btrfs: remove redundant btrfs_balance_control::fs_info
[sfrench/cifs-2.6.git] / fs / btrfs / volumes.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/sched.h>
7 #include <linux/bio.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/iocontext.h>
12 #include <linux/capability.h>
13 #include <linux/ratelimit.h>
14 #include <linux/kthread.h>
15 #include <linux/raid/pq.h>
16 #include <linux/semaphore.h>
17 #include <linux/uuid.h>
18 #include <linux/list_sort.h>
19 #include <asm/div64.h>
20 #include "ctree.h"
21 #include "extent_map.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24 #include "print-tree.h"
25 #include "volumes.h"
26 #include "raid56.h"
27 #include "async-thread.h"
28 #include "check-integrity.h"
29 #include "rcu-string.h"
30 #include "math.h"
31 #include "dev-replace.h"
32 #include "sysfs.h"
33
34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
35         [BTRFS_RAID_RAID10] = {
36                 .sub_stripes    = 2,
37                 .dev_stripes    = 1,
38                 .devs_max       = 0,    /* 0 == as many as possible */
39                 .devs_min       = 4,
40                 .tolerated_failures = 1,
41                 .devs_increment = 2,
42                 .ncopies        = 2,
43                 .raid_name      = "raid10",
44                 .bg_flag        = BTRFS_BLOCK_GROUP_RAID10,
45                 .mindev_error   = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
46         },
47         [BTRFS_RAID_RAID1] = {
48                 .sub_stripes    = 1,
49                 .dev_stripes    = 1,
50                 .devs_max       = 2,
51                 .devs_min       = 2,
52                 .tolerated_failures = 1,
53                 .devs_increment = 2,
54                 .ncopies        = 2,
55                 .raid_name      = "raid1",
56                 .bg_flag        = BTRFS_BLOCK_GROUP_RAID1,
57                 .mindev_error   = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
58         },
59         [BTRFS_RAID_DUP] = {
60                 .sub_stripes    = 1,
61                 .dev_stripes    = 2,
62                 .devs_max       = 1,
63                 .devs_min       = 1,
64                 .tolerated_failures = 0,
65                 .devs_increment = 1,
66                 .ncopies        = 2,
67                 .raid_name      = "dup",
68                 .bg_flag        = BTRFS_BLOCK_GROUP_DUP,
69                 .mindev_error   = 0,
70         },
71         [BTRFS_RAID_RAID0] = {
72                 .sub_stripes    = 1,
73                 .dev_stripes    = 1,
74                 .devs_max       = 0,
75                 .devs_min       = 2,
76                 .tolerated_failures = 0,
77                 .devs_increment = 1,
78                 .ncopies        = 1,
79                 .raid_name      = "raid0",
80                 .bg_flag        = BTRFS_BLOCK_GROUP_RAID0,
81                 .mindev_error   = 0,
82         },
83         [BTRFS_RAID_SINGLE] = {
84                 .sub_stripes    = 1,
85                 .dev_stripes    = 1,
86                 .devs_max       = 1,
87                 .devs_min       = 1,
88                 .tolerated_failures = 0,
89                 .devs_increment = 1,
90                 .ncopies        = 1,
91                 .raid_name      = "single",
92                 .bg_flag        = 0,
93                 .mindev_error   = 0,
94         },
95         [BTRFS_RAID_RAID5] = {
96                 .sub_stripes    = 1,
97                 .dev_stripes    = 1,
98                 .devs_max       = 0,
99                 .devs_min       = 2,
100                 .tolerated_failures = 1,
101                 .devs_increment = 1,
102                 .ncopies        = 2,
103                 .raid_name      = "raid5",
104                 .bg_flag        = BTRFS_BLOCK_GROUP_RAID5,
105                 .mindev_error   = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
106         },
107         [BTRFS_RAID_RAID6] = {
108                 .sub_stripes    = 1,
109                 .dev_stripes    = 1,
110                 .devs_max       = 0,
111                 .devs_min       = 3,
112                 .tolerated_failures = 2,
113                 .devs_increment = 1,
114                 .ncopies        = 3,
115                 .raid_name      = "raid6",
116                 .bg_flag        = BTRFS_BLOCK_GROUP_RAID6,
117                 .mindev_error   = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
118         },
119 };
120
121 const char *get_raid_name(enum btrfs_raid_types type)
122 {
123         if (type >= BTRFS_NR_RAID_TYPES)
124                 return NULL;
125
126         return btrfs_raid_array[type].raid_name;
127 }
128
129 static int init_first_rw_device(struct btrfs_trans_handle *trans,
130                                 struct btrfs_fs_info *fs_info);
131 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
132 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
133 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
134 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
135 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
136                              enum btrfs_map_op op,
137                              u64 logical, u64 *length,
138                              struct btrfs_bio **bbio_ret,
139                              int mirror_num, int need_raid_map);
140
141 /*
142  * Device locking
143  * ==============
144  *
145  * There are several mutexes that protect manipulation of devices and low-level
146  * structures like chunks but not block groups, extents or files
147  *
148  * uuid_mutex (global lock)
149  * ------------------------
150  * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
151  * the SCAN_DEV ioctl registration or from mount either implicitly (the first
152  * device) or requested by the device= mount option
153  *
154  * the mutex can be very coarse and can cover long-running operations
155  *
156  * protects: updates to fs_devices counters like missing devices, rw devices,
157  * seeding, structure cloning, openning/closing devices at mount/umount time
158  *
159  * global::fs_devs - add, remove, updates to the global list
160  *
161  * does not protect: manipulation of the fs_devices::devices list!
162  *
163  * btrfs_device::name - renames (write side), read is RCU
164  *
165  * fs_devices::device_list_mutex (per-fs, with RCU)
166  * ------------------------------------------------
167  * protects updates to fs_devices::devices, ie. adding and deleting
168  *
169  * simple list traversal with read-only actions can be done with RCU protection
170  *
171  * may be used to exclude some operations from running concurrently without any
172  * modifications to the list (see write_all_supers)
173  *
174  * balance_mutex
175  * -------------
176  * protects balance structures (status, state) and context accessed from
177  * several places (internally, ioctl)
178  *
179  * chunk_mutex
180  * -----------
181  * protects chunks, adding or removing during allocation, trim or when a new
182  * device is added/removed
183  *
184  * cleaner_mutex
185  * -------------
186  * a big lock that is held by the cleaner thread and prevents running subvolume
187  * cleaning together with relocation or delayed iputs
188  *
189  *
190  * Lock nesting
191  * ============
192  *
193  * uuid_mutex
194  *   volume_mutex
195  *     device_list_mutex
196  *       chunk_mutex
197  *     balance_mutex
198  *
199  *
200  * Exclusive operations, BTRFS_FS_EXCL_OP
201  * ======================================
202  *
203  * Maintains the exclusivity of the following operations that apply to the
204  * whole filesystem and cannot run in parallel.
205  *
206  * - Balance (*)
207  * - Device add
208  * - Device remove
209  * - Device replace (*)
210  * - Resize
211  *
212  * The device operations (as above) can be in one of the following states:
213  *
214  * - Running state
215  * - Paused state
216  * - Completed state
217  *
218  * Only device operations marked with (*) can go into the Paused state for the
219  * following reasons:
220  *
221  * - ioctl (only Balance can be Paused through ioctl)
222  * - filesystem remounted as read-only
223  * - filesystem unmounted and mounted as read-only
224  * - system power-cycle and filesystem mounted as read-only
225  * - filesystem or device errors leading to forced read-only
226  *
227  * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
228  * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
229  * A device operation in Paused or Running state can be canceled or resumed
230  * either by ioctl (Balance only) or when remounted as read-write.
231  * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
232  * completed.
233  */
234
235 DEFINE_MUTEX(uuid_mutex);
236 static LIST_HEAD(fs_uuids);
237 struct list_head *btrfs_get_fs_uuids(void)
238 {
239         return &fs_uuids;
240 }
241
242 /*
243  * alloc_fs_devices - allocate struct btrfs_fs_devices
244  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
245  *
246  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
247  * The returned struct is not linked onto any lists and can be destroyed with
248  * kfree() right away.
249  */
250 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
251 {
252         struct btrfs_fs_devices *fs_devs;
253
254         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
255         if (!fs_devs)
256                 return ERR_PTR(-ENOMEM);
257
258         mutex_init(&fs_devs->device_list_mutex);
259
260         INIT_LIST_HEAD(&fs_devs->devices);
261         INIT_LIST_HEAD(&fs_devs->resized_devices);
262         INIT_LIST_HEAD(&fs_devs->alloc_list);
263         INIT_LIST_HEAD(&fs_devs->fs_list);
264         if (fsid)
265                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
266
267         return fs_devs;
268 }
269
270 void btrfs_free_device(struct btrfs_device *device)
271 {
272         rcu_string_free(device->name);
273         bio_put(device->flush_bio);
274         kfree(device);
275 }
276
277 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
278 {
279         struct btrfs_device *device;
280         WARN_ON(fs_devices->opened);
281         while (!list_empty(&fs_devices->devices)) {
282                 device = list_entry(fs_devices->devices.next,
283                                     struct btrfs_device, dev_list);
284                 list_del(&device->dev_list);
285                 btrfs_free_device(device);
286         }
287         kfree(fs_devices);
288 }
289
290 static void btrfs_kobject_uevent(struct block_device *bdev,
291                                  enum kobject_action action)
292 {
293         int ret;
294
295         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
296         if (ret)
297                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
298                         action,
299                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
300                         &disk_to_dev(bdev->bd_disk)->kobj);
301 }
302
303 void __exit btrfs_cleanup_fs_uuids(void)
304 {
305         struct btrfs_fs_devices *fs_devices;
306
307         while (!list_empty(&fs_uuids)) {
308                 fs_devices = list_entry(fs_uuids.next,
309                                         struct btrfs_fs_devices, fs_list);
310                 list_del(&fs_devices->fs_list);
311                 free_fs_devices(fs_devices);
312         }
313 }
314
315 /*
316  * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
317  * Returned struct is not linked onto any lists and must be destroyed using
318  * btrfs_free_device.
319  */
320 static struct btrfs_device *__alloc_device(void)
321 {
322         struct btrfs_device *dev;
323
324         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
325         if (!dev)
326                 return ERR_PTR(-ENOMEM);
327
328         /*
329          * Preallocate a bio that's always going to be used for flushing device
330          * barriers and matches the device lifespan
331          */
332         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
333         if (!dev->flush_bio) {
334                 kfree(dev);
335                 return ERR_PTR(-ENOMEM);
336         }
337
338         INIT_LIST_HEAD(&dev->dev_list);
339         INIT_LIST_HEAD(&dev->dev_alloc_list);
340         INIT_LIST_HEAD(&dev->resized_list);
341
342         spin_lock_init(&dev->io_lock);
343
344         atomic_set(&dev->reada_in_flight, 0);
345         atomic_set(&dev->dev_stats_ccnt, 0);
346         btrfs_device_data_ordered_init(dev);
347         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
348         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
349
350         return dev;
351 }
352
353 /*
354  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
355  * return NULL.
356  *
357  * If devid and uuid are both specified, the match must be exact, otherwise
358  * only devid is used.
359  */
360 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
361                 u64 devid, const u8 *uuid)
362 {
363         struct btrfs_device *dev;
364
365         list_for_each_entry(dev, &fs_devices->devices, dev_list) {
366                 if (dev->devid == devid &&
367                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
368                         return dev;
369                 }
370         }
371         return NULL;
372 }
373
374 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
375 {
376         struct btrfs_fs_devices *fs_devices;
377
378         list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
379                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
380                         return fs_devices;
381         }
382         return NULL;
383 }
384
385 static int
386 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
387                       int flush, struct block_device **bdev,
388                       struct buffer_head **bh)
389 {
390         int ret;
391
392         *bdev = blkdev_get_by_path(device_path, flags, holder);
393
394         if (IS_ERR(*bdev)) {
395                 ret = PTR_ERR(*bdev);
396                 goto error;
397         }
398
399         if (flush)
400                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
401         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
402         if (ret) {
403                 blkdev_put(*bdev, flags);
404                 goto error;
405         }
406         invalidate_bdev(*bdev);
407         *bh = btrfs_read_dev_super(*bdev);
408         if (IS_ERR(*bh)) {
409                 ret = PTR_ERR(*bh);
410                 blkdev_put(*bdev, flags);
411                 goto error;
412         }
413
414         return 0;
415
416 error:
417         *bdev = NULL;
418         *bh = NULL;
419         return ret;
420 }
421
422 static void requeue_list(struct btrfs_pending_bios *pending_bios,
423                         struct bio *head, struct bio *tail)
424 {
425
426         struct bio *old_head;
427
428         old_head = pending_bios->head;
429         pending_bios->head = head;
430         if (pending_bios->tail)
431                 tail->bi_next = old_head;
432         else
433                 pending_bios->tail = tail;
434 }
435
436 /*
437  * we try to collect pending bios for a device so we don't get a large
438  * number of procs sending bios down to the same device.  This greatly
439  * improves the schedulers ability to collect and merge the bios.
440  *
441  * But, it also turns into a long list of bios to process and that is sure
442  * to eventually make the worker thread block.  The solution here is to
443  * make some progress and then put this work struct back at the end of
444  * the list if the block device is congested.  This way, multiple devices
445  * can make progress from a single worker thread.
446  */
447 static noinline void run_scheduled_bios(struct btrfs_device *device)
448 {
449         struct btrfs_fs_info *fs_info = device->fs_info;
450         struct bio *pending;
451         struct backing_dev_info *bdi;
452         struct btrfs_pending_bios *pending_bios;
453         struct bio *tail;
454         struct bio *cur;
455         int again = 0;
456         unsigned long num_run;
457         unsigned long batch_run = 0;
458         unsigned long last_waited = 0;
459         int force_reg = 0;
460         int sync_pending = 0;
461         struct blk_plug plug;
462
463         /*
464          * this function runs all the bios we've collected for
465          * a particular device.  We don't want to wander off to
466          * another device without first sending all of these down.
467          * So, setup a plug here and finish it off before we return
468          */
469         blk_start_plug(&plug);
470
471         bdi = device->bdev->bd_bdi;
472
473 loop:
474         spin_lock(&device->io_lock);
475
476 loop_lock:
477         num_run = 0;
478
479         /* take all the bios off the list at once and process them
480          * later on (without the lock held).  But, remember the
481          * tail and other pointers so the bios can be properly reinserted
482          * into the list if we hit congestion
483          */
484         if (!force_reg && device->pending_sync_bios.head) {
485                 pending_bios = &device->pending_sync_bios;
486                 force_reg = 1;
487         } else {
488                 pending_bios = &device->pending_bios;
489                 force_reg = 0;
490         }
491
492         pending = pending_bios->head;
493         tail = pending_bios->tail;
494         WARN_ON(pending && !tail);
495
496         /*
497          * if pending was null this time around, no bios need processing
498          * at all and we can stop.  Otherwise it'll loop back up again
499          * and do an additional check so no bios are missed.
500          *
501          * device->running_pending is used to synchronize with the
502          * schedule_bio code.
503          */
504         if (device->pending_sync_bios.head == NULL &&
505             device->pending_bios.head == NULL) {
506                 again = 0;
507                 device->running_pending = 0;
508         } else {
509                 again = 1;
510                 device->running_pending = 1;
511         }
512
513         pending_bios->head = NULL;
514         pending_bios->tail = NULL;
515
516         spin_unlock(&device->io_lock);
517
518         while (pending) {
519
520                 rmb();
521                 /* we want to work on both lists, but do more bios on the
522                  * sync list than the regular list
523                  */
524                 if ((num_run > 32 &&
525                     pending_bios != &device->pending_sync_bios &&
526                     device->pending_sync_bios.head) ||
527                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
528                     device->pending_bios.head)) {
529                         spin_lock(&device->io_lock);
530                         requeue_list(pending_bios, pending, tail);
531                         goto loop_lock;
532                 }
533
534                 cur = pending;
535                 pending = pending->bi_next;
536                 cur->bi_next = NULL;
537
538                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
539
540                 /*
541                  * if we're doing the sync list, record that our
542                  * plug has some sync requests on it
543                  *
544                  * If we're doing the regular list and there are
545                  * sync requests sitting around, unplug before
546                  * we add more
547                  */
548                 if (pending_bios == &device->pending_sync_bios) {
549                         sync_pending = 1;
550                 } else if (sync_pending) {
551                         blk_finish_plug(&plug);
552                         blk_start_plug(&plug);
553                         sync_pending = 0;
554                 }
555
556                 btrfsic_submit_bio(cur);
557                 num_run++;
558                 batch_run++;
559
560                 cond_resched();
561
562                 /*
563                  * we made progress, there is more work to do and the bdi
564                  * is now congested.  Back off and let other work structs
565                  * run instead
566                  */
567                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
568                     fs_info->fs_devices->open_devices > 1) {
569                         struct io_context *ioc;
570
571                         ioc = current->io_context;
572
573                         /*
574                          * the main goal here is that we don't want to
575                          * block if we're going to be able to submit
576                          * more requests without blocking.
577                          *
578                          * This code does two great things, it pokes into
579                          * the elevator code from a filesystem _and_
580                          * it makes assumptions about how batching works.
581                          */
582                         if (ioc && ioc->nr_batch_requests > 0 &&
583                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
584                             (last_waited == 0 ||
585                              ioc->last_waited == last_waited)) {
586                                 /*
587                                  * we want to go through our batch of
588                                  * requests and stop.  So, we copy out
589                                  * the ioc->last_waited time and test
590                                  * against it before looping
591                                  */
592                                 last_waited = ioc->last_waited;
593                                 cond_resched();
594                                 continue;
595                         }
596                         spin_lock(&device->io_lock);
597                         requeue_list(pending_bios, pending, tail);
598                         device->running_pending = 1;
599
600                         spin_unlock(&device->io_lock);
601                         btrfs_queue_work(fs_info->submit_workers,
602                                          &device->work);
603                         goto done;
604                 }
605         }
606
607         cond_resched();
608         if (again)
609                 goto loop;
610
611         spin_lock(&device->io_lock);
612         if (device->pending_bios.head || device->pending_sync_bios.head)
613                 goto loop_lock;
614         spin_unlock(&device->io_lock);
615
616 done:
617         blk_finish_plug(&plug);
618 }
619
620 static void pending_bios_fn(struct btrfs_work *work)
621 {
622         struct btrfs_device *device;
623
624         device = container_of(work, struct btrfs_device, work);
625         run_scheduled_bios(device);
626 }
627
628 /*
629  *  Search and remove all stale (devices which are not mounted) devices.
630  *  When both inputs are NULL, it will search and release all stale devices.
631  *  path:       Optional. When provided will it release all unmounted devices
632  *              matching this path only.
633  *  skip_dev:   Optional. Will skip this device when searching for the stale
634  *              devices.
635  */
636 static void btrfs_free_stale_devices(const char *path,
637                                      struct btrfs_device *skip_dev)
638 {
639         struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
640         struct btrfs_device *dev, *tmp_dev;
641
642         list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, fs_list) {
643
644                 if (fs_devs->opened)
645                         continue;
646
647                 list_for_each_entry_safe(dev, tmp_dev,
648                                          &fs_devs->devices, dev_list) {
649                         int not_found = 0;
650
651                         if (skip_dev && skip_dev == dev)
652                                 continue;
653                         if (path && !dev->name)
654                                 continue;
655
656                         rcu_read_lock();
657                         if (path)
658                                 not_found = strcmp(rcu_str_deref(dev->name),
659                                                    path);
660                         rcu_read_unlock();
661                         if (not_found)
662                                 continue;
663
664                         /* delete the stale device */
665                         if (fs_devs->num_devices == 1) {
666                                 btrfs_sysfs_remove_fsid(fs_devs);
667                                 list_del(&fs_devs->fs_list);
668                                 free_fs_devices(fs_devs);
669                                 break;
670                         } else {
671                                 fs_devs->num_devices--;
672                                 list_del(&dev->dev_list);
673                                 btrfs_free_device(dev);
674                         }
675                 }
676         }
677 }
678
679 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
680                         struct btrfs_device *device, fmode_t flags,
681                         void *holder)
682 {
683         struct request_queue *q;
684         struct block_device *bdev;
685         struct buffer_head *bh;
686         struct btrfs_super_block *disk_super;
687         u64 devid;
688         int ret;
689
690         if (device->bdev)
691                 return -EINVAL;
692         if (!device->name)
693                 return -EINVAL;
694
695         ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
696                                     &bdev, &bh);
697         if (ret)
698                 return ret;
699
700         disk_super = (struct btrfs_super_block *)bh->b_data;
701         devid = btrfs_stack_device_id(&disk_super->dev_item);
702         if (devid != device->devid)
703                 goto error_brelse;
704
705         if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
706                 goto error_brelse;
707
708         device->generation = btrfs_super_generation(disk_super);
709
710         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
711                 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
712                 fs_devices->seeding = 1;
713         } else {
714                 if (bdev_read_only(bdev))
715                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
716                 else
717                         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
718         }
719
720         q = bdev_get_queue(bdev);
721         if (!blk_queue_nonrot(q))
722                 fs_devices->rotating = 1;
723
724         device->bdev = bdev;
725         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
726         device->mode = flags;
727
728         fs_devices->open_devices++;
729         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
730             device->devid != BTRFS_DEV_REPLACE_DEVID) {
731                 fs_devices->rw_devices++;
732                 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
733         }
734         brelse(bh);
735
736         return 0;
737
738 error_brelse:
739         brelse(bh);
740         blkdev_put(bdev, flags);
741
742         return -EINVAL;
743 }
744
745 /*
746  * Add new device to list of registered devices
747  *
748  * Returns:
749  * device pointer which was just added or updated when successful
750  * error pointer when failed
751  */
752 static noinline struct btrfs_device *device_list_add(const char *path,
753                            struct btrfs_super_block *disk_super)
754 {
755         struct btrfs_device *device;
756         struct btrfs_fs_devices *fs_devices;
757         struct rcu_string *name;
758         u64 found_transid = btrfs_super_generation(disk_super);
759         u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
760
761         fs_devices = find_fsid(disk_super->fsid);
762         if (!fs_devices) {
763                 fs_devices = alloc_fs_devices(disk_super->fsid);
764                 if (IS_ERR(fs_devices))
765                         return ERR_CAST(fs_devices);
766
767                 list_add(&fs_devices->fs_list, &fs_uuids);
768
769                 device = NULL;
770         } else {
771                 device = find_device(fs_devices, devid,
772                                 disk_super->dev_item.uuid);
773         }
774
775         if (!device) {
776                 if (fs_devices->opened)
777                         return ERR_PTR(-EBUSY);
778
779                 device = btrfs_alloc_device(NULL, &devid,
780                                             disk_super->dev_item.uuid);
781                 if (IS_ERR(device)) {
782                         /* we can safely leave the fs_devices entry around */
783                         return device;
784                 }
785
786                 name = rcu_string_strdup(path, GFP_NOFS);
787                 if (!name) {
788                         btrfs_free_device(device);
789                         return ERR_PTR(-ENOMEM);
790                 }
791                 rcu_assign_pointer(device->name, name);
792
793                 mutex_lock(&fs_devices->device_list_mutex);
794                 list_add_rcu(&device->dev_list, &fs_devices->devices);
795                 fs_devices->num_devices++;
796                 mutex_unlock(&fs_devices->device_list_mutex);
797
798                 device->fs_devices = fs_devices;
799                 btrfs_free_stale_devices(path, device);
800
801                 if (disk_super->label[0])
802                         pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
803                                 disk_super->label, devid, found_transid, path);
804                 else
805                         pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
806                                 disk_super->fsid, devid, found_transid, path);
807
808         } else if (!device->name || strcmp(device->name->str, path)) {
809                 /*
810                  * When FS is already mounted.
811                  * 1. If you are here and if the device->name is NULL that
812                  *    means this device was missing at time of FS mount.
813                  * 2. If you are here and if the device->name is different
814                  *    from 'path' that means either
815                  *      a. The same device disappeared and reappeared with
816                  *         different name. or
817                  *      b. The missing-disk-which-was-replaced, has
818                  *         reappeared now.
819                  *
820                  * We must allow 1 and 2a above. But 2b would be a spurious
821                  * and unintentional.
822                  *
823                  * Further in case of 1 and 2a above, the disk at 'path'
824                  * would have missed some transaction when it was away and
825                  * in case of 2a the stale bdev has to be updated as well.
826                  * 2b must not be allowed at all time.
827                  */
828
829                 /*
830                  * For now, we do allow update to btrfs_fs_device through the
831                  * btrfs dev scan cli after FS has been mounted.  We're still
832                  * tracking a problem where systems fail mount by subvolume id
833                  * when we reject replacement on a mounted FS.
834                  */
835                 if (!fs_devices->opened && found_transid < device->generation) {
836                         /*
837                          * That is if the FS is _not_ mounted and if you
838                          * are here, that means there is more than one
839                          * disk with same uuid and devid.We keep the one
840                          * with larger generation number or the last-in if
841                          * generation are equal.
842                          */
843                         return ERR_PTR(-EEXIST);
844                 }
845
846                 name = rcu_string_strdup(path, GFP_NOFS);
847                 if (!name)
848                         return ERR_PTR(-ENOMEM);
849                 rcu_string_free(device->name);
850                 rcu_assign_pointer(device->name, name);
851                 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
852                         fs_devices->missing_devices--;
853                         clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
854                 }
855         }
856
857         /*
858          * Unmount does not free the btrfs_device struct but would zero
859          * generation along with most of the other members. So just update
860          * it back. We need it to pick the disk with largest generation
861          * (as above).
862          */
863         if (!fs_devices->opened)
864                 device->generation = found_transid;
865
866         fs_devices->total_devices = btrfs_super_num_devices(disk_super);
867
868         return device;
869 }
870
871 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
872 {
873         struct btrfs_fs_devices *fs_devices;
874         struct btrfs_device *device;
875         struct btrfs_device *orig_dev;
876
877         fs_devices = alloc_fs_devices(orig->fsid);
878         if (IS_ERR(fs_devices))
879                 return fs_devices;
880
881         mutex_lock(&orig->device_list_mutex);
882         fs_devices->total_devices = orig->total_devices;
883
884         /* We have held the volume lock, it is safe to get the devices. */
885         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
886                 struct rcu_string *name;
887
888                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
889                                             orig_dev->uuid);
890                 if (IS_ERR(device))
891                         goto error;
892
893                 /*
894                  * This is ok to do without rcu read locked because we hold the
895                  * uuid mutex so nothing we touch in here is going to disappear.
896                  */
897                 if (orig_dev->name) {
898                         name = rcu_string_strdup(orig_dev->name->str,
899                                         GFP_KERNEL);
900                         if (!name) {
901                                 btrfs_free_device(device);
902                                 goto error;
903                         }
904                         rcu_assign_pointer(device->name, name);
905                 }
906
907                 list_add(&device->dev_list, &fs_devices->devices);
908                 device->fs_devices = fs_devices;
909                 fs_devices->num_devices++;
910         }
911         mutex_unlock(&orig->device_list_mutex);
912         return fs_devices;
913 error:
914         mutex_unlock(&orig->device_list_mutex);
915         free_fs_devices(fs_devices);
916         return ERR_PTR(-ENOMEM);
917 }
918
919 /*
920  * After we have read the system tree and know devids belonging to
921  * this filesystem, remove the device which does not belong there.
922  */
923 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
924 {
925         struct btrfs_device *device, *next;
926         struct btrfs_device *latest_dev = NULL;
927
928         mutex_lock(&uuid_mutex);
929 again:
930         /* This is the initialized path, it is safe to release the devices. */
931         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
932                 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
933                                                         &device->dev_state)) {
934                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
935                              &device->dev_state) &&
936                              (!latest_dev ||
937                               device->generation > latest_dev->generation)) {
938                                 latest_dev = device;
939                         }
940                         continue;
941                 }
942
943                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
944                         /*
945                          * In the first step, keep the device which has
946                          * the correct fsid and the devid that is used
947                          * for the dev_replace procedure.
948                          * In the second step, the dev_replace state is
949                          * read from the device tree and it is known
950                          * whether the procedure is really active or
951                          * not, which means whether this device is
952                          * used or whether it should be removed.
953                          */
954                         if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
955                                                   &device->dev_state)) {
956                                 continue;
957                         }
958                 }
959                 if (device->bdev) {
960                         blkdev_put(device->bdev, device->mode);
961                         device->bdev = NULL;
962                         fs_devices->open_devices--;
963                 }
964                 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
965                         list_del_init(&device->dev_alloc_list);
966                         clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
967                         if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
968                                       &device->dev_state))
969                                 fs_devices->rw_devices--;
970                 }
971                 list_del_init(&device->dev_list);
972                 fs_devices->num_devices--;
973                 btrfs_free_device(device);
974         }
975
976         if (fs_devices->seed) {
977                 fs_devices = fs_devices->seed;
978                 goto again;
979         }
980
981         fs_devices->latest_bdev = latest_dev->bdev;
982
983         mutex_unlock(&uuid_mutex);
984 }
985
986 static void free_device_rcu(struct rcu_head *head)
987 {
988         struct btrfs_device *device;
989
990         device = container_of(head, struct btrfs_device, rcu);
991         btrfs_free_device(device);
992 }
993
994 static void btrfs_close_bdev(struct btrfs_device *device)
995 {
996         if (!device->bdev)
997                 return;
998
999         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1000                 sync_blockdev(device->bdev);
1001                 invalidate_bdev(device->bdev);
1002         }
1003
1004         blkdev_put(device->bdev, device->mode);
1005 }
1006
1007 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
1008 {
1009         struct btrfs_fs_devices *fs_devices = device->fs_devices;
1010         struct btrfs_device *new_device;
1011         struct rcu_string *name;
1012
1013         if (device->bdev)
1014                 fs_devices->open_devices--;
1015
1016         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1017             device->devid != BTRFS_DEV_REPLACE_DEVID) {
1018                 list_del_init(&device->dev_alloc_list);
1019                 fs_devices->rw_devices--;
1020         }
1021
1022         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1023                 fs_devices->missing_devices--;
1024
1025         new_device = btrfs_alloc_device(NULL, &device->devid,
1026                                         device->uuid);
1027         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1028
1029         /* Safe because we are under uuid_mutex */
1030         if (device->name) {
1031                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1032                 BUG_ON(!name); /* -ENOMEM */
1033                 rcu_assign_pointer(new_device->name, name);
1034         }
1035
1036         list_replace_rcu(&device->dev_list, &new_device->dev_list);
1037         new_device->fs_devices = device->fs_devices;
1038 }
1039
1040 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1041 {
1042         struct btrfs_device *device, *tmp;
1043         struct list_head pending_put;
1044
1045         INIT_LIST_HEAD(&pending_put);
1046
1047         if (--fs_devices->opened > 0)
1048                 return 0;
1049
1050         mutex_lock(&fs_devices->device_list_mutex);
1051         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1052                 btrfs_prepare_close_one_device(device);
1053                 list_add(&device->dev_list, &pending_put);
1054         }
1055         mutex_unlock(&fs_devices->device_list_mutex);
1056
1057         /*
1058          * btrfs_show_devname() is using the device_list_mutex,
1059          * sometimes call to blkdev_put() leads vfs calling
1060          * into this func. So do put outside of device_list_mutex,
1061          * as of now.
1062          */
1063         while (!list_empty(&pending_put)) {
1064                 device = list_first_entry(&pending_put,
1065                                 struct btrfs_device, dev_list);
1066                 list_del(&device->dev_list);
1067                 btrfs_close_bdev(device);
1068                 call_rcu(&device->rcu, free_device_rcu);
1069         }
1070
1071         WARN_ON(fs_devices->open_devices);
1072         WARN_ON(fs_devices->rw_devices);
1073         fs_devices->opened = 0;
1074         fs_devices->seeding = 0;
1075
1076         return 0;
1077 }
1078
1079 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1080 {
1081         struct btrfs_fs_devices *seed_devices = NULL;
1082         int ret;
1083
1084         mutex_lock(&uuid_mutex);
1085         ret = close_fs_devices(fs_devices);
1086         if (!fs_devices->opened) {
1087                 seed_devices = fs_devices->seed;
1088                 fs_devices->seed = NULL;
1089         }
1090         mutex_unlock(&uuid_mutex);
1091
1092         while (seed_devices) {
1093                 fs_devices = seed_devices;
1094                 seed_devices = fs_devices->seed;
1095                 close_fs_devices(fs_devices);
1096                 free_fs_devices(fs_devices);
1097         }
1098         return ret;
1099 }
1100
1101 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1102                                 fmode_t flags, void *holder)
1103 {
1104         struct btrfs_device *device;
1105         struct btrfs_device *latest_dev = NULL;
1106         int ret = 0;
1107
1108         flags |= FMODE_EXCL;
1109
1110         list_for_each_entry(device, &fs_devices->devices, dev_list) {
1111                 /* Just open everything we can; ignore failures here */
1112                 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1113                         continue;
1114
1115                 if (!latest_dev ||
1116                     device->generation > latest_dev->generation)
1117                         latest_dev = device;
1118         }
1119         if (fs_devices->open_devices == 0) {
1120                 ret = -EINVAL;
1121                 goto out;
1122         }
1123         fs_devices->opened = 1;
1124         fs_devices->latest_bdev = latest_dev->bdev;
1125         fs_devices->total_rw_bytes = 0;
1126 out:
1127         return ret;
1128 }
1129
1130 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1131 {
1132         struct btrfs_device *dev1, *dev2;
1133
1134         dev1 = list_entry(a, struct btrfs_device, dev_list);
1135         dev2 = list_entry(b, struct btrfs_device, dev_list);
1136
1137         if (dev1->devid < dev2->devid)
1138                 return -1;
1139         else if (dev1->devid > dev2->devid)
1140                 return 1;
1141         return 0;
1142 }
1143
1144 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1145                        fmode_t flags, void *holder)
1146 {
1147         int ret;
1148
1149         mutex_lock(&uuid_mutex);
1150         if (fs_devices->opened) {
1151                 fs_devices->opened++;
1152                 ret = 0;
1153         } else {
1154                 list_sort(NULL, &fs_devices->devices, devid_cmp);
1155                 ret = open_fs_devices(fs_devices, flags, holder);
1156         }
1157         mutex_unlock(&uuid_mutex);
1158         return ret;
1159 }
1160
1161 static void btrfs_release_disk_super(struct page *page)
1162 {
1163         kunmap(page);
1164         put_page(page);
1165 }
1166
1167 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1168                                  struct page **page,
1169                                  struct btrfs_super_block **disk_super)
1170 {
1171         void *p;
1172         pgoff_t index;
1173
1174         /* make sure our super fits in the device */
1175         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1176                 return 1;
1177
1178         /* make sure our super fits in the page */
1179         if (sizeof(**disk_super) > PAGE_SIZE)
1180                 return 1;
1181
1182         /* make sure our super doesn't straddle pages on disk */
1183         index = bytenr >> PAGE_SHIFT;
1184         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1185                 return 1;
1186
1187         /* pull in the page with our super */
1188         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1189                                    index, GFP_KERNEL);
1190
1191         if (IS_ERR_OR_NULL(*page))
1192                 return 1;
1193
1194         p = kmap(*page);
1195
1196         /* align our pointer to the offset of the super block */
1197         *disk_super = p + (bytenr & ~PAGE_MASK);
1198
1199         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1200             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1201                 btrfs_release_disk_super(*page);
1202                 return 1;
1203         }
1204
1205         if ((*disk_super)->label[0] &&
1206                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1207                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1208
1209         return 0;
1210 }
1211
1212 /*
1213  * Look for a btrfs signature on a device. This may be called out of the mount path
1214  * and we are not allowed to call set_blocksize during the scan. The superblock
1215  * is read via pagecache
1216  */
1217 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1218                           struct btrfs_fs_devices **fs_devices_ret)
1219 {
1220         struct btrfs_super_block *disk_super;
1221         struct btrfs_device *device;
1222         struct block_device *bdev;
1223         struct page *page;
1224         int ret = 0;
1225         u64 bytenr;
1226
1227         /*
1228          * we would like to check all the supers, but that would make
1229          * a btrfs mount succeed after a mkfs from a different FS.
1230          * So, we need to add a special mount option to scan for
1231          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1232          */
1233         bytenr = btrfs_sb_offset(0);
1234         flags |= FMODE_EXCL;
1235         mutex_lock(&uuid_mutex);
1236
1237         bdev = blkdev_get_by_path(path, flags, holder);
1238         if (IS_ERR(bdev)) {
1239                 ret = PTR_ERR(bdev);
1240                 goto error;
1241         }
1242
1243         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1244                 ret = -EINVAL;
1245                 goto error_bdev_put;
1246         }
1247
1248         device = device_list_add(path, disk_super);
1249         if (IS_ERR(device))
1250                 ret = PTR_ERR(device);
1251         else
1252                 *fs_devices_ret = device->fs_devices;
1253
1254         btrfs_release_disk_super(page);
1255
1256 error_bdev_put:
1257         blkdev_put(bdev, flags);
1258 error:
1259         mutex_unlock(&uuid_mutex);
1260         return ret;
1261 }
1262
1263 /* helper to account the used device space in the range */
1264 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1265                                    u64 end, u64 *length)
1266 {
1267         struct btrfs_key key;
1268         struct btrfs_root *root = device->fs_info->dev_root;
1269         struct btrfs_dev_extent *dev_extent;
1270         struct btrfs_path *path;
1271         u64 extent_end;
1272         int ret;
1273         int slot;
1274         struct extent_buffer *l;
1275
1276         *length = 0;
1277
1278         if (start >= device->total_bytes ||
1279                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1280                 return 0;
1281
1282         path = btrfs_alloc_path();
1283         if (!path)
1284                 return -ENOMEM;
1285         path->reada = READA_FORWARD;
1286
1287         key.objectid = device->devid;
1288         key.offset = start;
1289         key.type = BTRFS_DEV_EXTENT_KEY;
1290
1291         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1292         if (ret < 0)
1293                 goto out;
1294         if (ret > 0) {
1295                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1296                 if (ret < 0)
1297                         goto out;
1298         }
1299
1300         while (1) {
1301                 l = path->nodes[0];
1302                 slot = path->slots[0];
1303                 if (slot >= btrfs_header_nritems(l)) {
1304                         ret = btrfs_next_leaf(root, path);
1305                         if (ret == 0)
1306                                 continue;
1307                         if (ret < 0)
1308                                 goto out;
1309
1310                         break;
1311                 }
1312                 btrfs_item_key_to_cpu(l, &key, slot);
1313
1314                 if (key.objectid < device->devid)
1315                         goto next;
1316
1317                 if (key.objectid > device->devid)
1318                         break;
1319
1320                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1321                         goto next;
1322
1323                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1324                 extent_end = key.offset + btrfs_dev_extent_length(l,
1325                                                                   dev_extent);
1326                 if (key.offset <= start && extent_end > end) {
1327                         *length = end - start + 1;
1328                         break;
1329                 } else if (key.offset <= start && extent_end > start)
1330                         *length += extent_end - start;
1331                 else if (key.offset > start && extent_end <= end)
1332                         *length += extent_end - key.offset;
1333                 else if (key.offset > start && key.offset <= end) {
1334                         *length += end - key.offset + 1;
1335                         break;
1336                 } else if (key.offset > end)
1337                         break;
1338
1339 next:
1340                 path->slots[0]++;
1341         }
1342         ret = 0;
1343 out:
1344         btrfs_free_path(path);
1345         return ret;
1346 }
1347
1348 static int contains_pending_extent(struct btrfs_transaction *transaction,
1349                                    struct btrfs_device *device,
1350                                    u64 *start, u64 len)
1351 {
1352         struct btrfs_fs_info *fs_info = device->fs_info;
1353         struct extent_map *em;
1354         struct list_head *search_list = &fs_info->pinned_chunks;
1355         int ret = 0;
1356         u64 physical_start = *start;
1357
1358         if (transaction)
1359                 search_list = &transaction->pending_chunks;
1360 again:
1361         list_for_each_entry(em, search_list, list) {
1362                 struct map_lookup *map;
1363                 int i;
1364
1365                 map = em->map_lookup;
1366                 for (i = 0; i < map->num_stripes; i++) {
1367                         u64 end;
1368
1369                         if (map->stripes[i].dev != device)
1370                                 continue;
1371                         if (map->stripes[i].physical >= physical_start + len ||
1372                             map->stripes[i].physical + em->orig_block_len <=
1373                             physical_start)
1374                                 continue;
1375                         /*
1376                          * Make sure that while processing the pinned list we do
1377                          * not override our *start with a lower value, because
1378                          * we can have pinned chunks that fall within this
1379                          * device hole and that have lower physical addresses
1380                          * than the pending chunks we processed before. If we
1381                          * do not take this special care we can end up getting
1382                          * 2 pending chunks that start at the same physical
1383                          * device offsets because the end offset of a pinned
1384                          * chunk can be equal to the start offset of some
1385                          * pending chunk.
1386                          */
1387                         end = map->stripes[i].physical + em->orig_block_len;
1388                         if (end > *start) {
1389                                 *start = end;
1390                                 ret = 1;
1391                         }
1392                 }
1393         }
1394         if (search_list != &fs_info->pinned_chunks) {
1395                 search_list = &fs_info->pinned_chunks;
1396                 goto again;
1397         }
1398
1399         return ret;
1400 }
1401
1402
1403 /*
1404  * find_free_dev_extent_start - find free space in the specified device
1405  * @device:       the device which we search the free space in
1406  * @num_bytes:    the size of the free space that we need
1407  * @search_start: the position from which to begin the search
1408  * @start:        store the start of the free space.
1409  * @len:          the size of the free space. that we find, or the size
1410  *                of the max free space if we don't find suitable free space
1411  *
1412  * this uses a pretty simple search, the expectation is that it is
1413  * called very infrequently and that a given device has a small number
1414  * of extents
1415  *
1416  * @start is used to store the start of the free space if we find. But if we
1417  * don't find suitable free space, it will be used to store the start position
1418  * of the max free space.
1419  *
1420  * @len is used to store the size of the free space that we find.
1421  * But if we don't find suitable free space, it is used to store the size of
1422  * the max free space.
1423  */
1424 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1425                                struct btrfs_device *device, u64 num_bytes,
1426                                u64 search_start, u64 *start, u64 *len)
1427 {
1428         struct btrfs_fs_info *fs_info = device->fs_info;
1429         struct btrfs_root *root = fs_info->dev_root;
1430         struct btrfs_key key;
1431         struct btrfs_dev_extent *dev_extent;
1432         struct btrfs_path *path;
1433         u64 hole_size;
1434         u64 max_hole_start;
1435         u64 max_hole_size;
1436         u64 extent_end;
1437         u64 search_end = device->total_bytes;
1438         int ret;
1439         int slot;
1440         struct extent_buffer *l;
1441
1442         /*
1443          * We don't want to overwrite the superblock on the drive nor any area
1444          * used by the boot loader (grub for example), so we make sure to start
1445          * at an offset of at least 1MB.
1446          */
1447         search_start = max_t(u64, search_start, SZ_1M);
1448
1449         path = btrfs_alloc_path();
1450         if (!path)
1451                 return -ENOMEM;
1452
1453         max_hole_start = search_start;
1454         max_hole_size = 0;
1455
1456 again:
1457         if (search_start >= search_end ||
1458                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1459                 ret = -ENOSPC;
1460                 goto out;
1461         }
1462
1463         path->reada = READA_FORWARD;
1464         path->search_commit_root = 1;
1465         path->skip_locking = 1;
1466
1467         key.objectid = device->devid;
1468         key.offset = search_start;
1469         key.type = BTRFS_DEV_EXTENT_KEY;
1470
1471         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1472         if (ret < 0)
1473                 goto out;
1474         if (ret > 0) {
1475                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1476                 if (ret < 0)
1477                         goto out;
1478         }
1479
1480         while (1) {
1481                 l = path->nodes[0];
1482                 slot = path->slots[0];
1483                 if (slot >= btrfs_header_nritems(l)) {
1484                         ret = btrfs_next_leaf(root, path);
1485                         if (ret == 0)
1486                                 continue;
1487                         if (ret < 0)
1488                                 goto out;
1489
1490                         break;
1491                 }
1492                 btrfs_item_key_to_cpu(l, &key, slot);
1493
1494                 if (key.objectid < device->devid)
1495                         goto next;
1496
1497                 if (key.objectid > device->devid)
1498                         break;
1499
1500                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1501                         goto next;
1502
1503                 if (key.offset > search_start) {
1504                         hole_size = key.offset - search_start;
1505
1506                         /*
1507                          * Have to check before we set max_hole_start, otherwise
1508                          * we could end up sending back this offset anyway.
1509                          */
1510                         if (contains_pending_extent(transaction, device,
1511                                                     &search_start,
1512                                                     hole_size)) {
1513                                 if (key.offset >= search_start) {
1514                                         hole_size = key.offset - search_start;
1515                                 } else {
1516                                         WARN_ON_ONCE(1);
1517                                         hole_size = 0;
1518                                 }
1519                         }
1520
1521                         if (hole_size > max_hole_size) {
1522                                 max_hole_start = search_start;
1523                                 max_hole_size = hole_size;
1524                         }
1525
1526                         /*
1527                          * If this free space is greater than which we need,
1528                          * it must be the max free space that we have found
1529                          * until now, so max_hole_start must point to the start
1530                          * of this free space and the length of this free space
1531                          * is stored in max_hole_size. Thus, we return
1532                          * max_hole_start and max_hole_size and go back to the
1533                          * caller.
1534                          */
1535                         if (hole_size >= num_bytes) {
1536                                 ret = 0;
1537                                 goto out;
1538                         }
1539                 }
1540
1541                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1542                 extent_end = key.offset + btrfs_dev_extent_length(l,
1543                                                                   dev_extent);
1544                 if (extent_end > search_start)
1545                         search_start = extent_end;
1546 next:
1547                 path->slots[0]++;
1548                 cond_resched();
1549         }
1550
1551         /*
1552          * At this point, search_start should be the end of
1553          * allocated dev extents, and when shrinking the device,
1554          * search_end may be smaller than search_start.
1555          */
1556         if (search_end > search_start) {
1557                 hole_size = search_end - search_start;
1558
1559                 if (contains_pending_extent(transaction, device, &search_start,
1560                                             hole_size)) {
1561                         btrfs_release_path(path);
1562                         goto again;
1563                 }
1564
1565                 if (hole_size > max_hole_size) {
1566                         max_hole_start = search_start;
1567                         max_hole_size = hole_size;
1568                 }
1569         }
1570
1571         /* See above. */
1572         if (max_hole_size < num_bytes)
1573                 ret = -ENOSPC;
1574         else
1575                 ret = 0;
1576
1577 out:
1578         btrfs_free_path(path);
1579         *start = max_hole_start;
1580         if (len)
1581                 *len = max_hole_size;
1582         return ret;
1583 }
1584
1585 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1586                          struct btrfs_device *device, u64 num_bytes,
1587                          u64 *start, u64 *len)
1588 {
1589         /* FIXME use last free of some kind */
1590         return find_free_dev_extent_start(trans->transaction, device,
1591                                           num_bytes, 0, start, len);
1592 }
1593
1594 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1595                           struct btrfs_device *device,
1596                           u64 start, u64 *dev_extent_len)
1597 {
1598         struct btrfs_fs_info *fs_info = device->fs_info;
1599         struct btrfs_root *root = fs_info->dev_root;
1600         int ret;
1601         struct btrfs_path *path;
1602         struct btrfs_key key;
1603         struct btrfs_key found_key;
1604         struct extent_buffer *leaf = NULL;
1605         struct btrfs_dev_extent *extent = NULL;
1606
1607         path = btrfs_alloc_path();
1608         if (!path)
1609                 return -ENOMEM;
1610
1611         key.objectid = device->devid;
1612         key.offset = start;
1613         key.type = BTRFS_DEV_EXTENT_KEY;
1614 again:
1615         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1616         if (ret > 0) {
1617                 ret = btrfs_previous_item(root, path, key.objectid,
1618                                           BTRFS_DEV_EXTENT_KEY);
1619                 if (ret)
1620                         goto out;
1621                 leaf = path->nodes[0];
1622                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1623                 extent = btrfs_item_ptr(leaf, path->slots[0],
1624                                         struct btrfs_dev_extent);
1625                 BUG_ON(found_key.offset > start || found_key.offset +
1626                        btrfs_dev_extent_length(leaf, extent) < start);
1627                 key = found_key;
1628                 btrfs_release_path(path);
1629                 goto again;
1630         } else if (ret == 0) {
1631                 leaf = path->nodes[0];
1632                 extent = btrfs_item_ptr(leaf, path->slots[0],
1633                                         struct btrfs_dev_extent);
1634         } else {
1635                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1636                 goto out;
1637         }
1638
1639         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1640
1641         ret = btrfs_del_item(trans, root, path);
1642         if (ret) {
1643                 btrfs_handle_fs_error(fs_info, ret,
1644                                       "Failed to remove dev extent item");
1645         } else {
1646                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1647         }
1648 out:
1649         btrfs_free_path(path);
1650         return ret;
1651 }
1652
1653 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1654                                   struct btrfs_device *device,
1655                                   u64 chunk_offset, u64 start, u64 num_bytes)
1656 {
1657         int ret;
1658         struct btrfs_path *path;
1659         struct btrfs_fs_info *fs_info = device->fs_info;
1660         struct btrfs_root *root = fs_info->dev_root;
1661         struct btrfs_dev_extent *extent;
1662         struct extent_buffer *leaf;
1663         struct btrfs_key key;
1664
1665         WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1666         WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1667         path = btrfs_alloc_path();
1668         if (!path)
1669                 return -ENOMEM;
1670
1671         key.objectid = device->devid;
1672         key.offset = start;
1673         key.type = BTRFS_DEV_EXTENT_KEY;
1674         ret = btrfs_insert_empty_item(trans, root, path, &key,
1675                                       sizeof(*extent));
1676         if (ret)
1677                 goto out;
1678
1679         leaf = path->nodes[0];
1680         extent = btrfs_item_ptr(leaf, path->slots[0],
1681                                 struct btrfs_dev_extent);
1682         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1683                                         BTRFS_CHUNK_TREE_OBJECTID);
1684         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1685                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1686         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1687
1688         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1689         btrfs_mark_buffer_dirty(leaf);
1690 out:
1691         btrfs_free_path(path);
1692         return ret;
1693 }
1694
1695 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1696 {
1697         struct extent_map_tree *em_tree;
1698         struct extent_map *em;
1699         struct rb_node *n;
1700         u64 ret = 0;
1701
1702         em_tree = &fs_info->mapping_tree.map_tree;
1703         read_lock(&em_tree->lock);
1704         n = rb_last(&em_tree->map);
1705         if (n) {
1706                 em = rb_entry(n, struct extent_map, rb_node);
1707                 ret = em->start + em->len;
1708         }
1709         read_unlock(&em_tree->lock);
1710
1711         return ret;
1712 }
1713
1714 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1715                                     u64 *devid_ret)
1716 {
1717         int ret;
1718         struct btrfs_key key;
1719         struct btrfs_key found_key;
1720         struct btrfs_path *path;
1721
1722         path = btrfs_alloc_path();
1723         if (!path)
1724                 return -ENOMEM;
1725
1726         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1727         key.type = BTRFS_DEV_ITEM_KEY;
1728         key.offset = (u64)-1;
1729
1730         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1731         if (ret < 0)
1732                 goto error;
1733
1734         BUG_ON(ret == 0); /* Corruption */
1735
1736         ret = btrfs_previous_item(fs_info->chunk_root, path,
1737                                   BTRFS_DEV_ITEMS_OBJECTID,
1738                                   BTRFS_DEV_ITEM_KEY);
1739         if (ret) {
1740                 *devid_ret = 1;
1741         } else {
1742                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1743                                       path->slots[0]);
1744                 *devid_ret = found_key.offset + 1;
1745         }
1746         ret = 0;
1747 error:
1748         btrfs_free_path(path);
1749         return ret;
1750 }
1751
1752 /*
1753  * the device information is stored in the chunk root
1754  * the btrfs_device struct should be fully filled in
1755  */
1756 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1757                             struct btrfs_fs_info *fs_info,
1758                             struct btrfs_device *device)
1759 {
1760         struct btrfs_root *root = fs_info->chunk_root;
1761         int ret;
1762         struct btrfs_path *path;
1763         struct btrfs_dev_item *dev_item;
1764         struct extent_buffer *leaf;
1765         struct btrfs_key key;
1766         unsigned long ptr;
1767
1768         path = btrfs_alloc_path();
1769         if (!path)
1770                 return -ENOMEM;
1771
1772         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1773         key.type = BTRFS_DEV_ITEM_KEY;
1774         key.offset = device->devid;
1775
1776         ret = btrfs_insert_empty_item(trans, root, path, &key,
1777                                       sizeof(*dev_item));
1778         if (ret)
1779                 goto out;
1780
1781         leaf = path->nodes[0];
1782         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1783
1784         btrfs_set_device_id(leaf, dev_item, device->devid);
1785         btrfs_set_device_generation(leaf, dev_item, 0);
1786         btrfs_set_device_type(leaf, dev_item, device->type);
1787         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1788         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1789         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1790         btrfs_set_device_total_bytes(leaf, dev_item,
1791                                      btrfs_device_get_disk_total_bytes(device));
1792         btrfs_set_device_bytes_used(leaf, dev_item,
1793                                     btrfs_device_get_bytes_used(device));
1794         btrfs_set_device_group(leaf, dev_item, 0);
1795         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1796         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1797         btrfs_set_device_start_offset(leaf, dev_item, 0);
1798
1799         ptr = btrfs_device_uuid(dev_item);
1800         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1801         ptr = btrfs_device_fsid(dev_item);
1802         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1803         btrfs_mark_buffer_dirty(leaf);
1804
1805         ret = 0;
1806 out:
1807         btrfs_free_path(path);
1808         return ret;
1809 }
1810
1811 /*
1812  * Function to update ctime/mtime for a given device path.
1813  * Mainly used for ctime/mtime based probe like libblkid.
1814  */
1815 static void update_dev_time(const char *path_name)
1816 {
1817         struct file *filp;
1818
1819         filp = filp_open(path_name, O_RDWR, 0);
1820         if (IS_ERR(filp))
1821                 return;
1822         file_update_time(filp);
1823         filp_close(filp, NULL);
1824 }
1825
1826 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1827                              struct btrfs_device *device)
1828 {
1829         struct btrfs_root *root = fs_info->chunk_root;
1830         int ret;
1831         struct btrfs_path *path;
1832         struct btrfs_key key;
1833         struct btrfs_trans_handle *trans;
1834
1835         path = btrfs_alloc_path();
1836         if (!path)
1837                 return -ENOMEM;
1838
1839         trans = btrfs_start_transaction(root, 0);
1840         if (IS_ERR(trans)) {
1841                 btrfs_free_path(path);
1842                 return PTR_ERR(trans);
1843         }
1844         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1845         key.type = BTRFS_DEV_ITEM_KEY;
1846         key.offset = device->devid;
1847
1848         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1849         if (ret) {
1850                 if (ret > 0)
1851                         ret = -ENOENT;
1852                 btrfs_abort_transaction(trans, ret);
1853                 btrfs_end_transaction(trans);
1854                 goto out;
1855         }
1856
1857         ret = btrfs_del_item(trans, root, path);
1858         if (ret) {
1859                 btrfs_abort_transaction(trans, ret);
1860                 btrfs_end_transaction(trans);
1861         }
1862
1863 out:
1864         btrfs_free_path(path);
1865         if (!ret)
1866                 ret = btrfs_commit_transaction(trans);
1867         return ret;
1868 }
1869
1870 /*
1871  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1872  * filesystem. It's up to the caller to adjust that number regarding eg. device
1873  * replace.
1874  */
1875 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1876                 u64 num_devices)
1877 {
1878         u64 all_avail;
1879         unsigned seq;
1880         int i;
1881
1882         do {
1883                 seq = read_seqbegin(&fs_info->profiles_lock);
1884
1885                 all_avail = fs_info->avail_data_alloc_bits |
1886                             fs_info->avail_system_alloc_bits |
1887                             fs_info->avail_metadata_alloc_bits;
1888         } while (read_seqretry(&fs_info->profiles_lock, seq));
1889
1890         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1891                 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1892                         continue;
1893
1894                 if (num_devices < btrfs_raid_array[i].devs_min) {
1895                         int ret = btrfs_raid_array[i].mindev_error;
1896
1897                         if (ret)
1898                                 return ret;
1899                 }
1900         }
1901
1902         return 0;
1903 }
1904
1905 static struct btrfs_device * btrfs_find_next_active_device(
1906                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1907 {
1908         struct btrfs_device *next_device;
1909
1910         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1911                 if (next_device != device &&
1912                     !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1913                     && next_device->bdev)
1914                         return next_device;
1915         }
1916
1917         return NULL;
1918 }
1919
1920 /*
1921  * Helper function to check if the given device is part of s_bdev / latest_bdev
1922  * and replace it with the provided or the next active device, in the context
1923  * where this function called, there should be always be another device (or
1924  * this_dev) which is active.
1925  */
1926 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1927                 struct btrfs_device *device, struct btrfs_device *this_dev)
1928 {
1929         struct btrfs_device *next_device;
1930
1931         if (this_dev)
1932                 next_device = this_dev;
1933         else
1934                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1935                                                                 device);
1936         ASSERT(next_device);
1937
1938         if (fs_info->sb->s_bdev &&
1939                         (fs_info->sb->s_bdev == device->bdev))
1940                 fs_info->sb->s_bdev = next_device->bdev;
1941
1942         if (fs_info->fs_devices->latest_bdev == device->bdev)
1943                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1944 }
1945
1946 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1947                 u64 devid)
1948 {
1949         struct btrfs_device *device;
1950         struct btrfs_fs_devices *cur_devices;
1951         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1952         u64 num_devices;
1953         int ret = 0;
1954
1955         mutex_lock(&uuid_mutex);
1956
1957         num_devices = fs_devices->num_devices;
1958         btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1959         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1960                 WARN_ON(num_devices < 1);
1961                 num_devices--;
1962         }
1963         btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1964
1965         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1966         if (ret)
1967                 goto out;
1968
1969         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1970                                            &device);
1971         if (ret)
1972                 goto out;
1973
1974         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1975                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1976                 goto out;
1977         }
1978
1979         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1980             fs_info->fs_devices->rw_devices == 1) {
1981                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1982                 goto out;
1983         }
1984
1985         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1986                 mutex_lock(&fs_info->chunk_mutex);
1987                 list_del_init(&device->dev_alloc_list);
1988                 device->fs_devices->rw_devices--;
1989                 mutex_unlock(&fs_info->chunk_mutex);
1990         }
1991
1992         mutex_unlock(&uuid_mutex);
1993         ret = btrfs_shrink_device(device, 0);
1994         mutex_lock(&uuid_mutex);
1995         if (ret)
1996                 goto error_undo;
1997
1998         /*
1999          * TODO: the superblock still includes this device in its num_devices
2000          * counter although write_all_supers() is not locked out. This
2001          * could give a filesystem state which requires a degraded mount.
2002          */
2003         ret = btrfs_rm_dev_item(fs_info, device);
2004         if (ret)
2005                 goto error_undo;
2006
2007         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2008         btrfs_scrub_cancel_dev(fs_info, device);
2009
2010         /*
2011          * the device list mutex makes sure that we don't change
2012          * the device list while someone else is writing out all
2013          * the device supers. Whoever is writing all supers, should
2014          * lock the device list mutex before getting the number of
2015          * devices in the super block (super_copy). Conversely,
2016          * whoever updates the number of devices in the super block
2017          * (super_copy) should hold the device list mutex.
2018          */
2019
2020         cur_devices = device->fs_devices;
2021         mutex_lock(&fs_devices->device_list_mutex);
2022         list_del_rcu(&device->dev_list);
2023
2024         device->fs_devices->num_devices--;
2025         device->fs_devices->total_devices--;
2026
2027         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2028                 device->fs_devices->missing_devices--;
2029
2030         btrfs_assign_next_active_device(fs_info, device, NULL);
2031
2032         if (device->bdev) {
2033                 device->fs_devices->open_devices--;
2034                 /* remove sysfs entry */
2035                 btrfs_sysfs_rm_device_link(fs_devices, device);
2036         }
2037
2038         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2039         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2040         mutex_unlock(&fs_devices->device_list_mutex);
2041
2042         /*
2043          * at this point, the device is zero sized and detached from
2044          * the devices list.  All that's left is to zero out the old
2045          * supers and free the device.
2046          */
2047         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2048                 btrfs_scratch_superblocks(device->bdev, device->name->str);
2049
2050         btrfs_close_bdev(device);
2051         call_rcu(&device->rcu, free_device_rcu);
2052
2053         if (cur_devices->open_devices == 0) {
2054                 while (fs_devices) {
2055                         if (fs_devices->seed == cur_devices) {
2056                                 fs_devices->seed = cur_devices->seed;
2057                                 break;
2058                         }
2059                         fs_devices = fs_devices->seed;
2060                 }
2061                 cur_devices->seed = NULL;
2062                 close_fs_devices(cur_devices);
2063                 free_fs_devices(cur_devices);
2064         }
2065
2066 out:
2067         mutex_unlock(&uuid_mutex);
2068         return ret;
2069
2070 error_undo:
2071         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2072                 mutex_lock(&fs_info->chunk_mutex);
2073                 list_add(&device->dev_alloc_list,
2074                          &fs_devices->alloc_list);
2075                 device->fs_devices->rw_devices++;
2076                 mutex_unlock(&fs_info->chunk_mutex);
2077         }
2078         goto out;
2079 }
2080
2081 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2082                                         struct btrfs_device *srcdev)
2083 {
2084         struct btrfs_fs_devices *fs_devices;
2085
2086         lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2087
2088         /*
2089          * in case of fs with no seed, srcdev->fs_devices will point
2090          * to fs_devices of fs_info. However when the dev being replaced is
2091          * a seed dev it will point to the seed's local fs_devices. In short
2092          * srcdev will have its correct fs_devices in both the cases.
2093          */
2094         fs_devices = srcdev->fs_devices;
2095
2096         list_del_rcu(&srcdev->dev_list);
2097         list_del(&srcdev->dev_alloc_list);
2098         fs_devices->num_devices--;
2099         if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2100                 fs_devices->missing_devices--;
2101
2102         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2103                 fs_devices->rw_devices--;
2104
2105         if (srcdev->bdev)
2106                 fs_devices->open_devices--;
2107 }
2108
2109 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2110                                       struct btrfs_device *srcdev)
2111 {
2112         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2113
2114         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2115                 /* zero out the old super if it is writable */
2116                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2117         }
2118
2119         btrfs_close_bdev(srcdev);
2120         call_rcu(&srcdev->rcu, free_device_rcu);
2121
2122         /* if this is no devs we rather delete the fs_devices */
2123         if (!fs_devices->num_devices) {
2124                 struct btrfs_fs_devices *tmp_fs_devices;
2125
2126                 /*
2127                  * On a mounted FS, num_devices can't be zero unless it's a
2128                  * seed. In case of a seed device being replaced, the replace
2129                  * target added to the sprout FS, so there will be no more
2130                  * device left under the seed FS.
2131                  */
2132                 ASSERT(fs_devices->seeding);
2133
2134                 tmp_fs_devices = fs_info->fs_devices;
2135                 while (tmp_fs_devices) {
2136                         if (tmp_fs_devices->seed == fs_devices) {
2137                                 tmp_fs_devices->seed = fs_devices->seed;
2138                                 break;
2139                         }
2140                         tmp_fs_devices = tmp_fs_devices->seed;
2141                 }
2142                 fs_devices->seed = NULL;
2143                 close_fs_devices(fs_devices);
2144                 free_fs_devices(fs_devices);
2145         }
2146 }
2147
2148 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2149                                       struct btrfs_device *tgtdev)
2150 {
2151         mutex_lock(&uuid_mutex);
2152         WARN_ON(!tgtdev);
2153         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2154
2155         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2156
2157         if (tgtdev->bdev)
2158                 fs_info->fs_devices->open_devices--;
2159
2160         fs_info->fs_devices->num_devices--;
2161
2162         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2163
2164         list_del_rcu(&tgtdev->dev_list);
2165
2166         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2167         mutex_unlock(&uuid_mutex);
2168
2169         /*
2170          * The update_dev_time() with in btrfs_scratch_superblocks()
2171          * may lead to a call to btrfs_show_devname() which will try
2172          * to hold device_list_mutex. And here this device
2173          * is already out of device list, so we don't have to hold
2174          * the device_list_mutex lock.
2175          */
2176         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2177
2178         btrfs_close_bdev(tgtdev);
2179         call_rcu(&tgtdev->rcu, free_device_rcu);
2180 }
2181
2182 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2183                                      const char *device_path,
2184                                      struct btrfs_device **device)
2185 {
2186         int ret = 0;
2187         struct btrfs_super_block *disk_super;
2188         u64 devid;
2189         u8 *dev_uuid;
2190         struct block_device *bdev;
2191         struct buffer_head *bh;
2192
2193         *device = NULL;
2194         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2195                                     fs_info->bdev_holder, 0, &bdev, &bh);
2196         if (ret)
2197                 return ret;
2198         disk_super = (struct btrfs_super_block *)bh->b_data;
2199         devid = btrfs_stack_device_id(&disk_super->dev_item);
2200         dev_uuid = disk_super->dev_item.uuid;
2201         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2202         brelse(bh);
2203         if (!*device)
2204                 ret = -ENOENT;
2205         blkdev_put(bdev, FMODE_READ);
2206         return ret;
2207 }
2208
2209 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2210                                          const char *device_path,
2211                                          struct btrfs_device **device)
2212 {
2213         *device = NULL;
2214         if (strcmp(device_path, "missing") == 0) {
2215                 struct list_head *devices;
2216                 struct btrfs_device *tmp;
2217
2218                 devices = &fs_info->fs_devices->devices;
2219                 list_for_each_entry(tmp, devices, dev_list) {
2220                         if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2221                                         &tmp->dev_state) && !tmp->bdev) {
2222                                 *device = tmp;
2223                                 break;
2224                         }
2225                 }
2226
2227                 if (!*device)
2228                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2229
2230                 return 0;
2231         } else {
2232                 return btrfs_find_device_by_path(fs_info, device_path, device);
2233         }
2234 }
2235
2236 /*
2237  * Lookup a device given by device id, or the path if the id is 0.
2238  */
2239 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2240                                  const char *devpath,
2241                                  struct btrfs_device **device)
2242 {
2243         int ret;
2244
2245         if (devid) {
2246                 ret = 0;
2247                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2248                 if (!*device)
2249                         ret = -ENOENT;
2250         } else {
2251                 if (!devpath || !devpath[0])
2252                         return -EINVAL;
2253
2254                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2255                                                            device);
2256         }
2257         return ret;
2258 }
2259
2260 /*
2261  * does all the dirty work required for changing file system's UUID.
2262  */
2263 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2264 {
2265         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2266         struct btrfs_fs_devices *old_devices;
2267         struct btrfs_fs_devices *seed_devices;
2268         struct btrfs_super_block *disk_super = fs_info->super_copy;
2269         struct btrfs_device *device;
2270         u64 super_flags;
2271
2272         lockdep_assert_held(&uuid_mutex);
2273         if (!fs_devices->seeding)
2274                 return -EINVAL;
2275
2276         seed_devices = alloc_fs_devices(NULL);
2277         if (IS_ERR(seed_devices))
2278                 return PTR_ERR(seed_devices);
2279
2280         old_devices = clone_fs_devices(fs_devices);
2281         if (IS_ERR(old_devices)) {
2282                 kfree(seed_devices);
2283                 return PTR_ERR(old_devices);
2284         }
2285
2286         list_add(&old_devices->fs_list, &fs_uuids);
2287
2288         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2289         seed_devices->opened = 1;
2290         INIT_LIST_HEAD(&seed_devices->devices);
2291         INIT_LIST_HEAD(&seed_devices->alloc_list);
2292         mutex_init(&seed_devices->device_list_mutex);
2293
2294         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2295         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2296                               synchronize_rcu);
2297         list_for_each_entry(device, &seed_devices->devices, dev_list)
2298                 device->fs_devices = seed_devices;
2299
2300         mutex_lock(&fs_info->chunk_mutex);
2301         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2302         mutex_unlock(&fs_info->chunk_mutex);
2303
2304         fs_devices->seeding = 0;
2305         fs_devices->num_devices = 0;
2306         fs_devices->open_devices = 0;
2307         fs_devices->missing_devices = 0;
2308         fs_devices->rotating = 0;
2309         fs_devices->seed = seed_devices;
2310
2311         generate_random_uuid(fs_devices->fsid);
2312         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2313         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2314         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2315
2316         super_flags = btrfs_super_flags(disk_super) &
2317                       ~BTRFS_SUPER_FLAG_SEEDING;
2318         btrfs_set_super_flags(disk_super, super_flags);
2319
2320         return 0;
2321 }
2322
2323 /*
2324  * Store the expected generation for seed devices in device items.
2325  */
2326 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2327                                struct btrfs_fs_info *fs_info)
2328 {
2329         struct btrfs_root *root = fs_info->chunk_root;
2330         struct btrfs_path *path;
2331         struct extent_buffer *leaf;
2332         struct btrfs_dev_item *dev_item;
2333         struct btrfs_device *device;
2334         struct btrfs_key key;
2335         u8 fs_uuid[BTRFS_FSID_SIZE];
2336         u8 dev_uuid[BTRFS_UUID_SIZE];
2337         u64 devid;
2338         int ret;
2339
2340         path = btrfs_alloc_path();
2341         if (!path)
2342                 return -ENOMEM;
2343
2344         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2345         key.offset = 0;
2346         key.type = BTRFS_DEV_ITEM_KEY;
2347
2348         while (1) {
2349                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2350                 if (ret < 0)
2351                         goto error;
2352
2353                 leaf = path->nodes[0];
2354 next_slot:
2355                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2356                         ret = btrfs_next_leaf(root, path);
2357                         if (ret > 0)
2358                                 break;
2359                         if (ret < 0)
2360                                 goto error;
2361                         leaf = path->nodes[0];
2362                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2363                         btrfs_release_path(path);
2364                         continue;
2365                 }
2366
2367                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2368                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2369                     key.type != BTRFS_DEV_ITEM_KEY)
2370                         break;
2371
2372                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2373                                           struct btrfs_dev_item);
2374                 devid = btrfs_device_id(leaf, dev_item);
2375                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2376                                    BTRFS_UUID_SIZE);
2377                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2378                                    BTRFS_FSID_SIZE);
2379                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2380                 BUG_ON(!device); /* Logic error */
2381
2382                 if (device->fs_devices->seeding) {
2383                         btrfs_set_device_generation(leaf, dev_item,
2384                                                     device->generation);
2385                         btrfs_mark_buffer_dirty(leaf);
2386                 }
2387
2388                 path->slots[0]++;
2389                 goto next_slot;
2390         }
2391         ret = 0;
2392 error:
2393         btrfs_free_path(path);
2394         return ret;
2395 }
2396
2397 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2398 {
2399         struct btrfs_root *root = fs_info->dev_root;
2400         struct request_queue *q;
2401         struct btrfs_trans_handle *trans;
2402         struct btrfs_device *device;
2403         struct block_device *bdev;
2404         struct list_head *devices;
2405         struct super_block *sb = fs_info->sb;
2406         struct rcu_string *name;
2407         u64 tmp;
2408         int seeding_dev = 0;
2409         int ret = 0;
2410         bool unlocked = false;
2411
2412         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2413                 return -EROFS;
2414
2415         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2416                                   fs_info->bdev_holder);
2417         if (IS_ERR(bdev))
2418                 return PTR_ERR(bdev);
2419
2420         if (fs_info->fs_devices->seeding) {
2421                 seeding_dev = 1;
2422                 down_write(&sb->s_umount);
2423                 mutex_lock(&uuid_mutex);
2424         }
2425
2426         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2427
2428         devices = &fs_info->fs_devices->devices;
2429
2430         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2431         list_for_each_entry(device, devices, dev_list) {
2432                 if (device->bdev == bdev) {
2433                         ret = -EEXIST;
2434                         mutex_unlock(
2435                                 &fs_info->fs_devices->device_list_mutex);
2436                         goto error;
2437                 }
2438         }
2439         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2440
2441         device = btrfs_alloc_device(fs_info, NULL, NULL);
2442         if (IS_ERR(device)) {
2443                 /* we can safely leave the fs_devices entry around */
2444                 ret = PTR_ERR(device);
2445                 goto error;
2446         }
2447
2448         name = rcu_string_strdup(device_path, GFP_KERNEL);
2449         if (!name) {
2450                 ret = -ENOMEM;
2451                 goto error_free_device;
2452         }
2453         rcu_assign_pointer(device->name, name);
2454
2455         trans = btrfs_start_transaction(root, 0);
2456         if (IS_ERR(trans)) {
2457                 ret = PTR_ERR(trans);
2458                 goto error_free_device;
2459         }
2460
2461         q = bdev_get_queue(bdev);
2462         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2463         device->generation = trans->transid;
2464         device->io_width = fs_info->sectorsize;
2465         device->io_align = fs_info->sectorsize;
2466         device->sector_size = fs_info->sectorsize;
2467         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2468                                          fs_info->sectorsize);
2469         device->disk_total_bytes = device->total_bytes;
2470         device->commit_total_bytes = device->total_bytes;
2471         device->fs_info = fs_info;
2472         device->bdev = bdev;
2473         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2474         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2475         device->mode = FMODE_EXCL;
2476         device->dev_stats_valid = 1;
2477         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2478
2479         if (seeding_dev) {
2480                 sb->s_flags &= ~SB_RDONLY;
2481                 ret = btrfs_prepare_sprout(fs_info);
2482                 if (ret) {
2483                         btrfs_abort_transaction(trans, ret);
2484                         goto error_trans;
2485                 }
2486         }
2487
2488         device->fs_devices = fs_info->fs_devices;
2489
2490         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2491         mutex_lock(&fs_info->chunk_mutex);
2492         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2493         list_add(&device->dev_alloc_list,
2494                  &fs_info->fs_devices->alloc_list);
2495         fs_info->fs_devices->num_devices++;
2496         fs_info->fs_devices->open_devices++;
2497         fs_info->fs_devices->rw_devices++;
2498         fs_info->fs_devices->total_devices++;
2499         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2500
2501         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2502
2503         if (!blk_queue_nonrot(q))
2504                 fs_info->fs_devices->rotating = 1;
2505
2506         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2507         btrfs_set_super_total_bytes(fs_info->super_copy,
2508                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2509
2510         tmp = btrfs_super_num_devices(fs_info->super_copy);
2511         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2512
2513         /* add sysfs device entry */
2514         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2515
2516         /*
2517          * we've got more storage, clear any full flags on the space
2518          * infos
2519          */
2520         btrfs_clear_space_info_full(fs_info);
2521
2522         mutex_unlock(&fs_info->chunk_mutex);
2523         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2524
2525         if (seeding_dev) {
2526                 mutex_lock(&fs_info->chunk_mutex);
2527                 ret = init_first_rw_device(trans, fs_info);
2528                 mutex_unlock(&fs_info->chunk_mutex);
2529                 if (ret) {
2530                         btrfs_abort_transaction(trans, ret);
2531                         goto error_sysfs;
2532                 }
2533         }
2534
2535         ret = btrfs_add_dev_item(trans, fs_info, device);
2536         if (ret) {
2537                 btrfs_abort_transaction(trans, ret);
2538                 goto error_sysfs;
2539         }
2540
2541         if (seeding_dev) {
2542                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2543
2544                 ret = btrfs_finish_sprout(trans, fs_info);
2545                 if (ret) {
2546                         btrfs_abort_transaction(trans, ret);
2547                         goto error_sysfs;
2548                 }
2549
2550                 /* Sprouting would change fsid of the mounted root,
2551                  * so rename the fsid on the sysfs
2552                  */
2553                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2554                                                 fs_info->fsid);
2555                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2556                         btrfs_warn(fs_info,
2557                                    "sysfs: failed to create fsid for sprout");
2558         }
2559
2560         ret = btrfs_commit_transaction(trans);
2561
2562         if (seeding_dev) {
2563                 mutex_unlock(&uuid_mutex);
2564                 up_write(&sb->s_umount);
2565                 unlocked = true;
2566
2567                 if (ret) /* transaction commit */
2568                         return ret;
2569
2570                 ret = btrfs_relocate_sys_chunks(fs_info);
2571                 if (ret < 0)
2572                         btrfs_handle_fs_error(fs_info, ret,
2573                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2574                 trans = btrfs_attach_transaction(root);
2575                 if (IS_ERR(trans)) {
2576                         if (PTR_ERR(trans) == -ENOENT)
2577                                 return 0;
2578                         ret = PTR_ERR(trans);
2579                         trans = NULL;
2580                         goto error_sysfs;
2581                 }
2582                 ret = btrfs_commit_transaction(trans);
2583         }
2584
2585         /* Update ctime/mtime for libblkid */
2586         update_dev_time(device_path);
2587         return ret;
2588
2589 error_sysfs:
2590         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2591 error_trans:
2592         if (seeding_dev)
2593                 sb->s_flags |= SB_RDONLY;
2594         if (trans)
2595                 btrfs_end_transaction(trans);
2596 error_free_device:
2597         btrfs_free_device(device);
2598 error:
2599         blkdev_put(bdev, FMODE_EXCL);
2600         if (seeding_dev && !unlocked) {
2601                 mutex_unlock(&uuid_mutex);
2602                 up_write(&sb->s_umount);
2603         }
2604         return ret;
2605 }
2606
2607 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2608                                         struct btrfs_device *device)
2609 {
2610         int ret;
2611         struct btrfs_path *path;
2612         struct btrfs_root *root = device->fs_info->chunk_root;
2613         struct btrfs_dev_item *dev_item;
2614         struct extent_buffer *leaf;
2615         struct btrfs_key key;
2616
2617         path = btrfs_alloc_path();
2618         if (!path)
2619                 return -ENOMEM;
2620
2621         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2622         key.type = BTRFS_DEV_ITEM_KEY;
2623         key.offset = device->devid;
2624
2625         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2626         if (ret < 0)
2627                 goto out;
2628
2629         if (ret > 0) {
2630                 ret = -ENOENT;
2631                 goto out;
2632         }
2633
2634         leaf = path->nodes[0];
2635         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2636
2637         btrfs_set_device_id(leaf, dev_item, device->devid);
2638         btrfs_set_device_type(leaf, dev_item, device->type);
2639         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2640         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2641         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2642         btrfs_set_device_total_bytes(leaf, dev_item,
2643                                      btrfs_device_get_disk_total_bytes(device));
2644         btrfs_set_device_bytes_used(leaf, dev_item,
2645                                     btrfs_device_get_bytes_used(device));
2646         btrfs_mark_buffer_dirty(leaf);
2647
2648 out:
2649         btrfs_free_path(path);
2650         return ret;
2651 }
2652
2653 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2654                       struct btrfs_device *device, u64 new_size)
2655 {
2656         struct btrfs_fs_info *fs_info = device->fs_info;
2657         struct btrfs_super_block *super_copy = fs_info->super_copy;
2658         struct btrfs_fs_devices *fs_devices;
2659         u64 old_total;
2660         u64 diff;
2661
2662         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2663                 return -EACCES;
2664
2665         new_size = round_down(new_size, fs_info->sectorsize);
2666
2667         mutex_lock(&fs_info->chunk_mutex);
2668         old_total = btrfs_super_total_bytes(super_copy);
2669         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2670
2671         if (new_size <= device->total_bytes ||
2672             test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2673                 mutex_unlock(&fs_info->chunk_mutex);
2674                 return -EINVAL;
2675         }
2676
2677         fs_devices = fs_info->fs_devices;
2678
2679         btrfs_set_super_total_bytes(super_copy,
2680                         round_down(old_total + diff, fs_info->sectorsize));
2681         device->fs_devices->total_rw_bytes += diff;
2682
2683         btrfs_device_set_total_bytes(device, new_size);
2684         btrfs_device_set_disk_total_bytes(device, new_size);
2685         btrfs_clear_space_info_full(device->fs_info);
2686         if (list_empty(&device->resized_list))
2687                 list_add_tail(&device->resized_list,
2688                               &fs_devices->resized_devices);
2689         mutex_unlock(&fs_info->chunk_mutex);
2690
2691         return btrfs_update_device(trans, device);
2692 }
2693
2694 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2695                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2696 {
2697         struct btrfs_root *root = fs_info->chunk_root;
2698         int ret;
2699         struct btrfs_path *path;
2700         struct btrfs_key key;
2701
2702         path = btrfs_alloc_path();
2703         if (!path)
2704                 return -ENOMEM;
2705
2706         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2707         key.offset = chunk_offset;
2708         key.type = BTRFS_CHUNK_ITEM_KEY;
2709
2710         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2711         if (ret < 0)
2712                 goto out;
2713         else if (ret > 0) { /* Logic error or corruption */
2714                 btrfs_handle_fs_error(fs_info, -ENOENT,
2715                                       "Failed lookup while freeing chunk.");
2716                 ret = -ENOENT;
2717                 goto out;
2718         }
2719
2720         ret = btrfs_del_item(trans, root, path);
2721         if (ret < 0)
2722                 btrfs_handle_fs_error(fs_info, ret,
2723                                       "Failed to delete chunk item.");
2724 out:
2725         btrfs_free_path(path);
2726         return ret;
2727 }
2728
2729 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2730 {
2731         struct btrfs_super_block *super_copy = fs_info->super_copy;
2732         struct btrfs_disk_key *disk_key;
2733         struct btrfs_chunk *chunk;
2734         u8 *ptr;
2735         int ret = 0;
2736         u32 num_stripes;
2737         u32 array_size;
2738         u32 len = 0;
2739         u32 cur;
2740         struct btrfs_key key;
2741
2742         mutex_lock(&fs_info->chunk_mutex);
2743         array_size = btrfs_super_sys_array_size(super_copy);
2744
2745         ptr = super_copy->sys_chunk_array;
2746         cur = 0;
2747
2748         while (cur < array_size) {
2749                 disk_key = (struct btrfs_disk_key *)ptr;
2750                 btrfs_disk_key_to_cpu(&key, disk_key);
2751
2752                 len = sizeof(*disk_key);
2753
2754                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2755                         chunk = (struct btrfs_chunk *)(ptr + len);
2756                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2757                         len += btrfs_chunk_item_size(num_stripes);
2758                 } else {
2759                         ret = -EIO;
2760                         break;
2761                 }
2762                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2763                     key.offset == chunk_offset) {
2764                         memmove(ptr, ptr + len, array_size - (cur + len));
2765                         array_size -= len;
2766                         btrfs_set_super_sys_array_size(super_copy, array_size);
2767                 } else {
2768                         ptr += len;
2769                         cur += len;
2770                 }
2771         }
2772         mutex_unlock(&fs_info->chunk_mutex);
2773         return ret;
2774 }
2775
2776 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2777                                         u64 logical, u64 length)
2778 {
2779         struct extent_map_tree *em_tree;
2780         struct extent_map *em;
2781
2782         em_tree = &fs_info->mapping_tree.map_tree;
2783         read_lock(&em_tree->lock);
2784         em = lookup_extent_mapping(em_tree, logical, length);
2785         read_unlock(&em_tree->lock);
2786
2787         if (!em) {
2788                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2789                            logical, length);
2790                 return ERR_PTR(-EINVAL);
2791         }
2792
2793         if (em->start > logical || em->start + em->len < logical) {
2794                 btrfs_crit(fs_info,
2795                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2796                            logical, length, em->start, em->start + em->len);
2797                 free_extent_map(em);
2798                 return ERR_PTR(-EINVAL);
2799         }
2800
2801         /* callers are responsible for dropping em's ref. */
2802         return em;
2803 }
2804
2805 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2806                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2807 {
2808         struct extent_map *em;
2809         struct map_lookup *map;
2810         u64 dev_extent_len = 0;
2811         int i, ret = 0;
2812         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2813
2814         em = get_chunk_map(fs_info, chunk_offset, 1);
2815         if (IS_ERR(em)) {
2816                 /*
2817                  * This is a logic error, but we don't want to just rely on the
2818                  * user having built with ASSERT enabled, so if ASSERT doesn't
2819                  * do anything we still error out.
2820                  */
2821                 ASSERT(0);
2822                 return PTR_ERR(em);
2823         }
2824         map = em->map_lookup;
2825         mutex_lock(&fs_info->chunk_mutex);
2826         check_system_chunk(trans, fs_info, map->type);
2827         mutex_unlock(&fs_info->chunk_mutex);
2828
2829         /*
2830          * Take the device list mutex to prevent races with the final phase of
2831          * a device replace operation that replaces the device object associated
2832          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2833          */
2834         mutex_lock(&fs_devices->device_list_mutex);
2835         for (i = 0; i < map->num_stripes; i++) {
2836                 struct btrfs_device *device = map->stripes[i].dev;
2837                 ret = btrfs_free_dev_extent(trans, device,
2838                                             map->stripes[i].physical,
2839                                             &dev_extent_len);
2840                 if (ret) {
2841                         mutex_unlock(&fs_devices->device_list_mutex);
2842                         btrfs_abort_transaction(trans, ret);
2843                         goto out;
2844                 }
2845
2846                 if (device->bytes_used > 0) {
2847                         mutex_lock(&fs_info->chunk_mutex);
2848                         btrfs_device_set_bytes_used(device,
2849                                         device->bytes_used - dev_extent_len);
2850                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2851                         btrfs_clear_space_info_full(fs_info);
2852                         mutex_unlock(&fs_info->chunk_mutex);
2853                 }
2854
2855                 if (map->stripes[i].dev) {
2856                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2857                         if (ret) {
2858                                 mutex_unlock(&fs_devices->device_list_mutex);
2859                                 btrfs_abort_transaction(trans, ret);
2860                                 goto out;
2861                         }
2862                 }
2863         }
2864         mutex_unlock(&fs_devices->device_list_mutex);
2865
2866         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2867         if (ret) {
2868                 btrfs_abort_transaction(trans, ret);
2869                 goto out;
2870         }
2871
2872         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2873
2874         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2875                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2876                 if (ret) {
2877                         btrfs_abort_transaction(trans, ret);
2878                         goto out;
2879                 }
2880         }
2881
2882         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2883         if (ret) {
2884                 btrfs_abort_transaction(trans, ret);
2885                 goto out;
2886         }
2887
2888 out:
2889         /* once for us */
2890         free_extent_map(em);
2891         return ret;
2892 }
2893
2894 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2895 {
2896         struct btrfs_root *root = fs_info->chunk_root;
2897         struct btrfs_trans_handle *trans;
2898         int ret;
2899
2900         /*
2901          * Prevent races with automatic removal of unused block groups.
2902          * After we relocate and before we remove the chunk with offset
2903          * chunk_offset, automatic removal of the block group can kick in,
2904          * resulting in a failure when calling btrfs_remove_chunk() below.
2905          *
2906          * Make sure to acquire this mutex before doing a tree search (dev
2907          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2908          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2909          * we release the path used to search the chunk/dev tree and before
2910          * the current task acquires this mutex and calls us.
2911          */
2912         lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2913
2914         ret = btrfs_can_relocate(fs_info, chunk_offset);
2915         if (ret)
2916                 return -ENOSPC;
2917
2918         /* step one, relocate all the extents inside this chunk */
2919         btrfs_scrub_pause(fs_info);
2920         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2921         btrfs_scrub_continue(fs_info);
2922         if (ret)
2923                 return ret;
2924
2925         /*
2926          * We add the kobjects here (and after forcing data chunk creation)
2927          * since relocation is the only place we'll create chunks of a new
2928          * type at runtime.  The only place where we'll remove the last
2929          * chunk of a type is the call immediately below this one.  Even
2930          * so, we're protected against races with the cleaner thread since
2931          * we're covered by the delete_unused_bgs_mutex.
2932          */
2933         btrfs_add_raid_kobjects(fs_info);
2934
2935         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2936                                                      chunk_offset);
2937         if (IS_ERR(trans)) {
2938                 ret = PTR_ERR(trans);
2939                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2940                 return ret;
2941         }
2942
2943         /*
2944          * step two, delete the device extents and the
2945          * chunk tree entries
2946          */
2947         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2948         btrfs_end_transaction(trans);
2949         return ret;
2950 }
2951
2952 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2953 {
2954         struct btrfs_root *chunk_root = fs_info->chunk_root;
2955         struct btrfs_path *path;
2956         struct extent_buffer *leaf;
2957         struct btrfs_chunk *chunk;
2958         struct btrfs_key key;
2959         struct btrfs_key found_key;
2960         u64 chunk_type;
2961         bool retried = false;
2962         int failed = 0;
2963         int ret;
2964
2965         path = btrfs_alloc_path();
2966         if (!path)
2967                 return -ENOMEM;
2968
2969 again:
2970         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2971         key.offset = (u64)-1;
2972         key.type = BTRFS_CHUNK_ITEM_KEY;
2973
2974         while (1) {
2975                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2976                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2977                 if (ret < 0) {
2978                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2979                         goto error;
2980                 }
2981                 BUG_ON(ret == 0); /* Corruption */
2982
2983                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2984                                           key.type);
2985                 if (ret)
2986                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2987                 if (ret < 0)
2988                         goto error;
2989                 if (ret > 0)
2990                         break;
2991
2992                 leaf = path->nodes[0];
2993                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2994
2995                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2996                                        struct btrfs_chunk);
2997                 chunk_type = btrfs_chunk_type(leaf, chunk);
2998                 btrfs_release_path(path);
2999
3000                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3001                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3002                         if (ret == -ENOSPC)
3003                                 failed++;
3004                         else
3005                                 BUG_ON(ret);
3006                 }
3007                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3008
3009                 if (found_key.offset == 0)
3010                         break;
3011                 key.offset = found_key.offset - 1;
3012         }
3013         ret = 0;
3014         if (failed && !retried) {
3015                 failed = 0;
3016                 retried = true;
3017                 goto again;
3018         } else if (WARN_ON(failed && retried)) {
3019                 ret = -ENOSPC;
3020         }
3021 error:
3022         btrfs_free_path(path);
3023         return ret;
3024 }
3025
3026 /*
3027  * return 1 : allocate a data chunk successfully,
3028  * return <0: errors during allocating a data chunk,
3029  * return 0 : no need to allocate a data chunk.
3030  */
3031 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3032                                       u64 chunk_offset)
3033 {
3034         struct btrfs_block_group_cache *cache;
3035         u64 bytes_used;
3036         u64 chunk_type;
3037
3038         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3039         ASSERT(cache);
3040         chunk_type = cache->flags;
3041         btrfs_put_block_group(cache);
3042
3043         if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3044                 spin_lock(&fs_info->data_sinfo->lock);
3045                 bytes_used = fs_info->data_sinfo->bytes_used;
3046                 spin_unlock(&fs_info->data_sinfo->lock);
3047
3048                 if (!bytes_used) {
3049                         struct btrfs_trans_handle *trans;
3050                         int ret;
3051
3052                         trans = btrfs_join_transaction(fs_info->tree_root);
3053                         if (IS_ERR(trans))
3054                                 return PTR_ERR(trans);
3055
3056                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3057                                                       BTRFS_BLOCK_GROUP_DATA);
3058                         btrfs_end_transaction(trans);
3059                         if (ret < 0)
3060                                 return ret;
3061
3062                         btrfs_add_raid_kobjects(fs_info);
3063
3064                         return 1;
3065                 }
3066         }
3067         return 0;
3068 }
3069
3070 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3071                                struct btrfs_balance_control *bctl)
3072 {
3073         struct btrfs_root *root = fs_info->tree_root;
3074         struct btrfs_trans_handle *trans;
3075         struct btrfs_balance_item *item;
3076         struct btrfs_disk_balance_args disk_bargs;
3077         struct btrfs_path *path;
3078         struct extent_buffer *leaf;
3079         struct btrfs_key key;
3080         int ret, err;
3081
3082         path = btrfs_alloc_path();
3083         if (!path)
3084                 return -ENOMEM;
3085
3086         trans = btrfs_start_transaction(root, 0);
3087         if (IS_ERR(trans)) {
3088                 btrfs_free_path(path);
3089                 return PTR_ERR(trans);
3090         }
3091
3092         key.objectid = BTRFS_BALANCE_OBJECTID;
3093         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3094         key.offset = 0;
3095
3096         ret = btrfs_insert_empty_item(trans, root, path, &key,
3097                                       sizeof(*item));
3098         if (ret)
3099                 goto out;
3100
3101         leaf = path->nodes[0];
3102         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3103
3104         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3105
3106         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3107         btrfs_set_balance_data(leaf, item, &disk_bargs);
3108         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3109         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3110         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3111         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3112
3113         btrfs_set_balance_flags(leaf, item, bctl->flags);
3114
3115         btrfs_mark_buffer_dirty(leaf);
3116 out:
3117         btrfs_free_path(path);
3118         err = btrfs_commit_transaction(trans);
3119         if (err && !ret)
3120                 ret = err;
3121         return ret;
3122 }
3123
3124 static int del_balance_item(struct btrfs_fs_info *fs_info)
3125 {
3126         struct btrfs_root *root = fs_info->tree_root;
3127         struct btrfs_trans_handle *trans;
3128         struct btrfs_path *path;
3129         struct btrfs_key key;
3130         int ret, err;
3131
3132         path = btrfs_alloc_path();
3133         if (!path)
3134                 return -ENOMEM;
3135
3136         trans = btrfs_start_transaction(root, 0);
3137         if (IS_ERR(trans)) {
3138                 btrfs_free_path(path);
3139                 return PTR_ERR(trans);
3140         }
3141
3142         key.objectid = BTRFS_BALANCE_OBJECTID;
3143         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3144         key.offset = 0;
3145
3146         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3147         if (ret < 0)
3148                 goto out;
3149         if (ret > 0) {
3150                 ret = -ENOENT;
3151                 goto out;
3152         }
3153
3154         ret = btrfs_del_item(trans, root, path);
3155 out:
3156         btrfs_free_path(path);
3157         err = btrfs_commit_transaction(trans);
3158         if (err && !ret)
3159                 ret = err;
3160         return ret;
3161 }
3162
3163 /*
3164  * This is a heuristic used to reduce the number of chunks balanced on
3165  * resume after balance was interrupted.
3166  */
3167 static void update_balance_args(struct btrfs_balance_control *bctl)
3168 {
3169         /*
3170          * Turn on soft mode for chunk types that were being converted.
3171          */
3172         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3173                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3174         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3175                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3176         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3177                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3178
3179         /*
3180          * Turn on usage filter if is not already used.  The idea is
3181          * that chunks that we have already balanced should be
3182          * reasonably full.  Don't do it for chunks that are being
3183          * converted - that will keep us from relocating unconverted
3184          * (albeit full) chunks.
3185          */
3186         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3187             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3188             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3189                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3190                 bctl->data.usage = 90;
3191         }
3192         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3193             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3194             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3195                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3196                 bctl->sys.usage = 90;
3197         }
3198         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3199             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3200             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3201                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3202                 bctl->meta.usage = 90;
3203         }
3204 }
3205
3206 /*
3207  * Clear the balance status in fs_info and delete the balance item from disk.
3208  */
3209 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3210 {
3211         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3212         int ret;
3213
3214         BUG_ON(!fs_info->balance_ctl);
3215
3216         spin_lock(&fs_info->balance_lock);
3217         fs_info->balance_ctl = NULL;
3218         spin_unlock(&fs_info->balance_lock);
3219
3220         kfree(bctl);
3221         ret = del_balance_item(fs_info);
3222         if (ret)
3223                 btrfs_handle_fs_error(fs_info, ret, NULL);
3224 }
3225
3226 /*
3227  * Balance filters.  Return 1 if chunk should be filtered out
3228  * (should not be balanced).
3229  */
3230 static int chunk_profiles_filter(u64 chunk_type,
3231                                  struct btrfs_balance_args *bargs)
3232 {
3233         chunk_type = chunk_to_extended(chunk_type) &
3234                                 BTRFS_EXTENDED_PROFILE_MASK;
3235
3236         if (bargs->profiles & chunk_type)
3237                 return 0;
3238
3239         return 1;
3240 }
3241
3242 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3243                               struct btrfs_balance_args *bargs)