Merge tag 'microblaze-4.15-rc2' of git://git.monstr.eu/linux-2.6-microblaze
[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 /*
156  * alloc_fs_devices - allocate struct btrfs_fs_devices
157  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
158  *
159  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160  * The returned struct is not linked onto any lists and can be destroyed with
161  * kfree() right away.
162  */
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
164 {
165         struct btrfs_fs_devices *fs_devs;
166
167         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
168         if (!fs_devs)
169                 return ERR_PTR(-ENOMEM);
170
171         mutex_init(&fs_devs->device_list_mutex);
172
173         INIT_LIST_HEAD(&fs_devs->devices);
174         INIT_LIST_HEAD(&fs_devs->resized_devices);
175         INIT_LIST_HEAD(&fs_devs->alloc_list);
176         INIT_LIST_HEAD(&fs_devs->list);
177         if (fsid)
178                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
179
180         return fs_devs;
181 }
182
183 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
184 {
185         struct btrfs_device *device;
186         WARN_ON(fs_devices->opened);
187         while (!list_empty(&fs_devices->devices)) {
188                 device = list_entry(fs_devices->devices.next,
189                                     struct btrfs_device, dev_list);
190                 list_del(&device->dev_list);
191                 rcu_string_free(device->name);
192                 kfree(device);
193         }
194         kfree(fs_devices);
195 }
196
197 static void btrfs_kobject_uevent(struct block_device *bdev,
198                                  enum kobject_action action)
199 {
200         int ret;
201
202         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
203         if (ret)
204                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
205                         action,
206                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
207                         &disk_to_dev(bdev->bd_disk)->kobj);
208 }
209
210 void btrfs_cleanup_fs_uuids(void)
211 {
212         struct btrfs_fs_devices *fs_devices;
213
214         while (!list_empty(&fs_uuids)) {
215                 fs_devices = list_entry(fs_uuids.next,
216                                         struct btrfs_fs_devices, list);
217                 list_del(&fs_devices->list);
218                 free_fs_devices(fs_devices);
219         }
220 }
221
222 static struct btrfs_device *__alloc_device(void)
223 {
224         struct btrfs_device *dev;
225
226         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
227         if (!dev)
228                 return ERR_PTR(-ENOMEM);
229
230         /*
231          * Preallocate a bio that's always going to be used for flushing device
232          * barriers and matches the device lifespan
233          */
234         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
235         if (!dev->flush_bio) {
236                 kfree(dev);
237                 return ERR_PTR(-ENOMEM);
238         }
239         bio_get(dev->flush_bio);
240
241         INIT_LIST_HEAD(&dev->dev_list);
242         INIT_LIST_HEAD(&dev->dev_alloc_list);
243         INIT_LIST_HEAD(&dev->resized_list);
244
245         spin_lock_init(&dev->io_lock);
246
247         spin_lock_init(&dev->reada_lock);
248         atomic_set(&dev->reada_in_flight, 0);
249         atomic_set(&dev->dev_stats_ccnt, 0);
250         btrfs_device_data_ordered_init(dev);
251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253
254         return dev;
255 }
256
257 /*
258  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
259  * return NULL.
260  *
261  * If devid and uuid are both specified, the match must be exact, otherwise
262  * only devid is used.
263  */
264 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
265                 u64 devid, const u8 *uuid)
266 {
267         struct list_head *head = &fs_devices->devices;
268         struct btrfs_device *dev;
269
270         list_for_each_entry(dev, head, dev_list) {
271                 if (dev->devid == devid &&
272                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
273                         return dev;
274                 }
275         }
276         return NULL;
277 }
278
279 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
280 {
281         struct btrfs_fs_devices *fs_devices;
282
283         list_for_each_entry(fs_devices, &fs_uuids, list) {
284                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
285                         return fs_devices;
286         }
287         return NULL;
288 }
289
290 static int
291 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
292                       int flush, struct block_device **bdev,
293                       struct buffer_head **bh)
294 {
295         int ret;
296
297         *bdev = blkdev_get_by_path(device_path, flags, holder);
298
299         if (IS_ERR(*bdev)) {
300                 ret = PTR_ERR(*bdev);
301                 goto error;
302         }
303
304         if (flush)
305                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
306         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
307         if (ret) {
308                 blkdev_put(*bdev, flags);
309                 goto error;
310         }
311         invalidate_bdev(*bdev);
312         *bh = btrfs_read_dev_super(*bdev);
313         if (IS_ERR(*bh)) {
314                 ret = PTR_ERR(*bh);
315                 blkdev_put(*bdev, flags);
316                 goto error;
317         }
318
319         return 0;
320
321 error:
322         *bdev = NULL;
323         *bh = NULL;
324         return ret;
325 }
326
327 static void requeue_list(struct btrfs_pending_bios *pending_bios,
328                         struct bio *head, struct bio *tail)
329 {
330
331         struct bio *old_head;
332
333         old_head = pending_bios->head;
334         pending_bios->head = head;
335         if (pending_bios->tail)
336                 tail->bi_next = old_head;
337         else
338                 pending_bios->tail = tail;
339 }
340
341 /*
342  * we try to collect pending bios for a device so we don't get a large
343  * number of procs sending bios down to the same device.  This greatly
344  * improves the schedulers ability to collect and merge the bios.
345  *
346  * But, it also turns into a long list of bios to process and that is sure
347  * to eventually make the worker thread block.  The solution here is to
348  * make some progress and then put this work struct back at the end of
349  * the list if the block device is congested.  This way, multiple devices
350  * can make progress from a single worker thread.
351  */
352 static noinline void run_scheduled_bios(struct btrfs_device *device)
353 {
354         struct btrfs_fs_info *fs_info = device->fs_info;
355         struct bio *pending;
356         struct backing_dev_info *bdi;
357         struct btrfs_pending_bios *pending_bios;
358         struct bio *tail;
359         struct bio *cur;
360         int again = 0;
361         unsigned long num_run;
362         unsigned long batch_run = 0;
363         unsigned long last_waited = 0;
364         int force_reg = 0;
365         int sync_pending = 0;
366         struct blk_plug plug;
367
368         /*
369          * this function runs all the bios we've collected for
370          * a particular device.  We don't want to wander off to
371          * another device without first sending all of these down.
372          * So, setup a plug here and finish it off before we return
373          */
374         blk_start_plug(&plug);
375
376         bdi = device->bdev->bd_bdi;
377
378 loop:
379         spin_lock(&device->io_lock);
380
381 loop_lock:
382         num_run = 0;
383
384         /* take all the bios off the list at once and process them
385          * later on (without the lock held).  But, remember the
386          * tail and other pointers so the bios can be properly reinserted
387          * into the list if we hit congestion
388          */
389         if (!force_reg && device->pending_sync_bios.head) {
390                 pending_bios = &device->pending_sync_bios;
391                 force_reg = 1;
392         } else {
393                 pending_bios = &device->pending_bios;
394                 force_reg = 0;
395         }
396
397         pending = pending_bios->head;
398         tail = pending_bios->tail;
399         WARN_ON(pending && !tail);
400
401         /*
402          * if pending was null this time around, no bios need processing
403          * at all and we can stop.  Otherwise it'll loop back up again
404          * and do an additional check so no bios are missed.
405          *
406          * device->running_pending is used to synchronize with the
407          * schedule_bio code.
408          */
409         if (device->pending_sync_bios.head == NULL &&
410             device->pending_bios.head == NULL) {
411                 again = 0;
412                 device->running_pending = 0;
413         } else {
414                 again = 1;
415                 device->running_pending = 1;
416         }
417
418         pending_bios->head = NULL;
419         pending_bios->tail = NULL;
420
421         spin_unlock(&device->io_lock);
422
423         while (pending) {
424
425                 rmb();
426                 /* we want to work on both lists, but do more bios on the
427                  * sync list than the regular list
428                  */
429                 if ((num_run > 32 &&
430                     pending_bios != &device->pending_sync_bios &&
431                     device->pending_sync_bios.head) ||
432                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
433                     device->pending_bios.head)) {
434                         spin_lock(&device->io_lock);
435                         requeue_list(pending_bios, pending, tail);
436                         goto loop_lock;
437                 }
438
439                 cur = pending;
440                 pending = pending->bi_next;
441                 cur->bi_next = NULL;
442
443                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
444
445                 /*
446                  * if we're doing the sync list, record that our
447                  * plug has some sync requests on it
448                  *
449                  * If we're doing the regular list and there are
450                  * sync requests sitting around, unplug before
451                  * we add more
452                  */
453                 if (pending_bios == &device->pending_sync_bios) {
454                         sync_pending = 1;
455                 } else if (sync_pending) {
456                         blk_finish_plug(&plug);
457                         blk_start_plug(&plug);
458                         sync_pending = 0;
459                 }
460
461                 btrfsic_submit_bio(cur);
462                 num_run++;
463                 batch_run++;
464
465                 cond_resched();
466
467                 /*
468                  * we made progress, there is more work to do and the bdi
469                  * is now congested.  Back off and let other work structs
470                  * run instead
471                  */
472                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
473                     fs_info->fs_devices->open_devices > 1) {
474                         struct io_context *ioc;
475
476                         ioc = current->io_context;
477
478                         /*
479                          * the main goal here is that we don't want to
480                          * block if we're going to be able to submit
481                          * more requests without blocking.
482                          *
483                          * This code does two great things, it pokes into
484                          * the elevator code from a filesystem _and_
485                          * it makes assumptions about how batching works.
486                          */
487                         if (ioc && ioc->nr_batch_requests > 0 &&
488                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
489                             (last_waited == 0 ||
490                              ioc->last_waited == last_waited)) {
491                                 /*
492                                  * we want to go through our batch of
493                                  * requests and stop.  So, we copy out
494                                  * the ioc->last_waited time and test
495                                  * against it before looping
496                                  */
497                                 last_waited = ioc->last_waited;
498                                 cond_resched();
499                                 continue;
500                         }
501                         spin_lock(&device->io_lock);
502                         requeue_list(pending_bios, pending, tail);
503                         device->running_pending = 1;
504
505                         spin_unlock(&device->io_lock);
506                         btrfs_queue_work(fs_info->submit_workers,
507                                          &device->work);
508                         goto done;
509                 }
510         }
511
512         cond_resched();
513         if (again)
514                 goto loop;
515
516         spin_lock(&device->io_lock);
517         if (device->pending_bios.head || device->pending_sync_bios.head)
518                 goto loop_lock;
519         spin_unlock(&device->io_lock);
520
521 done:
522         blk_finish_plug(&plug);
523 }
524
525 static void pending_bios_fn(struct btrfs_work *work)
526 {
527         struct btrfs_device *device;
528
529         device = container_of(work, struct btrfs_device, work);
530         run_scheduled_bios(device);
531 }
532
533
534 static void btrfs_free_stale_device(struct btrfs_device *cur_dev)
535 {
536         struct btrfs_fs_devices *fs_devs;
537         struct btrfs_device *dev;
538
539         if (!cur_dev->name)
540                 return;
541
542         list_for_each_entry(fs_devs, &fs_uuids, list) {
543                 int del = 1;
544
545                 if (fs_devs->opened)
546                         continue;
547                 if (fs_devs->seeding)
548                         continue;
549
550                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
551
552                         if (dev == cur_dev)
553                                 continue;
554                         if (!dev->name)
555                                 continue;
556
557                         /*
558                          * Todo: This won't be enough. What if the same device
559                          * comes back (with new uuid and) with its mapper path?
560                          * But for now, this does help as mostly an admin will
561                          * either use mapper or non mapper path throughout.
562                          */
563                         rcu_read_lock();
564                         del = strcmp(rcu_str_deref(dev->name),
565                                                 rcu_str_deref(cur_dev->name));
566                         rcu_read_unlock();
567                         if (!del)
568                                 break;
569                 }
570
571                 if (!del) {
572                         /* delete the stale device */
573                         if (fs_devs->num_devices == 1) {
574                                 btrfs_sysfs_remove_fsid(fs_devs);
575                                 list_del(&fs_devs->list);
576                                 free_fs_devices(fs_devs);
577                         } else {
578                                 fs_devs->num_devices--;
579                                 list_del(&dev->dev_list);
580                                 rcu_string_free(dev->name);
581                                 kfree(dev);
582                         }
583                         break;
584                 }
585         }
586 }
587
588 /*
589  * Add new device to list of registered devices
590  *
591  * Returns:
592  * 1   - first time device is seen
593  * 0   - device already known
594  * < 0 - error
595  */
596 static noinline int device_list_add(const char *path,
597                            struct btrfs_super_block *disk_super,
598                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
599 {
600         struct btrfs_device *device;
601         struct btrfs_fs_devices *fs_devices;
602         struct rcu_string *name;
603         int ret = 0;
604         u64 found_transid = btrfs_super_generation(disk_super);
605
606         fs_devices = find_fsid(disk_super->fsid);
607         if (!fs_devices) {
608                 fs_devices = alloc_fs_devices(disk_super->fsid);
609                 if (IS_ERR(fs_devices))
610                         return PTR_ERR(fs_devices);
611
612                 list_add(&fs_devices->list, &fs_uuids);
613
614                 device = NULL;
615         } else {
616                 device = find_device(fs_devices, devid,
617                                 disk_super->dev_item.uuid);
618         }
619
620         if (!device) {
621                 if (fs_devices->opened)
622                         return -EBUSY;
623
624                 device = btrfs_alloc_device(NULL, &devid,
625                                             disk_super->dev_item.uuid);
626                 if (IS_ERR(device)) {
627                         /* we can safely leave the fs_devices entry around */
628                         return PTR_ERR(device);
629                 }
630
631                 name = rcu_string_strdup(path, GFP_NOFS);
632                 if (!name) {
633                         kfree(device);
634                         return -ENOMEM;
635                 }
636                 rcu_assign_pointer(device->name, name);
637
638                 mutex_lock(&fs_devices->device_list_mutex);
639                 list_add_rcu(&device->dev_list, &fs_devices->devices);
640                 fs_devices->num_devices++;
641                 mutex_unlock(&fs_devices->device_list_mutex);
642
643                 ret = 1;
644                 device->fs_devices = fs_devices;
645         } else if (!device->name || strcmp(device->name->str, path)) {
646                 /*
647                  * When FS is already mounted.
648                  * 1. If you are here and if the device->name is NULL that
649                  *    means this device was missing at time of FS mount.
650                  * 2. If you are here and if the device->name is different
651                  *    from 'path' that means either
652                  *      a. The same device disappeared and reappeared with
653                  *         different name. or
654                  *      b. The missing-disk-which-was-replaced, has
655                  *         reappeared now.
656                  *
657                  * We must allow 1 and 2a above. But 2b would be a spurious
658                  * and unintentional.
659                  *
660                  * Further in case of 1 and 2a above, the disk at 'path'
661                  * would have missed some transaction when it was away and
662                  * in case of 2a the stale bdev has to be updated as well.
663                  * 2b must not be allowed at all time.
664                  */
665
666                 /*
667                  * For now, we do allow update to btrfs_fs_device through the
668                  * btrfs dev scan cli after FS has been mounted.  We're still
669                  * tracking a problem where systems fail mount by subvolume id
670                  * when we reject replacement on a mounted FS.
671                  */
672                 if (!fs_devices->opened && found_transid < device->generation) {
673                         /*
674                          * That is if the FS is _not_ mounted and if you
675                          * are here, that means there is more than one
676                          * disk with same uuid and devid.We keep the one
677                          * with larger generation number or the last-in if
678                          * generation are equal.
679                          */
680                         return -EEXIST;
681                 }
682
683                 name = rcu_string_strdup(path, GFP_NOFS);
684                 if (!name)
685                         return -ENOMEM;
686                 rcu_string_free(device->name);
687                 rcu_assign_pointer(device->name, name);
688                 if (device->missing) {
689                         fs_devices->missing_devices--;
690                         device->missing = 0;
691                 }
692         }
693
694         /*
695          * Unmount does not free the btrfs_device struct but would zero
696          * generation along with most of the other members. So just update
697          * it back. We need it to pick the disk with largest generation
698          * (as above).
699          */
700         if (!fs_devices->opened)
701                 device->generation = found_transid;
702
703         /*
704          * if there is new btrfs on an already registered device,
705          * then remove the stale device entry.
706          */
707         if (ret > 0)
708                 btrfs_free_stale_device(device);
709
710         *fs_devices_ret = fs_devices;
711
712         return ret;
713 }
714
715 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
716 {
717         struct btrfs_fs_devices *fs_devices;
718         struct btrfs_device *device;
719         struct btrfs_device *orig_dev;
720
721         fs_devices = alloc_fs_devices(orig->fsid);
722         if (IS_ERR(fs_devices))
723                 return fs_devices;
724
725         mutex_lock(&orig->device_list_mutex);
726         fs_devices->total_devices = orig->total_devices;
727
728         /* We have held the volume lock, it is safe to get the devices. */
729         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
730                 struct rcu_string *name;
731
732                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
733                                             orig_dev->uuid);
734                 if (IS_ERR(device))
735                         goto error;
736
737                 /*
738                  * This is ok to do without rcu read locked because we hold the
739                  * uuid mutex so nothing we touch in here is going to disappear.
740                  */
741                 if (orig_dev->name) {
742                         name = rcu_string_strdup(orig_dev->name->str,
743                                         GFP_KERNEL);
744                         if (!name) {
745                                 kfree(device);
746                                 goto error;
747                         }
748                         rcu_assign_pointer(device->name, name);
749                 }
750
751                 list_add(&device->dev_list, &fs_devices->devices);
752                 device->fs_devices = fs_devices;
753                 fs_devices->num_devices++;
754         }
755         mutex_unlock(&orig->device_list_mutex);
756         return fs_devices;
757 error:
758         mutex_unlock(&orig->device_list_mutex);
759         free_fs_devices(fs_devices);
760         return ERR_PTR(-ENOMEM);
761 }
762
763 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
764 {
765         struct btrfs_device *device, *next;
766         struct btrfs_device *latest_dev = NULL;
767
768         mutex_lock(&uuid_mutex);
769 again:
770         /* This is the initialized path, it is safe to release the devices. */
771         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
772                 if (device->in_fs_metadata) {
773                         if (!device->is_tgtdev_for_dev_replace &&
774                             (!latest_dev ||
775                              device->generation > latest_dev->generation)) {
776                                 latest_dev = device;
777                         }
778                         continue;
779                 }
780
781                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
782                         /*
783                          * In the first step, keep the device which has
784                          * the correct fsid and the devid that is used
785                          * for the dev_replace procedure.
786                          * In the second step, the dev_replace state is
787                          * read from the device tree and it is known
788                          * whether the procedure is really active or
789                          * not, which means whether this device is
790                          * used or whether it should be removed.
791                          */
792                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
793                                 continue;
794                         }
795                 }
796                 if (device->bdev) {
797                         blkdev_put(device->bdev, device->mode);
798                         device->bdev = NULL;
799                         fs_devices->open_devices--;
800                 }
801                 if (device->writeable) {
802                         list_del_init(&device->dev_alloc_list);
803                         device->writeable = 0;
804                         if (!device->is_tgtdev_for_dev_replace)
805                                 fs_devices->rw_devices--;
806                 }
807                 list_del_init(&device->dev_list);
808                 fs_devices->num_devices--;
809                 rcu_string_free(device->name);
810                 kfree(device);
811         }
812
813         if (fs_devices->seed) {
814                 fs_devices = fs_devices->seed;
815                 goto again;
816         }
817
818         fs_devices->latest_bdev = latest_dev->bdev;
819
820         mutex_unlock(&uuid_mutex);
821 }
822
823 static void __free_device(struct work_struct *work)
824 {
825         struct btrfs_device *device;
826
827         device = container_of(work, struct btrfs_device, rcu_work);
828         rcu_string_free(device->name);
829         bio_put(device->flush_bio);
830         kfree(device);
831 }
832
833 static void free_device(struct rcu_head *head)
834 {
835         struct btrfs_device *device;
836
837         device = container_of(head, struct btrfs_device, rcu);
838
839         INIT_WORK(&device->rcu_work, __free_device);
840         schedule_work(&device->rcu_work);
841 }
842
843 static void btrfs_close_bdev(struct btrfs_device *device)
844 {
845         if (device->bdev && device->writeable) {
846                 sync_blockdev(device->bdev);
847                 invalidate_bdev(device->bdev);
848         }
849
850         if (device->bdev)
851                 blkdev_put(device->bdev, device->mode);
852 }
853
854 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
855 {
856         struct btrfs_fs_devices *fs_devices = device->fs_devices;
857         struct btrfs_device *new_device;
858         struct rcu_string *name;
859
860         if (device->bdev)
861                 fs_devices->open_devices--;
862
863         if (device->writeable &&
864             device->devid != BTRFS_DEV_REPLACE_DEVID) {
865                 list_del_init(&device->dev_alloc_list);
866                 fs_devices->rw_devices--;
867         }
868
869         if (device->missing)
870                 fs_devices->missing_devices--;
871
872         new_device = btrfs_alloc_device(NULL, &device->devid,
873                                         device->uuid);
874         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
875
876         /* Safe because we are under uuid_mutex */
877         if (device->name) {
878                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
879                 BUG_ON(!name); /* -ENOMEM */
880                 rcu_assign_pointer(new_device->name, name);
881         }
882
883         list_replace_rcu(&device->dev_list, &new_device->dev_list);
884         new_device->fs_devices = device->fs_devices;
885 }
886
887 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
888 {
889         struct btrfs_device *device, *tmp;
890         struct list_head pending_put;
891
892         INIT_LIST_HEAD(&pending_put);
893
894         if (--fs_devices->opened > 0)
895                 return 0;
896
897         mutex_lock(&fs_devices->device_list_mutex);
898         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
899                 btrfs_prepare_close_one_device(device);
900                 list_add(&device->dev_list, &pending_put);
901         }
902         mutex_unlock(&fs_devices->device_list_mutex);
903
904         /*
905          * btrfs_show_devname() is using the device_list_mutex,
906          * sometimes call to blkdev_put() leads vfs calling
907          * into this func. So do put outside of device_list_mutex,
908          * as of now.
909          */
910         while (!list_empty(&pending_put)) {
911                 device = list_first_entry(&pending_put,
912                                 struct btrfs_device, dev_list);
913                 list_del(&device->dev_list);
914                 btrfs_close_bdev(device);
915                 call_rcu(&device->rcu, free_device);
916         }
917
918         WARN_ON(fs_devices->open_devices);
919         WARN_ON(fs_devices->rw_devices);
920         fs_devices->opened = 0;
921         fs_devices->seeding = 0;
922
923         return 0;
924 }
925
926 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
927 {
928         struct btrfs_fs_devices *seed_devices = NULL;
929         int ret;
930
931         mutex_lock(&uuid_mutex);
932         ret = __btrfs_close_devices(fs_devices);
933         if (!fs_devices->opened) {
934                 seed_devices = fs_devices->seed;
935                 fs_devices->seed = NULL;
936         }
937         mutex_unlock(&uuid_mutex);
938
939         while (seed_devices) {
940                 fs_devices = seed_devices;
941                 seed_devices = fs_devices->seed;
942                 __btrfs_close_devices(fs_devices);
943                 free_fs_devices(fs_devices);
944         }
945         /*
946          * Wait for rcu kworkers under __btrfs_close_devices
947          * to finish all blkdev_puts so device is really
948          * free when umount is done.
949          */
950         rcu_barrier();
951         return ret;
952 }
953
954 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
955                                 fmode_t flags, void *holder)
956 {
957         struct request_queue *q;
958         struct block_device *bdev;
959         struct list_head *head = &fs_devices->devices;
960         struct btrfs_device *device;
961         struct btrfs_device *latest_dev = NULL;
962         struct buffer_head *bh;
963         struct btrfs_super_block *disk_super;
964         u64 devid;
965         int seeding = 1;
966         int ret = 0;
967
968         flags |= FMODE_EXCL;
969
970         list_for_each_entry(device, head, dev_list) {
971                 if (device->bdev)
972                         continue;
973                 if (!device->name)
974                         continue;
975
976                 /* Just open everything we can; ignore failures here */
977                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
978                                             &bdev, &bh))
979                         continue;
980
981                 disk_super = (struct btrfs_super_block *)bh->b_data;
982                 devid = btrfs_stack_device_id(&disk_super->dev_item);
983                 if (devid != device->devid)
984                         goto error_brelse;
985
986                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
987                            BTRFS_UUID_SIZE))
988                         goto error_brelse;
989
990                 device->generation = btrfs_super_generation(disk_super);
991                 if (!latest_dev ||
992                     device->generation > latest_dev->generation)
993                         latest_dev = device;
994
995                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
996                         device->writeable = 0;
997                 } else {
998                         device->writeable = !bdev_read_only(bdev);
999                         seeding = 0;
1000                 }
1001
1002                 q = bdev_get_queue(bdev);
1003                 if (blk_queue_discard(q))
1004                         device->can_discard = 1;
1005                 if (!blk_queue_nonrot(q))
1006                         fs_devices->rotating = 1;
1007
1008                 device->bdev = bdev;
1009                 device->in_fs_metadata = 0;
1010                 device->mode = flags;
1011
1012                 fs_devices->open_devices++;
1013                 if (device->writeable &&
1014                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1015                         fs_devices->rw_devices++;
1016                         list_add(&device->dev_alloc_list,
1017                                  &fs_devices->alloc_list);
1018                 }
1019                 brelse(bh);
1020                 continue;
1021
1022 error_brelse:
1023                 brelse(bh);
1024                 blkdev_put(bdev, flags);
1025                 continue;
1026         }
1027         if (fs_devices->open_devices == 0) {
1028                 ret = -EINVAL;
1029                 goto out;
1030         }
1031         fs_devices->seeding = seeding;
1032         fs_devices->opened = 1;
1033         fs_devices->latest_bdev = latest_dev->bdev;
1034         fs_devices->total_rw_bytes = 0;
1035 out:
1036         return ret;
1037 }
1038
1039 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1040                        fmode_t flags, void *holder)
1041 {
1042         int ret;
1043
1044         mutex_lock(&uuid_mutex);
1045         if (fs_devices->opened) {
1046                 fs_devices->opened++;
1047                 ret = 0;
1048         } else {
1049                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1050         }
1051         mutex_unlock(&uuid_mutex);
1052         return ret;
1053 }
1054
1055 static void btrfs_release_disk_super(struct page *page)
1056 {
1057         kunmap(page);
1058         put_page(page);
1059 }
1060
1061 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1062                                  struct page **page,
1063                                  struct btrfs_super_block **disk_super)
1064 {
1065         void *p;
1066         pgoff_t index;
1067
1068         /* make sure our super fits in the device */
1069         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1070                 return 1;
1071
1072         /* make sure our super fits in the page */
1073         if (sizeof(**disk_super) > PAGE_SIZE)
1074                 return 1;
1075
1076         /* make sure our super doesn't straddle pages on disk */
1077         index = bytenr >> PAGE_SHIFT;
1078         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1079                 return 1;
1080
1081         /* pull in the page with our super */
1082         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1083                                    index, GFP_KERNEL);
1084
1085         if (IS_ERR_OR_NULL(*page))
1086                 return 1;
1087
1088         p = kmap(*page);
1089
1090         /* align our pointer to the offset of the super block */
1091         *disk_super = p + (bytenr & ~PAGE_MASK);
1092
1093         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1094             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1095                 btrfs_release_disk_super(*page);
1096                 return 1;
1097         }
1098
1099         if ((*disk_super)->label[0] &&
1100                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1101                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1102
1103         return 0;
1104 }
1105
1106 /*
1107  * Look for a btrfs signature on a device. This may be called out of the mount path
1108  * and we are not allowed to call set_blocksize during the scan. The superblock
1109  * is read via pagecache
1110  */
1111 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1112                           struct btrfs_fs_devices **fs_devices_ret)
1113 {
1114         struct btrfs_super_block *disk_super;
1115         struct block_device *bdev;
1116         struct page *page;
1117         int ret = -EINVAL;
1118         u64 devid;
1119         u64 transid;
1120         u64 total_devices;
1121         u64 bytenr;
1122
1123         /*
1124          * we would like to check all the supers, but that would make
1125          * a btrfs mount succeed after a mkfs from a different FS.
1126          * So, we need to add a special mount option to scan for
1127          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1128          */
1129         bytenr = btrfs_sb_offset(0);
1130         flags |= FMODE_EXCL;
1131         mutex_lock(&uuid_mutex);
1132
1133         bdev = blkdev_get_by_path(path, flags, holder);
1134         if (IS_ERR(bdev)) {
1135                 ret = PTR_ERR(bdev);
1136                 goto error;
1137         }
1138
1139         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1140                 goto error_bdev_put;
1141
1142         devid = btrfs_stack_device_id(&disk_super->dev_item);
1143         transid = btrfs_super_generation(disk_super);
1144         total_devices = btrfs_super_num_devices(disk_super);
1145
1146         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1147         if (ret > 0) {
1148                 if (disk_super->label[0]) {
1149                         pr_info("BTRFS: device label %s ", disk_super->label);
1150                 } else {
1151                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1152                 }
1153
1154                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1155                 ret = 0;
1156         }
1157         if (!ret && fs_devices_ret)
1158                 (*fs_devices_ret)->total_devices = total_devices;
1159
1160         btrfs_release_disk_super(page);
1161
1162 error_bdev_put:
1163         blkdev_put(bdev, flags);
1164 error:
1165         mutex_unlock(&uuid_mutex);
1166         return ret;
1167 }
1168
1169 /* helper to account the used device space in the range */
1170 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1171                                    u64 end, u64 *length)
1172 {
1173         struct btrfs_key key;
1174         struct btrfs_root *root = device->fs_info->dev_root;
1175         struct btrfs_dev_extent *dev_extent;
1176         struct btrfs_path *path;
1177         u64 extent_end;
1178         int ret;
1179         int slot;
1180         struct extent_buffer *l;
1181
1182         *length = 0;
1183
1184         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1185                 return 0;
1186
1187         path = btrfs_alloc_path();
1188         if (!path)
1189                 return -ENOMEM;
1190         path->reada = READA_FORWARD;
1191
1192         key.objectid = device->devid;
1193         key.offset = start;
1194         key.type = BTRFS_DEV_EXTENT_KEY;
1195
1196         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1197         if (ret < 0)
1198                 goto out;
1199         if (ret > 0) {
1200                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1201                 if (ret < 0)
1202                         goto out;
1203         }
1204
1205         while (1) {
1206                 l = path->nodes[0];
1207                 slot = path->slots[0];
1208                 if (slot >= btrfs_header_nritems(l)) {
1209                         ret = btrfs_next_leaf(root, path);
1210                         if (ret == 0)
1211                                 continue;
1212                         if (ret < 0)
1213                                 goto out;
1214
1215                         break;
1216                 }
1217                 btrfs_item_key_to_cpu(l, &key, slot);
1218
1219                 if (key.objectid < device->devid)
1220                         goto next;
1221
1222                 if (key.objectid > device->devid)
1223                         break;
1224
1225                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1226                         goto next;
1227
1228                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1229                 extent_end = key.offset + btrfs_dev_extent_length(l,
1230                                                                   dev_extent);
1231                 if (key.offset <= start && extent_end > end) {
1232                         *length = end - start + 1;
1233                         break;
1234                 } else if (key.offset <= start && extent_end > start)
1235                         *length += extent_end - start;
1236                 else if (key.offset > start && extent_end <= end)
1237                         *length += extent_end - key.offset;
1238                 else if (key.offset > start && key.offset <= end) {
1239                         *length += end - key.offset + 1;
1240                         break;
1241                 } else if (key.offset > end)
1242                         break;
1243
1244 next:
1245                 path->slots[0]++;
1246         }
1247         ret = 0;
1248 out:
1249         btrfs_free_path(path);
1250         return ret;
1251 }
1252
1253 static int contains_pending_extent(struct btrfs_transaction *transaction,
1254                                    struct btrfs_device *device,
1255                                    u64 *start, u64 len)
1256 {
1257         struct btrfs_fs_info *fs_info = device->fs_info;
1258         struct extent_map *em;
1259         struct list_head *search_list = &fs_info->pinned_chunks;
1260         int ret = 0;
1261         u64 physical_start = *start;
1262
1263         if (transaction)
1264                 search_list = &transaction->pending_chunks;
1265 again:
1266         list_for_each_entry(em, search_list, list) {
1267                 struct map_lookup *map;
1268                 int i;
1269
1270                 map = em->map_lookup;
1271                 for (i = 0; i < map->num_stripes; i++) {
1272                         u64 end;
1273
1274                         if (map->stripes[i].dev != device)
1275                                 continue;
1276                         if (map->stripes[i].physical >= physical_start + len ||
1277                             map->stripes[i].physical + em->orig_block_len <=
1278                             physical_start)
1279                                 continue;
1280                         /*
1281                          * Make sure that while processing the pinned list we do
1282                          * not override our *start with a lower value, because
1283                          * we can have pinned chunks that fall within this
1284                          * device hole and that have lower physical addresses
1285                          * than the pending chunks we processed before. If we
1286                          * do not take this special care we can end up getting
1287                          * 2 pending chunks that start at the same physical
1288                          * device offsets because the end offset of a pinned
1289                          * chunk can be equal to the start offset of some
1290                          * pending chunk.
1291                          */
1292                         end = map->stripes[i].physical + em->orig_block_len;
1293                         if (end > *start) {
1294                                 *start = end;
1295                                 ret = 1;
1296                         }
1297                 }
1298         }
1299         if (search_list != &fs_info->pinned_chunks) {
1300                 search_list = &fs_info->pinned_chunks;
1301                 goto again;
1302         }
1303
1304         return ret;
1305 }
1306
1307
1308 /*
1309  * find_free_dev_extent_start - find free space in the specified device
1310  * @device:       the device which we search the free space in
1311  * @num_bytes:    the size of the free space that we need
1312  * @search_start: the position from which to begin the search
1313  * @start:        store the start of the free space.
1314  * @len:          the size of the free space. that we find, or the size
1315  *                of the max free space if we don't find suitable free space
1316  *
1317  * this uses a pretty simple search, the expectation is that it is
1318  * called very infrequently and that a given device has a small number
1319  * of extents
1320  *
1321  * @start is used to store the start of the free space if we find. But if we
1322  * don't find suitable free space, it will be used to store the start position
1323  * of the max free space.
1324  *
1325  * @len is used to store the size of the free space that we find.
1326  * But if we don't find suitable free space, it is used to store the size of
1327  * the max free space.
1328  */
1329 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1330                                struct btrfs_device *device, u64 num_bytes,
1331                                u64 search_start, u64 *start, u64 *len)
1332 {
1333         struct btrfs_fs_info *fs_info = device->fs_info;
1334         struct btrfs_root *root = fs_info->dev_root;
1335         struct btrfs_key key;
1336         struct btrfs_dev_extent *dev_extent;
1337         struct btrfs_path *path;
1338         u64 hole_size;
1339         u64 max_hole_start;
1340         u64 max_hole_size;
1341         u64 extent_end;
1342         u64 search_end = device->total_bytes;
1343         int ret;
1344         int slot;
1345         struct extent_buffer *l;
1346
1347         /*
1348          * We don't want to overwrite the superblock on the drive nor any area
1349          * used by the boot loader (grub for example), so we make sure to start
1350          * at an offset of at least 1MB.
1351          */
1352         search_start = max_t(u64, search_start, SZ_1M);
1353
1354         path = btrfs_alloc_path();
1355         if (!path)
1356                 return -ENOMEM;
1357
1358         max_hole_start = search_start;
1359         max_hole_size = 0;
1360
1361 again:
1362         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1363                 ret = -ENOSPC;
1364                 goto out;
1365         }
1366
1367         path->reada = READA_FORWARD;
1368         path->search_commit_root = 1;
1369         path->skip_locking = 1;
1370
1371         key.objectid = device->devid;
1372         key.offset = search_start;
1373         key.type = BTRFS_DEV_EXTENT_KEY;
1374
1375         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1376         if (ret < 0)
1377                 goto out;
1378         if (ret > 0) {
1379                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1380                 if (ret < 0)
1381                         goto out;
1382         }
1383
1384         while (1) {
1385                 l = path->nodes[0];
1386                 slot = path->slots[0];
1387                 if (slot >= btrfs_header_nritems(l)) {
1388                         ret = btrfs_next_leaf(root, path);
1389                         if (ret == 0)
1390                                 continue;
1391                         if (ret < 0)
1392                                 goto out;
1393
1394                         break;
1395                 }
1396                 btrfs_item_key_to_cpu(l, &key, slot);
1397
1398                 if (key.objectid < device->devid)
1399                         goto next;
1400
1401                 if (key.objectid > device->devid)
1402                         break;
1403
1404                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1405                         goto next;
1406
1407                 if (key.offset > search_start) {
1408                         hole_size = key.offset - search_start;
1409
1410                         /*
1411                          * Have to check before we set max_hole_start, otherwise
1412                          * we could end up sending back this offset anyway.
1413                          */
1414                         if (contains_pending_extent(transaction, device,
1415                                                     &search_start,
1416                                                     hole_size)) {
1417                                 if (key.offset >= search_start) {
1418                                         hole_size = key.offset - search_start;
1419                                 } else {
1420                                         WARN_ON_ONCE(1);
1421                                         hole_size = 0;
1422                                 }
1423                         }
1424
1425                         if (hole_size > max_hole_size) {
1426                                 max_hole_start = search_start;
1427                                 max_hole_size = hole_size;
1428                         }
1429
1430                         /*
1431                          * If this free space is greater than which we need,
1432                          * it must be the max free space that we have found
1433                          * until now, so max_hole_start must point to the start
1434                          * of this free space and the length of this free space
1435                          * is stored in max_hole_size. Thus, we return
1436                          * max_hole_start and max_hole_size and go back to the
1437                          * caller.
1438                          */
1439                         if (hole_size >= num_bytes) {
1440                                 ret = 0;
1441                                 goto out;
1442                         }
1443                 }
1444
1445                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1446                 extent_end = key.offset + btrfs_dev_extent_length(l,
1447                                                                   dev_extent);
1448                 if (extent_end > search_start)
1449                         search_start = extent_end;
1450 next:
1451                 path->slots[0]++;
1452                 cond_resched();
1453         }
1454
1455         /*
1456          * At this point, search_start should be the end of
1457          * allocated dev extents, and when shrinking the device,
1458          * search_end may be smaller than search_start.
1459          */
1460         if (search_end > search_start) {
1461                 hole_size = search_end - search_start;
1462
1463                 if (contains_pending_extent(transaction, device, &search_start,
1464                                             hole_size)) {
1465                         btrfs_release_path(path);
1466                         goto again;
1467                 }
1468
1469                 if (hole_size > max_hole_size) {
1470                         max_hole_start = search_start;
1471                         max_hole_size = hole_size;
1472                 }
1473         }
1474
1475         /* See above. */
1476         if (max_hole_size < num_bytes)
1477                 ret = -ENOSPC;
1478         else
1479                 ret = 0;
1480
1481 out:
1482         btrfs_free_path(path);
1483         *start = max_hole_start;
1484         if (len)
1485                 *len = max_hole_size;
1486         return ret;
1487 }
1488
1489 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1490                          struct btrfs_device *device, u64 num_bytes,
1491                          u64 *start, u64 *len)
1492 {
1493         /* FIXME use last free of some kind */
1494         return find_free_dev_extent_start(trans->transaction, device,
1495                                           num_bytes, 0, start, len);
1496 }
1497
1498 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1499                           struct btrfs_device *device,
1500                           u64 start, u64 *dev_extent_len)
1501 {
1502         struct btrfs_fs_info *fs_info = device->fs_info;
1503         struct btrfs_root *root = fs_info->dev_root;
1504         int ret;
1505         struct btrfs_path *path;
1506         struct btrfs_key key;
1507         struct btrfs_key found_key;
1508         struct extent_buffer *leaf = NULL;
1509         struct btrfs_dev_extent *extent = NULL;
1510
1511         path = btrfs_alloc_path();
1512         if (!path)
1513                 return -ENOMEM;
1514
1515         key.objectid = device->devid;
1516         key.offset = start;
1517         key.type = BTRFS_DEV_EXTENT_KEY;
1518 again:
1519         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1520         if (ret > 0) {
1521                 ret = btrfs_previous_item(root, path, key.objectid,
1522                                           BTRFS_DEV_EXTENT_KEY);
1523                 if (ret)
1524                         goto out;
1525                 leaf = path->nodes[0];
1526                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1527                 extent = btrfs_item_ptr(leaf, path->slots[0],
1528                                         struct btrfs_dev_extent);
1529                 BUG_ON(found_key.offset > start || found_key.offset +
1530                        btrfs_dev_extent_length(leaf, extent) < start);
1531                 key = found_key;
1532                 btrfs_release_path(path);
1533                 goto again;
1534         } else if (ret == 0) {
1535                 leaf = path->nodes[0];
1536                 extent = btrfs_item_ptr(leaf, path->slots[0],
1537                                         struct btrfs_dev_extent);
1538         } else {
1539                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1540                 goto out;
1541         }
1542
1543         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1544
1545         ret = btrfs_del_item(trans, root, path);
1546         if (ret) {
1547                 btrfs_handle_fs_error(fs_info, ret,
1548                                       "Failed to remove dev extent item");
1549         } else {
1550                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1551         }
1552 out:
1553         btrfs_free_path(path);
1554         return ret;
1555 }
1556
1557 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1558                                   struct btrfs_device *device,
1559                                   u64 chunk_offset, u64 start, u64 num_bytes)
1560 {
1561         int ret;
1562         struct btrfs_path *path;
1563         struct btrfs_fs_info *fs_info = device->fs_info;
1564         struct btrfs_root *root = fs_info->dev_root;
1565         struct btrfs_dev_extent *extent;
1566         struct extent_buffer *leaf;
1567         struct btrfs_key key;
1568
1569         WARN_ON(!device->in_fs_metadata);
1570         WARN_ON(device->is_tgtdev_for_dev_replace);
1571         path = btrfs_alloc_path();
1572         if (!path)
1573                 return -ENOMEM;
1574
1575         key.objectid = device->devid;
1576         key.offset = start;
1577         key.type = BTRFS_DEV_EXTENT_KEY;
1578         ret = btrfs_insert_empty_item(trans, root, path, &key,
1579                                       sizeof(*extent));
1580         if (ret)
1581                 goto out;
1582
1583         leaf = path->nodes[0];
1584         extent = btrfs_item_ptr(leaf, path->slots[0],
1585                                 struct btrfs_dev_extent);
1586         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1587                                         BTRFS_CHUNK_TREE_OBJECTID);
1588         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1589                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1590         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1591
1592         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1593         btrfs_mark_buffer_dirty(leaf);
1594 out:
1595         btrfs_free_path(path);
1596         return ret;
1597 }
1598
1599 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1600 {
1601         struct extent_map_tree *em_tree;
1602         struct extent_map *em;
1603         struct rb_node *n;
1604         u64 ret = 0;
1605
1606         em_tree = &fs_info->mapping_tree.map_tree;
1607         read_lock(&em_tree->lock);
1608         n = rb_last(&em_tree->map);
1609         if (n) {
1610                 em = rb_entry(n, struct extent_map, rb_node);
1611                 ret = em->start + em->len;
1612         }
1613         read_unlock(&em_tree->lock);
1614
1615         return ret;
1616 }
1617
1618 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1619                                     u64 *devid_ret)
1620 {
1621         int ret;
1622         struct btrfs_key key;
1623         struct btrfs_key found_key;
1624         struct btrfs_path *path;
1625
1626         path = btrfs_alloc_path();
1627         if (!path)
1628                 return -ENOMEM;
1629
1630         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1631         key.type = BTRFS_DEV_ITEM_KEY;
1632         key.offset = (u64)-1;
1633
1634         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1635         if (ret < 0)
1636                 goto error;
1637
1638         BUG_ON(ret == 0); /* Corruption */
1639
1640         ret = btrfs_previous_item(fs_info->chunk_root, path,
1641                                   BTRFS_DEV_ITEMS_OBJECTID,
1642                                   BTRFS_DEV_ITEM_KEY);
1643         if (ret) {
1644                 *devid_ret = 1;
1645         } else {
1646                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1647                                       path->slots[0]);
1648                 *devid_ret = found_key.offset + 1;
1649         }
1650         ret = 0;
1651 error:
1652         btrfs_free_path(path);
1653         return ret;
1654 }
1655
1656 /*
1657  * the device information is stored in the chunk root
1658  * the btrfs_device struct should be fully filled in
1659  */
1660 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1661                             struct btrfs_fs_info *fs_info,
1662                             struct btrfs_device *device)
1663 {
1664         struct btrfs_root *root = fs_info->chunk_root;
1665         int ret;
1666         struct btrfs_path *path;
1667         struct btrfs_dev_item *dev_item;
1668         struct extent_buffer *leaf;
1669         struct btrfs_key key;
1670         unsigned long ptr;
1671
1672         path = btrfs_alloc_path();
1673         if (!path)
1674                 return -ENOMEM;
1675
1676         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1677         key.type = BTRFS_DEV_ITEM_KEY;
1678         key.offset = device->devid;
1679
1680         ret = btrfs_insert_empty_item(trans, root, path, &key,
1681                                       sizeof(*dev_item));
1682         if (ret)
1683                 goto out;
1684
1685         leaf = path->nodes[0];
1686         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1687
1688         btrfs_set_device_id(leaf, dev_item, device->devid);
1689         btrfs_set_device_generation(leaf, dev_item, 0);
1690         btrfs_set_device_type(leaf, dev_item, device->type);
1691         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1692         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1693         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1694         btrfs_set_device_total_bytes(leaf, dev_item,
1695                                      btrfs_device_get_disk_total_bytes(device));
1696         btrfs_set_device_bytes_used(leaf, dev_item,
1697                                     btrfs_device_get_bytes_used(device));
1698         btrfs_set_device_group(leaf, dev_item, 0);
1699         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1700         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1701         btrfs_set_device_start_offset(leaf, dev_item, 0);
1702
1703         ptr = btrfs_device_uuid(dev_item);
1704         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1705         ptr = btrfs_device_fsid(dev_item);
1706         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1707         btrfs_mark_buffer_dirty(leaf);
1708
1709         ret = 0;
1710 out:
1711         btrfs_free_path(path);
1712         return ret;
1713 }
1714
1715 /*
1716  * Function to update ctime/mtime for a given device path.
1717  * Mainly used for ctime/mtime based probe like libblkid.
1718  */
1719 static void update_dev_time(const char *path_name)
1720 {
1721         struct file *filp;
1722
1723         filp = filp_open(path_name, O_RDWR, 0);
1724         if (IS_ERR(filp))
1725                 return;
1726         file_update_time(filp);
1727         filp_close(filp, NULL);
1728 }
1729
1730 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1731                              struct btrfs_device *device)
1732 {
1733         struct btrfs_root *root = fs_info->chunk_root;
1734         int ret;
1735         struct btrfs_path *path;
1736         struct btrfs_key key;
1737         struct btrfs_trans_handle *trans;
1738
1739         path = btrfs_alloc_path();
1740         if (!path)
1741                 return -ENOMEM;
1742
1743         trans = btrfs_start_transaction(root, 0);
1744         if (IS_ERR(trans)) {
1745                 btrfs_free_path(path);
1746                 return PTR_ERR(trans);
1747         }
1748         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1749         key.type = BTRFS_DEV_ITEM_KEY;
1750         key.offset = device->devid;
1751
1752         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1753         if (ret < 0)
1754                 goto out;
1755
1756         if (ret > 0) {
1757                 ret = -ENOENT;
1758                 goto out;
1759         }
1760
1761         ret = btrfs_del_item(trans, root, path);
1762         if (ret)
1763                 goto out;
1764 out:
1765         btrfs_free_path(path);
1766         btrfs_commit_transaction(trans);
1767         return ret;
1768 }
1769
1770 /*
1771  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1772  * filesystem. It's up to the caller to adjust that number regarding eg. device
1773  * replace.
1774  */
1775 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1776                 u64 num_devices)
1777 {
1778         u64 all_avail;
1779         unsigned seq;
1780         int i;
1781
1782         do {
1783                 seq = read_seqbegin(&fs_info->profiles_lock);
1784
1785                 all_avail = fs_info->avail_data_alloc_bits |
1786                             fs_info->avail_system_alloc_bits |
1787                             fs_info->avail_metadata_alloc_bits;
1788         } while (read_seqretry(&fs_info->profiles_lock, seq));
1789
1790         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1791                 if (!(all_avail & btrfs_raid_group[i]))
1792                         continue;
1793
1794                 if (num_devices < btrfs_raid_array[i].devs_min) {
1795                         int ret = btrfs_raid_mindev_error[i];
1796
1797                         if (ret)
1798                                 return ret;
1799                 }
1800         }
1801
1802         return 0;
1803 }
1804
1805 static struct btrfs_device * btrfs_find_next_active_device(
1806                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1807 {
1808         struct btrfs_device *next_device;
1809
1810         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1811                 if (next_device != device &&
1812                         !next_device->missing && next_device->bdev)
1813                         return next_device;
1814         }
1815
1816         return NULL;
1817 }
1818
1819 /*
1820  * Helper function to check if the given device is part of s_bdev / latest_bdev
1821  * and replace it with the provided or the next active device, in the context
1822  * where this function called, there should be always be another device (or
1823  * this_dev) which is active.
1824  */
1825 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1826                 struct btrfs_device *device, struct btrfs_device *this_dev)
1827 {
1828         struct btrfs_device *next_device;
1829
1830         if (this_dev)
1831                 next_device = this_dev;
1832         else
1833                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1834                                                                 device);
1835         ASSERT(next_device);
1836
1837         if (fs_info->sb->s_bdev &&
1838                         (fs_info->sb->s_bdev == device->bdev))
1839                 fs_info->sb->s_bdev = next_device->bdev;
1840
1841         if (fs_info->fs_devices->latest_bdev == device->bdev)
1842                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1843 }
1844
1845 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1846                 u64 devid)
1847 {
1848         struct btrfs_device *device;
1849         struct btrfs_fs_devices *cur_devices;
1850         u64 num_devices;
1851         int ret = 0;
1852
1853         mutex_lock(&uuid_mutex);
1854
1855         num_devices = fs_info->fs_devices->num_devices;
1856         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1857         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1858                 WARN_ON(num_devices < 1);
1859                 num_devices--;
1860         }
1861         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1862
1863         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1864         if (ret)
1865                 goto out;
1866
1867         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1868                                            &device);
1869         if (ret)
1870                 goto out;
1871
1872         if (device->is_tgtdev_for_dev_replace) {
1873                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1874                 goto out;
1875         }
1876
1877         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1878                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1879                 goto out;
1880         }
1881
1882         if (device->writeable) {
1883                 mutex_lock(&fs_info->chunk_mutex);
1884                 list_del_init(&device->dev_alloc_list);
1885                 device->fs_devices->rw_devices--;
1886                 mutex_unlock(&fs_info->chunk_mutex);
1887         }
1888
1889         mutex_unlock(&uuid_mutex);
1890         ret = btrfs_shrink_device(device, 0);
1891         mutex_lock(&uuid_mutex);
1892         if (ret)
1893                 goto error_undo;
1894
1895         /*
1896          * TODO: the superblock still includes this device in its num_devices
1897          * counter although write_all_supers() is not locked out. This
1898          * could give a filesystem state which requires a degraded mount.
1899          */
1900         ret = btrfs_rm_dev_item(fs_info, device);
1901         if (ret)
1902                 goto error_undo;
1903
1904         device->in_fs_metadata = 0;
1905         btrfs_scrub_cancel_dev(fs_info, device);
1906
1907         /*
1908          * the device list mutex makes sure that we don't change
1909          * the device list while someone else is writing out all
1910          * the device supers. Whoever is writing all supers, should
1911          * lock the device list mutex before getting the number of
1912          * devices in the super block (super_copy). Conversely,
1913          * whoever updates the number of devices in the super block
1914          * (super_copy) should hold the device list mutex.
1915          */
1916
1917         cur_devices = device->fs_devices;
1918         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1919         list_del_rcu(&device->dev_list);
1920
1921         device->fs_devices->num_devices--;
1922         device->fs_devices->total_devices--;
1923
1924         if (device->missing)
1925                 device->fs_devices->missing_devices--;
1926
1927         btrfs_assign_next_active_device(fs_info, device, NULL);
1928
1929         if (device->bdev) {
1930                 device->fs_devices->open_devices--;
1931                 /* remove sysfs entry */
1932                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1933         }
1934
1935         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1936         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1937         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1938
1939         /*
1940          * at this point, the device is zero sized and detached from
1941          * the devices list.  All that's left is to zero out the old
1942          * supers and free the device.
1943          */
1944         if (device->writeable)
1945                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1946
1947         btrfs_close_bdev(device);
1948         call_rcu(&device->rcu, free_device);
1949
1950         if (cur_devices->open_devices == 0) {
1951                 struct btrfs_fs_devices *fs_devices;
1952                 fs_devices = fs_info->fs_devices;
1953                 while (fs_devices) {
1954                         if (fs_devices->seed == cur_devices) {
1955                                 fs_devices->seed = cur_devices->seed;
1956                                 break;
1957                         }
1958                         fs_devices = fs_devices->seed;
1959                 }
1960                 cur_devices->seed = NULL;
1961                 __btrfs_close_devices(cur_devices);
1962                 free_fs_devices(cur_devices);
1963         }
1964
1965 out:
1966         mutex_unlock(&uuid_mutex);
1967         return ret;
1968
1969 error_undo:
1970         if (device->writeable) {
1971                 mutex_lock(&fs_info->chunk_mutex);
1972                 list_add(&device->dev_alloc_list,
1973                          &fs_info->fs_devices->alloc_list);
1974                 device->fs_devices->rw_devices++;
1975                 mutex_unlock(&fs_info->chunk_mutex);
1976         }
1977         goto out;
1978 }
1979
1980 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1981                                         struct btrfs_device *srcdev)
1982 {
1983         struct btrfs_fs_devices *fs_devices;
1984
1985         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1986
1987         /*
1988          * in case of fs with no seed, srcdev->fs_devices will point
1989          * to fs_devices of fs_info. However when the dev being replaced is
1990          * a seed dev it will point to the seed's local fs_devices. In short
1991          * srcdev will have its correct fs_devices in both the cases.
1992          */
1993         fs_devices = srcdev->fs_devices;
1994
1995         list_del_rcu(&srcdev->dev_list);
1996         list_del_rcu(&srcdev->dev_alloc_list);
1997         fs_devices->num_devices--;
1998         if (srcdev->missing)
1999                 fs_devices->missing_devices--;
2000
2001         if (srcdev->writeable)
2002                 fs_devices->rw_devices--;
2003
2004         if (srcdev->bdev)
2005                 fs_devices->open_devices--;
2006 }
2007
2008 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2009                                       struct btrfs_device *srcdev)
2010 {
2011         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2012
2013         if (srcdev->writeable) {
2014                 /* zero out the old super if it is writable */
2015                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2016         }
2017
2018         btrfs_close_bdev(srcdev);
2019         call_rcu(&srcdev->rcu, free_device);
2020
2021         /* if this is no devs we rather delete the fs_devices */
2022         if (!fs_devices->num_devices) {
2023                 struct btrfs_fs_devices *tmp_fs_devices;
2024
2025                 /*
2026                  * On a mounted FS, num_devices can't be zero unless it's a
2027                  * seed. In case of a seed device being replaced, the replace
2028                  * target added to the sprout FS, so there will be no more
2029                  * device left under the seed FS.
2030                  */
2031                 ASSERT(fs_devices->seeding);
2032
2033                 tmp_fs_devices = fs_info->fs_devices;
2034                 while (tmp_fs_devices) {
2035                         if (tmp_fs_devices->seed == fs_devices) {
2036                                 tmp_fs_devices->seed = fs_devices->seed;
2037                                 break;
2038                         }
2039                         tmp_fs_devices = tmp_fs_devices->seed;
2040                 }
2041                 fs_devices->seed = NULL;
2042                 __btrfs_close_devices(fs_devices);
2043                 free_fs_devices(fs_devices);
2044         }
2045 }
2046
2047 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2048                                       struct btrfs_device *tgtdev)
2049 {
2050         mutex_lock(&uuid_mutex);
2051         WARN_ON(!tgtdev);
2052         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2053
2054         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2055
2056         if (tgtdev->bdev)
2057                 fs_info->fs_devices->open_devices--;
2058
2059         fs_info->fs_devices->num_devices--;
2060
2061         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2062
2063         list_del_rcu(&tgtdev->dev_list);
2064
2065         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2066         mutex_unlock(&uuid_mutex);
2067
2068         /*
2069          * The update_dev_time() with in btrfs_scratch_superblocks()
2070          * may lead to a call to btrfs_show_devname() which will try
2071          * to hold device_list_mutex. And here this device
2072          * is already out of device list, so we don't have to hold
2073          * the device_list_mutex lock.
2074          */
2075         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2076
2077         btrfs_close_bdev(tgtdev);
2078         call_rcu(&tgtdev->rcu, free_device);
2079 }
2080
2081 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2082                                      const char *device_path,
2083                                      struct btrfs_device **device)
2084 {
2085         int ret = 0;
2086         struct btrfs_super_block *disk_super;
2087         u64 devid;
2088         u8 *dev_uuid;
2089         struct block_device *bdev;
2090         struct buffer_head *bh;
2091
2092         *device = NULL;
2093         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2094                                     fs_info->bdev_holder, 0, &bdev, &bh);
2095         if (ret)
2096                 return ret;
2097         disk_super = (struct btrfs_super_block *)bh->b_data;
2098         devid = btrfs_stack_device_id(&disk_super->dev_item);
2099         dev_uuid = disk_super->dev_item.uuid;
2100         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2101         brelse(bh);
2102         if (!*device)
2103                 ret = -ENOENT;
2104         blkdev_put(bdev, FMODE_READ);
2105         return ret;
2106 }
2107
2108 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2109                                          const char *device_path,
2110                                          struct btrfs_device **device)
2111 {
2112         *device = NULL;
2113         if (strcmp(device_path, "missing") == 0) {
2114                 struct list_head *devices;
2115                 struct btrfs_device *tmp;
2116
2117                 devices = &fs_info->fs_devices->devices;
2118                 /*
2119                  * It is safe to read the devices since the volume_mutex
2120                  * is held by the caller.
2121                  */
2122                 list_for_each_entry(tmp, devices, dev_list) {
2123                         if (tmp->in_fs_metadata && !tmp->bdev) {
2124                                 *device = tmp;
2125                                 break;
2126                         }
2127                 }
2128
2129                 if (!*device)
2130                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2131
2132                 return 0;
2133         } else {
2134                 return btrfs_find_device_by_path(fs_info, device_path, device);
2135         }
2136 }
2137
2138 /*
2139  * Lookup a device given by device id, or the path if the id is 0.
2140  */
2141 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2142                                  const char *devpath,
2143                                  struct btrfs_device **device)
2144 {
2145         int ret;
2146
2147         if (devid) {
2148                 ret = 0;
2149                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2150                 if (!*device)
2151                         ret = -ENOENT;
2152         } else {
2153                 if (!devpath || !devpath[0])
2154                         return -EINVAL;
2155
2156                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2157                                                            device);
2158         }
2159         return ret;
2160 }
2161
2162 /*
2163  * does all the dirty work required for changing file system's UUID.
2164  */
2165 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2166 {
2167         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2168         struct btrfs_fs_devices *old_devices;
2169         struct btrfs_fs_devices *seed_devices;
2170         struct btrfs_super_block *disk_super = fs_info->super_copy;
2171         struct btrfs_device *device;
2172         u64 super_flags;
2173
2174         BUG_ON(!mutex_is_locked(&uuid_mutex));
2175         if (!fs_devices->seeding)
2176                 return -EINVAL;
2177
2178         seed_devices = alloc_fs_devices(NULL);
2179         if (IS_ERR(seed_devices))
2180                 return PTR_ERR(seed_devices);
2181
2182         old_devices = clone_fs_devices(fs_devices);
2183         if (IS_ERR(old_devices)) {
2184                 kfree(seed_devices);
2185                 return PTR_ERR(old_devices);
2186         }
2187
2188         list_add(&old_devices->list, &fs_uuids);
2189
2190         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2191         seed_devices->opened = 1;
2192         INIT_LIST_HEAD(&seed_devices->devices);
2193         INIT_LIST_HEAD(&seed_devices->alloc_list);
2194         mutex_init(&seed_devices->device_list_mutex);
2195
2196         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2197         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2198                               synchronize_rcu);
2199         list_for_each_entry(device, &seed_devices->devices, dev_list)
2200                 device->fs_devices = seed_devices;
2201
2202         mutex_lock(&fs_info->chunk_mutex);
2203         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2204         mutex_unlock(&fs_info->chunk_mutex);
2205
2206         fs_devices->seeding = 0;
2207         fs_devices->num_devices = 0;
2208         fs_devices->open_devices = 0;
2209         fs_devices->missing_devices = 0;
2210         fs_devices->rotating = 0;
2211         fs_devices->seed = seed_devices;
2212
2213         generate_random_uuid(fs_devices->fsid);
2214         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2215         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2216         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2217
2218         super_flags = btrfs_super_flags(disk_super) &
2219                       ~BTRFS_SUPER_FLAG_SEEDING;
2220         btrfs_set_super_flags(disk_super, super_flags);
2221
2222         return 0;
2223 }
2224
2225 /*
2226  * Store the expected generation for seed devices in device items.
2227  */
2228 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2229                                struct btrfs_fs_info *fs_info)
2230 {
2231         struct btrfs_root *root = fs_info->chunk_root;
2232         struct btrfs_path *path;
2233         struct extent_buffer *leaf;
2234         struct btrfs_dev_item *dev_item;
2235         struct btrfs_device *device;
2236         struct btrfs_key key;
2237         u8 fs_uuid[BTRFS_FSID_SIZE];
2238         u8 dev_uuid[BTRFS_UUID_SIZE];
2239         u64 devid;
2240         int ret;
2241
2242         path = btrfs_alloc_path();
2243         if (!path)
2244                 return -ENOMEM;
2245
2246         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2247         key.offset = 0;
2248         key.type = BTRFS_DEV_ITEM_KEY;
2249
2250         while (1) {
2251                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2252                 if (ret < 0)
2253                         goto error;
2254
2255                 leaf = path->nodes[0];
2256 next_slot:
2257                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2258                         ret = btrfs_next_leaf(root, path);
2259                         if (ret > 0)
2260                                 break;
2261                         if (ret < 0)
2262                                 goto error;
2263                         leaf = path->nodes[0];
2264                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2265                         btrfs_release_path(path);
2266                         continue;
2267                 }
2268
2269                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2270                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2271                     key.type != BTRFS_DEV_ITEM_KEY)
2272                         break;
2273
2274                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2275                                           struct btrfs_dev_item);
2276                 devid = btrfs_device_id(leaf, dev_item);
2277                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2278                                    BTRFS_UUID_SIZE);
2279                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2280                                    BTRFS_FSID_SIZE);
2281                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2282                 BUG_ON(!device); /* Logic error */
2283
2284                 if (device->fs_devices->seeding) {
2285                         btrfs_set_device_generation(leaf, dev_item,
2286                                                     device->generation);
2287                         btrfs_mark_buffer_dirty(leaf);
2288                 }
2289
2290                 path->slots[0]++;
2291                 goto next_slot;
2292         }
2293         ret = 0;
2294 error:
2295         btrfs_free_path(path);
2296         return ret;
2297 }
2298
2299 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2300 {
2301         struct btrfs_root *root = fs_info->dev_root;
2302         struct request_queue *q;
2303         struct btrfs_trans_handle *trans;
2304         struct btrfs_device *device;
2305         struct block_device *bdev;
2306         struct list_head *devices;
2307         struct super_block *sb = fs_info->sb;
2308         struct rcu_string *name;
2309         u64 tmp;
2310         int seeding_dev = 0;
2311         int ret = 0;
2312         bool unlocked = false;
2313
2314         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2315                 return -EROFS;
2316
2317         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2318                                   fs_info->bdev_holder);
2319         if (IS_ERR(bdev))
2320                 return PTR_ERR(bdev);
2321
2322         if (fs_info->fs_devices->seeding) {
2323                 seeding_dev = 1;
2324                 down_write(&sb->s_umount);
2325                 mutex_lock(&uuid_mutex);
2326         }
2327
2328         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2329
2330         devices = &fs_info->fs_devices->devices;
2331
2332         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2333         list_for_each_entry(device, devices, dev_list) {
2334                 if (device->bdev == bdev) {
2335                         ret = -EEXIST;
2336                         mutex_unlock(
2337                                 &fs_info->fs_devices->device_list_mutex);
2338                         goto error;
2339                 }
2340         }
2341         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2342
2343         device = btrfs_alloc_device(fs_info, NULL, NULL);
2344         if (IS_ERR(device)) {
2345                 /* we can safely leave the fs_devices entry around */
2346                 ret = PTR_ERR(device);
2347                 goto error;
2348         }
2349
2350         name = rcu_string_strdup(device_path, GFP_KERNEL);
2351         if (!name) {
2352                 kfree(device);
2353                 ret = -ENOMEM;
2354                 goto error;
2355         }
2356         rcu_assign_pointer(device->name, name);
2357
2358         trans = btrfs_start_transaction(root, 0);
2359         if (IS_ERR(trans)) {
2360                 rcu_string_free(device->name);
2361                 kfree(device);
2362                 ret = PTR_ERR(trans);
2363                 goto error;
2364         }
2365
2366         q = bdev_get_queue(bdev);
2367         if (blk_queue_discard(q))
2368                 device->can_discard = 1;
2369         device->writeable = 1;
2370         device->generation = trans->transid;
2371         device->io_width = fs_info->sectorsize;
2372         device->io_align = fs_info->sectorsize;
2373         device->sector_size = fs_info->sectorsize;
2374         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2375                                          fs_info->sectorsize);
2376         device->disk_total_bytes = device->total_bytes;
2377         device->commit_total_bytes = device->total_bytes;
2378         device->fs_info = fs_info;
2379         device->bdev = bdev;
2380         device->in_fs_metadata = 1;
2381         device->is_tgtdev_for_dev_replace = 0;
2382         device->mode = FMODE_EXCL;
2383         device->dev_stats_valid = 1;
2384         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2385
2386         if (seeding_dev) {
2387                 sb->s_flags &= ~SB_RDONLY;
2388                 ret = btrfs_prepare_sprout(fs_info);
2389                 if (ret) {
2390                         btrfs_abort_transaction(trans, ret);
2391                         goto error_trans;
2392                 }
2393         }
2394
2395         device->fs_devices = fs_info->fs_devices;
2396
2397         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2398         mutex_lock(&fs_info->chunk_mutex);
2399         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2400         list_add(&device->dev_alloc_list,
2401                  &fs_info->fs_devices->alloc_list);
2402         fs_info->fs_devices->num_devices++;
2403         fs_info->fs_devices->open_devices++;
2404         fs_info->fs_devices->rw_devices++;
2405         fs_info->fs_devices->total_devices++;
2406         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2407
2408         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2409
2410         if (!blk_queue_nonrot(q))
2411                 fs_info->fs_devices->rotating = 1;
2412
2413         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2414         btrfs_set_super_total_bytes(fs_info->super_copy,
2415                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2416
2417         tmp = btrfs_super_num_devices(fs_info->super_copy);
2418         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2419
2420         /* add sysfs device entry */
2421         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2422
2423         /*
2424          * we've got more storage, clear any full flags on the space
2425          * infos
2426          */
2427         btrfs_clear_space_info_full(fs_info);
2428
2429         mutex_unlock(&fs_info->chunk_mutex);
2430         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2431
2432         if (seeding_dev) {
2433                 mutex_lock(&fs_info->chunk_mutex);
2434                 ret = init_first_rw_device(trans, fs_info);
2435                 mutex_unlock(&fs_info->chunk_mutex);
2436                 if (ret) {
2437                         btrfs_abort_transaction(trans, ret);
2438                         goto error_sysfs;
2439                 }
2440         }
2441
2442         ret = btrfs_add_device(trans, fs_info, device);
2443         if (ret) {
2444                 btrfs_abort_transaction(trans, ret);
2445                 goto error_sysfs;
2446         }
2447
2448         if (seeding_dev) {
2449                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2450
2451                 ret = btrfs_finish_sprout(trans, fs_info);
2452                 if (ret) {
2453                         btrfs_abort_transaction(trans, ret);
2454                         goto error_sysfs;
2455                 }
2456
2457                 /* Sprouting would change fsid of the mounted root,
2458                  * so rename the fsid on the sysfs
2459                  */
2460                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2461                                                 fs_info->fsid);
2462                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2463                         btrfs_warn(fs_info,
2464                                    "sysfs: failed to create fsid for sprout");
2465         }
2466
2467         ret = btrfs_commit_transaction(trans);
2468
2469         if (seeding_dev) {
2470                 mutex_unlock(&uuid_mutex);
2471                 up_write(&sb->s_umount);
2472                 unlocked = true;
2473
2474                 if (ret) /* transaction commit */
2475                         return ret;
2476
2477                 ret = btrfs_relocate_sys_chunks(fs_info);
2478                 if (ret < 0)
2479                         btrfs_handle_fs_error(fs_info, ret,
2480                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2481                 trans = btrfs_attach_transaction(root);
2482                 if (IS_ERR(trans)) {
2483                         if (PTR_ERR(trans) == -ENOENT)
2484                                 return 0;
2485                         ret = PTR_ERR(trans);
2486                         trans = NULL;
2487                         goto error_sysfs;
2488                 }
2489                 ret = btrfs_commit_transaction(trans);
2490         }
2491
2492         /* Update ctime/mtime for libblkid */
2493         update_dev_time(device_path);
2494         return ret;
2495
2496 error_sysfs:
2497         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2498 error_trans:
2499         if (seeding_dev)
2500                 sb->s_flags |= SB_RDONLY;
2501         if (trans)
2502                 btrfs_end_transaction(trans);
2503         rcu_string_free(device->name);
2504         kfree(device);
2505 error:
2506         blkdev_put(bdev, FMODE_EXCL);
2507         if (seeding_dev && !unlocked) {
2508                 mutex_unlock(&uuid_mutex);
2509                 up_write(&sb->s_umount);
2510         }
2511         return ret;
2512 }
2513
2514 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2515                                   const char *device_path,
2516                                   struct btrfs_device *srcdev,
2517                                   struct btrfs_device **device_out)
2518 {
2519         struct request_queue *q;
2520         struct btrfs_device *device;
2521         struct block_device *bdev;
2522         struct list_head *devices;
2523         struct rcu_string *name;
2524         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2525         int ret = 0;
2526
2527         *device_out = NULL;
2528         if (fs_info->fs_devices->seeding) {
2529                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2530                 return -EINVAL;
2531         }
2532
2533         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2534                                   fs_info->bdev_holder);
2535         if (IS_ERR(bdev)) {
2536                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2537                 return PTR_ERR(bdev);
2538         }
2539
2540         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2541
2542         devices = &fs_info->fs_devices->devices;
2543         list_for_each_entry(device, devices, dev_list) {
2544                 if (device->bdev == bdev) {
2545                         btrfs_err(fs_info,
2546                                   "target device is in the filesystem!");
2547                         ret = -EEXIST;
2548                         goto error;
2549                 }
2550         }
2551
2552
2553         if (i_size_read(bdev->bd_inode) <
2554             btrfs_device_get_total_bytes(srcdev)) {
2555                 btrfs_err(fs_info,
2556                           "target device is smaller than source device!");
2557                 ret = -EINVAL;
2558                 goto error;
2559         }
2560
2561
2562         device = btrfs_alloc_device(NULL, &devid, NULL);
2563         if (IS_ERR(device)) {
2564                 ret = PTR_ERR(device);
2565                 goto error;
2566         }
2567
2568         name = rcu_string_strdup(device_path, GFP_KERNEL);
2569         if (!name) {
2570                 kfree(device);
2571                 ret = -ENOMEM;
2572                 goto error;
2573         }
2574         rcu_assign_pointer(device->name, name);
2575
2576         q = bdev_get_queue(bdev);
2577         if (blk_queue_discard(q))
2578                 device->can_discard = 1;
2579         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2580         device->writeable = 1;
2581         device->generation = 0;
2582         device->io_width = fs_info->sectorsize;
2583         device->io_align = fs_info->sectorsize;
2584         device->sector_size = fs_info->sectorsize;
2585         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2586         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2587         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2588         ASSERT(list_empty(&srcdev->resized_list));
2589         device->commit_total_bytes = srcdev->commit_total_bytes;
2590         device->commit_bytes_used = device->bytes_used;
2591         device->fs_info = fs_info;
2592         device->bdev = bdev;
2593         device->in_fs_metadata = 1;
2594         device->is_tgtdev_for_dev_replace = 1;
2595         device->mode = FMODE_EXCL;
2596         device->dev_stats_valid = 1;
2597         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2598         device->fs_devices = fs_info->fs_devices;
2599         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2600         fs_info->fs_devices->num_devices++;
2601         fs_info->fs_devices->open_devices++;
2602         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2603
2604         *device_out = device;
2605         return ret;
2606
2607 error:
2608         blkdev_put(bdev, FMODE_EXCL);
2609         return ret;
2610 }
2611
2612 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2613                                               struct btrfs_device *tgtdev)
2614 {
2615         u32 sectorsize = fs_info->sectorsize;
2616
2617         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2618         tgtdev->io_width = sectorsize;
2619         tgtdev->io_align = sectorsize;
2620         tgtdev->sector_size = sectorsize;
2621         tgtdev->fs_info = fs_info;
2622         tgtdev->in_fs_metadata = 1;
2623 }
2624
2625 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2626                                         struct btrfs_device *device)
2627 {
2628         int ret;
2629         struct btrfs_path *path;
2630         struct btrfs_root *root = device->fs_info->chunk_root;
2631         struct btrfs_dev_item *dev_item;
2632         struct extent_buffer *leaf;
2633         struct btrfs_key key;
2634
2635         path = btrfs_alloc_path();
2636         if (!path)
2637                 return -ENOMEM;
2638
2639         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2640         key.type = BTRFS_DEV_ITEM_KEY;
2641         key.offset = device->devid;
2642
2643         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2644         if (ret < 0)
2645                 goto out;
2646
2647         if (ret > 0) {
2648                 ret = -ENOENT;
2649                 goto out;
2650         }
2651
2652         leaf = path->nodes[0];
2653         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2654
2655         btrfs_set_device_id(leaf, dev_item, device->devid);
2656         btrfs_set_device_type(leaf, dev_item, device->type);
2657         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2658         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2659         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2660         btrfs_set_device_total_bytes(leaf, dev_item,
2661                                      btrfs_device_get_disk_total_bytes(device));
2662         btrfs_set_device_bytes_used(leaf, dev_item,
2663                                     btrfs_device_get_bytes_used(device));
2664         btrfs_mark_buffer_dirty(leaf);
2665
2666 out:
2667         btrfs_free_path(path);
2668         return ret;
2669 }
2670
2671 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2672                       struct btrfs_device *device, u64 new_size)
2673 {
2674         struct btrfs_fs_info *fs_info = device->fs_info;
2675         struct btrfs_super_block *super_copy = fs_info->super_copy;
2676         struct btrfs_fs_devices *fs_devices;
2677         u64 old_total;
2678         u64 diff;
2679
2680         if (!device->writeable)
2681                 return -EACCES;
2682
2683         new_size = round_down(new_size, fs_info->sectorsize);
2684
2685         mutex_lock(&fs_info->chunk_mutex);
2686         old_total = btrfs_super_total_bytes(super_copy);
2687         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2688
2689         if (new_size <= device->total_bytes ||
2690             device->is_tgtdev_for_dev_replace) {
2691                 mutex_unlock(&fs_info->chunk_mutex);
2692                 return -EINVAL;
2693         }
2694
2695         fs_devices = fs_info->fs_devices;
2696
2697         btrfs_set_super_total_bytes(super_copy,
2698                         round_down(old_total + diff, fs_info->sectorsize));
2699         device->fs_devices->total_rw_bytes += diff;
2700
2701         btrfs_device_set_total_bytes(device, new_size);
2702         btrfs_device_set_disk_total_bytes(device, new_size);
2703         btrfs_clear_space_info_full(device->fs_info);
2704         if (list_empty(&device->resized_list))
2705                 list_add_tail(&device->resized_list,
2706                               &fs_devices->resized_devices);
2707         mutex_unlock(&fs_info->chunk_mutex);
2708
2709         return btrfs_update_device(trans, device);
2710 }
2711
2712 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2713                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2714 {
2715         struct btrfs_root *root = fs_info->chunk_root;
2716         int ret;
2717         struct btrfs_path *path;
2718         struct btrfs_key key;
2719
2720         path = btrfs_alloc_path();
2721         if (!path)
2722                 return -ENOMEM;
2723
2724         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2725         key.offset = chunk_offset;
2726         key.type = BTRFS_CHUNK_ITEM_KEY;
2727
2728         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2729         if (ret < 0)
2730                 goto out;
2731         else if (ret > 0) { /* Logic error or corruption */
2732                 btrfs_handle_fs_error(fs_info, -ENOENT,
2733                                       "Failed lookup while freeing chunk.");
2734                 ret = -ENOENT;
2735                 goto out;
2736         }
2737
2738         ret = btrfs_del_item(trans, root, path);
2739         if (ret < 0)
2740                 btrfs_handle_fs_error(fs_info, ret,
2741                                       "Failed to delete chunk item.");
2742 out:
2743         btrfs_free_path(path);
2744         return ret;
2745 }
2746
2747 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2748 {
2749         struct btrfs_super_block *super_copy = fs_info->super_copy;
2750         struct btrfs_disk_key *disk_key;
2751         struct btrfs_chunk *chunk;
2752         u8 *ptr;
2753         int ret = 0;
2754         u32 num_stripes;
2755         u32 array_size;
2756         u32 len = 0;
2757         u32 cur;
2758         struct btrfs_key key;
2759
2760         mutex_lock(&fs_info->chunk_mutex);
2761         array_size = btrfs_super_sys_array_size(super_copy);
2762
2763         ptr = super_copy->sys_chunk_array;
2764         cur = 0;
2765
2766         while (cur < array_size) {
2767                 disk_key = (struct btrfs_disk_key *)ptr;
2768                 btrfs_disk_key_to_cpu(&key, disk_key);
2769
2770                 len = sizeof(*disk_key);
2771
2772                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2773                         chunk = (struct btrfs_chunk *)(ptr + len);
2774                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2775                         len += btrfs_chunk_item_size(num_stripes);
2776                 } else {
2777                         ret = -EIO;
2778                         break;
2779                 }
2780                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2781                     key.offset == chunk_offset) {
2782                         memmove(ptr, ptr + len, array_size - (cur + len));
2783                         array_size -= len;
2784                         btrfs_set_super_sys_array_size(super_copy, array_size);
2785                 } else {
2786                         ptr += len;
2787                         cur += len;
2788                 }
2789         }
2790         mutex_unlock(&fs_info->chunk_mutex);
2791         return ret;
2792 }
2793
2794 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2795                                         u64 logical, u64 length)
2796 {
2797         struct extent_map_tree *em_tree;
2798         struct extent_map *em;
2799
2800         em_tree = &fs_info->mapping_tree.map_tree;
2801         read_lock(&em_tree->lock);
2802         em = lookup_extent_mapping(em_tree, logical, length);
2803         read_unlock(&em_tree->lock);
2804
2805         if (!em) {
2806                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2807                            logical, length);
2808                 return ERR_PTR(-EINVAL);
2809         }
2810
2811         if (em->start > logical || em->start + em->len < logical) {
2812                 btrfs_crit(fs_info,
2813                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2814                            logical, length, em->start, em->start + em->len);
2815                 free_extent_map(em);
2816                 return ERR_PTR(-EINVAL);
2817         }
2818
2819         /* callers are responsible for dropping em's ref. */
2820         return em;
2821 }
2822
2823 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2824                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2825 {
2826         struct extent_map *em;
2827         struct map_lookup *map;
2828         u64 dev_extent_len = 0;
2829         int i, ret = 0;
2830         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2831
2832         em = get_chunk_map(fs_info, chunk_offset, 1);
2833         if (IS_ERR(em)) {
2834                 /*
2835                  * This is a logic error, but we don't want to just rely on the
2836                  * user having built with ASSERT enabled, so if ASSERT doesn't
2837                  * do anything we still error out.
2838                  */
2839                 ASSERT(0);
2840                 return PTR_ERR(em);
2841         }
2842         map = em->map_lookup;
2843         mutex_lock(&fs_info->chunk_mutex);
2844         check_system_chunk(trans, fs_info, map->type);
2845         mutex_unlock(&fs_info->chunk_mutex);
2846
2847         /*
2848          * Take the device list mutex to prevent races with the final phase of
2849          * a device replace operation that replaces the device object associated
2850          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2851          */
2852         mutex_lock(&fs_devices->device_list_mutex);
2853         for (i = 0; i < map->num_stripes; i++) {
2854                 struct btrfs_device *device = map->stripes[i].dev;
2855                 ret = btrfs_free_dev_extent(trans, device,
2856                                             map->stripes[i].physical,
2857                                             &dev_extent_len);
2858                 if (ret) {
2859                         mutex_unlock(&fs_devices->device_list_mutex);
2860                         btrfs_abort_transaction(trans, ret);
2861                         goto out;
2862                 }
2863
2864                 if (device->bytes_used > 0) {
2865                         mutex_lock(&fs_info->chunk_mutex);
2866                         btrfs_device_set_bytes_used(device,
2867                                         device->bytes_used - dev_extent_len);
2868                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2869                         btrfs_clear_space_info_full(fs_info);
2870                         mutex_unlock(&fs_info->chunk_mutex);
2871                 }
2872
2873                 if (map->stripes[i].dev) {
2874                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2875                         if (ret) {
2876                                 mutex_unlock(&fs_devices->device_list_mutex);
2877                                 btrfs_abort_transaction(trans, ret);
2878                                 goto out;
2879                         }
2880                 }
2881         }
2882         mutex_unlock(&fs_devices->device_list_mutex);
2883
2884         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2885         if (ret) {
2886                 btrfs_abort_transaction(trans, ret);
2887                 goto out;
2888         }
2889
2890         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2891
2892         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2893                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2894                 if (ret) {
2895                         btrfs_abort_transaction(trans, ret);
2896                         goto out;
2897                 }
2898         }
2899
2900         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2901         if (ret) {
2902                 btrfs_abort_transaction(trans, ret);
2903                 goto out;
2904         }
2905
2906 out:
2907         /* once for us */
2908         free_extent_map(em);
2909         return ret;
2910 }
2911
2912 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2913 {
2914         struct btrfs_root *root = fs_info->chunk_root;
2915         struct btrfs_trans_handle *trans;
2916         int ret;
2917
2918         /*
2919          * Prevent races with automatic removal of unused block groups.
2920          * After we relocate and before we remove the chunk with offset
2921          * chunk_offset, automatic removal of the block group can kick in,
2922          * resulting in a failure when calling btrfs_remove_chunk() below.
2923          *
2924          * Make sure to acquire this mutex before doing a tree search (dev
2925          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2926          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2927          * we release the path used to search the chunk/dev tree and before
2928          * the current task acquires this mutex and calls us.
2929          */
2930         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2931
2932         ret = btrfs_can_relocate(fs_info, chunk_offset);
2933         if (ret)
2934                 return -ENOSPC;
2935
2936         /* step one, relocate all the extents inside this chunk */
2937         btrfs_scrub_pause(fs_info);
2938         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2939         btrfs_scrub_continue(fs_info);
2940         if (ret)
2941                 return ret;
2942
2943         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2944                                                      chunk_offset);
2945         if (IS_ERR(trans)) {
2946                 ret = PTR_ERR(trans);
2947                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2948                 return ret;
2949         }
2950
2951         /*
2952          * step two, delete the device extents and the
2953          * chunk tree entries
2954          */
2955         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2956         btrfs_end_transaction(trans);
2957         return ret;
2958 }
2959
2960 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2961 {
2962         struct btrfs_root *chunk_root = fs_info->chunk_root;
2963         struct btrfs_path *path;
2964         struct extent_buffer *leaf;
2965         struct btrfs_chunk *chunk;
2966         struct btrfs_key key;
2967         struct btrfs_key found_key;
2968         u64 chunk_type;
2969         bool retried = false;
2970         int failed = 0;
2971         int ret;
2972
2973         path = btrfs_alloc_path();
2974         if (!path)
2975                 return -ENOMEM;
2976
2977 again:
2978         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2979         key.offset = (u64)-1;
2980         key.type = BTRFS_CHUNK_ITEM_KEY;
2981
2982         while (1) {
2983                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2984                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2985                 if (ret < 0) {
2986                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2987                         goto error;
2988                 }
2989                 BUG_ON(ret == 0); /* Corruption */
2990
2991                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2992                                           key.type);
2993                 if (ret)
2994                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2995                 if (ret < 0)
2996                         goto error;
2997                 if (ret > 0)
2998                         break;
2999
3000                 leaf = path->nodes[0];
3001                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3002
3003                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3004                                        struct btrfs_chunk);
3005                 chunk_type = btrfs_chunk_type(leaf, chunk);
3006                 btrfs_release_path(path);
3007
3008                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3009                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3010                         if (ret == -ENOSPC)
3011                                 failed++;
3012                         else
3013                                 BUG_ON(ret);
3014                 }
3015                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3016
3017                 if (found_key.offset == 0)
3018                         break;
3019                 key.offset = found_key.offset - 1;
3020         }
3021         ret = 0;
3022         if (failed && !retried) {
3023                 failed = 0;
3024                 retried = true;
3025                 goto again;
3026         } else if (WARN_ON(failed && retried)) {
3027                 ret = -ENOSPC;
3028         }
3029 error:
3030         btrfs_free_path(path);
3031         return ret;
3032 }
3033
3034 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3035                                struct btrfs_balance_control *bctl)
3036 {
3037         struct btrfs_root *root = fs_info->tree_root;
3038         struct btrfs_trans_handle *trans;
3039         struct btrfs_balance_item *item;
3040         struct btrfs_disk_balance_args disk_bargs;
3041         struct btrfs_path *path;
3042         struct extent_buffer *leaf;
3043         struct btrfs_key key;
3044         int ret, err;
3045
3046         path = btrfs_alloc_path();
3047         if (!path)
3048                 return -ENOMEM;
3049
3050         trans = btrfs_start_transaction(root, 0);
3051         if (IS_ERR(trans)) {
3052                 btrfs_free_path(path);
3053                 return PTR_ERR(trans);
3054         }
3055
3056         key.objectid = BTRFS_BALANCE_OBJECTID;
3057         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3058         key.offset = 0;
3059
3060         ret = btrfs_insert_empty_item(trans, root, path, &key,
3061                                       sizeof(*item));
3062         if (ret)
3063                 goto out;
3064
3065         leaf = path->nodes[0];
3066         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3067
3068         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3069
3070         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3071         btrfs_set_balance_data(leaf, item, &disk_bargs);
3072         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3073         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3074         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3075         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3076
3077         btrfs_set_balance_flags(leaf, item, bctl->flags);
3078
3079         btrfs_mark_buffer_dirty(leaf);
3080 out:
3081         btrfs_free_path(path);
3082         err = btrfs_commit_transaction(trans);
3083         if (err && !ret)
3084                 ret = err;
3085         return ret;
3086 }
3087
3088 static int del_balance_item(struct btrfs_fs_info *fs_info)
3089 {
3090         struct btrfs_root *root = fs_info->tree_root;
3091         struct btrfs_trans_handle *trans;
3092         struct btrfs_path *path;
3093         struct btrfs_key key;
3094         int ret, err;
3095
3096         path = btrfs_alloc_path();
3097         if (!path)
3098                 return -ENOMEM;
3099
3100         trans = btrfs_start_transaction(root, 0);
3101         if (IS_ERR(trans)) {
3102                 btrfs_free_path(path);
3103                 return PTR_ERR(trans);
3104         }
3105
3106         key.objectid = BTRFS_BALANCE_OBJECTID;
3107         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3108         key.offset = 0;
3109
3110         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3111         if (ret < 0)
3112                 goto out;
3113         if (ret > 0) {
3114                 ret = -ENOENT;
3115                 goto out;
3116         }
3117
3118         ret = btrfs_del_item(trans, root, path);
3119 out:
3120         btrfs_free_path(path);
3121         err = btrfs_commit_transaction(trans);
3122         if (err && !ret)
3123                 ret = err;
3124         return ret;
3125 }
3126
3127 /*
3128  * This is a heuristic used to reduce the number of chunks balanced on
3129  * resume after balance was interrupted.
3130  */
3131 static void update_balance_args(struct btrfs_balance_control *bctl)
3132 {
3133         /*
3134          * Turn on soft mode for chunk types that were being converted.
3135          */
3136         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3140         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3141                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3142
3143         /*
3144          * Turn on usage filter if is not already used.  The idea is
3145          * that chunks that we have already balanced should be
3146          * reasonably full.  Don't do it for chunks that are being
3147          * converted - that will keep us from relocating unconverted
3148          * (albeit full) chunks.
3149          */
3150         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3151             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3152             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3153                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3154                 bctl->data.usage = 90;
3155         }
3156         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3157             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3158             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3159                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3160                 bctl->sys.usage = 90;
3161         }
3162         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3163             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3164             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3165                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3166                 bctl->meta.usage = 90;
3167         }
3168 }
3169
3170 /*
3171  * Should be called with both balance and volume mutexes held to
3172  * serialize other volume operations (add_dev/rm_dev/resize) with
3173  * restriper.  Same goes for unset_balance_control.
3174  */
3175 static void set_balance_control(struct btrfs_balance_control *bctl)
3176 {
3177         struct btrfs_fs_info *fs_info = bctl->fs_info;
3178
3179         BUG_ON(fs_info->balance_ctl);
3180
3181         spin_lock(&fs_info->balance_lock);
3182         fs_info->balance_ctl = bctl;
3183         spin_unlock(&fs_info->balance_lock);
3184 }
3185
3186 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3187 {
3188         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3189
3190         BUG_ON(!fs_info->balance_ctl);
3191
3192         spin_lock(&fs_info->balance_lock);
3193         fs_info->balance_ctl = NULL;
3194         spin_unlock(&fs_info->balance_lock);
3195
3196         kfree(bctl);
3197 }
3198
3199 /*
3200  * Balance filters.  Return 1 if chunk should be filtered out
3201  * (should not be balanced).
3202  */
3203 static int chunk_profiles_filter(u64 chunk_type,
3204                                  struct btrfs_balance_args *bargs)
3205 {
3206         chunk_type = chunk_to_extended(chunk_type) &
3207                                 BTRFS_EXTENDED_PROFILE_MASK;
3208
3209         if (bargs->profiles & chunk_type)
3210                 return 0;
3211
3212         return 1;
3213 }
3214
3215 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3216                               struct btrfs_balance_args *bargs)
3217 {
3218         struct btrfs_block_group_cache *cache;
3219         u64 chunk_used;
3220         u64 user_thresh_min;
3221         u64 user_thresh_max;
3222         int ret = 1;
3223
3224         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3225         chunk_used = btrfs_block_group_used(&cache->item);
3226
3227         if (bargs->usage_min == 0)
3228                 user_thresh_min = 0;
3229         else
3230                 user_thresh_min = div_factor_fine(cache->key.offset,
3231                                         bargs->usage_min);
3232
3233         if (bargs->usage_max == 0)
3234                 user_thresh_max = 1;
3235         else if (bargs->usage_max > 100)
3236                 user_thresh_max = cache->key.offset;
3237         else
3238                 user_thresh_max = div_factor_fine(cache->key.offset,
3239                                         bargs->usage_max);
3240