btrfs: use existing cur_devices, cleanup btrfs_rm_device
[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
1236         bdev = blkdev_get_by_path(path, flags, holder);
1237         if (IS_ERR(bdev))
1238                 return PTR_ERR(bdev);
1239
1240         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1241                 ret = -EINVAL;
1242                 goto error_bdev_put;
1243         }
1244
1245         mutex_lock(&uuid_mutex);
1246         device = device_list_add(path, disk_super);
1247         if (IS_ERR(device))
1248                 ret = PTR_ERR(device);
1249         else
1250                 *fs_devices_ret = device->fs_devices;
1251         mutex_unlock(&uuid_mutex);
1252
1253         btrfs_release_disk_super(page);
1254
1255 error_bdev_put:
1256         blkdev_put(bdev, flags);
1257
1258         return ret;
1259 }
1260
1261 /* helper to account the used device space in the range */
1262 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1263                                    u64 end, u64 *length)
1264 {
1265         struct btrfs_key key;
1266         struct btrfs_root *root = device->fs_info->dev_root;
1267         struct btrfs_dev_extent *dev_extent;
1268         struct btrfs_path *path;
1269         u64 extent_end;
1270         int ret;
1271         int slot;
1272         struct extent_buffer *l;
1273
1274         *length = 0;
1275
1276         if (start >= device->total_bytes ||
1277                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1278                 return 0;
1279
1280         path = btrfs_alloc_path();
1281         if (!path)
1282                 return -ENOMEM;
1283         path->reada = READA_FORWARD;
1284
1285         key.objectid = device->devid;
1286         key.offset = start;
1287         key.type = BTRFS_DEV_EXTENT_KEY;
1288
1289         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1290         if (ret < 0)
1291                 goto out;
1292         if (ret > 0) {
1293                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1294                 if (ret < 0)
1295                         goto out;
1296         }
1297
1298         while (1) {
1299                 l = path->nodes[0];
1300                 slot = path->slots[0];
1301                 if (slot >= btrfs_header_nritems(l)) {
1302                         ret = btrfs_next_leaf(root, path);
1303                         if (ret == 0)
1304                                 continue;
1305                         if (ret < 0)
1306                                 goto out;
1307
1308                         break;
1309                 }
1310                 btrfs_item_key_to_cpu(l, &key, slot);
1311
1312                 if (key.objectid < device->devid)
1313                         goto next;
1314
1315                 if (key.objectid > device->devid)
1316                         break;
1317
1318                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1319                         goto next;
1320
1321                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1322                 extent_end = key.offset + btrfs_dev_extent_length(l,
1323                                                                   dev_extent);
1324                 if (key.offset <= start && extent_end > end) {
1325                         *length = end - start + 1;
1326                         break;
1327                 } else if (key.offset <= start && extent_end > start)
1328                         *length += extent_end - start;
1329                 else if (key.offset > start && extent_end <= end)
1330                         *length += extent_end - key.offset;
1331                 else if (key.offset > start && key.offset <= end) {
1332                         *length += end - key.offset + 1;
1333                         break;
1334                 } else if (key.offset > end)
1335                         break;
1336
1337 next:
1338                 path->slots[0]++;
1339         }
1340         ret = 0;
1341 out:
1342         btrfs_free_path(path);
1343         return ret;
1344 }
1345
1346 static int contains_pending_extent(struct btrfs_transaction *transaction,
1347                                    struct btrfs_device *device,
1348                                    u64 *start, u64 len)
1349 {
1350         struct btrfs_fs_info *fs_info = device->fs_info;
1351         struct extent_map *em;
1352         struct list_head *search_list = &fs_info->pinned_chunks;
1353         int ret = 0;
1354         u64 physical_start = *start;
1355
1356         if (transaction)
1357                 search_list = &transaction->pending_chunks;
1358 again:
1359         list_for_each_entry(em, search_list, list) {
1360                 struct map_lookup *map;
1361                 int i;
1362
1363                 map = em->map_lookup;
1364                 for (i = 0; i < map->num_stripes; i++) {
1365                         u64 end;
1366
1367                         if (map->stripes[i].dev != device)
1368                                 continue;
1369                         if (map->stripes[i].physical >= physical_start + len ||
1370                             map->stripes[i].physical + em->orig_block_len <=
1371                             physical_start)
1372                                 continue;
1373                         /*
1374                          * Make sure that while processing the pinned list we do
1375                          * not override our *start with a lower value, because
1376                          * we can have pinned chunks that fall within this
1377                          * device hole and that have lower physical addresses
1378                          * than the pending chunks we processed before. If we
1379                          * do not take this special care we can end up getting
1380                          * 2 pending chunks that start at the same physical
1381                          * device offsets because the end offset of a pinned
1382                          * chunk can be equal to the start offset of some
1383                          * pending chunk.
1384                          */
1385                         end = map->stripes[i].physical + em->orig_block_len;
1386                         if (end > *start) {
1387                                 *start = end;
1388                                 ret = 1;
1389                         }
1390                 }
1391         }
1392         if (search_list != &fs_info->pinned_chunks) {
1393                 search_list = &fs_info->pinned_chunks;
1394                 goto again;
1395         }
1396
1397         return ret;
1398 }
1399
1400
1401 /*
1402  * find_free_dev_extent_start - find free space in the specified device
1403  * @device:       the device which we search the free space in
1404  * @num_bytes:    the size of the free space that we need
1405  * @search_start: the position from which to begin the search
1406  * @start:        store the start of the free space.
1407  * @len:          the size of the free space. that we find, or the size
1408  *                of the max free space if we don't find suitable free space
1409  *
1410  * this uses a pretty simple search, the expectation is that it is
1411  * called very infrequently and that a given device has a small number
1412  * of extents
1413  *
1414  * @start is used to store the start of the free space if we find. But if we
1415  * don't find suitable free space, it will be used to store the start position
1416  * of the max free space.
1417  *
1418  * @len is used to store the size of the free space that we find.
1419  * But if we don't find suitable free space, it is used to store the size of
1420  * the max free space.
1421  */
1422 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1423                                struct btrfs_device *device, u64 num_bytes,
1424                                u64 search_start, u64 *start, u64 *len)
1425 {
1426         struct btrfs_fs_info *fs_info = device->fs_info;
1427         struct btrfs_root *root = fs_info->dev_root;
1428         struct btrfs_key key;
1429         struct btrfs_dev_extent *dev_extent;
1430         struct btrfs_path *path;
1431         u64 hole_size;
1432         u64 max_hole_start;
1433         u64 max_hole_size;
1434         u64 extent_end;
1435         u64 search_end = device->total_bytes;
1436         int ret;
1437         int slot;
1438         struct extent_buffer *l;
1439
1440         /*
1441          * We don't want to overwrite the superblock on the drive nor any area
1442          * used by the boot loader (grub for example), so we make sure to start
1443          * at an offset of at least 1MB.
1444          */
1445         search_start = max_t(u64, search_start, SZ_1M);
1446
1447         path = btrfs_alloc_path();
1448         if (!path)
1449                 return -ENOMEM;
1450
1451         max_hole_start = search_start;
1452         max_hole_size = 0;
1453
1454 again:
1455         if (search_start >= search_end ||
1456                 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1457                 ret = -ENOSPC;
1458                 goto out;
1459         }
1460
1461         path->reada = READA_FORWARD;
1462         path->search_commit_root = 1;
1463         path->skip_locking = 1;
1464
1465         key.objectid = device->devid;
1466         key.offset = search_start;
1467         key.type = BTRFS_DEV_EXTENT_KEY;
1468
1469         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1470         if (ret < 0)
1471                 goto out;
1472         if (ret > 0) {
1473                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1474                 if (ret < 0)
1475                         goto out;
1476         }
1477
1478         while (1) {
1479                 l = path->nodes[0];
1480                 slot = path->slots[0];
1481                 if (slot >= btrfs_header_nritems(l)) {
1482                         ret = btrfs_next_leaf(root, path);
1483                         if (ret == 0)
1484                                 continue;
1485                         if (ret < 0)
1486                                 goto out;
1487
1488                         break;
1489                 }
1490                 btrfs_item_key_to_cpu(l, &key, slot);
1491
1492                 if (key.objectid < device->devid)
1493                         goto next;
1494
1495                 if (key.objectid > device->devid)
1496                         break;
1497
1498                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1499                         goto next;
1500
1501                 if (key.offset > search_start) {
1502                         hole_size = key.offset - search_start;
1503
1504                         /*
1505                          * Have to check before we set max_hole_start, otherwise
1506                          * we could end up sending back this offset anyway.
1507                          */
1508                         if (contains_pending_extent(transaction, device,
1509                                                     &search_start,
1510                                                     hole_size)) {
1511                                 if (key.offset >= search_start) {
1512                                         hole_size = key.offset - search_start;
1513                                 } else {
1514                                         WARN_ON_ONCE(1);
1515                                         hole_size = 0;
1516                                 }
1517                         }
1518
1519                         if (hole_size > max_hole_size) {
1520                                 max_hole_start = search_start;
1521                                 max_hole_size = hole_size;
1522                         }
1523
1524                         /*
1525                          * If this free space is greater than which we need,
1526                          * it must be the max free space that we have found
1527                          * until now, so max_hole_start must point to the start
1528                          * of this free space and the length of this free space
1529                          * is stored in max_hole_size. Thus, we return
1530                          * max_hole_start and max_hole_size and go back to the
1531                          * caller.
1532                          */
1533                         if (hole_size >= num_bytes) {
1534                                 ret = 0;
1535                                 goto out;
1536                         }
1537                 }
1538
1539                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1540                 extent_end = key.offset + btrfs_dev_extent_length(l,
1541                                                                   dev_extent);
1542                 if (extent_end > search_start)
1543                         search_start = extent_end;
1544 next:
1545                 path->slots[0]++;
1546                 cond_resched();
1547         }
1548
1549         /*
1550          * At this point, search_start should be the end of
1551          * allocated dev extents, and when shrinking the device,
1552          * search_end may be smaller than search_start.
1553          */
1554         if (search_end > search_start) {
1555                 hole_size = search_end - search_start;
1556
1557                 if (contains_pending_extent(transaction, device, &search_start,
1558                                             hole_size)) {
1559                         btrfs_release_path(path);
1560                         goto again;
1561                 }
1562
1563                 if (hole_size > max_hole_size) {
1564                         max_hole_start = search_start;
1565                         max_hole_size = hole_size;
1566                 }
1567         }
1568
1569         /* See above. */
1570         if (max_hole_size < num_bytes)
1571                 ret = -ENOSPC;
1572         else
1573                 ret = 0;
1574
1575 out:
1576         btrfs_free_path(path);
1577         *start = max_hole_start;
1578         if (len)
1579                 *len = max_hole_size;
1580         return ret;
1581 }
1582
1583 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1584                          struct btrfs_device *device, u64 num_bytes,
1585                          u64 *start, u64 *len)
1586 {
1587         /* FIXME use last free of some kind */
1588         return find_free_dev_extent_start(trans->transaction, device,
1589                                           num_bytes, 0, start, len);
1590 }
1591
1592 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1593                           struct btrfs_device *device,
1594                           u64 start, u64 *dev_extent_len)
1595 {
1596         struct btrfs_fs_info *fs_info = device->fs_info;
1597         struct btrfs_root *root = fs_info->dev_root;
1598         int ret;
1599         struct btrfs_path *path;
1600         struct btrfs_key key;
1601         struct btrfs_key found_key;
1602         struct extent_buffer *leaf = NULL;
1603         struct btrfs_dev_extent *extent = NULL;
1604
1605         path = btrfs_alloc_path();
1606         if (!path)
1607                 return -ENOMEM;
1608
1609         key.objectid = device->devid;
1610         key.offset = start;
1611         key.type = BTRFS_DEV_EXTENT_KEY;
1612 again:
1613         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1614         if (ret > 0) {
1615                 ret = btrfs_previous_item(root, path, key.objectid,
1616                                           BTRFS_DEV_EXTENT_KEY);
1617                 if (ret)
1618                         goto out;
1619                 leaf = path->nodes[0];
1620                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1621                 extent = btrfs_item_ptr(leaf, path->slots[0],
1622                                         struct btrfs_dev_extent);
1623                 BUG_ON(found_key.offset > start || found_key.offset +
1624                        btrfs_dev_extent_length(leaf, extent) < start);
1625                 key = found_key;
1626                 btrfs_release_path(path);
1627                 goto again;
1628         } else if (ret == 0) {
1629                 leaf = path->nodes[0];
1630                 extent = btrfs_item_ptr(leaf, path->slots[0],
1631                                         struct btrfs_dev_extent);
1632         } else {
1633                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1634                 goto out;
1635         }
1636
1637         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1638
1639         ret = btrfs_del_item(trans, root, path);
1640         if (ret) {
1641                 btrfs_handle_fs_error(fs_info, ret,
1642                                       "Failed to remove dev extent item");
1643         } else {
1644                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1645         }
1646 out:
1647         btrfs_free_path(path);
1648         return ret;
1649 }
1650
1651 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1652                                   struct btrfs_device *device,
1653                                   u64 chunk_offset, u64 start, u64 num_bytes)
1654 {
1655         int ret;
1656         struct btrfs_path *path;
1657         struct btrfs_fs_info *fs_info = device->fs_info;
1658         struct btrfs_root *root = fs_info->dev_root;
1659         struct btrfs_dev_extent *extent;
1660         struct extent_buffer *leaf;
1661         struct btrfs_key key;
1662
1663         WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1664         WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1665         path = btrfs_alloc_path();
1666         if (!path)
1667                 return -ENOMEM;
1668
1669         key.objectid = device->devid;
1670         key.offset = start;
1671         key.type = BTRFS_DEV_EXTENT_KEY;
1672         ret = btrfs_insert_empty_item(trans, root, path, &key,
1673                                       sizeof(*extent));
1674         if (ret)
1675                 goto out;
1676
1677         leaf = path->nodes[0];
1678         extent = btrfs_item_ptr(leaf, path->slots[0],
1679                                 struct btrfs_dev_extent);
1680         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1681                                         BTRFS_CHUNK_TREE_OBJECTID);
1682         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1683                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1684         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1685
1686         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1687         btrfs_mark_buffer_dirty(leaf);
1688 out:
1689         btrfs_free_path(path);
1690         return ret;
1691 }
1692
1693 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1694 {
1695         struct extent_map_tree *em_tree;
1696         struct extent_map *em;
1697         struct rb_node *n;
1698         u64 ret = 0;
1699
1700         em_tree = &fs_info->mapping_tree.map_tree;
1701         read_lock(&em_tree->lock);
1702         n = rb_last(&em_tree->map);
1703         if (n) {
1704                 em = rb_entry(n, struct extent_map, rb_node);
1705                 ret = em->start + em->len;
1706         }
1707         read_unlock(&em_tree->lock);
1708
1709         return ret;
1710 }
1711
1712 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1713                                     u64 *devid_ret)
1714 {
1715         int ret;
1716         struct btrfs_key key;
1717         struct btrfs_key found_key;
1718         struct btrfs_path *path;
1719
1720         path = btrfs_alloc_path();
1721         if (!path)
1722                 return -ENOMEM;
1723
1724         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1725         key.type = BTRFS_DEV_ITEM_KEY;
1726         key.offset = (u64)-1;
1727
1728         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1729         if (ret < 0)
1730                 goto error;
1731
1732         BUG_ON(ret == 0); /* Corruption */
1733
1734         ret = btrfs_previous_item(fs_info->chunk_root, path,
1735                                   BTRFS_DEV_ITEMS_OBJECTID,
1736                                   BTRFS_DEV_ITEM_KEY);
1737         if (ret) {
1738                 *devid_ret = 1;
1739         } else {
1740                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1741                                       path->slots[0]);
1742                 *devid_ret = found_key.offset + 1;
1743         }
1744         ret = 0;
1745 error:
1746         btrfs_free_path(path);
1747         return ret;
1748 }
1749
1750 /*
1751  * the device information is stored in the chunk root
1752  * the btrfs_device struct should be fully filled in
1753  */
1754 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1755                             struct btrfs_fs_info *fs_info,
1756                             struct btrfs_device *device)
1757 {
1758         struct btrfs_root *root = fs_info->chunk_root;
1759         int ret;
1760         struct btrfs_path *path;
1761         struct btrfs_dev_item *dev_item;
1762         struct extent_buffer *leaf;
1763         struct btrfs_key key;
1764         unsigned long ptr;
1765
1766         path = btrfs_alloc_path();
1767         if (!path)
1768                 return -ENOMEM;
1769
1770         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1771         key.type = BTRFS_DEV_ITEM_KEY;
1772         key.offset = device->devid;
1773
1774         ret = btrfs_insert_empty_item(trans, root, path, &key,
1775                                       sizeof(*dev_item));
1776         if (ret)
1777                 goto out;
1778
1779         leaf = path->nodes[0];
1780         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1781
1782         btrfs_set_device_id(leaf, dev_item, device->devid);
1783         btrfs_set_device_generation(leaf, dev_item, 0);
1784         btrfs_set_device_type(leaf, dev_item, device->type);
1785         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1786         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1787         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1788         btrfs_set_device_total_bytes(leaf, dev_item,
1789                                      btrfs_device_get_disk_total_bytes(device));
1790         btrfs_set_device_bytes_used(leaf, dev_item,
1791                                     btrfs_device_get_bytes_used(device));
1792         btrfs_set_device_group(leaf, dev_item, 0);
1793         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1794         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1795         btrfs_set_device_start_offset(leaf, dev_item, 0);
1796
1797         ptr = btrfs_device_uuid(dev_item);
1798         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1799         ptr = btrfs_device_fsid(dev_item);
1800         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1801         btrfs_mark_buffer_dirty(leaf);
1802
1803         ret = 0;
1804 out:
1805         btrfs_free_path(path);
1806         return ret;
1807 }
1808
1809 /*
1810  * Function to update ctime/mtime for a given device path.
1811  * Mainly used for ctime/mtime based probe like libblkid.
1812  */
1813 static void update_dev_time(const char *path_name)
1814 {
1815         struct file *filp;
1816
1817         filp = filp_open(path_name, O_RDWR, 0);
1818         if (IS_ERR(filp))
1819                 return;
1820         file_update_time(filp);
1821         filp_close(filp, NULL);
1822 }
1823
1824 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1825                              struct btrfs_device *device)
1826 {
1827         struct btrfs_root *root = fs_info->chunk_root;
1828         int ret;
1829         struct btrfs_path *path;
1830         struct btrfs_key key;
1831         struct btrfs_trans_handle *trans;
1832
1833         path = btrfs_alloc_path();
1834         if (!path)
1835                 return -ENOMEM;
1836
1837         trans = btrfs_start_transaction(root, 0);
1838         if (IS_ERR(trans)) {
1839                 btrfs_free_path(path);
1840                 return PTR_ERR(trans);
1841         }
1842         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1843         key.type = BTRFS_DEV_ITEM_KEY;
1844         key.offset = device->devid;
1845
1846         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1847         if (ret) {
1848                 if (ret > 0)
1849                         ret = -ENOENT;
1850                 btrfs_abort_transaction(trans, ret);
1851                 btrfs_end_transaction(trans);
1852                 goto out;
1853         }
1854
1855         ret = btrfs_del_item(trans, root, path);
1856         if (ret) {
1857                 btrfs_abort_transaction(trans, ret);
1858                 btrfs_end_transaction(trans);
1859         }
1860
1861 out:
1862         btrfs_free_path(path);
1863         if (!ret)
1864                 ret = btrfs_commit_transaction(trans);
1865         return ret;
1866 }
1867
1868 /*
1869  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1870  * filesystem. It's up to the caller to adjust that number regarding eg. device
1871  * replace.
1872  */
1873 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1874                 u64 num_devices)
1875 {
1876         u64 all_avail;
1877         unsigned seq;
1878         int i;
1879
1880         do {
1881                 seq = read_seqbegin(&fs_info->profiles_lock);
1882
1883                 all_avail = fs_info->avail_data_alloc_bits |
1884                             fs_info->avail_system_alloc_bits |
1885                             fs_info->avail_metadata_alloc_bits;
1886         } while (read_seqretry(&fs_info->profiles_lock, seq));
1887
1888         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1889                 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1890                         continue;
1891
1892                 if (num_devices < btrfs_raid_array[i].devs_min) {
1893                         int ret = btrfs_raid_array[i].mindev_error;
1894
1895                         if (ret)
1896                                 return ret;
1897                 }
1898         }
1899
1900         return 0;
1901 }
1902
1903 static struct btrfs_device * btrfs_find_next_active_device(
1904                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1905 {
1906         struct btrfs_device *next_device;
1907
1908         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1909                 if (next_device != device &&
1910                     !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1911                     && next_device->bdev)
1912                         return next_device;
1913         }
1914
1915         return NULL;
1916 }
1917
1918 /*
1919  * Helper function to check if the given device is part of s_bdev / latest_bdev
1920  * and replace it with the provided or the next active device, in the context
1921  * where this function called, there should be always be another device (or
1922  * this_dev) which is active.
1923  */
1924 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1925                 struct btrfs_device *device, struct btrfs_device *this_dev)
1926 {
1927         struct btrfs_device *next_device;
1928
1929         if (this_dev)
1930                 next_device = this_dev;
1931         else
1932                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1933                                                                 device);
1934         ASSERT(next_device);
1935
1936         if (fs_info->sb->s_bdev &&
1937                         (fs_info->sb->s_bdev == device->bdev))
1938                 fs_info->sb->s_bdev = next_device->bdev;
1939
1940         if (fs_info->fs_devices->latest_bdev == device->bdev)
1941                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1942 }
1943
1944 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1945                 u64 devid)
1946 {
1947         struct btrfs_device *device;
1948         struct btrfs_fs_devices *cur_devices;
1949         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1950         u64 num_devices;
1951         int ret = 0;
1952
1953         mutex_lock(&uuid_mutex);
1954
1955         num_devices = fs_devices->num_devices;
1956         btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1957         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1958                 WARN_ON(num_devices < 1);
1959                 num_devices--;
1960         }
1961         btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1962
1963         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1964         if (ret)
1965                 goto out;
1966
1967         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1968                                            &device);
1969         if (ret)
1970                 goto out;
1971
1972         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1973                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1974                 goto out;
1975         }
1976
1977         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1978             fs_info->fs_devices->rw_devices == 1) {
1979                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1980                 goto out;
1981         }
1982
1983         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1984                 mutex_lock(&fs_info->chunk_mutex);
1985                 list_del_init(&device->dev_alloc_list);
1986                 device->fs_devices->rw_devices--;
1987                 mutex_unlock(&fs_info->chunk_mutex);
1988         }
1989
1990         mutex_unlock(&uuid_mutex);
1991         ret = btrfs_shrink_device(device, 0);
1992         mutex_lock(&uuid_mutex);
1993         if (ret)
1994                 goto error_undo;
1995
1996         /*
1997          * TODO: the superblock still includes this device in its num_devices
1998          * counter although write_all_supers() is not locked out. This
1999          * could give a filesystem state which requires a degraded mount.
2000          */
2001         ret = btrfs_rm_dev_item(fs_info, device);
2002         if (ret)
2003                 goto error_undo;
2004
2005         clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2006         btrfs_scrub_cancel_dev(fs_info, device);
2007
2008         /*
2009          * the device list mutex makes sure that we don't change
2010          * the device list while someone else is writing out all
2011          * the device supers. Whoever is writing all supers, should
2012          * lock the device list mutex before getting the number of
2013          * devices in the super block (super_copy). Conversely,
2014          * whoever updates the number of devices in the super block
2015          * (super_copy) should hold the device list mutex.
2016          */
2017
2018         /*
2019          * In normal cases the cur_devices == fs_devices. But in case
2020          * of deleting a seed device, the cur_devices should point to
2021          * its own fs_devices listed under the fs_devices->seed.
2022          */
2023         cur_devices = device->fs_devices;
2024         mutex_lock(&fs_devices->device_list_mutex);
2025         list_del_rcu(&device->dev_list);
2026
2027         cur_devices->num_devices--;
2028         cur_devices->total_devices--;
2029
2030         if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2031                 cur_devices->missing_devices--;
2032
2033         btrfs_assign_next_active_device(fs_info, device, NULL);
2034
2035         if (device->bdev) {
2036                 cur_devices->open_devices--;
2037                 /* remove sysfs entry */
2038                 btrfs_sysfs_rm_device_link(fs_devices, device);
2039         }
2040
2041         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2042         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2043         mutex_unlock(&fs_devices->device_list_mutex);
2044
2045         /*
2046          * at this point, the device is zero sized and detached from
2047          * the devices list.  All that's left is to zero out the old
2048          * supers and free the device.
2049          */
2050         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2051                 btrfs_scratch_superblocks(device->bdev, device->name->str);
2052
2053         btrfs_close_bdev(device);
2054         call_rcu(&device->rcu, free_device_rcu);
2055
2056         if (cur_devices->open_devices == 0) {
2057                 while (fs_devices) {
2058                         if (fs_devices->seed == cur_devices) {
2059                                 fs_devices->seed = cur_devices->seed;
2060                                 break;
2061                         }
2062                         fs_devices = fs_devices->seed;
2063                 }
2064                 cur_devices->seed = NULL;
2065                 close_fs_devices(cur_devices);
2066                 free_fs_devices(cur_devices);
2067         }
2068
2069 out:
2070         mutex_unlock(&uuid_mutex);
2071         return ret;
2072
2073 error_undo:
2074         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2075                 mutex_lock(&fs_info->chunk_mutex);
2076                 list_add(&device->dev_alloc_list,
2077                          &fs_devices->alloc_list);
2078                 device->fs_devices->rw_devices++;
2079                 mutex_unlock(&fs_info->chunk_mutex);
2080         }
2081         goto out;
2082 }
2083
2084 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2085                                         struct btrfs_device *srcdev)
2086 {
2087         struct btrfs_fs_devices *fs_devices;
2088
2089         lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2090
2091         /*
2092          * in case of fs with no seed, srcdev->fs_devices will point
2093          * to fs_devices of fs_info. However when the dev being replaced is
2094          * a seed dev it will point to the seed's local fs_devices. In short
2095          * srcdev will have its correct fs_devices in both the cases.
2096          */
2097         fs_devices = srcdev->fs_devices;
2098
2099         list_del_rcu(&srcdev->dev_list);
2100         list_del(&srcdev->dev_alloc_list);
2101         fs_devices->num_devices--;
2102         if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2103                 fs_devices->missing_devices--;
2104
2105         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2106                 fs_devices->rw_devices--;
2107
2108         if (srcdev->bdev)
2109                 fs_devices->open_devices--;
2110 }
2111
2112 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2113                                       struct btrfs_device *srcdev)
2114 {
2115         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2116
2117         if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2118                 /* zero out the old super if it is writable */
2119                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2120         }
2121
2122         btrfs_close_bdev(srcdev);
2123         call_rcu(&srcdev->rcu, free_device_rcu);
2124
2125         /* if this is no devs we rather delete the fs_devices */
2126         if (!fs_devices->num_devices) {
2127                 struct btrfs_fs_devices *tmp_fs_devices;
2128
2129                 /*
2130                  * On a mounted FS, num_devices can't be zero unless it's a
2131                  * seed. In case of a seed device being replaced, the replace
2132                  * target added to the sprout FS, so there will be no more
2133                  * device left under the seed FS.
2134                  */
2135                 ASSERT(fs_devices->seeding);
2136
2137                 tmp_fs_devices = fs_info->fs_devices;
2138                 while (tmp_fs_devices) {
2139                         if (tmp_fs_devices->seed == fs_devices) {
2140                                 tmp_fs_devices->seed = fs_devices->seed;
2141                                 break;
2142                         }
2143                         tmp_fs_devices = tmp_fs_devices->seed;
2144                 }
2145                 fs_devices->seed = NULL;
2146                 close_fs_devices(fs_devices);
2147                 free_fs_devices(fs_devices);
2148         }
2149 }
2150
2151 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2152                                       struct btrfs_device *tgtdev)
2153 {
2154         mutex_lock(&uuid_mutex);
2155         WARN_ON(!tgtdev);
2156         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2157
2158         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2159
2160         if (tgtdev->bdev)
2161                 fs_info->fs_devices->open_devices--;
2162
2163         fs_info->fs_devices->num_devices--;
2164
2165         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2166
2167         list_del_rcu(&tgtdev->dev_list);
2168
2169         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2170         mutex_unlock(&uuid_mutex);
2171
2172         /*
2173          * The update_dev_time() with in btrfs_scratch_superblocks()
2174          * may lead to a call to btrfs_show_devname() which will try
2175          * to hold device_list_mutex. And here this device
2176          * is already out of device list, so we don't have to hold
2177          * the device_list_mutex lock.
2178          */
2179         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2180
2181         btrfs_close_bdev(tgtdev);
2182         call_rcu(&tgtdev->rcu, free_device_rcu);
2183 }
2184
2185 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2186                                      const char *device_path,
2187                                      struct btrfs_device **device)
2188 {
2189         int ret = 0;
2190         struct btrfs_super_block *disk_super;
2191         u64 devid;
2192         u8 *dev_uuid;
2193         struct block_device *bdev;
2194         struct buffer_head *bh;
2195
2196         *device = NULL;
2197         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2198                                     fs_info->bdev_holder, 0, &bdev, &bh);
2199         if (ret)
2200                 return ret;
2201         disk_super = (struct btrfs_super_block *)bh->b_data;
2202         devid = btrfs_stack_device_id(&disk_super->dev_item);
2203         dev_uuid = disk_super->dev_item.uuid;
2204         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2205         brelse(bh);
2206         if (!*device)
2207                 ret = -ENOENT;
2208         blkdev_put(bdev, FMODE_READ);
2209         return ret;
2210 }
2211
2212 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2213                                          const char *device_path,
2214                                          struct btrfs_device **device)
2215 {
2216         *device = NULL;
2217         if (strcmp(device_path, "missing") == 0) {
2218                 struct list_head *devices;
2219                 struct btrfs_device *tmp;
2220
2221                 devices = &fs_info->fs_devices->devices;
2222                 list_for_each_entry(tmp, devices, dev_list) {
2223                         if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2224                                         &tmp->dev_state) && !tmp->bdev) {
2225                                 *device = tmp;
2226                                 break;
2227                         }
2228                 }
2229
2230                 if (!*device)
2231                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2232
2233                 return 0;
2234         } else {
2235                 return btrfs_find_device_by_path(fs_info, device_path, device);
2236         }
2237 }
2238
2239 /*
2240  * Lookup a device given by device id, or the path if the id is 0.
2241  */
2242 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2243                                  const char *devpath,
2244                                  struct btrfs_device **device)
2245 {
2246         int ret;
2247
2248         if (devid) {
2249                 ret = 0;
2250                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2251                 if (!*device)
2252                         ret = -ENOENT;
2253         } else {
2254                 if (!devpath || !devpath[0])
2255                         return -EINVAL;
2256
2257                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2258                                                            device);
2259         }
2260         return ret;
2261 }
2262
2263 /*
2264  * does all the dirty work required for changing file system's UUID.
2265  */
2266 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2267 {
2268         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2269         struct btrfs_fs_devices *old_devices;
2270         struct btrfs_fs_devices *seed_devices;
2271         struct btrfs_super_block *disk_super = fs_info->super_copy;
2272         struct btrfs_device *device;
2273         u64 super_flags;
2274
2275         lockdep_assert_held(&uuid_mutex);
2276         if (!fs_devices->seeding)
2277                 return -EINVAL;
2278
2279         seed_devices = alloc_fs_devices(NULL);
2280         if (IS_ERR(seed_devices))
2281                 return PTR_ERR(seed_devices);
2282
2283         old_devices = clone_fs_devices(fs_devices);
2284         if (IS_ERR(old_devices)) {
2285                 kfree(seed_devices);
2286                 return PTR_ERR(old_devices);
2287         }
2288
2289         list_add(&old_devices->fs_list, &fs_uuids);
2290
2291         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2292         seed_devices->opened = 1;
2293         INIT_LIST_HEAD(&seed_devices->devices);
2294         INIT_LIST_HEAD(&seed_devices->alloc_list);
2295         mutex_init(&seed_devices->device_list_mutex);
2296
2297         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2298         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2299                               synchronize_rcu);
2300         list_for_each_entry(device, &seed_devices->devices, dev_list)
2301                 device->fs_devices = seed_devices;
2302
2303         mutex_lock(&fs_info->chunk_mutex);
2304         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2305         mutex_unlock(&fs_info->chunk_mutex);
2306
2307         fs_devices->seeding = 0;
2308         fs_devices->num_devices = 0;
2309         fs_devices->open_devices = 0;
2310         fs_devices->missing_devices = 0;
2311         fs_devices->rotating = 0;
2312         fs_devices->seed = seed_devices;
2313
2314         generate_random_uuid(fs_devices->fsid);
2315         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2316         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2317         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2318
2319         super_flags = btrfs_super_flags(disk_super) &
2320                       ~BTRFS_SUPER_FLAG_SEEDING;
2321         btrfs_set_super_flags(disk_super, super_flags);
2322
2323         return 0;
2324 }
2325
2326 /*
2327  * Store the expected generation for seed devices in device items.
2328  */
2329 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2330                                struct btrfs_fs_info *fs_info)
2331 {
2332         struct btrfs_root *root = fs_info->chunk_root;
2333         struct btrfs_path *path;
2334         struct extent_buffer *leaf;
2335         struct btrfs_dev_item *dev_item;
2336         struct btrfs_device *device;
2337         struct btrfs_key key;
2338         u8 fs_uuid[BTRFS_FSID_SIZE];
2339         u8 dev_uuid[BTRFS_UUID_SIZE];
2340         u64 devid;
2341         int ret;
2342
2343         path = btrfs_alloc_path();
2344         if (!path)
2345                 return -ENOMEM;
2346
2347         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2348         key.offset = 0;
2349         key.type = BTRFS_DEV_ITEM_KEY;
2350
2351         while (1) {
2352                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2353                 if (ret < 0)
2354                         goto error;
2355
2356                 leaf = path->nodes[0];
2357 next_slot:
2358                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2359                         ret = btrfs_next_leaf(root, path);
2360                         if (ret > 0)
2361                                 break;
2362                         if (ret < 0)
2363                                 goto error;
2364                         leaf = path->nodes[0];
2365                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2366                         btrfs_release_path(path);
2367                         continue;
2368                 }
2369
2370                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2371                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2372                     key.type != BTRFS_DEV_ITEM_KEY)
2373                         break;
2374
2375                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2376                                           struct btrfs_dev_item);
2377                 devid = btrfs_device_id(leaf, dev_item);
2378                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2379                                    BTRFS_UUID_SIZE);
2380                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2381                                    BTRFS_FSID_SIZE);
2382                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2383                 BUG_ON(!device); /* Logic error */
2384
2385                 if (device->fs_devices->seeding) {
2386                         btrfs_set_device_generation(leaf, dev_item,
2387                                                     device->generation);
2388                         btrfs_mark_buffer_dirty(leaf);
2389                 }
2390
2391                 path->slots[0]++;
2392                 goto next_slot;
2393         }
2394         ret = 0;
2395 error:
2396         btrfs_free_path(path);
2397         return ret;
2398 }
2399
2400 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2401 {
2402         struct btrfs_root *root = fs_info->dev_root;
2403         struct request_queue *q;
2404         struct btrfs_trans_handle *trans;
2405         struct btrfs_device *device;
2406         struct block_device *bdev;
2407         struct list_head *devices;
2408         struct super_block *sb = fs_info->sb;
2409         struct rcu_string *name;
2410         u64 tmp;
2411         int seeding_dev = 0;
2412         int ret = 0;
2413         bool unlocked = false;
2414
2415         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2416                 return -EROFS;
2417
2418         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2419                                   fs_info->bdev_holder);
2420         if (IS_ERR(bdev))
2421                 return PTR_ERR(bdev);
2422
2423         if (fs_info->fs_devices->seeding) {
2424                 seeding_dev = 1;
2425                 down_write(&sb->s_umount);
2426                 mutex_lock(&uuid_mutex);
2427         }
2428
2429         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2430
2431         devices = &fs_info->fs_devices->devices;
2432
2433         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2434         list_for_each_entry(device, devices, dev_list) {
2435                 if (device->bdev == bdev) {
2436                         ret = -EEXIST;
2437                         mutex_unlock(
2438                                 &fs_info->fs_devices->device_list_mutex);
2439                         goto error;
2440                 }
2441         }
2442         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2443
2444         device = btrfs_alloc_device(fs_info, NULL, NULL);
2445         if (IS_ERR(device)) {
2446                 /* we can safely leave the fs_devices entry around */
2447                 ret = PTR_ERR(device);
2448                 goto error;
2449         }
2450
2451         name = rcu_string_strdup(device_path, GFP_KERNEL);
2452         if (!name) {
2453                 ret = -ENOMEM;
2454                 goto error_free_device;
2455         }
2456         rcu_assign_pointer(device->name, name);
2457
2458         trans = btrfs_start_transaction(root, 0);
2459         if (IS_ERR(trans)) {
2460                 ret = PTR_ERR(trans);
2461                 goto error_free_device;
2462         }
2463
2464         q = bdev_get_queue(bdev);
2465         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2466         device->generation = trans->transid;
2467         device->io_width = fs_info->sectorsize;
2468         device->io_align = fs_info->sectorsize;
2469         device->sector_size = fs_info->sectorsize;
2470         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2471                                          fs_info->sectorsize);
2472         device->disk_total_bytes = device->total_bytes;
2473         device->commit_total_bytes = device->total_bytes;
2474         device->fs_info = fs_info;
2475         device->bdev = bdev;
2476         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2477         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2478         device->mode = FMODE_EXCL;
2479         device->dev_stats_valid = 1;
2480         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2481
2482         if (seeding_dev) {
2483                 sb->s_flags &= ~SB_RDONLY;
2484                 ret = btrfs_prepare_sprout(fs_info);
2485                 if (ret) {
2486                         btrfs_abort_transaction(trans, ret);
2487                         goto error_trans;
2488                 }
2489         }
2490
2491         device->fs_devices = fs_info->fs_devices;
2492
2493         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2494         mutex_lock(&fs_info->chunk_mutex);
2495         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2496         list_add(&device->dev_alloc_list,
2497                  &fs_info->fs_devices->alloc_list);
2498         fs_info->fs_devices->num_devices++;
2499         fs_info->fs_devices->open_devices++;
2500         fs_info->fs_devices->rw_devices++;
2501         fs_info->fs_devices->total_devices++;
2502         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2503
2504         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2505
2506         if (!blk_queue_nonrot(q))
2507                 fs_info->fs_devices->rotating = 1;
2508
2509         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2510         btrfs_set_super_total_bytes(fs_info->super_copy,
2511                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2512
2513         tmp = btrfs_super_num_devices(fs_info->super_copy);
2514         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2515
2516         /* add sysfs device entry */
2517         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2518
2519         /*
2520          * we've got more storage, clear any full flags on the space
2521          * infos
2522          */
2523         btrfs_clear_space_info_full(fs_info);
2524
2525         mutex_unlock(&fs_info->chunk_mutex);
2526         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2527
2528         if (seeding_dev) {
2529                 mutex_lock(&fs_info->chunk_mutex);
2530                 ret = init_first_rw_device(trans, fs_info);
2531                 mutex_unlock(&fs_info->chunk_mutex);
2532                 if (ret) {
2533                         btrfs_abort_transaction(trans, ret);
2534                         goto error_sysfs;
2535                 }
2536         }
2537
2538         ret = btrfs_add_dev_item(trans, fs_info, device);
2539         if (ret) {
2540                 btrfs_abort_transaction(trans, ret);
2541                 goto error_sysfs;
2542         }
2543
2544         if (seeding_dev) {
2545                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2546
2547                 ret = btrfs_finish_sprout(trans, fs_info);
2548                 if (ret) {
2549                         btrfs_abort_transaction(trans, ret);
2550                         goto error_sysfs;
2551                 }
2552
2553                 /* Sprouting would change fsid of the mounted root,
2554                  * so rename the fsid on the sysfs
2555                  */
2556                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2557                                                 fs_info->fsid);
2558                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2559                         btrfs_warn(fs_info,
2560                                    "sysfs: failed to create fsid for sprout");
2561         }
2562
2563         ret = btrfs_commit_transaction(trans);
2564
2565         if (seeding_dev) {
2566                 mutex_unlock(&uuid_mutex);
2567                 up_write(&sb->s_umount);
2568                 unlocked = true;
2569
2570                 if (ret) /* transaction commit */
2571                         return ret;
2572
2573                 ret = btrfs_relocate_sys_chunks(fs_info);
2574                 if (ret < 0)
2575                         btrfs_handle_fs_error(fs_info, ret,
2576                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2577                 trans = btrfs_attach_transaction(root);
2578                 if (IS_ERR(trans)) {
2579                         if (PTR_ERR(trans) == -ENOENT)
2580                                 return 0;
2581                         ret = PTR_ERR(trans);
2582                         trans = NULL;
2583                         goto error_sysfs;
2584                 }
2585                 ret = btrfs_commit_transaction(trans);
2586         }
2587
2588         /* Update ctime/mtime for libblkid */
2589         update_dev_time(device_path);
2590         return ret;
2591
2592 error_sysfs:
2593         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2594 error_trans:
2595         if (seeding_dev)
2596                 sb->s_flags |= SB_RDONLY;
2597         if (trans)
2598                 btrfs_end_transaction(trans);
2599 error_free_device:
2600         btrfs_free_device(device);
2601 error:
2602         blkdev_put(bdev, FMODE_EXCL);
2603         if (seeding_dev && !unlocked) {
2604                 mutex_unlock(&uuid_mutex);
2605                 up_write(&sb->s_umount);
2606         }
2607         return ret;
2608 }
2609
2610 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2611                                         struct btrfs_device *device)
2612 {
2613         int ret;
2614         struct btrfs_path *path;
2615         struct btrfs_root *root = device->fs_info->chunk_root;
2616         struct btrfs_dev_item *dev_item;
2617         struct extent_buffer *leaf;
2618         struct btrfs_key key;
2619
2620         path = btrfs_alloc_path();
2621         if (!path)
2622                 return -ENOMEM;
2623
2624         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2625         key.type = BTRFS_DEV_ITEM_KEY;
2626         key.offset = device->devid;
2627
2628         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2629         if (ret < 0)
2630                 goto out;
2631
2632         if (ret > 0) {
2633                 ret = -ENOENT;
2634                 goto out;
2635         }
2636
2637         leaf = path->nodes[0];
2638         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2639
2640         btrfs_set_device_id(leaf, dev_item, device->devid);
2641         btrfs_set_device_type(leaf, dev_item, device->type);
2642         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2643         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2644         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2645         btrfs_set_device_total_bytes(leaf, dev_item,
2646                                      btrfs_device_get_disk_total_bytes(device));
2647         btrfs_set_device_bytes_used(leaf, dev_item,
2648                                     btrfs_device_get_bytes_used(device));
2649         btrfs_mark_buffer_dirty(leaf);
2650
2651 out:
2652         btrfs_free_path(path);
2653         return ret;
2654 }
2655
2656 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2657                       struct btrfs_device *device, u64 new_size)
2658 {
2659         struct btrfs_fs_info *fs_info = device->fs_info;
2660         struct btrfs_super_block *super_copy = fs_info->super_copy;
2661         struct btrfs_fs_devices *fs_devices;
2662         u64 old_total;
2663         u64 diff;
2664
2665         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2666                 return -EACCES;
2667
2668         new_size = round_down(new_size, fs_info->sectorsize);
2669
2670         mutex_lock(&fs_info->chunk_mutex);
2671         old_total = btrfs_super_total_bytes(super_copy);
2672         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2673
2674         if (new_size <= device->total_bytes ||
2675             test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2676                 mutex_unlock(&fs_info->chunk_mutex);
2677                 return -EINVAL;
2678         }
2679
2680         fs_devices = fs_info->fs_devices;
2681
2682         btrfs_set_super_total_bytes(super_copy,
2683                         round_down(old_total + diff, fs_info->sectorsize));
2684         device->fs_devices->total_rw_bytes += diff;
2685
2686         btrfs_device_set_total_bytes(device, new_size);
2687         btrfs_device_set_disk_total_bytes(device, new_size);
2688         btrfs_clear_space_info_full(device->fs_info);
2689         if (list_empty(&device->resized_list))
2690                 list_add_tail(&device->resized_list,
2691                               &fs_devices->resized_devices);
2692         mutex_unlock(&fs_info->chunk_mutex);
2693
2694         return btrfs_update_device(trans, device);
2695 }
2696
2697 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2698                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2699 {
2700         struct btrfs_root *root = fs_info->chunk_root;
2701         int ret;
2702         struct btrfs_path *path;
2703         struct btrfs_key key;
2704
2705         path = btrfs_alloc_path();
2706         if (!path)
2707                 return -ENOMEM;
2708
2709         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2710         key.offset = chunk_offset;
2711         key.type = BTRFS_CHUNK_ITEM_KEY;
2712
2713         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2714         if (ret < 0)
2715                 goto out;
2716         else if (ret > 0) { /* Logic error or corruption */
2717                 btrfs_handle_fs_error(fs_info, -ENOENT,
2718                                       "Failed lookup while freeing chunk.");
2719                 ret = -ENOENT;
2720                 goto out;
2721         }
2722
2723         ret = btrfs_del_item(trans, root, path);
2724         if (ret < 0)
2725                 btrfs_handle_fs_error(fs_info, ret,
2726                                       "Failed to delete chunk item.");
2727 out:
2728         btrfs_free_path(path);
2729         return ret;
2730 }
2731
2732 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2733 {
2734         struct btrfs_super_block *super_copy = fs_info->super_copy;
2735         struct btrfs_disk_key *disk_key;
2736         struct btrfs_chunk *chunk;
2737         u8 *ptr;
2738         int ret = 0;
2739         u32 num_stripes;
2740         u32 array_size;
2741         u32 len = 0;
2742         u32 cur;
2743         struct btrfs_key key;
2744
2745         mutex_lock(&fs_info->chunk_mutex);
2746         array_size = btrfs_super_sys_array_size(super_copy);
2747
2748         ptr = super_copy->sys_chunk_array;
2749         cur = 0;
2750
2751         while (cur < array_size) {
2752                 disk_key = (struct btrfs_disk_key *)ptr;
2753                 btrfs_disk_key_to_cpu(&key, disk_key);
2754
2755                 len = sizeof(*disk_key);
2756
2757                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2758                         chunk = (struct btrfs_chunk *)(ptr + len);
2759                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2760                         len += btrfs_chunk_item_size(num_stripes);
2761                 } else {
2762                         ret = -EIO;
2763                         break;
2764                 }
2765                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2766                     key.offset == chunk_offset) {
2767                         memmove(ptr, ptr + len, array_size - (cur + len));
2768                         array_size -= len;
2769                         btrfs_set_super_sys_array_size(super_copy, array_size);
2770                 } else {
2771                         ptr += len;
2772                         cur += len;
2773                 }
2774         }
2775         mutex_unlock(&fs_info->chunk_mutex);
2776         return ret;
2777 }
2778
2779 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2780                                         u64 logical, u64 length)
2781 {
2782         struct extent_map_tree *em_tree;
2783         struct extent_map *em;
2784
2785         em_tree = &fs_info->mapping_tree.map_tree;
2786         read_lock(&em_tree->lock);
2787         em = lookup_extent_mapping(em_tree, logical, length);
2788         read_unlock(&em_tree->lock);
2789
2790         if (!em) {
2791                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2792                            logical, length);
2793                 return ERR_PTR(-EINVAL);
2794         }
2795
2796         if (em->start > logical || em->start + em->len < logical) {
2797                 btrfs_crit(fs_info,
2798                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2799                            logical, length, em->start, em->start + em->len);
2800                 free_extent_map(em);
2801                 return ERR_PTR(-EINVAL);
2802         }
2803
2804         /* callers are responsible for dropping em's ref. */
2805         return em;
2806 }
2807
2808 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2809                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2810 {
2811         struct extent_map *em;
2812         struct map_lookup *map;
2813         u64 dev_extent_len = 0;
2814         int i, ret = 0;
2815         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2816
2817         em = get_chunk_map(fs_info, chunk_offset, 1);
2818         if (IS_ERR(em)) {
2819                 /*
2820                  * This is a logic error, but we don't want to just rely on the
2821                  * user having built with ASSERT enabled, so if ASSERT doesn't
2822                  * do anything we still error out.
2823                  */
2824                 ASSERT(0);
2825                 return PTR_ERR(em);
2826         }
2827         map = em->map_lookup;
2828         mutex_lock(&fs_info->chunk_mutex);
2829         check_system_chunk(trans, fs_info, map->type);
2830         mutex_unlock(&fs_info->chunk_mutex);
2831
2832         /*
2833          * Take the device list mutex to prevent races with the final phase of
2834          * a device replace operation that replaces the device object associated
2835          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2836          */
2837         mutex_lock(&fs_devices->device_list_mutex);
2838         for (i = 0; i < map->num_stripes; i++) {
2839                 struct btrfs_device *device = map->stripes[i].dev;
2840                 ret = btrfs_free_dev_extent(trans, device,
2841                                             map->stripes[i].physical,
2842                                             &dev_extent_len);
2843                 if (ret) {
2844                         mutex_unlock(&fs_devices->device_list_mutex);
2845                         btrfs_abort_transaction(trans, ret);
2846                         goto out;
2847                 }
2848
2849                 if (device->bytes_used > 0) {
2850                         mutex_lock(&fs_info->chunk_mutex);
2851                         btrfs_device_set_bytes_used(device,
2852                                         device->bytes_used - dev_extent_len);
2853                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2854                         btrfs_clear_space_info_full(fs_info);
2855                         mutex_unlock(&fs_info->chunk_mutex);
2856                 }
2857
2858                 if (map->stripes[i].dev) {
2859                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2860                         if (ret) {
2861                                 mutex_unlock(&fs_devices->device_list_mutex);
2862                                 btrfs_abort_transaction(trans, ret);
2863                                 goto out;
2864                         }
2865                 }
2866         }
2867         mutex_unlock(&fs_devices->device_list_mutex);
2868
2869         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2870         if (ret) {
2871                 btrfs_abort_transaction(trans, ret);
2872                 goto out;
2873         }
2874
2875         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2876
2877         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2878                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2879                 if (ret) {
2880                         btrfs_abort_transaction(trans, ret);
2881                         goto out;
2882                 }
2883         }
2884
2885         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2886         if (ret) {
2887                 btrfs_abort_transaction(trans, ret);
2888                 goto out;
2889         }
2890
2891 out:
2892         /* once for us */
2893         free_extent_map(em);
2894         return ret;
2895 }
2896
2897 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2898 {
2899         struct btrfs_root *root = fs_info->chunk_root;
2900         struct btrfs_trans_handle *trans;
2901         int ret;
2902
2903         /*
2904          * Prevent races with automatic removal of unused block groups.
2905          * After we relocate and before we remove the chunk with offset
2906          * chunk_offset, automatic removal of the block group can kick in,
2907          * resulting in a failure when calling btrfs_remove_chunk() below.
2908          *
2909          * Make sure to acquire this mutex before doing a tree search (dev
2910          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2911          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2912          * we release the path used to search the chunk/dev tree and before
2913          * the current task acquires this mutex and calls us.
2914          */
2915         lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2916
2917         ret = btrfs_can_relocate(fs_info, chunk_offset);
2918         if (ret)
2919                 return -ENOSPC;
2920
2921         /* step one, relocate all the extents inside this chunk */
2922         btrfs_scrub_pause(fs_info);
2923         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2924         btrfs_scrub_continue(fs_info);
2925         if (ret)
2926                 return ret;
2927
2928         /*
2929          * We add the kobjects here (and after forcing data chunk creation)
2930          * since relocation is the only place we'll create chunks of a new
2931          * type at runtime.  The only place where we'll remove the last
2932          * chunk of a type is the call immediately below this one.  Even
2933          * so, we're protected against races with the cleaner thread since
2934          * we're covered by the delete_unused_bgs_mutex.
2935          */
2936         btrfs_add_raid_kobjects(fs_info);
2937
2938         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2939                                                      chunk_offset);
2940         if (IS_ERR(trans)) {
2941                 ret = PTR_ERR(trans);
2942                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2943                 return ret;
2944         }
2945
2946         /*
2947          * step two, delete the device extents and the
2948          * chunk tree entries
2949          */
2950         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2951         btrfs_end_transaction(trans);
2952         return ret;
2953 }
2954
2955 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2956 {
2957         struct btrfs_root *chunk_root = fs_info->chunk_root;
2958         struct btrfs_path *path;
2959         struct extent_buffer *leaf;
2960         struct btrfs_chunk *chunk;
2961         struct btrfs_key key;
2962         struct btrfs_key found_key;
2963         u64 chunk_type;
2964         bool retried = false;
2965         int failed = 0;
2966         int ret;
2967
2968         path = btrfs_alloc_path();
2969         if (!path)
2970                 return -ENOMEM;
2971
2972 again:
2973         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2974         key.offset = (u64)-1;
2975         key.type = BTRFS_CHUNK_ITEM_KEY;
2976
2977         while (1) {
2978                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2979                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2980                 if (ret < 0) {
2981                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2982                         goto error;
2983                 }
2984                 BUG_ON(ret == 0); /* Corruption */
2985
2986                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2987                                           key.type);
2988                 if (ret)
2989                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2990                 if (ret < 0)
2991                         goto error;
2992                 if (ret > 0)
2993                         break;
2994
2995                 leaf = path->nodes[0];
2996                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2997
2998                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2999                                        struct btrfs_chunk);
3000                 chunk_type = btrfs_chunk_type(leaf, chunk);
3001                 btrfs_release_path(path);
3002
3003                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3004                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3005                         if (ret == -ENOSPC)
3006                                 failed++;
3007                         else
3008                                 BUG_ON(ret);
3009                 }
3010                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3011
3012                 if (found_key.offset == 0)
3013                         break;
3014                 key.offset = found_key.offset - 1;
3015         }
3016         ret = 0;
3017         if (failed && !retried) {
3018                 failed = 0;
3019                 retried = true;
3020                 goto again;
3021         } else if (WARN_ON(failed && retried)) {
3022                 ret = -ENOSPC;
3023         }
3024 error:
3025         btrfs_free_path(path);
3026         return ret;
3027 }
3028
3029 /*
3030  * return 1 : allocate a data chunk successfully,
3031  * return <0: errors during allocating a data chunk,
3032  * return 0 : no need to allocate a data chunk.
3033  */
3034 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3035                                       u64 chunk_offset)
3036 {
3037         struct btrfs_block_group_cache *cache;
3038         u64 bytes_used;
3039         u64 chunk_type;
3040
3041         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3042         ASSERT(cache);
3043         chunk_type = cache->flags;
3044         btrfs_put_block_group(cache);
3045
3046         if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3047                 spin_lock(&fs_info->data_sinfo->lock);
3048                 bytes_used = fs_info->data_sinfo->bytes_used;
3049                 spin_unlock(&fs_info->data_sinfo->lock);
3050
3051                 if (!bytes_used) {
3052                         struct btrfs_trans_handle *trans;
3053                         int ret;
3054
3055                         trans = btrfs_join_transaction(fs_info->tree_root);
3056                         if (IS_ERR(trans))
3057                                 return PTR_ERR(trans);
3058
3059                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3060                                                       BTRFS_BLOCK_GROUP_DATA);
3061                         btrfs_end_transaction(trans);
3062                         if (ret < 0)
3063                                 return ret;
3064
3065                         btrfs_add_raid_kobjects(fs_info);
3066
3067                         return 1;
3068                 }
3069         }
3070         return 0;
3071 }
3072
3073 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3074                                struct btrfs_balance_control *bctl)
3075 {
3076         struct btrfs_root *root = fs_info->tree_root;
3077         struct btrfs_trans_handle *trans;
3078         struct btrfs_balance_item *item;
3079         struct btrfs_disk_balance_args disk_bargs;
3080         struct btrfs_path *path;
3081         struct extent_buffer *leaf;
3082         struct btrfs_key key;
3083         int ret, err;
3084
3085         path = btrfs_alloc_path();
3086         if (!path)
3087                 return -ENOMEM;
3088
3089         trans = btrfs_start_transaction(root, 0);
3090         if (IS_ERR(trans)) {
3091                 btrfs_free_path(path);
3092                 return PTR_ERR(trans);
3093         }
3094
3095         key.objectid = BTRFS_BALANCE_OBJECTID;
3096         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3097         key.offset = 0;
3098
3099         ret = btrfs_insert_empty_item(trans, root, path, &key,
3100                                       sizeof(*item));
3101         if (ret)
3102                 goto out;
3103
3104         leaf = path->nodes[0];
3105         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3106
3107         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3108
3109         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3110         btrfs_set_balance_data(leaf, item, &disk_bargs);
3111         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3112         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3113         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3114         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3115
3116         btrfs_set_balance_flags(leaf, item, bctl->flags);
3117
3118         btrfs_mark_buffer_dirty(leaf);
3119 out:
3120         btrfs_free_path(path);
3121         err = btrfs_commit_transaction(trans);
3122         if (err && !ret)
3123                 ret = err;
3124         return ret;
3125 }
3126
3127 static int del_balance_item(struct btrfs_fs_info *fs_info)
3128 {
3129         struct btrfs_root *root = fs_info->tree_root;
3130         struct btrfs_trans_handle *trans;
3131         struct btrfs_path *path;
3132         struct btrfs_key key;
3133         int ret, err;
3134
3135         path = btrfs_alloc_path();
3136         if (!path)
3137                 return -ENOMEM;
3138
3139         trans = btrfs_start_transaction(root, 0);
3140         if (IS_ERR(trans)) {
3141                 btrfs_free_path(path);
3142                 return PTR_ERR(trans);
3143         }
3144
3145         key.objectid = BTRFS_BALANCE_OBJECTID;
3146         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3147         key.offset = 0;
3148
3149         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3150         if (ret < 0)
3151                 goto out;
3152         if (ret > 0) {
3153                 ret = -ENOENT;
3154                 goto out;
3155         }
3156
3157         ret = btrfs_del_item(trans, root, path);
3158 out:
3159         btrfs_free_path(path);
3160         err = btrfs_commit_transaction(trans);
3161         if (err && !ret)
3162                 ret = err;
3163         return ret;
3164 }
3165
3166 /*
3167  * This is a heuristic used to reduce the number of chunks balanced on
3168  * resume after balance was interrupted.
3169  */
3170 static void update_balance_args(struct btrfs_balance_control *bctl)
3171 {
3172         /*
3173          * Turn on soft mode for chunk types that were being converted.
3174          */
3175         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3176                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3177         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3178                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3179         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3180                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3181
3182         /*
3183          * Turn on usage filter if is not already used.  The idea is
3184          * that chunks that we have already balanced should be
3185          * reasonably full.  Don't do it for chunks that are being
3186          * converted - that will keep us from relocating unconverted
3187          * (albeit full) chunks.
3188          */
3189         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3190             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3191             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3192                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3193                 bctl->data.usage = 90;
3194         }
3195         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3196             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3197             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3198                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3199                 bctl->sys.usage = 90;
3200         }
3201         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3202             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3203             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3204                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3205                 bctl->meta.usage = 90;
3206         }
3207 }
3208
3209 /*
3210  * Clear the balance status in fs_info and delete the balance item from disk.
3211  */
3212 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3213 {
3214         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3215         int ret;
3216
3217         BUG_ON(!fs_info->balance_ctl);
3218
3219         spin_lock(&fs_info->balance_lock);
3220         fs_info->balance_ctl = NULL;
3221         spin_unlock(&fs_info->balance_lock);
3222
3223         kfree(bctl);
3224         ret = del_balance_item(fs_info);
3225         if (ret)
3226                 btrfs_handle_fs_error(fs_info, ret, NULL);
3227 }
3228
3229 /*
3230  * Balance filters.  Return 1 if chunk should be filtered out
3231  * (should not be balanced).
3232  */
3233 static int chunk_profiles_filter(u64 chunk_type,
3234                                  struct btrfs_balance_args *bargs)
3235 {
3236         chunk_type = chunk_to_extended(chunk_type) &
3237                                 BTRFS_EXTENDED_PROFILE_MASK;
3238
3239         if (bargs->profiles & chunk_type)
3240                 return 0;
3241
3242         return 1;
3243 }