b2d05c6b1c5672a638ba8752333c10508e366885
[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 /*
149  * Device locking
150  * ==============
151  *
152  * There are several mutexes that protect manipulation of devices and low-level
153  * structures like chunks but not block groups, extents or files
154  *
155  * uuid_mutex (global lock)
156  * ------------------------
157  * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
158  * the SCAN_DEV ioctl registration or from mount either implicitly (the first
159  * device) or requested by the device= mount option
160  *
161  * the mutex can be very coarse and can cover long-running operations
162  *
163  * protects: updates to fs_devices counters like missing devices, rw devices,
164  * seeding, structure cloning, openning/closing devices at mount/umount time
165  *
166  * global::fs_devs - add, remove, updates to the global list
167  *
168  * does not protect: manipulation of the fs_devices::devices list!
169  *
170  * btrfs_device::name - renames (write side), read is RCU
171  *
172  * fs_devices::device_list_mutex (per-fs, with RCU)
173  * ------------------------------------------------
174  * protects updates to fs_devices::devices, ie. adding and deleting
175  *
176  * simple list traversal with read-only actions can be done with RCU protection
177  *
178  * may be used to exclude some operations from running concurrently without any
179  * modifications to the list (see write_all_supers)
180  *
181  * volume_mutex
182  * ------------
183  * coarse lock owned by a mounted filesystem; used to exclude some operations
184  * that cannot run in parallel and affect the higher-level properties of the
185  * filesystem like: device add/deleting/resize/replace, or balance
186  *
187  * balance_mutex
188  * -------------
189  * protects balance structures (status, state) and context accessed from
190  * several places (internally, ioctl)
191  *
192  * chunk_mutex
193  * -----------
194  * protects chunks, adding or removing during allocation, trim or when a new
195  * device is added/removed
196  *
197  * cleaner_mutex
198  * -------------
199  * a big lock that is held by the cleaner thread and prevents running subvolume
200  * cleaning together with relocation or delayed iputs
201  *
202  *
203  * Lock nesting
204  * ============
205  *
206  * uuid_mutex
207  *   volume_mutex
208  *     device_list_mutex
209  *       chunk_mutex
210  *     balance_mutex
211  */
212
213 DEFINE_MUTEX(uuid_mutex);
214 static LIST_HEAD(fs_uuids);
215 struct list_head *btrfs_get_fs_uuids(void)
216 {
217         return &fs_uuids;
218 }
219
220 /*
221  * alloc_fs_devices - allocate struct btrfs_fs_devices
222  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
223  *
224  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
225  * The returned struct is not linked onto any lists and can be destroyed with
226  * kfree() right away.
227  */
228 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
229 {
230         struct btrfs_fs_devices *fs_devs;
231
232         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
233         if (!fs_devs)
234                 return ERR_PTR(-ENOMEM);
235
236         mutex_init(&fs_devs->device_list_mutex);
237
238         INIT_LIST_HEAD(&fs_devs->devices);
239         INIT_LIST_HEAD(&fs_devs->resized_devices);
240         INIT_LIST_HEAD(&fs_devs->alloc_list);
241         INIT_LIST_HEAD(&fs_devs->list);
242         if (fsid)
243                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
244
245         return fs_devs;
246 }
247
248 static void free_device(struct btrfs_device *device)
249 {
250         rcu_string_free(device->name);
251         bio_put(device->flush_bio);
252         kfree(device);
253 }
254
255 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
256 {
257         struct btrfs_device *device;
258         WARN_ON(fs_devices->opened);
259         while (!list_empty(&fs_devices->devices)) {
260                 device = list_entry(fs_devices->devices.next,
261                                     struct btrfs_device, dev_list);
262                 list_del(&device->dev_list);
263                 free_device(device);
264         }
265         kfree(fs_devices);
266 }
267
268 static void btrfs_kobject_uevent(struct block_device *bdev,
269                                  enum kobject_action action)
270 {
271         int ret;
272
273         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
274         if (ret)
275                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
276                         action,
277                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
278                         &disk_to_dev(bdev->bd_disk)->kobj);
279 }
280
281 void btrfs_cleanup_fs_uuids(void)
282 {
283         struct btrfs_fs_devices *fs_devices;
284
285         while (!list_empty(&fs_uuids)) {
286                 fs_devices = list_entry(fs_uuids.next,
287                                         struct btrfs_fs_devices, list);
288                 list_del(&fs_devices->list);
289                 free_fs_devices(fs_devices);
290         }
291 }
292
293 /*
294  * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
295  * Returned struct is not linked onto any lists and must be destroyed using
296  * free_device.
297  */
298 static struct btrfs_device *__alloc_device(void)
299 {
300         struct btrfs_device *dev;
301
302         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
303         if (!dev)
304                 return ERR_PTR(-ENOMEM);
305
306         /*
307          * Preallocate a bio that's always going to be used for flushing device
308          * barriers and matches the device lifespan
309          */
310         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
311         if (!dev->flush_bio) {
312                 kfree(dev);
313                 return ERR_PTR(-ENOMEM);
314         }
315
316         INIT_LIST_HEAD(&dev->dev_list);
317         INIT_LIST_HEAD(&dev->dev_alloc_list);
318         INIT_LIST_HEAD(&dev->resized_list);
319
320         spin_lock_init(&dev->io_lock);
321
322         atomic_set(&dev->reada_in_flight, 0);
323         atomic_set(&dev->dev_stats_ccnt, 0);
324         btrfs_device_data_ordered_init(dev);
325         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
326         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
327
328         return dev;
329 }
330
331 /*
332  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
333  * return NULL.
334  *
335  * If devid and uuid are both specified, the match must be exact, otherwise
336  * only devid is used.
337  */
338 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
339                 u64 devid, const u8 *uuid)
340 {
341         struct list_head *head = &fs_devices->devices;
342         struct btrfs_device *dev;
343
344         list_for_each_entry(dev, head, dev_list) {
345                 if (dev->devid == devid &&
346                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
347                         return dev;
348                 }
349         }
350         return NULL;
351 }
352
353 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
354 {
355         struct btrfs_fs_devices *fs_devices;
356
357         list_for_each_entry(fs_devices, &fs_uuids, list) {
358                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
359                         return fs_devices;
360         }
361         return NULL;
362 }
363
364 static int
365 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
366                       int flush, struct block_device **bdev,
367                       struct buffer_head **bh)
368 {
369         int ret;
370
371         *bdev = blkdev_get_by_path(device_path, flags, holder);
372
373         if (IS_ERR(*bdev)) {
374                 ret = PTR_ERR(*bdev);
375                 goto error;
376         }
377
378         if (flush)
379                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
380         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
381         if (ret) {
382                 blkdev_put(*bdev, flags);
383                 goto error;
384         }
385         invalidate_bdev(*bdev);
386         *bh = btrfs_read_dev_super(*bdev);
387         if (IS_ERR(*bh)) {
388                 ret = PTR_ERR(*bh);
389                 blkdev_put(*bdev, flags);
390                 goto error;
391         }
392
393         return 0;
394
395 error:
396         *bdev = NULL;
397         *bh = NULL;
398         return ret;
399 }
400
401 static void requeue_list(struct btrfs_pending_bios *pending_bios,
402                         struct bio *head, struct bio *tail)
403 {
404
405         struct bio *old_head;
406
407         old_head = pending_bios->head;
408         pending_bios->head = head;
409         if (pending_bios->tail)
410                 tail->bi_next = old_head;
411         else
412                 pending_bios->tail = tail;
413 }
414
415 /*
416  * we try to collect pending bios for a device so we don't get a large
417  * number of procs sending bios down to the same device.  This greatly
418  * improves the schedulers ability to collect and merge the bios.
419  *
420  * But, it also turns into a long list of bios to process and that is sure
421  * to eventually make the worker thread block.  The solution here is to
422  * make some progress and then put this work struct back at the end of
423  * the list if the block device is congested.  This way, multiple devices
424  * can make progress from a single worker thread.
425  */
426 static noinline void run_scheduled_bios(struct btrfs_device *device)
427 {
428         struct btrfs_fs_info *fs_info = device->fs_info;
429         struct bio *pending;
430         struct backing_dev_info *bdi;
431         struct btrfs_pending_bios *pending_bios;
432         struct bio *tail;
433         struct bio *cur;
434         int again = 0;
435         unsigned long num_run;
436         unsigned long batch_run = 0;
437         unsigned long last_waited = 0;
438         int force_reg = 0;
439         int sync_pending = 0;
440         struct blk_plug plug;
441
442         /*
443          * this function runs all the bios we've collected for
444          * a particular device.  We don't want to wander off to
445          * another device without first sending all of these down.
446          * So, setup a plug here and finish it off before we return
447          */
448         blk_start_plug(&plug);
449
450         bdi = device->bdev->bd_bdi;
451
452 loop:
453         spin_lock(&device->io_lock);
454
455 loop_lock:
456         num_run = 0;
457
458         /* take all the bios off the list at once and process them
459          * later on (without the lock held).  But, remember the
460          * tail and other pointers so the bios can be properly reinserted
461          * into the list if we hit congestion
462          */
463         if (!force_reg && device->pending_sync_bios.head) {
464                 pending_bios = &device->pending_sync_bios;
465                 force_reg = 1;
466         } else {
467                 pending_bios = &device->pending_bios;
468                 force_reg = 0;
469         }
470
471         pending = pending_bios->head;
472         tail = pending_bios->tail;
473         WARN_ON(pending && !tail);
474
475         /*
476          * if pending was null this time around, no bios need processing
477          * at all and we can stop.  Otherwise it'll loop back up again
478          * and do an additional check so no bios are missed.
479          *
480          * device->running_pending is used to synchronize with the
481          * schedule_bio code.
482          */
483         if (device->pending_sync_bios.head == NULL &&
484             device->pending_bios.head == NULL) {
485                 again = 0;
486                 device->running_pending = 0;
487         } else {
488                 again = 1;
489                 device->running_pending = 1;
490         }
491
492         pending_bios->head = NULL;
493         pending_bios->tail = NULL;
494
495         spin_unlock(&device->io_lock);
496
497         while (pending) {
498
499                 rmb();
500                 /* we want to work on both lists, but do more bios on the
501                  * sync list than the regular list
502                  */
503                 if ((num_run > 32 &&
504                     pending_bios != &device->pending_sync_bios &&
505                     device->pending_sync_bios.head) ||
506                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
507                     device->pending_bios.head)) {
508                         spin_lock(&device->io_lock);
509                         requeue_list(pending_bios, pending, tail);
510                         goto loop_lock;
511                 }
512
513                 cur = pending;
514                 pending = pending->bi_next;
515                 cur->bi_next = NULL;
516
517                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
518
519                 /*
520                  * if we're doing the sync list, record that our
521                  * plug has some sync requests on it
522                  *
523                  * If we're doing the regular list and there are
524                  * sync requests sitting around, unplug before
525                  * we add more
526                  */
527                 if (pending_bios == &device->pending_sync_bios) {
528                         sync_pending = 1;
529                 } else if (sync_pending) {
530                         blk_finish_plug(&plug);
531                         blk_start_plug(&plug);
532                         sync_pending = 0;
533                 }
534
535                 btrfsic_submit_bio(cur);
536                 num_run++;
537                 batch_run++;
538
539                 cond_resched();
540
541                 /*
542                  * we made progress, there is more work to do and the bdi
543                  * is now congested.  Back off and let other work structs
544                  * run instead
545                  */
546                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
547                     fs_info->fs_devices->open_devices > 1) {
548                         struct io_context *ioc;
549
550                         ioc = current->io_context;
551
552                         /*
553                          * the main goal here is that we don't want to
554                          * block if we're going to be able to submit
555                          * more requests without blocking.
556                          *
557                          * This code does two great things, it pokes into
558                          * the elevator code from a filesystem _and_
559                          * it makes assumptions about how batching works.
560                          */
561                         if (ioc && ioc->nr_batch_requests > 0 &&
562                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
563                             (last_waited == 0 ||
564                              ioc->last_waited == last_waited)) {
565                                 /*
566                                  * we want to go through our batch of
567                                  * requests and stop.  So, we copy out
568                                  * the ioc->last_waited time and test
569                                  * against it before looping
570                                  */
571                                 last_waited = ioc->last_waited;
572                                 cond_resched();
573                                 continue;
574                         }
575                         spin_lock(&device->io_lock);
576                         requeue_list(pending_bios, pending, tail);
577                         device->running_pending = 1;
578
579                         spin_unlock(&device->io_lock);
580                         btrfs_queue_work(fs_info->submit_workers,
581                                          &device->work);
582                         goto done;
583                 }
584         }
585
586         cond_resched();
587         if (again)
588                 goto loop;
589
590         spin_lock(&device->io_lock);
591         if (device->pending_bios.head || device->pending_sync_bios.head)
592                 goto loop_lock;
593         spin_unlock(&device->io_lock);
594
595 done:
596         blk_finish_plug(&plug);
597 }
598
599 static void pending_bios_fn(struct btrfs_work *work)
600 {
601         struct btrfs_device *device;
602
603         device = container_of(work, struct btrfs_device, work);
604         run_scheduled_bios(device);
605 }
606
607 /*
608  *  Search and remove all stale (devices which are not mounted) devices.
609  *  When both inputs are NULL, it will search and release all stale devices.
610  *  path:       Optional. When provided will it release all unmounted devices
611  *              matching this path only.
612  *  skip_dev:   Optional. Will skip this device when searching for the stale
613  *              devices.
614  */
615 static void btrfs_free_stale_devices(const char *path,
616                                      struct btrfs_device *skip_dev)
617 {
618         struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
619         struct btrfs_device *dev, *tmp_dev;
620
621         list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, list) {
622
623                 if (fs_devs->opened)
624                         continue;
625
626                 list_for_each_entry_safe(dev, tmp_dev,
627                                          &fs_devs->devices, dev_list) {
628                         int not_found = 0;
629
630                         if (skip_dev && skip_dev == dev)
631                                 continue;
632                         if (path && !dev->name)
633                                 continue;
634
635                         rcu_read_lock();
636                         if (path)
637                                 not_found = strcmp(rcu_str_deref(dev->name),
638                                                    path);
639                         rcu_read_unlock();
640                         if (not_found)
641                                 continue;
642
643                         /* delete the stale device */
644                         if (fs_devs->num_devices == 1) {
645                                 btrfs_sysfs_remove_fsid(fs_devs);
646                                 list_del(&fs_devs->list);
647                                 free_fs_devices(fs_devs);
648                                 break;
649                         } else {
650                                 fs_devs->num_devices--;
651                                 list_del(&dev->dev_list);
652                                 free_device(dev);
653                         }
654                 }
655         }
656 }
657
658 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
659                         struct btrfs_device *device, fmode_t flags,
660                         void *holder)
661 {
662         struct request_queue *q;
663         struct block_device *bdev;
664         struct buffer_head *bh;
665         struct btrfs_super_block *disk_super;
666         u64 devid;
667         int ret;
668
669         if (device->bdev)
670                 return -EINVAL;
671         if (!device->name)
672                 return -EINVAL;
673
674         ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
675                                     &bdev, &bh);
676         if (ret)
677                 return ret;
678
679         disk_super = (struct btrfs_super_block *)bh->b_data;
680         devid = btrfs_stack_device_id(&disk_super->dev_item);
681         if (devid != device->devid)
682                 goto error_brelse;
683
684         if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
685                 goto error_brelse;
686
687         device->generation = btrfs_super_generation(disk_super);
688
689         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
690                 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
691                 fs_devices->seeding = 1;
692         } else {
693                 if (bdev_read_only(bdev))
694                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
695                 else
696                         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
697         }
698
699         q = bdev_get_queue(bdev);
700         if (!blk_queue_nonrot(q))
701                 fs_devices->rotating = 1;
702
703         device->bdev = bdev;
704         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
705         device->mode = flags;
706
707         fs_devices->open_devices++;
708         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
709             device->devid != BTRFS_DEV_REPLACE_DEVID) {
710                 fs_devices->rw_devices++;
711                 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
712         }
713         brelse(bh);
714
715         return 0;
716
717 error_brelse:
718         brelse(bh);
719         blkdev_put(bdev, flags);
720
721         return -EINVAL;
722 }
723
724 /*
725  * Add new device to list of registered devices
726  *
727  * Returns:
728  * device pointer which was just added or updated when successful
729  * error pointer when failed
730  */
731 static noinline struct btrfs_device *device_list_add(const char *path,
732                            struct btrfs_super_block *disk_super)
733 {
734         struct btrfs_device *device;
735         struct btrfs_fs_devices *fs_devices;
736         struct rcu_string *name;
737         u64 found_transid = btrfs_super_generation(disk_super);
738         u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
739
740         fs_devices = find_fsid(disk_super->fsid);
741         if (!fs_devices) {
742                 fs_devices = alloc_fs_devices(disk_super->fsid);
743                 if (IS_ERR(fs_devices))
744                         return ERR_CAST(fs_devices);
745
746                 list_add(&fs_devices->list, &fs_uuids);
747
748                 device = NULL;
749         } else {
750                 device = find_device(fs_devices, devid,
751                                 disk_super->dev_item.uuid);
752         }
753
754         if (!device) {
755                 if (fs_devices->opened)
756                         return ERR_PTR(-EBUSY);
757
758                 device = btrfs_alloc_device(NULL, &devid,
759                                             disk_super->dev_item.uuid);
760                 if (IS_ERR(device)) {
761                         /* we can safely leave the fs_devices entry around */
762                         return device;
763                 }
764
765                 name = rcu_string_strdup(path, GFP_NOFS);
766                 if (!name) {
767                         free_device(device);
768                         return ERR_PTR(-ENOMEM);
769                 }
770                 rcu_assign_pointer(device->name, name);
771
772                 mutex_lock(&fs_devices->device_list_mutex);
773                 list_add_rcu(&device->dev_list, &fs_devices->devices);
774                 fs_devices->num_devices++;
775                 mutex_unlock(&fs_devices->device_list_mutex);
776
777                 device->fs_devices = fs_devices;
778                 btrfs_free_stale_devices(path, device);
779
780                 if (disk_super->label[0])
781                         pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
782                                 disk_super->label, devid, found_transid, path);
783                 else
784                         pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
785                                 disk_super->fsid, devid, found_transid, path);
786
787         } else if (!device->name || strcmp(device->name->str, path)) {
788                 /*
789                  * When FS is already mounted.
790                  * 1. If you are here and if the device->name is NULL that
791                  *    means this device was missing at time of FS mount.
792                  * 2. If you are here and if the device->name is different
793                  *    from 'path' that means either
794                  *      a. The same device disappeared and reappeared with
795                  *         different name. or
796                  *      b. The missing-disk-which-was-replaced, has
797                  *         reappeared now.
798                  *
799                  * We must allow 1 and 2a above. But 2b would be a spurious
800                  * and unintentional.
801                  *
802                  * Further in case of 1 and 2a above, the disk at 'path'
803                  * would have missed some transaction when it was away and
804                  * in case of 2a the stale bdev has to be updated as well.
805                  * 2b must not be allowed at all time.
806                  */
807
808                 /*
809                  * For now, we do allow update to btrfs_fs_device through the
810                  * btrfs dev scan cli after FS has been mounted.  We're still
811                  * tracking a problem where systems fail mount by subvolume id
812                  * when we reject replacement on a mounted FS.
813                  */
814                 if (!fs_devices->opened && found_transid < device->generation) {
815                         /*
816                          * That is if the FS is _not_ mounted and if you
817                          * are here, that means there is more than one
818                          * disk with same uuid and devid.We keep the one
819                          * with larger generation number or the last-in if
820                          * generation are equal.
821                          */
822                         return ERR_PTR(-EEXIST);
823                 }
824
825                 name = rcu_string_strdup(path, GFP_NOFS);
826                 if (!name)
827                         return ERR_PTR(-ENOMEM);
828                 rcu_string_free(device->name);
829                 rcu_assign_pointer(device->name, name);
830                 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
831                         fs_devices->missing_devices--;
832                         clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
833                 }
834         }
835
836         /*
837          * Unmount does not free the btrfs_device struct but would zero
838          * generation along with most of the other members. So just update
839          * it back. We need it to pick the disk with largest generation
840          * (as above).
841          */
842         if (!fs_devices->opened)
843                 device->generation = found_transid;
844
845         fs_devices->total_devices = btrfs_super_num_devices(disk_super);
846
847         return device;
848 }
849
850 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
851 {
852         struct btrfs_fs_devices *fs_devices;
853         struct btrfs_device *device;
854         struct btrfs_device *orig_dev;
855
856         fs_devices = alloc_fs_devices(orig->fsid);
857         if (IS_ERR(fs_devices))
858                 return fs_devices;
859
860         mutex_lock(&orig->device_list_mutex);
861         fs_devices->total_devices = orig->total_devices;
862
863         /* We have held the volume lock, it is safe to get the devices. */
864         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
865                 struct rcu_string *name;
866
867                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
868                                             orig_dev->uuid);
869                 if (IS_ERR(device))
870                         goto error;
871
872                 /*
873                  * This is ok to do without rcu read locked because we hold the
874                  * uuid mutex so nothing we touch in here is going to disappear.
875                  */
876                 if (orig_dev->name) {
877                         name = rcu_string_strdup(orig_dev->name->str,
878                                         GFP_KERNEL);
879                         if (!name) {
880                                 free_device(device);
881                                 goto error;
882                         }
883                         rcu_assign_pointer(device->name, name);
884                 }
885
886                 list_add(&device->dev_list, &fs_devices->devices);
887                 device->fs_devices = fs_devices;
888                 fs_devices->num_devices++;
889         }
890         mutex_unlock(&orig->device_list_mutex);
891         return fs_devices;
892 error:
893         mutex_unlock(&orig->device_list_mutex);
894         free_fs_devices(fs_devices);
895         return ERR_PTR(-ENOMEM);
896 }
897
898 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
899 {
900         struct btrfs_device *device, *next;
901         struct btrfs_device *latest_dev = NULL;
902
903         mutex_lock(&uuid_mutex);
904 again:
905         /* This is the initialized path, it is safe to release the devices. */
906         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
907                 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
908                                                         &device->dev_state)) {
909                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
910                              &device->dev_state) &&
911                              (!latest_dev ||
912                               device->generation > latest_dev->generation)) {
913                                 latest_dev = device;
914                         }
915                         continue;
916                 }
917
918                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
919                         /*
920                          * In the first step, keep the device which has
921                          * the correct fsid and the devid that is used
922                          * for the dev_replace procedure.
923                          * In the second step, the dev_replace state is
924                          * read from the device tree and it is known
925                          * whether the procedure is really active or
926                          * not, which means whether this device is
927                          * used or whether it should be removed.
928                          */
929                         if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
930                                                   &device->dev_state)) {
931                                 continue;
932                         }
933                 }
934                 if (device->bdev) {
935                         blkdev_put(device->bdev, device->mode);
936                         device->bdev = NULL;
937                         fs_devices->open_devices--;
938                 }
939                 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
940                         list_del_init(&device->dev_alloc_list);
941                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
942                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
943                                       &device->dev_state))
944                                 fs_devices->rw_devices--;
945                 }
946                 list_del_init(&device->dev_list);
947                 fs_devices->num_devices--;
948                 free_device(device);
949         }
950
951         if (fs_devices->seed) {
952                 fs_devices = fs_devices->seed;
953                 goto again;
954         }
955
956         fs_devices->latest_bdev = latest_dev->bdev;
957
958         mutex_unlock(&uuid_mutex);
959 }
960
961 static void free_device_rcu(struct rcu_head *head)
962 {
963         struct btrfs_device *device;
964
965         device = container_of(head, struct btrfs_device, rcu);
966         free_device(device);
967 }
968
969 static void btrfs_close_bdev(struct btrfs_device *device)
970 {
971         if (!device->bdev)
972                 return;
973
974         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
975                 sync_blockdev(device->bdev);
976                 invalidate_bdev(device->bdev);
977         }
978
979         blkdev_put(device->bdev, device->mode);
980 }
981
982 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
983 {
984         struct btrfs_fs_devices *fs_devices = device->fs_devices;
985         struct btrfs_device *new_device;
986         struct rcu_string *name;
987
988         if (device->bdev)
989                 fs_devices->open_devices--;
990
991         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
992             device->devid != BTRFS_DEV_REPLACE_DEVID) {
993                 list_del_init(&device->dev_alloc_list);
994                 fs_devices->rw_devices--;
995         }
996
997         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
998                 fs_devices->missing_devices--;
999
1000         new_device = btrfs_alloc_device(NULL, &device->devid,
1001                                         device->uuid);
1002         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1003
1004         /* Safe because we are under uuid_mutex */
1005         if (device->name) {
1006                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1007                 BUG_ON(!name); /* -ENOMEM */
1008                 rcu_assign_pointer(new_device->name, name);
1009         }
1010
1011         list_replace_rcu(&device->dev_list, &new_device->dev_list);
1012         new_device->fs_devices = device->fs_devices;
1013 }
1014
1015 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1016 {
1017         struct btrfs_device *device, *tmp;
1018         struct list_head pending_put;
1019
1020         INIT_LIST_HEAD(&pending_put);
1021
1022         if (--fs_devices->opened > 0)
1023                 return 0;
1024
1025         mutex_lock(&fs_devices->device_list_mutex);
1026         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1027                 btrfs_prepare_close_one_device(device);
1028                 list_add(&device->dev_list, &pending_put);
1029         }
1030         mutex_unlock(&fs_devices->device_list_mutex);
1031
1032         /*
1033          * btrfs_show_devname() is using the device_list_mutex,
1034          * sometimes call to blkdev_put() leads vfs calling
1035          * into this func. So do put outside of device_list_mutex,
1036          * as of now.
1037          */
1038         while (!list_empty(&pending_put)) {
1039                 device = list_first_entry(&pending_put,
1040                                 struct btrfs_device, dev_list);
1041                 list_del(&device->dev_list);
1042                 btrfs_close_bdev(device);
1043                 call_rcu(&device->rcu, free_device_rcu);
1044         }
1045
1046         WARN_ON(fs_devices->open_devices);
1047         WARN_ON(fs_devices->rw_devices);
1048         fs_devices->opened = 0;
1049         fs_devices->seeding = 0;
1050
1051         return 0;
1052 }
1053
1054 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1055 {
1056         struct btrfs_fs_devices *seed_devices = NULL;
1057         int ret;
1058
1059         mutex_lock(&uuid_mutex);
1060         ret = __btrfs_close_devices(fs_devices);
1061         if (!fs_devices->opened) {
1062                 seed_devices = fs_devices->seed;
1063                 fs_devices->seed = NULL;
1064         }
1065         mutex_unlock(&uuid_mutex);
1066
1067         while (seed_devices) {
1068                 fs_devices = seed_devices;
1069                 seed_devices = fs_devices->seed;
1070                 __btrfs_close_devices(fs_devices);
1071                 free_fs_devices(fs_devices);
1072         }
1073         return ret;
1074 }
1075
1076 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1077                                 fmode_t flags, void *holder)
1078 {
1079         struct list_head *head = &fs_devices->devices;
1080         struct btrfs_device *device;
1081         struct btrfs_device *latest_dev = NULL;
1082         int ret = 0;
1083
1084         flags |= FMODE_EXCL;
1085
1086         list_for_each_entry(device, head, dev_list) {
1087                 /* Just open everything we can; ignore failures here */
1088                 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1089                         continue;
1090
1091                 if (!latest_dev ||
1092                     device->generation > latest_dev->generation)
1093                         latest_dev = device;
1094         }
1095         if (fs_devices->open_devices == 0) {
1096                 ret = -EINVAL;
1097                 goto out;
1098         }
1099         fs_devices->opened = 1;
1100         fs_devices->latest_bdev = latest_dev->bdev;
1101         fs_devices->total_rw_bytes = 0;
1102 out:
1103         return ret;
1104 }
1105
1106 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1107                        fmode_t flags, void *holder)
1108 {
1109         int ret;
1110
1111         mutex_lock(&uuid_mutex);
1112         if (fs_devices->opened) {
1113                 fs_devices->opened++;
1114                 ret = 0;
1115         } else {
1116                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1117         }
1118         mutex_unlock(&uuid_mutex);
1119         return ret;
1120 }
1121
1122 static void btrfs_release_disk_super(struct page *page)
1123 {
1124         kunmap(page);
1125         put_page(page);
1126 }
1127
1128 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1129                                  struct page **page,
1130                                  struct btrfs_super_block **disk_super)
1131 {
1132         void *p;
1133         pgoff_t index;
1134
1135         /* make sure our super fits in the device */
1136         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1137                 return 1;
1138
1139         /* make sure our super fits in the page */
1140         if (sizeof(**disk_super) > PAGE_SIZE)
1141                 return 1;
1142
1143         /* make sure our super doesn't straddle pages on disk */
1144         index = bytenr >> PAGE_SHIFT;
1145         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1146                 return 1;
1147
1148         /* pull in the page with our super */
1149         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1150                                    index, GFP_KERNEL);
1151
1152         if (IS_ERR_OR_NULL(*page))
1153                 return 1;
1154
1155         p = kmap(*page);
1156
1157         /* align our pointer to the offset of the super block */
1158         *disk_super = p + (bytenr & ~PAGE_MASK);
1159
1160         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1161             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1162                 btrfs_release_disk_super(*page);
1163                 return 1;
1164         }
1165
1166         if ((*disk_super)->label[0] &&
1167                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1168                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Look for a btrfs signature on a device. This may be called out of the mount path
1175  * and we are not allowed to call set_blocksize during the scan. The superblock
1176  * is read via pagecache
1177  */
1178 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1179                           struct btrfs_fs_devices **fs_devices_ret)
1180 {
1181         struct btrfs_super_block *disk_super;
1182         struct btrfs_device *device;
1183         struct block_device *bdev;
1184         struct page *page;
1185         int ret = 0;
1186         u64 bytenr;
1187
1188         /*
1189          * we would like to check all the supers, but that would make
1190          * a btrfs mount succeed after a mkfs from a different FS.
1191          * So, we need to add a special mount option to scan for
1192          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1193          */
1194         bytenr = btrfs_sb_offset(0);
1195         flags |= FMODE_EXCL;
1196         mutex_lock(&uuid_mutex);
1197
1198         bdev = blkdev_get_by_path(path, flags, holder);
1199         if (IS_ERR(bdev)) {
1200                 ret = PTR_ERR(bdev);
1201                 goto error;
1202         }
1203
1204         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1205                 ret = -EINVAL;
1206                 goto error_bdev_put;
1207         }
1208
1209         device = device_list_add(path, disk_super);
1210         if (IS_ERR(device))
1211                 ret = PTR_ERR(device);
1212         else
1213                 *fs_devices_ret = device->fs_devices;
1214
1215         btrfs_release_disk_super(page);
1216
1217 error_bdev_put:
1218         blkdev_put(bdev, flags);
1219 error:
1220         mutex_unlock(&uuid_mutex);
1221         return ret;
1222 }
1223
1224 /* helper to account the used device space in the range */
1225 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1226                                    u64 end, u64 *length)
1227 {
1228         struct btrfs_key key;
1229         struct btrfs_root *root = device->fs_info->dev_root;
1230         struct btrfs_dev_extent *dev_extent;
1231         struct btrfs_path *path;
1232         u64 extent_end;
1233         int ret;
1234         int slot;
1235         struct extent_buffer *l;
1236
1237         *length = 0;
1238
1239         if (start >= device->total_bytes ||
1240                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1241                 return 0;
1242
1243         path = btrfs_alloc_path();
1244         if (!path)
1245                 return -ENOMEM;
1246         path->reada = READA_FORWARD;
1247
1248         key.objectid = device->devid;
1249         key.offset = start;
1250         key.type = BTRFS_DEV_EXTENT_KEY;
1251
1252         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1253         if (ret < 0)
1254                 goto out;
1255         if (ret > 0) {
1256                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1257                 if (ret < 0)
1258                         goto out;
1259         }
1260
1261         while (1) {
1262                 l = path->nodes[0];
1263                 slot = path->slots[0];
1264                 if (slot >= btrfs_header_nritems(l)) {
1265                         ret = btrfs_next_leaf(root, path);
1266                         if (ret == 0)
1267                                 continue;
1268                         if (ret < 0)
1269                                 goto out;
1270
1271                         break;
1272                 }
1273                 btrfs_item_key_to_cpu(l, &key, slot);
1274
1275                 if (key.objectid < device->devid)
1276                         goto next;
1277
1278                 if (key.objectid > device->devid)
1279                         break;
1280
1281                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1282                         goto next;
1283
1284                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1285                 extent_end = key.offset + btrfs_dev_extent_length(l,
1286                                                                   dev_extent);
1287                 if (key.offset <= start && extent_end > end) {
1288                         *length = end - start + 1;
1289                         break;
1290                 } else if (key.offset <= start && extent_end > start)
1291                         *length += extent_end - start;
1292                 else if (key.offset > start && extent_end <= end)
1293                         *length += extent_end - key.offset;
1294                 else if (key.offset > start && key.offset <= end) {
1295                         *length += end - key.offset + 1;
1296                         break;
1297                 } else if (key.offset > end)
1298                         break;
1299
1300 next:
1301                 path->slots[0]++;
1302         }
1303         ret = 0;
1304 out:
1305         btrfs_free_path(path);
1306         return ret;
1307 }
1308
1309 static int contains_pending_extent(struct btrfs_transaction *transaction,
1310                                    struct btrfs_device *device,
1311                                    u64 *start, u64 len)
1312 {
1313         struct btrfs_fs_info *fs_info = device->fs_info;
1314         struct extent_map *em;
1315         struct list_head *search_list = &fs_info->pinned_chunks;
1316         int ret = 0;
1317         u64 physical_start = *start;
1318
1319         if (transaction)
1320                 search_list = &transaction->pending_chunks;
1321 again:
1322         list_for_each_entry(em, search_list, list) {
1323                 struct map_lookup *map;
1324                 int i;
1325
1326                 map = em->map_lookup;
1327                 for (i = 0; i < map->num_stripes; i++) {
1328                         u64 end;
1329
1330                         if (map->stripes[i].dev != device)
1331                                 continue;
1332                         if (map->stripes[i].physical >= physical_start + len ||
1333                             map->stripes[i].physical + em->orig_block_len <=
1334                             physical_start)
1335                                 continue;
1336                         /*
1337                          * Make sure that while processing the pinned list we do
1338                          * not override our *start with a lower value, because
1339                          * we can have pinned chunks that fall within this
1340                          * device hole and that have lower physical addresses
1341                          * than the pending chunks we processed before. If we
1342                          * do not take this special care we can end up getting
1343                          * 2 pending chunks that start at the same physical
1344                          * device offsets because the end offset of a pinned
1345                          * chunk can be equal to the start offset of some
1346                          * pending chunk.
1347                          */
1348                         end = map->stripes[i].physical + em->orig_block_len;
1349                         if (end > *start) {
1350                                 *start = end;
1351                                 ret = 1;
1352                         }
1353                 }
1354         }
1355         if (search_list != &fs_info->pinned_chunks) {
1356                 search_list = &fs_info->pinned_chunks;
1357                 goto again;
1358         }
1359
1360         return ret;
1361 }
1362
1363
1364 /*
1365  * find_free_dev_extent_start - find free space in the specified device
1366  * @device:       the device which we search the free space in
1367  * @num_bytes:    the size of the free space that we need
1368  * @search_start: the position from which to begin the search
1369  * @start:        store the start of the free space.
1370  * @len:          the size of the free space. that we find, or the size
1371  *                of the max free space if we don't find suitable free space
1372  *
1373  * this uses a pretty simple search, the expectation is that it is
1374  * called very infrequently and that a given device has a small number
1375  * of extents
1376  *
1377  * @start is used to store the start of the free space if we find. But if we
1378  * don't find suitable free space, it will be used to store the start position
1379  * of the max free space.
1380  *
1381  * @len is used to store the size of the free space that we find.
1382  * But if we don't find suitable free space, it is used to store the size of
1383  * the max free space.
1384  */
1385 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1386                                struct btrfs_device *device, u64 num_bytes,
1387                                u64 search_start, u64 *start, u64 *len)
1388 {
1389         struct btrfs_fs_info *fs_info = device->fs_info;
1390         struct btrfs_root *root = fs_info->dev_root;
1391         struct btrfs_key key;
1392         struct btrfs_dev_extent *dev_extent;
1393         struct btrfs_path *path;
1394         u64 hole_size;
1395         u64 max_hole_start;
1396         u64 max_hole_size;
1397         u64 extent_end;
1398         u64 search_end = device->total_bytes;
1399         int ret;
1400         int slot;
1401         struct extent_buffer *l;
1402
1403         /*
1404          * We don't want to overwrite the superblock on the drive nor any area
1405          * used by the boot loader (grub for example), so we make sure to start
1406          * at an offset of at least 1MB.
1407          */
1408         search_start = max_t(u64, search_start, SZ_1M);
1409
1410         path = btrfs_alloc_path();
1411         if (!path)
1412                 return -ENOMEM;
1413
1414         max_hole_start = search_start;
1415         max_hole_size = 0;
1416
1417 again:
1418         if (search_start >= search_end ||
1419                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1420                 ret = -ENOSPC;
1421                 goto out;
1422         }
1423
1424         path->reada = READA_FORWARD;
1425         path->search_commit_root = 1;
1426         path->skip_locking = 1;
1427
1428         key.objectid = device->devid;
1429         key.offset = search_start;
1430         key.type = BTRFS_DEV_EXTENT_KEY;
1431
1432         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1433         if (ret < 0)
1434                 goto out;
1435         if (ret > 0) {
1436                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1437                 if (ret < 0)
1438                         goto out;
1439         }
1440
1441         while (1) {
1442                 l = path->nodes[0];
1443                 slot = path->slots[0];
1444                 if (slot >= btrfs_header_nritems(l)) {
1445                         ret = btrfs_next_leaf(root, path);
1446                         if (ret == 0)
1447                                 continue;
1448                         if (ret < 0)
1449                                 goto out;
1450
1451                         break;
1452                 }
1453                 btrfs_item_key_to_cpu(l, &key, slot);
1454
1455                 if (key.objectid < device->devid)
1456                         goto next;
1457
1458                 if (key.objectid > device->devid)
1459                         break;
1460
1461                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1462                         goto next;
1463
1464                 if (key.offset > search_start) {
1465                         hole_size = key.offset - search_start;
1466
1467                         /*
1468                          * Have to check before we set max_hole_start, otherwise
1469                          * we could end up sending back this offset anyway.
1470                          */
1471                         if (contains_pending_extent(transaction, device,
1472                                                     &search_start,
1473                                                     hole_size)) {
1474                                 if (key.offset >= search_start) {
1475                                         hole_size = key.offset - search_start;
1476                                 } else {
1477                                         WARN_ON_ONCE(1);
1478                                         hole_size = 0;
1479                                 }
1480                         }
1481
1482                         if (hole_size > max_hole_size) {
1483                                 max_hole_start = search_start;
1484                                 max_hole_size = hole_size;
1485                         }
1486
1487                         /*
1488                          * If this free space is greater than which we need,
1489                          * it must be the max free space that we have found
1490                          * until now, so max_hole_start must point to the start
1491                          * of this free space and the length of this free space
1492                          * is stored in max_hole_size. Thus, we return
1493                          * max_hole_start and max_hole_size and go back to the
1494                          * caller.
1495                          */
1496                         if (hole_size >= num_bytes) {
1497                                 ret = 0;
1498                                 goto out;
1499                         }
1500                 }
1501
1502                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1503                 extent_end = key.offset + btrfs_dev_extent_length(l,
1504                                                                   dev_extent);
1505                 if (extent_end > search_start)
1506                         search_start = extent_end;
1507 next:
1508                 path->slots[0]++;
1509                 cond_resched();
1510         }
1511
1512         /*
1513          * At this point, search_start should be the end of
1514          * allocated dev extents, and when shrinking the device,
1515          * search_end may be smaller than search_start.
1516          */
1517         if (search_end > search_start) {
1518                 hole_size = search_end - search_start;
1519
1520                 if (contains_pending_extent(transaction, device, &search_start,
1521                                             hole_size)) {
1522                         btrfs_release_path(path);
1523                         goto again;
1524                 }
1525
1526                 if (hole_size > max_hole_size) {
1527                         max_hole_start = search_start;
1528                         max_hole_size = hole_size;
1529                 }
1530         }
1531
1532         /* See above. */
1533         if (max_hole_size < num_bytes)
1534                 ret = -ENOSPC;
1535         else
1536                 ret = 0;
1537
1538 out:
1539         btrfs_free_path(path);
1540         *start = max_hole_start;
1541         if (len)
1542                 *len = max_hole_size;
1543         return ret;
1544 }
1545
1546 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1547                          struct btrfs_device *device, u64 num_bytes,
1548                          u64 *start, u64 *len)
1549 {
1550         /* FIXME use last free of some kind */
1551         return find_free_dev_extent_start(trans->transaction, device,
1552                                           num_bytes, 0, start, len);
1553 }
1554
1555 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1556                           struct btrfs_device *device,
1557                           u64 start, u64 *dev_extent_len)
1558 {
1559         struct btrfs_fs_info *fs_info = device->fs_info;
1560         struct btrfs_root *root = fs_info->dev_root;
1561         int ret;
1562         struct btrfs_path *path;
1563         struct btrfs_key key;
1564         struct btrfs_key found_key;
1565         struct extent_buffer *leaf = NULL;
1566         struct btrfs_dev_extent *extent = NULL;
1567
1568         path = btrfs_alloc_path();
1569         if (!path)
1570                 return -ENOMEM;
1571
1572         key.objectid = device->devid;
1573         key.offset = start;
1574         key.type = BTRFS_DEV_EXTENT_KEY;
1575 again:
1576         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1577         if (ret > 0) {
1578                 ret = btrfs_previous_item(root, path, key.objectid,
1579                                           BTRFS_DEV_EXTENT_KEY);
1580                 if (ret)
1581                         goto out;
1582                 leaf = path->nodes[0];
1583                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1584                 extent = btrfs_item_ptr(leaf, path->slots[0],
1585                                         struct btrfs_dev_extent);
1586                 BUG_ON(found_key.offset > start || found_key.offset +
1587                        btrfs_dev_extent_length(leaf, extent) < start);
1588                 key = found_key;
1589                 btrfs_release_path(path);
1590                 goto again;
1591         } else if (ret == 0) {
1592                 leaf = path->nodes[0];
1593                 extent = btrfs_item_ptr(leaf, path->slots[0],
1594                                         struct btrfs_dev_extent);
1595         } else {
1596                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1597                 goto out;
1598         }
1599
1600         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1601
1602         ret = btrfs_del_item(trans, root, path);
1603         if (ret) {
1604                 btrfs_handle_fs_error(fs_info, ret,
1605                                       "Failed to remove dev extent item");
1606         } else {
1607                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1608         }
1609 out:
1610         btrfs_free_path(path);
1611         return ret;
1612 }
1613
1614 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1615                                   struct btrfs_device *device,
1616                                   u64 chunk_offset, u64 start, u64 num_bytes)
1617 {
1618         int ret;
1619         struct btrfs_path *path;
1620         struct btrfs_fs_info *fs_info = device->fs_info;
1621         struct btrfs_root *root = fs_info->dev_root;
1622         struct btrfs_dev_extent *extent;
1623         struct extent_buffer *leaf;
1624         struct btrfs_key key;
1625
1626         WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1627         WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1628         path = btrfs_alloc_path();
1629         if (!path)
1630                 return -ENOMEM;
1631
1632         key.objectid = device->devid;
1633         key.offset = start;
1634         key.type = BTRFS_DEV_EXTENT_KEY;
1635         ret = btrfs_insert_empty_item(trans, root, path, &key,
1636                                       sizeof(*extent));
1637         if (ret)
1638                 goto out;
1639
1640         leaf = path->nodes[0];
1641         extent = btrfs_item_ptr(leaf, path->slots[0],
1642                                 struct btrfs_dev_extent);
1643         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1644                                         BTRFS_CHUNK_TREE_OBJECTID);
1645         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1646                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1647         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1648
1649         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1650         btrfs_mark_buffer_dirty(leaf);
1651 out:
1652         btrfs_free_path(path);
1653         return ret;
1654 }
1655
1656 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1657 {
1658         struct extent_map_tree *em_tree;
1659         struct extent_map *em;
1660         struct rb_node *n;
1661         u64 ret = 0;
1662
1663         em_tree = &fs_info->mapping_tree.map_tree;
1664         read_lock(&em_tree->lock);
1665         n = rb_last(&em_tree->map);
1666         if (n) {
1667                 em = rb_entry(n, struct extent_map, rb_node);
1668                 ret = em->start + em->len;
1669         }
1670         read_unlock(&em_tree->lock);
1671
1672         return ret;
1673 }
1674
1675 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1676                                     u64 *devid_ret)
1677 {
1678         int ret;
1679         struct btrfs_key key;
1680         struct btrfs_key found_key;
1681         struct btrfs_path *path;
1682
1683         path = btrfs_alloc_path();
1684         if (!path)
1685                 return -ENOMEM;
1686
1687         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1688         key.type = BTRFS_DEV_ITEM_KEY;
1689         key.offset = (u64)-1;
1690
1691         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1692         if (ret < 0)
1693                 goto error;
1694
1695         BUG_ON(ret == 0); /* Corruption */
1696
1697         ret = btrfs_previous_item(fs_info->chunk_root, path,
1698                                   BTRFS_DEV_ITEMS_OBJECTID,
1699                                   BTRFS_DEV_ITEM_KEY);
1700         if (ret) {
1701                 *devid_ret = 1;
1702         } else {
1703                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1704                                       path->slots[0]);
1705                 *devid_ret = found_key.offset + 1;
1706         }
1707         ret = 0;
1708 error:
1709         btrfs_free_path(path);
1710         return ret;
1711 }
1712
1713 /*
1714  * the device information is stored in the chunk root
1715  * the btrfs_device struct should be fully filled in
1716  */
1717 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1718                             struct btrfs_fs_info *fs_info,
1719                             struct btrfs_device *device)
1720 {
1721         struct btrfs_root *root = fs_info->chunk_root;
1722         int ret;
1723         struct btrfs_path *path;
1724         struct btrfs_dev_item *dev_item;
1725         struct extent_buffer *leaf;
1726         struct btrfs_key key;
1727         unsigned long ptr;
1728
1729         path = btrfs_alloc_path();
1730         if (!path)
1731                 return -ENOMEM;
1732
1733         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1734         key.type = BTRFS_DEV_ITEM_KEY;
1735         key.offset = device->devid;
1736
1737         ret = btrfs_insert_empty_item(trans, root, path, &key,
1738                                       sizeof(*dev_item));
1739         if (ret)
1740                 goto out;
1741
1742         leaf = path->nodes[0];
1743         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1744
1745         btrfs_set_device_id(leaf, dev_item, device->devid);
1746         btrfs_set_device_generation(leaf, dev_item, 0);
1747         btrfs_set_device_type(leaf, dev_item, device->type);
1748         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1749         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1750         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1751         btrfs_set_device_total_bytes(leaf, dev_item,
1752                                      btrfs_device_get_disk_total_bytes(device));
1753         btrfs_set_device_bytes_used(leaf, dev_item,
1754                                     btrfs_device_get_bytes_used(device));
1755         btrfs_set_device_group(leaf, dev_item, 0);
1756         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1757         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1758         btrfs_set_device_start_offset(leaf, dev_item, 0);
1759
1760         ptr = btrfs_device_uuid(dev_item);
1761         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1762         ptr = btrfs_device_fsid(dev_item);
1763         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1764         btrfs_mark_buffer_dirty(leaf);
1765
1766         ret = 0;
1767 out:
1768         btrfs_free_path(path);
1769         return ret;
1770 }
1771
1772 /*
1773  * Function to update ctime/mtime for a given device path.
1774  * Mainly used for ctime/mtime based probe like libblkid.
1775  */
1776 static void update_dev_time(const char *path_name)
1777 {
1778         struct file *filp;
1779
1780         filp = filp_open(path_name, O_RDWR, 0);
1781         if (IS_ERR(filp))
1782                 return;
1783         file_update_time(filp);
1784         filp_close(filp, NULL);
1785 }
1786
1787 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1788                              struct btrfs_device *device)
1789 {
1790         struct btrfs_root *root = fs_info->chunk_root;
1791         int ret;
1792         struct btrfs_path *path;
1793         struct btrfs_key key;
1794         struct btrfs_trans_handle *trans;
1795
1796         path = btrfs_alloc_path();
1797         if (!path)
1798                 return -ENOMEM;
1799
1800         trans = btrfs_start_transaction(root, 0);
1801         if (IS_ERR(trans)) {
1802                 btrfs_free_path(path);
1803                 return PTR_ERR(trans);
1804         }
1805         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1806         key.type = BTRFS_DEV_ITEM_KEY;
1807         key.offset = device->devid;
1808
1809         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1810         if (ret) {
1811                 if (ret > 0)
1812                         ret = -ENOENT;
1813                 btrfs_abort_transaction(trans, ret);
1814                 btrfs_end_transaction(trans);
1815                 goto out;
1816         }
1817
1818         ret = btrfs_del_item(trans, root, path);
1819         if (ret) {
1820                 btrfs_abort_transaction(trans, ret);
1821                 btrfs_end_transaction(trans);
1822         }
1823
1824 out:
1825         btrfs_free_path(path);
1826         if (!ret)
1827                 ret = btrfs_commit_transaction(trans);
1828         return ret;
1829 }
1830
1831 /*
1832  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1833  * filesystem. It's up to the caller to adjust that number regarding eg. device
1834  * replace.
1835  */
1836 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1837                 u64 num_devices)
1838 {
1839         u64 all_avail;
1840         unsigned seq;
1841         int i;
1842
1843         do {
1844                 seq = read_seqbegin(&fs_info->profiles_lock);
1845
1846                 all_avail = fs_info->avail_data_alloc_bits |
1847                             fs_info->avail_system_alloc_bits |
1848                             fs_info->avail_metadata_alloc_bits;
1849         } while (read_seqretry(&fs_info->profiles_lock, seq));
1850
1851         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1852                 if (!(all_avail & btrfs_raid_group[i]))
1853                         continue;
1854
1855                 if (num_devices < btrfs_raid_array[i].devs_min) {
1856                         int ret = btrfs_raid_mindev_error[i];
1857
1858                         if (ret)
1859                                 return ret;
1860                 }
1861         }
1862
1863         return 0;
1864 }
1865
1866 static struct btrfs_device * btrfs_find_next_active_device(
1867                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1868 {
1869         struct btrfs_device *next_device;
1870
1871         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1872                 if (next_device != device &&
1873                     !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1874                     && next_device->bdev)
1875                         return next_device;
1876         }
1877
1878         return NULL;
1879 }
1880
1881 /*
1882  * Helper function to check if the given device is part of s_bdev / latest_bdev
1883  * and replace it with the provided or the next active device, in the context
1884  * where this function called, there should be always be another device (or
1885  * this_dev) which is active.
1886  */
1887 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1888                 struct btrfs_device *device, struct btrfs_device *this_dev)
1889 {
1890         struct btrfs_device *next_device;
1891
1892         if (this_dev)
1893                 next_device = this_dev;
1894         else
1895                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1896                                                                 device);
1897         ASSERT(next_device);
1898
1899         if (fs_info->sb->s_bdev &&
1900                         (fs_info->sb->s_bdev == device->bdev))
1901                 fs_info->sb->s_bdev = next_device->bdev;
1902
1903         if (fs_info->fs_devices->latest_bdev == device->bdev)
1904                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1905 }
1906
1907 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1908                 u64 devid)
1909 {
1910         struct btrfs_device *device;
1911         struct btrfs_fs_devices *cur_devices;
1912         u64 num_devices;
1913         int ret = 0;
1914
1915         mutex_lock(&fs_info->volume_mutex);
1916         mutex_lock(&uuid_mutex);
1917
1918         num_devices = fs_info->fs_devices->num_devices;
1919         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1920         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1921                 WARN_ON(num_devices < 1);
1922                 num_devices--;
1923         }
1924         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1925
1926         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1927         if (ret)
1928                 goto out;
1929
1930         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1931                                            &device);
1932         if (ret)
1933                 goto out;
1934
1935         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1936                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1937                 goto out;
1938         }
1939
1940         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1941             fs_info->fs_devices->rw_devices == 1) {
1942                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1943                 goto out;
1944         }
1945
1946         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1947                 mutex_lock(&fs_info->chunk_mutex);
1948                 list_del_init(&device->dev_alloc_list);
1949                 device->fs_devices->rw_devices--;
1950                 mutex_unlock(&fs_info->chunk_mutex);
1951         }
1952
1953         mutex_unlock(&uuid_mutex);
1954         ret = btrfs_shrink_device(device, 0);
1955         mutex_lock(&uuid_mutex);
1956         if (ret)
1957                 goto error_undo;
1958
1959         /*
1960          * TODO: the superblock still includes this device in its num_devices
1961          * counter although write_all_supers() is not locked out. This
1962          * could give a filesystem state which requires a degraded mount.
1963          */
1964         ret = btrfs_rm_dev_item(fs_info, device);
1965         if (ret)
1966                 goto error_undo;
1967
1968         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1969         btrfs_scrub_cancel_dev(fs_info, device);
1970
1971         /*
1972          * the device list mutex makes sure that we don't change
1973          * the device list while someone else is writing out all
1974          * the device supers. Whoever is writing all supers, should
1975          * lock the device list mutex before getting the number of
1976          * devices in the super block (super_copy). Conversely,
1977          * whoever updates the number of devices in the super block
1978          * (super_copy) should hold the device list mutex.
1979          */
1980
1981         cur_devices = device->fs_devices;
1982         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1983         list_del_rcu(&device->dev_list);
1984
1985         device->fs_devices->num_devices--;
1986         device->fs_devices->total_devices--;
1987
1988         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1989                 device->fs_devices->missing_devices--;
1990
1991         btrfs_assign_next_active_device(fs_info, device, NULL);
1992
1993         if (device->bdev) {
1994                 device->fs_devices->open_devices--;
1995                 /* remove sysfs entry */
1996                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1997         }
1998
1999         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2000         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2001         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2002
2003         /*
2004          * at this point, the device is zero sized and detached from
2005          * the devices list.  All that's left is to zero out the old
2006          * supers and free the device.
2007          */
2008         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2009                 btrfs_scratch_superblocks(device->bdev, device->name->str);
2010
2011         btrfs_close_bdev(device);
2012         call_rcu(&device->rcu, free_device_rcu);
2013
2014         if (cur_devices->open_devices == 0) {
2015                 struct btrfs_fs_devices *fs_devices;
2016                 fs_devices = fs_info->fs_devices;
2017                 while (fs_devices) {
2018                         if (fs_devices->seed == cur_devices) {
2019                                 fs_devices->seed = cur_devices->seed;
2020                                 break;
2021                         }
2022                         fs_devices = fs_devices->seed;
2023                 }
2024                 cur_devices->seed = NULL;
2025                 __btrfs_close_devices(cur_devices);
2026                 free_fs_devices(cur_devices);
2027         }
2028
2029 out:
2030         mutex_unlock(&uuid_mutex);
2031         mutex_unlock(&fs_info->volume_mutex);
2032         return ret;
2033
2034 error_undo:
2035         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2036                 mutex_lock(&fs_info->chunk_mutex);
2037                 list_add(&device->dev_alloc_list,
2038                          &fs_info->fs_devices->alloc_list);
2039                 device->fs_devices->rw_devices++;
2040                 mutex_unlock(&fs_info->chunk_mutex);
2041         }
2042         goto out;
2043 }
2044
2045 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2046                                         struct btrfs_device *srcdev)
2047 {
2048         struct btrfs_fs_devices *fs_devices;
2049
2050         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2051
2052         /*
2053          * in case of fs with no seed, srcdev->fs_devices will point
2054          * to fs_devices of fs_info. However when the dev being replaced is
2055          * a seed dev it will point to the seed's local fs_devices. In short
2056          * srcdev will have its correct fs_devices in both the cases.
2057          */
2058         fs_devices = srcdev->fs_devices;
2059
2060         list_del_rcu(&srcdev->dev_list);
2061         list_del(&srcdev->dev_alloc_list);
2062         fs_devices->num_devices--;
2063         if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2064                 fs_devices->missing_devices--;
2065
2066         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2067                 fs_devices->rw_devices--;
2068
2069         if (srcdev->bdev)
2070                 fs_devices->open_devices--;
2071 }
2072
2073 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2074                                       struct btrfs_device *srcdev)
2075 {
2076         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2077
2078         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2079                 /* zero out the old super if it is writable */
2080                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2081         }
2082
2083         btrfs_close_bdev(srcdev);
2084         call_rcu(&srcdev->rcu, free_device_rcu);
2085
2086         /* if this is no devs we rather delete the fs_devices */
2087         if (!fs_devices->num_devices) {
2088                 struct btrfs_fs_devices *tmp_fs_devices;
2089
2090                 /*
2091                  * On a mounted FS, num_devices can't be zero unless it's a
2092                  * seed. In case of a seed device being replaced, the replace
2093                  * target added to the sprout FS, so there will be no more
2094                  * device left under the seed FS.
2095                  */
2096                 ASSERT(fs_devices->seeding);
2097
2098                 tmp_fs_devices = fs_info->fs_devices;
2099                 while (tmp_fs_devices) {
2100                         if (tmp_fs_devices->seed == fs_devices) {
2101                                 tmp_fs_devices->seed = fs_devices->seed;
2102                                 break;
2103                         }
2104                         tmp_fs_devices = tmp_fs_devices->seed;
2105                 }
2106                 fs_devices->seed = NULL;
2107                 __btrfs_close_devices(fs_devices);
2108                 free_fs_devices(fs_devices);
2109         }
2110 }
2111
2112 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2113                                       struct btrfs_device *tgtdev)
2114 {
2115         mutex_lock(&uuid_mutex);
2116         WARN_ON(!tgtdev);
2117         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2118
2119         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2120
2121         if (tgtdev->bdev)
2122                 fs_info->fs_devices->open_devices--;
2123
2124         fs_info->fs_devices->num_devices--;
2125
2126         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2127
2128         list_del_rcu(&tgtdev->dev_list);
2129
2130         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2131         mutex_unlock(&uuid_mutex);
2132
2133         /*
2134          * The update_dev_time() with in btrfs_scratch_superblocks()
2135          * may lead to a call to btrfs_show_devname() which will try
2136          * to hold device_list_mutex. And here this device
2137          * is already out of device list, so we don't have to hold
2138          * the device_list_mutex lock.
2139          */
2140         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2141
2142         btrfs_close_bdev(tgtdev);
2143         call_rcu(&tgtdev->rcu, free_device_rcu);
2144 }
2145
2146 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2147                                      const char *device_path,
2148                                      struct btrfs_device **device)
2149 {
2150         int ret = 0;
2151         struct btrfs_super_block *disk_super;
2152         u64 devid;
2153         u8 *dev_uuid;
2154         struct block_device *bdev;
2155         struct buffer_head *bh;
2156
2157         *device = NULL;
2158         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2159                                     fs_info->bdev_holder, 0, &bdev, &bh);
2160         if (ret)
2161                 return ret;
2162         disk_super = (struct btrfs_super_block *)bh->b_data;
2163         devid = btrfs_stack_device_id(&disk_super->dev_item);
2164         dev_uuid = disk_super->dev_item.uuid;
2165         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2166         brelse(bh);
2167         if (!*device)
2168                 ret = -ENOENT;
2169         blkdev_put(bdev, FMODE_READ);
2170         return ret;
2171 }
2172
2173 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2174                                          const char *device_path,
2175                                          struct btrfs_device **device)
2176 {
2177         *device = NULL;
2178         if (strcmp(device_path, "missing") == 0) {
2179                 struct list_head *devices;
2180                 struct btrfs_device *tmp;
2181
2182                 devices = &fs_info->fs_devices->devices;
2183                 /*
2184                  * It is safe to read the devices since the volume_mutex
2185                  * is held by the caller.
2186                  */
2187                 list_for_each_entry(tmp, devices, dev_list) {
2188                         if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2189                                         &tmp->dev_state) && !tmp->bdev) {
2190                                 *device = tmp;
2191                                 break;
2192                         }
2193                 }
2194
2195                 if (!*device)
2196                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2197
2198                 return 0;
2199         } else {
2200                 return btrfs_find_device_by_path(fs_info, device_path, device);
2201         }
2202 }
2203
2204 /*
2205  * Lookup a device given by device id, or the path if the id is 0.
2206  */
2207 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2208                                  const char *devpath,
2209                                  struct btrfs_device **device)
2210 {
2211         int ret;
2212
2213         if (devid) {
2214                 ret = 0;
2215                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2216                 if (!*device)
2217                         ret = -ENOENT;
2218         } else {
2219                 if (!devpath || !devpath[0])
2220                         return -EINVAL;
2221
2222                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2223                                                            device);
2224         }
2225         return ret;
2226 }
2227
2228 /*
2229  * does all the dirty work required for changing file system's UUID.
2230  */
2231 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2232 {
2233         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2234         struct btrfs_fs_devices *old_devices;
2235         struct btrfs_fs_devices *seed_devices;
2236         struct btrfs_super_block *disk_super = fs_info->super_copy;
2237         struct btrfs_device *device;
2238         u64 super_flags;
2239
2240         BUG_ON(!mutex_is_locked(&uuid_mutex));
2241         if (!fs_devices->seeding)
2242                 return -EINVAL;
2243
2244         seed_devices = alloc_fs_devices(NULL);
2245         if (IS_ERR(seed_devices))
2246                 return PTR_ERR(seed_devices);
2247
2248         old_devices = clone_fs_devices(fs_devices);
2249         if (IS_ERR(old_devices)) {
2250                 kfree(seed_devices);
2251                 return PTR_ERR(old_devices);
2252         }
2253
2254         list_add(&old_devices->list, &fs_uuids);
2255
2256         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2257         seed_devices->opened = 1;
2258         INIT_LIST_HEAD(&seed_devices->devices);
2259         INIT_LIST_HEAD(&seed_devices->alloc_list);
2260         mutex_init(&seed_devices->device_list_mutex);
2261
2262         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2263         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2264                               synchronize_rcu);
2265         list_for_each_entry(device, &seed_devices->devices, dev_list)
2266                 device->fs_devices = seed_devices;
2267
2268         mutex_lock(&fs_info->chunk_mutex);
2269         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2270         mutex_unlock(&fs_info->chunk_mutex);
2271
2272         fs_devices->seeding = 0;
2273         fs_devices->num_devices = 0;
2274         fs_devices->open_devices = 0;
2275         fs_devices->missing_devices = 0;
2276         fs_devices->rotating = 0;
2277         fs_devices->seed = seed_devices;
2278
2279         generate_random_uuid(fs_devices->fsid);
2280         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2281         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2282         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2283
2284         super_flags = btrfs_super_flags(disk_super) &
2285                       ~BTRFS_SUPER_FLAG_SEEDING;
2286         btrfs_set_super_flags(disk_super, super_flags);
2287
2288         return 0;
2289 }
2290
2291 /*
2292  * Store the expected generation for seed devices in device items.
2293  */
2294 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2295                                struct btrfs_fs_info *fs_info)
2296 {
2297         struct btrfs_root *root = fs_info->chunk_root;
2298         struct btrfs_path *path;
2299         struct extent_buffer *leaf;
2300         struct btrfs_dev_item *dev_item;
2301         struct btrfs_device *device;
2302         struct btrfs_key key;
2303         u8 fs_uuid[BTRFS_FSID_SIZE];
2304         u8 dev_uuid[BTRFS_UUID_SIZE];
2305         u64 devid;
2306         int ret;
2307
2308         path = btrfs_alloc_path();
2309         if (!path)
2310                 return -ENOMEM;
2311
2312         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2313         key.offset = 0;
2314         key.type = BTRFS_DEV_ITEM_KEY;
2315
2316         while (1) {
2317                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2318                 if (ret < 0)
2319                         goto error;
2320
2321                 leaf = path->nodes[0];
2322 next_slot:
2323                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2324                         ret = btrfs_next_leaf(root, path);
2325                         if (ret > 0)
2326                                 break;
2327                         if (ret < 0)
2328                                 goto error;
2329                         leaf = path->nodes[0];
2330                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2331                         btrfs_release_path(path);
2332                         continue;
2333                 }
2334
2335                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2336                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2337                     key.type != BTRFS_DEV_ITEM_KEY)
2338                         break;
2339
2340                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2341                                           struct btrfs_dev_item);
2342                 devid = btrfs_device_id(leaf, dev_item);
2343                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2344                                    BTRFS_UUID_SIZE);
2345                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2346                                    BTRFS_FSID_SIZE);
2347                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2348                 BUG_ON(!device); /* Logic error */
2349
2350                 if (device->fs_devices->seeding) {
2351                         btrfs_set_device_generation(leaf, dev_item,
2352                                                     device->generation);
2353                         btrfs_mark_buffer_dirty(leaf);
2354                 }
2355
2356                 path->slots[0]++;
2357                 goto next_slot;
2358         }
2359         ret = 0;
2360 error:
2361         btrfs_free_path(path);
2362         return ret;
2363 }
2364
2365 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2366 {
2367         struct btrfs_root *root = fs_info->dev_root;
2368         struct request_queue *q;
2369         struct btrfs_trans_handle *trans;
2370         struct btrfs_device *device;
2371         struct block_device *bdev;
2372         struct list_head *devices;
2373         struct super_block *sb = fs_info->sb;
2374         struct rcu_string *name;
2375         u64 tmp;
2376         int seeding_dev = 0;
2377         int ret = 0;
2378         bool unlocked = false;
2379
2380         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2381                 return -EROFS;
2382
2383         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2384                                   fs_info->bdev_holder);
2385         if (IS_ERR(bdev))
2386                 return PTR_ERR(bdev);
2387
2388         if (fs_info->fs_devices->seeding) {
2389                 seeding_dev = 1;
2390                 down_write(&sb->s_umount);
2391                 mutex_lock(&uuid_mutex);
2392         }
2393
2394         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2395
2396         devices = &fs_info->fs_devices->devices;
2397
2398         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2399         list_for_each_entry(device, devices, dev_list) {
2400                 if (device->bdev == bdev) {
2401                         ret = -EEXIST;
2402                         mutex_unlock(
2403                                 &fs_info->fs_devices->device_list_mutex);
2404                         goto error;
2405                 }
2406         }
2407         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2408
2409         device = btrfs_alloc_device(fs_info, NULL, NULL);
2410         if (IS_ERR(device)) {
2411                 /* we can safely leave the fs_devices entry around */
2412                 ret = PTR_ERR(device);
2413                 goto error;
2414         }
2415
2416         name = rcu_string_strdup(device_path, GFP_KERNEL);
2417         if (!name) {
2418                 ret = -ENOMEM;
2419                 goto error_free_device;
2420         }
2421         rcu_assign_pointer(device->name, name);
2422
2423         trans = btrfs_start_transaction(root, 0);
2424         if (IS_ERR(trans)) {
2425                 ret = PTR_ERR(trans);
2426                 goto error_free_device;
2427         }
2428
2429         q = bdev_get_queue(bdev);
2430         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2431         device->generation = trans->transid;
2432         device->io_width = fs_info->sectorsize;
2433         device->io_align = fs_info->sectorsize;
2434         device->sector_size = fs_info->sectorsize;
2435         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2436                                          fs_info->sectorsize);
2437         device->disk_total_bytes = device->total_bytes;
2438         device->commit_total_bytes = device->total_bytes;
2439         device->fs_info = fs_info;
2440         device->bdev = bdev;
2441         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2442         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2443         device->mode = FMODE_EXCL;
2444         device->dev_stats_valid = 1;
2445         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2446
2447         if (seeding_dev) {
2448                 sb->s_flags &= ~SB_RDONLY;
2449                 ret = btrfs_prepare_sprout(fs_info);
2450                 if (ret) {
2451                         btrfs_abort_transaction(trans, ret);
2452                         goto error_trans;
2453                 }
2454         }
2455
2456         device->fs_devices = fs_info->fs_devices;
2457
2458         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2459         mutex_lock(&fs_info->chunk_mutex);
2460         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2461         list_add(&device->dev_alloc_list,
2462                  &fs_info->fs_devices->alloc_list);
2463         fs_info->fs_devices->num_devices++;
2464         fs_info->fs_devices->open_devices++;
2465         fs_info->fs_devices->rw_devices++;
2466         fs_info->fs_devices->total_devices++;
2467         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2468
2469         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2470
2471         if (!blk_queue_nonrot(q))
2472                 fs_info->fs_devices->rotating = 1;
2473
2474         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2475         btrfs_set_super_total_bytes(fs_info->super_copy,
2476                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2477
2478         tmp = btrfs_super_num_devices(fs_info->super_copy);
2479         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2480
2481         /* add sysfs device entry */
2482         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2483
2484         /*
2485          * we've got more storage, clear any full flags on the space
2486          * infos
2487          */
2488         btrfs_clear_space_info_full(fs_info);
2489
2490         mutex_unlock(&fs_info->chunk_mutex);
2491         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2492
2493         if (seeding_dev) {
2494                 mutex_lock(&fs_info->chunk_mutex);
2495                 ret = init_first_rw_device(trans, fs_info);
2496                 mutex_unlock(&fs_info->chunk_mutex);
2497                 if (ret) {
2498                         btrfs_abort_transaction(trans, ret);
2499                         goto error_sysfs;
2500                 }
2501         }
2502
2503         ret = btrfs_add_dev_item(trans, fs_info, device);
2504         if (ret) {
2505                 btrfs_abort_transaction(trans, ret);
2506                 goto error_sysfs;
2507         }
2508
2509         if (seeding_dev) {
2510                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2511
2512                 ret = btrfs_finish_sprout(trans, fs_info);
2513                 if (ret) {
2514                         btrfs_abort_transaction(trans, ret);
2515                         goto error_sysfs;
2516                 }
2517
2518                 /* Sprouting would change fsid of the mounted root,
2519                  * so rename the fsid on the sysfs
2520                  */
2521                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2522                                                 fs_info->fsid);
2523                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2524                         btrfs_warn(fs_info,
2525                                    "sysfs: failed to create fsid for sprout");
2526         }
2527
2528         ret = btrfs_commit_transaction(trans);
2529
2530         if (seeding_dev) {
2531                 mutex_unlock(&uuid_mutex);
2532                 up_write(&sb->s_umount);
2533                 unlocked = true;
2534
2535                 if (ret) /* transaction commit */
2536                         return ret;
2537
2538                 ret = btrfs_relocate_sys_chunks(fs_info);
2539                 if (ret < 0)
2540                         btrfs_handle_fs_error(fs_info, ret,
2541                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2542                 trans = btrfs_attach_transaction(root);
2543                 if (IS_ERR(trans)) {
2544                         if (PTR_ERR(trans) == -ENOENT)
2545                                 return 0;
2546                         ret = PTR_ERR(trans);
2547                         trans = NULL;
2548                         goto error_sysfs;
2549                 }
2550                 ret = btrfs_commit_transaction(trans);
2551         }
2552
2553         /* Update ctime/mtime for libblkid */
2554         update_dev_time(device_path);
2555         return ret;
2556
2557 error_sysfs:
2558         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2559 error_trans:
2560         if (seeding_dev)
2561                 sb->s_flags |= SB_RDONLY;
2562         if (trans)
2563                 btrfs_end_transaction(trans);
2564 error_free_device:
2565         free_device(device);
2566 error:
2567         blkdev_put(bdev, FMODE_EXCL);
2568         if (seeding_dev && !unlocked) {
2569                 mutex_unlock(&uuid_mutex);
2570                 up_write(&sb->s_umount);
2571         }
2572         return ret;
2573 }
2574
2575 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2576                                   const char *device_path,
2577                                   struct btrfs_device *srcdev,
2578                                   struct btrfs_device **device_out)
2579 {
2580         struct btrfs_device *device;
2581         struct block_device *bdev;
2582         struct list_head *devices;
2583         struct rcu_string *name;
2584         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2585         int ret = 0;
2586
2587         *device_out = NULL;
2588         if (fs_info->fs_devices->seeding) {
2589                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2590                 return -EINVAL;
2591         }
2592
2593         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2594                                   fs_info->bdev_holder);
2595         if (IS_ERR(bdev)) {
2596                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2597                 return PTR_ERR(bdev);
2598         }
2599
2600         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2601
2602         devices = &fs_info->fs_devices->devices;
2603         list_for_each_entry(device, devices, dev_list) {
2604                 if (device->bdev == bdev) {
2605                         btrfs_err(fs_info,
2606                                   "target device is in the filesystem!");
2607                         ret = -EEXIST;
2608                         goto error;
2609                 }
2610         }
2611
2612
2613         if (i_size_read(bdev->bd_inode) <
2614             btrfs_device_get_total_bytes(srcdev)) {
2615                 btrfs_err(fs_info,
2616                           "target device is smaller than source device!");
2617                 ret = -EINVAL;
2618                 goto error;
2619         }
2620
2621
2622         device = btrfs_alloc_device(NULL, &devid, NULL);
2623         if (IS_ERR(device)) {
2624                 ret = PTR_ERR(device);
2625                 goto error;
2626         }
2627
2628         name = rcu_string_strdup(device_path, GFP_KERNEL);
2629         if (!name) {
2630                 free_device(device);
2631                 ret = -ENOMEM;
2632                 goto error;
2633         }
2634         rcu_assign_pointer(device->name, name);
2635
2636         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2637         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2638         device->generation = 0;
2639         device->io_width = fs_info->sectorsize;
2640         device->io_align = fs_info->sectorsize;
2641         device->sector_size = fs_info->sectorsize;
2642         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2643         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2644         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2645         ASSERT(list_empty(&srcdev->resized_list));
2646         device->commit_total_bytes = srcdev->commit_total_bytes;
2647         device->commit_bytes_used = device->bytes_used;
2648         device->fs_info = fs_info;
2649         device->bdev = bdev;
2650         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2651         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2652         device->mode = FMODE_EXCL;
2653         device->dev_stats_valid = 1;
2654         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2655         device->fs_devices = fs_info->fs_devices;
2656         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2657         fs_info->fs_devices->num_devices++;
2658         fs_info->fs_devices->open_devices++;
2659         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2660
2661         *device_out = device;
2662         return ret;
2663
2664 error:
2665         blkdev_put(bdev, FMODE_EXCL);
2666         return ret;
2667 }
2668
2669 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2670                                               struct btrfs_device *tgtdev)
2671 {
2672         u32 sectorsize = fs_info->sectorsize;
2673
2674         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2675         tgtdev->io_width = sectorsize;
2676         tgtdev->io_align = sectorsize;
2677         tgtdev->sector_size = sectorsize;
2678         tgtdev->fs_info = fs_info;
2679         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &tgtdev->dev_state);
2680 }
2681
2682 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2683                                         struct btrfs_device *device)
2684 {
2685         int ret;
2686         struct btrfs_path *path;
2687         struct btrfs_root *root = device->fs_info->chunk_root;
2688         struct btrfs_dev_item *dev_item;
2689         struct extent_buffer *leaf;
2690         struct btrfs_key key;
2691
2692         path = btrfs_alloc_path();
2693         if (!path)
2694                 return -ENOMEM;
2695
2696         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2697         key.type = BTRFS_DEV_ITEM_KEY;
2698         key.offset = device->devid;
2699
2700         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2701         if (ret < 0)
2702                 goto out;
2703
2704         if (ret > 0) {
2705                 ret = -ENOENT;
2706                 goto out;
2707         }
2708
2709         leaf = path->nodes[0];
2710         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2711
2712         btrfs_set_device_id(leaf, dev_item, device->devid);
2713         btrfs_set_device_type(leaf, dev_item, device->type);
2714         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2715         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2716         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2717         btrfs_set_device_total_bytes(leaf, dev_item,
2718                                      btrfs_device_get_disk_total_bytes(device));
2719         btrfs_set_device_bytes_used(leaf, dev_item,
2720                                     btrfs_device_get_bytes_used(device));
2721         btrfs_mark_buffer_dirty(leaf);
2722
2723 out:
2724         btrfs_free_path(path);
2725         return ret;
2726 }
2727
2728 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2729                       struct btrfs_device *device, u64 new_size)
2730 {
2731         struct btrfs_fs_info *fs_info = device->fs_info;
2732         struct btrfs_super_block *super_copy = fs_info->super_copy;
2733         struct btrfs_fs_devices *fs_devices;
2734         u64 old_total;
2735         u64 diff;
2736
2737         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2738                 return -EACCES;
2739
2740         new_size = round_down(new_size, fs_info->sectorsize);
2741
2742         mutex_lock(&fs_info->chunk_mutex);
2743         old_total = btrfs_super_total_bytes(super_copy);
2744         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2745
2746         if (new_size <= device->total_bytes ||
2747             test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2748                 mutex_unlock(&fs_info->chunk_mutex);
2749                 return -EINVAL;
2750         }
2751
2752         fs_devices = fs_info->fs_devices;
2753
2754         btrfs_set_super_total_bytes(super_copy,
2755                         round_down(old_total + diff, fs_info->sectorsize));
2756         device->fs_devices->total_rw_bytes += diff;
2757
2758         btrfs_device_set_total_bytes(device, new_size);
2759         btrfs_device_set_disk_total_bytes(device, new_size);
2760         btrfs_clear_space_info_full(device->fs_info);
2761         if (list_empty(&device->resized_list))
2762                 list_add_tail(&device->resized_list,
2763                               &fs_devices->resized_devices);
2764         mutex_unlock(&fs_info->chunk_mutex);
2765
2766         return btrfs_update_device(trans, device);
2767 }
2768
2769 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2770                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2771 {
2772         struct btrfs_root *root = fs_info->chunk_root;
2773         int ret;
2774         struct btrfs_path *path;
2775         struct btrfs_key key;
2776
2777         path = btrfs_alloc_path();
2778         if (!path)
2779                 return -ENOMEM;
2780
2781         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2782         key.offset = chunk_offset;
2783         key.type = BTRFS_CHUNK_ITEM_KEY;
2784
2785         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2786         if (ret < 0)
2787                 goto out;
2788         else if (ret > 0) { /* Logic error or corruption */
2789                 btrfs_handle_fs_error(fs_info, -ENOENT,
2790                                       "Failed lookup while freeing chunk.");
2791                 ret = -ENOENT;
2792                 goto out;
2793         }
2794
2795         ret = btrfs_del_item(trans, root, path);
2796         if (ret < 0)
2797                 btrfs_handle_fs_error(fs_info, ret,
2798                                       "Failed to delete chunk item.");
2799 out:
2800         btrfs_free_path(path);
2801         return ret;
2802 }
2803
2804 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2805 {
2806         struct btrfs_super_block *super_copy = fs_info->super_copy;
2807         struct btrfs_disk_key *disk_key;
2808         struct btrfs_chunk *chunk;
2809         u8 *ptr;
2810         int ret = 0;
2811         u32 num_stripes;
2812         u32 array_size;
2813         u32 len = 0;
2814         u32 cur;
2815         struct btrfs_key key;
2816
2817         mutex_lock(&fs_info->chunk_mutex);
2818         array_size = btrfs_super_sys_array_size(super_copy);
2819
2820         ptr = super_copy->sys_chunk_array;
2821         cur = 0;
2822
2823         while (cur < array_size) {
2824                 disk_key = (struct btrfs_disk_key *)ptr;
2825                 btrfs_disk_key_to_cpu(&key, disk_key);
2826
2827                 len = sizeof(*disk_key);
2828
2829                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2830                         chunk = (struct btrfs_chunk *)(ptr + len);
2831                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2832                         len += btrfs_chunk_item_size(num_stripes);
2833                 } else {
2834                         ret = -EIO;
2835                         break;
2836                 }
2837                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2838                     key.offset == chunk_offset) {
2839                         memmove(ptr, ptr + len, array_size - (cur + len));
2840                         array_size -= len;
2841                         btrfs_set_super_sys_array_size(super_copy, array_size);
2842                 } else {
2843                         ptr += len;
2844                         cur += len;
2845                 }
2846         }
2847         mutex_unlock(&fs_info->chunk_mutex);
2848         return ret;
2849 }
2850
2851 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2852                                         u64 logical, u64 length)
2853 {
2854         struct extent_map_tree *em_tree;
2855         struct extent_map *em;
2856
2857         em_tree = &fs_info->mapping_tree.map_tree;
2858         read_lock(&em_tree->lock);
2859         em = lookup_extent_mapping(em_tree, logical, length);
2860         read_unlock(&em_tree->lock);
2861
2862         if (!em) {
2863                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2864                            logical, length);
2865                 return ERR_PTR(-EINVAL);
2866         }
2867
2868         if (em->start > logical || em->start + em->len < logical) {
2869                 btrfs_crit(fs_info,
2870                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2871                            logical, length, em->start, em->start + em->len);
2872                 free_extent_map(em);
2873                 return ERR_PTR(-EINVAL);
2874         }
2875
2876         /* callers are responsible for dropping em's ref. */
2877         return em;
2878 }
2879
2880 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2881                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2882 {
2883         struct extent_map *em;
2884         struct map_lookup *map;
2885         u64 dev_extent_len = 0;
2886         int i, ret = 0;
2887         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2888
2889         em = get_chunk_map(fs_info, chunk_offset, 1);
2890         if (IS_ERR(em)) {
2891                 /*
2892                  * This is a logic error, but we don't want to just rely on the
2893                  * user having built with ASSERT enabled, so if ASSERT doesn't
2894                  * do anything we still error out.
2895                  */
2896                 ASSERT(0);
2897                 return PTR_ERR(em);
2898         }
2899         map = em->map_lookup;
2900         mutex_lock(&fs_info->chunk_mutex);
2901         check_system_chunk(trans, fs_info, map->type);
2902         mutex_unlock(&fs_info->chunk_mutex);
2903
2904         /*
2905          * Take the device list mutex to prevent races with the final phase of
2906          * a device replace operation that replaces the device object associated
2907          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2908          */
2909         mutex_lock(&fs_devices->device_list_mutex);
2910         for (i = 0; i < map->num_stripes; i++) {
2911                 struct btrfs_device *device = map->stripes[i].dev;
2912                 ret = btrfs_free_dev_extent(trans, device,
2913                                             map->stripes[i].physical,
2914                                             &dev_extent_len);
2915                 if (ret) {
2916                         mutex_unlock(&fs_devices->device_list_mutex);
2917                         btrfs_abort_transaction(trans, ret);
2918                         goto out;
2919                 }
2920
2921                 if (device->bytes_used > 0) {
2922                         mutex_lock(&fs_info->chunk_mutex);
2923                         btrfs_device_set_bytes_used(device,
2924                                         device->bytes_used - dev_extent_len);
2925                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2926                         btrfs_clear_space_info_full(fs_info);
2927                         mutex_unlock(&fs_info->chunk_mutex);
2928                 }
2929
2930                 if (map->stripes[i].dev) {
2931                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2932                         if (ret) {
2933                                 mutex_unlock(&fs_devices->device_list_mutex);
2934                                 btrfs_abort_transaction(trans, ret);
2935                                 goto out;
2936                         }
2937                 }
2938         }
2939         mutex_unlock(&fs_devices->device_list_mutex);
2940
2941         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2942         if (ret) {
2943                 btrfs_abort_transaction(trans, ret);
2944                 goto out;
2945         }
2946
2947         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2948
2949         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2950                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2951                 if (ret) {
2952                         btrfs_abort_transaction(trans, ret);
2953                         goto out;
2954                 }
2955         }
2956
2957         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2958         if (ret) {
2959                 btrfs_abort_transaction(trans, ret);
2960                 goto out;
2961         }
2962
2963 out:
2964         /* once for us */
2965         free_extent_map(em);
2966         return ret;
2967 }
2968
2969 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2970 {
2971         struct btrfs_root *root = fs_info->chunk_root;
2972         struct btrfs_trans_handle *trans;
2973         int ret;
2974
2975         /*
2976          * Prevent races with automatic removal of unused block groups.
2977          * After we relocate and before we remove the chunk with offset
2978          * chunk_offset, automatic removal of the block group can kick in,
2979          * resulting in a failure when calling btrfs_remove_chunk() below.
2980          *
2981          * Make sure to acquire this mutex before doing a tree search (dev
2982          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2983          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2984          * we release the path used to search the chunk/dev tree and before
2985          * the current task acquires this mutex and calls us.
2986          */
2987         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2988
2989         ret = btrfs_can_relocate(fs_info, chunk_offset);
2990         if (ret)
2991                 return -ENOSPC;
2992
2993         /* step one, relocate all the extents inside this chunk */
2994         btrfs_scrub_pause(fs_info);
2995         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2996         btrfs_scrub_continue(fs_info);
2997         if (ret)
2998                 return ret;
2999
3000         trans = btrfs_start_trans_remove_block_group(root->fs_info,
3001                                                      chunk_offset);
3002         if (IS_ERR(trans)) {
3003                 ret = PTR_ERR(trans);
3004                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3005                 return ret;
3006         }
3007
3008         /*
3009          * step two, delete the device extents and the
3010          * chunk tree entries
3011          */
3012         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3013         btrfs_end_transaction(trans);
3014         return ret;
3015 }
3016
3017 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3018 {
3019         struct btrfs_root *chunk_root = fs_info->chunk_root;
3020         struct btrfs_path *path;
3021         struct extent_buffer *leaf;
3022         struct btrfs_chunk *chunk;
3023         struct btrfs_key key;
3024         struct btrfs_key found_key;
3025         u64 chunk_type;
3026         bool retried = false;
3027         int failed = 0;
3028         int ret;
3029
3030         path = btrfs_alloc_path();
3031         if (!path)
3032                 return -ENOMEM;
3033
3034 again:
3035         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3036         key.offset = (u64)-1;
3037         key.type = BTRFS_CHUNK_ITEM_KEY;
3038
3039         while (1) {
3040                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3041                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3042                 if (ret < 0) {
3043                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3044                         goto error;
3045                 }
3046                 BUG_ON(ret == 0); /* Corruption */
3047
3048                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3049                                           key.type);
3050                 if (ret)
3051                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3052                 if (ret < 0)
3053                         goto error;
3054                 if (ret > 0)
3055                         break;
3056
3057                 leaf = path->nodes[0];
3058                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3059
3060                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3061                                        struct btrfs_chunk);
3062                 chunk_type = btrfs_chunk_type(leaf, chunk);
3063                 btrfs_release_path(path);
3064
3065                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3066                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3067                         if (ret == -ENOSPC)
3068                                 failed++;
3069                         else
3070                                 BUG_ON(ret);
3071                 }
3072                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3073
3074                 if (found_key.offset == 0)
3075                         break;
3076                 key.offset = found_key.offset - 1;
3077         }
3078         ret = 0;
3079         if (failed && !retried) {
3080                 failed = 0;
3081                 retried = true;
3082                 goto again;
3083         } else if (WARN_ON(failed && retried)) {
3084                 ret = -ENOSPC;
3085         }
3086 error:
3087         btrfs_free_path(path);
3088         return ret;
3089 }
3090
3091 /*
3092  * return 1 : allocate a data chunk successfully,
3093  * return <0: errors during allocating a data chunk,
3094  * return 0 : no need to allocate a data chunk.
3095  */
3096 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3097                                       u64 chunk_offset)
3098 {
3099         struct btrfs_block_group_cache *cache;
3100         u64 bytes_used;
3101         u64 chunk_type;
3102
3103         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3104         ASSERT(cache);
3105         chunk_type = cache->flags;
3106         btrfs_put_block_group(cache);
3107
3108         if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3109                 spin_lock(&fs_info->data_sinfo->lock);
3110                 bytes_used = fs_info->data_sinfo->bytes_used;
3111                 spin_unlock(&fs_info->data_sinfo->lock);
3112
3113                 if (!bytes_used) {
3114                         struct btrfs_trans_handle *trans;
3115                         int ret;
3116
3117                         trans = btrfs_join_transaction(fs_info->tree_root);
3118                         if (IS_ERR(trans))
3119                                 return PTR_ERR(trans);
3120
3121                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3122                                                       BTRFS_BLOCK_GROUP_DATA);
3123                         btrfs_end_transaction(trans);
3124                         if (ret < 0)
3125                                 return ret;
3126
3127                         return 1;
3128                 }
3129         }
3130         return 0;
3131 }
3132
3133 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3134                                struct btrfs_balance_control *bctl)
3135 {
3136         struct btrfs_root *root = fs_info->tree_root;
3137         struct btrfs_trans_handle *trans;
3138         struct btrfs_balance_item *item;
3139         struct btrfs_disk_balance_args disk_bargs;
3140         struct btrfs_path *path;
3141         struct extent_buffer *leaf;
3142         struct btrfs_key key;
3143         int ret, err;
3144
3145         path = btrfs_alloc_path();
3146         if (!path)
3147                 return -ENOMEM;
3148
3149         trans = btrfs_start_transaction(root, 0);
3150         if (IS_ERR(trans)) {
3151                 btrfs_free_path(path);
3152                 return PTR_ERR(trans);
3153         }
3154
3155         key.objectid = BTRFS_BALANCE_OBJECTID;
3156         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3157         key.offset = 0;
3158
3159         ret = btrfs_insert_empty_item(trans, root, path, &key,
3160                                       sizeof(*item));
3161         if (ret)
3162                 goto out;
3163
3164         leaf = path->nodes[0];
3165         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3166
3167         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3168
3169         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3170         btrfs_set_balance_data(leaf, item, &disk_bargs);
3171         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3172         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3173         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3174         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3175
3176         btrfs_set_balance_flags(leaf, item, bctl->flags);
3177
3178         btrfs_mark_buffer_dirty(leaf);
3179 out:
3180         btrfs_free_path(path);
3181         err = btrfs_commit_transaction(trans);
3182         if (err && !ret)
3183                 ret = err;
3184         return ret;
3185 }
3186
3187 static int del_balance_item(struct btrfs_fs_info *fs_info)
3188 {
3189         struct btrfs_root *root = fs_info->tree_root;
3190         struct btrfs_trans_handle *trans;
3191         struct btrfs_path *path;
3192         struct btrfs_key key;
3193         int ret, err;
3194
3195         path = btrfs_alloc_path();
3196         if (!path)
3197                 return -ENOMEM;
3198
3199         trans = btrfs_start_transaction(root, 0);
3200         if (IS_ERR(trans)) {
3201                 btrfs_free_path(path);
3202                 return PTR_ERR(trans);
3203         }
3204
3205         key.objectid = BTRFS_BALANCE_OBJECTID;
3206         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3207         key.offset = 0;
3208
3209         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3210         if (ret < 0)
3211                 goto out;
3212         if (ret > 0) {
3213                 ret = -ENOENT;
3214                 goto out;
3215         }
3216
3217         ret = btrfs_del_item(trans, root, path);
3218 out:
3219         btrfs_free_path(path);
3220         err = btrfs_commit_transaction(trans);
3221         if (err && !ret)
3222                 ret = err;
3223         return ret;
3224 }
3225
3226 /*
3227  * This is a heuristic used to reduce the number of chunks balanced on
3228  * resume after balance was interrupted.
3229  */
3230 static void update_balance_args(struct btrfs_balance_control *bctl)
3231 {
3232         /*
3233          * Turn on soft mode for chunk types that were being converted.
3234          */
3235         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3236                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3237         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3238                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3239         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3240                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3241
3242         /*
3243          * Turn on usage filter if is not already used.  The idea is
3244          * that chunks that we have already balanced should be
3245          * reasonably full.  Don't do it for chunks that are being
3246          * converted - that will keep us from relocating unconverted
3247          * (albeit full) chunks.
3248          */
3249         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3250             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3251             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3252                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3253                 bctl->data.usage = 90;
3254         }
3255         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3256             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3257             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3258                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3259                 bctl->sys.usage = 90;
3260         }
3261         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3262             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3263             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3264                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3265                 bctl->meta.usage = 90;
3266         }
3267 }
3268
3269 /*
3270  * Should be called with both balance and volume mutexes held to
3271  * serialize other volume operations (add_dev/rm_dev/resize) with
3272  * restriper.  Same goes for unset_balance_control.
3273  */
3274 static void set_balance_control(struct btrfs_balance_control *bctl)
3275 {
3276         struct btrfs_fs_info *fs_info = bctl->fs_info;
3277
3278         BUG_ON(fs_info->balance_ctl);
3279
3280         spin_lock(&fs_info->balance_lock);
3281         fs_info->balance_ctl = bctl;
3282         spin_unlock(&fs_info->balance_lock);
3283 }
3284
3285 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3286 {
3287         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3288
3289         BUG_ON(!fs_info->balance_ctl);
3290
3291         spin_lock(&fs_info->balance_lock);
3292         fs_info->balance_ctl = NULL;
3293         spin_unlock(&fs_info->balance_lock);
3294
3295         kfree(bctl);
3296 }
3297
3298 /*
3299  * Balance filters.  Return 1 if chunk should be filtered out
3300  * (should not be balanced).
3301  */
3302 static int chunk_profiles_filter(u64 chunk_type,
3303                                  struct btrfs_balance_args *bargs)
3304 {
3305         chunk_type = chunk_to_extended(chunk_type) &
3306                                 BTRFS_EXTENDED_PROFILE_MASK;
3307
3308         if (bargs->profiles & chunk_type)
3309                 return 0;
3310
3311         return 1;
3312 }
3313
3314 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3315                               struct btrfs_balance_args *bargs)
3316 {
3317         struct btrfs_block_group_cache *cache;
3318         u64 chunk_used;
3319         u64 user_thresh_min;
3320         u64 user_thresh_max;
3321         int ret = 1;
3322
3323         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3324         chunk_used = btrfs_block_group_used(&cache->item);
3325
3326         if (bargs->usage_min == 0)
3327                 user_thresh_min = 0;
3328         else
3329                 user_thresh_min = div_factor_fine(cache->key.offset,
3330                                         bargs->usage_min);
3331
3332         if (bargs->usage_max == 0)
3333                 user_thresh_max = 1;
3334         else if (bargs->usage_max > 100)
3335                 user_thresh_max = cache->key.offset;
3336         else
3337                 user_thresh_max = div_factor_fine(cache->key.offset,
3338                                         bargs->usage_max);
3339
3340         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3341                 ret = 0;
3342
3343         btrfs_put_block_group(cache);
3344         return ret;
3345 }
3346
3347 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3348                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3349 {
3350         struct btrfs_block_group_cache *cache;
3351         u64 chunk_used, user_thresh;
3352         int ret = 1;
3353
3354         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3355         chunk_used = btrfs_block_group_used(&cache->item);
3356
3357         if (bargs->usage_min == 0)
3358                 user_thresh = 1;
3359         else if (bargs->usage > 100)
3360                 user_thresh = cache->key.offset;
3361         else
3362                 user_thresh = div_factor_fine(cache->key.offset,
3363                                               bargs->usage);
3364
3365         if (chunk_used < user_thresh)
3366                 ret = 0;
3367
3368         btrfs_put_block_group(cache);
3369         return ret;
3370 }
3371
3372 static int chunk_devid_filter(struct extent_buffer *leaf,
3373                               struct btrfs_chunk *chunk,
3374                               struct btrfs_balance_args *bargs)
3375 {
3376         struct btrfs_stripe *stripe;
3377         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3378         int i;
3379
3380         for (i = 0; i < num_stripes; i++) {
3381                 stripe = btrfs_stripe_nr(chunk, i);
3382                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3383                         return 0;
3384         }
3385
3386         return 1;
3387 }
3388
3389 /* [pstart, pend) */
3390 static int chunk_drange_filter(struct extent_buffer *leaf,
3391                                struct btrfs_chunk *chunk,
3392                                struct btrfs_balance_args *bargs)
3393 {
3394         struct btrfs_stripe *stripe;
3395         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3396         u64 stripe_offset;
3397         u64 stripe_length;
3398         int factor;
3399         int i;
3400
3401         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3402                 return 0;
3403
3404         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3405              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3406                 factor = num_stripes / 2;
3407         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3408                 factor = num_stripes - 1;
3409         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3410                 factor = num_stripes - 2;
3411         } else {
3412                 factor = num_stripes;
3413         }
3414
3415         for (i = 0; i < num_stripes; i++) {
3416                 stripe = btrfs_stripe_nr(chunk, i);
3417                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3418                         continue;
3419
3420                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3421                 stripe_length = btrfs_chunk_length(leaf, chunk);
3422                 stripe_length = div_u64(stripe_length, factor);
3423
3424                 if (stripe_offset < bargs->pend &&
3425                     stripe_offset + stripe_length > bargs->pstart)
3426                         return 0;
3427         }
3428
3429         return 1;
3430 }
3431
3432 /* [vstart, vend) */
3433 static int chunk_vrange_filter(struct extent_buffer *leaf,
3434                                struct btrfs_chunk *chunk,
3435                                u64 chunk_offset,
3436                                struct btrfs_balance_args *bargs)
3437 {
3438         if (chunk_offset < bargs->vend &&
3439             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3440                 /* at least part of the chunk is inside this vrange */
3441                 return 0;
3442
3443         return 1;
3444 }
3445
3446 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3447                                struct btrfs_chunk *chunk,
3448                                struct btrfs_balance_args *bargs)
3449 {
3450         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3451
3452         if (bargs->stripes_min <= num_stripes
3453                         && num_stripes <= bargs->stripes_max)
3454                 return 0;
3455
3456         return 1;
3457 }
3458
3459 static int chunk_soft_convert_filter(u64 chunk_type,
3460                                      struct btrfs_balance_args *bargs)
3461 {
3462         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3463                 return 0;
3464
3465         chunk_type = chunk_to_extended(chunk_type) &
3466                                 BTRFS_EXTENDED_PROFILE_MASK;
3467
3468         if (bargs->target == chunk_type)
3469                 return 1;
3470
3471         return 0;
3472 }
3473
3474 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3475                                 struct extent_buffer *leaf,
3476                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3477 {
3478         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3479         struct btrfs_balance_args *bargs = NULL;
3480         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3481
3482         /* type filter */
3483         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3484               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3485                 return 0;
3486         }
3487
3488         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3489                 bargs = &bctl->data;
3490         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3491                 bargs = &bctl->sys;
3492         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3493                 bargs = &bctl->meta;
3494
3495         /* profiles filter */
3496         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3497             chunk_profiles_filter(chunk_type, bargs)) {
3498                 return 0;
3499         }
3500
3501         /* usage filter */
3502         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3503             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3504                 return 0;
3505         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3506             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3507                 return 0;
3508         }
3509
3510         /* devid filter */
3511         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3512             chunk_devid_filter(leaf, chunk, bargs)) {
3513                 return 0;
3514         }
3515
3516         /* drange filter, makes sense only with devid filter */
3517         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3518             chunk_drange_filter(leaf, chunk, bargs)) {
3519                 return 0;
3520         }
3521
3522         /* vrange filter */
3523         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3524             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3525                 return 0;
3526         }
3527
3528         /* stripes filter */
3529         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3530             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3531                 return 0;
3532         }
3533
3534         /* soft profile changing mode */
3535         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3536             chunk_soft_convert_filter(chunk_type, bargs)) {
3537                 return 0;
3538         }
3539
3540         /*
3541          * limited by count, must be the last filter
3542          */
3543         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3544                 if (bargs->limit == 0)
3545                         return 0;
3546                 else
3547                         bargs->limit--;
3548         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3549                 /*
3550                  * Same logic as the 'limit' filter; the minimum cannot be
3551                  * determined here because we do not have the global information
3552                  * about the count of all chunks that satisfy the filters.
3553                  */
3554                 if (bargs->limit_max == 0)
3555                         return 0;
3556                 else
3557                         bargs->limit_max--;
3558         }
3559
3560         return 1;
3561 }
3562
3563 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3564 {
3565         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3566         struct btrfs_root *chunk_root = fs_info->chunk_root;
3567         struct btrfs_root *dev_root = fs_info->dev_root;
3568         struct list_head *devices;
3569         struct btrfs_device *device;
3570         u64 old_size;
3571         u64 size_to_free;
3572         u64 chunk_type;
3573         struct btrfs_chunk *chunk;
3574         struct btrfs_path *path = NULL;
3575         struct btrfs_key key;
3576         struct btrfs_key found_key;
3577         struct btrfs_trans_handle *trans;
3578         struct extent_buffer *leaf;
3579         int slot;
3580         int ret;
3581         int enospc_errors = 0;
3582         bool counting = true;
3583         /* The single value limit and min/max limits use the same bytes in the */
3584         u64 limit_data = bctl->data.limit;
3585         u64 limit_meta = bctl->meta.limit;
3586         u64 limit_sys = bctl->sys.limit;
3587         u32 count_data = 0;
3588         u32 count_meta = 0;
3589         u32 count_sys = 0;
3590         int chunk_reserved = 0;
3591
3592         /* step one make some room on all the devices */
3593         devices = &fs_info->fs_devices->devices;
3594         list_for_each_entry(device, devices, dev_list) {
3595                 old_size = btrfs_device_get_total_bytes(device);
3596                 size_to_free = div_factor(old_size, 1);
3597                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3598                 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3599                     btrfs_device_get_total_bytes(device) -
3600                     btrfs_device_get_bytes_used(device) > size_to_free ||
3601                     test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3602                         continue;
3603
3604                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3605                 if (ret == -ENOSPC)
3606                         break;
3607                 if (ret) {
3608                         /* btrfs_shrink_device never returns ret > 0 */
3609                         WARN_ON(ret > 0);
3610                         goto error;
3611                 }
3612
3613                 trans = btrfs_start_transaction(dev_root, 0);
3614                 if (IS_ERR(trans)) {
3615                         ret = PTR_ERR(trans);
3616                         btrfs_info_in_rcu(fs_info,
3617                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3618                                           rcu_str_deref(device->name), ret,
3619                                           old_size, old_size - size_to_free);
3620                         goto error;
3621                 }
3622
3623                 ret = btrfs_grow_device(trans, device, old_size);
3624                 if (ret) {
3625                         btrfs_end_transaction(trans);
3626                         /* btrfs_grow_device never returns ret > 0 */
3627                         WARN_ON(ret > 0);
3628                         btrfs_info_in_rcu(fs_info,
3629                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3630                                           rcu_str_deref(device->name), ret,
3631                                           old_size, old_size - size_to_free);
3632                         goto error;
3633                 }
3634
3635                 btrfs_end_transaction(trans);
3636         }
3637
3638         /* step two, relocate all the chunks */
3639         path = btrfs_alloc_path();
3640         if (!path) {
3641                 ret = -ENOMEM;
3642                 goto error;
3643         }
3644
3645         /* zero out stat counters */
3646         spin_lock(&fs_info->balance_lock);
3647         memset(&bctl->stat, 0, sizeof(bctl->stat));
3648         spin_unlock(&fs_info->balance_lock);
3649 again:
3650         if (!counting) {
3651                 /*
3652                  * The single value limit and min/max limits use the same bytes
3653                  * in the
3654                  */
3655                 bctl->data.limit = limit_data;
3656                 bctl->meta.limit = limit_meta;
3657                 bctl->sys.limit = limit_sys;
3658         }
3659         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3660         key.offset = (u64)-1;
3661         key.type = BTRFS_CHUNK_ITEM_KEY;
3662
3663         while (1) {
3664                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3665                     atomic_read(&fs_info->balance_cancel_req)) {
3666                         ret = -ECANCELED;
3667                         goto error;
3668                 }
3669
3670                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3671                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3672                 if (ret < 0) {
3673                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3674                         goto error;
3675                 }
3676
3677                 /*
3678                  * this shouldn't happen, it means the last relocate
3679                  * failed
3680                  */
3681                 if (ret == 0)
3682                         BUG(); /* FIXME break ? */
3683
3684                 ret = btrfs_previous_item(chunk_root, path, 0,
3685                                           BTRFS_CHUNK_ITEM_KEY);
3686                 if (ret) {
3687                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3688                         ret = 0;
3689                         break;
3690                 }
3691
3692                 leaf = path->nodes[0];
3693                 slot = path->slots[0];
3694                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3695
3696                 if (found_key.objectid != key.objectid) {
3697                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3698                         break;
3699                 }
3700
3701                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3702                 chunk_type = btrfs_chunk_type(leaf, chunk);
3703
3704                 if (!counting) {
3705                         spin_lock(&fs_info->balance_lock);
3706                         bctl->stat.considered++;
3707                         spin_unlock(&fs_info->balance_lock);
3708                 }
3709
3710                 ret = should_balance_chunk(fs_info, leaf, chunk,
3711                                            found_key.offset);
3712
3713                 btrfs_release_path(path);
3714                 if (!ret) {
3715                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3716                         goto loop;
3717                 }
3718
3719                 if (counting) {
3720                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3721                         spin_lock(&fs_info->balance_lock);
3722                         bctl->stat.expected++;
3723                         spin_unlock(&fs_info->balance_lock);
3724
3725                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3726                                 count_data++;
3727                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3728                                 count_sys++;
3729                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3730                                 count_meta++;
3731
3732                         goto loop;
3733                 }
3734
3735                 /*
3736                  * Apply limit_min filter, no need to check if the LIMITS
3737                  * filter is used, limit_min is 0 by default
3738                  */
3739                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3740                                         count_data < bctl->data.limit_min)
3741                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3742                                         count_meta < bctl->meta.limit_min)
3743                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3744                                         count_sys < bctl->sys.limit_min)) {
3745                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3746                         goto loop;
3747 &