~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/btrfs/volumes.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (C) 2007 Oracle.  All rights reserved.
  3  *
  4  * This program is free software; you can redistribute it and/or
  5  * modify it under the terms of the GNU General Public
  6  * License v2 as published by the Free Software Foundation.
  7  *
  8  * This program is distributed in the hope that it will be useful,
  9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11  * General Public License for more details.
 12  *
 13  * You should have received a copy of the GNU General Public
 14  * License along with this program; if not, write to the
 15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16  * Boston, MA 021110-1307, USA.
 17  */
 18 #include <linux/sched.h>
 19 #include <linux/bio.h>
 20 #include <linux/slab.h>
 21 #include <linux/buffer_head.h>
 22 #include <linux/blkdev.h>
 23 #include <linux/iocontext.h>
 24 #include <linux/capability.h>
 25 #include <linux/ratelimit.h>
 26 #include <linux/kthread.h>
 27 #include <linux/raid/pq.h>
 28 #include <linux/semaphore.h>
 29 #include <linux/uuid.h>
 30 #include <asm/div64.h>
 31 #include "ctree.h"
 32 #include "extent_map.h"
 33 #include "disk-io.h"
 34 #include "transaction.h"
 35 #include "print-tree.h"
 36 #include "volumes.h"
 37 #include "raid56.h"
 38 #include "async-thread.h"
 39 #include "check-integrity.h"
 40 #include "rcu-string.h"
 41 #include "math.h"
 42 #include "dev-replace.h"
 43 #include "sysfs.h"
 44 
 45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
 46         [BTRFS_RAID_RAID10] = {
 47                 .sub_stripes    = 2,
 48                 .dev_stripes    = 1,
 49                 .devs_max       = 0,    /* 0 == as many as possible */
 50                 .devs_min       = 4,
 51                 .tolerated_failures = 1,
 52                 .devs_increment = 2,
 53                 .ncopies        = 2,
 54         },
 55         [BTRFS_RAID_RAID1] = {
 56                 .sub_stripes    = 1,
 57                 .dev_stripes    = 1,
 58                 .devs_max       = 2,
 59                 .devs_min       = 2,
 60                 .tolerated_failures = 1,
 61                 .devs_increment = 2,
 62                 .ncopies        = 2,
 63         },
 64         [BTRFS_RAID_DUP] = {
 65                 .sub_stripes    = 1,
 66                 .dev_stripes    = 2,
 67                 .devs_max       = 1,
 68                 .devs_min       = 1,
 69                 .tolerated_failures = 0,
 70                 .devs_increment = 1,
 71                 .ncopies        = 2,
 72         },
 73         [BTRFS_RAID_RAID0] = {
 74                 .sub_stripes    = 1,
 75                 .dev_stripes    = 1,
 76                 .devs_max       = 0,
 77                 .devs_min       = 2,
 78                 .tolerated_failures = 0,
 79                 .devs_increment = 1,
 80                 .ncopies        = 1,
 81         },
 82         [BTRFS_RAID_SINGLE] = {
 83                 .sub_stripes    = 1,
 84                 .dev_stripes    = 1,
 85                 .devs_max       = 1,
 86                 .devs_min       = 1,
 87                 .tolerated_failures = 0,
 88                 .devs_increment = 1,
 89                 .ncopies        = 1,
 90         },
 91         [BTRFS_RAID_RAID5] = {
 92                 .sub_stripes    = 1,
 93                 .dev_stripes    = 1,
 94                 .devs_max       = 0,
 95                 .devs_min       = 2,
 96                 .tolerated_failures = 1,
 97                 .devs_increment = 1,
 98                 .ncopies        = 2,
 99         },
100         [BTRFS_RAID_RAID6] = {
101                 .sub_stripes    = 1,
102                 .dev_stripes    = 1,
103                 .devs_max       = 0,
104                 .devs_min       = 3,
105                 .tolerated_failures = 2,
106                 .devs_increment = 1,
107                 .ncopies        = 3,
108         },
109 };
110 
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112         [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113         [BTRFS_RAID_RAID1]  = BTRFS_BLOCK_GROUP_RAID1,
114         [BTRFS_RAID_DUP]    = BTRFS_BLOCK_GROUP_DUP,
115         [BTRFS_RAID_RAID0]  = BTRFS_BLOCK_GROUP_RAID0,
116         [BTRFS_RAID_SINGLE] = 0,
117         [BTRFS_RAID_RAID5]  = BTRFS_BLOCK_GROUP_RAID5,
118         [BTRFS_RAID_RAID6]  = BTRFS_BLOCK_GROUP_RAID6,
119 };
120 
121 /*
122  * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123  * condition is not met. Zero means there's no corresponding
124  * BTRFS_ERROR_DEV_*_NOT_MET value.
125  */
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127         [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128         [BTRFS_RAID_RAID1]  = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129         [BTRFS_RAID_DUP]    = 0,
130         [BTRFS_RAID_RAID0]  = 0,
131         [BTRFS_RAID_SINGLE] = 0,
132         [BTRFS_RAID_RAID5]  = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133         [BTRFS_RAID_RAID6]  = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134 };
135 
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137                                 struct btrfs_fs_info *fs_info);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
143                              enum btrfs_map_op op,
144                              u64 logical, u64 *length,
145                              struct btrfs_bio **bbio_ret,
146                              int mirror_num, int need_raid_map);
147 
148 DEFINE_MUTEX(uuid_mutex);
149 static LIST_HEAD(fs_uuids);
150 struct list_head *btrfs_get_fs_uuids(void)
151 {
152         return &fs_uuids;
153 }
154 
155 /*
156  * alloc_fs_devices - allocate struct btrfs_fs_devices
157  * @fsid:       if not NULL, copy the uuid to fs_devices::fsid
158  *
159  * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160  * The returned struct is not linked onto any lists and can be destroyed with
161  * kfree() right away.
162  */
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
164 {
165         struct btrfs_fs_devices *fs_devs;
166 
167         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
168         if (!fs_devs)
169                 return ERR_PTR(-ENOMEM);
170 
171         mutex_init(&fs_devs->device_list_mutex);
172 
173         INIT_LIST_HEAD(&fs_devs->devices);
174         INIT_LIST_HEAD(&fs_devs->resized_devices);
175         INIT_LIST_HEAD(&fs_devs->alloc_list);
176         INIT_LIST_HEAD(&fs_devs->list);
177         if (fsid)
178                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
179 
180         return fs_devs;
181 }
182 
183 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
184 {
185         struct btrfs_device *device;
186         WARN_ON(fs_devices->opened);
187         while (!list_empty(&fs_devices->devices)) {
188                 device = list_entry(fs_devices->devices.next,
189                                     struct btrfs_device, dev_list);
190                 list_del(&device->dev_list);
191                 rcu_string_free(device->name);
192                 bio_put(device->flush_bio);
193                 kfree(device);
194         }
195         kfree(fs_devices);
196 }
197 
198 static void btrfs_kobject_uevent(struct block_device *bdev,
199                                  enum kobject_action action)
200 {
201         int ret;
202 
203         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
204         if (ret)
205                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
206                         action,
207                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
208                         &disk_to_dev(bdev->bd_disk)->kobj);
209 }
210 
211 void btrfs_cleanup_fs_uuids(void)
212 {
213         struct btrfs_fs_devices *fs_devices;
214 
215         while (!list_empty(&fs_uuids)) {
216                 fs_devices = list_entry(fs_uuids.next,
217                                         struct btrfs_fs_devices, list);
218                 list_del(&fs_devices->list);
219                 free_fs_devices(fs_devices);
220         }
221 }
222 
223 static struct btrfs_device *__alloc_device(void)
224 {
225         struct btrfs_device *dev;
226 
227         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
228         if (!dev)
229                 return ERR_PTR(-ENOMEM);
230 
231         /*
232          * Preallocate a bio that's always going to be used for flushing device
233          * barriers and matches the device lifespan
234          */
235         dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
236         if (!dev->flush_bio) {
237                 kfree(dev);
238                 return ERR_PTR(-ENOMEM);
239         }
240 
241         INIT_LIST_HEAD(&dev->dev_list);
242         INIT_LIST_HEAD(&dev->dev_alloc_list);
243         INIT_LIST_HEAD(&dev->resized_list);
244 
245         spin_lock_init(&dev->io_lock);
246 
247         spin_lock_init(&dev->reada_lock);
248         atomic_set(&dev->reada_in_flight, 0);
249         atomic_set(&dev->dev_stats_ccnt, 0);
250         btrfs_device_data_ordered_init(dev);
251         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253 
254         return dev;
255 }
256 
257 /*
258  * Find a device specified by @devid or @uuid in the list of @fs_devices, or
259  * return NULL.
260  *
261  * If devid and uuid are both specified, the match must be exact, otherwise
262  * only devid is used.
263  */
264 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
265                 u64 devid, const u8 *uuid)
266 {
267         struct list_head *head = &fs_devices->devices;
268         struct btrfs_device *dev;
269 
270         list_for_each_entry(dev, head, dev_list) {
271                 if (dev->devid == devid &&
272                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
273                         return dev;
274                 }
275         }
276         return NULL;
277 }
278 
279 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
280 {
281         struct btrfs_fs_devices *fs_devices;
282 
283         list_for_each_entry(fs_devices, &fs_uuids, list) {
284                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
285                         return fs_devices;
286         }
287         return NULL;
288 }
289 
290 static int
291 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
292                       int flush, struct block_device **bdev,
293                       struct buffer_head **bh)
294 {
295         int ret;
296 
297         *bdev = blkdev_get_by_path(device_path, flags, holder);
298 
299         if (IS_ERR(*bdev)) {
300                 ret = PTR_ERR(*bdev);
301                 goto error;
302         }
303 
304         if (flush)
305                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
306         ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
307         if (ret) {
308                 blkdev_put(*bdev, flags);
309                 goto error;
310         }
311         invalidate_bdev(*bdev);
312         *bh = btrfs_read_dev_super(*bdev);
313         if (IS_ERR(*bh)) {
314                 ret = PTR_ERR(*bh);
315                 blkdev_put(*bdev, flags);
316                 goto error;
317         }
318 
319         return 0;
320 
321 error:
322         *bdev = NULL;
323         *bh = NULL;
324         return ret;
325 }
326 
327 static void requeue_list(struct btrfs_pending_bios *pending_bios,
328                         struct bio *head, struct bio *tail)
329 {
330 
331         struct bio *old_head;
332 
333         old_head = pending_bios->head;
334         pending_bios->head = head;
335         if (pending_bios->tail)
336                 tail->bi_next = old_head;
337         else
338                 pending_bios->tail = tail;
339 }
340 
341 /*
342  * we try to collect pending bios for a device so we don't get a large
343  * number of procs sending bios down to the same device.  This greatly
344  * improves the schedulers ability to collect and merge the bios.
345  *
346  * But, it also turns into a long list of bios to process and that is sure
347  * to eventually make the worker thread block.  The solution here is to
348  * make some progress and then put this work struct back at the end of
349  * the list if the block device is congested.  This way, multiple devices
350  * can make progress from a single worker thread.
351  */
352 static noinline void run_scheduled_bios(struct btrfs_device *device)
353 {
354         struct btrfs_fs_info *fs_info = device->fs_info;
355         struct bio *pending;
356         struct backing_dev_info *bdi;
357         struct btrfs_pending_bios *pending_bios;
358         struct bio *tail;
359         struct bio *cur;
360         int again = 0;
361         unsigned long num_run;
362         unsigned long batch_run = 0;
363         unsigned long last_waited = 0;
364         int force_reg = 0;
365         int sync_pending = 0;
366         struct blk_plug plug;
367 
368         /*
369          * this function runs all the bios we've collected for
370          * a particular device.  We don't want to wander off to
371          * another device without first sending all of these down.
372          * So, setup a plug here and finish it off before we return
373          */
374         blk_start_plug(&plug);
375 
376         bdi = device->bdev->bd_bdi;
377 
378 loop:
379         spin_lock(&device->io_lock);
380 
381 loop_lock:
382         num_run = 0;
383 
384         /* take all the bios off the list at once and process them
385          * later on (without the lock held).  But, remember the
386          * tail and other pointers so the bios can be properly reinserted
387          * into the list if we hit congestion
388          */
389         if (!force_reg && device->pending_sync_bios.head) {
390                 pending_bios = &device->pending_sync_bios;
391                 force_reg = 1;
392         } else {
393                 pending_bios = &device->pending_bios;
394                 force_reg = 0;
395         }
396 
397         pending = pending_bios->head;
398         tail = pending_bios->tail;
399         WARN_ON(pending && !tail);
400 
401         /*
402          * if pending was null this time around, no bios need processing
403          * at all and we can stop.  Otherwise it'll loop back up again
404          * and do an additional check so no bios are missed.
405          *
406          * device->running_pending is used to synchronize with the
407          * schedule_bio code.
408          */
409         if (device->pending_sync_bios.head == NULL &&
410             device->pending_bios.head == NULL) {
411                 again = 0;
412                 device->running_pending = 0;
413         } else {
414                 again = 1;
415                 device->running_pending = 1;
416         }
417 
418         pending_bios->head = NULL;
419         pending_bios->tail = NULL;
420 
421         spin_unlock(&device->io_lock);
422 
423         while (pending) {
424 
425                 rmb();
426                 /* we want to work on both lists, but do more bios on the
427                  * sync list than the regular list
428                  */
429                 if ((num_run > 32 &&
430                     pending_bios != &device->pending_sync_bios &&
431                     device->pending_sync_bios.head) ||
432                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
433                     device->pending_bios.head)) {
434                         spin_lock(&device->io_lock);
435                         requeue_list(pending_bios, pending, tail);
436                         goto loop_lock;
437                 }
438 
439                 cur = pending;
440                 pending = pending->bi_next;
441                 cur->bi_next = NULL;
442 
443                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
444 
445                 /*
446                  * if we're doing the sync list, record that our
447                  * plug has some sync requests on it
448                  *
449                  * If we're doing the regular list and there are
450                  * sync requests sitting around, unplug before
451                  * we add more
452                  */
453                 if (pending_bios == &device->pending_sync_bios) {
454                         sync_pending = 1;
455                 } else if (sync_pending) {
456                         blk_finish_plug(&plug);
457                         blk_start_plug(&plug);
458                         sync_pending = 0;
459                 }
460 
461                 btrfsic_submit_bio(cur);
462                 num_run++;
463                 batch_run++;
464 
465                 cond_resched();
466 
467                 /*
468                  * we made progress, there is more work to do and the bdi
469                  * is now congested.  Back off and let other work structs
470                  * run instead
471                  */
472                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
473                     fs_info->fs_devices->open_devices > 1) {
474                         struct io_context *ioc;
475 
476                         ioc = current->io_context;
477 
478                         /*
479                          * the main goal here is that we don't want to
480                          * block if we're going to be able to submit
481                          * more requests without blocking.
482                          *
483                          * This code does two great things, it pokes into
484                          * the elevator code from a filesystem _and_
485                          * it makes assumptions about how batching works.
486                          */
487                         if (ioc && ioc->nr_batch_requests > 0 &&
488                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
489                             (last_waited == 0 ||
490                              ioc->last_waited == last_waited)) {
491                                 /*
492                                  * we want to go through our batch of
493                                  * requests and stop.  So, we copy out
494                                  * the ioc->last_waited time and test
495                                  * against it before looping
496                                  */
497                                 last_waited = ioc->last_waited;
498                                 cond_resched();
499                                 continue;
500                         }
501                         spin_lock(&device->io_lock);
502                         requeue_list(pending_bios, pending, tail);
503                         device->running_pending = 1;
504 
505                         spin_unlock(&device->io_lock);
506                         btrfs_queue_work(fs_info->submit_workers,
507                                          &device->work);
508                         goto done;
509                 }
510         }
511 
512         cond_resched();
513         if (again)
514                 goto loop;
515 
516         spin_lock(&device->io_lock);
517         if (device->pending_bios.head || device->pending_sync_bios.head)
518                 goto loop_lock;
519         spin_unlock(&device->io_lock);
520 
521 done:
522         blk_finish_plug(&plug);
523 }
524 
525 static void pending_bios_fn(struct btrfs_work *work)
526 {
527         struct btrfs_device *device;
528 
529         device = container_of(work, struct btrfs_device, work);
530         run_scheduled_bios(device);
531 }
532 
533 
534 static void btrfs_free_stale_device(struct btrfs_device *cur_dev)
535 {
536         struct btrfs_fs_devices *fs_devs;
537         struct btrfs_device *dev;
538 
539         if (!cur_dev->name)
540                 return;
541 
542         list_for_each_entry(fs_devs, &fs_uuids, list) {
543                 int del = 1;
544 
545                 if (fs_devs->opened)
546                         continue;
547                 if (fs_devs->seeding)
548                         continue;
549 
550                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
551 
552                         if (dev == cur_dev)
553                                 continue;
554                         if (!dev->name)
555                                 continue;
556 
557                         /*
558                          * Todo: This won't be enough. What if the same device
559                          * comes back (with new uuid and) with its mapper path?
560                          * But for now, this does help as mostly an admin will
561                          * either use mapper or non mapper path throughout.
562                          */
563                         rcu_read_lock();
564                         del = strcmp(rcu_str_deref(dev->name),
565                                                 rcu_str_deref(cur_dev->name));
566                         rcu_read_unlock();
567                         if (!del)
568                                 break;
569                 }
570 
571                 if (!del) {
572                         /* delete the stale device */
573                         if (fs_devs->num_devices == 1) {
574                                 btrfs_sysfs_remove_fsid(fs_devs);
575                                 list_del(&fs_devs->list);
576                                 free_fs_devices(fs_devs);
577                                 break;
578                         } else {
579                                 fs_devs->num_devices--;
580                                 list_del(&dev->dev_list);
581                                 rcu_string_free(dev->name);
582                                 bio_put(dev->flush_bio);
583                                 kfree(dev);
584                         }
585                         break;
586                 }
587         }
588 }
589 
590 /*
591  * Add new device to list of registered devices
592  *
593  * Returns:
594  * 1   - first time device is seen
595  * 0   - device already known
596  * < 0 - error
597  */
598 static noinline int device_list_add(const char *path,
599                            struct btrfs_super_block *disk_super,
600                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
601 {
602         struct btrfs_device *device;
603         struct btrfs_fs_devices *fs_devices;
604         struct rcu_string *name;
605         int ret = 0;
606         u64 found_transid = btrfs_super_generation(disk_super);
607 
608         fs_devices = find_fsid(disk_super->fsid);
609         if (!fs_devices) {
610                 fs_devices = alloc_fs_devices(disk_super->fsid);
611                 if (IS_ERR(fs_devices))
612                         return PTR_ERR(fs_devices);
613 
614                 list_add(&fs_devices->list, &fs_uuids);
615 
616                 device = NULL;
617         } else {
618                 device = find_device(fs_devices, devid,
619                                 disk_super->dev_item.uuid);
620         }
621 
622         if (!device) {
623                 if (fs_devices->opened)
624                         return -EBUSY;
625 
626                 device = btrfs_alloc_device(NULL, &devid,
627                                             disk_super->dev_item.uuid);
628                 if (IS_ERR(device)) {
629                         /* we can safely leave the fs_devices entry around */
630                         return PTR_ERR(device);
631                 }
632 
633                 name = rcu_string_strdup(path, GFP_NOFS);
634                 if (!name) {
635                         bio_put(device->flush_bio);
636                         kfree(device);
637                         return -ENOMEM;
638                 }
639                 rcu_assign_pointer(device->name, name);
640 
641                 mutex_lock(&fs_devices->device_list_mutex);
642                 list_add_rcu(&device->dev_list, &fs_devices->devices);
643                 fs_devices->num_devices++;
644                 mutex_unlock(&fs_devices->device_list_mutex);
645 
646                 ret = 1;
647                 device->fs_devices = fs_devices;
648         } else if (!device->name || strcmp(device->name->str, path)) {
649                 /*
650                  * When FS is already mounted.
651                  * 1. If you are here and if the device->name is NULL that
652                  *    means this device was missing at time of FS mount.
653                  * 2. If you are here and if the device->name is different
654                  *    from 'path' that means either
655                  *      a. The same device disappeared and reappeared with
656                  *         different name. or
657                  *      b. The missing-disk-which-was-replaced, has
658                  *         reappeared now.
659                  *
660                  * We must allow 1 and 2a above. But 2b would be a spurious
661                  * and unintentional.
662                  *
663                  * Further in case of 1 and 2a above, the disk at 'path'
664                  * would have missed some transaction when it was away and
665                  * in case of 2a the stale bdev has to be updated as well.
666                  * 2b must not be allowed at all time.
667                  */
668 
669                 /*
670                  * For now, we do allow update to btrfs_fs_device through the
671                  * btrfs dev scan cli after FS has been mounted.  We're still
672                  * tracking a problem where systems fail mount by subvolume id
673                  * when we reject replacement on a mounted FS.
674                  */
675                 if (!fs_devices->opened && found_transid < device->generation) {
676                         /*
677                          * That is if the FS is _not_ mounted and if you
678                          * are here, that means there is more than one
679                          * disk with same uuid and devid.We keep the one
680                          * with larger generation number or the last-in if
681                          * generation are equal.
682                          */
683                         return -EEXIST;
684                 }
685 
686                 name = rcu_string_strdup(path, GFP_NOFS);
687                 if (!name)
688                         return -ENOMEM;
689                 rcu_string_free(device->name);
690                 rcu_assign_pointer(device->name, name);
691                 if (device->missing) {
692                         fs_devices->missing_devices--;
693                         device->missing = 0;
694                 }
695         }
696 
697         /*
698          * Unmount does not free the btrfs_device struct but would zero
699          * generation along with most of the other members. So just update
700          * it back. We need it to pick the disk with largest generation
701          * (as above).
702          */
703         if (!fs_devices->opened)
704                 device->generation = found_transid;
705 
706         /*
707          * if there is new btrfs on an already registered device,
708          * then remove the stale device entry.
709          */
710         if (ret > 0)
711                 btrfs_free_stale_device(device);
712 
713         *fs_devices_ret = fs_devices;
714 
715         return ret;
716 }
717 
718 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
719 {
720         struct btrfs_fs_devices *fs_devices;
721         struct btrfs_device *device;
722         struct btrfs_device *orig_dev;
723 
724         fs_devices = alloc_fs_devices(orig->fsid);
725         if (IS_ERR(fs_devices))
726                 return fs_devices;
727 
728         mutex_lock(&orig->device_list_mutex);
729         fs_devices->total_devices = orig->total_devices;
730 
731         /* We have held the volume lock, it is safe to get the devices. */
732         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
733                 struct rcu_string *name;
734 
735                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
736                                             orig_dev->uuid);
737                 if (IS_ERR(device))
738                         goto error;
739 
740                 /*
741                  * This is ok to do without rcu read locked because we hold the
742                  * uuid mutex so nothing we touch in here is going to disappear.
743                  */
744                 if (orig_dev->name) {
745                         name = rcu_string_strdup(orig_dev->name->str,
746                                         GFP_KERNEL);
747                         if (!name) {
748                                 bio_put(device->flush_bio);
749                                 kfree(device);
750                                 goto error;
751                         }
752                         rcu_assign_pointer(device->name, name);
753                 }
754 
755                 list_add(&device->dev_list, &fs_devices->devices);
756                 device->fs_devices = fs_devices;
757                 fs_devices->num_devices++;
758         }
759         mutex_unlock(&orig->device_list_mutex);
760         return fs_devices;
761 error:
762         mutex_unlock(&orig->device_list_mutex);
763         free_fs_devices(fs_devices);
764         return ERR_PTR(-ENOMEM);
765 }
766 
767 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
768 {
769         struct btrfs_device *device, *next;
770         struct btrfs_device *latest_dev = NULL;
771 
772         mutex_lock(&uuid_mutex);
773 again:
774         /* This is the initialized path, it is safe to release the devices. */
775         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
776                 if (device->in_fs_metadata) {
777                         if (!device->is_tgtdev_for_dev_replace &&
778                             (!latest_dev ||
779                              device->generation > latest_dev->generation)) {
780                                 latest_dev = device;
781                         }
782                         continue;
783                 }
784 
785                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
786                         /*
787                          * In the first step, keep the device which has
788                          * the correct fsid and the devid that is used
789                          * for the dev_replace procedure.
790                          * In the second step, the dev_replace state is
791                          * read from the device tree and it is known
792                          * whether the procedure is really active or
793                          * not, which means whether this device is
794                          * used or whether it should be removed.
795                          */
796                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
797                                 continue;
798                         }
799                 }
800                 if (device->bdev) {
801                         blkdev_put(device->bdev, device->mode);
802                         device->bdev = NULL;
803                         fs_devices->open_devices--;
804                 }
805                 if (device->writeable) {
806                         list_del_init(&device->dev_alloc_list);
807                         device->writeable = 0;
808                         if (!device->is_tgtdev_for_dev_replace)
809                                 fs_devices->rw_devices--;
810                 }
811                 list_del_init(&device->dev_list);
812                 fs_devices->num_devices--;
813                 rcu_string_free(device->name);
814                 bio_put(device->flush_bio);
815                 kfree(device);
816         }
817 
818         if (fs_devices->seed) {
819                 fs_devices = fs_devices->seed;
820                 goto again;
821         }
822 
823         fs_devices->latest_bdev = latest_dev->bdev;
824 
825         mutex_unlock(&uuid_mutex);
826 }
827 
828 static void __free_device(struct work_struct *work)
829 {
830         struct btrfs_device *device;
831 
832         device = container_of(work, struct btrfs_device, rcu_work);
833         rcu_string_free(device->name);
834         bio_put(device->flush_bio);
835         kfree(device);
836 }
837 
838 static void free_device(struct rcu_head *head)
839 {
840         struct btrfs_device *device;
841 
842         device = container_of(head, struct btrfs_device, rcu);
843 
844         INIT_WORK(&device->rcu_work, __free_device);
845         schedule_work(&device->rcu_work);
846 }
847 
848 static void btrfs_close_bdev(struct btrfs_device *device)
849 {
850         if (device->bdev && device->writeable) {
851                 sync_blockdev(device->bdev);
852                 invalidate_bdev(device->bdev);
853         }
854 
855         if (device->bdev)
856                 blkdev_put(device->bdev, device->mode);
857 }
858 
859 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
860 {
861         struct btrfs_fs_devices *fs_devices = device->fs_devices;
862         struct btrfs_device *new_device;
863         struct rcu_string *name;
864 
865         if (device->bdev)
866                 fs_devices->open_devices--;
867 
868         if (device->writeable &&
869             device->devid != BTRFS_DEV_REPLACE_DEVID) {
870                 list_del_init(&device->dev_alloc_list);
871                 fs_devices->rw_devices--;
872         }
873 
874         if (device->missing)
875                 fs_devices->missing_devices--;
876 
877         new_device = btrfs_alloc_device(NULL, &device->devid,
878                                         device->uuid);
879         BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
880 
881         /* Safe because we are under uuid_mutex */
882         if (device->name) {
883                 name = rcu_string_strdup(device->name->str, GFP_NOFS);
884                 BUG_ON(!name); /* -ENOMEM */
885                 rcu_assign_pointer(new_device->name, name);
886         }
887 
888         list_replace_rcu(&device->dev_list, &new_device->dev_list);
889         new_device->fs_devices = device->fs_devices;
890 }
891 
892 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
893 {
894         struct btrfs_device *device, *tmp;
895         struct list_head pending_put;
896 
897         INIT_LIST_HEAD(&pending_put);
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_prepare_close_one_device(device);
905                 list_add(&device->dev_list, &pending_put);
906         }
907         mutex_unlock(&fs_devices->device_list_mutex);
908 
909         /*
910          * btrfs_show_devname() is using the device_list_mutex,
911          * sometimes call to blkdev_put() leads vfs calling
912          * into this func. So do put outside of device_list_mutex,
913          * as of now.
914          */
915         while (!list_empty(&pending_put)) {
916                 device = list_first_entry(&pending_put,
917                                 struct btrfs_device, dev_list);
918                 list_del(&device->dev_list);
919                 btrfs_close_bdev(device);
920                 call_rcu(&device->rcu, free_device);
921         }
922 
923         WARN_ON(fs_devices->open_devices);
924         WARN_ON(fs_devices->rw_devices);
925         fs_devices->opened = 0;
926         fs_devices->seeding = 0;
927 
928         return 0;
929 }
930 
931 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
932 {
933         struct btrfs_fs_devices *seed_devices = NULL;
934         int ret;
935 
936         mutex_lock(&uuid_mutex);
937         ret = __btrfs_close_devices(fs_devices);
938         if (!fs_devices->opened) {
939                 seed_devices = fs_devices->seed;
940                 fs_devices->seed = NULL;
941         }
942         mutex_unlock(&uuid_mutex);
943 
944         while (seed_devices) {
945                 fs_devices = seed_devices;
946                 seed_devices = fs_devices->seed;
947                 __btrfs_close_devices(fs_devices);
948                 free_fs_devices(fs_devices);
949         }
950         /*
951          * Wait for rcu kworkers under __btrfs_close_devices
952          * to finish all blkdev_puts so device is really
953          * free when umount is done.
954          */
955         rcu_barrier();
956         return ret;
957 }
958 
959 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
960                                 fmode_t flags, void *holder)
961 {
962         struct request_queue *q;
963         struct block_device *bdev;
964         struct list_head *head = &fs_devices->devices;
965         struct btrfs_device *device;
966         struct btrfs_device *latest_dev = NULL;
967         struct buffer_head *bh;
968         struct btrfs_super_block *disk_super;
969         u64 devid;
970         int seeding = 1;
971         int ret = 0;
972 
973         flags |= FMODE_EXCL;
974 
975         list_for_each_entry(device, head, dev_list) {
976                 if (device->bdev)
977                         continue;
978                 if (!device->name)
979                         continue;
980 
981                 /* Just open everything we can; ignore failures here */
982                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
983                                             &bdev, &bh))
984                         continue;
985 
986                 disk_super = (struct btrfs_super_block *)bh->b_data;
987                 devid = btrfs_stack_device_id(&disk_super->dev_item);
988                 if (devid != device->devid)
989                         goto error_brelse;
990 
991                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
992                            BTRFS_UUID_SIZE))
993                         goto error_brelse;
994 
995                 device->generation = btrfs_super_generation(disk_super);
996                 if (!latest_dev ||
997                     device->generation > latest_dev->generation)
998                         latest_dev = device;
999 
1000                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1001                         device->writeable = 0;
1002                 } else {
1003                         device->writeable = !bdev_read_only(bdev);
1004                         seeding = 0;
1005                 }
1006 
1007                 q = bdev_get_queue(bdev);
1008                 if (blk_queue_discard(q))
1009                         device->can_discard = 1;
1010                 if (!blk_queue_nonrot(q))
1011                         fs_devices->rotating = 1;
1012 
1013                 device->bdev = bdev;
1014                 device->in_fs_metadata = 0;
1015                 device->mode = flags;
1016 
1017                 fs_devices->open_devices++;
1018                 if (device->writeable &&
1019                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
1020                         fs_devices->rw_devices++;
1021                         list_add(&device->dev_alloc_list,
1022                                  &fs_devices->alloc_list);
1023                 }
1024                 brelse(bh);
1025                 continue;
1026 
1027 error_brelse:
1028                 brelse(bh);
1029                 blkdev_put(bdev, flags);
1030                 continue;
1031         }
1032         if (fs_devices->open_devices == 0) {
1033                 ret = -EINVAL;
1034                 goto out;
1035         }
1036         fs_devices->seeding = seeding;
1037         fs_devices->opened = 1;
1038         fs_devices->latest_bdev = latest_dev->bdev;
1039         fs_devices->total_rw_bytes = 0;
1040 out:
1041         return ret;
1042 }
1043 
1044 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1045                        fmode_t flags, void *holder)
1046 {
1047         int ret;
1048 
1049         mutex_lock(&uuid_mutex);
1050         if (fs_devices->opened) {
1051                 fs_devices->opened++;
1052                 ret = 0;
1053         } else {
1054                 ret = __btrfs_open_devices(fs_devices, flags, holder);
1055         }
1056         mutex_unlock(&uuid_mutex);
1057         return ret;
1058 }
1059 
1060 static void btrfs_release_disk_super(struct page *page)
1061 {
1062         kunmap(page);
1063         put_page(page);
1064 }
1065 
1066 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1067                                  struct page **page,
1068                                  struct btrfs_super_block **disk_super)
1069 {
1070         void *p;
1071         pgoff_t index;
1072 
1073         /* make sure our super fits in the device */
1074         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1075                 return 1;
1076 
1077         /* make sure our super fits in the page */
1078         if (sizeof(**disk_super) > PAGE_SIZE)
1079                 return 1;
1080 
1081         /* make sure our super doesn't straddle pages on disk */
1082         index = bytenr >> PAGE_SHIFT;
1083         if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1084                 return 1;
1085 
1086         /* pull in the page with our super */
1087         *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1088                                    index, GFP_KERNEL);
1089 
1090         if (IS_ERR_OR_NULL(*page))
1091                 return 1;
1092 
1093         p = kmap(*page);
1094 
1095         /* align our pointer to the offset of the super block */
1096         *disk_super = p + (bytenr & ~PAGE_MASK);
1097 
1098         if (btrfs_super_bytenr(*disk_super) != bytenr ||
1099             btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1100                 btrfs_release_disk_super(*page);
1101                 return 1;
1102         }
1103 
1104         if ((*disk_super)->label[0] &&
1105                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1106                 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1107 
1108         return 0;
1109 }
1110 
1111 /*
1112  * Look for a btrfs signature on a device. This may be called out of the mount path
1113  * and we are not allowed to call set_blocksize during the scan. The superblock
1114  * is read via pagecache
1115  */
1116 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1117                           struct btrfs_fs_devices **fs_devices_ret)
1118 {
1119         struct btrfs_super_block *disk_super;
1120         struct block_device *bdev;
1121         struct page *page;
1122         int ret = -EINVAL;
1123         u64 devid;
1124         u64 transid;
1125         u64 total_devices;
1126         u64 bytenr;
1127 
1128         /*
1129          * we would like to check all the supers, but that would make
1130          * a btrfs mount succeed after a mkfs from a different FS.
1131          * So, we need to add a special mount option to scan for
1132          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1133          */
1134         bytenr = btrfs_sb_offset(0);
1135         flags |= FMODE_EXCL;
1136         mutex_lock(&uuid_mutex);
1137 
1138         bdev = blkdev_get_by_path(path, flags, holder);
1139         if (IS_ERR(bdev)) {
1140                 ret = PTR_ERR(bdev);
1141                 goto error;
1142         }
1143 
1144         if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1145                 goto error_bdev_put;
1146 
1147         devid = btrfs_stack_device_id(&disk_super->dev_item);
1148         transid = btrfs_super_generation(disk_super);
1149         total_devices = btrfs_super_num_devices(disk_super);
1150 
1151         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1152         if (ret > 0) {
1153                 if (disk_super->label[0]) {
1154                         pr_info("BTRFS: device label %s ", disk_super->label);
1155                 } else {
1156                         pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1157                 }
1158 
1159                 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1160                 ret = 0;
1161         }
1162         if (!ret && fs_devices_ret)
1163                 (*fs_devices_ret)->total_devices = total_devices;
1164 
1165         btrfs_release_disk_super(page);
1166 
1167 error_bdev_put:
1168         blkdev_put(bdev, flags);
1169 error:
1170         mutex_unlock(&uuid_mutex);
1171         return ret;
1172 }
1173 
1174 /* helper to account the used device space in the range */
1175 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1176                                    u64 end, u64 *length)
1177 {
1178         struct btrfs_key key;
1179         struct btrfs_root *root = device->fs_info->dev_root;
1180         struct btrfs_dev_extent *dev_extent;
1181         struct btrfs_path *path;
1182         u64 extent_end;
1183         int ret;
1184         int slot;
1185         struct extent_buffer *l;
1186 
1187         *length = 0;
1188 
1189         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1190                 return 0;
1191 
1192         path = btrfs_alloc_path();
1193         if (!path)
1194                 return -ENOMEM;
1195         path->reada = READA_FORWARD;
1196 
1197         key.objectid = device->devid;
1198         key.offset = start;
1199         key.type = BTRFS_DEV_EXTENT_KEY;
1200 
1201         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1202         if (ret < 0)
1203                 goto out;
1204         if (ret > 0) {
1205                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1206                 if (ret < 0)
1207                         goto out;
1208         }
1209 
1210         while (1) {
1211                 l = path->nodes[0];
1212                 slot = path->slots[0];
1213                 if (slot >= btrfs_header_nritems(l)) {
1214                         ret = btrfs_next_leaf(root, path);
1215                         if (ret == 0)
1216                                 continue;
1217                         if (ret < 0)
1218                                 goto out;
1219 
1220                         break;
1221                 }
1222                 btrfs_item_key_to_cpu(l, &key, slot);
1223 
1224                 if (key.objectid < device->devid)
1225                         goto next;
1226 
1227                 if (key.objectid > device->devid)
1228                         break;
1229 
1230                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1231                         goto next;
1232 
1233                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1234                 extent_end = key.offset + btrfs_dev_extent_length(l,
1235                                                                   dev_extent);
1236                 if (key.offset <= start && extent_end > end) {
1237                         *length = end - start + 1;
1238                         break;
1239                 } else if (key.offset <= start && extent_end > start)
1240                         *length += extent_end - start;
1241                 else if (key.offset > start && extent_end <= end)
1242                         *length += extent_end - key.offset;
1243                 else if (key.offset > start && key.offset <= end) {
1244                         *length += end - key.offset + 1;
1245                         break;
1246                 } else if (key.offset > end)
1247                         break;
1248 
1249 next:
1250                 path->slots[0]++;
1251         }
1252         ret = 0;
1253 out:
1254         btrfs_free_path(path);
1255         return ret;
1256 }
1257 
1258 static int contains_pending_extent(struct btrfs_transaction *transaction,
1259                                    struct btrfs_device *device,
1260                                    u64 *start, u64 len)
1261 {
1262         struct btrfs_fs_info *fs_info = device->fs_info;
1263         struct extent_map *em;
1264         struct list_head *search_list = &fs_info->pinned_chunks;
1265         int ret = 0;
1266         u64 physical_start = *start;
1267 
1268         if (transaction)
1269                 search_list = &transaction->pending_chunks;
1270 again:
1271         list_for_each_entry(em, search_list, list) {
1272                 struct map_lookup *map;
1273                 int i;
1274 
1275                 map = em->map_lookup;
1276                 for (i = 0; i < map->num_stripes; i++) {
1277                         u64 end;
1278 
1279                         if (map->stripes[i].dev != device)
1280                                 continue;
1281                         if (map->stripes[i].physical >= physical_start + len ||
1282                             map->stripes[i].physical + em->orig_block_len <=
1283                             physical_start)
1284                                 continue;
1285                         /*
1286                          * Make sure that while processing the pinned list we do
1287                          * not override our *start with a lower value, because
1288                          * we can have pinned chunks that fall within this
1289                          * device hole and that have lower physical addresses
1290                          * than the pending chunks we processed before. If we
1291                          * do not take this special care we can end up getting
1292                          * 2 pending chunks that start at the same physical
1293                          * device offsets because the end offset of a pinned
1294                          * chunk can be equal to the start offset of some
1295                          * pending chunk.
1296                          */
1297                         end = map->stripes[i].physical + em->orig_block_len;
1298                         if (end > *start) {
1299                                 *start = end;
1300                                 ret = 1;
1301                         }
1302                 }
1303         }
1304         if (search_list != &fs_info->pinned_chunks) {
1305                 search_list = &fs_info->pinned_chunks;
1306                 goto again;
1307         }
1308 
1309         return ret;
1310 }
1311 
1312 
1313 /*
1314  * find_free_dev_extent_start - find free space in the specified device
1315  * @device:       the device which we search the free space in
1316  * @num_bytes:    the size of the free space that we need
1317  * @search_start: the position from which to begin the search
1318  * @start:        store the start of the free space.
1319  * @len:          the size of the free space. that we find, or the size
1320  *                of the max free space if we don't find suitable free space
1321  *
1322  * this uses a pretty simple search, the expectation is that it is
1323  * called very infrequently and that a given device has a small number
1324  * of extents
1325  *
1326  * @start is used to store the start of the free space if we find. But if we
1327  * don't find suitable free space, it will be used to store the start position
1328  * of the max free space.
1329  *
1330  * @len is used to store the size of the free space that we find.
1331  * But if we don't find suitable free space, it is used to store the size of
1332  * the max free space.
1333  */
1334 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1335                                struct btrfs_device *device, u64 num_bytes,
1336                                u64 search_start, u64 *start, u64 *len)
1337 {
1338         struct btrfs_fs_info *fs_info = device->fs_info;
1339         struct btrfs_root *root = fs_info->dev_root;
1340         struct btrfs_key key;
1341         struct btrfs_dev_extent *dev_extent;
1342         struct btrfs_path *path;
1343         u64 hole_size;
1344         u64 max_hole_start;
1345         u64 max_hole_size;
1346         u64 extent_end;
1347         u64 search_end = device->total_bytes;
1348         int ret;
1349         int slot;
1350         struct extent_buffer *l;
1351 
1352         /*
1353          * We don't want to overwrite the superblock on the drive nor any area
1354          * used by the boot loader (grub for example), so we make sure to start
1355          * at an offset of at least 1MB.
1356          */
1357         search_start = max_t(u64, search_start, SZ_1M);
1358 
1359         path = btrfs_alloc_path();
1360         if (!path)
1361                 return -ENOMEM;
1362 
1363         max_hole_start = search_start;
1364         max_hole_size = 0;
1365 
1366 again:
1367         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1368                 ret = -ENOSPC;
1369                 goto out;
1370         }
1371 
1372         path->reada = READA_FORWARD;
1373         path->search_commit_root = 1;
1374         path->skip_locking = 1;
1375 
1376         key.objectid = device->devid;
1377         key.offset = search_start;
1378         key.type = BTRFS_DEV_EXTENT_KEY;
1379 
1380         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1381         if (ret < 0)
1382                 goto out;
1383         if (ret > 0) {
1384                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1385                 if (ret < 0)
1386                         goto out;
1387         }
1388 
1389         while (1) {
1390                 l = path->nodes[0];
1391                 slot = path->slots[0];
1392                 if (slot >= btrfs_header_nritems(l)) {
1393                         ret = btrfs_next_leaf(root, path);
1394                         if (ret == 0)
1395                                 continue;
1396                         if (ret < 0)
1397                                 goto out;
1398 
1399                         break;
1400                 }
1401                 btrfs_item_key_to_cpu(l, &key, slot);
1402 
1403                 if (key.objectid < device->devid)
1404                         goto next;
1405 
1406                 if (key.objectid > device->devid)
1407                         break;
1408 
1409                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1410                         goto next;
1411 
1412                 if (key.offset > search_start) {
1413                         hole_size = key.offset - search_start;
1414 
1415                         /*
1416                          * Have to check before we set max_hole_start, otherwise
1417                          * we could end up sending back this offset anyway.
1418                          */
1419                         if (contains_pending_extent(transaction, device,
1420                                                     &search_start,
1421                                                     hole_size)) {
1422                                 if (key.offset >= search_start) {
1423                                         hole_size = key.offset - search_start;
1424                                 } else {
1425                                         WARN_ON_ONCE(1);
1426                                         hole_size = 0;
1427                                 }
1428                         }
1429 
1430                         if (hole_size > max_hole_size) {
1431                                 max_hole_start = search_start;
1432                                 max_hole_size = hole_size;
1433                         }
1434 
1435                         /*
1436                          * If this free space is greater than which we need,
1437                          * it must be the max free space that we have found
1438                          * until now, so max_hole_start must point to the start
1439                          * of this free space and the length of this free space
1440                          * is stored in max_hole_size. Thus, we return
1441                          * max_hole_start and max_hole_size and go back to the
1442                          * caller.
1443                          */
1444                         if (hole_size >= num_bytes) {
1445                                 ret = 0;
1446                                 goto out;
1447                         }
1448                 }
1449 
1450                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1451                 extent_end = key.offset + btrfs_dev_extent_length(l,
1452                                                                   dev_extent);
1453                 if (extent_end > search_start)
1454                         search_start = extent_end;
1455 next:
1456                 path->slots[0]++;
1457                 cond_resched();
1458         }
1459 
1460         /*
1461          * At this point, search_start should be the end of
1462          * allocated dev extents, and when shrinking the device,
1463          * search_end may be smaller than search_start.
1464          */
1465         if (search_end > search_start) {
1466                 hole_size = search_end - search_start;
1467 
1468                 if (contains_pending_extent(transaction, device, &search_start,
1469                                             hole_size)) {
1470                         btrfs_release_path(path);
1471                         goto again;
1472                 }
1473 
1474                 if (hole_size > max_hole_size) {
1475                         max_hole_start = search_start;
1476                         max_hole_size = hole_size;
1477                 }
1478         }
1479 
1480         /* See above. */
1481         if (max_hole_size < num_bytes)
1482                 ret = -ENOSPC;
1483         else
1484                 ret = 0;
1485 
1486 out:
1487         btrfs_free_path(path);
1488         *start = max_hole_start;
1489         if (len)
1490                 *len = max_hole_size;
1491         return ret;
1492 }
1493 
1494 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1495                          struct btrfs_device *device, u64 num_bytes,
1496                          u64 *start, u64 *len)
1497 {
1498         /* FIXME use last free of some kind */
1499         return find_free_dev_extent_start(trans->transaction, device,
1500                                           num_bytes, 0, start, len);
1501 }
1502 
1503 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1504                           struct btrfs_device *device,
1505                           u64 start, u64 *dev_extent_len)
1506 {
1507         struct btrfs_fs_info *fs_info = device->fs_info;
1508         struct btrfs_root *root = fs_info->dev_root;
1509         int ret;
1510         struct btrfs_path *path;
1511         struct btrfs_key key;
1512         struct btrfs_key found_key;
1513         struct extent_buffer *leaf = NULL;
1514         struct btrfs_dev_extent *extent = NULL;
1515 
1516         path = btrfs_alloc_path();
1517         if (!path)
1518                 return -ENOMEM;
1519 
1520         key.objectid = device->devid;
1521         key.offset = start;
1522         key.type = BTRFS_DEV_EXTENT_KEY;
1523 again:
1524         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1525         if (ret > 0) {
1526                 ret = btrfs_previous_item(root, path, key.objectid,
1527                                           BTRFS_DEV_EXTENT_KEY);
1528                 if (ret)
1529                         goto out;
1530                 leaf = path->nodes[0];
1531                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1532                 extent = btrfs_item_ptr(leaf, path->slots[0],
1533                                         struct btrfs_dev_extent);
1534                 BUG_ON(found_key.offset > start || found_key.offset +
1535                        btrfs_dev_extent_length(leaf, extent) < start);
1536                 key = found_key;
1537                 btrfs_release_path(path);
1538                 goto again;
1539         } else if (ret == 0) {
1540                 leaf = path->nodes[0];
1541                 extent = btrfs_item_ptr(leaf, path->slots[0],
1542                                         struct btrfs_dev_extent);
1543         } else {
1544                 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1545                 goto out;
1546         }
1547 
1548         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1549 
1550         ret = btrfs_del_item(trans, root, path);
1551         if (ret) {
1552                 btrfs_handle_fs_error(fs_info, ret,
1553                                       "Failed to remove dev extent item");
1554         } else {
1555                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1556         }
1557 out:
1558         btrfs_free_path(path);
1559         return ret;
1560 }
1561 
1562 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1563                                   struct btrfs_device *device,
1564                                   u64 chunk_offset, u64 start, u64 num_bytes)
1565 {
1566         int ret;
1567         struct btrfs_path *path;
1568         struct btrfs_fs_info *fs_info = device->fs_info;
1569         struct btrfs_root *root = fs_info->dev_root;
1570         struct btrfs_dev_extent *extent;
1571         struct extent_buffer *leaf;
1572         struct btrfs_key key;
1573 
1574         WARN_ON(!device->in_fs_metadata);
1575         WARN_ON(device->is_tgtdev_for_dev_replace);
1576         path = btrfs_alloc_path();
1577         if (!path)
1578                 return -ENOMEM;
1579 
1580         key.objectid = device->devid;
1581         key.offset = start;
1582         key.type = BTRFS_DEV_EXTENT_KEY;
1583         ret = btrfs_insert_empty_item(trans, root, path, &key,
1584                                       sizeof(*extent));
1585         if (ret)
1586                 goto out;
1587 
1588         leaf = path->nodes[0];
1589         extent = btrfs_item_ptr(leaf, path->slots[0],
1590                                 struct btrfs_dev_extent);
1591         btrfs_set_dev_extent_chunk_tree(leaf, extent,
1592                                         BTRFS_CHUNK_TREE_OBJECTID);
1593         btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1594                                             BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1595         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1596 
1597         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1598         btrfs_mark_buffer_dirty(leaf);
1599 out:
1600         btrfs_free_path(path);
1601         return ret;
1602 }
1603 
1604 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1605 {
1606         struct extent_map_tree *em_tree;
1607         struct extent_map *em;
1608         struct rb_node *n;
1609         u64 ret = 0;
1610 
1611         em_tree = &fs_info->mapping_tree.map_tree;
1612         read_lock(&em_tree->lock);
1613         n = rb_last(&em_tree->map);
1614         if (n) {
1615                 em = rb_entry(n, struct extent_map, rb_node);
1616                 ret = em->start + em->len;
1617         }
1618         read_unlock(&em_tree->lock);
1619 
1620         return ret;
1621 }
1622 
1623 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1624                                     u64 *devid_ret)
1625 {
1626         int ret;
1627         struct btrfs_key key;
1628         struct btrfs_key found_key;
1629         struct btrfs_path *path;
1630 
1631         path = btrfs_alloc_path();
1632         if (!path)
1633                 return -ENOMEM;
1634 
1635         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1636         key.type = BTRFS_DEV_ITEM_KEY;
1637         key.offset = (u64)-1;
1638 
1639         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1640         if (ret < 0)
1641                 goto error;
1642 
1643         BUG_ON(ret == 0); /* Corruption */
1644 
1645         ret = btrfs_previous_item(fs_info->chunk_root, path,
1646                                   BTRFS_DEV_ITEMS_OBJECTID,
1647                                   BTRFS_DEV_ITEM_KEY);
1648         if (ret) {
1649                 *devid_ret = 1;
1650         } else {
1651                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1652                                       path->slots[0]);
1653                 *devid_ret = found_key.offset + 1;
1654         }
1655         ret = 0;
1656 error:
1657         btrfs_free_path(path);
1658         return ret;
1659 }
1660 
1661 /*
1662  * the device information is stored in the chunk root
1663  * the btrfs_device struct should be fully filled in
1664  */
1665 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1666                             struct btrfs_fs_info *fs_info,
1667                             struct btrfs_device *device)
1668 {
1669         struct btrfs_root *root = fs_info->chunk_root;
1670         int ret;
1671         struct btrfs_path *path;
1672         struct btrfs_dev_item *dev_item;
1673         struct extent_buffer *leaf;
1674         struct btrfs_key key;
1675         unsigned long ptr;
1676 
1677         path = btrfs_alloc_path();
1678         if (!path)
1679                 return -ENOMEM;
1680 
1681         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1682         key.type = BTRFS_DEV_ITEM_KEY;
1683         key.offset = device->devid;
1684 
1685         ret = btrfs_insert_empty_item(trans, root, path, &key,
1686                                       sizeof(*dev_item));
1687         if (ret)
1688                 goto out;
1689 
1690         leaf = path->nodes[0];
1691         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1692 
1693         btrfs_set_device_id(leaf, dev_item, device->devid);
1694         btrfs_set_device_generation(leaf, dev_item, 0);
1695         btrfs_set_device_type(leaf, dev_item, device->type);
1696         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1697         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1698         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1699         btrfs_set_device_total_bytes(leaf, dev_item,
1700                                      btrfs_device_get_disk_total_bytes(device));
1701         btrfs_set_device_bytes_used(leaf, dev_item,
1702                                     btrfs_device_get_bytes_used(device));
1703         btrfs_set_device_group(leaf, dev_item, 0);
1704         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1705         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1706         btrfs_set_device_start_offset(leaf, dev_item, 0);
1707 
1708         ptr = btrfs_device_uuid(dev_item);
1709         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1710         ptr = btrfs_device_fsid(dev_item);
1711         write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1712         btrfs_mark_buffer_dirty(leaf);
1713 
1714         ret = 0;
1715 out:
1716         btrfs_free_path(path);
1717         return ret;
1718 }
1719 
1720 /*
1721  * Function to update ctime/mtime for a given device path.
1722  * Mainly used for ctime/mtime based probe like libblkid.
1723  */
1724 static void update_dev_time(const char *path_name)
1725 {
1726         struct file *filp;
1727 
1728         filp = filp_open(path_name, O_RDWR, 0);
1729         if (IS_ERR(filp))
1730                 return;
1731         file_update_time(filp);
1732         filp_close(filp, NULL);
1733 }
1734 
1735 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1736                              struct btrfs_device *device)
1737 {
1738         struct btrfs_root *root = fs_info->chunk_root;
1739         int ret;
1740         struct btrfs_path *path;
1741         struct btrfs_key key;
1742         struct btrfs_trans_handle *trans;
1743 
1744         path = btrfs_alloc_path();
1745         if (!path)
1746                 return -ENOMEM;
1747 
1748         trans = btrfs_start_transaction(root, 0);
1749         if (IS_ERR(trans)) {
1750                 btrfs_free_path(path);
1751                 return PTR_ERR(trans);
1752         }
1753         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1754         key.type = BTRFS_DEV_ITEM_KEY;
1755         key.offset = device->devid;
1756 
1757         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1758         if (ret) {
1759                 if (ret > 0)
1760                         ret = -ENOENT;
1761                 btrfs_abort_transaction(trans, ret);
1762                 btrfs_end_transaction(trans);
1763                 goto out;
1764         }
1765 
1766         ret = btrfs_del_item(trans, root, path);
1767         if (ret) {
1768                 btrfs_abort_transaction(trans, ret);
1769                 btrfs_end_transaction(trans);
1770         }
1771 
1772 out:
1773         btrfs_free_path(path);
1774         if (!ret)
1775                 ret = btrfs_commit_transaction(trans);
1776         return ret;
1777 }
1778 
1779 /*
1780  * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781  * filesystem. It's up to the caller to adjust that number regarding eg. device
1782  * replace.
1783  */
1784 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1785                 u64 num_devices)
1786 {
1787         u64 all_avail;
1788         unsigned seq;
1789         int i;
1790 
1791         do {
1792                 seq = read_seqbegin(&fs_info->profiles_lock);
1793 
1794                 all_avail = fs_info->avail_data_alloc_bits |
1795                             fs_info->avail_system_alloc_bits |
1796                             fs_info->avail_metadata_alloc_bits;
1797         } while (read_seqretry(&fs_info->profiles_lock, seq));
1798 
1799         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1800                 if (!(all_avail & btrfs_raid_group[i]))
1801                         continue;
1802 
1803                 if (num_devices < btrfs_raid_array[i].devs_min) {
1804                         int ret = btrfs_raid_mindev_error[i];
1805 
1806                         if (ret)
1807                                 return ret;
1808                 }
1809         }
1810 
1811         return 0;
1812 }
1813 
1814 static struct btrfs_device * btrfs_find_next_active_device(
1815                 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1816 {
1817         struct btrfs_device *next_device;
1818 
1819         list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1820                 if (next_device != device &&
1821                         !next_device->missing && next_device->bdev)
1822                         return next_device;
1823         }
1824 
1825         return NULL;
1826 }
1827 
1828 /*
1829  * Helper function to check if the given device is part of s_bdev / latest_bdev
1830  * and replace it with the provided or the next active device, in the context
1831  * where this function called, there should be always be another device (or
1832  * this_dev) which is active.
1833  */
1834 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1835                 struct btrfs_device *device, struct btrfs_device *this_dev)
1836 {
1837         struct btrfs_device *next_device;
1838 
1839         if (this_dev)
1840                 next_device = this_dev;
1841         else
1842                 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1843                                                                 device);
1844         ASSERT(next_device);
1845 
1846         if (fs_info->sb->s_bdev &&
1847                         (fs_info->sb->s_bdev == device->bdev))
1848                 fs_info->sb->s_bdev = next_device->bdev;
1849 
1850         if (fs_info->fs_devices->latest_bdev == device->bdev)
1851                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1852 }
1853 
1854 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1855                 u64 devid)
1856 {
1857         struct btrfs_device *device;
1858         struct btrfs_fs_devices *cur_devices;
1859         u64 num_devices;
1860         int ret = 0;
1861 
1862         mutex_lock(&uuid_mutex);
1863 
1864         num_devices = fs_info->fs_devices->num_devices;
1865         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1866         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1867                 WARN_ON(num_devices < 1);
1868                 num_devices--;
1869         }
1870         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1871 
1872         ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1873         if (ret)
1874                 goto out;
1875 
1876         ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1877                                            &device);
1878         if (ret)
1879                 goto out;
1880 
1881         if (device->is_tgtdev_for_dev_replace) {
1882                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1883                 goto out;
1884         }
1885 
1886         if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1887                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1888                 goto out;
1889         }
1890 
1891         if (device->writeable) {
1892                 mutex_lock(&fs_info->chunk_mutex);
1893                 list_del_init(&device->dev_alloc_list);
1894                 device->fs_devices->rw_devices--;
1895                 mutex_unlock(&fs_info->chunk_mutex);
1896         }
1897 
1898         mutex_unlock(&uuid_mutex);
1899         ret = btrfs_shrink_device(device, 0);
1900         mutex_lock(&uuid_mutex);
1901         if (ret)
1902                 goto error_undo;
1903 
1904         /*
1905          * TODO: the superblock still includes this device in its num_devices
1906          * counter although write_all_supers() is not locked out. This
1907          * could give a filesystem state which requires a degraded mount.
1908          */
1909         ret = btrfs_rm_dev_item(fs_info, device);
1910         if (ret)
1911                 goto error_undo;
1912 
1913         device->in_fs_metadata = 0;
1914         btrfs_scrub_cancel_dev(fs_info, device);
1915 
1916         /*
1917          * the device list mutex makes sure that we don't change
1918          * the device list while someone else is writing out all
1919          * the device supers. Whoever is writing all supers, should
1920          * lock the device list mutex before getting the number of
1921          * devices in the super block (super_copy). Conversely,
1922          * whoever updates the number of devices in the super block
1923          * (super_copy) should hold the device list mutex.
1924          */
1925 
1926         cur_devices = device->fs_devices;
1927         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1928         list_del_rcu(&device->dev_list);
1929 
1930         device->fs_devices->num_devices--;
1931         device->fs_devices->total_devices--;
1932 
1933         if (device->missing)
1934                 device->fs_devices->missing_devices--;
1935 
1936         btrfs_assign_next_active_device(fs_info, device, NULL);
1937 
1938         if (device->bdev) {
1939                 device->fs_devices->open_devices--;
1940                 /* remove sysfs entry */
1941                 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1942         }
1943 
1944         num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1945         btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1946         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1947 
1948         /*
1949          * at this point, the device is zero sized and detached from
1950          * the devices list.  All that's left is to zero out the old
1951          * supers and free the device.
1952          */
1953         if (device->writeable)
1954                 btrfs_scratch_superblocks(device->bdev, device->name->str);
1955 
1956         btrfs_close_bdev(device);
1957         call_rcu(&device->rcu, free_device);
1958 
1959         if (cur_devices->open_devices == 0) {
1960                 struct btrfs_fs_devices *fs_devices;
1961                 fs_devices = fs_info->fs_devices;
1962                 while (fs_devices) {
1963                         if (fs_devices->seed == cur_devices) {
1964                                 fs_devices->seed = cur_devices->seed;
1965                                 break;
1966                         }
1967                         fs_devices = fs_devices->seed;
1968                 }
1969                 cur_devices->seed = NULL;
1970                 __btrfs_close_devices(cur_devices);
1971                 free_fs_devices(cur_devices);
1972         }
1973 
1974 out:
1975         mutex_unlock(&uuid_mutex);
1976         return ret;
1977 
1978 error_undo:
1979         if (device->writeable) {
1980                 mutex_lock(&fs_info->chunk_mutex);
1981                 list_add(&device->dev_alloc_list,
1982                          &fs_info->fs_devices->alloc_list);
1983                 device->fs_devices->rw_devices++;
1984                 mutex_unlock(&fs_info->chunk_mutex);
1985         }
1986         goto out;
1987 }
1988 
1989 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1990                                         struct btrfs_device *srcdev)
1991 {
1992         struct btrfs_fs_devices *fs_devices;
1993 
1994         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1995 
1996         /*
1997          * in case of fs with no seed, srcdev->fs_devices will point
1998          * to fs_devices of fs_info. However when the dev being replaced is
1999          * a seed dev it will point to the seed's local fs_devices. In short
2000          * srcdev will have its correct fs_devices in both the cases.
2001          */
2002         fs_devices = srcdev->fs_devices;
2003 
2004         list_del_rcu(&srcdev->dev_list);
2005         list_del(&srcdev->dev_alloc_list);
2006         fs_devices->num_devices--;
2007         if (srcdev->missing)
2008                 fs_devices->missing_devices--;
2009 
2010         if (srcdev->writeable)
2011                 fs_devices->rw_devices--;
2012 
2013         if (srcdev->bdev)
2014                 fs_devices->open_devices--;
2015 }
2016 
2017 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2018                                       struct btrfs_device *srcdev)
2019 {
2020         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2021 
2022         if (srcdev->writeable) {
2023                 /* zero out the old super if it is writable */
2024                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2025         }
2026 
2027         btrfs_close_bdev(srcdev);
2028         call_rcu(&srcdev->rcu, free_device);
2029 
2030         /* if this is no devs we rather delete the fs_devices */
2031         if (!fs_devices->num_devices) {
2032                 struct btrfs_fs_devices *tmp_fs_devices;
2033 
2034                 /*
2035                  * On a mounted FS, num_devices can't be zero unless it's a
2036                  * seed. In case of a seed device being replaced, the replace
2037                  * target added to the sprout FS, so there will be no more
2038                  * device left under the seed FS.
2039                  */
2040                 ASSERT(fs_devices->seeding);
2041 
2042                 tmp_fs_devices = fs_info->fs_devices;
2043                 while (tmp_fs_devices) {
2044                         if (tmp_fs_devices->seed == fs_devices) {
2045                                 tmp_fs_devices->seed = fs_devices->seed;
2046                                 break;
2047                         }
2048                         tmp_fs_devices = tmp_fs_devices->seed;
2049                 }
2050                 fs_devices->seed = NULL;
2051                 __btrfs_close_devices(fs_devices);
2052                 free_fs_devices(fs_devices);
2053         }
2054 }
2055 
2056 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2057                                       struct btrfs_device *tgtdev)
2058 {
2059         mutex_lock(&uuid_mutex);
2060         WARN_ON(!tgtdev);
2061         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2062 
2063         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2064 
2065         if (tgtdev->bdev)
2066                 fs_info->fs_devices->open_devices--;
2067 
2068         fs_info->fs_devices->num_devices--;
2069 
2070         btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2071 
2072         list_del_rcu(&tgtdev->dev_list);
2073 
2074         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2075         mutex_unlock(&uuid_mutex);
2076 
2077         /*
2078          * The update_dev_time() with in btrfs_scratch_superblocks()
2079          * may lead to a call to btrfs_show_devname() which will try
2080          * to hold device_list_mutex. And here this device
2081          * is already out of device list, so we don't have to hold
2082          * the device_list_mutex lock.
2083          */
2084         btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2085 
2086         btrfs_close_bdev(tgtdev);
2087         call_rcu(&tgtdev->rcu, free_device);
2088 }
2089 
2090 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2091                                      const char *device_path,
2092                                      struct btrfs_device **device)
2093 {
2094         int ret = 0;
2095         struct btrfs_super_block *disk_super;
2096         u64 devid;
2097         u8 *dev_uuid;
2098         struct block_device *bdev;
2099         struct buffer_head *bh;
2100 
2101         *device = NULL;
2102         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2103                                     fs_info->bdev_holder, 0, &bdev, &bh);
2104         if (ret)
2105                 return ret;
2106         disk_super = (struct btrfs_super_block *)bh->b_data;
2107         devid = btrfs_stack_device_id(&disk_super->dev_item);
2108         dev_uuid = disk_super->dev_item.uuid;
2109         *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2110         brelse(bh);
2111         if (!*device)
2112                 ret = -ENOENT;
2113         blkdev_put(bdev, FMODE_READ);
2114         return ret;
2115 }
2116 
2117 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2118                                          const char *device_path,
2119                                          struct btrfs_device **device)
2120 {
2121         *device = NULL;
2122         if (strcmp(device_path, "missing") == 0) {
2123                 struct list_head *devices;
2124                 struct btrfs_device *tmp;
2125 
2126                 devices = &fs_info->fs_devices->devices;
2127                 /*
2128                  * It is safe to read the devices since the volume_mutex
2129                  * is held by the caller.
2130                  */
2131                 list_for_each_entry(tmp, devices, dev_list) {
2132                         if (tmp->in_fs_metadata && !tmp->bdev) {
2133                                 *device = tmp;
2134                                 break;
2135                         }
2136                 }
2137 
2138                 if (!*device)
2139                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2140 
2141                 return 0;
2142         } else {
2143                 return btrfs_find_device_by_path(fs_info, device_path, device);
2144         }
2145 }
2146 
2147 /*
2148  * Lookup a device given by device id, or the path if the id is 0.
2149  */
2150 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2151                                  const char *devpath,
2152                                  struct btrfs_device **device)
2153 {
2154         int ret;
2155 
2156         if (devid) {
2157                 ret = 0;
2158                 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2159                 if (!*device)
2160                         ret = -ENOENT;
2161         } else {
2162                 if (!devpath || !devpath[0])
2163                         return -EINVAL;
2164 
2165                 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2166                                                            device);
2167         }
2168         return ret;
2169 }
2170 
2171 /*
2172  * does all the dirty work required for changing file system's UUID.
2173  */
2174 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2175 {
2176         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2177         struct btrfs_fs_devices *old_devices;
2178         struct btrfs_fs_devices *seed_devices;
2179         struct btrfs_super_block *disk_super = fs_info->super_copy;
2180         struct btrfs_device *device;
2181         u64 super_flags;
2182 
2183         BUG_ON(!mutex_is_locked(&uuid_mutex));
2184         if (!fs_devices->seeding)
2185                 return -EINVAL;
2186 
2187         seed_devices = alloc_fs_devices(NULL);
2188         if (IS_ERR(seed_devices))
2189                 return PTR_ERR(seed_devices);
2190 
2191         old_devices = clone_fs_devices(fs_devices);
2192         if (IS_ERR(old_devices)) {
2193                 kfree(seed_devices);
2194                 return PTR_ERR(old_devices);
2195         }
2196 
2197         list_add(&old_devices->list, &fs_uuids);
2198 
2199         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2200         seed_devices->opened = 1;
2201         INIT_LIST_HEAD(&seed_devices->devices);
2202         INIT_LIST_HEAD(&seed_devices->alloc_list);
2203         mutex_init(&seed_devices->device_list_mutex);
2204 
2205         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2206         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2207                               synchronize_rcu);
2208         list_for_each_entry(device, &seed_devices->devices, dev_list)
2209                 device->fs_devices = seed_devices;
2210 
2211         mutex_lock(&fs_info->chunk_mutex);
2212         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2213         mutex_unlock(&fs_info->chunk_mutex);
2214 
2215         fs_devices->seeding = 0;
2216         fs_devices->num_devices = 0;
2217         fs_devices->open_devices = 0;
2218         fs_devices->missing_devices = 0;
2219         fs_devices->rotating = 0;
2220         fs_devices->seed = seed_devices;
2221 
2222         generate_random_uuid(fs_devices->fsid);
2223         memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2224         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2225         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2226 
2227         super_flags = btrfs_super_flags(disk_super) &
2228                       ~BTRFS_SUPER_FLAG_SEEDING;
2229         btrfs_set_super_flags(disk_super, super_flags);
2230 
2231         return 0;
2232 }
2233 
2234 /*
2235  * Store the expected generation for seed devices in device items.
2236  */
2237 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2238                                struct btrfs_fs_info *fs_info)
2239 {
2240         struct btrfs_root *root = fs_info->chunk_root;
2241         struct btrfs_path *path;
2242         struct extent_buffer *leaf;
2243         struct btrfs_dev_item *dev_item;
2244         struct btrfs_device *device;
2245         struct btrfs_key key;
2246         u8 fs_uuid[BTRFS_FSID_SIZE];
2247         u8 dev_uuid[BTRFS_UUID_SIZE];
2248         u64 devid;
2249         int ret;
2250 
2251         path = btrfs_alloc_path();
2252         if (!path)
2253                 return -ENOMEM;
2254 
2255         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2256         key.offset = 0;
2257         key.type = BTRFS_DEV_ITEM_KEY;
2258 
2259         while (1) {
2260                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2261                 if (ret < 0)
2262                         goto error;
2263 
2264                 leaf = path->nodes[0];
2265 next_slot:
2266                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2267                         ret = btrfs_next_leaf(root, path);
2268                         if (ret > 0)
2269                                 break;
2270                         if (ret < 0)
2271                                 goto error;
2272                         leaf = path->nodes[0];
2273                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2274                         btrfs_release_path(path);
2275                         continue;
2276                 }
2277 
2278                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2279                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2280                     key.type != BTRFS_DEV_ITEM_KEY)
2281                         break;
2282 
2283                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2284                                           struct btrfs_dev_item);
2285                 devid = btrfs_device_id(leaf, dev_item);
2286                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2287                                    BTRFS_UUID_SIZE);
2288                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2289                                    BTRFS_FSID_SIZE);
2290                 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2291                 BUG_ON(!device); /* Logic error */
2292 
2293                 if (device->fs_devices->seeding) {
2294                         btrfs_set_device_generation(leaf, dev_item,
2295                                                     device->generation);
2296                         btrfs_mark_buffer_dirty(leaf);
2297                 }
2298 
2299                 path->slots[0]++;
2300                 goto next_slot;
2301         }
2302         ret = 0;
2303 error:
2304         btrfs_free_path(path);
2305         return ret;
2306 }
2307 
2308 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2309 {
2310         struct btrfs_root *root = fs_info->dev_root;
2311         struct request_queue *q;
2312         struct btrfs_trans_handle *trans;
2313         struct btrfs_device *device;
2314         struct block_device *bdev;
2315         struct list_head *devices;
2316         struct super_block *sb = fs_info->sb;
2317         struct rcu_string *name;
2318         u64 tmp;
2319         int seeding_dev = 0;
2320         int ret = 0;
2321         bool unlocked = false;
2322 
2323         if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2324                 return -EROFS;
2325 
2326         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2327                                   fs_info->bdev_holder);
2328         if (IS_ERR(bdev))
2329                 return PTR_ERR(bdev);
2330 
2331         if (fs_info->fs_devices->seeding) {
2332                 seeding_dev = 1;
2333                 down_write(&sb->s_umount);
2334                 mutex_lock(&uuid_mutex);
2335         }
2336 
2337         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2338 
2339         devices = &fs_info->fs_devices->devices;
2340 
2341         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2342         list_for_each_entry(device, devices, dev_list) {
2343                 if (device->bdev == bdev) {
2344                         ret = -EEXIST;
2345                         mutex_unlock(
2346                                 &fs_info->fs_devices->device_list_mutex);
2347                         goto error;
2348                 }
2349         }
2350         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2351 
2352         device = btrfs_alloc_device(fs_info, NULL, NULL);
2353         if (IS_ERR(device)) {
2354                 /* we can safely leave the fs_devices entry around */
2355                 ret = PTR_ERR(device);
2356                 goto error;
2357         }
2358 
2359         name = rcu_string_strdup(device_path, GFP_KERNEL);
2360         if (!name) {
2361                 bio_put(device->flush_bio);
2362                 kfree(device);
2363                 ret = -ENOMEM;
2364                 goto error;
2365         }
2366         rcu_assign_pointer(device->name, name);
2367 
2368         trans = btrfs_start_transaction(root, 0);
2369         if (IS_ERR(trans)) {
2370                 rcu_string_free(device->name);
2371                 bio_put(device->flush_bio);
2372                 kfree(device);
2373                 ret = PTR_ERR(trans);
2374                 goto error;
2375         }
2376 
2377         q = bdev_get_queue(bdev);
2378         if (blk_queue_discard(q))
2379                 device->can_discard = 1;
2380         device->writeable = 1;
2381         device->generation = trans->transid;
2382         device->io_width = fs_info->sectorsize;
2383         device->io_align = fs_info->sectorsize;
2384         device->sector_size = fs_info->sectorsize;
2385         device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2386                                          fs_info->sectorsize);
2387         device->disk_total_bytes = device->total_bytes;
2388         device->commit_total_bytes = device->total_bytes;
2389         device->fs_info = fs_info;
2390         device->bdev = bdev;
2391         device->in_fs_metadata = 1;
2392         device->is_tgtdev_for_dev_replace = 0;
2393         device->mode = FMODE_EXCL;
2394         device->dev_stats_valid = 1;
2395         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2396 
2397         if (seeding_dev) {
2398                 sb->s_flags &= ~SB_RDONLY;
2399                 ret = btrfs_prepare_sprout(fs_info);
2400                 if (ret) {
2401                         btrfs_abort_transaction(trans, ret);
2402                         goto error_trans;
2403                 }
2404         }
2405 
2406         device->fs_devices = fs_info->fs_devices;
2407 
2408         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2409         mutex_lock(&fs_info->chunk_mutex);
2410         list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2411         list_add(&device->dev_alloc_list,
2412                  &fs_info->fs_devices->alloc_list);
2413         fs_info->fs_devices->num_devices++;
2414         fs_info->fs_devices->open_devices++;
2415         fs_info->fs_devices->rw_devices++;
2416         fs_info->fs_devices->total_devices++;
2417         fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2418 
2419         atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2420 
2421         if (!blk_queue_nonrot(q))
2422                 fs_info->fs_devices->rotating = 1;
2423 
2424         tmp = btrfs_super_total_bytes(fs_info->super_copy);
2425         btrfs_set_super_total_bytes(fs_info->super_copy,
2426                 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2427 
2428         tmp = btrfs_super_num_devices(fs_info->super_copy);
2429         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2430 
2431         /* add sysfs device entry */
2432         btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2433 
2434         /*
2435          * we've got more storage, clear any full flags on the space
2436          * infos
2437          */
2438         btrfs_clear_space_info_full(fs_info);
2439 
2440         mutex_unlock(&fs_info->chunk_mutex);
2441         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2442 
2443         if (seeding_dev) {
2444                 mutex_lock(&fs_info->chunk_mutex);
2445                 ret = init_first_rw_device(trans, fs_info);
2446                 mutex_unlock(&fs_info->chunk_mutex);
2447                 if (ret) {
2448                         btrfs_abort_transaction(trans, ret);
2449                         goto error_sysfs;
2450                 }
2451         }
2452 
2453         ret = btrfs_add_device(trans, fs_info, device);
2454         if (ret) {
2455                 btrfs_abort_transaction(trans, ret);
2456                 goto error_sysfs;
2457         }
2458 
2459         if (seeding_dev) {
2460                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2461 
2462                 ret = btrfs_finish_sprout(trans, fs_info);
2463                 if (ret) {
2464                         btrfs_abort_transaction(trans, ret);
2465                         goto error_sysfs;
2466                 }
2467 
2468                 /* Sprouting would change fsid of the mounted root,
2469                  * so rename the fsid on the sysfs
2470                  */
2471                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2472                                                 fs_info->fsid);
2473                 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2474                         btrfs_warn(fs_info,
2475                                    "sysfs: failed to create fsid for sprout");
2476         }
2477 
2478         ret = btrfs_commit_transaction(trans);
2479 
2480         if (seeding_dev) {
2481                 mutex_unlock(&uuid_mutex);
2482                 up_write(&sb->s_umount);
2483                 unlocked = true;
2484 
2485                 if (ret) /* transaction commit */
2486                         return ret;
2487 
2488                 ret = btrfs_relocate_sys_chunks(fs_info);
2489                 if (ret < 0)
2490                         btrfs_handle_fs_error(fs_info, ret,
2491                                     "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2492                 trans = btrfs_attach_transaction(root);
2493                 if (IS_ERR(trans)) {
2494                         if (PTR_ERR(trans) == -ENOENT)
2495                                 return 0;
2496                         ret = PTR_ERR(trans);
2497                         trans = NULL;
2498                         goto error_sysfs;
2499                 }
2500                 ret = btrfs_commit_transaction(trans);
2501         }
2502 
2503         /* Update ctime/mtime for libblkid */
2504         update_dev_time(device_path);
2505         return ret;
2506 
2507 error_sysfs:
2508         btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2509 error_trans:
2510         if (seeding_dev)
2511                 sb->s_flags |= SB_RDONLY;
2512         if (trans)
2513                 btrfs_end_transaction(trans);
2514         rcu_string_free(device->name);
2515         bio_put(device->flush_bio);
2516         kfree(device);
2517 error:
2518         blkdev_put(bdev, FMODE_EXCL);
2519         if (seeding_dev && !unlocked) {
2520                 mutex_unlock(&uuid_mutex);
2521                 up_write(&sb->s_umount);
2522         }
2523         return ret;
2524 }
2525 
2526 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2527                                   const char *device_path,
2528                                   struct btrfs_device *srcdev,
2529                                   struct btrfs_device **device_out)
2530 {
2531         struct request_queue *q;
2532         struct btrfs_device *device;
2533         struct block_device *bdev;
2534         struct list_head *devices;
2535         struct rcu_string *name;
2536         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2537         int ret = 0;
2538 
2539         *device_out = NULL;
2540         if (fs_info->fs_devices->seeding) {
2541                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2542                 return -EINVAL;
2543         }
2544 
2545         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2546                                   fs_info->bdev_holder);
2547         if (IS_ERR(bdev)) {
2548                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2549                 return PTR_ERR(bdev);
2550         }
2551 
2552         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2553 
2554         devices = &fs_info->fs_devices->devices;
2555         list_for_each_entry(device, devices, dev_list) {
2556                 if (device->bdev == bdev) {
2557                         btrfs_err(fs_info,
2558                                   "target device is in the filesystem!");
2559                         ret = -EEXIST;
2560                         goto error;
2561                 }
2562         }
2563 
2564 
2565         if (i_size_read(bdev->bd_inode) <
2566             btrfs_device_get_total_bytes(srcdev)) {
2567                 btrfs_err(fs_info,
2568                           "target device is smaller than source device!");
2569                 ret = -EINVAL;
2570                 goto error;
2571         }
2572 
2573 
2574         device = btrfs_alloc_device(NULL, &devid, NULL);
2575         if (IS_ERR(device)) {
2576                 ret = PTR_ERR(device);
2577                 goto error;
2578         }
2579 
2580         name = rcu_string_strdup(device_path, GFP_KERNEL);
2581         if (!name) {
2582                 bio_put(device->flush_bio);
2583                 kfree(device);
2584                 ret = -ENOMEM;
2585                 goto error;
2586         }
2587         rcu_assign_pointer(device->name, name);
2588 
2589         q = bdev_get_queue(bdev);
2590         if (blk_queue_discard(q))
2591                 device->can_discard = 1;
2592         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2593         device->writeable = 1;
2594         device->generation = 0;
2595         device->io_width = fs_info->sectorsize;
2596         device->io_align = fs_info->sectorsize;
2597         device->sector_size = fs_info->sectorsize;
2598         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2599         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2600         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2601         ASSERT(list_empty(&srcdev->resized_list));
2602         device->commit_total_bytes = srcdev->commit_total_bytes;
2603         device->commit_bytes_used = device->bytes_used;
2604         device->fs_info = fs_info;
2605         device->bdev = bdev;
2606         device->in_fs_metadata = 1;
2607         device->is_tgtdev_for_dev_replace = 1;
2608         device->mode = FMODE_EXCL;
2609         device->dev_stats_valid = 1;
2610         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2611         device->fs_devices = fs_info->fs_devices;
2612         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2613         fs_info->fs_devices->num_devices++;
2614         fs_info->fs_devices->open_devices++;
2615         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2616 
2617         *device_out = device;
2618         return ret;
2619 
2620 error:
2621         blkdev_put(bdev, FMODE_EXCL);
2622         return ret;
2623 }
2624 
2625 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2626                                               struct btrfs_device *tgtdev)
2627 {
2628         u32 sectorsize = fs_info->sectorsize;
2629 
2630         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2631         tgtdev->io_width = sectorsize;
2632         tgtdev->io_align = sectorsize;
2633         tgtdev->sector_size = sectorsize;
2634         tgtdev->fs_info = fs_info;
2635         tgtdev->in_fs_metadata = 1;
2636 }
2637 
2638 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2639                                         struct btrfs_device *device)
2640 {
2641         int ret;
2642         struct btrfs_path *path;
2643         struct btrfs_root *root = device->fs_info->chunk_root;
2644         struct btrfs_dev_item *dev_item;
2645         struct extent_buffer *leaf;
2646         struct btrfs_key key;
2647 
2648         path = btrfs_alloc_path();
2649         if (!path)
2650                 return -ENOMEM;
2651 
2652         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2653         key.type = BTRFS_DEV_ITEM_KEY;
2654         key.offset = device->devid;
2655 
2656         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2657         if (ret < 0)
2658                 goto out;
2659 
2660         if (ret > 0) {
2661                 ret = -ENOENT;
2662                 goto out;
2663         }
2664 
2665         leaf = path->nodes[0];
2666         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2667 
2668         btrfs_set_device_id(leaf, dev_item, device->devid);
2669         btrfs_set_device_type(leaf, dev_item, device->type);
2670         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2671         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2672         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2673         btrfs_set_device_total_bytes(leaf, dev_item,
2674                                      btrfs_device_get_disk_total_bytes(device));
2675         btrfs_set_device_bytes_used(leaf, dev_item,
2676                                     btrfs_device_get_bytes_used(device));
2677         btrfs_mark_buffer_dirty(leaf);
2678 
2679 out:
2680         btrfs_free_path(path);
2681         return ret;
2682 }
2683 
2684 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2685                       struct btrfs_device *device, u64 new_size)
2686 {
2687         struct btrfs_fs_info *fs_info = device->fs_info;
2688         struct btrfs_super_block *super_copy = fs_info->super_copy;
2689         struct btrfs_fs_devices *fs_devices;
2690         u64 old_total;
2691         u64 diff;
2692 
2693         if (!device->writeable)
2694                 return -EACCES;
2695 
2696         new_size = round_down(new_size, fs_info->sectorsize);
2697 
2698         mutex_lock(&fs_info->chunk_mutex);
2699         old_total = btrfs_super_total_bytes(super_copy);
2700         diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2701 
2702         if (new_size <= device->total_bytes ||
2703             device->is_tgtdev_for_dev_replace) {
2704                 mutex_unlock(&fs_info->chunk_mutex);
2705                 return -EINVAL;
2706         }
2707 
2708         fs_devices = fs_info->fs_devices;
2709 
2710         btrfs_set_super_total_bytes(super_copy,
2711                         round_down(old_total + diff, fs_info->sectorsize));
2712         device->fs_devices->total_rw_bytes += diff;
2713 
2714         btrfs_device_set_total_bytes(device, new_size);
2715         btrfs_device_set_disk_total_bytes(device, new_size);
2716         btrfs_clear_space_info_full(device->fs_info);
2717         if (list_empty(&device->resized_list))
2718                 list_add_tail(&device->resized_list,
2719                               &fs_devices->resized_devices);
2720         mutex_unlock(&fs_info->chunk_mutex);
2721 
2722         return btrfs_update_device(trans, device);
2723 }
2724 
2725 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2726                             struct btrfs_fs_info *fs_info, u64 chunk_offset)
2727 {
2728         struct btrfs_root *root = fs_info->chunk_root;
2729         int ret;
2730         struct btrfs_path *path;
2731         struct btrfs_key key;
2732 
2733         path = btrfs_alloc_path();
2734         if (!path)
2735                 return -ENOMEM;
2736 
2737         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2738         key.offset = chunk_offset;
2739         key.type = BTRFS_CHUNK_ITEM_KEY;
2740 
2741         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2742         if (ret < 0)
2743                 goto out;
2744         else if (ret > 0) { /* Logic error or corruption */
2745                 btrfs_handle_fs_error(fs_info, -ENOENT,
2746                                       "Failed lookup while freeing chunk.");
2747                 ret = -ENOENT;
2748                 goto out;
2749         }
2750 
2751         ret = btrfs_del_item(trans, root, path);
2752         if (ret < 0)
2753                 btrfs_handle_fs_error(fs_info, ret,
2754                                       "Failed to delete chunk item.");
2755 out:
2756         btrfs_free_path(path);
2757         return ret;
2758 }
2759 
2760 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2761 {
2762         struct btrfs_super_block *super_copy = fs_info->super_copy;
2763         struct btrfs_disk_key *disk_key;
2764         struct btrfs_chunk *chunk;
2765         u8 *ptr;
2766         int ret = 0;
2767         u32 num_stripes;
2768         u32 array_size;
2769         u32 len = 0;
2770         u32 cur;
2771         struct btrfs_key key;
2772 
2773         mutex_lock(&fs_info->chunk_mutex);
2774         array_size = btrfs_super_sys_array_size(super_copy);
2775 
2776         ptr = super_copy->sys_chunk_array;
2777         cur = 0;
2778 
2779         while (cur < array_size) {
2780                 disk_key = (struct btrfs_disk_key *)ptr;
2781                 btrfs_disk_key_to_cpu(&key, disk_key);
2782 
2783                 len = sizeof(*disk_key);
2784 
2785                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2786                         chunk = (struct btrfs_chunk *)(ptr + len);
2787                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2788                         len += btrfs_chunk_item_size(num_stripes);
2789                 } else {
2790                         ret = -EIO;
2791                         break;
2792                 }
2793                 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2794                     key.offset == chunk_offset) {
2795                         memmove(ptr, ptr + len, array_size - (cur + len));
2796                         array_size -= len;
2797                         btrfs_set_super_sys_array_size(super_copy, array_size);
2798                 } else {
2799                         ptr += len;
2800                         cur += len;
2801                 }
2802         }
2803         mutex_unlock(&fs_info->chunk_mutex);
2804         return ret;
2805 }
2806 
2807 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2808                                         u64 logical, u64 length)
2809 {
2810         struct extent_map_tree *em_tree;
2811         struct extent_map *em;
2812 
2813         em_tree = &fs_info->mapping_tree.map_tree;
2814         read_lock(&em_tree->lock);
2815         em = lookup_extent_mapping(em_tree, logical, length);
2816         read_unlock(&em_tree->lock);
2817 
2818         if (!em) {
2819                 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2820                            logical, length);
2821                 return ERR_PTR(-EINVAL);
2822         }
2823 
2824         if (em->start > logical || em->start + em->len < logical) {
2825                 btrfs_crit(fs_info,
2826                            "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2827                            logical, length, em->start, em->start + em->len);
2828                 free_extent_map(em);
2829                 return ERR_PTR(-EINVAL);
2830         }
2831 
2832         /* callers are responsible for dropping em's ref. */
2833         return em;
2834 }
2835 
2836 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2837                        struct btrfs_fs_info *fs_info, u64 chunk_offset)
2838 {
2839         struct extent_map *em;
2840         struct map_lookup *map;
2841         u64 dev_extent_len = 0;
2842         int i, ret = 0;
2843         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2844 
2845         em = get_chunk_map(fs_info, chunk_offset, 1);
2846         if (IS_ERR(em)) {
2847                 /*
2848                  * This is a logic error, but we don't want to just rely on the
2849                  * user having built with ASSERT enabled, so if ASSERT doesn't
2850                  * do anything we still error out.
2851                  */
2852                 ASSERT(0);
2853                 return PTR_ERR(em);
2854         }
2855         map = em->map_lookup;
2856         mutex_lock(&fs_info->chunk_mutex);
2857         check_system_chunk(trans, fs_info, map->type);
2858         mutex_unlock(&fs_info->chunk_mutex);
2859 
2860         /*
2861          * Take the device list mutex to prevent races with the final phase of
2862          * a device replace operation that replaces the device object associated
2863          * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2864          */
2865         mutex_lock(&fs_devices->device_list_mutex);
2866         for (i = 0; i < map->num_stripes; i++) {
2867                 struct btrfs_device *device = map->stripes[i].dev;
2868                 ret = btrfs_free_dev_extent(trans, device,
2869                                             map->stripes[i].physical,
2870                                             &dev_extent_len);
2871                 if (ret) {
2872                         mutex_unlock(&fs_devices->device_list_mutex);
2873                         btrfs_abort_transaction(trans, ret);
2874                         goto out;
2875                 }
2876 
2877                 if (device->bytes_used > 0) {
2878                         mutex_lock(&fs_info->chunk_mutex);
2879                         btrfs_device_set_bytes_used(device,
2880                                         device->bytes_used - dev_extent_len);
2881                         atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2882                         btrfs_clear_space_info_full(fs_info);
2883                         mutex_unlock(&fs_info->chunk_mutex);
2884                 }
2885 
2886                 if (map->stripes[i].dev) {
2887                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2888                         if (ret) {
2889                                 mutex_unlock(&fs_devices->device_list_mutex);
2890                                 btrfs_abort_transaction(trans, ret);
2891                                 goto out;
2892                         }
2893                 }
2894         }
2895         mutex_unlock(&fs_devices->device_list_mutex);
2896 
2897         ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2898         if (ret) {
2899                 btrfs_abort_transaction(trans, ret);
2900                 goto out;
2901         }
2902 
2903         trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2904 
2905         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2906                 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2907                 if (ret) {
2908                         btrfs_abort_transaction(trans, ret);
2909                         goto out;
2910                 }
2911         }
2912 
2913         ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2914         if (ret) {
2915                 btrfs_abort_transaction(trans, ret);
2916                 goto out;
2917         }
2918 
2919 out:
2920         /* once for us */
2921         free_extent_map(em);
2922         return ret;
2923 }
2924 
2925 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2926 {
2927         struct btrfs_root *root = fs_info->chunk_root;
2928         struct btrfs_trans_handle *trans;
2929         int ret;
2930 
2931         /*
2932          * Prevent races with automatic removal of unused block groups.
2933          * After we relocate and before we remove the chunk with offset
2934          * chunk_offset, automatic removal of the block group can kick in,
2935          * resulting in a failure when calling btrfs_remove_chunk() below.
2936          *
2937          * Make sure to acquire this mutex before doing a tree search (dev
2938          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2939          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2940          * we release the path used to search the chunk/dev tree and before
2941          * the current task acquires this mutex and calls us.
2942          */
2943         ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2944 
2945         ret = btrfs_can_relocate(fs_info, chunk_offset);
2946         if (ret)
2947                 return -ENOSPC;
2948 
2949         /* step one, relocate all the extents inside this chunk */
2950         btrfs_scrub_pause(fs_info);
2951         ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2952         btrfs_scrub_continue(fs_info);
2953         if (ret)
2954                 return ret;
2955 
2956         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2957                                                      chunk_offset);
2958         if (IS_ERR(trans)) {
2959                 ret = PTR_ERR(trans);
2960                 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2961                 return ret;
2962         }
2963 
2964         /*
2965          * step two, delete the device extents and the
2966          * chunk tree entries
2967          */
2968         ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2969         btrfs_end_transaction(trans);
2970         return ret;
2971 }
2972 
2973 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2974 {
2975         struct btrfs_root *chunk_root = fs_info->chunk_root;
2976         struct btrfs_path *path;
2977         struct extent_buffer *leaf;
2978         struct btrfs_chunk *chunk;
2979         struct btrfs_key key;
2980         struct btrfs_key found_key;
2981         u64 chunk_type;
2982         bool retried = false;
2983         int failed = 0;
2984         int ret;
2985 
2986         path = btrfs_alloc_path();
2987         if (!path)
2988                 return -ENOMEM;
2989 
2990 again:
2991         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2992         key.offset = (u64)-1;
2993         key.type = BTRFS_CHUNK_ITEM_KEY;
2994 
2995         while (1) {
2996                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2997                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2998                 if (ret < 0) {
2999                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3000                         goto error;
3001                 }
3002                 BUG_ON(ret == 0); /* Corruption */
3003 
3004                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3005                                           key.type);
3006                 if (ret)
3007                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3008                 if (ret < 0)
3009                         goto error;
3010                 if (ret > 0)
3011                         break;
3012 
3013                 leaf = path->nodes[0];
3014                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3015 
3016                 chunk = btrfs_item_ptr(leaf, path->slots[0],
3017                                        struct btrfs_chunk);
3018                 chunk_type = btrfs_chunk_type(leaf, chunk);
3019                 btrfs_release_path(path);
3020 
3021                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3022                         ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3023                         if (ret == -ENOSPC)
3024                                 failed++;
3025                         else
3026                                 BUG_ON(ret);
3027                 }
3028                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3029 
3030                 if (found_key.offset == 0)
3031                         break;
3032                 key.offset = found_key.offset - 1;
3033         }
3034         ret = 0;
3035         if (failed && !retried) {
3036                 failed = 0;
3037                 retried = true;
3038                 goto again;
3039         } else if (WARN_ON(failed && retried)) {
3040                 ret = -ENOSPC;
3041         }
3042 error:
3043         btrfs_free_path(path);
3044         return ret;
3045 }
3046 
3047 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3048                                struct btrfs_balance_control *bctl)
3049 {
3050         struct btrfs_root *root = fs_info->tree_root;
3051         struct btrfs_trans_handle *trans;
3052         struct btrfs_balance_item *item;
3053         struct btrfs_disk_balance_args disk_bargs;
3054         struct btrfs_path *path;
3055         struct extent_buffer *leaf;
3056         struct btrfs_key key;
3057         int ret, err;
3058 
3059         path = btrfs_alloc_path();
3060         if (!path)
3061                 return -ENOMEM;
3062 
3063         trans = btrfs_start_transaction(root, 0);
3064         if (IS_ERR(trans)) {
3065                 btrfs_free_path(path);
3066                 return PTR_ERR(trans);
3067         }
3068 
3069         key.objectid = BTRFS_BALANCE_OBJECTID;
3070         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3071         key.offset = 0;
3072 
3073         ret = btrfs_insert_empty_item(trans, root, path, &key,
3074                                       sizeof(*item));
3075         if (ret)
3076                 goto out;
3077 
3078         leaf = path->nodes[0];
3079         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3080 
3081         memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3082 
3083         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3084         btrfs_set_balance_data(leaf, item, &disk_bargs);
3085         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3086         btrfs_set_balance_meta(leaf, item, &disk_bargs);
3087         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3088         btrfs_set_balance_sys(leaf, item, &disk_bargs);
3089 
3090         btrfs_set_balance_flags(leaf, item, bctl->flags);
3091 
3092         btrfs_mark_buffer_dirty(leaf);
3093 out:
3094         btrfs_free_path(path);
3095         err = btrfs_commit_transaction(trans);
3096         if (err && !ret)
3097                 ret = err;
3098         return ret;
3099 }
3100 
3101 static int del_balance_item(struct btrfs_fs_info *fs_info)
3102 {
3103         struct btrfs_root *root = fs_info->tree_root;
3104         struct btrfs_trans_handle *trans;
3105         struct btrfs_path *path;
3106         struct btrfs_key key;
3107         int ret, err;
3108 
3109         path = btrfs_alloc_path();
3110         if (!path)
3111                 return -ENOMEM;
3112 
3113         trans = btrfs_start_transaction(root, 0);
3114         if (IS_ERR(trans)) {
3115                 btrfs_free_path(path);
3116                 return PTR_ERR(trans);
3117         }
3118 
3119         key.objectid = BTRFS_BALANCE_OBJECTID;
3120         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3121         key.offset = 0;
3122 
3123         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3124         if (ret < 0)
3125                 goto out;
3126         if (ret > 0) {
3127                 ret = -ENOENT;
3128                 goto out;
3129         }
3130 
3131         ret = btrfs_del_item(trans, root, path);
3132 out:
3133         btrfs_free_path(path);
3134         err = btrfs_commit_transaction(trans);
3135         if (err && !ret)
3136                 ret = err;
3137         return ret;
3138 }
3139 
3140 /*
3141  * This is a heuristic used to reduce the number of chunks balanced on
3142  * resume after balance was interrupted.
3143  */
3144 static void update_balance_args(struct btrfs_balance_control *bctl)
3145 {
3146         /*
3147          * Turn on soft mode for chunk types that were being converted.
3148          */
3149         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3150                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3151         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3152                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3153         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3154                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3155 
3156         /*
3157          * Turn on usage filter if is not already used.  The idea is
3158          * that chunks that we have already balanced should be
3159          * reasonably full.  Don't do it for chunks that are being
3160          * converted - that will keep us from relocating unconverted
3161          * (albeit full) chunks.
3162          */
3163         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3164             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3165             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3166                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3167                 bctl->data.usage = 90;
3168         }
3169         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3170             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3171             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3172                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3173                 bctl->sys.usage = 90;
3174         }
3175         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3176             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3177             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3178                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3179                 bctl->meta.usage = 90;
3180         }
3181 }
3182 
3183 /*
3184  * Should be called with both balance and volume mutexes held to
3185  * serialize other volume operations (add_dev/rm_dev/resize) with
3186  * restriper.  Same goes for unset_balance_control.
3187  */
3188 static void set_balance_control(struct btrfs_balance_control *bctl)
3189 {
3190         struct btrfs_fs_info *fs_info = bctl->fs_info;
3191 
3192         BUG_ON(fs_info->balance_ctl);
3193 
3194         spin_lock(&fs_info->balance_lock);
3195         fs_info->balance_ctl = bctl;
3196         spin_unlock(&fs_info->balance_lock);
3197 }
3198 
3199 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3200 {
3201         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3202 
3203         BUG_ON(!fs_info->balance_ctl);
3204 
3205         spin_lock(&fs_info->balance_lock);
3206         fs_info->balance_ctl = NULL;
3207         spin_unlock(&fs_info->balance_lock);
3208 
3209         kfree(bctl);
3210 }
3211 
3212 /*
3213  * Balance filters.  Return 1 if chunk should be filtered out
3214  * (should not be balanced).
3215  */
3216 static int chunk_profiles_filter(u64 chunk_type,
3217                                  struct btrfs_balance_args *bargs)
3218 {
3219         chunk_type = chunk_to_extended(chunk_type) &
3220                                 BTRFS_EXTENDED_PROFILE_MASK;
3221 
3222         if (bargs->profiles & chunk_type)
3223                 return 0;
3224 
3225         return 1;
3226 }
3227 
3228 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3229                               struct btrfs_balance_args *bargs)
3230 {
3231         struct btrfs_block_group_cache *cache;
3232         u64 chunk_used;
3233         u64 user_thresh_min;
3234         u64 user_thresh_max;
3235         int ret = 1;
3236 
3237         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3238         chunk_used = btrfs_block_group_used(&cache->item);
3239 
3240         if (bargs->usage_min == 0)
3241                 user_thresh_min = 0;
3242         else
3243                 user_thresh_min = div_factor_fine(cache->key.offset,
3244                                         bargs->usage_min);
3245 
3246         if (bargs->usage_max == 0)
3247                 user_thresh_max = 1;
3248         else if (bargs->usage_max > 100)
3249                 user_thresh_max = cache->key.offset;
3250         else
3251                 user_thresh_max = div_factor_fine(cache->key.offset,
3252                                         bargs->usage_max);
3253 
3254         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3255                 ret = 0;
3256 
3257         btrfs_put_block_group(cache);
3258         return ret;
3259 }
3260 
3261 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3262                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3263 {
3264         struct btrfs_block_group_cache *cache;
3265         u64 chunk_used, user_thresh;
3266         int ret = 1;
3267 
3268         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3269         chunk_used = btrfs_block_group_used(&cache->item);
3270 
3271         if (bargs->usage_min == 0)
3272                 user_thresh = 1;
3273         else if (bargs->usage > 100)
3274                 user_thresh = cache->key.offset;
3275         else
3276                 user_thresh = div_factor_fine(cache->key.offset,
3277                                               bargs->usage);
3278 
3279         if (chunk_used < user_thresh)
3280                 ret = 0;
3281 
3282         btrfs_put_block_group(cache);
3283         return ret;
3284 }
3285 
3286 static int chunk_devid_filter(struct extent_buffer *leaf,
3287                               struct btrfs_chunk *chunk,
3288                               struct btrfs_balance_args *bargs)
3289 {
3290         struct btrfs_stripe *stripe;
3291         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3292         int i;
3293 
3294         for (i = 0; i < num_stripes; i++) {
3295                 stripe = btrfs_stripe_nr(chunk, i);
3296                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3297                         return 0;
3298         }
3299 
3300         return 1;
3301 }
3302 
3303 /* [pstart, pend) */
3304 static int chunk_drange_filter(struct extent_buffer *leaf,
3305                                struct btrfs_chunk *chunk,
3306                                struct btrfs_balance_args *bargs)
3307 {
3308         struct btrfs_stripe *stripe;
3309         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3310         u64 stripe_offset;
3311         u64 stripe_length;
3312         int factor;
3313         int i;
3314 
3315         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3316                 return 0;
3317 
3318         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3319              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3320                 factor = num_stripes / 2;
3321         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3322                 factor = num_stripes - 1;
3323         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3324                 factor = num_stripes - 2;
3325         } else {
3326                 factor = num_stripes;
3327         }
3328 
3329         for (i = 0; i < num_stripes; i++) {
3330                 stripe = btrfs_stripe_nr(chunk, i);
3331                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3332                         continue;
3333 
3334                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3335                 stripe_length = btrfs_chunk_length(leaf, chunk);
3336                 stripe_length = div_u64(stripe_length, factor);
3337 
3338                 if (stripe_offset < bargs->pend &&
3339                     stripe_offset + stripe_length > bargs->pstart)
3340                         return 0;
3341         }
3342 
3343         return 1;
3344 }
3345 
3346 /* [vstart, vend) */
3347 static int chunk_vrange_filter(struct extent_buffer *leaf,
3348                                struct btrfs_chunk *chunk,
3349                                u64 chunk_offset,
3350                                struct btrfs_balance_args *bargs)
3351 {
3352         if (chunk_offset < bargs->vend &&
3353             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3354                 /* at least part of the chunk is inside this vrange */
3355                 return 0;
3356 
3357         return 1;
3358 }
3359 
3360 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3361                                struct btrfs_chunk *chunk,
3362                                struct btrfs_balance_args *bargs)
3363 {
3364         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3365 
3366         if (bargs->stripes_min <= num_stripes
3367                         && num_stripes <= bargs->stripes_max)
3368                 return 0;
3369 
3370         return 1;
3371 }
3372 
3373 static int chunk_soft_convert_filter(u64 chunk_type,
3374                                      struct btrfs_balance_args *bargs)
3375 {
3376         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3377                 return 0;
3378 
3379         chunk_type = chunk_to_extended(chunk_type) &
3380                                 BTRFS_EXTENDED_PROFILE_MASK;
3381 
3382         if (bargs->target == chunk_type)
3383                 return 1;
3384 
3385         return 0;
3386 }
3387 
3388 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3389                                 struct extent_buffer *leaf,
3390                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3391 {
3392         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3393         struct btrfs_balance_args *bargs = NULL;
3394         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3395 
3396         /* type filter */
3397         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3398               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3399                 return 0;
3400         }
3401 
3402         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3403                 bargs = &bctl->data;
3404         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3405                 bargs = &bctl->sys;
3406         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3407                 bargs = &bctl->meta;
3408 
3409         /* profiles filter */
3410         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3411             chunk_profiles_filter(chunk_type, bargs)) {
3412                 return 0;
3413         }
3414 
3415         /* usage filter */
3416         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3417             chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3418                 return 0;
3419         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3420             chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3421                 return 0;
3422         }
3423 
3424         /* devid filter */
3425         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3426             chunk_devid_filter(leaf, chunk, bargs)) {
3427                 return 0;
3428         }
3429 
3430         /* drange filter, makes sense only with devid filter */
3431         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3432             chunk_drange_filter(leaf, chunk, bargs)) {
3433                 return 0;
3434         }
3435 
3436         /* vrange filter */
3437         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3438             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3439                 return 0;
3440         }
3441 
3442         /* stripes filter */
3443         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3444             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3445                 return 0;
3446         }
3447 
3448         /* soft profile changing mode */
3449         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3450             chunk_soft_convert_filter(chunk_type, bargs)) {
3451                 return 0;
3452         }
3453 
3454         /*
3455          * limited by count, must be the last filter
3456          */
3457         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3458                 if (bargs->limit == 0)
3459                         return 0;
3460                 else
3461                         bargs->limit--;
3462         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3463                 /*
3464                  * Same logic as the 'limit' filter; the minimum cannot be
3465                  * determined here because we do not have the global information
3466                  * about the count of all chunks that satisfy the filters.
3467                  */
3468                 if (bargs->limit_max == 0)
3469                         return 0;
3470                 else
3471                         bargs->limit_max--;
3472         }
3473 
3474         return 1;
3475 }
3476 
3477 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3478 {
3479         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3480         struct btrfs_root *chunk_root = fs_info->chunk_root;
3481         struct btrfs_root *dev_root = fs_info->dev_root;
3482         struct list_head *devices;
3483         struct btrfs_device *device;
3484         u64 old_size;
3485         u64 size_to_free;
3486         u64 chunk_type;
3487         struct btrfs_chunk *chunk;
3488         struct btrfs_path *path = NULL;
3489         struct btrfs_key key;
3490         struct btrfs_key found_key;
3491         struct btrfs_trans_handle *trans;
3492         struct extent_buffer *leaf;
3493         int slot;
3494         int ret;
3495         int enospc_errors = 0;
3496         bool counting = true;
3497         /* The single value limit and min/max limits use the same bytes in the */
3498         u64 limit_data = bctl->data.limit;
3499         u64 limit_meta = bctl->meta.limit;
3500         u64 limit_sys = bctl->sys.limit;
3501         u32 count_data = 0;
3502         u32 count_meta = 0;
3503         u32 count_sys = 0;
3504         int chunk_reserved = 0;
3505         u64 bytes_used = 0;
3506 
3507         /* step one make some room on all the devices */
3508         devices = &fs_info->fs_devices->devices;
3509         list_for_each_entry(device, devices, dev_list) {
3510                 old_size = btrfs_device_get_total_bytes(device);
3511                 size_to_free = div_factor(old_size, 1);
3512                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3513                 if (!device->writeable ||
3514                     btrfs_device_get_total_bytes(device) -
3515                     btrfs_device_get_bytes_used(device) > size_to_free ||
3516                     device->is_tgtdev_for_dev_replace)
3517                         continue;
3518 
3519                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3520                 if (ret == -ENOSPC)
3521                         break;
3522                 if (ret) {
3523                         /* btrfs_shrink_device never returns ret > 0 */
3524                         WARN_ON(ret > 0);
3525                         goto error;
3526                 }
3527 
3528                 trans = btrfs_start_transaction(dev_root, 0);
3529                 if (IS_ERR(trans)) {
3530                         ret = PTR_ERR(trans);
3531                         btrfs_info_in_rcu(fs_info,
3532                  "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3533                                           rcu_str_deref(device->name), ret,
3534                                           old_size, old_size - size_to_free);
3535                         goto error;
3536                 }
3537 
3538                 ret = btrfs_grow_device(trans, device, old_size);
3539                 if (ret) {
3540                         btrfs_end_transaction(trans);
3541                         /* btrfs_grow_device never returns ret > 0 */
3542                         WARN_ON(ret > 0);
3543                         btrfs_info_in_rcu(fs_info,
3544                  "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3545                                           rcu_str_deref(device->name), ret,
3546                                           old_size, old_size - size_to_free);
3547                         goto error;
3548                 }
3549 
3550                 btrfs_end_transaction(trans);
3551         }
3552 
3553         /* step two, relocate all the chunks */
3554         path = btrfs_alloc_path();
3555         if (!path) {
3556                 ret = -ENOMEM;
3557                 goto error;
3558         }
3559 
3560         /* zero out stat counters */
3561         spin_lock(&fs_info->balance_lock);
3562         memset(&bctl->stat, 0, sizeof(bctl->stat));
3563         spin_unlock(&fs_info->balance_lock);
3564 again:
3565         if (!counting) {
3566                 /*
3567                  * The single value limit and min/max limits use the same bytes
3568                  * in the
3569                  */
3570                 bctl->data.limit = limit_data;
3571                 bctl->meta.limit = limit_meta;
3572                 bctl->sys.limit = limit_sys;
3573         }
3574         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3575         key.offset = (u64)-1;
3576         key.type = BTRFS_CHUNK_ITEM_KEY;
3577 
3578         while (1) {
3579                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3580                     atomic_read(&fs_info->balance_cancel_req)) {
3581                         ret = -ECANCELED;
3582                         goto error;
3583                 }
3584 
3585                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3586                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3587                 if (ret < 0) {
3588                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3589                         goto error;
3590                 }
3591 
3592                 /*
3593                  * this shouldn't happen, it means the last relocate
3594                  * failed
3595                  */
3596                 if (ret == 0)
3597                         BUG(); /* FIXME break ? */
3598 
3599                 ret = btrfs_previous_item(chunk_root, path, 0,
3600                                           BTRFS_CHUNK_ITEM_KEY);
3601                 if (ret) {
3602                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3603                         ret = 0;
3604                         break;
3605                 }
3606 
3607                 leaf = path->nodes[0];
3608                 slot = path->slots[0];
3609                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3610 
3611                 if (found_key.objectid != key.objectid) {
3612                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3613                         break;
3614                 }
3615 
3616                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3617                 chunk_type = btrfs_chunk_type(leaf, chunk);
3618 
3619                 if (!counting) {
3620                         spin_lock(&fs_info->balance_lock);
3621                         bctl->stat.considered++;
3622                         spin_unlock(&fs_info->balance_lock);
3623                 }
3624 
3625                 ret = should_balance_chunk(fs_info, leaf, chunk,
3626                                            found_key.offset);
3627 
3628                 btrfs_release_path(path);
3629                 if (!ret) {
3630                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631                         goto loop;
3632                 }
3633 
3634                 if (counting) {
3635                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3636                         spin_lock(&fs_info->balance_lock);
3637                         bctl->stat.expected++;
3638                         spin_unlock(&fs_info->balance_lock);
3639 
3640                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3641                                 count_data++;
3642                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3643                                 count_sys++;
3644                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3645                                 count_meta++;
3646 
3647                         goto loop;
3648                 }
3649 
3650                 /*
3651                  * Apply limit_min filter, no need to check if the LIMITS
3652                  * filter is used, limit_min is 0 by default
3653                  */
3654                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3655                                         count_data < bctl->data.limit_min)
3656                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3657                                         count_meta < bctl->meta.limit_min)
3658                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3659                                         count_sys < bctl->sys.limit_min)) {
3660                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3661                         goto loop;
3662                 }
3663 
3664                 ASSERT(fs_info->data_sinfo);
3665                 spin_lock(&fs_info->data_sinfo->lock);
3666                 bytes_used = fs_info->data_sinfo->bytes_used;
3667                 spin_unlock(&fs_info->data_sinfo->lock);
3668 
3669                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3670                     !chunk_reserved && !bytes_used) {
3671                         trans = btrfs_start_transaction(chunk_root, 0);
3672                         if (IS_ERR(trans)) {
3673                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3674                                 ret = PTR_ERR(trans);
3675                                 goto error;
3676                         }
3677 
3678                         ret = btrfs_force_chunk_alloc(trans, fs_info,
3679                                                       BTRFS_BLOCK_GROUP_DATA);
3680                         btrfs_end_transaction(trans);
3681                         if (ret < 0) {
3682                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3683                                 goto error;
3684                         }
3685                         chunk_reserved = 1;
3686                 }
3687 
3688                 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3689                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3690                 if (ret && ret != -ENOSPC)
3691                         goto error;
3692                 if (ret == -ENOSPC) {
3693                         enospc_errors++;
3694                 } else {
3695                         spin_lock(&fs_info->balance_lock);
3696                         bctl->stat.completed++;
3697                         spin_unlock(&fs_info->balance_lock);
3698                 }
3699 loop:
3700                 if (found_key.offset == 0)
3701                         break;
3702                 key.offset = found_key.offset - 1;
3703         }
3704 
3705         if (counting) {
3706                 btrfs_release_path(path);
3707                 counting = false;
3708                 goto again;
3709         }
3710 error:
3711         btrfs_free_path(path);
3712         if (enospc_errors) {
3713                 btrfs_info(fs_info, "%d enospc errors during balance",
3714                            enospc_errors);
3715                 if (!ret)
3716                         ret = -ENOSPC;
3717         }
3718 
3719         return ret;
3720 }
3721 
3722 /**
3723  * alloc_profile_is_valid - see if a given profile is valid and reduced
3724  * @flags: profile to validate
3725  * @extended: if true @flags is treated as an extended profile
3726  */
3727 static int alloc_profile_is_valid(u64 flags, int extended)
3728 {
3729         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3730                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3731 
3732         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3733 
3734         /* 1) check that all other bits are zeroed */
3735         if (flags & ~mask)
3736                 return 0;
3737 
3738         /* 2) see if profile is reduced */
3739         if (flags == 0)
3740                 return !extended; /* "" is valid for usual profiles */
3741 
3742         /* true if exactly one bit set */
3743         return (flags & (flags - 1)) == 0;
3744 }
3745 
3746 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3747 {
3748         /* cancel requested || normal exit path */
3749         return atomic_read(&fs_info->balance_cancel_req) ||
3750                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3751                  atomic_read(&fs_info->balance_cancel_req) == 0);
3752 }
3753 
3754 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3755 {
3756         int ret;
3757 
3758         unset_balance_control(fs_info);
3759         ret = del_balance_item(fs_info);
3760         if (ret)
3761                 btrfs_handle_fs_error(fs_info, ret, NULL);
3762 
3763         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3764 }
3765 
3766 /* Non-zero return value signifies invalidity */
3767 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3768                 u64 allowed)
3769 {
3770         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3771                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3772                  (bctl_arg->target & ~allowed)));
3773 }
3774 
3775 /*
3776  * Should be called with both balance and volume mutexes held
3777  */
3778 int btrfs_balance(struct btrfs_balance_control *bctl,
3779                   struct btrfs_ioctl_balance_args *bargs)
3780 {
3781         struct btrfs_fs_info *fs_info = bctl->fs_info;
3782         u64 meta_target, data_target;
3783         u64 allowed;
3784         int mixed = 0;
3785         int ret;
3786         u64 num_devices;
3787         unsigned seq;
3788 
3789         if (btrfs_fs_closing(fs_info) ||
3790             atomic_read(&fs_info->balance_pause_req) ||
3791             atomic_read(&fs_info->balance_cancel_req)) {
3792                 ret = -EINVAL;
3793                 goto out;
3794         }
3795 
3796         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3797         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3798                 mixed = 1;
3799 
3800         /*
3801          * In case of mixed groups both data and meta should be picked,
3802          * and identical options should be given for both of them.
3803          */
3804         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3805         if (mixed && (bctl->flags & allowed)) {
3806                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3807                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3808                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3809                         btrfs_err(fs_info,
3810                                   "with mixed groups data and metadata balance options must be the same");
3811                         ret = -EINVAL;
3812                         goto out;
3813                 }
3814         }
3815 
3816         num_devices = fs_info->fs_devices->num_devices;
3817         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3818         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3819                 BUG_ON(num_devices < 1);
3820                 num_devices--;
3821         }
3822         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3823         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3824         if (num_devices > 1)
3825                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3826         if (num_devices > 2)
3827                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3828         if (num_devices > 3)
3829                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3830                             BTRFS_BLOCK_GROUP_RAID6);
3831         if (validate_convert_profile(&bctl->data, allowed)) {
3832                 btrfs_err(fs_info,
3833                           "unable to start balance with target data profile %llu",
3834                           bctl->data.target);
3835                 ret = -EINVAL;
3836                 goto out;
3837         }
3838         if (validate_convert_profile(&bctl->meta, allowed)) {
3839                 btrfs_err(fs_info,
3840                           "unable to start balance with target metadata profile %llu",
3841                           bctl->meta.target);
3842                 ret = -EINVAL;
3843                 goto out;
3844         }
3845         if (validate_convert_profile(&bctl->sys, allowed)) {
3846                 btrfs_err(fs_info,
3847                           "unable to start balance with target system profile %llu",
3848                           bctl->sys.target);
3849                 ret = -EINVAL;
3850                 goto out;
3851         }
3852 
3853         /* allow to reduce meta or sys integrity only if force set */
3854         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3855                         BTRFS_BLOCK_GROUP_RAID10 |
3856                         BTRFS_BLOCK_GROUP_RAID5 |
3857                         BTRFS_BLOCK_GROUP_RAID6;
3858         do {
3859                 seq = read_seqbegin(&fs_info->profiles_lock);
3860 
3861                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3862                      (fs_info->avail_system_alloc_bits & allowed) &&
3863                      !(bctl->sys.target & allowed)) ||
3864                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3865                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3866                      !(bctl->meta.target & allowed))) {
3867                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3868                                 btrfs_info(fs_info,
3869                                            "force reducing metadata integrity");
3870                         } else {
3871                                 btrfs_err(fs_info,
3872                                           "balance will reduce metadata integrity, use force if you want this");
3873                                 ret = -EINVAL;
3874                                 goto out;
3875                         }
3876                 }
3877         } while (read_seqretry(&fs_info->profiles_lock, seq));
3878 
3879         /* if we're not converting, the target field is uninitialized */
3880         meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3881                 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3882         data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3883                 bctl->data.target : fs_info->avail_data_alloc_bits;
3884         if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3885                 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3886                 btrfs_warn(fs_info,
3887                            "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3888                            meta_target, data_target);
3889         }
3890 
3891         ret = insert_balance_item(fs_info, bctl);
3892         if (ret && ret != -EEXIST)
3893                 goto out;
3894 
3895         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3896                 BUG_ON(ret == -EEXIST);
3897                 set_balance_control(bctl);
3898         } else {
3899                 BUG_ON(ret != -EEXIST);
3900                 spin_lock(&fs_info->balance_lock);
3901                 update_balance_args(bctl);
3902                 spin_unlock(&fs_info->balance_lock);
3903         }
3904 
3905         atomic_inc(&fs_info->balance_running);
3906         mutex_unlock(&fs_info->balance_mutex);
3907 
3908         ret = __btrfs_balance(fs_info);
3909 
3910         mutex_lock(&fs_info->balance_mutex);
3911         atomic_dec(&fs_info->balance_running);
3912 
3913         if (bargs) {
3914                 memset(bargs, 0, sizeof(*bargs));
3915                 update_ioctl_balance_args(fs_info, 0, bargs);
3916         }
3917 
3918         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3919             balance_need_close(fs_info)) {
3920                 __cancel_balance(fs_info);
3921         }
3922 
3923         wake_up(&fs_info->balance_wait_q);
3924 
3925         return ret;
3926 out:
3927         if (bctl->flags & BTRFS_BALANCE_RESUME)
3928                 __cancel_balance(fs_info);
3929         else {
3930                 kfree(bctl);
3931                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3932         }
3933         return ret;
3934 }
3935 
3936 static int balance_kthread(void *data)
3937 {
3938         struct btrfs_fs_info *fs_info = data;
3939         int ret = 0;
3940 
3941         mutex_lock(&fs_info->volume_mutex);
3942         mutex_lock(&fs_info->balance_mutex);
3943 
3944         if (fs_info->balance_ctl) {
3945                 btrfs_info(fs_info, "continuing balance");
3946                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3947         }
3948 
3949         mutex_unlock(&fs_info->balance_mutex);
3950         mutex_unlock(&fs_info->volume_mutex);
3951 
3952         return ret;
3953 }
3954 
3955 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3956 {
3957         struct task_struct *tsk;
3958 
3959         spin_lock(&fs_info->balance_lock);
3960         if (!fs_info->balance_ctl) {
3961                 spin_unlock(&fs_info->balance_lock);
3962                 return 0;
3963         }
3964         spin_unlock(&fs_info->balance_lock);
3965 
3966         if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3967                 btrfs_info(fs_info, "force skipping balance");
3968                 return 0;
3969         }
3970 
3971         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3972         return PTR_ERR_OR_ZERO(tsk);
3973 }
3974 
3975 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3976 {
3977         struct btrfs_balance_control *bctl;
3978         struct btrfs_balance_item *item;
3979         struct btrfs_disk_balance_args disk_bargs;
3980         struct btrfs_path *path;
3981         struct extent_buffer *leaf;
3982         struct btrfs_key key;
3983         int ret;
3984 
3985         path = btrfs_alloc_path();
3986         if (!path)
3987                 return -ENOMEM;
3988 
3989         key.objectid = BTRFS_BALANCE_OBJECTID;
3990         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3991         key.offset = 0;
3992 
3993         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3994         if (ret < 0)
3995                 goto out;
3996         if (ret > 0) { /* ret = -ENOENT; */
3997                 ret = 0;
3998                 goto out;
3999         }
4000 
4001         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4002         if (!bctl) {
4003                 ret = -ENOMEM;
4004                 goto out;
4005         }
4006 
4007         leaf = path->nodes[0];
4008         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4009 
4010         bctl->fs_info = fs_info;
4011         bctl->flags = btrfs_balance_flags(leaf, item);
4012         bctl->flags |= BTRFS_BALANCE_RESUME;
4013 
4014         btrfs_balance_data(leaf, item, &disk_bargs);
4015         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4016         btrfs_balance_meta(leaf, item, &disk_bargs);
4017         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4018         btrfs_balance_sys(leaf, item, &disk_bargs);
4019         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4020 
4021         WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4022 
4023         mutex_lock(&fs_info->volume_mutex);
4024         mutex_lock(&fs_info->balance_mutex);
4025 
4026         set_balance_control(bctl);
4027 
4028         mutex_unlock(&fs_info->balance_mutex);
4029         mutex_unlock(&fs_info->volume_mutex);
4030 out:
4031         btrfs_free_path(path);
4032         return ret;
4033 }
4034 
4035 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4036 {
4037         int ret = 0;
4038 
4039         mutex_lock(&fs_info->balance_mutex);
4040         if (!fs_info->balance_ctl) {
4041                 mutex_unlock(&fs_info->balance_mutex);
4042                 return -ENOTCONN;
4043         }
4044 
4045         if (atomic_read(&fs_info->balance_running)) {
4046                 atomic_inc(&fs_info->balance_pause_req);
4047                 mutex_unlock(&fs_info->balance_mutex);
4048 
4049                 wait_event(fs_info->balance_wait_q,
4050                            atomic_read(&fs_info->balance_running) == 0);
4051 
4052                 mutex_lock(&fs_info->balance_mutex);
4053                 /* we are good with balance_ctl ripped off from under us */
4054                 BUG_ON(atomic_read(&fs_info->balance_running));
4055                 atomic_dec(&fs_info->balance_pause_req);
4056         } else {
4057                 ret = -ENOTCONN;
4058         }
4059 
4060         mutex_unlock(&fs_info->balance_mutex);
4061         return ret;
4062 }
4063 
4064 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4065 {
4066         if (sb_rdonly(fs_info->sb))
4067                 return -EROFS;
4068 
4069         mutex_lock(&fs_info->balance_mutex);
4070         if (!fs_info->balance_ctl) {
4071                 mutex_unlock(&fs_info->balance_mutex);
4072                 return -ENOTCONN;
4073         }
4074 
4075         atomic_inc(&fs_info->balance_cancel_req);
4076         /*
4077          * if we are running just wait and return, balance item is
4078          * deleted in btrfs_balance in this case
4079          */
4080         if (atomic_read(&fs_info->balance_running)) {
4081                 mutex_unlock(&fs_info->balance_mutex);
4082                 wait_event(fs_info->balance_wait_q,
4083                            atomic_read(&fs_info->balance_running) == 0);
4084                 mutex_lock(&fs_info->balance_mutex);
4085         } else {
4086                 /* __cancel_balance needs volume_mutex */
4087                 mutex_unlock(&fs_info->balance_mutex);
4088                 mutex_lock(&fs_info->volume_mutex);
4089                 mutex_lock(&fs_info->balance_mutex);
4090 
4091                 if (fs_info->balance_ctl)
4092                         __cancel_balance(fs_info);
4093 
4094                 mutex_unlock(&fs_info->volume_mutex);
4095         }
4096 
4097         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4098         atomic_dec(&fs_info->balance_cancel_req);
4099         mutex_unlock(&fs_info->balance_mutex);
4100         return 0;
4101 }
4102 
4103 static int btrfs_uuid_scan_kthread(void *data)
4104 {
4105         struct btrfs_fs_info *fs_info = data;
4106         struct btrfs_root *root = fs_info->tree_root;
4107         struct btrfs_key key;
4108         struct btrfs_path *path = NULL;
4109         int ret = 0;
4110         struct extent_buffer *eb;
4111         int slot;
4112         struct btrfs_root_item root_item;
4113         u32 item_size;
4114         struct btrfs_trans_handle *trans = NULL;
4115 
4116         path = btrfs_alloc_path();
4117         if (!path) {
4118                 ret = -ENOMEM;
4119                 goto out;
4120         }
4121 
4122         key.objectid = 0;
4123         key.type = BTRFS_ROOT_ITEM_KEY;
4124         key.offset = 0;
4125 
4126         while (1) {
4127                 ret = btrfs_search_forward(root, &key, path, 0);
4128                 if (ret) {
4129                         if (ret > 0)
4130                                 ret = 0;
4131                         break;
4132                 }
4133 
4134                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4135                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4136                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4137                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4138                         goto skip;
4139 
4140                 eb = path->nodes[0];
4141                 slot = path->slots[0];
4142                 item_size = btrfs_item_size_nr(eb, slot);
4143                 if (item_size < sizeof(root_item))
4144                         goto skip;
4145 
4146                 read_extent_buffer(eb, &root_item,
4147                                    btrfs_item_ptr_offset(eb, slot),
4148                                    (int)sizeof(root_item));
4149                 if (btrfs_root_refs(&root_item) == 0)
4150                         goto skip;
4151 
4152                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4153                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4154                         if (trans)
4155                                 goto update_tree;
4156 
4157                         btrfs_release_path(path);
4158                         /*
4159                          * 1 - subvol uuid item
4160                          * 1 - received_subvol uuid item
4161                          */
4162                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4163                         if (IS_ERR(trans)) {
4164                                 ret = PTR_ERR(trans);
4165                                 break;
4166                         }
4167                         continue;
4168                 } else {
4169                         goto skip;
4170                 }
4171 update_tree:
4172                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4173                         ret = btrfs_uuid_tree_add(trans, fs_info,
4174                                                   root_item.uuid,
4175                                                   BTRFS_UUID_KEY_SUBVOL,
4176                                                   key.objectid);
4177                         if (ret < 0) {
4178                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4179                                         ret);
4180                                 break;
4181                         }
4182                 }
4183 
4184                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4185                         ret = btrfs_uuid_tree_add(trans, fs_info,
4186                                                   root_item.received_uuid,
4187                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4188                                                   key.objectid);
4189                         if (ret < 0) {
4190                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4191                                         ret);
4192                                 break;
4193                         }
4194                 }
4195 
4196 skip:
4197                 if (trans) {
4198                         ret = btrfs_end_transaction(trans);
4199                         trans = NULL;
4200                         if (ret)
4201                                 break;
4202                 }
4203 
4204                 btrfs_release_path(path);
4205                 if (key.offset < (u64)-1) {
4206                         key.offset++;
4207                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4208                         key.offset = 0;
4209                         key.type = BTRFS_ROOT_ITEM_KEY;
4210                 } else if (key.objectid < (u64)-1) {
4211                         key.offset = 0;
4212                         key.type = BTRFS_ROOT_ITEM_KEY;
4213                         key.objectid++;
4214                 } else {
4215                         break;
4216                 }
4217                 cond_resched();
4218         }
4219 
4220 out:
4221         btrfs_free_path(path);
4222         if (trans && !IS_ERR(trans))
4223                 btrfs_end_transaction(trans);
4224         if (ret)
4225                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4226         else
4227                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4228         up(&fs_info->uuid_tree_rescan_sem);
4229         return 0;
4230 }
4231 
4232 /*
4233  * Callback for btrfs_uuid_tree_iterate().
4234  * returns:
4235  * 0    check succeeded, the entry is not outdated.
4236  * < 0  if an error occurred.
4237  * > 0  if the check failed, which means the caller shall remove the entry.
4238  */
4239 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4240                                        u8 *uuid, u8 type, u64 subid)
4241 {
4242         struct btrfs_key key;
4243         int ret = 0;
4244         struct btrfs_root *subvol_root;
4245 
4246         if (type != BTRFS_UUID_KEY_SUBVOL &&
4247             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4248                 goto out;
4249 
4250         key.objectid = subid;
4251         key.type = BTRFS_ROOT_ITEM_KEY;
4252         key.offset = (u64)-1;
4253         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4254         if (IS_ERR(subvol_root)) {
4255                 ret = PTR_ERR(subvol_root);
4256                 if (ret == -ENOENT)
4257                         ret = 1;
4258                 goto out;
4259         }
4260 
4261         switch (type) {
4262         case BTRFS_UUID_KEY_SUBVOL:
4263                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4264                         ret = 1;
4265                 break;
4266         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4267                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4268                            BTRFS_UUID_SIZE))
4269                         ret = 1;
4270                 break;
4271         }
4272 
4273 out:
4274         return ret;
4275 }
4276 
4277 static int btrfs_uuid_rescan_kthread(void *data)
4278 {
4279         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4280         int ret;
4281 
4282         /*
4283          * 1st step is to iterate through the existing UUID tree and
4284          * to delete all entries that contain outdated data.
4285          * 2nd step is to add all missing entries to the UUID tree.
4286          */
4287         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4288         if (ret < 0) {
4289                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4290                 up(&fs_info->uuid_tree_rescan_sem);
4291                 return ret;
4292         }
4293         return btrfs_uuid_scan_kthread(data);
4294 }
4295 
4296 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4297 {
4298         struct btrfs_trans_handle *trans;
4299         struct btrfs_root *tree_root = fs_info->tree_root;
4300         struct btrfs_root *uuid_root;
4301         struct task_struct *task;
4302         int ret;
4303 
4304         /*
4305          * 1 - root node
4306          * 1 - root item
4307          */
4308         trans = btrfs_start_transaction(tree_root, 2);
4309         if (IS_ERR(trans))
4310                 return PTR_ERR(trans);
4311 
4312         uuid_root = btrfs_create_tree(trans, fs_info,
4313                                       BTRFS_UUID_TREE_OBJECTID);
4314         if (IS_ERR(uuid_root)) {
4315                 ret = PTR_ERR(uuid_root);
4316                 btrfs_abort_transaction(trans, ret);
4317                 btrfs_end_transaction(trans);
4318                 return ret;
4319         }
4320 
4321         fs_info->uuid_root = uuid_root;
4322 
4323         ret = btrfs_commit_transaction(trans);
4324         if (ret)
4325                 return ret;
4326 
4327         down(&fs_info->uuid_tree_rescan_sem);
4328         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4329         if (IS_ERR(task)) {
4330                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4331                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4332                 up(&fs_info->uuid_tree_rescan_sem);
4333                 return PTR_ERR(task);
4334         }
4335 
4336         return 0;
4337 }
4338 
4339 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4340 {
4341         struct task_struct *task;
4342 
4343         down(&fs_info->uuid_tree_rescan_sem);
4344         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4345         if (IS_ERR(task)) {
4346                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4347                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4348                 up(&fs_info->uuid_tree_rescan_sem);
4349                 return PTR_ERR(task);
4350         }
4351 
4352         return 0;
4353 }
4354 
4355 /*
4356  * shrinking a device means finding all of the device extents past
4357  * the new size, and then following the back refs to the chunks.
4358  * The chunk relocation code actually frees the device extent
4359  */
4360 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4361 {
4362         struct btrfs_fs_info *fs_info = device->fs_info;
4363         struct btrfs_root *root = fs_info->dev_root;
4364         struct btrfs_trans_handle *trans;
4365         struct btrfs_dev_extent *dev_extent = NULL;
4366         struct btrfs_path *path;
4367         u64 length;
4368         u64 chunk_offset;
4369         int ret;
4370         int slot;
4371         int failed = 0;
4372         bool retried = false;
4373         bool checked_pending_chunks = false;
4374         struct extent_buffer *l;
4375         struct btrfs_key key;
4376         struct btrfs_super_block *super_copy = fs_info->super_copy;
4377         u64 old_total = btrfs_super_total_bytes(super_copy);
4378         u64 old_size = btrfs_device_get_total_bytes(device);
4379         u64 diff;
4380 
4381         new_size = round_down(new_size, fs_info->sectorsize);
4382         diff = round_down(old_size - new_size, fs_info->sectorsize);
4383 
4384         if (device->is_tgtdev_for_dev_replace)
4385                 return -EINVAL;
4386 
4387         path = btrfs_alloc_path();
4388         if (!path)
4389                 return -ENOMEM;
4390 
4391         path->reada = READA_FORWARD;
4392 
4393         mutex_lock(&fs_info->chunk_mutex);
4394 
4395         btrfs_device_set_total_bytes(device, new_size);
4396         if (device->writeable) {
4397                 device->fs_devices->total_rw_bytes -= diff;
4398                 atomic64_sub(diff, &fs_info->free_chunk_space);
4399         }
4400         mutex_unlock(&fs_info->chunk_mutex);
4401 
4402 again:
4403         key.objectid = device->devid;
4404         key.offset = (u64)-1;
4405         key.type = BTRFS_DEV_EXTENT_KEY;
4406 
4407         do {
4408                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4409                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4410                 if (ret < 0) {
4411                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4412                         goto done;
4413                 }
4414 
4415                 ret = btrfs_previous_item(root, path, 0, key.type);
4416                 if (ret)
4417                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4418                 if (ret < 0)
4419                         goto done;
4420                 if (ret) {
4421                         ret = 0;
4422                         btrfs_release_path(path);
4423                         break;
4424                 }
4425 
4426                 l = path->nodes[0];
4427                 slot = path->slots[0];
4428                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4429 
4430                 if (key.objectid != device->devid) {
4431                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4432                         btrfs_release_path(path);
4433                         break;
4434                 }
4435 
4436                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4437                 length = btrfs_dev_extent_length(l, dev_extent);
4438 
4439                 if (key.offset + length <= new_size) {
4440                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4441                         btrfs_release_path(path);
4442                         break;
4443                 }
4444 
4445                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4446                 btrfs_release_path(path);
4447 
4448                 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4449                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4450                 if (ret && ret != -ENOSPC)
4451                         goto done;
4452                 if (ret == -ENOSPC)
4453                         failed++;
4454         } while (key.offset-- > 0);
4455 
4456         if (failed && !retried) {
4457                 failed = 0;
4458                 retried = true;
4459                 goto again;
4460         } else if (failed && retried) {
4461                 ret = -ENOSPC;
4462                 goto done;
4463         }
4464 
4465         /* Shrinking succeeded, else we would be at "done". */
4466         trans = btrfs_start_transaction(root, 0);
4467         if (IS_ERR(trans)) {
4468                 ret = PTR_ERR(trans);
4469                 goto done;
4470         }
4471 
4472         mutex_lock(&fs_info->chunk_mutex);
4473 
4474         /*
4475          * We checked in the above loop all device extents that were already in
4476          * the device tree. However before we have updated the device's
4477          * total_bytes to the new size, we might have had chunk allocations that
4478          * have not complete yet (new block groups attached to transaction
4479          * handles), and therefore their device extents were not yet in the
4480          * device tree and we missed them in the loop above. So if we have any
4481          * pending chunk using a device extent that overlaps the device range
4482          * that we can not use anymore, commit the current transaction and
4483          * repeat the search on the device tree - this way we guarantee we will
4484          * not have chunks using device extents that end beyond 'new_size'.
4485          */
4486         if (!checked_pending_chunks) {
4487                 u64 start = new_size;
4488                 u64 len = old_size - new_size;
4489 
4490                 if (contains_pending_extent(trans->transaction, device,
4491                                             &start, len)) {
4492                         mutex_unlock(&fs_info->chunk_mutex);
4493                         checked_pending_chunks = true;
4494                         failed = 0;
4495                         retried = false;
4496                         ret = btrfs_commit_transaction(trans);
4497                         if (ret)
4498                                 goto done;
4499                         goto again;
4500                 }
4501         }
4502 
4503         btrfs_device_set_disk_total_bytes(device, new_size);
4504         if (list_empty(&device->resized_list))
4505                 list_add_tail(&device->resized_list,
4506                               &fs_info->fs_devices->resized_devices);
4507 
4508         WARN_ON(diff > old_total);
4509         btrfs_set_super_total_bytes(super_copy,
4510                         round_down(old_total - diff, fs_info->sectorsize));
4511         mutex_unlock(&fs_info->chunk_mutex);
4512 
4513         /* Now btrfs_update_device() will change the on-disk size. */
4514         ret = btrfs_update_device(trans, device);
4515         btrfs_end_transaction(trans);
4516 done:
4517         btrfs_free_path(path);
4518         if (ret) {
4519                 mutex_lock(&fs_info->chunk_mutex);
4520                 btrfs_device_set_total_bytes(device, old_size);
4521                 if (device->writeable)
4522                         device->fs_devices->total_rw_bytes += diff;
4523                 atomic64_add(diff, &fs_info->free_chunk_space);
4524                 mutex_unlock(&fs_info->chunk_mutex);
4525         }
4526         return ret;
4527 }
4528 
4529 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4530                            struct btrfs_key *key,
4531                            struct btrfs_chunk *chunk, int item_size)
4532 {
4533         struct btrfs_super_block *super_copy = fs_info->super_copy;
4534         struct btrfs_disk_key disk_key;
4535         u32 array_size;
4536         u8 *ptr;
4537 
4538         mutex_lock(&fs_info->chunk_mutex);
4539         array_size = btrfs_super_sys_array_size(super_copy);
4540         if (array_size + item_size + sizeof(disk_key)
4541                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4542                 mutex_unlock(&fs_info->chunk_mutex);
4543                 return -EFBIG;
4544         }
4545 
4546         ptr = super_copy->sys_chunk_array + array_size;
4547         btrfs_cpu_key_to_disk(&disk_key, key);
4548         memcpy(ptr, &disk_key, sizeof(disk_key));
4549         ptr += sizeof(disk_key);
4550         memcpy(ptr, chunk, item_size);
4551         item_size += sizeof(disk_key);
4552         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4553         mutex_unlock(&fs_info->chunk_mutex);
4554 
4555         return 0;
4556 }
4557 
4558 /*
4559  * sort the devices in descending order by max_avail, total_avail
4560  */
4561 static int btrfs_cmp_device_info(const void *a, const void *b)
4562 {
4563         const struct btrfs_device_info *di_a = a;
4564         const struct btrfs_device_info *di_b = b;
4565 
4566         if (di_a->max_avail > di_b->max_avail)
4567                 return -1;
4568         if (di_a->max_avail < di_b->max_avail)
4569                 return 1;
4570         if (di_a->total_avail > di_b->total_avail)
4571                 return -1;
4572         if (di_a->total_avail < di_b->total_avail)
4573                 return 1;
4574         return 0;
4575 }
4576 
4577 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4578 {
4579         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4580                 return;
4581 
4582         btrfs_set_fs_incompat(info, RAID56);
4583 }
4584 
4585 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)             \
4586                         - sizeof(struct btrfs_chunk))           \
4587                         / sizeof(struct btrfs_stripe) + 1)
4588 
4589 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4590                                 - 2 * sizeof(struct btrfs_disk_key)     \
4591                                 - 2 * sizeof(struct btrfs_chunk))       \
4592                                 / sizeof(struct btrfs_stripe) + 1)
4593 
4594 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4595                                u64 start, u64 type)
4596 {
4597         struct btrfs_fs_info *info = trans->fs_info;
4598         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4599         struct btrfs_device *device;
4600         struct map_lookup *map = NULL;
4601         struct extent_map_tree *em_tree;
4602         struct extent_map *em;
4603         struct btrfs_device_info *devices_info = NULL;
4604         u64 total_avail;
4605         int num_stripes;        /* total number of stripes to allocate */
4606         int data_stripes;       /* number of stripes that count for
4607                                    block group size */
4608         int sub_stripes;        /* sub_stripes info for map */
4609         int dev_stripes;        /* stripes per dev */
4610         int devs_max;           /* max devs to use */
4611         int devs_min;           /* min devs needed */
4612         int devs_increment;     /* ndevs has to be a multiple of this */
4613         int ncopies;            /* how many copies to data has */
4614         int ret;
4615         u64 max_stripe_size;
4616         u64 max_chunk_size;
4617         u64 stripe_size;
4618         u64 num_bytes;
4619         int ndevs;
4620         int i;
4621         int j;
4622         int index;
4623 
4624         BUG_ON(!alloc_profile_is_valid(type, 0));
4625 
4626         if (list_empty(&fs_devices->alloc_list))
4627                 return -ENOSPC;
4628 
4629         index = __get_raid_index(type);
4630 
4631         sub_stripes = btrfs_raid_array[index].sub_stripes;
4632         dev_stripes = btrfs_raid_array[index].dev_stripes;
4633         devs_max = btrfs_raid_array[index].devs_max;
4634         devs_min = btrfs_raid_array[index].devs_min;
4635         devs_increment = btrfs_raid_array[index].devs_increment;
4636         ncopies = btrfs_raid_array[index].ncopies;
4637 
4638         if (type & BTRFS_BLOCK_GROUP_DATA) {
4639                 max_stripe_size = SZ_1G;
4640                 max_chunk_size = 10 * max_stripe_size;
4641                 if (!devs_max)
4642                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4643         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4644                 /* for larger filesystems, use larger metadata chunks */
4645                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4646                         max_stripe_size = SZ_1G;
4647                 else
4648                         max_stripe_size = SZ_256M;
4649                 max_chunk_size = max_stripe_size;
4650                 if (!devs_max)
4651                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4652         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4653                 max_stripe_size = SZ_32M;
4654                 max_chunk_size = 2 * max_stripe_size;
4655                 if (!devs_max)
4656                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4657         } else {
4658                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4659                        type);
4660                 BUG_ON(1);
4661         }
4662 
4663         /* we don't want a chunk larger than 10% of writeable space */
4664         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4665                              max_chunk_size);
4666 
4667         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4668                                GFP_NOFS);
4669         if (!devices_info)
4670                 return -ENOMEM;
4671 
4672         /*
4673          * in the first pass through the devices list, we gather information
4674          * about the available holes on each device.
4675          */
4676         ndevs = 0;
4677         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4678                 u64 max_avail;
4679                 u64 dev_offset;
4680 
4681                 if (!device->writeable) {
4682                         WARN(1, KERN_ERR
4683                                "BTRFS: read-only device in alloc_list\n");
4684                         continue;
4685                 }
4686 
4687                 if (!device->in_fs_metadata ||
4688                     device->is_tgtdev_for_dev_replace)
4689                         continue;
4690 
4691                 if (device->total_bytes > device->bytes_used)
4692                         total_avail = device->total_bytes - device->bytes_used;
4693                 else
4694                         total_avail = 0;
4695 
4696                 /* If there is no space on this device, skip it. */
4697                 if (total_avail == 0)
4698                         continue;
4699 
4700                 ret = find_free_dev_extent(trans, device,
4701                                            max_stripe_size * dev_stripes,
4702                                            &dev_offset, &max_avail);
4703                 if (ret && ret != -ENOSPC)
4704                         goto error;
4705 
4706                 if (ret == 0)
4707                         max_avail = max_stripe_size * dev_stripes;
4708 
4709                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4710                         continue;
4711 
4712                 if (ndevs == fs_devices->rw_devices) {
4713                         WARN(1, "%s: found more than %llu devices\n",
4714                              __func__, fs_devices->rw_devices);
4715                         break;
4716                 }
4717                 devices_info[ndevs].dev_offset = dev_offset;
4718                 devices_info[ndevs].max_avail = max_avail;
4719                 devices_info[ndevs].total_avail = total_avail;
4720                 devices_info[ndevs].dev = device;
4721                 ++ndevs;
4722         }
4723 
4724         /*
4725          * now sort the devices by hole size / available space
4726          */
4727         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4728              btrfs_cmp_device_info, NULL);
4729 
4730         /* round down to number of usable stripes */
4731         ndevs = round_down(ndevs, devs_increment);
4732 
4733         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4734                 ret = -ENOSPC;
4735                 goto error;
4736         }
4737 
4738         ndevs = min(ndevs, devs_max);
4739 
4740         /*
4741          * The primary goal is to maximize the number of stripes, so use as
4742          * many devices as possible, even if the stripes are not maximum sized.
4743          *
4744          * The DUP profile stores more than one stripe per device, the
4745          * max_avail is the total size so we have to adjust.
4746          */
4747         stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4748         num_stripes = ndevs * dev_stripes;
4749 
4750         /*
4751          * this will have to be fixed for RAID1 and RAID10 over
4752          * more drives
4753          */
4754         data_stripes = num_stripes / ncopies;
4755 
4756         if (type & BTRFS_BLOCK_GROUP_RAID5)
4757                 data_stripes = num_stripes - 1;
4758 
4759         if (type & BTRFS_BLOCK_GROUP_RAID6)
4760                 data_stripes = num_stripes - 2;
4761 
4762         /*
4763          * Use the number of data stripes to figure out how big this chunk
4764          * is really going to be in terms of logical address space,
4765          * and compare that answer with the max chunk size
4766          */
4767         if (stripe_size * data_stripes > max_chunk_size) {
4768                 u64 mask = (1ULL << 24) - 1;
4769 
4770                 stripe_size = div_u64(max_chunk_size, data_stripes);
4771 
4772                 /* bump the answer up to a 16MB boundary */
4773                 stripe_size = (stripe_size + mask) & ~mask;
4774 
4775                 /* but don't go higher than the limits we found
4776                  * while searching for free extents
4777                  */
4778                 if (stripe_size > devices_info[ndevs-1].max_avail)
4779                         stripe_size = devices_info[ndevs-1].max_avail;
4780         }
4781 
4782         /* align to BTRFS_STRIPE_LEN */
4783         stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4784 
4785         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4786         if (!map) {
4787                 ret = -ENOMEM;
4788                 goto error;
4789         }
4790         map->num_stripes = num_stripes;
4791 
4792         for (i = 0; i < ndevs; ++i) {
4793                 for (j = 0; j < dev_stripes; ++j) {
4794                         int s = i * dev_stripes + j;
4795                         map->stripes[s].dev = devices_info[i].dev;
4796                         map->stripes[s].physical = devices_info[i].dev_offset +
4797                                                    j * stripe_size;
4798                 }
4799         }
4800         map->stripe_len = BTRFS_STRIPE_LEN;
4801         map->io_align = BTRFS_STRIPE_LEN;
4802         map->io_width = BTRFS_STRIPE_LEN;
4803         map->type = type;
4804         map->sub_stripes = sub_stripes;
4805 
4806         num_bytes = stripe_size * data_stripes;
4807 
4808         trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4809 
4810         em = alloc_extent_map();
4811         if (!em) {
4812                 kfree(map);
4813                 ret = -ENOMEM;
4814                 goto error;
4815         }
4816         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4817         em->map_lookup = map;
4818         em->start = start;
4819         em->len = num_bytes;
4820         em->block_start = 0;
4821         em->block_len = em->len;
4822         em->orig_block_len = stripe_size;
4823 
4824         em_tree = &info->mapping_tree.map_tree;
4825         write_lock(&em_tree->lock);
4826         ret = add_extent_mapping(em_tree, em, 0);
4827         if (ret) {
4828                 write_unlock(&em_tree->lock);
4829                 free_extent_map(em);
4830                 goto error;
4831         }
4832 
4833         list_add_tail(&em->list, &trans->transaction->pending_chunks);
4834         refcount_inc(&em->refs);
4835         write_unlock(&em_tree->lock);
4836 
4837         ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4838         if (ret)
4839                 goto error_del_extent;
4840 
4841         for (i = 0; i < map->num_stripes; i++) {
4842                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4843                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4844         }
4845 
4846         atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4847 
4848         free_extent_map(em);
4849         check_raid56_incompat_flag(info, type);
4850 
4851         kfree(devices_info);
4852         return 0;
4853 
4854 error_del_extent:
4855         write_lock(&em_tree->lock);
4856         remove_extent_mapping(em_tree, em);
4857         write_unlock(&em_tree->lock);
4858 
4859         /* One for our allocation */
4860         free_extent_map(em);
4861         /* One for the tree reference */
4862         free_extent_map(em);
4863         /* One for the pending_chunks list reference */
4864         free_extent_map(em);
4865 error:
4866         kfree(devices_info);
4867         return ret;
4868 }
4869 
4870 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4871                                 struct btrfs_fs_info *fs_info,
4872                                 u64 chunk_offset, u64 chunk_size)
4873 {
4874         struct btrfs_root *extent_root = fs_info->extent_root;
4875         struct btrfs_root *chunk_root = fs_info->chunk_root;
4876         struct btrfs_key key;
4877         struct btrfs_device *device;
4878         struct btrfs_chunk *chunk;
4879         struct btrfs_stripe *stripe;
4880         struct extent_map *em;
4881         struct map_lookup *map;
4882         size_t item_size;
4883         u64 dev_offset;
4884         u64 stripe_size;
4885         int i = 0;
4886         int ret = 0;
4887 
4888         em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4889         if (IS_ERR(em))
4890                 return PTR_ERR(em);
4891 
4892         map = em->map_lookup;
4893         item_size = btrfs_chunk_item_size(map->num_stripes);
4894         stripe_size = em->orig_block_len;
4895 
4896         chunk = kzalloc(item_size, GFP_NOFS);
4897         if (!chunk) {
4898                 ret = -ENOMEM;
4899                 goto out;
4900         }
4901 
4902         /*
4903          * Take the device list mutex to prevent races with the final phase of
4904          * a device replace operation that replaces the device object associated
4905          * with the map's stripes, because the device object's id can change
4906          * at any time during that final phase of the device replace operation
4907          * (dev-replace.c:btrfs_dev_replace_finishing()).
4908          */
4909         mutex_lock(&fs_info->fs_devices->device_list_mutex);
4910         for (i = 0; i < map->num_stripes; i++) {
4911                 device = map->stripes[i].dev;
4912                 dev_offset = map->stripes[i].physical;
4913 
4914                 ret = btrfs_update_device(trans, device);
4915                 if (ret)
4916                         break;
4917                 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4918                                              dev_offset, stripe_size);
4919                 if (ret)
4920                         break;
4921         }
4922         if (ret) {
4923                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4924                 goto out;
4925         }
4926 
4927         stripe = &chunk->stripe;
4928         for (i = 0; i < map->num_stripes; i++) {
4929                 device = map->stripes[i].dev;
4930                 dev_offset = map->stripes[i].physical;
4931 
4932                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4933                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4934                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4935                 stripe++;
4936         }
4937         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4938 
4939         btrfs_set_stack_chunk_length(chunk, chunk_size);
4940         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4941         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4942         btrfs_set_stack_chunk_type(chunk, map->type);
4943         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4944         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4945         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4946         btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4947         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4948 
4949         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4950         key.type = BTRFS_CHUNK_ITEM_KEY;
4951         key.offset = chunk_offset;
4952 
4953         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4954         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4955                 /*
4956                  * TODO: Cleanup of inserted chunk root in case of
4957                  * failure.
4958                  */
4959                 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4960         }
4961 
4962 out:
4963         kfree(chunk);
4964         free_extent_map(em);
4965         return ret;
4966 }
4967 
4968 /*
4969  * Chunk allocation falls into two parts. The first part does works
4970  * that make the new allocated chunk useable, but not do any operation
4971  * that modifies the chunk tree. The second part does the works that
4972  * require modifying the chunk tree. This division is important for the
4973  * bootstrap process of adding storage to a seed btrfs.
4974  */
4975 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4976                       struct btrfs_fs_info *fs_info, u64 type)
4977 {
4978         u64 chunk_offset;
4979 
4980         ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4981         chunk_offset = find_next_chunk(fs_info);
4982         return __btrfs_alloc_chunk(trans, chunk_offset, type);
4983 }
4984 
4985 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4986                                          struct btrfs_fs_info *fs_info)
4987 {
4988         u64 chunk_offset;
4989         u64 sys_chunk_offset;
4990         u64 alloc_profile;
4991         int ret;
4992 
4993         chunk_offset = find_next_chunk(fs_info);
4994         alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4995         ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4996         if (ret)
4997                 return ret;
4998 
4999         sys_chunk_offset = find_next_chunk(fs_info);
5000         alloc_profile = btrfs_system_alloc_profile(fs_info);
5001         ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5002         return ret;
5003 }
5004 
5005 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5006 {
5007         int max_errors;
5008 
5009         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5010                          BTRFS_BLOCK_GROUP_RAID10 |
5011                          BTRFS_BLOCK_GROUP_RAID5 |
5012                          BTRFS_BLOCK_GROUP_DUP)) {
5013                 max_errors = 1;
5014         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5015                 max_errors = 2;
5016         } else {
5017                 max_errors = 0;
5018         }
5019 
5020         return max_errors;
5021 }
5022 
5023 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5024 {
5025         struct extent_map *em;
5026         struct map_lookup *map;
5027         int readonly = 0;
5028         int miss_ndevs = 0;
5029         int i;
5030 
5031         em = get_chunk_map(fs_info, chunk_offset, 1);
5032         if (IS_ERR(em))
5033                 return 1;
5034 
5035         map = em->map_lookup;
5036         for (i = 0; i < map->num_stripes; i++) {
5037                 if (map->stripes[i].dev->missing) {
5038                         miss_ndevs++;
5039                         continue;
5040                 }
5041 
5042                 if (!map->stripes[i].dev->writeable) {
5043                         readonly = 1;
5044                         goto end;
5045                 }
5046         }
5047 
5048         /*
5049          * If the number of missing devices is larger than max errors,
5050          * we can not write the data into that chunk successfully, so
5051          * set it readonly.
5052          */
5053         if (miss_ndevs > btrfs_chunk_max_errors(map))
5054                 readonly = 1;
5055 end:
5056         free_extent_map(em);
5057         return readonly;
5058 }
5059 
5060 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5061 {
5062         extent_map_tree_init(&tree->map_tree);
5063 }
5064 
5065 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5066 {
5067         struct extent_map *em;
5068 
5069         while (1) {
5070                 write_lock(&tree->map_tree.lock);
5071                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5072                 if (em)
5073                         remove_extent_mapping(&tree->map_tree, em);
5074                 write_unlock(&tree->map_tree.lock);
5075                 if (!em)
5076                         break;
5077                 /* once for us */
5078                 free_extent_map(em);
5079                 /* once for the tree */
5080                 free_extent_map(em);
5081         }
5082 }
5083 
5084 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5085 {
5086         struct extent_map *em;
5087         struct map_lookup *map;
5088         int ret;
5089 
5090         em = get_chunk_map(fs_info, logical, len);
5091         if (IS_ERR(em))
5092                 /*
5093                  * We could return errors for these cases, but that could get
5094                  * ugly and we'd probably do the same thing which is just not do
5095                  * anything else and exit, so return 1 so the callers don't try
5096                  * to use other copies.
5097                  */
5098                 return 1;
5099 
5100         map = em->map_lookup;
5101         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5102                 ret = map->num_stripes;
5103         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5104                 ret = map->sub_stripes;
5105         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5106                 ret = 2;
5107         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5108                 ret = 3;
5109         else
5110                 ret = 1;
5111         free_extent_map(em);
5112 
5113         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5114         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5115             fs_info->dev_replace.tgtdev)
5116                 ret++;
5117         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5118 
5119         return ret;
5120 }
5121 
5122 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5123                                     u64 logical)
5124 {
5125         struct extent_map *em;
5126         struct map_lookup *map;
5127         unsigned long len = fs_info->sectorsize;
5128 
5129         em = get_chunk_map(fs_info, logical, len);
5130 
5131         if (!WARN_ON(IS_ERR(em))) {
5132                 map = em->map_lookup;
5133                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5134                         len = map->stripe_len * nr_data_stripes(map);
5135                 free_extent_map(em);
5136         }
5137         return len;
5138 }
5139 
5140 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5141 {
5142         struct extent_map *em;
5143         struct map_lookup *map;
5144         int ret = 0;
5145 
5146         em = get_chunk_map(fs_info, logical, len);
5147 
5148         if(!WARN_ON(IS_ERR(em))) {
5149                 map = em->map_lookup;
5150                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5151                         ret = 1;
5152                 free_extent_map(em);
5153         }
5154         return ret;
5155 }
5156 
5157 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5158                             struct map_lookup *map, int first, int num,
5159                             int optimal, int dev_replace_is_ongoing)
5160 {
5161         int i;
5162         int tolerance;
5163         struct btrfs_device *srcdev;
5164 
5165         if (dev_replace_is_ongoing &&
5166             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5167              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5168                 srcdev = fs_info->dev_replace.srcdev;
5169         else
5170                 srcdev = NULL;
5171 
5172         /*
5173          * try to avoid the drive that is the source drive for a
5174          * dev-replace procedure, only choose it if no other non-missing
5175          * mirror is available
5176          */
5177         for (tolerance = 0; tolerance < 2; tolerance++) {
5178                 if (map->stripes[optimal].dev->bdev &&
5179                     (tolerance || map->stripes[optimal].dev != srcdev))
5180                         return optimal;
5181                 for (i = first; i < first + num; i++) {
5182                         if (map->stripes[i].dev->bdev &&
5183                             (tolerance || map->stripes[i].dev != srcdev))
5184                                 return i;
5185                 }
5186         }
5187 
5188         /* we couldn't find one that doesn't fail.  Just return something
5189          * and the io error handling code will clean up eventually
5190          */
5191         return optimal;
5192 }
5193 
5194 static inline int parity_smaller(u64 a, u64 b)
5195 {
5196         return a > b;
5197 }
5198 
5199 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5200 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5201 {
5202         struct btrfs_bio_stripe s;
5203         int i;
5204         u64 l;
5205         int again = 1;
5206 
5207         while (again) {
5208                 again = 0;
5209                 for (i = 0; i < num_stripes - 1; i++) {
5210                         if (parity_smaller(bbio->raid_map[i],
5211                                            bbio->raid_map[i+1])) {
5212                                 s = bbio->stripes[i];
5213                                 l = bbio->raid_map[i];
5214                                 bbio->stripes[i] = bbio->stripes[i+1];
5215                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5216                                 bbio->stripes[i+1] = s;
5217                                 bbio->raid_map[i+1] = l;
5218 
5219                                 again = 1;
5220                         }
5221                 }
5222         }
5223 }
5224 
5225 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5226 {
5227         struct btrfs_bio *bbio = kzalloc(
5228                  /* the size of the btrfs_bio */
5229                 sizeof(struct btrfs_bio) +
5230                 /* plus the variable array for the stripes */
5231                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5232                 /* plus the variable array for the tgt dev */
5233                 sizeof(int) * (real_stripes) +
5234                 /*
5235                  * plus the raid_map, which includes both the tgt dev
5236                  * and the stripes
5237                  */
5238                 sizeof(u64) * (total_stripes),
5239                 GFP_NOFS|__GFP_NOFAIL);
5240 
5241         atomic_set(&bbio->error, 0);
5242         refcount_set(&bbio->refs, 1);
5243 
5244         return bbio;
5245 }
5246 
5247 void btrfs_get_bbio(struct btrfs_bio *bbio)
5248 {
5249         WARN_ON(!refcount_read(&bbio->refs));
5250         refcount_inc(&bbio->refs);
5251 }
5252 
5253 void btrfs_put_bbio(struct btrfs_bio *bbio)
5254 {
5255         if (!bbio)
5256                 return;
5257         if (refcount_dec_and_test(&bbio->refs))
5258                 kfree(bbio);
5259 }
5260 
5261 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5262 /*
5263  * Please note that, discard won't be sent to target device of device
5264  * replace.
5265  */
5266 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5267                                          u64 logical, u64 length,
5268                                          struct btrfs_bio **bbio_ret)
5269 {
5270         struct extent_map *em;
5271         struct map_lookup *map;
5272         struct btrfs_bio *bbio;
5273         u64 offset;
5274         u64 stripe_nr;
5275         u64 stripe_nr_end;
5276         u64 stripe_end_offset;
5277         u64 stripe_cnt;
5278         u64 stripe_len;
5279         u64 stripe_offset;
5280         u64 num_stripes;
5281         u32 stripe_index;
5282         u32 factor = 0;
5283         u32 sub_stripes = 0;
5284         u64 stripes_per_dev = 0;
5285         u32 remaining_stripes = 0;
5286         u32 last_stripe = 0;
5287         int ret = 0;
5288         int i;
5289 
5290         /* discard always return a bbio */
5291         ASSERT(bbio_ret);
5292 
5293         em = get_chunk_map(fs_info, logical, length);
5294         if (IS_ERR(em))
5295                 return PTR_ERR(em);
5296 
5297         map = em->map_lookup;
5298         /* we don't discard raid56 yet */
5299         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5300                 ret = -EOPNOTSUPP;
5301                 goto out;
5302         }
5303 
5304         offset = logical - em->start;
5305         length = min_t(u64, em->len - offset, length);
5306 
5307         stripe_len = map->stripe_len;
5308         /*
5309          * stripe_nr counts the total number of stripes we have to stride
5310          * to get to this block
5311          */
5312         stripe_nr = div64_u64(offset, stripe_len);
5313 
5314         /* stripe_offset is the offset of this block in its stripe */
5315         stripe_offset = offset - stripe_nr * stripe_len;
5316 
5317         stripe_nr_end = round_up(offset + length, map->stripe_len);
5318         stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5319         stripe_cnt = stripe_nr_end - stripe_nr;
5320         stripe_end_offset = stripe_nr_end * map->stripe_len -
5321                             (offset + length);
5322         /*
5323          * after this, stripe_nr is the number of stripes on this
5324          * device we have to walk to find the data, and stripe_index is
5325          * the number of our device in the stripe array
5326          */
5327         num_stripes = 1;
5328         stripe_index = 0;
5329         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5330                          BTRFS_BLOCK_GROUP_RAID10)) {
5331                 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5332                         sub_stripes = 1;
5333                 else
5334                         sub_stripes = map->sub_stripes;
5335 
5336                 factor = map->num_stripes / sub_stripes;
5337                 num_stripes = min_t(u64, map->num_stripes,
5338                                     sub_stripes * stripe_cnt);
5339                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5340                 stripe_index *= sub_stripes;
5341                 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5342                                               &remaining_stripes);
5343                 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5344                 last_stripe *= sub_stripes;
5345         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5346                                 BTRFS_BLOCK_GROUP_DUP)) {
5347                 num_stripes = map->num_stripes;
5348         } else {
5349                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5350                                         &stripe_index);
5351         }
5352 
5353         bbio = alloc_btrfs_bio(num_stripes, 0);
5354         if (!bbio) {
5355                 ret = -ENOMEM;
5356                 goto out;
5357         }
5358 
5359         for (i = 0; i < num_stripes; i++) {
5360                 bbio->stripes[i].physical =
5361                         map->stripes[stripe_index].physical +
5362                         stripe_offset + stripe_nr * map->stripe_len;
5363                 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5364 
5365                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5366                                  BTRFS_BLOCK_GROUP_RAID10)) {
5367                         bbio->stripes[i].length = stripes_per_dev *
5368                                 map->stripe_len;
5369 
5370                         if (i / sub_stripes < remaining_stripes)
5371                                 bbio->stripes[i].length +=
5372                                         map->stripe_len;
5373 
5374                         /*
5375                          * Special for the first stripe and
5376                          * the last stripe:
5377                          *
5378                          * |-------|...|-------|
5379                          *     |----------|
5380                          *    off     end_off
5381                          */
5382                         if (i < sub_stripes)
5383                                 bbio->stripes[i].length -=
5384                                         stripe_offset;
5385 
5386                         if (stripe_index >= last_stripe &&
5387                             stripe_index <= (last_stripe +
5388                                              sub_stripes - 1))
5389                                 bbio->stripes[i].length -=
5390                                         stripe_end_offset;
5391 
5392                         if (i == sub_stripes - 1)
5393                                 stripe_offset = 0;
5394                 } else {
5395                         bbio->stripes[i].length = length;
5396                 }
5397 
5398                 stripe_index++;
5399                 if (stripe_index == map->num_stripes) {
5400                         stripe_index = 0;
5401                         stripe_nr++;
5402                 }
5403         }
5404 
5405         *bbio_ret = bbio;
5406         bbio->map_type = map->type;
5407         bbio->num_stripes = num_stripes;
5408 out:
5409         free_extent_map(em);
5410         return ret;
5411 }
5412 
5413 /*
5414  * In dev-replace case, for repair case (that's the only case where the mirror
5415  * is selected explicitly when calling btrfs_map_block), blocks left of the
5416  * left cursor can also be read from the target drive.
5417  *
5418  * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5419  * array of stripes.
5420  * For READ, it also needs to be supported using the same mirror number.
5421  *
5422  * If the requested block is not left of the left cursor, EIO is returned. This
5423  * can happen because btrfs_num_copies() returns one more in the dev-replace
5424  * case.
5425  */
5426 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5427                                          u64 logical, u64 length,
5428                                          u64 srcdev_devid, int *mirror_num,
5429                                          u64 *physical)
5430 {
5431         struct btrfs_bio *bbio = NULL;
5432         int num_stripes;
5433         int index_srcdev = 0;
5434         int found = 0;
5435         u64 physical_of_found = 0;
5436         int i;
5437         int ret = 0;
5438 
5439         ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5440                                 logical, &length, &bbio, 0, 0);
5441         if (ret) {
5442                 ASSERT(bbio == NULL);
5443                 return ret;
5444         }
5445 
5446         num_stripes = bbio->num_stripes;
5447         if (*mirror_num > num_stripes) {
5448                 /*
5449                  * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5450                  * that means that the requested area is not left of the left
5451                  * cursor
5452                  */
5453                 btrfs_put_bbio(bbio);
5454                 return -EIO;
5455         }
5456 
5457         /*
5458          * process the rest of the function using the mirror_num of the source
5459          * drive. Therefore look it up first.  At the end, patch the device
5460          * pointer to the one of the target drive.
5461          */
5462         for (i = 0; i < num_stripes; i++) {
5463                 if (bbio->stripes[i].dev->devid != srcdev_devid)
5464                         continue;
5465 
5466                 /*
5467                  * In case of DUP, in order to keep it simple, only add the
5468                  * mirror with the lowest physical address
5469                  */
5470                 if (found &&
5471                     physical_of_found <= bbio->stripes[i].physical)
5472                         continue;
5473 
5474                 index_srcdev = i;
5475                 found = 1;
5476                 physical_of_found = bbio->stripes[i].physical;
5477         }
5478 
5479         btrfs_put_bbio(bbio);
5480 
5481         ASSERT(found);
5482         if (!found)
5483                 return -EIO;
5484 
5485         *mirror_num = index_srcdev + 1;
5486         *physical = physical_of_found;
5487         return ret;
5488 }
5489 
5490 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5491                                       struct btrfs_bio **bbio_ret,
5492                                       struct btrfs_dev_replace *dev_replace,
5493                                       int *num_stripes_ret, int *max_errors_ret)
5494 {
5495         struct btrfs_bio *bbio = *bbio_ret;
5496         u64 srcdev_devid = dev_replace->srcdev->devid;
5497         int tgtdev_indexes = 0;
5498         int num_stripes = *num_stripes_ret;
5499         int max_errors = *max_errors_ret;
5500         int i;
5501 
5502         if (op == BTRFS_MAP_WRITE) {
5503                 int index_where_to_add;
5504 
5505                 /*
5506                  * duplicate the write operations while the dev replace
5507                  * procedure is running. Since the copying of the old disk to
5508                  * the new disk takes place at run time while the filesystem is
5509                  * mounted writable, the regular write operations to the old
5510                  * disk have to be duplicated to go to the new disk as well.
5511                  *
5512                  * Note that device->missing is handled by the caller, and that
5513                  * the write to the old disk is already set up in the stripes
5514                  * array.
5515                  */
5516                 index_where_to_add = num_stripes;
5517                 for (i = 0; i < num_stripes; i++) {
5518                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5519                                 /* write to new disk, too */
5520                                 struct btrfs_bio_stripe *new =
5521                                         bbio->stripes + index_where_to_add;
5522                                 struct btrfs_bio_stripe *old =
5523                                         bbio->stripes + i;
5524 
5525                                 new->physical = old->physical;
5526                                 new->length = old->length;
5527                                 new->dev = dev_replace->tgtdev;
5528                                 bbio->tgtdev_map[i] = index_where_to_add;
5529                                 index_where_to_add++;
5530                                 max_errors++;
5531                                 tgtdev_indexes++;
5532                         }
5533                 }
5534                 num_stripes = index_where_to_add;
5535         } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5536                 int index_srcdev = 0;
5537                 int found = 0;
5538                 u64 physical_of_found = 0;
5539 
5540                 /*
5541                  * During the dev-replace procedure, the target drive can also
5542                  * be used to read data in case it is needed to repair a corrupt
5543                  * block elsewhere. This is possible if the requested area is
5544                  * left of the left cursor. In this area, the target drive is a
5545                  * full copy of the source drive.
5546                  */
5547                 for (i = 0; i < num_stripes; i++) {
5548                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5549                                 /*
5550                                  * In case of DUP, in order to keep it simple,
5551                                  * only add the mirror with the lowest physical
5552                                  * address
5553                                  */
5554                                 if (found &&
5555                                     physical_of_found <=
5556                                      bbio->stripes[i].physical)
5557                                         continue;
5558                                 index_srcdev = i;
5559                                 found = 1;
5560                                 physical_of_found = bbio->stripes[i].physical;
5561                         }
5562                 }
5563                 if (found) {
5564                         struct btrfs_bio_stripe *tgtdev_stripe =
5565                                 bbio->stripes + num_stripes;
5566 
5567                         tgtdev_stripe->physical = physical_of_found;
5568                         tgtdev_stripe->length =
5569                                 bbio->stripes[index_srcdev].length;
5570                         tgtdev_stripe->dev = dev_replace->tgtdev;
5571                         bbio->tgtdev_map[index_srcdev] = num_stripes;
5572 
5573                         tgtdev_indexes++;
5574                         num_stripes++;
5575                 }
5576         }
5577 
5578         *num_stripes_ret = num_stripes;
5579         *max_errors_ret = max_errors;
5580         bbio->num_tgtdevs = tgtdev_indexes;
5581         *bbio_ret = bbio;
5582 }
5583 
5584 static bool need_full_stripe(enum btrfs_map_op op)
5585 {
5586         return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5587 }
5588 
5589 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5590                              enum btrfs_map_op op,
5591                              u64 logical, u64 *length,
5592                              struct btrfs_bio **bbio_ret,
5593                              int mirror_num, int need_raid_map)
5594 {
5595         struct extent_map *em;
5596         struct map_lookup *map;
5597         u64 offset;
5598         u64 stripe_offset;
5599         u64 stripe_nr;
5600         u64 stripe_len;
5601         u32 stripe_index;
5602         int i;
5603         int ret = 0;
5604         int num_stripes;
5605         int max_errors = 0;
5606         int tgtdev_indexes = 0;
5607         struct btrfs_bio *bbio = NULL;
5608         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5609         int dev_replace_is_ongoing = 0;
5610         int num_alloc_stripes;
5611         int patch_the_first_stripe_for_dev_replace = 0;
5612         u64 physical_to_patch_in_first_stripe = 0;
5613         u64 raid56_full_stripe_start = (u64)-1;
5614 
5615         if (op == BTRFS_MAP_DISCARD)
5616                 return __btrfs_map_block_for_discard(fs_info, logical,
5617                                                      *length, bbio_ret);
5618 
5619         em = get_chunk_map(fs_info, logical, *length);
5620         if (IS_ERR(em))
5621                 return PTR_ERR(em);
5622 
5623         map = em->map_lookup;
5624         offset = logical - em->start;
5625 
5626         stripe_len = map->stripe_len;
5627         stripe_nr = offset;
5628         /*
5629          * stripe_nr counts the total number of stripes we have to stride
5630          * to get to this block
5631          */
5632         stripe_nr = div64_u64(stripe_nr, stripe_len);
5633 
5634         stripe_offset = stripe_nr * stripe_len;
5635         if (offset < stripe_offset) {
5636                 btrfs_crit(fs_info,
5637                            "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5638                            stripe_offset, offset, em->start, logical,
5639                            stripe_len);
5640                 free_extent_map(em);
5641                 return -EINVAL;
5642         }
5643 
5644         /* stripe_offset is the offset of this block in its stripe*/
5645         stripe_offset = offset - stripe_offset;
5646 
5647         /* if we're here for raid56, we need to know the stripe aligned start */
5648         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5649                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5650                 raid56_full_stripe_start = offset;
5651 
5652                 /* allow a write of a full stripe, but make sure we don't
5653                  * allow straddling of stripes
5654                  */
5655                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5656                                 full_stripe_len);
5657                 raid56_full_stripe_start *= full_stripe_len;
5658         }
5659 
5660         if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5661                 u64 max_len;
5662                 /* For writes to RAID[56], allow a full stripeset across all disks.
5663                    For other RAID types and for RAID[56] reads, just allow a single
5664                    stripe (on a single disk). */
5665                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5666                     (op == BTRFS_MAP_WRITE)) {
5667                         max_len = stripe_len * nr_data_stripes(map) -
5668                                 (offset - raid56_full_stripe_start);
5669                 } else {
5670                         /* we limit the length of each bio to what fits in a stripe */
5671                         max_len = stripe_len - stripe_offset;
5672                 }
5673                 *length = min_t(u64, em->len - offset, max_len);
5674         } else {
5675                 *length = em->len - offset;
5676         }
5677 
5678         /* This is for when we're called from btrfs_merge_bio_hook() and all
5679            it cares about is the length */
5680         if (!bbio_ret)
5681                 goto out;
5682 
5683         btrfs_dev_replace_lock(dev_replace, 0);
5684         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5685         if (!dev_replace_is_ongoing)
5686                 btrfs_dev_replace_unlock(dev_replace, 0);
5687         else
5688                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5689 
5690         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5691             !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5692                 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5693