_replace procedure.
809                          * In the second step, the dev_replace state is
810                          * read from the device tree and it is known
811                          * whether the procedure is really active or
812                          * not, which means whether this device is
813                          * used or whether it should be removed.
814                          */
815                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
816                                 continue;
817                         }
818                 }
819                 if (device->bdev) {
820                         blkdev_put(device->bdev, device->mode);
821                         device->bdev = NULL;
822                         fs_devices->open_devices--;
823                 }
824                 if (device->writeable) {
825                         list_del_init(&device->dev_alloc_list);
826                         device->writeable = 0;
827                         if (!device->is_tgtdev_for_dev_replace)
828                                 fs_devices->rw_devices--;
829                 }
830                 list_del_init(&device->dev_list);
831                 fs_devices->num_devices--;
832                 rcu_string_free(device->name);
833                 kfree(device);
834         }
835
836         if (fs_devices->seed) {
837                 fs_devices = fs_devices->seed;
838                 goto again;
839         }
840
841         fs_devices->latest_bdev = latest_dev->bdev;
842
843         mutex_unlock(&uuid_mutex);
844 }
845
846 static void __free_device(struct work_struct *work)
847 {
848         struct btrfs_device *device;
849
850         device = container_of(work, struct btrfs_device, rcu_work);
851         rcu_string_free(device->name);
852         bio_put(device->flush_bio);
853         kfree(device);
854 }
855
856 static void free_device(struct rcu_head *head)
857 {
858         struct btrfs_device *device;
859
860         device = container_of(head, struct btrfs_device, rcu);
861
862         INIT_WORK(&device->rcu_work, __free_device);
863         schedule_work(&device->rcu_work);
864 }
865
866 static void btrfs_close_bdev(struct btrfs_device *device)
867 {
868         if (device->bdev && device->writeable) {
869                 sync_blockdev(device->bdev);
870                 invalidate_bdev(device->bdev);
871         }
872
873         if (device->bdev)
874                 blkdev_put(device->bdev, device->mode);
875 }
876
877 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
878 {
879         struct btrfs_fs_devices *fs_devices = device->fs_devices;
880         struct btrfs_device *new_device;
881         struct rcu_string *name;
882
883         if (device->bdev)
884                 fs_devices->open_devices--;
885
886         if (device->writeable &&
887             device->devid != BTRFS_DEV_REPLACE_DEVID) {
888                 list_del_init(&device->dev_alloc_list);
889                 fs_devices->rw_devices--;
890         }
891
892         if (device->missing)
893                 fs_devices->missing_devices--;
894
895         new_device = btrfs_alloc_device(NULL, &device->devid,
896                                         device->uuid);
897         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
898
899         /* Safe because we are under uuid_mutex */
900         if (device->name) {
901                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
902                 BUG_ON(!name); /* -ENOMEM */
903                 rcu_assign_pointer(new_device->name, name);
904         }
905
906         list_replace_rcu(&device->dev_list, &new_device->dev_list);
907         new_device->fs_devices = device->fs_devices;
908 }
909
910 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
911 {
912         struct btrfs_device *device, *tmp;
913         struct list_head pending_put;
914
915         INIT_LIST_HEAD(&pending_put);
916
917         if (--fs_devices->opened > 0)
918                 return 0;
919
920         mutex_lock(&fs_devices->device_list_mutex);
921         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
922                 btrfs_prepare_close_one_device(device);
923                 list_add(&device->dev_list, &pending_put);
924         }
925         mutex_unlock(&fs_devices->device_list_mutex);
926
927         /*
928          * btrfs_show_devname() is using the device_list_mutex,
929          * sometimes call to blkdev_put() leads vfs calling
930          * into this func. So do put outside of device_list_mutex,
931          * as of now.
932          */
933         while (!list_empty(&pending_put)) {
934                 device = list_first_entry(&pending_put,
935                                 struct btrfs_device, dev_list);
936                 list_del(&device->dev_list);
937                 btrfs_close_bdev(device);
938                 call_rcu(&device->rcu, free_device);
939         }
940
941         WARN_ON(fs_devices->open_devices);
942         WARN_ON(fs_devices->rw_devices);
943         fs_devices->opened = 0;
944         fs_devices->seeding = 0;
945
946         return 0;
947 }
948
949 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
950 {
951         struct btrfs_fs_devices *seed_devices = NULL;
952         int ret;
953
954         mutex_lock(&uuid_mutex);
955         ret = __btrfs_close_devices(fs_devices);
956         if (!fs_devices->opened) {
957                 seed_devices = fs_devices->seed;
958                 fs_devices->seed = NULL;
959         }
960         mutex_unlock(&uuid_mutex);
961
962         while (seed_devices) {
963                 fs_devices = seed_devices;
964                 seed_devices = fs_devices->seed;
965                 __btrfs_close_devices(fs_devices);
966                 free_fs_devices(fs_devices);
967         }
968         /*
969          * Wait for rcu kworkers under __btrfs_close_devices
970          * to finish all blkdev_puts so device is really
971          * free when umount is done.
972          */
973         rcu_barrier();
974         return ret;
975 }
976
977 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
978                                 fmode_t flags, void *holder)
979 {
980         struct request_queue *q;
981         struct block_device *bdev;
982         struct list_head *head = &fs_devices->devices;
983         struct btrfs_device *device;
984         struct btrfs_device *latest_dev = NULL;
985         struct buffer_head *bh;
986         struct btrfs_super_block *disk_super;
987         u64 devid;
988         int seeding = 1;
989         int ret = 0;
990
991         flags |= FMODE_EXCL;
992
993         list_for_each_entry(device, head, dev_list) {
994                 if (device->bdev)
995                         continue;
996                 if (!device->name)
997                         continue;
998
999                 /* Just open everything we can; ignore failures here */
1000                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
1001                                             &bdev, &bh))
1002                         continue;
1003
1004                 disk_super = (struct btrfs_super_block *)bh->b_data;
1005                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1006                 if (devid != device->devid)
1007                         goto error_brelse;
1008
1009                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
1010                            BTRFS_UUID_SIZE))
1011                         goto error_brelse;
1012
1013                 device->generation = btrfs_super_generation(disk_super);
1014                 if (!latest_dev ||
1015                     device->generation > latest_dev->generation)
1016                         latest_dev = device;
1017
1018                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1019                         device->writeable = 0;
1020                 } else {
1021                         device->writeable = !bdev_read_only(bdev);
1022                         seeding = 0;
1023                 }
1024
1025                 q = bdev_get_queue(bdev);
1026                 if (blk_queue_discard(q))
1027                         device->can_discard = 1;
1028                 if (!blk_queue_nonrot(q))
1029                         fs_devices->rotating = 1;
1030
1031                 device->bdev = bdev;
1032                 device->in_fs_metadata = 0;
1033                 device->mode = flags;
1034
1035                 fs_devices->open_devices++;
1036                 if (device->writeable &&
1037                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1038                         fs_devices->rw_devices++;
1039                         list_add(&device->dev_alloc_list,
1040                                  &fs_devices->alloc_list);
1041                 }
1042                 brelse(bh);
1043                 continue;
1044
1045 error_brelse:
1046                 brelse(bh);
1047                 blkdev_put(bdev, flags);
1048                 continue;
1049         }
1050         if (fs_devices->open_devices == 0) {
1051                 ret = -EINVAL;
1052                 goto out;
1053         }
1054         fs_devices->seeding = seeding;
1055         fs_devices->opened = 1;
1056         fs_devices->latest_bdev = latest_dev->bdev;
1057         fs_devices->total_rw_bytes = 0;
1058 out:
1059         return ret;
1060 }
1061
1062 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1063                        fmode_t flags, void *holder)
1064 {
1065         int ret;
1066
1067         mutex_lock(&uuid_mutex);
1068         if (fs_devices->opened) {
1069                 fs_devices->opened++;
1070                 ret = 0;
1071         } else {
1072                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1073         }
1074         mutex_unlock(&uuid_mutex);
1075         return ret;
1076 }
1077
1078 void btrfs_release_disk_super(struct page *page)
1079 {
1080         kunmap(page);
1081         put_page(page);
1082 }
1083
1084 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1085                 struct page **page, struct btrfs_super_block **disk_super)
1086 {
1087         void *p;
1088         pgoff_t index;
1089
1090         /* make sure our super fits in the device */
1091         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1092                 return 1;
1093
1094         /* make sure our super fits in the page */
1095         if (sizeof(**disk_super) > PAGE_SIZE)
1096                 return 1;
1097
1098         /* make sure our super doesn't straddle pages on disk */
1099         index = bytenr >> PAGE_SHIFT;
1100         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1101                 return 1;
1102
1103         /* pull in the page with our super */
1104         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1105                                    index, GFP_KERNEL);
1106
1107         if (IS_ERR_OR_NULL(*page))
1108                 return 1;
1109
1110         p = kmap(*page);
1111
1112         /* align our pointer to the offset of the super block */
1113         *disk_super = p + (bytenr & ~PAGE_MASK);
1114
1115         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1116             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1117                 btrfs_release_disk_super(*page);
1118                 return 1;
1119         }
1120
1121         if ((*disk_super)->label[0] &&
1122                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1123                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1124
1125         return 0;
1126 }
1127
1128 /*
1129  * Look for a btrfs signature on a device. This may be called out of the mount path
1130  * and we are not allowed to call set_blocksize during the scan. The superblock
1131  * is read via pagecache
1132  */
1133 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1134                           struct btrfs_fs_devices **fs_devices_ret)
1135 {
1136         struct btrfs_super_block *disk_super;
1137         struct block_device *bdev;
1138         struct page *page;
1139         int ret = -EINVAL;
1140         u64 devid;
1141         u64 transid;
1142         u64 total_devices;
1143         u64 bytenr;
1144
1145         /*
1146          * we would like to check all the supers, but that would make
1147          * a btrfs mount succeed after a mkfs from a different FS.
1148          * So, we need to add a special mount option to scan for
1149          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1150          */
1151         bytenr = btrfs_sb_offset(0);
1152         flags |= FMODE_EXCL;
1153         mutex_lock(&uuid_mutex);
1154
1155         bdev = blkdev_get_by_path(path, flags, holder);
1156         if (IS_ERR(bdev)) {
1157                 ret = PTR_ERR(bdev);
1158                 goto error;
1159         }
1160
1161         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1162                 goto error_bdev_put;
1163
1164         devid = btrfs_stack_device_id(&disk_super->dev_item);
1165         transid = btrfs_super_generation(disk_super);
1166         total_devices = btrfs_super_num_devices(disk_super);
1167
1168         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1169         if (ret > 0) {
1170                 if (disk_super->label[0]) {
1171                         pr_info("BTRFS: device label %s ", disk_super->label);
1172                 } else {
1173                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1174                 }
1175
1176                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1177                 ret = 0;
1178         }
1179         if (!ret && fs_devices_ret)
1180                 (*fs_devices_ret)->total_devices = total_devices;
1181
1182         btrfs_release_disk_super(page);
1183
1184 error_bdev_put:
1185         blkdev_put(bdev, flags);
1186 error:
1187         mutex_unlock(&uuid_mutex);
1188         return ret;
1189 }
1190
1191 /* helper to account the used device space in the range */
1192 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1193                                    u64 end, u64 *length)
1194 {
1195         struct btrfs_key key;
1196         struct btrfs_root *root = device->fs_info->dev_root;
1197         struct btrfs_dev_extent *dev_extent;
1198         struct btrfs_path *path;
1199         u64 extent_end;
1200         int ret;
1201         int slot;
1202         struct extent_buffer *l;
1203
1204         *length = 0;
1205
1206         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1207                 return 0;
1208
1209         path = btrfs_alloc_path();
1210         if (!path)
1211                 return -ENOMEM;
1212         path->reada = READA_FORWARD;
1213
1214         key.objectid = device->devid;
1215         key.offset = start;
1216         key.type = BTRFS_DEV_EXTENT_KEY;
1217
1218         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1219         if (ret < 0)
1220                 goto out;
1221         if (ret > 0) {
1222                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1223                 if (ret < 0)
1224                         goto out;
1225         }
1226
1227         while (1) {
1228                 l = path->nodes[0];
1229                 slot = path->slots[0];
1230                 if (slot >= btrfs_header_nritems(l)) {
1231                         ret = btrfs_next_leaf(root, path);
1232                         if (ret == 0)
1233                                 continue;
1234                         if (ret < 0)
1235                                 goto out;
1236
1237                         break;
1238                 }
1239                 btrfs_item_key_to_cpu(l, &key, slot);
1240
1241                 if (key.objectid < device->devid)
1242                         goto next;
1243
1244                 if (key.objectid > device->devid)
1245                         break;
1246
1247                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1248                         goto next;
1249
1250                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1251                 extent_end = key.offset + btrfs_dev_extent_length(l,
1252                                                                   dev_extent);
1253                 if (key.offset <= start && extent_end > end) {
1254                         *length = end - start + 1;
1255                         break;
1256                 } else if (key.offset <= start && extent_end > start)
1257                         *length += extent_end - start;
1258                 else if (key.offset > start && extent_end <= end)
1259                         *length += extent_end - key.offset;
1260                 else if (key.offset > start && key.offset <= end) {
1261                         *length += end - key.offset + 1;
1262                         break;
1263                 } else if (key.offset > end)
1264                         break;
1265
1266 next:
1267                 path->slots[0]++;
1268         }
1269         ret = 0;
1270 out:
1271         btrfs_free_path(path);
1272         return ret;
1273 }
1274
1275 static int contains_pending_extent(struct btrfs_transaction *transaction,
1276                                    struct btrfs_device *device,
1277                                    u64 *start, u64 len)
1278 {
1279         struct btrfs_fs_info *fs_info = device->fs_info;
1280         struct extent_map *em;
1281         struct list_head *search_list = &fs_info->pinned_chunks;
1282         int ret = 0;
1283         u64 physical_start = *start;
1284
1285         if (transaction)
1286                 search_list = &transaction->pending_chunks;
1287 again:
1288         list_for_each_entry(em, search_list, list) {
1289                 struct map_lookup *map;
1290                 int i;
1291
1292                 map = em->map_lookup;
1293                 for (i = 0; i < map->num_stripes; i++) {
1294                         u64 end;
1295
1296                         if (map->stripes[i].dev != device)
1297                                 continue;
1298                         if (map->stripes[i].physical >= physical_start + len ||
1299                             map->stripes[i].physical + em->orig_block_len <=
1300                             physical_start)
1301                                 continue;
1302                         /*
1303                          * Make sure that while processing the pinned list we do
1304                          * not override our *start with a lower value, because
1305                          * we can have pinned chunks that fall within this
1306                          * device hole and that have lower physical addresses
1307                          * than the pending chunks we processed before. If we
1308                          * do not take this special care we can end up getting
1309                          * 2 pending chunks that start at the same physical
1310                          * device offsets because the end offset of a pinned
1311                          * chunk can be equal to the start offset of some
1312                          * pending chunk.
1313                          */
1314                         end = map->stripes[i].physical + em->orig_block_len;
1315                         if (end > *start) {
1316                                 *start = end;
1317                                 ret = 1;
1318                         }
1319                 }
1320         }
1321         if (search_list != &fs_info->pinned_chunks) {
1322                 search_list = &fs_info->pinned_chunks;
1323                 goto again;
1324         }
1325
1326         return ret;
1327 }
1328
1329
1330 /*
1331  * find_free_dev_extent_start - find free space in the specified device
1332  * @device:       the device which we search the free space in
1333  * @num_bytes:    the size of the free space that we need
1334  * @search_start: the position from which to begin the search
1335  * @start:        store the start of the free space.
1336  * @len:          the size of the free space. that we find, or the size
1337  *                of the max free space if we don't find suitable free space
1338  *
1339  * this uses a pretty simple search, the expectation is that it is
1340  * called very infrequently and that a given device has a small number
1341  * of extents
1342  *
1343  * @start is used to store the start of the free space if we find. But if we
1344  * don't find suitable free space, it will be used to store the start position
1345  * of the max free space.
1346  *
1347  * @len is used to store the size of the free space that we find.
1348  * But if we don't find suitable free space, it is used to store the size of
1349  * the max free space.
1350  */
1351 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1352                                struct btrfs_device *device, u64 num_bytes,
1353                                u64 search_start, u64 *start, u64 *len)
1354 {
1355         struct btrfs_fs_info *fs_info = device->fs_info;
1356         struct btrfs_root *root = fs_info->dev_root;
1357         struct btrfs_key key;
1358         struct btrfs_dev_extent *dev_extent;
1359         struct btrfs_path *path;
1360         u64 hole_size;
1361         u64 max_hole_start;
1362         u64 max_hole_size;
1363         u64 extent_end;
1364         u64 search_end = device->total_bytes;
1365         int ret;
1366         int slot;
1367         struct extent_buffer *l;
1368
1369         /*
1370          * We don't want to overwrite the superblock on the drive nor any area
1371          * used by the boot loader (grub for example), so we make sure to start
1372          * at an offset of at least 1MB.
1373          */
1374         search_start = max_t(u64, search_start, SZ_1M);
1375
1376         path = btrfs_alloc_path();
1377         if (!path)
1378                 return -ENOMEM;
1379
1380         max_hole_start = search_start;
1381         max_hole_size = 0;
1382
1383 again:
1384         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1385                 ret = -ENOSPC;
1386                 goto out;
1387         }
1388
1389         path->reada = READA_FORWARD;
1390         path->search_commit_root = 1;
1391         path->skip_locking = 1;
1392
1393         key.objectid = device->devid;
1394         key.offset = search_start;
1395         key.type = BTRFS_DEV_EXTENT_KEY;
1396
1397         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1398         if (ret < 0)
1399                 goto out;
1400         if (ret > 0) {
1401                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1402                 if (ret < 0)
1403                         goto out;
1404         }
1405
1406         while (1) {
1407                 l = path->nodes[0];
1408                 slot = path->slots[0];
1409                 if (slot >= btrfs_header_nritems(l)) {
1410                         ret = btrfs_next_leaf(root, path);
1411                         if (ret == 0)
1412                                 continue;
1413                         if (ret < 0)
1414                                 goto out;
1415
1416                         break;
1417                 }
1418                 btrfs_item_key_to_cpu(l, &key, slot);
1419
1420                 if (key.objectid < device->devid)
1421                         goto next;
1422
1423                 if (key.objectid > device->devid)
1424                         break;
1425
1426                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1427                         goto next;
1428
1429                 if (key.offset > search_start) {
1430                         hole_size = key.offset - search_start;
1431
1432                         /*
1433                          * Have to check before we set max_hole_start, otherwise
1434                          * we could end up sending back this offset anyway.
1435                          */
1436                         if (contains_pending_extent(transaction, device,
1437                                                     &search_start,
1438                                                     hole_size)) {
1439                                 if (key.offset >= search_start) {
1440                                         hole_size = key.offset - search_start;
1441                                 } else {
1442                                         WARN_ON_ONCE(1);
1443                                         hole_size = 0;
1444                                 }
1445                         }
1446
1447                         if (hole_size > max_hole_size) {
1448                                 max_hole_start = search_start;
1449                                 max_hole_size = hole_size;
1450                         }
1451
1452                         /*
1453                          * If this free space is greater than which we need,
1454                          * it must be the max free space that we have found
1455                          * until now, so max_hole_start must point to the start
1456                          * of this free space and the length of this free space
1457                          * is stored in max_hole_size. Thus, we return
1458                          * max_hole_start and max_hole_size and go back to the
1459                          * caller.
1460                          */
1461                         if (hole_size >= num_bytes) {
1462                                 ret = 0;
1463                                 goto out;
1464                         }
1465                 }
1466
1467                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1468                 extent_end = key.offset + btrfs_dev_extent_length(l,
1469                                                                   dev_extent);
1470                 if (extent_end > search_start)
1471                         search_start = extent_end;
1472 next:
1473                 path->slots[0]++;
1474                 cond_resched();
1475         }
1476
1477         /*
1478          * At this point, search_start should be the end of
1479          * allocated dev extents, and when shrinking the device,
1480          * search_end may be smaller than search_start.
1481          */
1482         if (search_end > search_start) {
1483                 hole_size = search_end - search_start;
1484
1485                 if (contains_pending_extent(transaction, device, &search_start,
1486                                             hole_size)) {
1487                         btrfs_release_path(path);
1488                         goto again;
1489                 }
1490
1491                 if (hole_size > max_hole_size) {
1492                         max_hole_start = search_start;
1493                         max_hole_size = hole_size;
1494                 }
1495         }
1496
1497         /* See above. */
1498         if (max_hole_size < num_bytes)
1499                 ret = -ENOSPC;
1500         else
1501                 ret = 0;
1502
1503 out:
1504         btrfs_free_path(path);
1505         *start = max_hole_start;
1506         if (len)
1507                 *len = max_hole_size;
1508         return ret;
1509 }
1510
1511 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1512                          struct btrfs_device *device, u64 num_bytes,
1513                          u64 *start, u64 *len)
1514 {
1515         /* FIXME use last free of some kind */
1516         return find_free_dev_extent_start(trans->transaction, device,
1517                                           num_bytes, 0, start, len);
1518 }
1519
1520 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1521                           struct btrfs_device *device,
1522                           u64 start, u64 *dev_extent_len)
1523 {
1524         struct btrfs_fs_info *fs_info = device->fs_info;
1525         struct btrfs_root *root = fs_info->dev_root;
1526         int ret;
1527         struct btrfs_path *path;
1528         struct btrfs_key key;
1529         struct btrfs_key found_key;
1530         struct extent_buffer *leaf = NULL;
1531         struct btrfs_dev_extent *extent = NULL;
1532
1533         path = btrfs_alloc_path();
1534         if (!path)
1535                 return -ENOMEM;
1536
1537         key.objectid = device->devid;
1538         key.offset = start;
1539         key.type = BTRFS_DEV_EXTENT_KEY;
1540 again:
1541         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1542         if (ret > 0) {
1543                 ret = btrfs_previous_item(root, path, key.objectid,
1544                                           BTRFS_DEV_EXTENT_KEY);
1545                 if (ret)
1546                         goto out;
1547                 leaf = path->nodes[0];
1548                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1549                 extent = btrfs_item_ptr(leaf, path->slots[0],
1550                                         struct btrfs_dev_extent);
1551                 BUG_ON(found_key.offset > start || found_key.offset +
1552                        btrfs_dev_extent_length(leaf, extent) < start);
1553                 key = found_key;
1554                 btrfs_release_path(path);
1555                 goto again;
1556         } else if (ret == 0) {
1557                 leaf = path->nodes[0];
1558                 extent = btrfs_item_ptr(leaf, path->slots[0],
1559                                         struct btrfs_dev_extent);
1560         } else {
1561                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1562                 goto out;
1563         }
1564
1565         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1566
1567         ret = btrfs_del_item(trans, root, path);
1568         if (ret) {
1569                 btrfs_handle_fs_error(fs_info, ret,
1570                                       "Failed to remove dev extent item");
1571         } else {
1572                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1573         }
1574 out:
1575         btrfs_free_path(path);
1576         return ret;
1577 }
1578
1579 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1580                                   struct btrfs_device *device,
1581                                   u64 chunk_tree, u64 chunk_objectid,
1582                                   u64 chunk_offset, u64 start, u64 num_bytes)
1583 {
1584         int ret;
1585         struct btrfs_path *path;
1586         struct btrfs_fs_info *fs_info = device->fs_info;
1587         struct btrfs_root *root = fs_info->dev_root;
1588         struct btrfs_dev_extent *extent;
1589         struct extent_buffer *leaf;
1590         struct btrfs_key key;
1591
1592         WARN_ON(!device->in_fs_metadata);
1593         WARN_ON(device->is_tgtdev_for_dev_replace);
1594         path = btrfs_alloc_path();
1595         if (!path)
1596                 return -ENOMEM;
1597
1598         key.objectid = device->devid;
1599         key.offset = start;
1600         key.type = BTRFS_DEV_EXTENT_KEY;
1601         ret = btrfs_insert_empty_item(trans, root, path, &key,
1602                                       sizeof(*extent));
1603         if (ret)
1604                 goto out;
1605
1606         leaf = path->nodes[0];
1607         extent = btrfs_item_ptr(leaf, path->slots[0],
1608                                 struct btrfs_dev_extent);
1609         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1610         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1611         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1612
1613         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1614
1615         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1616         btrfs_mark_buffer_dirty(leaf);
1617 out:
1618         btrfs_free_path(path);
1619         return ret;
1620 }
1621
1622 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1623 {
1624         struct extent_map_tree *em_tree;
1625         struct extent_map *em;
1626         struct rb_node *n;
1627         u64 ret = 0;
1628
1629         em_tree = &fs_info->mapping_tree.map_tree;
1630         read_lock(&em_tree->lock);
1631         n = rb_last(&em_tree->map);
1632         if (n) {
1633                 em = rb_entry(n, struct extent_map, rb_node);
1634                 ret = em->start + em->len;
1635         }
1636         read_unlock(&em_tree->lock);
1637
1638         return ret;
1639 }
1640
1641 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1642                                     u64 *devid_ret)
1643 {
1644         int ret;
1645         struct btrfs_key key;
1646         struct btrfs_key found_key;
1647         struct btrfs_path *path;
1648
1649         path = btrfs_alloc_path();
1650         if (!path)
1651                 return -ENOMEM;
1652
1653         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1654         key.type = BTRFS_DEV_ITEM_KEY;
1655         key.offset = (u64)-1;
1656
1657         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1658         if (ret < 0)
1659                 goto error;
1660
1661         BUG_ON(ret == 0); /* Corruption */
1662
1663         ret = btrfs_previous_item(fs_info->chunk_root, path,
1664                                   BTRFS_DEV_ITEMS_OBJECTID,
1665                                   BTRFS_DEV_ITEM_KEY);
1666         if (ret) {
1667                 *devid_ret = 1;
1668         } else {
1669                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1670                                       path->slots[0]);
1671                 *devid_ret = found_key.offset + 1;
1672         }
1673         ret = 0;
1674 error:
1675         btrfs_free_path(path);
1676         return ret;
1677 }
1678
1679 /*
1680  * the device information is stored in the chunk root
1681  * the btrfs_device struct should be fully filled in
1682  */
1683 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1684                             struct btrfs_fs_info *fs_info,
1685                             struct btrfs_device *device)
1686 {
1687         struct btrfs_root *root = fs_info->chunk_root;
1688         int ret;
1689         struct btrfs_path *path;
1690         struct btrfs_dev_item *dev_item;
1691         struct extent_buffer *leaf;
1692         struct btrfs_key key;
1693         unsigned long ptr;
1694
1695         path = btrfs_alloc_path();
1696         if (!path)
1697                 return -ENOMEM;
1698
1699         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1700         key.type = BTRFS_DEV_ITEM_KEY;
1701         key.offset = device->devid;
1702
1703         ret = btrfs_insert_empty_item(trans, root, path, &key,
1704                                       sizeof(*dev_item));
1705         if (ret)
1706                 goto out;
1707
1708         leaf = path->nodes[0];
1709         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1710
1711         btrfs_set_device_id(leaf, dev_item, device->devid);
1712         btrfs_set_device_generation(leaf, dev_item, 0);
1713         btrfs_set_device_type(leaf, dev_item, device->type);
1714         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1715         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1716         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1717         btrfs_set_device_total_bytes(leaf, dev_item,
1718                                      btrfs_device_get_disk_total_bytes(device));
1719         btrfs_set_device_bytes_used(leaf, dev_item,
1720                                     btrfs_device_get_bytes_used(device));
1721         btrfs_set_device_group(leaf, dev_item, 0);
1722         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1723         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1724         btrfs_set_device_start_offset(leaf, dev_item, 0);
1725
1726         ptr = btrfs_device_uuid(dev_item);
1727         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1728         ptr = btrfs_device_fsid(dev_item);
1729         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1730         btrfs_mark_buffer_dirty(leaf);
1731
1732         ret = 0;
1733 out:
1734         btrfs_free_path(path);
1735         return ret;
1736 }
1737
1738 /*
1739  * Function to update ctime/mtime for a given device path.
1740  * Mainly used for ctime/mtime based probe like libblkid.
1741  */
1742 static void update_dev_time(const char *path_name)
1743 {
1744         struct file *filp;
1745
1746         filp = filp_open(path_name, O_RDWR, 0);
1747         if (IS_ERR(filp))
1748                 return;
1749         file_update_time(filp);
1750         filp_close(filp, NULL);
1751 }
1752
1753 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1754                              struct btrfs_device *device)
1755 {
1756         struct btrfs_root *root = fs_info->chunk_root;
1757         int ret;
1758         struct btrfs_path *path;
1759         struct btrfs_key key;
1760         struct btrfs_trans_handle *trans;
1761
1762         path = btrfs_alloc_path();
1763         if (!path)
1764                 return -ENOMEM;
1765
1766         trans = btrfs_start_transaction(root, 0);
1767         if (IS_ERR(trans)) {
1768                 btrfs_free_path(path);
1769                 return PTR_ERR(trans);
1770         }
1771         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1772         key.type = BTRFS_DEV_ITEM_KEY;
1773         key.offset = device->devid;
1774
1775         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1776         if (ret < 0)
1777                 goto out;
1778
1779         if (ret > 0) {
1780                 ret = -ENOENT;
1781                 goto out;
1782         }
1783
1784         ret = btrfs_del_item(trans, root, path);
1785         if (ret)
1786                 goto out;
1787 out:
1788         btrfs_free_path(path);
1789         btrfs_commit_transaction(trans);
1790         return ret;
1791 }
1792
1793 /*
1794  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1795  * filesystem. It's up to the caller to adjust that number regarding eg. device
1796  * replace.
1797  */
1798 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1799                 u64 num_devices)
1800 {
1801         u64 all_avail;
1802         unsigned seq;
1803         int i;
1804
1805         do {
1806                 seq = read_seqbegin(&fs_info->profiles_lock);
1807
1808                 all_avail = fs_info->avail_data_alloc_bits |
1809                             fs_info->avail_system_alloc_bits |
1810                             fs_info->avail_metadata_alloc_bits;
1811         } while (read_seqretry(&fs_info->profiles_lock, seq));
1812
1813         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1814                 if (!(all_avail & btrfs_raid_group[i]))
1815                         continue;
1816
1817                 if (num_devices < btrfs_raid_array[i].devs_min) {
1818                         int ret = btrfs_raid_mindev_error[i];
1819
1820                         if (ret)
1821                                 return ret;
1822                 }
1823         }
1824
1825         return 0;
1826 }
1827
1828 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1829                                         struct btrfs_device *device)
1830 {
1831         struct btrfs_device *next_device;
1832
1833         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1834                 if (next_device != device &&
1835                         !next_device->missing && next_device->bdev)
1836                         return next_device;
1837         }
1838
1839         return NULL;
1840 }
1841
1842 /*
1843  * Helper function to check if the given device is part of s_bdev / latest_bdev
1844  * and replace it with the provided or the next active device, in the context
1845  * where this function called, there should be always be another device (or
1846  * this_dev) which is active.
1847  */
1848 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1849                 struct btrfs_device *device, struct btrfs_device *this_dev)
1850 {
1851         struct btrfs_device *next_device;
1852
1853         if (this_dev)
1854                 next_device = this_dev;
1855         else
1856                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1857                                                                 device);
1858         ASSERT(next_device);
1859
1860         if (fs_info->sb->s_bdev &&
1861                         (fs_info->sb->s_bdev == device->bdev))
1862                 fs_info->sb->s_bdev = next_device->bdev;
1863
1864         if (fs_info->fs_devices->latest_bdev == device->bdev)
1865                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1866 }
1867
1868 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1869                 u64 devid)
1870 {
1871         struct btrfs_device *device;
1872         struct btrfs_fs_devices *cur_devices;
1873         u64 num_devices;
1874         int ret = 0;
1875         bool clear_super = false;
1876
1877         mutex_lock(&uuid_mutex);
1878
1879         num_devices = fs_info->fs_devices->num_devices;
1880         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1881         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1882                 WARN_ON(num_devices < 1);
1883                 num_devices--;
1884         }
1885         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1886
1887         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1888         if (ret)
1889                 goto out;
1890
1891         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1892                                            &device);
1893         if (ret)
1894                 goto out;
1895
1896         if (device->is_tgtdev_for_dev_replace) {
1897                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1898                 goto out;
1899         }
1900
1901         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1902                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1903                 goto out;
1904         }
1905
1906         if (device->writeable) {
1907                 mutex_lock(&fs_info->chunk_mutex);
1908                 list_del_init(&device->dev_alloc_list);
1909                 device->fs_devices->rw_devices--;
1910                 mutex_unlock(&fs_info->chunk_mutex);
1911                 clear_super = true;
1912         }
1913
1914         mutex_unlock(&uuid_mutex);
1915         ret = btrfs_shrink_device(device, 0);
1916         mutex_lock(&uuid_mutex);
1917         if (ret)
1918                 goto error_undo;
1919
1920         /*
1921          * TODO: the superblock still includes this device in its num_devices
1922          * counter although write_all_supers() is not locked out. This
1923          * could give a filesystem state which requires a degraded mount.
1924          */
1925         ret = btrfs_rm_dev_item(fs_info, device);
1926         if (ret)
1927                 goto error_undo;
1928
1929         device->in_fs_metadata = 0;
1930         btrfs_scrub_cancel_dev(fs_info, device);
1931
1932         /*
1933          * the device list mutex makes sure that we don't change
1934          * the device list while someone else is writing out all
1935          * the device supers. Whoever is writing all supers, should
1936          * lock the device list mutex before getting the number of
1937          * devices in the super block (super_copy). Conversely,
1938          * whoever updates the number of devices in the super block
1939          * (super_copy) should hold the device list mutex.
1940          */
1941
1942         cur_devices = device->fs_devices;
1943         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1944         list_del_rcu(&device->dev_list);
1945
1946         device->fs_devices->num_devices--;
1947         device->fs_devices->total_devices--;
1948
1949         if (device->missing)
1950                 device->fs_devices->missing_devices--;
1951
1952         btrfs_assign_next_active_device(fs_info, device, NULL);
1953
1954         if (device->bdev) {
1955                 device->fs_devices->open_devices--;
1956                 /* remove sysfs entry */
1957                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1958         }
1959
1960         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1961         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1962         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1963
1964         /*
1965          * at this point, the device is zero sized and detached from
1966          * the devices list.  All that's left is to zero out the old
1967          * supers and free the device.
1968          */
1969         if (device->writeable)
1970                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1971
1972         btrfs_close_bdev(device);
1973         call_rcu(&device->rcu, free_device);
1974
1975         if (cur_devices->open_devices == 0) {
1976                 struct btrfs_fs_devices *fs_devices;
1977                 fs_devices = fs_info->fs_devices;
1978                 while (fs_devices) {
1979                         if (fs_devices->seed == cur_devices) {
1980                                 fs_devices->seed = cur_devices->seed;
1981                                 break;
1982                         }
1983                         fs_devices = fs_devices->seed;
1984                 }
1985                 cur_devices->seed = NULL;
1986                 __btrfs_close_devices(cur_devices);
1987                 free_fs_devices(cur_devices);
1988         }
1989
1990         fs_info->num_tolerated_disk_barrier_failures =
1991                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1992
1993 out:
1994         mutex_unlock(&uuid_mutex);
1995         return ret;
1996
1997 error_undo:
1998         if (device->writeable) {
1999                 mutex_lock(&fs_info->chunk_mutex);
2000                 list_add(&device->dev_alloc_list,
2001                          &fs_info->fs_devices->alloc_list);
2002                 device->fs_devices->rw_devices++;
2003                 mutex_unlock(&fs_info->chunk_mutex);
2004         }
2005         goto out;
2006 }
2007
2008 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2009                                         struct btrfs_device *srcdev)
2010 {
2011         struct btrfs_fs_devices *fs_devices;
2012
2013         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2014
2015         /*
2016          * in case of fs with no seed, srcdev->fs_devices will point
2017          * to fs_devices of fs_info. However when the dev being replaced is
2018          * a seed dev it will point to the seed's local fs_devices. In short
2019          * srcdev will have its correct fs_devices in both the cases.
2020          */
2021         fs_devices = srcdev->fs_devices;
2022
2023         list_del_rcu(&srcdev->dev_list);
2024         list_del_rcu(&srcdev->dev_alloc_list);
2025         fs_devices->num_devices--;
2026         if (srcdev->missing)
2027                 fs_devices->missing_devices--;
2028
2029         if (srcdev->writeable)
2030                 fs_devices->rw_devices--;
2031
2032         if (srcdev->bdev)
2033                 fs_devices->open_devices--;
2034 }
2035
2036 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2037                                       struct btrfs_device *srcdev)
2038 {
2039         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2040
2041         if (srcdev->writeable) {
2042                 /* zero out the old super if it is writable */
2043                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2044         }
2045
2046         btrfs_close_bdev(srcdev);
2047
2048         call_rcu(&srcdev->rcu, free_device);
2049
2050         /*
2051          * unless fs_devices is seed fs, num_devices shouldn't go
2052          * zero
2053          */
2054         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2055
2056         /* if this is no devs we rather delete the fs_devices */
2057         if (!fs_devices->num_devices) {
2058                 struct btrfs_fs_devices *tmp_fs_devices;
2059
2060                 tmp_fs_devices = fs_info->fs_devices;
2061                 while (tmp_fs_devices) {
2062                         if (tmp_fs_devices->seed == fs_devices) {
2063                                 tmp_fs_devices->seed = fs_devices->seed;
2064                                 break;
2065                         }
2066                         tmp_fs_devices = tmp_fs_devices->seed;
2067                 }
2068                 fs_devices->seed = NULL;
2069                 __btrfs_close_devices(fs_devices);
2070                 free_fs_devices(fs_devices);
2071         }
2072 }
2073
2074 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2075                                       struct btrfs_device *tgtdev)
2076 {
2077         mutex_lock(&uuid_mutex);
2078         WARN_ON(!tgtdev);
2079         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2080
2081         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2082
2083         if (tgtdev->bdev)
2084                 fs_info->fs_devices->open_devices--;
2085
2086         fs_info->fs_devices->num_devices--;
2087
2088         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2089
2090         list_del_rcu(&tgtdev->dev_list);
2091
2092         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2093         mutex_unlock(&uuid_mutex);
2094
2095         /*
2096          * The update_dev_time() with in btrfs_scratch_superblocks()
2097          * may lead to a call to btrfs_show_devname() which will try
2098          * to hold device_list_mutex. And here this device
2099          * is already out of device list, so we don't have to hold
2100          * the device_list_mutex lock.
2101          */
2102         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2103
2104         btrfs_close_bdev(tgtdev);
2105         call_rcu(&tgtdev->rcu, free_device);
2106 }
2107
2108 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2109                                      const char *device_path,
2110                                      struct btrfs_device **device)
2111 {
2112         int ret = 0;
2113         struct btrfs_super_block *disk_super;
2114         u64 devid;
2115         u8 *dev_uuid;
2116         struct block_device *bdev;
2117         struct buffer_head *bh;
2118
2119         *device = NULL;
2120         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2121                                     fs_info->bdev_holder, 0, &bdev, &bh);
2122         if (ret)
2123                 return ret;
2124         disk_super = (struct btrfs_super_block *)bh->b_data;
2125         devid = btrfs_stack_device_id(&disk_super->dev_item);
2126         dev_uuid = disk_super->dev_item.uuid;
2127         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2128         brelse(bh);
2129         if (!*device)
2130                 ret = -ENOENT;
2131         blkdev_put(bdev, FMODE_READ);
2132         return ret;
2133 }
2134
2135 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2136                                          const char *device_path,
2137                                          struct btrfs_device **device)
2138 {
2139         *device = NULL;
2140         if (strcmp(device_path, "missing") == 0) {
2141                 struct list_head *devices;
2142                 struct btrfs_device *tmp;
2143
2144                 devices = &fs_info->fs_devices->devices;
2145                 /*
2146                  * It is safe to read the devices since the volume_mutex
2147                  * is held by the caller.
2148                  */
2149                 list_for_each_entry(tmp, devices, dev_list) {
2150                         if (tmp->in_fs_metadata && !tmp->bdev) {
2151                                 *device = tmp;
2152                                 break;
2153                         }
2154                 }
2155
2156                 if (!*device)
2157                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2158
2159                 return 0;
2160         } else {
2161                 return btrfs_find_device_by_path(fs_info, device_path, device);
2162         }
2163 }
2164
2165 /*
2166  * Lookup a device given by device id, or the path if the id is 0.
2167  */
2168 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2169                                  const char *devpath,
2170                                  struct btrfs_device **device)
2171 {
2172         int ret;
2173
2174         if (devid) {
2175                 ret = 0;
2176                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2177                 if (!*device)
2178                         ret = -ENOENT;
2179         } else {
2180                 if (!devpath || !devpath[0])
2181                         return -EINVAL;
2182
2183                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2184                                                            device);
2185         }
2186         return ret;
2187 }
2188
2189 /*
2190  * does all the dirty work required for changing file system's UUID.
2191  */
2192 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2193 {
2194         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2195         struct btrfs_fs_devices *old_devices;
2196         struct btrfs_fs_devices *seed_devices;
2197         struct btrfs_super_block *disk_super = fs_info->super_copy;
2198         struct btrfs_device *device;
2199         u64 super_flags;
2200
2201         BUG_ON(!mutex_is_locked(&uuid_mutex));
2202         if (!fs_devices->seeding)
2203                 return -EINVAL;
2204
2205         seed_devices = __alloc_fs_devices();
2206         if (IS_ERR(seed_devices))
2207                 return PTR_ERR(seed_devices);
2208
2209         old_devices = clone_fs_devices(fs_devices);
2210         if (IS_ERR(old_devices)) {
2211                 kfree(seed_devices);
2212                 return PTR_ERR(old_devices);
2213         }
2214
2215         list_add(&old_devices->list, &fs_uuids);
2216
2217         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2218         seed_devices->opened = 1;
2219         INIT_LIST_HEAD(&seed_devices->devices);
2220         INIT_LIST_HEAD(&seed_devices->alloc_list);
2221         mutex_init(&seed_devices->device_list_mutex);
2222
2223         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2224         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2225                               synchronize_rcu);
2226         list_for_each_entry(device, &seed_devices->devices, dev_list)
2227                 device->fs_devices = seed_devices;
2228
2229         mutex_lock(&fs_info->chunk_mutex);
2230         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2231         mutex_unlock(&fs_info->chunk_mutex);
2232
2233         fs_devices->seeding = 0;
2234         fs_devices->num_devices = 0;
2235         fs_devices->open_devices = 0;
2236         fs_devices->missing_devices = 0;
2237         fs_devices->rotating = 0;
2238         fs_devices->seed = seed_devices;
2239
2240         generate_random_uuid(fs_devices->fsid);
2241         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2242         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2243         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2244
2245         super_flags = btrfs_super_flags(disk_super) &
2246                       ~BTRFS_SUPER_FLAG_SEEDING;
2247         btrfs_set_super_flags(disk_super, super_flags);
2248
2249         return 0;
2250 }
2251
2252 /*
2253  * Store the expected generation for seed devices in device items.
2254  */
2255 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2256                                struct btrfs_fs_info *fs_info)
2257 {
2258         struct btrfs_root *root = fs_info->chunk_root;
2259         struct btrfs_path *path;
2260         struct extent_buffer *leaf;
2261         struct btrfs_dev_item *dev_item;
2262         struct btrfs_device *device;
2263         struct btrfs_key key;
2264         u8 fs_uuid[BTRFS_UUID_SIZE];
2265         u8 dev_uuid[BTRFS_UUID_SIZE];
2266         u64 devid;
2267         int ret;
2268
2269         path = btrfs_alloc_path();
2270         if (!path)
2271                 return -ENOMEM;
2272
2273         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2274         key.offset = 0;
2275         key.type = BTRFS_DEV_ITEM_KEY;
2276
2277         while (1) {
2278                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2279                 if (ret < 0)
2280                         goto error;
2281
2282                 leaf = path->nodes[0];
2283 next_slot:
2284                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2285                         ret = btrfs_next_leaf(root, path);
2286                         if (ret > 0)
2287                                 break;
2288                         if (ret < 0)
2289                                 goto error;
2290                         leaf = path->nodes[0];
2291                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2292                         btrfs_release_path(path);
2293                         continue;
2294                 }
2295
2296                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2297                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2298                     key.type != BTRFS_DEV_ITEM_KEY)
2299                         break;
2300
2301                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2302                                           struct btrfs_dev_item);
2303                 devid = btrfs_device_id(leaf, dev_item);
2304                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2305                                    BTRFS_UUID_SIZE);
2306                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2307                                    BTRFS_UUID_SIZE);
2308                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2309                 BUG_ON(!device); /* Logic error */
2310
2311                 if (device->fs_devices->seeding) {
2312                         btrfs_set_device_generation(leaf, dev_item,
2313                                                     device->generation);
2314                         btrfs_mark_buffer_dirty(leaf);
2315                 }
2316
2317                 path->slots[0]++;
2318                 goto next_slot;
2319         }
2320         ret = 0;
2321 error:
2322         btrfs_free_path(path);
2323         return ret;
2324 }
2325
2326 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2327 {
2328         struct btrfs_root *root = fs_info->dev_root;
2329         struct request_queue *q;
2330         struct btrfs_trans_handle *trans;
2331         struct btrfs_device *device;
2332         struct block_device *bdev;
2333         struct list_head *devices;
2334         struct super_block *sb = fs_info->sb;
2335         struct rcu_string *name;
2336         u64 tmp;
2337         int seeding_dev = 0;
2338         int ret = 0;
2339
2340         if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2341                 return -EROFS;
2342
2343         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2344                                   fs_info->bdev_holder);
2345         if (IS_ERR(bdev))
2346                 return PTR_ERR(bdev);
2347
2348         if (fs_info->fs_devices->seeding) {
2349                 seeding_dev = 1;
2350                 down_write(&sb->s_umount);
2351                 mutex_lock(&uuid_mutex);
2352         }
2353
2354         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2355
2356         devices = &fs_info->fs_devices->devices;
2357
2358         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2359         list_for_each_entry(device, devices, dev_list) {
2360                 if (device->bdev == bdev) {
2361                         ret = -EEXIST;
2362                         mutex_unlock(
2363                                 &fs_info->fs_devices->device_list_mutex);
2364                         goto error;
2365                 }
2366         }
2367         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2368
2369         device = btrfs_alloc_device(fs_info, NULL, NULL);
2370         if (IS_ERR(device)) {
2371                 /* we can safely leave the fs_devices entry around */
2372                 ret = PTR_ERR(device);
2373                 goto error;
2374         }
2375
2376         name = rcu_string_strdup(device_path, GFP_KERNEL);
2377         if (!name) {
2378                 kfree(device);
2379                 ret = -ENOMEM;
2380                 goto error;
2381         }
2382         rcu_assign_pointer(device->name, name);
2383
2384         trans = btrfs_start_transaction(root, 0);
2385         if (IS_ERR(trans)) {
2386                 rcu_string_free(device->name);
2387                 kfree(device);
2388                 ret = PTR_ERR(trans);
2389                 goto error;
2390         }
2391
2392         q = bdev_get_queue(bdev);
2393         if (blk_queue_discard(q))
2394                 device->can_discard = 1;
2395         device->writeable = 1;
2396         device->generation = trans->transid;
2397         device->io_width = fs_info->sectorsize;
2398         device->io_align = fs_info->sectorsize;
2399         device->sector_size = fs_info->sectorsize;
2400         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2401                                          fs_info->sectorsize);
2402         device->disk_total_bytes = device->total_bytes;
2403         device->commit_total_bytes = device->total_bytes;
2404         device->fs_info = fs_info;
2405         device->bdev = bdev;
2406         device->in_fs_metadata = 1;
2407         device->is_tgtdev_for_dev_replace = 0;
2408         device->mode = FMODE_EXCL;
2409         device->dev_stats_valid = 1;
2410         set_blocksize(device->bdev, 4096);
2411
2412         if (seeding_dev) {
2413                 sb->s_flags &= ~MS_RDONLY;
2414                 ret = btrfs_prepare_sprout(fs_info);
2415                 BUG_ON(ret); /* -ENOMEM */
2416         }
2417
2418         device->fs_devices = fs_info->fs_devices;
2419
2420         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2421         mutex_lock(&fs_info->chunk_mutex);
2422         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2423         list_add(&device->dev_alloc_list,
2424                  &fs_info->fs_devices->alloc_list);
2425         fs_info->fs_devices->num_devices++;
2426         fs_info->fs_devices->open_devices++;
2427         fs_info->fs_devices->rw_devices++;
2428         fs_info->fs_devices->total_devices++;
2429         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2430
2431         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2432
2433         if (!blk_queue_nonrot(q))
2434                 fs_info->fs_devices->rotating = 1;
2435
2436         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2437         btrfs_set_super_total_bytes(fs_info->super_copy,
2438                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2439
2440         tmp = btrfs_super_num_devices(fs_info->super_copy);
2441         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2442
2443         /* add sysfs device entry */
2444         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2445
2446         /*
2447          * we've got more storage, clear any full flags on the space
2448          * infos
2449          */
2450         btrfs_clear_space_info_full(fs_info);
2451
2452         mutex_unlock(&fs_info->chunk_mutex);
2453         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2454
2455         if (seeding_dev) {
2456                 mutex_lock(&fs_info->chunk_mutex);
2457                 ret = init_first_rw_device(trans, fs_info);
2458                 mutex_unlock(&fs_info->chunk_mutex);
2459                 if (ret) {
2460                         btrfs_abort_transaction(trans, ret);
2461                         goto error_trans;
2462                 }
2463         }
2464
2465         ret = btrfs_add_device(trans, fs_info, device);
2466         if (ret) {
2467                 btrfs_abort_transaction(trans, ret);
2468                 goto error_trans;
2469         }
2470
2471         if (seeding_dev) {
2472                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2473
2474                 ret = btrfs_finish_sprout(trans, fs_info);
2475                 if (ret) {
2476                         btrfs_abort_transaction(trans, ret);
2477                         goto error_trans;
2478                 }
2479
2480                 /* Sprouting would change fsid of the mounted root,
2481                  * so rename the fsid on the sysfs
2482                  */
2483                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2484                                                 fs_info->fsid);
2485                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2486                         btrfs_warn(fs_info,
2487                                    "sysfs: failed to create fsid for sprout");
2488         }
2489
2490         fs_info->num_tolerated_disk_barrier_failures =
2491                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2492         ret = btrfs_commit_transaction(trans);
2493
2494         if (seeding_dev) {
2495                 mutex_unlock(&uuid_mutex);
2496                 up_write(&sb->s_umount);
2497
2498                 if (ret) /* transaction commit */
2499                         return ret;
2500
2501                 ret = btrfs_relocate_sys_chunks(fs_info);
2502                 if (ret < 0)
2503                         btrfs_handle_fs_error(fs_info, ret,
2504                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2505                 trans = btrfs_attach_transaction(root);
2506                 if (IS_ERR(trans)) {
2507                         if (PTR_ERR(trans) == -ENOENT)
2508                                 return 0;
2509                         return PTR_ERR(trans);
2510                 }
2511                 ret = btrfs_commit_transaction(trans);
2512         }
2513
2514         /* Update ctime/mtime for libblkid */
2515         update_dev_time(device_path);
2516         return ret;
2517
2518 error_trans:
2519         btrfs_end_transaction(trans);
2520         rcu_string_free(device->name);
2521         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2522         kfree(device);
2523 error:
2524         blkdev_put(bdev, FMODE_EXCL);
2525         if (seeding_dev) {
2526                 mutex_unlock(&uuid_mutex);
2527                 up_write(&sb->s_umount);
2528         }
2529         return ret;
2530 }
2531
2532 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2533                                   const char *device_path,
2534                                   struct btrfs_device *srcdev,
2535                                   struct btrfs_device **device_out)
2536 {
2537         struct request_queue *q;
2538         struct btrfs_device *device;
2539         struct block_device *bdev;
2540         struct list_head *devices;
2541         struct rcu_string *name;
2542         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2543         int ret = 0;
2544
2545         *device_out = NULL;
2546         if (fs_info->fs_devices->seeding) {
2547                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2548                 return -EINVAL;
2549         }
2550
2551         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2552                                   fs_info->bdev_holder);
2553         if (IS_ERR(bdev)) {
2554                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2555                 return PTR_ERR(bdev);
2556         }
2557
2558         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2559
2560         devices = &fs_info->fs_devices->devices;
2561         list_for_each_entry(device, devices, dev_list) {
2562                 if (device->bdev == bdev) {
2563                         btrfs_err(fs_info,
2564                                   "target device is in the filesystem!");
2565                         ret = -EEXIST;
2566                         goto error;
2567                 }
2568         }
2569
2570
2571         if (i_size_read(bdev->bd_inode) <
2572             btrfs_device_get_total_bytes(srcdev)) {
2573                 btrfs_err(fs_info,
2574                           "target device is smaller than source device!");
2575                 ret = -EINVAL;
2576                 goto error;
2577         }
2578
2579
2580         device = btrfs_alloc_device(NULL, &devid, NULL);
2581         if (IS_ERR(device)) {
2582                 ret = PTR_ERR(device);
2583                 goto error;
2584         }
2585
2586         name = rcu_string_strdup(device_path, GFP_KERNEL);
2587         if (!name) {
2588                 kfree(device);
2589                 ret = -ENOMEM;
2590                 goto error;
2591         }
2592         rcu_assign_pointer(device->name, name);
2593
2594         q = bdev_get_queue(bdev);
2595         if (blk_queue_discard(q))
2596                 device->can_discard = 1;
2597         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2598         device->writeable = 1;
2599         device->generation = 0;
2600         device->io_width = fs_info->sectorsize;
2601         device->io_align = fs_info->sectorsize;
2602         device->sector_size = fs_info->sectorsize;
2603         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2604         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2605         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2606         ASSERT(list_empty(&srcdev->resized_list));
2607         device->commit_total_bytes = srcdev->commit_total_bytes;
2608         device->commit_bytes_used = device->bytes_used;
2609         device->fs_info = fs_info;
2610         device->bdev = bdev;
2611         device->in_fs_metadata = 1;
2612         device->is_tgtdev_for_dev_replace = 1;
2613         device->mode = FMODE_EXCL;
2614         device->dev_stats_valid = 1;
2615         set_blocksize(device->bdev, 4096);
2616         device->fs_devices = fs_info->fs_devices;
2617         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2618         fs_info->fs_devices->num_devices++;
2619         fs_info->fs_devices->open_devices++;
2620         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2621
2622         *device_out = device;
2623         return ret;
2624
2625 error:
2626         blkdev_put(bdev, FMODE_EXCL);
2627         return ret;
2628 }
2629
2630 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2631                                               struct btrfs_device *tgtdev)
2632 {
2633         u32 sectorsize = fs_info->sectorsize;
2634
2635         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2636         tgtdev->io_width = sectorsize;
2637         tgtdev->io_align = sectorsize;
2638         tgtdev->sector_size = sectorsize;
2639         tgtdev->fs_info = fs_info;
2640         tgtdev->in_fs_metadata = 1;
2641 }
2642
2643 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2644                                         struct btrfs_device *device)
2645 {
2646         int ret;
2647         struct btrfs_path *path;
2648         struct btrfs_root *root = device->fs_info->chunk_root;
2649         struct btrfs_dev_item *dev_item;
2650         struct extent_buffer *leaf;
2651         struct btrfs_key key;
2652
2653         path = btrfs_alloc_path();
2654         if (!path)
2655                 return -ENOMEM;
2656
2657         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2658         key.type = BTRFS_DEV_ITEM_KEY;
2659         key.offset = device->devid;
2660
2661         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2662         if (ret < 0)
2663                 goto out;
2664
2665         if (ret > 0) {
2666                 ret = -ENOENT;
2667                 goto out;
2668         }
2669
2670         leaf = path->nodes[0];
2671         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2672
2673         btrfs_set_device_id(leaf, dev_item, device->devid);
2674         btrfs_set_device_type(leaf, dev_item, device->type);
2675         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2676         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2677         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2678         btrfs_set_device_total_bytes(leaf, dev_item,
2679                                      btrfs_device_get_disk_total_bytes(device));
2680         btrfs_set_device_bytes_used(leaf, dev_item,
2681                                     btrfs_device_get_bytes_used(device));
2682         btrfs_mark_buffer_dirty(leaf);
2683
2684 out:
2685         btrfs_free_path(path);
2686         return ret;
2687 }
2688
2689 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2690                       struct btrfs_device *device, u64 new_size)
2691 {
2692         struct btrfs_fs_info *fs_info = device->fs_info;
2693         struct btrfs_super_block *super_copy = fs_info->super_copy;
2694         struct btrfs_fs_devices *fs_devices;
2695         u64 old_total;
2696         u64 diff;
2697
2698         if (!device->writeable)
2699                 return -EACCES;
2700
2701         new_size = round_down(new_size, fs_info->sectorsize);
2702
2703         mutex_lock(&fs_info->chunk_mutex);
2704         old_total = btrfs_super_total_bytes(super_copy);
2705         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2706
2707         if (new_size <= device->total_bytes ||
2708             device->is_tgtdev_for_dev_replace) {
2709                 mutex_unlock(&fs_info->chunk_mutex);
2710                 return -EINVAL;
2711         }
2712
2713         fs_devices = fs_info->fs_devices;
2714
2715         btrfs_set_super_total_bytes(super_copy,
2716                         round_down(old_total + diff, fs_info->sectorsize));
2717         device->fs_devices->total_rw_bytes += diff;
2718
2719         btrfs_device_set_total_bytes(device, new_size);
2720         btrfs_device_set_disk_total_bytes(device, new_size);
2721         btrfs_clear_space_info_full(device->fs_info);
2722         if (list_empty(&device->resized_list))
2723                 list_add_tail(&device->resized_list,
2724                               &fs_devices->resized_devices);
2725         mutex_unlock(&fs_info->chunk_mutex);
2726
2727         return btrfs_update_device(trans, device);
2728 }
2729
2730 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2731                             struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2732                             u64 chunk_offset)
2733 {
2734         struct btrfs_root *root = fs_info->chunk_root;
2735         int ret;
2736         struct btrfs_path *path;
2737         struct btrfs_key key;
2738
2739         path = btrfs_alloc_path();
2740         if (!path)
2741                 return -ENOMEM;
2742
2743         key.objectid = chunk_objectid;
2744         key.offset = chunk_offset;
2745         key.type = BTRFS_CHUNK_ITEM_KEY;
2746
2747         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2748         if (ret < 0)
2749                 goto out;
2750         else if (ret > 0) { /* Logic error or corruption */
2751                 btrfs_handle_fs_error(fs_info, -ENOENT,
2752                                       "Failed lookup while freeing chunk.");
2753                 ret = -ENOENT;
2754                 goto out;
2755         }
2756
2757         ret = btrfs_del_item(trans, root, path);
2758         if (ret < 0)
2759                 btrfs_handle_fs_error(fs_info, ret,
2760                                       "Failed to delete chunk item.");
2761 out:
2762         btrfs_free_path(path);
2763         return ret;
2764 }
2765
2766 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2767                                u64 chunk_objectid, u64 chunk_offset)
2768 {
2769         struct btrfs_super_block *super_copy = fs_info->super_copy;
2770         struct btrfs_disk_key *disk_key;
2771         struct btrfs_chunk *chunk;
2772         u8 *ptr;
2773         int ret = 0;
2774         u32 num_stripes;
2775         u32 array_size;
2776         u32 len = 0;
2777         u32 cur;
2778         struct btrfs_key key;
2779
2780         mutex_lock(&fs_info->chunk_mutex);
2781         array_size = btrfs_super_sys_array_size(super_copy);
2782
2783         ptr = super_copy->sys_chunk_array;
2784         cur = 0;
2785
2786         while (cur < array_size) {
2787                 disk_key = (struct btrfs_disk_key *)ptr;
2788                 btrfs_disk_key_to_cpu(&key, disk_key);
2789
2790                 len = sizeof(*disk_key);
2791
2792                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2793                         chunk = (struct btrfs_chunk *)(ptr + len);
2794                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2795                         len += btrfs_chunk_item_size(num_stripes);
2796                 } else {
2797                         ret = -EIO;
2798                         break;
2799                 }
2800                 if (key.objectid == chunk_objectid &&
2801                     key.offset == chunk_offset) {
2802                         memmove(ptr, ptr + len, array_size - (cur + len));
2803                         array_size -= len;
2804                         btrfs_set_super_sys_array_size(super_copy, array_size);
2805                 } else {
2806                         ptr += len;
2807                         cur += len;
2808                 }
2809         }
2810         mutex_unlock(&fs_info->chunk_mutex);
2811         return ret;
2812 }
2813
2814 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2815                                         u64 logical, u64 length)
2816 {
2817         struct extent_map_tree *em_tree;
2818         struct extent_map *em;
2819
2820         em_tree = &fs_info->mapping_tree.map_tree;
2821         read_lock(&em_tree->lock);
2822         em = lookup_extent_mapping(em_tree, logical, length);
2823         read_unlock(&em_tree->lock);
2824
2825         if (!em) {
2826                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2827                            logical, length);
2828                 return ERR_PTR(-EINVAL);
2829         }
2830
2831         if (em->start > logical || em->start + em->len < logical) {
2832                 btrfs_crit(fs_info,
2833                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2834                            logical, length, em->start, em->start + em->len);
2835                 free_extent_map(em);
2836                 return ERR_PTR(-EINVAL);
2837         }
2838
2839         /* callers are responsible for dropping em's ref. */
2840         return em;
2841 }
2842
2843 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2844                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2845 {
2846         struct extent_map *em;
2847         struct map_lookup *map;
2848         u64 dev_extent_len = 0;
2849         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2850         int i, ret = 0;
2851         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2852
2853         em = get_chunk_map(fs_info, chunk_offset, 1);
2854         if (IS_ERR(em)) {
2855                 /*
2856                  * This is a logic error, but we don't want to just rely on the
2857                  * user having built with ASSERT enabled, so if ASSERT doesn't
2858                  * do anything we still error out.
2859                  */
2860                 ASSERT(0);
2861                 return PTR_ERR(em);
2862         }
2863         map = em->map_lookup;
2864         mutex_lock(&fs_info->chunk_mutex);
2865         check_system_chunk(trans, fs_info, map->type);
2866         mutex_unlock(&fs_info->chunk_mutex);
2867
2868         /*
2869          * Take the device list mutex to prevent races with the final phase of
2870          * a device replace operation that replaces the device object associated
2871          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2872          */
2873         mutex_lock(&fs_devices->device_list_mutex);
2874         for (i = 0; i < map->num_stripes; i++) {
2875                 struct btrfs_device *device = map->stripes[i].dev;
2876                 ret = btrfs_free_dev_extent(trans, device,
2877                                             map->stripes[i].physical,
2878                                             &dev_extent_len);
2879                 if (ret) {
2880                         mutex_unlock(&fs_devices->device_list_mutex);
2881                         btrfs_abort_transaction(trans, ret);
2882                         goto out;
2883                 }
2884
2885                 if (device->bytes_used > 0) {
2886                         mutex_lock(&fs_info->chunk_mutex);
2887                         btrfs_device_set_bytes_used(device,
2888                                         device->bytes_used - dev_extent_len);
2889                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2890                         btrfs_clear_space_info_full(fs_info);
2891                         mutex_unlock(&fs_info->chunk_mutex);
2892                 }
2893
2894                 if (map->stripes[i].dev) {
2895                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2896                         if (ret) {
2897                                 mutex_unlock(&fs_devices->device_list_mutex);
2898                                 btrfs_abort_transaction(trans, ret);
2899                                 goto out;
2900                         }
2901                 }
2902         }
2903         mutex_unlock(&fs_devices->device_list_mutex);
2904
2905         ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2906         if (ret) {
2907                 btrfs_abort_transaction(trans, ret);
2908                 goto out;
2909         }
2910
2911         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2912
2913         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2914                 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2915                                           chunk_offset);
2916                 if (ret) {
2917                         btrfs_abort_transaction(trans, ret);
2918                         goto out;
2919                 }
2920         }
2921
2922         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2923         if (ret) {
2924                 btrfs_abort_transaction(trans, ret);
2925                 goto out;
2926         }
2927
2928 out:
2929         /* once for us */
2930         free_extent_map(em);
2931         return ret;
2932 }
2933
2934 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2935 {
2936         struct btrfs_root *root = fs_info->chunk_root;
2937         struct btrfs_trans_handle *trans;
2938         int ret;
2939
2940         /*
2941          * Prevent races with automatic removal of unused block groups.
2942          * After we relocate and before we remove the chunk with offset
2943          * chunk_offset, automatic removal of the block group can kick in,
2944          * resulting in a failure when calling btrfs_remove_chunk() below.
2945          *
2946          * Make sure to acquire this mutex before doing a tree search (dev
2947          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2948          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2949          * we release the path used to search the chunk/dev tree and before
2950          * the current task acquires this mutex and calls us.
2951          */
2952         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2953
2954         ret = btrfs_can_relocate(fs_info, chunk_offset);
2955         if (ret)
2956                 return -ENOSPC;
2957
2958         /* step one, relocate all the extents inside this chunk */
2959         btrfs_scrub_pause(fs_info);
2960         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2961         btrfs_scrub_continue(fs_info);
2962         if (ret)
2963                 return ret;
2964
2965         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2966                                                      chunk_offset);
2967         if (IS_ERR(trans)) {
2968                 ret = PTR_ERR(trans);
2969                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2970                 return ret;
2971         }
2972
2973         /*
2974          * step two, delete the device extents and the
2975          * chunk tree entries
2976          */
2977         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2978         btrfs_end_transaction(trans);
2979         return ret;
2980 }
2981
2982 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2983 {
2984         struct btrfs_root *chunk_root = fs_info->chunk_root;
2985         struct btrfs_path *path;
2986         struct extent_buffer *leaf;
2987         struct btrfs_chunk *chunk;
2988         struct btrfs_key key;
2989         struct btrfs_key found_key;
2990         u64 chunk_type;
2991         bool retried = false;
2992         int failed = 0;
2993         int ret;
2994
2995         path = btrfs_alloc_path();
2996         if (!path)
2997                 return -ENOMEM;
2998
2999 again:
3000         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3001         key.offset = (u64)-1;
3002         key.type = BTRFS_CHUNK_ITEM_KEY;
3003
3004         while (1) {
3005                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3006                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3007                 if (ret < 0) {
3008                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3009                         goto error;
3010                 }
3011                 BUG_ON(ret == 0); /* Corruption */
3012
3013                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3014                                           key.type);
3015                 if (ret)
3016                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3017                 if (ret < 0)
3018                         goto error;
3019                 if (ret > 0)
3020                         break;
3021
3022                 leaf = path->nodes[0];
3023                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3024
3025                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3026                                        struct btrfs_chunk);
3027                 chunk_type = btrfs_chunk_type(leaf, chunk);
3028                 btrfs_release_path(path);
3029
3030                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3031                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3032                         if (ret == -ENOSPC)
3033                                 failed++;
3034                         else
3035                                 BUG_ON(ret);
3036                 }
3037                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3038
3039                 if (found_key.offset == 0)
3040                         break;
3041                 key.offset = found_key.offset - 1;
3042         }
3043         ret = 0;
3044         if (failed && !retried) {
3045                 failed = 0;
3046                 retried = true;
3047                 goto again;
3048         } else if (WARN_ON(failed && retried)) {
3049                 ret = -ENOSPC;
3050         }
3051 error:
3052         btrfs_free_path(path);
3053         return ret;
3054 }
3055
3056 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3057                                struct btrfs_balance_control *bctl)
3058 {
3059         struct btrfs_root *root = fs_info->tree_root;
3060         struct btrfs_trans_handle *trans;
3061         struct btrfs_balance_item *item;
3062         struct btrfs_disk_balance_args disk_bargs;
3063         struct btrfs_path *path;
3064         struct extent_buffer *leaf;
3065         struct btrfs_key key;
3066         int ret, err;
3067
3068         path = btrfs_alloc_path();
3069         if (!path)
3070                 return -ENOMEM;
3071
3072         trans = btrfs_start_transaction(root, 0);
3073         if (IS_ERR(trans)) {
3074                 btrfs_free_path(path);
3075                 return PTR_ERR(trans);
3076         }
3077
3078         key.objectid = BTRFS_BALANCE_OBJECTID;
3079         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3080         key.offset = 0;
3081
3082         ret = btrfs_insert_empty_item(trans, root, path, &key,
3083                                       sizeof(*item));
3084         if (ret)
3085                 goto out;
3086
3087         leaf = path->nodes[0];
3088         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3089
3090         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3091
3092         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3093         btrfs_set_balance_data(leaf, item, &disk_bargs);
3094         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3095         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3096         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3097         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3098
3099         btrfs_set_balance_flags(leaf, item, bctl->flags);
3100
3101         btrfs_mark_buffer_dirty(leaf);
3102 out:
3103         btrfs_free_path(path);
3104         err = btrfs_commit_transaction(trans);
3105         if (err && !ret)
3106                 ret = err;
3107         return ret;
3108 }
3109
3110 static int del_balance_item(struct btrfs_fs_info *fs_info)
3111 {
3112         struct btrfs_root *root = fs_info->tree_root;
3113         struct btrfs_trans_handle *trans;
3114         struct btrfs_path *path;
3115         struct btrfs_key key;
3116         int ret, err;
3117
3118         path = btrfs_alloc_path();
3119         if (!path)
3120                 return -ENOMEM;
3121
3122         trans = btrfs_start_transaction(root, 0);
3123         if (IS_ERR(trans)) {
3124                 btrfs_free_path(path);
3125                 return PTR_ERR(trans);
3126         }
3127
3128         key.objectid = BTRFS_BALANCE_OBJECTID;
3129         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3130         key.offset = 0;
3131
3132         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3133         if (ret < 0)
3134                 goto out;
3135         if (ret > 0) {
3136                 ret = -ENOENT;
3137                 goto out;
3138         }
3139
3140         ret = btrfs_del_item(trans, root, path);
3141 out:
3142         btrfs_free_path(path);
3143         err = btrfs_commit_transaction(trans);
3144         if (err && !ret)
3145                 ret = err;
3146         return ret;
3147 }
3148
3149 /*
3150  * This is a heuristic used to reduce the number of chunks balanced on
3151  * resume after balance was interrupted.
3152  */
3153 static void update_balance_args(struct btrfs_balance_control *bctl)
3154 {
3155         /*
3156          * Turn on soft mode for chunk types that were being converted.
3157          */
3158         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3159                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3160         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3161                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3162         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3163                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3164
3165         /*
3166          * Turn on usage filter if is not already used.  The idea is
3167          * that chunks that we have already balanced should be
3168          * reasonably full.  Don't do it for chunks that are being
3169          * converted - that will keep us from relocating unconverted
3170          * (albeit full) chunks.
3171          */
3172         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3173             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3174             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3175                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3176                 bctl->data.usage = 90;
3177         }
3178         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3179             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3180             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3181                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3182                 bctl->sys.usage = 90;
3183         }
3184         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3185             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3186             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3187                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3188                 bctl->meta.usage = 90;
3189         }
3190 }
3191
3192 /*
3193  * Should be called with both balance and volume mutexes held to
3194  * serialize other volume operations (add_dev/rm_dev/resize) with
3195  * restriper.  Same goes for unset_balance_control.
3196  */
3197 static void set_balance_control(struct btrfs_balance_control *bctl)
3198 {
3199         struct btrfs_fs_info *fs_info = bctl->fs_info;
3200
3201         BUG_ON(fs_info->balance_ctl);
3202
3203         spin_lock(&fs_info->balance_lock);
3204         fs_info->balance_ctl = bctl;
3205         spin_unlock(&fs_info->balance_lock);
3206 }
3207
3208 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3209 {
3210         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3211
3212         BUG_ON(!fs_info->balance_ctl);
3213
3214         spin_lock(&fs_info->balance_lock);
3215         fs_info->balance_ctl = NULL;
3216         spin_unlock(&fs_info->balance_lock);
3217
3218         kfree(bctl);
3219 }
3220
3221 /*
3222  * Balance filters.  Return 1 if chunk should be filtered out
3223  * (should not be balanced).
3224  */
3225 static int chunk_profiles_filter(u64 chunk_type,
3226                                  struct btrfs_balance_args *bargs)
3227 {
3228         chunk_type = chunk_to_extended(chunk_type) &
3229                                 BTRFS_EXTENDED_PROFILE_MASK;
3230
3231         if (bargs->profiles & chunk_type)
3232                 return 0;
3233
3234         return 1;
3235 }
3236
3237 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3238                               struct btrfs_balance_args *bargs)
3239 {
3240         struct btrfs_block_group_cache *cache;
3241         u64 chunk_u