bb0addce755865dbb3fb3726998e2792c90205ed
[sfrench/cifs-2.6.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
31 #include "ctree.h"
32 #include "extent_map.h"
33 #include "disk-io.h"
34 #include "transaction.h"
35 #include "print-tree.h"
36 #include "volumes.h"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43 #include "sysfs.h"
44
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46         [BTRFS_RAID_RAID10] = {
47                 .sub_stripes    = 2,
48                 .dev_stripes    = 1,
49                 .devs_max       = 0,    /* 0 == as many as possible */
50                 .devs_min       = 4,
51                 .tolerated_failures = 1,
52                 .devs_increment = 2,
53                 .ncopies        = 2,
54         },
55         [BTRFS_RAID_RAID1] = {
56                 .sub_stripes    = 1,
57                 .dev_stripes    = 1,
58                 .devs_max       = 2,
59                 .devs_min       = 2,
60                 .tolerated_failures = 1,
61                 .devs_increment = 2,
62                 .ncopies        = 2,
63         },
64         [BTRFS_RAID_DUP] = {
65                 .sub_stripes    = 1,
66                 .dev_stripes    = 2,
67                 .devs_max       = 1,
68                 .devs_min       = 1,
69                 .tolerated_failures = 0,
70                 .devs_increment = 1,
71                 .ncopies        = 2,
72         },
73         [BTRFS_RAID_RAID0] = {
74                 .sub_stripes    = 1,
75                 .dev_stripes    = 1,
76                 .devs_max       = 0,
77                 .devs_min       = 2,
78                 .tolerated_failures = 0,
79                 .devs_increment = 1,
80                 .ncopies        = 1,
81         },
82         [BTRFS_RAID_SINGLE] = {
83                 .sub_stripes    = 1,
84                 .dev_stripes    = 1,
85                 .devs_max       = 1,
86                 .devs_min       = 1,
87                 .tolerated_failures = 0,
88                 .devs_increment = 1,
89                 .ncopies        = 1,
90         },
91         [BTRFS_RAID_RAID5] = {
92                 .sub_stripes    = 1,
93                 .dev_stripes    = 1,
94                 .devs_max       = 0,
95                 .devs_min       = 2,
96                 .tolerated_failures = 1,
97                 .devs_increment = 1,
98                 .ncopies        = 2,
99         },
100         [BTRFS_RAID_RAID6] = {
101                 .sub_stripes    = 1,
102                 .dev_stripes    = 1,
103                 .devs_max       = 0,
104                 .devs_min       = 3,
105                 .tolerated_failures = 2,
106                 .devs_increment = 1,
107                 .ncopies        = 3,
108         },
109 };
110
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
116         [BTRFS_RAID_SINGLE] = 0,
117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
119 };
120
121 /*
122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123  * condition is not met. Zero means there's no corresponding
124  * BTRFS_ERROR_DEV_*_NOT_MET value.
125  */
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129         [BTRFS_RAID_DUP]    = 0,
130         [BTRFS_RAID_RAID0]  = 0,
131         [BTRFS_RAID_SINGLE] = 0,
132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134 };
135
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137                                 struct btrfs_root *root,
138                                 struct btrfs_device *device);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143
144 DEFINE_MUTEX(uuid_mutex);
145 static LIST_HEAD(fs_uuids);
146 struct list_head *btrfs_get_fs_uuids(void)
147 {
148         return &fs_uuids;
149 }
150
151 static struct btrfs_fs_devices *__alloc_fs_devices(void)
152 {
153         struct btrfs_fs_devices *fs_devs;
154
155         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
156         if (!fs_devs)
157                 return ERR_PTR(-ENOMEM);
158
159         mutex_init(&fs_devs->device_list_mutex);
160
161         INIT_LIST_HEAD(&fs_devs->devices);
162         INIT_LIST_HEAD(&fs_devs->resized_devices);
163         INIT_LIST_HEAD(&fs_devs->alloc_list);
164         INIT_LIST_HEAD(&fs_devs->list);
165
166         return fs_devs;
167 }
168
169 /**
170  * alloc_fs_devices - allocate struct btrfs_fs_devices
171  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
172  *              generated.
173  *
174  * Return: a pointer to a new &struct btrfs_fs_devices on success;
175  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
176  * can be destroyed with kfree() right away.
177  */
178 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
179 {
180         struct btrfs_fs_devices *fs_devs;
181
182         fs_devs = __alloc_fs_devices();
183         if (IS_ERR(fs_devs))
184                 return fs_devs;
185
186         if (fsid)
187                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
188         else
189                 generate_random_uuid(fs_devs->fsid);
190
191         return fs_devs;
192 }
193
194 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
195 {
196         struct btrfs_device *device;
197         WARN_ON(fs_devices->opened);
198         while (!list_empty(&fs_devices->devices)) {
199                 device = list_entry(fs_devices->devices.next,
200                                     struct btrfs_device, dev_list);
201                 list_del(&device->dev_list);
202                 rcu_string_free(device->name);
203                 kfree(device);
204         }
205         kfree(fs_devices);
206 }
207
208 static void btrfs_kobject_uevent(struct block_device *bdev,
209                                  enum kobject_action action)
210 {
211         int ret;
212
213         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
214         if (ret)
215                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
216                         action,
217                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218                         &disk_to_dev(bdev->bd_disk)->kobj);
219 }
220
221 void btrfs_cleanup_fs_uuids(void)
222 {
223         struct btrfs_fs_devices *fs_devices;
224
225         while (!list_empty(&fs_uuids)) {
226                 fs_devices = list_entry(fs_uuids.next,
227                                         struct btrfs_fs_devices, list);
228                 list_del(&fs_devices->list);
229                 free_fs_devices(fs_devices);
230         }
231 }
232
233 static struct btrfs_device *__alloc_device(void)
234 {
235         struct btrfs_device *dev;
236
237         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
238         if (!dev)
239                 return ERR_PTR(-ENOMEM);
240
241         INIT_LIST_HEAD(&dev->dev_list);
242         INIT_LIST_HEAD(&dev->dev_alloc_list);
243         INIT_LIST_HEAD(&dev->resized_list);
244
245         spin_lock_init(&dev->io_lock);
246
247         spin_lock_init(&dev->reada_lock);
248         atomic_set(&dev->reada_in_flight, 0);
249         atomic_set(&dev->dev_stats_ccnt, 0);
250         btrfs_device_data_ordered_init(dev);
251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253
254         return dev;
255 }
256
257 static noinline struct btrfs_device *__find_device(struct list_head *head,
258                                                    u64 devid, u8 *uuid)
259 {
260         struct btrfs_device *dev;
261
262         list_for_each_entry(dev, head, dev_list) {
263                 if (dev->devid == devid &&
264                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
265                         return dev;
266                 }
267         }
268         return NULL;
269 }
270
271 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
272 {
273         struct btrfs_fs_devices *fs_devices;
274
275         list_for_each_entry(fs_devices, &fs_uuids, list) {
276                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
277                         return fs_devices;
278         }
279         return NULL;
280 }
281
282 static int
283 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284                       int flush, struct block_device **bdev,
285                       struct buffer_head **bh)
286 {
287         int ret;
288
289         *bdev = blkdev_get_by_path(device_path, flags, holder);
290
291         if (IS_ERR(*bdev)) {
292                 ret = PTR_ERR(*bdev);
293                 goto error;
294         }
295
296         if (flush)
297                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298         ret = set_blocksize(*bdev, 4096);
299         if (ret) {
300                 blkdev_put(*bdev, flags);
301                 goto error;
302         }
303         invalidate_bdev(*bdev);
304         *bh = btrfs_read_dev_super(*bdev);
305         if (IS_ERR(*bh)) {
306                 ret = PTR_ERR(*bh);
307                 blkdev_put(*bdev, flags);
308                 goto error;
309         }
310
311         return 0;
312
313 error:
314         *bdev = NULL;
315         *bh = NULL;
316         return ret;
317 }
318
319 static void requeue_list(struct btrfs_pending_bios *pending_bios,
320                         struct bio *head, struct bio *tail)
321 {
322
323         struct bio *old_head;
324
325         old_head = pending_bios->head;
326         pending_bios->head = head;
327         if (pending_bios->tail)
328                 tail->bi_next = old_head;
329         else
330                 pending_bios->tail = tail;
331 }
332
333 /*
334  * we try to collect pending bios for a device so we don't get a large
335  * number of procs sending bios down to the same device.  This greatly
336  * improves the schedulers ability to collect and merge the bios.
337  *
338  * But, it also turns into a long list of bios to process and that is sure
339  * to eventually make the worker thread block.  The solution here is to
340  * make some progress and then put this work struct back at the end of
341  * the list if the block device is congested.  This way, multiple devices
342  * can make progress from a single worker thread.
343  */
344 static noinline void run_scheduled_bios(struct btrfs_device *device)
345 {
346         struct bio *pending;
347         struct backing_dev_info *bdi;
348         struct btrfs_fs_info *fs_info;
349         struct btrfs_pending_bios *pending_bios;
350         struct bio *tail;
351         struct bio *cur;
352         int again = 0;
353         unsigned long num_run;
354         unsigned long batch_run = 0;
355         unsigned long limit;
356         unsigned long last_waited = 0;
357         int force_reg = 0;
358         int sync_pending = 0;
359         struct blk_plug plug;
360
361         /*
362          * this function runs all the bios we've collected for
363          * a particular device.  We don't want to wander off to
364          * another device without first sending all of these down.
365          * So, setup a plug here and finish it off before we return
366          */
367         blk_start_plug(&plug);
368
369         bdi = blk_get_backing_dev_info(device->bdev);
370         fs_info = device->dev_root->fs_info;
371         limit = btrfs_async_submit_limit(fs_info);
372         limit = limit * 2 / 3;
373
374 loop:
375         spin_lock(&device->io_lock);
376
377 loop_lock:
378         num_run = 0;
379
380         /* take all the bios off the list at once and process them
381          * later on (without the lock held).  But, remember the
382          * tail and other pointers so the bios can be properly reinserted
383          * into the list if we hit congestion
384          */
385         if (!force_reg && device->pending_sync_bios.head) {
386                 pending_bios = &device->pending_sync_bios;
387                 force_reg = 1;
388         } else {
389                 pending_bios = &device->pending_bios;
390                 force_reg = 0;
391         }
392
393         pending = pending_bios->head;
394         tail = pending_bios->tail;
395         WARN_ON(pending && !tail);
396
397         /*
398          * if pending was null this time around, no bios need processing
399          * at all and we can stop.  Otherwise it'll loop back up again
400          * and do an additional check so no bios are missed.
401          *
402          * device->running_pending is used to synchronize with the
403          * schedule_bio code.
404          */
405         if (device->pending_sync_bios.head == NULL &&
406             device->pending_bios.head == NULL) {
407                 again = 0;
408                 device->running_pending = 0;
409         } else {
410                 again = 1;
411                 device->running_pending = 1;
412         }
413
414         pending_bios->head = NULL;
415         pending_bios->tail = NULL;
416
417         spin_unlock(&device->io_lock);
418
419         while (pending) {
420
421                 rmb();
422                 /* we want to work on both lists, but do more bios on the
423                  * sync list than the regular list
424                  */
425                 if ((num_run > 32 &&
426                     pending_bios != &device->pending_sync_bios &&
427                     device->pending_sync_bios.head) ||
428                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
429                     device->pending_bios.head)) {
430                         spin_lock(&device->io_lock);
431                         requeue_list(pending_bios, pending, tail);
432                         goto loop_lock;
433                 }
434
435                 cur = pending;
436                 pending = pending->bi_next;
437                 cur->bi_next = NULL;
438
439                 /*
440                  * atomic_dec_return implies a barrier for waitqueue_active
441                  */
442                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
443                     waitqueue_active(&fs_info->async_submit_wait))
444                         wake_up(&fs_info->async_submit_wait);
445
446                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
447
448                 /*
449                  * if we're doing the sync list, record that our
450                  * plug has some sync requests on it
451                  *
452                  * If we're doing the regular list and there are
453                  * sync requests sitting around, unplug before
454                  * we add more
455                  */
456                 if (pending_bios == &device->pending_sync_bios) {
457                         sync_pending = 1;
458                 } else if (sync_pending) {
459                         blk_finish_plug(&plug);
460                         blk_start_plug(&plug);
461                         sync_pending = 0;
462                 }
463
464                 btrfsic_submit_bio(cur);
465                 num_run++;
466                 batch_run++;
467
468                 cond_resched();
469
470                 /*
471                  * we made progress, there is more work to do and the bdi
472                  * is now congested.  Back off and let other work structs
473                  * run instead
474                  */
475                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
476                     fs_info->fs_devices->open_devices > 1) {
477                         struct io_context *ioc;
478
479                         ioc = current->io_context;
480
481                         /*
482                          * the main goal here is that we don't want to
483                          * block if we're going to be able to submit
484                          * more requests without blocking.
485                          *
486                          * This code does two great things, it pokes into
487                          * the elevator code from a filesystem _and_
488                          * it makes assumptions about how batching works.
489                          */
490                         if (ioc && ioc->nr_batch_requests > 0 &&
491                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
492                             (last_waited == 0 ||
493                              ioc->last_waited == last_waited)) {
494                                 /*
495                                  * we want to go through our batch of
496                                  * requests and stop.  So, we copy out
497                                  * the ioc->last_waited time and test
498                                  * against it before looping
499                                  */
500                                 last_waited = ioc->last_waited;
501                                 cond_resched();
502                                 continue;
503                         }
504                         spin_lock(&device->io_lock);
505                         requeue_list(pending_bios, pending, tail);
506                         device->running_pending = 1;
507
508                         spin_unlock(&device->io_lock);
509                         btrfs_queue_work(fs_info->submit_workers,
510                                          &device->work);
511                         goto done;
512                 }
513                 /* unplug every 64 requests just for good measure */
514                 if (batch_run % 64 == 0) {
515                         blk_finish_plug(&plug);
516                         blk_start_plug(&plug);
517                         sync_pending = 0;
518                 }
519         }
520
521         cond_resched();
522         if (again)
523                 goto loop;
524
525         spin_lock(&device->io_lock);
526         if (device->pending_bios.head || device->pending_sync_bios.head)
527                 goto loop_lock;
528         spin_unlock(&device->io_lock);
529
530 done:
531         blk_finish_plug(&plug);
532 }
533
534 static void pending_bios_fn(struct btrfs_work *work)
535 {
536         struct btrfs_device *device;
537
538         device = container_of(work, struct btrfs_device, work);
539         run_scheduled_bios(device);
540 }
541
542
543 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
544 {
545         struct btrfs_fs_devices *fs_devs;
546         struct btrfs_device *dev;
547
548         if (!cur_dev->name)
549                 return;
550
551         list_for_each_entry(fs_devs, &fs_uuids, list) {
552                 int del = 1;
553
554                 if (fs_devs->opened)
555                         continue;
556                 if (fs_devs->seeding)
557                         continue;
558
559                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
560
561                         if (dev == cur_dev)
562                                 continue;
563                         if (!dev->name)
564                                 continue;
565
566                         /*
567                          * Todo: This won't be enough. What if the same device
568                          * comes back (with new uuid and) with its mapper path?
569                          * But for now, this does help as mostly an admin will
570                          * either use mapper or non mapper path throughout.
571                          */
572                         rcu_read_lock();
573                         del = strcmp(rcu_str_deref(dev->name),
574                                                 rcu_str_deref(cur_dev->name));
575                         rcu_read_unlock();
576                         if (!del)
577                                 break;
578                 }
579
580                 if (!del) {
581                         /* delete the stale device */
582                         if (fs_devs->num_devices == 1) {
583                                 btrfs_sysfs_remove_fsid(fs_devs);
584                                 list_del(&fs_devs->list);
585                                 free_fs_devices(fs_devs);
586                         } else {
587                                 fs_devs->num_devices--;
588                                 list_del(&dev->dev_list);
589                                 rcu_string_free(dev->name);
590                                 kfree(dev);
591                         }
592                         break;
593                 }
594         }
595 }
596
597 /*
598  * Add new device to list of registered devices
599  *
600  * Returns:
601  * 1   - first time device is seen
602  * 0   - device already known
603  * < 0 - error
604  */
605 static noinline int device_list_add(const char *path,
606                            struct btrfs_super_block *disk_super,
607                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
608 {
609         struct btrfs_device *device;
610         struct btrfs_fs_devices *fs_devices;
611         struct rcu_string *name;
612         int ret = 0;
613         u64 found_transid = btrfs_super_generation(disk_super);
614
615         fs_devices = find_fsid(disk_super->fsid);
616         if (!fs_devices) {
617                 fs_devices = alloc_fs_devices(disk_super->fsid);
618                 if (IS_ERR(fs_devices))
619                         return PTR_ERR(fs_devices);
620
621                 list_add(&fs_devices->list, &fs_uuids);
622
623                 device = NULL;
624         } else {
625                 device = __find_device(&fs_devices->devices, devid,
626                                        disk_super->dev_item.uuid);
627         }
628
629         if (!device) {
630                 if (fs_devices->opened)
631                         return -EBUSY;
632
633                 device = btrfs_alloc_device(NULL, &devid,
634                                             disk_super->dev_item.uuid);
635                 if (IS_ERR(device)) {
636                         /* we can safely leave the fs_devices entry around */
637                         return PTR_ERR(device);
638                 }
639
640                 name = rcu_string_strdup(path, GFP_NOFS);
641                 if (!name) {
642                         kfree(device);
643                         return -ENOMEM;
644                 }
645                 rcu_assign_pointer(device->name, name);
646
647                 mutex_lock(&fs_devices->device_list_mutex);
648                 list_add_rcu(&device->dev_list, &fs_devices->devices);
649                 fs_devices->num_devices++;
650                 mutex_unlock(&fs_devices->device_list_mutex);
651
652                 ret = 1;
653                 device->fs_devices = fs_devices;
654         } else if (!device->name || strcmp(device->name->str, path)) {
655                 /*
656                  * When FS is already mounted.
657                  * 1. If you are here and if the device->name is NULL that
658                  *    means this device was missing at time of FS mount.
659                  * 2. If you are here and if the device->name is different
660                  *    from 'path' that means either
661                  *      a. The same device disappeared and reappeared with
662                  *         different name. or
663                  *      b. The missing-disk-which-was-replaced, has
664                  *         reappeared now.
665                  *
666                  * We must allow 1 and 2a above. But 2b would be a spurious
667                  * and unintentional.
668                  *
669                  * Further in case of 1 and 2a above, the disk at 'path'
670                  * would have missed some transaction when it was away and
671                  * in case of 2a the stale bdev has to be updated as well.
672                  * 2b must not be allowed at all time.
673                  */
674
675                 /*
676                  * For now, we do allow update to btrfs_fs_device through the
677                  * btrfs dev scan cli after FS has been mounted.  We're still
678                  * tracking a problem where systems fail mount by subvolume id
679                  * when we reject replacement on a mounted FS.
680                  */
681                 if (!fs_devices->opened && found_transid < device->generation) {
682                         /*
683                          * That is if the FS is _not_ mounted and if you
684                          * are here, that means there is more than one
685                          * disk with same uuid and devid.We keep the one
686                          * with larger generation number or the last-in if
687                          * generation are equal.
688                          */
689                         return -EEXIST;
690                 }
691
692                 name = rcu_string_strdup(path, GFP_NOFS);
693                 if (!name)
694                         return -ENOMEM;
695                 rcu_string_free(device->name);
696                 rcu_assign_pointer(device->name, name);
697                 if (device->missing) {
698                         fs_devices->missing_devices--;
699                         device->missing = 0;
700                 }
701         }
702
703         /*
704          * Unmount does not free the btrfs_device struct but would zero
705          * generation along with most of the other members. So just update
706          * it back. We need it to pick the disk with largest generation
707          * (as above).
708          */
709         if (!fs_devices->opened)
710                 device->generation = found_transid;
711
712         /*
713          * if there is new btrfs on an already registered device,
714          * then remove the stale device entry.
715          */
716         if (ret > 0)
717                 btrfs_free_stale_device(device);
718
719         *fs_devices_ret = fs_devices;
720
721         return ret;
722 }
723
724 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
725 {
726         struct btrfs_fs_devices *fs_devices;
727         struct btrfs_device *device;
728         struct btrfs_device *orig_dev;
729
730         fs_devices = alloc_fs_devices(orig->fsid);
731         if (IS_ERR(fs_devices))
732                 return fs_devices;
733
734         mutex_lock(&orig->device_list_mutex);
735         fs_devices->total_devices = orig->total_devices;
736
737         /* We have held the volume lock, it is safe to get the devices. */
738         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
739                 struct rcu_string *name;
740
741                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
742                                             orig_dev->uuid);
743                 if (IS_ERR(device))
744                         goto error;
745
746                 /*
747                  * This is ok to do without rcu read locked because we hold the
748                  * uuid mutex so nothing we touch in here is going to disappear.
749                  */
750                 if (orig_dev->name) {
751                         name = rcu_string_strdup(orig_dev->name->str,
752                                         GFP_KERNEL);
753                         if (!name) {
754                                 kfree(device);
755                                 goto error;
756                         }
757                         rcu_assign_pointer(device->name, name);
758                 }
759
760                 list_add(&device->dev_list, &fs_devices->devices);
761                 device->fs_devices = fs_devices;
762                 fs_devices->num_devices++;
763         }
764         mutex_unlock(&orig->device_list_mutex);
765         return fs_devices;
766 error:
767         mutex_unlock(&orig->device_list_mutex);
768         free_fs_devices(fs_devices);
769         return ERR_PTR(-ENOMEM);
770 }
771
772 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
773 {
774         struct btrfs_device *device, *next;
775         struct btrfs_device *latest_dev = NULL;
776
777         mutex_lock(&uuid_mutex);
778 again:
779         /* This is the initialized path, it is safe to release the devices. */
780         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
781                 if (device->in_fs_metadata) {
782                         if (!device->is_tgtdev_for_dev_replace &&
783                             (!latest_dev ||
784                              device->generation > latest_dev->generation)) {
785                                 latest_dev = device;
786                         }
787                         continue;
788                 }
789
790                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
791                         /*
792                          * In the first step, keep the device which has
793                          * the correct fsid and the devid that is used
794                          * for the dev_replace procedure.
795                          * In the second step, the dev_replace state is
796                          * read from the device tree and it is known
797                          * whether the procedure is really active or
798                          * not, which means whether this device is
799                          * used or whether it should be removed.
800                          */
801                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
802                                 continue;
803                         }
804                 }
805                 if (device->bdev) {
806                         blkdev_put(device->bdev, device->mode);
807                         device->bdev = NULL;
808                         fs_devices->open_devices--;
809                 }
810                 if (device->writeable) {
811                         list_del_init(&device->dev_alloc_list);
812                         device->writeable = 0;
813                         if (!device->is_tgtdev_for_dev_replace)
814                                 fs_devices->rw_devices--;
815                 }
816                 list_del_init(&device->dev_list);
817                 fs_devices->num_devices--;
818                 rcu_string_free(device->name);
819                 kfree(device);
820         }
821
822         if (fs_devices->seed) {
823                 fs_devices = fs_devices->seed;
824                 goto again;
825         }
826
827         fs_devices->latest_bdev = latest_dev->bdev;
828
829         mutex_unlock(&uuid_mutex);
830 }
831
832 static void __free_device(struct work_struct *work)
833 {
834         struct btrfs_device *device;
835
836         device = container_of(work, struct btrfs_device, rcu_work);
837
838         if (device->bdev)
839                 blkdev_put(device->bdev, device->mode);
840
841         rcu_string_free(device->name);
842         kfree(device);
843 }
844
845 static void free_device(struct rcu_head *head)
846 {
847         struct btrfs_device *device;
848
849         device = container_of(head, struct btrfs_device, rcu);
850
851         INIT_WORK(&device->rcu_work, __free_device);
852         schedule_work(&device->rcu_work);
853 }
854
855 static void btrfs_close_one_device(struct btrfs_device *device)
856 {
857         struct btrfs_fs_devices *fs_devices = device->fs_devices;
858         struct btrfs_device *new_device;
859         struct rcu_string *name;
860
861         if (device->bdev)
862                 fs_devices->open_devices--;
863
864         if (device->writeable &&
865             device->devid != BTRFS_DEV_REPLACE_DEVID) {
866                 list_del_init(&device->dev_alloc_list);
867                 fs_devices->rw_devices--;
868         }
869
870         if (device->missing)
871                 fs_devices->missing_devices--;
872
873         if (device->bdev && device->writeable) {
874                 sync_blockdev(device->bdev);
875                 invalidate_bdev(device->bdev);
876         }
877
878         new_device = btrfs_alloc_device(NULL, &device->devid,
879                                         device->uuid);
880         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
881
882         /* Safe because we are under uuid_mutex */
883         if (device->name) {
884                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
885                 BUG_ON(!name); /* -ENOMEM */
886                 rcu_assign_pointer(new_device->name, name);
887         }
888
889         list_replace_rcu(&device->dev_list, &new_device->dev_list);
890         new_device->fs_devices = device->fs_devices;
891
892         call_rcu(&device->rcu, free_device);
893 }
894
895 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
896 {
897         struct btrfs_device *device, *tmp;
898
899         if (--fs_devices->opened > 0)
900                 return 0;
901
902         mutex_lock(&fs_devices->device_list_mutex);
903         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
904                 btrfs_close_one_device(device);
905         }
906         mutex_unlock(&fs_devices->device_list_mutex);
907
908         WARN_ON(fs_devices->open_devices);
909         WARN_ON(fs_devices->rw_devices);
910         fs_devices->opened = 0;
911         fs_devices->seeding = 0;
912
913         return 0;
914 }
915
916 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
917 {
918         struct btrfs_fs_devices *seed_devices = NULL;
919         int ret;
920
921         mutex_lock(&uuid_mutex);
922         ret = __btrfs_close_devices(fs_devices);
923         if (!fs_devices->opened) {
924                 seed_devices = fs_devices->seed;
925                 fs_devices->seed = NULL;
926         }
927         mutex_unlock(&uuid_mutex);
928
929         while (seed_devices) {
930                 fs_devices = seed_devices;
931                 seed_devices = fs_devices->seed;
932                 __btrfs_close_devices(fs_devices);
933                 free_fs_devices(fs_devices);
934         }
935         /*
936          * Wait for rcu kworkers under __btrfs_close_devices
937          * to finish all blkdev_puts so device is really
938          * free when umount is done.
939          */
940         rcu_barrier();
941         return ret;
942 }
943
944 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
945                                 fmode_t flags, void *holder)
946 {
947         struct request_queue *q;
948         struct block_device *bdev;
949         struct list_head *head = &fs_devices->devices;
950         struct btrfs_device *device;
951         struct btrfs_device *latest_dev = NULL;
952         struct buffer_head *bh;
953         struct btrfs_super_block *disk_super;
954         u64 devid;
955         int seeding = 1;
956         int ret = 0;
957
958         flags |= FMODE_EXCL;
959
960         list_for_each_entry(device, head, dev_list) {
961                 if (device->bdev)
962                         continue;
963                 if (!device->name)
964                         continue;
965
966                 /* Just open everything we can; ignore failures here */
967                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
968                                             &bdev, &bh))
969                         continue;
970
971                 disk_super = (struct btrfs_super_block *)bh->b_data;
972                 devid = btrfs_stack_device_id(&disk_super->dev_item);
973                 if (devid != device->devid)
974                         goto error_brelse;
975
976                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
977                            BTRFS_UUID_SIZE))
978                         goto error_brelse;
979
980                 device->generation = btrfs_super_generation(disk_super);
981                 if (!latest_dev ||
982                     device->generation > latest_dev->generation)
983                         latest_dev = device;
984
985                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
986                         device->writeable = 0;
987                 } else {
988                         device->writeable = !bdev_read_only(bdev);
989                         seeding = 0;
990                 }
991
992                 q = bdev_get_queue(bdev);
993                 if (blk_queue_discard(q))
994                         device->can_discard = 1;
995
996                 device->bdev = bdev;
997                 device->in_fs_metadata = 0;
998                 device->mode = flags;
999
1000                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1001                         fs_devices->rotating = 1;
1002
1003                 fs_devices->open_devices++;
1004                 if (device->writeable &&
1005                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1006                         fs_devices->rw_devices++;
1007                         list_add(&device->dev_alloc_list,
1008                                  &fs_devices->alloc_list);
1009                 }
1010                 brelse(bh);
1011                 continue;
1012
1013 error_brelse:
1014                 brelse(bh);
1015                 blkdev_put(bdev, flags);
1016                 continue;
1017         }
1018         if (fs_devices->open_devices == 0) {
1019                 ret = -EINVAL;
1020                 goto out;
1021         }
1022         fs_devices->seeding = seeding;
1023         fs_devices->opened = 1;
1024         fs_devices->latest_bdev = latest_dev->bdev;
1025         fs_devices->total_rw_bytes = 0;
1026 out:
1027         return ret;
1028 }
1029
1030 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1031                        fmode_t flags, void *holder)
1032 {
1033         int ret;
1034
1035         mutex_lock(&uuid_mutex);
1036         if (fs_devices->opened) {
1037                 fs_devices->opened++;
1038                 ret = 0;
1039         } else {
1040                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1041         }
1042         mutex_unlock(&uuid_mutex);
1043         return ret;
1044 }
1045
1046 void btrfs_release_disk_super(struct page *page)
1047 {
1048         kunmap(page);
1049         put_page(page);
1050 }
1051
1052 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1053                 struct page **page, struct btrfs_super_block **disk_super)
1054 {
1055         void *p;
1056         pgoff_t index;
1057
1058         /* make sure our super fits in the device */
1059         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1060                 return 1;
1061
1062         /* make sure our super fits in the page */
1063         if (sizeof(**disk_super) > PAGE_SIZE)
1064                 return 1;
1065
1066         /* make sure our super doesn't straddle pages on disk */
1067         index = bytenr >> PAGE_SHIFT;
1068         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1069                 return 1;
1070
1071         /* pull in the page with our super */
1072         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1073                                    index, GFP_KERNEL);
1074
1075         if (IS_ERR_OR_NULL(*page))
1076                 return 1;
1077
1078         p = kmap(*page);
1079
1080         /* align our pointer to the offset of the super block */
1081         *disk_super = p + (bytenr & ~PAGE_MASK);
1082
1083         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1084             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1085                 btrfs_release_disk_super(*page);
1086                 return 1;
1087         }
1088
1089         if ((*disk_super)->label[0] &&
1090                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1091                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1092
1093         return 0;
1094 }
1095
1096 /*
1097  * Look for a btrfs signature on a device. This may be called out of the mount path
1098  * and we are not allowed to call set_blocksize during the scan. The superblock
1099  * is read via pagecache
1100  */
1101 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1102                           struct btrfs_fs_devices **fs_devices_ret)
1103 {
1104         struct btrfs_super_block *disk_super;
1105         struct block_device *bdev;
1106         struct page *page;
1107         int ret = -EINVAL;
1108         u64 devid;
1109         u64 transid;
1110         u64 total_devices;
1111         u64 bytenr;
1112
1113         /*
1114          * we would like to check all the supers, but that would make
1115          * a btrfs mount succeed after a mkfs from a different FS.
1116          * So, we need to add a special mount option to scan for
1117          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1118          */
1119         bytenr = btrfs_sb_offset(0);
1120         flags |= FMODE_EXCL;
1121         mutex_lock(&uuid_mutex);
1122
1123         bdev = blkdev_get_by_path(path, flags, holder);
1124         if (IS_ERR(bdev)) {
1125                 ret = PTR_ERR(bdev);
1126                 goto error;
1127         }
1128
1129         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1130                 goto error_bdev_put;
1131
1132         devid = btrfs_stack_device_id(&disk_super->dev_item);
1133         transid = btrfs_super_generation(disk_super);
1134         total_devices = btrfs_super_num_devices(disk_super);
1135
1136         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1137         if (ret > 0) {
1138                 if (disk_super->label[0]) {
1139                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1140                 } else {
1141                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1142                 }
1143
1144                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1145                 ret = 0;
1146         }
1147         if (!ret && fs_devices_ret)
1148                 (*fs_devices_ret)->total_devices = total_devices;
1149
1150         btrfs_release_disk_super(page);
1151
1152 error_bdev_put:
1153         blkdev_put(bdev, flags);
1154 error:
1155         mutex_unlock(&uuid_mutex);
1156         return ret;
1157 }
1158
1159 /* helper to account the used device space in the range */
1160 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1161                                    u64 end, u64 *length)
1162 {
1163         struct btrfs_key key;
1164         struct btrfs_root *root = device->dev_root;
1165         struct btrfs_dev_extent *dev_extent;
1166         struct btrfs_path *path;
1167         u64 extent_end;
1168         int ret;
1169         int slot;
1170         struct extent_buffer *l;
1171
1172         *length = 0;
1173
1174         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1175                 return 0;
1176
1177         path = btrfs_alloc_path();
1178         if (!path)
1179                 return -ENOMEM;
1180         path->reada = READA_FORWARD;
1181
1182         key.objectid = device->devid;
1183         key.offset = start;
1184         key.type = BTRFS_DEV_EXTENT_KEY;
1185
1186         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1187         if (ret < 0)
1188                 goto out;
1189         if (ret > 0) {
1190                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1191                 if (ret < 0)
1192                         goto out;
1193         }
1194
1195         while (1) {
1196                 l = path->nodes[0];
1197                 slot = path->slots[0];
1198                 if (slot >= btrfs_header_nritems(l)) {
1199                         ret = btrfs_next_leaf(root, path);
1200                         if (ret == 0)
1201                                 continue;
1202                         if (ret < 0)
1203                                 goto out;
1204
1205                         break;
1206                 }
1207                 btrfs_item_key_to_cpu(l, &key, slot);
1208
1209                 if (key.objectid < device->devid)
1210                         goto next;
1211
1212                 if (key.objectid > device->devid)
1213                         break;
1214
1215                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1216                         goto next;
1217
1218                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1219                 extent_end = key.offset + btrfs_dev_extent_length(l,
1220                                                                   dev_extent);
1221                 if (key.offset <= start && extent_end > end) {
1222                         *length = end - start + 1;
1223                         break;
1224                 } else if (key.offset <= start && extent_end > start)
1225                         *length += extent_end - start;
1226                 else if (key.offset > start && extent_end <= end)
1227                         *length += extent_end - key.offset;
1228                 else if (key.offset > start && key.offset <= end) {
1229                         *length += end - key.offset + 1;
1230                         break;
1231                 } else if (key.offset > end)
1232                         break;
1233
1234 next:
1235                 path->slots[0]++;
1236         }
1237         ret = 0;
1238 out:
1239         btrfs_free_path(path);
1240         return ret;
1241 }
1242
1243 static int contains_pending_extent(struct btrfs_transaction *transaction,
1244                                    struct btrfs_device *device,
1245                                    u64 *start, u64 len)
1246 {
1247         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1248         struct extent_map *em;
1249         struct list_head *search_list = &fs_info->pinned_chunks;
1250         int ret = 0;
1251         u64 physical_start = *start;
1252
1253         if (transaction)
1254                 search_list = &transaction->pending_chunks;
1255 again:
1256         list_for_each_entry(em, search_list, list) {
1257                 struct map_lookup *map;
1258                 int i;
1259
1260                 map = em->map_lookup;
1261                 for (i = 0; i < map->num_stripes; i++) {
1262                         u64 end;
1263
1264                         if (map->stripes[i].dev != device)
1265                                 continue;
1266                         if (map->stripes[i].physical >= physical_start + len ||
1267                             map->stripes[i].physical + em->orig_block_len <=
1268                             physical_start)
1269                                 continue;
1270                         /*
1271                          * Make sure that while processing the pinned list we do
1272                          * not override our *start with a lower value, because
1273                          * we can have pinned chunks that fall within this
1274                          * device hole and that have lower physical addresses
1275                          * than the pending chunks we processed before. If we
1276                          * do not take this special care we can end up getting
1277                          * 2 pending chunks that start at the same physical
1278                          * device offsets because the end offset of a pinned
1279                          * chunk can be equal to the start offset of some
1280                          * pending chunk.
1281                          */
1282                         end = map->stripes[i].physical + em->orig_block_len;
1283                         if (end > *start) {
1284                                 *start = end;
1285                                 ret = 1;
1286                         }
1287                 }
1288         }
1289         if (search_list != &fs_info->pinned_chunks) {
1290                 search_list = &fs_info->pinned_chunks;
1291                 goto again;
1292         }
1293
1294         return ret;
1295 }
1296
1297
1298 /*
1299  * find_free_dev_extent_start - find free space in the specified device
1300  * @device:       the device which we search the free space in
1301  * @num_bytes:    the size of the free space that we need
1302  * @search_start: the position from which to begin the search
1303  * @start:        store the start of the free space.
1304  * @len:          the size of the free space. that we find, or the size
1305  *                of the max free space if we don't find suitable free space
1306  *
1307  * this uses a pretty simple search, the expectation is that it is
1308  * called very infrequently and that a given device has a small number
1309  * of extents
1310  *
1311  * @start is used to store the start of the free space if we find. But if we
1312  * don't find suitable free space, it will be used to store the start position
1313  * of the max free space.
1314  *
1315  * @len is used to store the size of the free space that we find.
1316  * But if we don't find suitable free space, it is used to store the size of
1317  * the max free space.
1318  */
1319 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1320                                struct btrfs_device *device, u64 num_bytes,
1321                                u64 search_start, u64 *start, u64 *len)
1322 {
1323         struct btrfs_key key;
1324         struct btrfs_root *root = device->dev_root;
1325         struct btrfs_dev_extent *dev_extent;
1326         struct btrfs_path *path;
1327         u64 hole_size;
1328         u64 max_hole_start;
1329         u64 max_hole_size;
1330         u64 extent_end;
1331         u64 search_end = device->total_bytes;
1332         int ret;
1333         int slot;
1334         struct extent_buffer *l;
1335         u64 min_search_start;
1336
1337         /*
1338          * We don't want to overwrite the superblock on the drive nor any area
1339          * used by the boot loader (grub for example), so we make sure to start
1340          * at an offset of at least 1MB.
1341          */
1342         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1343         search_start = max(search_start, min_search_start);
1344
1345         path = btrfs_alloc_path();
1346         if (!path)
1347                 return -ENOMEM;
1348
1349         max_hole_start = search_start;
1350         max_hole_size = 0;
1351
1352 again:
1353         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1354                 ret = -ENOSPC;
1355                 goto out;
1356         }
1357
1358         path->reada = READA_FORWARD;
1359         path->search_commit_root = 1;
1360         path->skip_locking = 1;
1361
1362         key.objectid = device->devid;
1363         key.offset = search_start;
1364         key.type = BTRFS_DEV_EXTENT_KEY;
1365
1366         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1367         if (ret < 0)
1368                 goto out;
1369         if (ret > 0) {
1370                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1371                 if (ret < 0)
1372                         goto out;
1373         }
1374
1375         while (1) {
1376                 l = path->nodes[0];
1377                 slot = path->slots[0];
1378                 if (slot >= btrfs_header_nritems(l)) {
1379                         ret = btrfs_next_leaf(root, path);
1380                         if (ret == 0)
1381                                 continue;
1382                         if (ret < 0)
1383                                 goto out;
1384
1385                         break;
1386                 }
1387                 btrfs_item_key_to_cpu(l, &key, slot);
1388
1389                 if (key.objectid < device->devid)
1390                         goto next;
1391
1392                 if (key.objectid > device->devid)
1393                         break;
1394
1395                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1396                         goto next;
1397
1398                 if (key.offset > search_start) {
1399                         hole_size = key.offset - search_start;
1400
1401                         /*
1402                          * Have to check before we set max_hole_start, otherwise
1403                          * we could end up sending back this offset anyway.
1404                          */
1405                         if (contains_pending_extent(transaction, device,
1406                                                     &search_start,
1407                                                     hole_size)) {
1408                                 if (key.offset >= search_start) {
1409                                         hole_size = key.offset - search_start;
1410                                 } else {
1411                                         WARN_ON_ONCE(1);
1412                                         hole_size = 0;
1413                                 }
1414                         }
1415
1416                         if (hole_size > max_hole_size) {
1417                                 max_hole_start = search_start;
1418                                 max_hole_size = hole_size;
1419                         }
1420
1421                         /*
1422                          * If this free space is greater than which we need,
1423                          * it must be the max free space that we have found
1424                          * until now, so max_hole_start must point to the start
1425                          * of this free space and the length of this free space
1426                          * is stored in max_hole_size. Thus, we return
1427                          * max_hole_start and max_hole_size and go back to the
1428                          * caller.
1429                          */
1430                         if (hole_size >= num_bytes) {
1431                                 ret = 0;
1432                                 goto out;
1433                         }
1434                 }
1435
1436                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1437                 extent_end = key.offset + btrfs_dev_extent_length(l,
1438                                                                   dev_extent);
1439                 if (extent_end > search_start)
1440                         search_start = extent_end;
1441 next:
1442                 path->slots[0]++;
1443                 cond_resched();
1444         }
1445
1446         /*
1447          * At this point, search_start should be the end of
1448          * allocated dev extents, and when shrinking the device,
1449          * search_end may be smaller than search_start.
1450          */
1451         if (search_end > search_start) {
1452                 hole_size = search_end - search_start;
1453
1454                 if (contains_pending_extent(transaction, device, &search_start,
1455                                             hole_size)) {
1456                         btrfs_release_path(path);
1457                         goto again;
1458                 }
1459
1460                 if (hole_size > max_hole_size) {
1461                         max_hole_start = search_start;
1462                         max_hole_size = hole_size;
1463                 }
1464         }
1465
1466         /* See above. */
1467         if (max_hole_size < num_bytes)
1468                 ret = -ENOSPC;
1469         else
1470                 ret = 0;
1471
1472 out:
1473         btrfs_free_path(path);
1474         *start = max_hole_start;
1475         if (len)
1476                 *len = max_hole_size;
1477         return ret;
1478 }
1479
1480 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1481                          struct btrfs_device *device, u64 num_bytes,
1482                          u64 *start, u64 *len)
1483 {
1484         /* FIXME use last free of some kind */
1485         return find_free_dev_extent_start(trans->transaction, device,
1486                                           num_bytes, 0, start, len);
1487 }
1488
1489 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1490                           struct btrfs_device *device,
1491                           u64 start, u64 *dev_extent_len)
1492 {
1493         int ret;
1494         struct btrfs_path *path;
1495         struct btrfs_root *root = device->dev_root;
1496         struct btrfs_key key;
1497         struct btrfs_key found_key;
1498         struct extent_buffer *leaf = NULL;
1499         struct btrfs_dev_extent *extent = NULL;
1500
1501         path = btrfs_alloc_path();
1502         if (!path)
1503                 return -ENOMEM;
1504
1505         key.objectid = device->devid;
1506         key.offset = start;
1507         key.type = BTRFS_DEV_EXTENT_KEY;
1508 again:
1509         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1510         if (ret > 0) {
1511                 ret = btrfs_previous_item(root, path, key.objectid,
1512                                           BTRFS_DEV_EXTENT_KEY);
1513                 if (ret)
1514                         goto out;
1515                 leaf = path->nodes[0];
1516                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1517                 extent = btrfs_item_ptr(leaf, path->slots[0],
1518                                         struct btrfs_dev_extent);
1519                 BUG_ON(found_key.offset > start || found_key.offset +
1520                        btrfs_dev_extent_length(leaf, extent) < start);
1521                 key = found_key;
1522                 btrfs_release_path(path);
1523                 goto again;
1524         } else if (ret == 0) {
1525                 leaf = path->nodes[0];
1526                 extent = btrfs_item_ptr(leaf, path->slots[0],
1527                                         struct btrfs_dev_extent);
1528         } else {
1529                 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1530                 goto out;
1531         }
1532
1533         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1534
1535         ret = btrfs_del_item(trans, root, path);
1536         if (ret) {
1537                 btrfs_handle_fs_error(root->fs_info, ret,
1538                             "Failed to remove dev extent item");
1539         } else {
1540                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1541         }
1542 out:
1543         btrfs_free_path(path);
1544         return ret;
1545 }
1546
1547 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1548                                   struct btrfs_device *device,
1549                                   u64 chunk_tree, u64 chunk_objectid,
1550                                   u64 chunk_offset, u64 start, u64 num_bytes)
1551 {
1552         int ret;
1553         struct btrfs_path *path;
1554         struct btrfs_root *root = device->dev_root;
1555         struct btrfs_dev_extent *extent;
1556         struct extent_buffer *leaf;
1557         struct btrfs_key key;
1558
1559         WARN_ON(!device->in_fs_metadata);
1560         WARN_ON(device->is_tgtdev_for_dev_replace);
1561         path = btrfs_alloc_path();
1562         if (!path)
1563                 return -ENOMEM;
1564
1565         key.objectid = device->devid;
1566         key.offset = start;
1567         key.type = BTRFS_DEV_EXTENT_KEY;
1568         ret = btrfs_insert_empty_item(trans, root, path, &key,
1569                                       sizeof(*extent));
1570         if (ret)
1571                 goto out;
1572
1573         leaf = path->nodes[0];
1574         extent = btrfs_item_ptr(leaf, path->slots[0],
1575                                 struct btrfs_dev_extent);
1576         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1577         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1578         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1579
1580         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1581                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1582
1583         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1584         btrfs_mark_buffer_dirty(leaf);
1585 out:
1586         btrfs_free_path(path);
1587         return ret;
1588 }
1589
1590 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1591 {
1592         struct extent_map_tree *em_tree;
1593         struct extent_map *em;
1594         struct rb_node *n;
1595         u64 ret = 0;
1596
1597         em_tree = &fs_info->mapping_tree.map_tree;
1598         read_lock(&em_tree->lock);
1599         n = rb_last(&em_tree->map);
1600         if (n) {
1601                 em = rb_entry(n, struct extent_map, rb_node);
1602                 ret = em->start + em->len;
1603         }
1604         read_unlock(&em_tree->lock);
1605
1606         return ret;
1607 }
1608
1609 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1610                                     u64 *devid_ret)
1611 {
1612         int ret;
1613         struct btrfs_key key;
1614         struct btrfs_key found_key;
1615         struct btrfs_path *path;
1616
1617         path = btrfs_alloc_path();
1618         if (!path)
1619                 return -ENOMEM;
1620
1621         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1622         key.type = BTRFS_DEV_ITEM_KEY;
1623         key.offset = (u64)-1;
1624
1625         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1626         if (ret < 0)
1627                 goto error;
1628
1629         BUG_ON(ret == 0); /* Corruption */
1630
1631         ret = btrfs_previous_item(fs_info->chunk_root, path,
1632                                   BTRFS_DEV_ITEMS_OBJECTID,
1633                                   BTRFS_DEV_ITEM_KEY);
1634         if (ret) {
1635                 *devid_ret = 1;
1636         } else {
1637                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1638                                       path->slots[0]);
1639                 *devid_ret = found_key.offset + 1;
1640         }
1641         ret = 0;
1642 error:
1643         btrfs_free_path(path);
1644         return ret;
1645 }
1646
1647 /*
1648  * the device information is stored in the chunk root
1649  * the btrfs_device struct should be fully filled in
1650  */
1651 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1652                             struct btrfs_root *root,
1653                             struct btrfs_device *device)
1654 {
1655         int ret;
1656         struct btrfs_path *path;
1657         struct btrfs_dev_item *dev_item;
1658         struct extent_buffer *leaf;
1659         struct btrfs_key key;
1660         unsigned long ptr;
1661
1662         root = root->fs_info->chunk_root;
1663
1664         path = btrfs_alloc_path();
1665         if (!path)
1666                 return -ENOMEM;
1667
1668         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1669         key.type = BTRFS_DEV_ITEM_KEY;
1670         key.offset = device->devid;
1671
1672         ret = btrfs_insert_empty_item(trans, root, path, &key,
1673                                       sizeof(*dev_item));
1674         if (ret)
1675                 goto out;
1676
1677         leaf = path->nodes[0];
1678         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1679
1680         btrfs_set_device_id(leaf, dev_item, device->devid);
1681         btrfs_set_device_generation(leaf, dev_item, 0);
1682         btrfs_set_device_type(leaf, dev_item, device->type);
1683         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1684         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1685         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1686         btrfs_set_device_total_bytes(leaf, dev_item,
1687                                      btrfs_device_get_disk_total_bytes(device));
1688         btrfs_set_device_bytes_used(leaf, dev_item,
1689                                     btrfs_device_get_bytes_used(device));
1690         btrfs_set_device_group(leaf, dev_item, 0);
1691         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1692         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1693         btrfs_set_device_start_offset(leaf, dev_item, 0);
1694
1695         ptr = btrfs_device_uuid(dev_item);
1696         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1697         ptr = btrfs_device_fsid(dev_item);
1698         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1699         btrfs_mark_buffer_dirty(leaf);
1700
1701         ret = 0;
1702 out:
1703         btrfs_free_path(path);
1704         return ret;
1705 }
1706
1707 /*
1708  * Function to update ctime/mtime for a given device path.
1709  * Mainly used for ctime/mtime based probe like libblkid.
1710  */
1711 static void update_dev_time(char *path_name)
1712 {
1713         struct file *filp;
1714
1715         filp = filp_open(path_name, O_RDWR, 0);
1716         if (IS_ERR(filp))
1717                 return;
1718         file_update_time(filp);
1719         filp_close(filp, NULL);
1720 }
1721
1722 static int btrfs_rm_dev_item(struct btrfs_root *root,
1723                              struct btrfs_device *device)
1724 {
1725         int ret;
1726         struct btrfs_path *path;
1727         struct btrfs_key key;
1728         struct btrfs_trans_handle *trans;
1729
1730         root = root->fs_info->chunk_root;
1731
1732         path = btrfs_alloc_path();
1733         if (!path)
1734                 return -ENOMEM;
1735
1736         trans = btrfs_start_transaction(root, 0);
1737         if (IS_ERR(trans)) {
1738                 btrfs_free_path(path);
1739                 return PTR_ERR(trans);
1740         }
1741         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1742         key.type = BTRFS_DEV_ITEM_KEY;
1743         key.offset = device->devid;
1744
1745         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1746         if (ret < 0)
1747                 goto out;
1748
1749         if (ret > 0) {
1750                 ret = -ENOENT;
1751                 goto out;
1752         }
1753
1754         ret = btrfs_del_item(trans, root, path);
1755         if (ret)
1756                 goto out;
1757 out:
1758         btrfs_free_path(path);
1759         btrfs_commit_transaction(trans, root);
1760         return ret;
1761 }
1762
1763 /*
1764  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1765  * filesystem. It's up to the caller to adjust that number regarding eg. device
1766  * replace.
1767  */
1768 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1769                 u64 num_devices)
1770 {
1771         u64 all_avail;
1772         unsigned seq;
1773         int i;
1774
1775         do {
1776                 seq = read_seqbegin(&fs_info->profiles_lock);
1777
1778                 all_avail = fs_info->avail_data_alloc_bits |
1779                             fs_info->avail_system_alloc_bits |
1780                             fs_info->avail_metadata_alloc_bits;
1781         } while (read_seqretry(&fs_info->profiles_lock, seq));
1782
1783         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1784                 if (!(all_avail & btrfs_raid_group[i]))
1785                         continue;
1786
1787                 if (num_devices < btrfs_raid_array[i].devs_min) {
1788                         int ret = btrfs_raid_mindev_error[i];
1789
1790                         if (ret)
1791                                 return ret;
1792                 }
1793         }
1794
1795         return 0;
1796 }
1797
1798 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1799                                         struct btrfs_device *device)
1800 {
1801         struct btrfs_device *next_device;
1802
1803         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1804                 if (next_device != device &&
1805                         !next_device->missing && next_device->bdev)
1806                         return next_device;
1807         }
1808
1809         return NULL;
1810 }
1811
1812 /*
1813  * Helper function to check if the given device is part of s_bdev / latest_bdev
1814  * and replace it with the provided or the next active device, in the context
1815  * where this function called, there should be always be another device (or
1816  * this_dev) which is active.
1817  */
1818 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1819                 struct btrfs_device *device, struct btrfs_device *this_dev)
1820 {
1821         struct btrfs_device *next_device;
1822
1823         if (this_dev)
1824                 next_device = this_dev;
1825         else
1826                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1827                                                                 device);
1828         ASSERT(next_device);
1829
1830         if (fs_info->sb->s_bdev &&
1831                         (fs_info->sb->s_bdev == device->bdev))
1832                 fs_info->sb->s_bdev = next_device->bdev;
1833
1834         if (fs_info->fs_devices->latest_bdev == device->bdev)
1835                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1836 }
1837
1838 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1839 {
1840         struct btrfs_device *device;
1841         struct btrfs_fs_devices *cur_devices;
1842         u64 num_devices;
1843         int ret = 0;
1844         bool clear_super = false;
1845         char *dev_name = NULL;
1846
1847         mutex_lock(&uuid_mutex);
1848
1849         num_devices = root->fs_info->fs_devices->num_devices;
1850         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1851         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1852                 WARN_ON(num_devices < 1);
1853                 num_devices--;
1854         }
1855         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1856
1857         ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1858         if (ret)
1859                 goto out;
1860
1861         ret = btrfs_find_device_by_devspec(root, devid, device_path,
1862                                 &device);
1863         if (ret)
1864                 goto out;
1865
1866         if (device->is_tgtdev_for_dev_replace) {
1867                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1868                 goto out;
1869         }
1870
1871         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1872                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1873                 goto out;
1874         }
1875
1876         if (device->writeable) {
1877                 lock_chunks(root);
1878                 list_del_init(&device->dev_alloc_list);
1879                 device->fs_devices->rw_devices--;
1880                 unlock_chunks(root);
1881                 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1882                 if (!dev_name) {
1883                         ret = -ENOMEM;
1884                         goto error_undo;
1885                 }
1886                 clear_super = true;
1887         }
1888
1889         mutex_unlock(&uuid_mutex);
1890         ret = btrfs_shrink_device(device, 0);
1891         mutex_lock(&uuid_mutex);
1892         if (ret)
1893                 goto error_undo;
1894
1895         /*
1896          * TODO: the superblock still includes this device in its num_devices
1897          * counter although write_all_supers() is not locked out. This
1898          * could give a filesystem state which requires a degraded mount.
1899          */
1900         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1901         if (ret)
1902                 goto error_undo;
1903
1904         device->in_fs_metadata = 0;
1905         btrfs_scrub_cancel_dev(root->fs_info, device);
1906
1907         /*
1908          * the device list mutex makes sure that we don't change
1909          * the device list while someone else is writing out all
1910          * the device supers. Whoever is writing all supers, should
1911          * lock the device list mutex before getting the number of
1912          * devices in the super block (super_copy). Conversely,
1913          * whoever updates the number of devices in the super block
1914          * (super_copy) should hold the device list mutex.
1915          */
1916
1917         cur_devices = device->fs_devices;
1918         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1919         list_del_rcu(&device->dev_list);
1920
1921         device->fs_devices->num_devices--;
1922         device->fs_devices->total_devices--;
1923
1924         if (device->missing)
1925                 device->fs_devices->missing_devices--;
1926
1927         btrfs_assign_next_active_device(root->fs_info, device, NULL);
1928
1929         if (device->bdev) {
1930                 device->fs_devices->open_devices--;
1931                 /* remove sysfs entry */
1932                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1933         }
1934
1935         call_rcu(&device->rcu, free_device);
1936
1937         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1938         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1939         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1940
1941         if (cur_devices->open_devices == 0) {
1942                 struct btrfs_fs_devices *fs_devices;
1943                 fs_devices = root->fs_info->fs_devices;
1944                 while (fs_devices) {
1945                         if (fs_devices->seed == cur_devices) {
1946                                 fs_devices->seed = cur_devices->seed;
1947                                 break;
1948                         }
1949                         fs_devices = fs_devices->seed;
1950                 }
1951                 cur_devices->seed = NULL;
1952                 __btrfs_close_devices(cur_devices);
1953                 free_fs_devices(cur_devices);
1954         }
1955
1956         root->fs_info->num_tolerated_disk_barrier_failures =
1957                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1958
1959         /*
1960          * at this point, the device is zero sized.  We want to
1961          * remove it from the devices list and zero out the old super
1962          */
1963         if (clear_super) {
1964                 struct block_device *bdev;
1965
1966                 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1967                                                 root->fs_info->bdev_holder);
1968                 if (!IS_ERR(bdev)) {
1969                         btrfs_scratch_superblocks(bdev, dev_name);
1970                         blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1971                 }
1972         }
1973
1974 out:
1975         kfree(dev_name);
1976
1977         mutex_unlock(&uuid_mutex);
1978         return ret;
1979
1980 error_undo:
1981         if (device->writeable) {
1982                 lock_chunks(root);
1983                 list_add(&device->dev_alloc_list,
1984                          &root->fs_info->fs_devices->alloc_list);
1985                 device->fs_devices->rw_devices++;
1986                 unlock_chunks(root);
1987         }
1988         goto out;
1989 }
1990
1991 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1992                                         struct btrfs_device *srcdev)
1993 {
1994         struct btrfs_fs_devices *fs_devices;
1995
1996         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1997
1998         /*
1999          * in case of fs with no seed, srcdev->fs_devices will point
2000          * to fs_devices of fs_info. However when the dev being replaced is
2001          * a seed dev it will point to the seed's local fs_devices. In short
2002          * srcdev will have its correct fs_devices in both the cases.
2003          */
2004         fs_devices = srcdev->fs_devices;
2005
2006         list_del_rcu(&srcdev->dev_list);
2007         list_del_rcu(&srcdev->dev_alloc_list);
2008         fs_devices->num_devices--;
2009         if (srcdev->missing)
2010                 fs_devices->missing_devices--;
2011
2012         if (srcdev->writeable)
2013                 fs_devices->rw_devices--;
2014
2015         if (srcdev->bdev)
2016                 fs_devices->open_devices--;
2017 }
2018
2019 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2020                                       struct btrfs_device *srcdev)
2021 {
2022         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2023
2024         if (srcdev->writeable) {
2025                 /* zero out the old super if it is writable */
2026                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2027         }
2028         call_rcu(&srcdev->rcu, free_device);
2029
2030         /*
2031          * unless fs_devices is seed fs, num_devices shouldn't go
2032          * zero
2033          */
2034         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2035
2036         /* if this is no devs we rather delete the fs_devices */
2037         if (!fs_devices->num_devices) {
2038                 struct btrfs_fs_devices *tmp_fs_devices;
2039
2040                 tmp_fs_devices = fs_info->fs_devices;
2041                 while (tmp_fs_devices) {
2042                         if (tmp_fs_devices->seed == fs_devices) {
2043                                 tmp_fs_devices->seed = fs_devices->seed;
2044                                 break;
2045                         }
2046                         tmp_fs_devices = tmp_fs_devices->seed;
2047                 }
2048                 fs_devices->seed = NULL;
2049                 __btrfs_close_devices(fs_devices);
2050                 free_fs_devices(fs_devices);
2051         }
2052 }
2053
2054 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2055                                       struct btrfs_device *tgtdev)
2056 {
2057         mutex_lock(&uuid_mutex);
2058         WARN_ON(!tgtdev);
2059         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2060
2061         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2062
2063         if (tgtdev->bdev)
2064                 fs_info->fs_devices->open_devices--;
2065
2066         fs_info->fs_devices->num_devices--;
2067
2068         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2069
2070         list_del_rcu(&tgtdev->dev_list);
2071
2072         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2073         mutex_unlock(&uuid_mutex);
2074
2075         /*
2076          * The update_dev_time() with in btrfs_scratch_superblocks()
2077          * may lead to a call to btrfs_show_devname() which will try
2078          * to hold device_list_mutex. And here this device
2079          * is already out of device list, so we don't have to hold
2080          * the device_list_mutex lock.
2081          */
2082         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2083         call_rcu(&tgtdev->rcu, free_device);
2084 }
2085
2086 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2087                                      struct btrfs_device **device)
2088 {
2089         int ret = 0;
2090         struct btrfs_super_block *disk_super;
2091         u64 devid;
2092         u8 *dev_uuid;
2093         struct block_device *bdev;
2094         struct buffer_head *bh;
2095
2096         *device = NULL;
2097         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2098                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2099         if (ret)
2100                 return ret;
2101         disk_super = (struct btrfs_super_block *)bh->b_data;
2102         devid = btrfs_stack_device_id(&disk_super->dev_item);
2103         dev_uuid = disk_super->dev_item.uuid;
2104         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2105                                     disk_super->fsid);
2106         brelse(bh);
2107         if (!*device)
2108                 ret = -ENOENT;
2109         blkdev_put(bdev, FMODE_READ);
2110         return ret;
2111 }
2112
2113 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2114                                          char *device_path,
2115                                          struct btrfs_device **device)
2116 {
2117         *device = NULL;
2118         if (strcmp(device_path, "missing") == 0) {
2119                 struct list_head *devices;
2120                 struct btrfs_device *tmp;
2121
2122                 devices = &root->fs_info->fs_devices->devices;
2123                 /*
2124                  * It is safe to read the devices since the volume_mutex
2125                  * is held by the caller.
2126                  */
2127                 list_for_each_entry(tmp, devices, dev_list) {
2128                         if (tmp->in_fs_metadata && !tmp->bdev) {
2129                                 *device = tmp;
2130                                 break;
2131                         }
2132                 }
2133
2134                 if (!*device)
2135                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2136
2137                 return 0;
2138         } else {
2139                 return btrfs_find_device_by_path(root, device_path, device);
2140         }
2141 }
2142
2143 /*
2144  * Lookup a device given by device id, or the path if the id is 0.
2145  */
2146 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2147                                          char *devpath,
2148                                          struct btrfs_device **device)
2149 {
2150         int ret;
2151
2152         if (devid) {
2153                 ret = 0;
2154                 *device = btrfs_find_device(root->fs_info, devid, NULL,
2155                                             NULL);
2156                 if (!*device)
2157                         ret = -ENOENT;
2158         } else {
2159                 if (!devpath || !devpath[0])
2160                         return -EINVAL;
2161
2162                 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2163                                                            device);
2164         }
2165         return ret;
2166 }
2167
2168 /*
2169  * does all the dirty work required for changing file system's UUID.
2170  */
2171 static int btrfs_prepare_sprout(struct btrfs_root *root)
2172 {
2173         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2174         struct btrfs_fs_devices *old_devices;
2175         struct btrfs_fs_devices *seed_devices;
2176         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2177         struct btrfs_device *device;
2178         u64 super_flags;
2179
2180         BUG_ON(!mutex_is_locked(&uuid_mutex));
2181         if (!fs_devices->seeding)
2182                 return -EINVAL;
2183
2184         seed_devices = __alloc_fs_devices();
2185         if (IS_ERR(seed_devices))
2186                 return PTR_ERR(seed_devices);
2187
2188         old_devices = clone_fs_devices(fs_devices);
2189         if (IS_ERR(old_devices)) {
2190                 kfree(seed_devices);
2191                 return PTR_ERR(old_devices);
2192         }
2193
2194         list_add(&old_devices->list, &fs_uuids);
2195
2196         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2197         seed_devices->opened = 1;
2198         INIT_LIST_HEAD(&seed_devices->devices);
2199         INIT_LIST_HEAD(&seed_devices->alloc_list);
2200         mutex_init(&seed_devices->device_list_mutex);
2201
2202         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2203         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2204                               synchronize_rcu);
2205         list_for_each_entry(device, &seed_devices->devices, dev_list)
2206                 device->fs_devices = seed_devices;
2207
2208         lock_chunks(root);
2209         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2210         unlock_chunks(root);
2211
2212         fs_devices->seeding = 0;
2213         fs_devices->num_devices = 0;
2214         fs_devices->open_devices = 0;
2215         fs_devices->missing_devices = 0;
2216         fs_devices->rotating = 0;
2217         fs_devices->seed = seed_devices;
2218
2219         generate_random_uuid(fs_devices->fsid);
2220         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2221         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2222         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2223
2224         super_flags = btrfs_super_flags(disk_super) &
2225                       ~BTRFS_SUPER_FLAG_SEEDING;
2226         btrfs_set_super_flags(disk_super, super_flags);
2227
2228         return 0;
2229 }
2230
2231 /*
2232  * Store the expected generation for seed devices in device items.
2233  */
2234 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2235                                struct btrfs_root *root)
2236 {
2237         struct btrfs_path *path;
2238         struct extent_buffer *leaf;
2239         struct btrfs_dev_item *dev_item;
2240         struct btrfs_device *device;
2241         struct btrfs_key key;
2242         u8 fs_uuid[BTRFS_UUID_SIZE];
2243         u8 dev_uuid[BTRFS_UUID_SIZE];
2244         u64 devid;
2245         int ret;
2246
2247         path = btrfs_alloc_path();
2248         if (!path)
2249                 return -ENOMEM;
2250
2251         root = root->fs_info->chunk_root;
2252         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2253         key.offset = 0;
2254         key.type = BTRFS_DEV_ITEM_KEY;
2255
2256         while (1) {
2257                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2258                 if (ret < 0)
2259                         goto error;
2260
2261                 leaf = path->nodes[0];
2262 next_slot:
2263                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2264                         ret = btrfs_next_leaf(root, path);
2265                         if (ret > 0)
2266                                 break;
2267                         if (ret < 0)
2268                                 goto error;
2269                         leaf = path->nodes[0];
2270                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2271                         btrfs_release_path(path);
2272                         continue;
2273                 }
2274
2275                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2276                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2277                     key.type != BTRFS_DEV_ITEM_KEY)
2278                         break;
2279
2280                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2281                                           struct btrfs_dev_item);
2282                 devid = btrfs_device_id(leaf, dev_item);
2283                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2284                                    BTRFS_UUID_SIZE);
2285                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2286                                    BTRFS_UUID_SIZE);
2287                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2288                                            fs_uuid);
2289                 BUG_ON(!device); /* Logic error */
2290
2291                 if (device->fs_devices->seeding) {
2292                         btrfs_set_device_generation(leaf, dev_item,
2293                                                     device->generation);
2294                         btrfs_mark_buffer_dirty(leaf);
2295                 }
2296
2297                 path->slots[0]++;
2298                 goto next_slot;
2299         }
2300         ret = 0;
2301 error:
2302         btrfs_free_path(path);
2303         return ret;
2304 }
2305
2306 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2307 {
2308         struct request_queue *q;
2309         struct btrfs_trans_handle *trans;
2310         struct btrfs_device *device;
2311         struct block_device *bdev;
2312         struct list_head *devices;
2313         struct super_block *sb = root->fs_info->sb;
2314         struct rcu_string *name;
2315         u64 tmp;
2316         int seeding_dev = 0;
2317         int ret = 0;
2318
2319         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2320                 return -EROFS;
2321
2322         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2323                                   root->fs_info->bdev_holder);
2324         if (IS_ERR(bdev))
2325                 return PTR_ERR(bdev);
2326
2327         if (root->fs_info->fs_devices->seeding) {
2328                 seeding_dev = 1;
2329                 down_write(&sb->s_umount);
2330                 mutex_lock(&uuid_mutex);
2331         }
2332
2333         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2334
2335         devices = &root->fs_info->fs_devices->devices;
2336
2337         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2338         list_for_each_entry(device, devices, dev_list) {
2339                 if (device->bdev == bdev) {
2340                         ret = -EEXIST;
2341                         mutex_unlock(
2342                                 &root->fs_info->fs_devices->device_list_mutex);
2343                         goto error;
2344                 }
2345         }
2346         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2347
2348         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2349         if (IS_ERR(device)) {
2350                 /* we can safely leave the fs_devices entry around */
2351                 ret = PTR_ERR(device);
2352                 goto error;
2353         }
2354
2355         name = rcu_string_strdup(device_path, GFP_KERNEL);
2356         if (!name) {
2357                 kfree(device);
2358                 ret = -ENOMEM;
2359                 goto error;
2360         }
2361         rcu_assign_pointer(device->name, name);
2362
2363         trans = btrfs_start_transaction(root, 0);
2364         if (IS_ERR(trans)) {
2365                 rcu_string_free(device->name);
2366                 kfree(device);
2367                 ret = PTR_ERR(trans);
2368                 goto error;
2369         }
2370
2371         q = bdev_get_queue(bdev);
2372         if (blk_queue_discard(q))
2373                 device->can_discard = 1;
2374         device->writeable = 1;
2375         device->generation = trans->transid;
2376         device->io_width = root->sectorsize;
2377         device->io_align = root->sectorsize;
2378         device->sector_size = root->sectorsize;
2379         device->total_bytes = i_size_read(bdev->bd_inode);
2380         device->disk_total_bytes = device->total_bytes;
2381         device->commit_total_bytes = device->total_bytes;
2382         device->dev_root = root->fs_info->dev_root;
2383         device->bdev = bdev;
2384         device->in_fs_metadata = 1;
2385         device->is_tgtdev_for_dev_replace = 0;
2386         device->mode = FMODE_EXCL;
2387         device->dev_stats_valid = 1;
2388         set_blocksize(device->bdev, 4096);
2389
2390         if (seeding_dev) {
2391                 sb->s_flags &= ~MS_RDONLY;
2392                 ret = btrfs_prepare_sprout(root);
2393                 BUG_ON(ret); /* -ENOMEM */
2394         }
2395
2396         device->fs_devices = root->fs_info->fs_devices;
2397
2398         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2399         lock_chunks(root);
2400         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2401         list_add(&device->dev_alloc_list,
2402                  &root->fs_info->fs_devices->alloc_list);
2403         root->fs_info->fs_devices->num_devices++;
2404         root->fs_info->fs_devices->open_devices++;
2405         root->fs_info->fs_devices->rw_devices++;
2406         root->fs_info->fs_devices->total_devices++;
2407         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2408
2409         spin_lock(&root->fs_info->free_chunk_lock);
2410         root->fs_info->free_chunk_space += device->total_bytes;
2411         spin_unlock(&root->fs_info->free_chunk_lock);
2412
2413         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2414                 root->fs_info->fs_devices->rotating = 1;
2415
2416         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2417         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2418                                     tmp + device->total_bytes);
2419
2420         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2421         btrfs_set_super_num_devices(root->fs_info->super_copy,
2422                                     tmp + 1);
2423
2424         /* add sysfs device entry */
2425         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2426
2427         /*
2428          * we've got more storage, clear any full flags on the space
2429          * infos
2430          */
2431         btrfs_clear_space_info_full(root->fs_info);
2432
2433         unlock_chunks(root);
2434         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2435
2436         if (seeding_dev) {
2437                 lock_chunks(root);
2438                 ret = init_first_rw_device(trans, root, device);
2439                 unlock_chunks(root);
2440                 if (ret) {
2441                         btrfs_abort_transaction(trans, ret);
2442                         goto error_trans;
2443                 }
2444         }
2445
2446         ret = btrfs_add_device(trans, root, device);
2447         if (ret) {
2448                 btrfs_abort_transaction(trans, ret);
2449                 goto error_trans;
2450         }
2451
2452         if (seeding_dev) {
2453                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2454
2455                 ret = btrfs_finish_sprout(trans, root);
2456                 if (ret) {
2457                         btrfs_abort_transaction(trans, ret);
2458                         goto error_trans;
2459                 }
2460
2461                 /* Sprouting would change fsid of the mounted root,
2462                  * so rename the fsid on the sysfs
2463                  */
2464                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2465                                                 root->fs_info->fsid);
2466                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2467                                                                 fsid_buf))
2468                         btrfs_warn(root->fs_info,
2469                                 "sysfs: failed to create fsid for sprout");
2470         }
2471
2472         root->fs_info->num_tolerated_disk_barrier_failures =
2473                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2474         ret = btrfs_commit_transaction(trans, root);
2475
2476         if (seeding_dev) {
2477                 mutex_unlock(&uuid_mutex);
2478                 up_write(&sb->s_umount);
2479
2480                 if (ret) /* transaction commit */
2481                         return ret;
2482
2483                 ret = btrfs_relocate_sys_chunks(root);
2484                 if (ret < 0)
2485                         btrfs_handle_fs_error(root->fs_info, ret,
2486                                     "Failed to relocate sys chunks after "
2487                                     "device initialization. This can be fixed "
2488                                     "using the \"btrfs balance\" command.");
2489                 trans = btrfs_attach_transaction(root);
2490                 if (IS_ERR(trans)) {
2491                         if (PTR_ERR(trans) == -ENOENT)
2492                                 return 0;
2493                         return PTR_ERR(trans);
2494                 }
2495                 ret = btrfs_commit_transaction(trans, root);
2496         }
2497
2498         /* Update ctime/mtime for libblkid */
2499         update_dev_time(device_path);
2500         return ret;
2501
2502 error_trans:
2503         btrfs_end_transaction(trans, root);
2504         rcu_string_free(device->name);
2505         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2506         kfree(device);
2507 error:
2508         blkdev_put(bdev, FMODE_EXCL);
2509         if (seeding_dev) {
2510                 mutex_unlock(&uuid_mutex);
2511                 up_write(&sb->s_umount);
2512         }
2513         return ret;
2514 }
2515
2516 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2517                                   struct btrfs_device *srcdev,
2518                                   struct btrfs_device **device_out)
2519 {
2520         struct request_queue *q;
2521         struct btrfs_device *device;
2522         struct block_device *bdev;
2523         struct btrfs_fs_info *fs_info = root->fs_info;
2524         struct list_head *devices;
2525         struct rcu_string *name;
2526         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2527         int ret = 0;
2528
2529         *device_out = NULL;
2530         if (fs_info->fs_devices->seeding) {
2531                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2532                 return -EINVAL;
2533         }
2534
2535         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2536                                   fs_info->bdev_holder);
2537         if (IS_ERR(bdev)) {
2538                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2539                 return PTR_ERR(bdev);
2540         }
2541
2542         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2543
2544         devices = &fs_info->fs_devices->devices;
2545         list_for_each_entry(device, devices, dev_list) {
2546                 if (device->bdev == bdev) {
2547                         btrfs_err(fs_info, "target device is in the filesystem!");
2548                         ret = -EEXIST;
2549                         goto error;
2550                 }
2551         }
2552
2553
2554         if (i_size_read(bdev->bd_inode) <
2555             btrfs_device_get_total_bytes(srcdev)) {
2556                 btrfs_err(fs_info, "target device is smaller than source device!");
2557                 ret = -EINVAL;
2558                 goto error;
2559         }
2560
2561
2562         device = btrfs_alloc_device(NULL, &devid, NULL);
2563         if (IS_ERR(device)) {
2564                 ret = PTR_ERR(device);
2565                 goto error;
2566         }
2567
2568         name = rcu_string_strdup(device_path, GFP_NOFS);
2569         if (!name) {
2570                 kfree(device);
2571                 ret = -ENOMEM;
2572                 goto error;
2573         }
2574         rcu_assign_pointer(device->name, name);
2575
2576         q = bdev_get_queue(bdev);
2577         if (blk_queue_discard(q))
2578                 device->can_discard = 1;
2579         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2580         device->writeable = 1;
2581         device->generation = 0;
2582         device->io_width = root->sectorsize;
2583         device->io_align = root->sectorsize;
2584         device->sector_size = root->sectorsize;
2585         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2586         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2587         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2588         ASSERT(list_empty(&srcdev->resized_list));
2589         device->commit_total_bytes = srcdev->commit_total_bytes;
2590         device->commit_bytes_used = device->bytes_used;
2591         device->dev_root = fs_info->dev_root;
2592         device->bdev = bdev;
2593         device->in_fs_metadata = 1;
2594         device->is_tgtdev_for_dev_replace = 1;
2595         device->mode = FMODE_EXCL;
2596         device->dev_stats_valid = 1;
2597         set_blocksize(device->bdev, 4096);
2598         device->fs_devices = fs_info->fs_devices;
2599         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2600         fs_info->fs_devices->num_devices++;
2601         fs_info->fs_devices->open_devices++;
2602         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2603
2604         *device_out = device;
2605         return ret;
2606
2607 error:
2608         blkdev_put(bdev, FMODE_EXCL);
2609         return ret;
2610 }
2611
2612 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2613                                               struct btrfs_device *tgtdev)
2614 {
2615         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2616         tgtdev->io_width = fs_info->dev_root->sectorsize;
2617         tgtdev->io_align = fs_info->dev_root->sectorsize;
2618         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2619         tgtdev->dev_root = fs_info->dev_root;
2620         tgtdev->in_fs_metadata = 1;
2621 }
2622
2623 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2624                                         struct btrfs_device *device)
2625 {
2626         int ret;
2627         struct btrfs_path *path;
2628         struct btrfs_root *root;
2629         struct btrfs_dev_item *dev_item;
2630         struct extent_buffer *leaf;
2631         struct btrfs_key key;
2632
2633         root = device->dev_root->fs_info->chunk_root;
2634
2635         path = btrfs_alloc_path();
2636         if (!path)
2637                 return -ENOMEM;
2638
2639         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2640         key.type = BTRFS_DEV_ITEM_KEY;
2641         key.offset = device->devid;
2642
2643         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2644         if (ret < 0)
2645                 goto out;
2646
2647         if (ret > 0) {
2648                 ret = -ENOENT;
2649                 goto out;
2650         }
2651
2652         leaf = path->nodes[0];
2653         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2654
2655         btrfs_set_device_id(leaf, dev_item, device->devid);
2656         btrfs_set_device_type(leaf, dev_item, device->type);
2657         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2658         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2659         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2660         btrfs_set_device_total_bytes(leaf, dev_item,
2661                                      btrfs_device_get_disk_total_bytes(device));
2662         btrfs_set_device_bytes_used(leaf, dev_item,
2663                                     btrfs_device_get_bytes_used(device));
2664         btrfs_mark_buffer_dirty(leaf);
2665
2666 out:
2667         btrfs_free_path(path);
2668         return ret;
2669 }
2670
2671 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2672                       struct btrfs_device *device, u64 new_size)
2673 {
2674         struct btrfs_super_block *super_copy =
2675                 device->dev_root->fs_info->super_copy;
2676         struct btrfs_fs_devices *fs_devices;
2677         u64 old_total;
2678         u64 diff;
2679
2680         if (!device->writeable)
2681                 return -EACCES;
2682
2683         lock_chunks(device->dev_root);
2684         old_total = btrfs_super_total_bytes(super_copy);
2685         diff = new_size - device->total_bytes;
2686
2687         if (new_size <= device->total_bytes ||
2688             device->is_tgtdev_for_dev_replace) {
2689                 unlock_chunks(device->dev_root);
2690                 return -EINVAL;
2691         }
2692
2693         fs_devices = device->dev_root->fs_info->fs_devices;
2694
2695         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2696         device->fs_devices->total_rw_bytes += diff;
2697
2698         btrfs_device_set_total_bytes(device, new_size);
2699         btrfs_device_set_disk_total_bytes(device, new_size);
2700         btrfs_clear_space_info_full(device->dev_root->fs_info);
2701         if (list_empty(&device->resized_list))
2702                 list_add_tail(&device->resized_list,
2703                               &fs_devices->resized_devices);
2704         unlock_chunks(device->dev_root);
2705
2706         return btrfs_update_device(trans, device);
2707 }
2708
2709 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2710                             struct btrfs_root *root, u64 chunk_objectid,
2711                             u64 chunk_offset)
2712 {
2713         int ret;
2714         struct btrfs_path *path;
2715         struct btrfs_key key;
2716
2717         root = root->fs_info->chunk_root;
2718         path = btrfs_alloc_path();
2719         if (!path)
2720                 return -ENOMEM;
2721
2722         key.objectid = chunk_objectid;
2723         key.offset = chunk_offset;
2724         key.type = BTRFS_CHUNK_ITEM_KEY;
2725
2726         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2727         if (ret < 0)
2728                 goto out;
2729         else if (ret > 0) { /* Logic error or corruption */
2730                 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2731                             "Failed lookup while freeing chunk.");
2732                 ret = -ENOENT;
2733                 goto out;
2734         }
2735
2736         ret = btrfs_del_item(trans, root, path);
2737         if (ret < 0)
2738                 btrfs_handle_fs_error(root->fs_info, ret,
2739                             "Failed to delete chunk item.");
2740 out:
2741         btrfs_free_path(path);
2742         return ret;
2743 }
2744
2745 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2746                         chunk_offset)
2747 {
2748         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2749         struct btrfs_disk_key *disk_key;
2750         struct btrfs_chunk *chunk;
2751         u8 *ptr;
2752         int ret = 0;
2753         u32 num_stripes;
2754         u32 array_size;
2755         u32 len = 0;
2756         u32 cur;
2757         struct btrfs_key key;
2758
2759         lock_chunks(root);
2760         array_size = btrfs_super_sys_array_size(super_copy);
2761
2762         ptr = super_copy->sys_chunk_array;
2763         cur = 0;
2764
2765         while (cur < array_size) {
2766                 disk_key = (struct btrfs_disk_key *)ptr;
2767                 btrfs_disk_key_to_cpu(&key, disk_key);
2768
2769                 len = sizeof(*disk_key);
2770
2771                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2772                         chunk = (struct btrfs_chunk *)(ptr + len);
2773                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2774                         len += btrfs_chunk_item_size(num_stripes);
2775                 } else {
2776                         ret = -EIO;
2777                         break;
2778                 }
2779                 if (key.objectid == chunk_objectid &&
2780                     key.offset == chunk_offset) {
2781                         memmove(ptr, ptr + len, array_size - (cur + len));
2782                         array_size -= len;
2783                         btrfs_set_super_sys_array_size(super_copy, array_size);
2784                 } else {
2785                         ptr += len;
2786                         cur += len;
2787                 }
2788         }
2789         unlock_chunks(root);
2790         return ret;
2791 }
2792
2793 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2794                        struct btrfs_root *root, u64 chunk_offset)
2795 {
2796         struct extent_map_tree *em_tree;
2797         struct extent_map *em;
2798         struct btrfs_root *extent_root = root->fs_info->extent_root;
2799         struct map_lookup *map;
2800         u64 dev_extent_len = 0;
2801         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2802         int i, ret = 0;
2803         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2804
2805         /* Just in case */
2806         root = root->fs_info->chunk_root;
2807         em_tree = &root->fs_info->mapping_tree.map_tree;
2808
2809         read_lock(&em_tree->lock);
2810         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2811         read_unlock(&em_tree->lock);
2812
2813         if (!em || em->start > chunk_offset ||
2814             em->start + em->len < chunk_offset) {
2815                 /*
2816                  * This is a logic error, but we don't want to just rely on the
2817                  * user having built with ASSERT enabled, so if ASSERT doesn't
2818                  * do anything we still error out.
2819                  */
2820                 ASSERT(0);
2821                 if (em)
2822                         free_extent_map(em);
2823                 return -EINVAL;
2824         }
2825         map = em->map_lookup;
2826         lock_chunks(root->fs_info->chunk_root);
2827         check_system_chunk(trans, extent_root, map->type);
2828         unlock_chunks(root->fs_info->chunk_root);
2829
2830         /*
2831          * Take the device list mutex to prevent races with the final phase of
2832          * a device replace operation that replaces the device object associated
2833          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2834          */
2835         mutex_lock(&fs_devices->device_list_mutex);
2836         for (i = 0; i < map->num_stripes; i++) {
2837                 struct btrfs_device *device = map->stripes[i].dev;
2838                 ret = btrfs_free_dev_extent(trans, device,
2839                                             map->stripes[i].physical,
2840                                             &dev_extent_len);
2841                 if (ret) {
2842                         mutex_unlock(&fs_devices->device_list_mutex);
2843                         btrfs_abort_transaction(trans, ret);
2844                         goto out;
2845                 }
2846
2847                 if (device->bytes_used > 0) {
2848                         lock_chunks(root);
2849                         btrfs_device_set_bytes_used(device,
2850                                         device->bytes_used - dev_extent_len);
2851                         spin_lock(&root->fs_info->free_chunk_lock);
2852                         root->fs_info->free_chunk_space += dev_extent_len;
2853                         spin_unlock(&root->fs_info->free_chunk_lock);
2854                         btrfs_clear_space_info_full(root->fs_info);
2855                         unlock_chunks(root);
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, root, chunk_objectid, chunk_offset);
2870         if (ret) {
2871                 btrfs_abort_transaction(trans, ret);
2872                 goto out;
2873         }
2874
2875         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2876
2877         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2878                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2879                 if (ret) {
2880                         btrfs_abort_transaction(trans, ret);
2881                         goto out;
2882                 }
2883         }
2884
2885         ret = btrfs_remove_block_group(trans, extent_root, 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_root *root, u64 chunk_offset)
2898 {
2899         struct btrfs_root *extent_root;
2900         struct btrfs_trans_handle *trans;
2901         int ret;
2902
2903         root = root->fs_info->chunk_root;
2904         extent_root = root->fs_info->extent_root;
2905
2906         /*
2907          * Prevent races with automatic removal of unused block groups.
2908          * After we relocate and before we remove the chunk with offset
2909          * chunk_offset, automatic removal of the block group can kick in,
2910          * resulting in a failure when calling btrfs_remove_chunk() below.
2911          *
2912          * Make sure to acquire this mutex before doing a tree search (dev
2913          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915          * we release the path used to search the chunk/dev tree and before
2916          * the current task acquires this mutex and calls us.
2917          */
2918         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2919
2920         ret = btrfs_can_relocate(extent_root, chunk_offset);
2921         if (ret)
2922                 return -ENOSPC;
2923
2924         /* step one, relocate all the extents inside this chunk */
2925         btrfs_scrub_pause(root);
2926         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2927         btrfs_scrub_continue(root);
2928         if (ret)
2929                 return ret;
2930
2931         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2932                                                      chunk_offset);
2933         if (IS_ERR(trans)) {
2934                 ret = PTR_ERR(trans);
2935                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2936                 return ret;
2937         }
2938
2939         /*
2940          * step two, delete the device extents and the
2941          * chunk tree entries
2942          */
2943         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2944         btrfs_end_transaction(trans, extent_root);
2945         return ret;
2946 }
2947
2948 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2949 {
2950         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2951         struct btrfs_path *path;
2952         struct extent_buffer *leaf;
2953         struct btrfs_chunk *chunk;
2954         struct btrfs_key key;
2955         struct btrfs_key found_key;
2956         u64 chunk_type;
2957         bool retried = false;
2958         int failed = 0;
2959         int ret;
2960
2961         path = btrfs_alloc_path();
2962         if (!path)
2963                 return -ENOMEM;
2964
2965 again:
2966         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967         key.offset = (u64)-1;
2968         key.type = BTRFS_CHUNK_ITEM_KEY;
2969
2970         while (1) {
2971                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2972                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2973                 if (ret < 0) {
2974                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2975                         goto error;
2976                 }
2977                 BUG_ON(ret == 0); /* Corruption */
2978
2979                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2980                                           key.type);
2981                 if (ret)
2982                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2983                 if (ret < 0)
2984                         goto error;
2985                 if (ret > 0)
2986                         break;
2987
2988                 leaf = path->nodes[0];
2989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2990
2991                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992                                        struct btrfs_chunk);
2993                 chunk_type = btrfs_chunk_type(leaf, chunk);
2994                 btrfs_release_path(path);
2995
2996                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997                         ret = btrfs_relocate_chunk(chunk_root,
2998                                                    found_key.offset);
2999                         if (ret == -ENOSPC)
3000                                 failed++;
3001                         else
3002                                 BUG_ON(ret);
3003                 }
3004                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3005
3006                 if (found_key.offset == 0)
3007                         break;
3008                 key.offset = found_key.offset - 1;
3009         }
3010         ret = 0;
3011         if (failed && !retried) {
3012                 failed = 0;
3013                 retried = true;
3014                 goto again;
3015         } else if (WARN_ON(failed && retried)) {
3016                 ret = -ENOSPC;
3017         }
3018 error:
3019         btrfs_free_path(path);
3020         return ret;
3021 }
3022
3023 static int insert_balance_item(struct btrfs_root *root,
3024                                struct btrfs_balance_control *bctl)
3025 {
3026         struct btrfs_trans_handle *trans;
3027         struct btrfs_balance_item *item;
3028         struct btrfs_disk_balance_args disk_bargs;
3029         struct btrfs_path *path;
3030         struct extent_buffer *leaf;
3031         struct btrfs_key key;
3032         int ret, err;
3033
3034         path = btrfs_alloc_path();
3035         if (!path)
3036                 return -ENOMEM;
3037
3038         trans = btrfs_start_transaction(root, 0);
3039         if (IS_ERR(trans)) {
3040                 btrfs_free_path(path);
3041                 return PTR_ERR(trans);
3042         }
3043
3044         key.objectid = BTRFS_BALANCE_OBJECTID;
3045         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3046         key.offset = 0;
3047
3048         ret = btrfs_insert_empty_item(trans, root, path, &key,
3049                                       sizeof(*item));
3050         if (ret)
3051                 goto out;
3052
3053         leaf = path->nodes[0];
3054         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3055
3056         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3057
3058         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3059         btrfs_set_balance_data(leaf, item, &disk_bargs);
3060         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3061         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3062         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3063         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3064
3065         btrfs_set_balance_flags(leaf, item, bctl->flags);
3066
3067         btrfs_mark_buffer_dirty(leaf);
3068 out:
3069         btrfs_free_path(path);
3070         err = btrfs_commit_transaction(trans, root);
3071         if (err && !ret)
3072                 ret = err;
3073         return ret;
3074 }
3075
3076 static int del_balance_item(struct btrfs_root *root)
3077 {
3078         struct btrfs_trans_handle *trans;
3079         struct btrfs_path *path;
3080         struct btrfs_key key;
3081         int ret, err;
3082
3083         path = btrfs_alloc_path();
3084         if (!path)
3085                 return -ENOMEM;
3086
3087         trans = btrfs_start_transaction(root, 0);
3088         if (IS_ERR(trans)) {
3089                 btrfs_free_path(path);
3090                 return PTR_ERR(trans);
3091         }
3092
3093         key.objectid = BTRFS_BALANCE_OBJECTID;
3094         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3095         key.offset = 0;
3096
3097         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3098         if (ret < 0)
3099                 goto out;
3100         if (ret > 0) {
3101                 ret = -ENOENT;
3102                 goto out;
3103         }
3104
3105         ret = btrfs_del_item(trans, root, path);
3106 out:
3107         btrfs_free_path(path);
3108         err = btrfs_commit_transaction(trans, root);
3109         if (err && !ret)
3110                 ret = err;
3111         return ret;
3112 }
3113
3114 /*
3115  * This is a heuristic used to reduce the number of chunks balanced on
3116  * resume after balance was interrupted.
3117  */
3118 static void update_balance_args(struct btrfs_balance_control *bctl)
3119 {
3120         /*
3121          * Turn on soft mode for chunk types that were being converted.
3122          */
3123         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3124                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3125         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3126                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3127         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3128                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3129
3130         /*
3131          * Turn on usage filter if is not already used.  The idea is
3132          * that chunks that we have already balanced should be
3133          * reasonably full.  Don't do it for chunks that are being
3134          * converted - that will keep us from relocating unconverted
3135          * (albeit full) chunks.
3136          */
3137         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3138             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3139             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3140                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3141                 bctl->data.usage = 90;
3142         }
3143         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3144             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3145             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3146                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3147                 bctl->sys.usage = 90;
3148         }
3149         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3150             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3151             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3152                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3153                 bctl->meta.usage = 90;
3154         }
3155 }
3156
3157 /*
3158  * Should be called with both balance and volume mutexes held to
3159  * serialize other volume operations (add_dev/rm_dev/resize) with
3160  * restriper.  Same goes for unset_balance_control.
3161  */
3162 static void set_balance_control(struct btrfs_balance_control *bctl)
3163 {
3164         struct btrfs_fs_info *fs_info = bctl->fs_info;
3165
3166         BUG_ON(fs_info->balance_ctl);
3167
3168         spin_lock(&fs_info->balance_lock);
3169         fs_info->balance_ctl = bctl;
3170         spin_unlock(&fs_info->balance_lock);
3171 }
3172
3173 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3174 {
3175         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3176
3177         BUG_ON(!fs_info->balance_ctl);
3178
3179         spin_lock(&fs_info->balance_lock);
3180         fs_info->balance_ctl = NULL;
3181         spin_unlock(&fs_info->balance_lock);
3182
3183         kfree(bctl);
3184 }
3185
3186 /*
3187  * Balance filters.  Return 1 if chunk should be filtered out
3188  * (should not be balanced).
3189  */
3190 static int chunk_profiles_filter(u64 chunk_type,
3191                                  struct btrfs_balance_args *bargs)
3192 {
3193         chunk_type = chunk_to_extended(chunk_type) &
3194                                 BTRFS_EXTENDED_PROFILE_MASK;
3195
3196         if (bargs->profiles & chunk_type)
3197                 return 0;
3198
3199         return 1;
3200 }
3201
3202 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3203                               struct btrfs_balance_args *bargs)
3204 {
3205         struct btrfs_block_group_cache *cache;
3206         u64 chunk_used;
3207         u64 user_thresh_min;
3208         u64 user_thresh_max;
3209         int ret = 1;
3210
3211         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3212         chunk_used = btrfs_block_group_used(&cache->item);
3213
3214         if (bargs->usage_min == 0)
3215                 user_thresh_min = 0;
3216         else
3217                 user_thresh_min = div_factor_fine(cache->key.offset,
3218                                         bargs->usage_min);
3219
3220         if (bargs->usage_max == 0)
3221                 user_thresh_max = 1;
3222         else if (bargs->usage_max > 100)
3223                 user_thresh_max = cache->key.offset;
3224         else
3225                 user_thresh_max = div_factor_fine(cache->key.offset,
3226                                         bargs->usage_max);
3227
3228         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3229                 ret = 0;
3230
3231         btrfs_put_block_group(cache);
3232         return ret;
3233 }
3234
3235 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3236                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3237 {
3238         struct btrfs_block_group_cache *cache;
3239         u64 chunk_used, user_thresh;
3240         int ret = 1;
3241
3242         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3243      &n