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

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

Version: ~ [ 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/random.h>
 24 #include <linux/iocontext.h>
 25 #include <linux/capability.h>
 26 #include <linux/ratelimit.h>
 27 #include <linux/kthread.h>
 28 #include <linux/raid/pq.h>
 29 #include <linux/semaphore.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 static int init_first_rw_device(struct btrfs_trans_handle *trans,
122                                 struct btrfs_root *root,
123                                 struct btrfs_device *device);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
128 static void btrfs_close_one_device(struct btrfs_device *device);
129 
130 DEFINE_MUTEX(uuid_mutex);
131 static LIST_HEAD(fs_uuids);
132 struct list_head *btrfs_get_fs_uuids(void)
133 {
134         return &fs_uuids;
135 }
136 
137 static struct btrfs_fs_devices *__alloc_fs_devices(void)
138 {
139         struct btrfs_fs_devices *fs_devs;
140 
141         fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
142         if (!fs_devs)
143                 return ERR_PTR(-ENOMEM);
144 
145         mutex_init(&fs_devs->device_list_mutex);
146 
147         INIT_LIST_HEAD(&fs_devs->devices);
148         INIT_LIST_HEAD(&fs_devs->resized_devices);
149         INIT_LIST_HEAD(&fs_devs->alloc_list);
150         INIT_LIST_HEAD(&fs_devs->list);
151 
152         return fs_devs;
153 }
154 
155 /**
156  * alloc_fs_devices - allocate struct btrfs_fs_devices
157  * @fsid:       a pointer to UUID for this FS.  If NULL a new UUID is
158  *              generated.
159  *
160  * Return: a pointer to a new &struct btrfs_fs_devices on success;
161  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
162  * can be destroyed with kfree() right away.
163  */
164 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
165 {
166         struct btrfs_fs_devices *fs_devs;
167 
168         fs_devs = __alloc_fs_devices();
169         if (IS_ERR(fs_devs))
170                 return fs_devs;
171 
172         if (fsid)
173                 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
174         else
175                 generate_random_uuid(fs_devs->fsid);
176 
177         return fs_devs;
178 }
179 
180 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
181 {
182         struct btrfs_device *device;
183         WARN_ON(fs_devices->opened);
184         while (!list_empty(&fs_devices->devices)) {
185                 device = list_entry(fs_devices->devices.next,
186                                     struct btrfs_device, dev_list);
187                 list_del(&device->dev_list);
188                 rcu_string_free(device->name);
189                 kfree(device);
190         }
191         kfree(fs_devices);
192 }
193 
194 static void btrfs_kobject_uevent(struct block_device *bdev,
195                                  enum kobject_action action)
196 {
197         int ret;
198 
199         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
200         if (ret)
201                 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
202                         action,
203                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
204                         &disk_to_dev(bdev->bd_disk)->kobj);
205 }
206 
207 void btrfs_cleanup_fs_uuids(void)
208 {
209         struct btrfs_fs_devices *fs_devices;
210 
211         while (!list_empty(&fs_uuids)) {
212                 fs_devices = list_entry(fs_uuids.next,
213                                         struct btrfs_fs_devices, list);
214                 list_del(&fs_devices->list);
215                 free_fs_devices(fs_devices);
216         }
217 }
218 
219 static struct btrfs_device *__alloc_device(void)
220 {
221         struct btrfs_device *dev;
222 
223         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
224         if (!dev)
225                 return ERR_PTR(-ENOMEM);
226 
227         INIT_LIST_HEAD(&dev->dev_list);
228         INIT_LIST_HEAD(&dev->dev_alloc_list);
229         INIT_LIST_HEAD(&dev->resized_list);
230 
231         spin_lock_init(&dev->io_lock);
232 
233         spin_lock_init(&dev->reada_lock);
234         atomic_set(&dev->reada_in_flight, 0);
235         atomic_set(&dev->dev_stats_ccnt, 0);
236         btrfs_device_data_ordered_init(dev);
237         INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
238         INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
239 
240         return dev;
241 }
242 
243 static noinline struct btrfs_device *__find_device(struct list_head *head,
244                                                    u64 devid, u8 *uuid)
245 {
246         struct btrfs_device *dev;
247 
248         list_for_each_entry(dev, head, dev_list) {
249                 if (dev->devid == devid &&
250                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
251                         return dev;
252                 }
253         }
254         return NULL;
255 }
256 
257 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
258 {
259         struct btrfs_fs_devices *fs_devices;
260 
261         list_for_each_entry(fs_devices, &fs_uuids, list) {
262                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
263                         return fs_devices;
264         }
265         return NULL;
266 }
267 
268 static int
269 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
270                       int flush, struct block_device **bdev,
271                       struct buffer_head **bh)
272 {
273         int ret;
274 
275         *bdev = blkdev_get_by_path(device_path, flags, holder);
276 
277         if (IS_ERR(*bdev)) {
278                 ret = PTR_ERR(*bdev);
279                 goto error;
280         }
281 
282         if (flush)
283                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
284         ret = set_blocksize(*bdev, 4096);
285         if (ret) {
286                 blkdev_put(*bdev, flags);
287                 goto error;
288         }
289         invalidate_bdev(*bdev);
290         *bh = btrfs_read_dev_super(*bdev);
291         if (IS_ERR(*bh)) {
292                 ret = PTR_ERR(*bh);
293                 blkdev_put(*bdev, flags);
294                 goto error;
295         }
296 
297         return 0;
298 
299 error:
300         *bdev = NULL;
301         *bh = NULL;
302         return ret;
303 }
304 
305 static void requeue_list(struct btrfs_pending_bios *pending_bios,
306                         struct bio *head, struct bio *tail)
307 {
308 
309         struct bio *old_head;
310 
311         old_head = pending_bios->head;
312         pending_bios->head = head;
313         if (pending_bios->tail)
314                 tail->bi_next = old_head;
315         else
316                 pending_bios->tail = tail;
317 }
318 
319 /*
320  * we try to collect pending bios for a device so we don't get a large
321  * number of procs sending bios down to the same device.  This greatly
322  * improves the schedulers ability to collect and merge the bios.
323  *
324  * But, it also turns into a long list of bios to process and that is sure
325  * to eventually make the worker thread block.  The solution here is to
326  * make some progress and then put this work struct back at the end of
327  * the list if the block device is congested.  This way, multiple devices
328  * can make progress from a single worker thread.
329  */
330 static noinline void run_scheduled_bios(struct btrfs_device *device)
331 {
332         struct bio *pending;
333         struct backing_dev_info *bdi;
334         struct btrfs_fs_info *fs_info;
335         struct btrfs_pending_bios *pending_bios;
336         struct bio *tail;
337         struct bio *cur;
338         int again = 0;
339         unsigned long num_run;
340         unsigned long batch_run = 0;
341         unsigned long limit;
342         unsigned long last_waited = 0;
343         int force_reg = 0;
344         int sync_pending = 0;
345         struct blk_plug plug;
346 
347         /*
348          * this function runs all the bios we've collected for
349          * a particular device.  We don't want to wander off to
350          * another device without first sending all of these down.
351          * So, setup a plug here and finish it off before we return
352          */
353         blk_start_plug(&plug);
354 
355         bdi = blk_get_backing_dev_info(device->bdev);
356         fs_info = device->dev_root->fs_info;
357         limit = btrfs_async_submit_limit(fs_info);
358         limit = limit * 2 / 3;
359 
360 loop:
361         spin_lock(&device->io_lock);
362 
363 loop_lock:
364         num_run = 0;
365 
366         /* take all the bios off the list at once and process them
367          * later on (without the lock held).  But, remember the
368          * tail and other pointers so the bios can be properly reinserted
369          * into the list if we hit congestion
370          */
371         if (!force_reg && device->pending_sync_bios.head) {
372                 pending_bios = &device->pending_sync_bios;
373                 force_reg = 1;
374         } else {
375                 pending_bios = &device->pending_bios;
376                 force_reg = 0;
377         }
378 
379         pending = pending_bios->head;
380         tail = pending_bios->tail;
381         WARN_ON(pending && !tail);
382 
383         /*
384          * if pending was null this time around, no bios need processing
385          * at all and we can stop.  Otherwise it'll loop back up again
386          * and do an additional check so no bios are missed.
387          *
388          * device->running_pending is used to synchronize with the
389          * schedule_bio code.
390          */
391         if (device->pending_sync_bios.head == NULL &&
392             device->pending_bios.head == NULL) {
393                 again = 0;
394                 device->running_pending = 0;
395         } else {
396                 again = 1;
397                 device->running_pending = 1;
398         }
399 
400         pending_bios->head = NULL;
401         pending_bios->tail = NULL;
402 
403         spin_unlock(&device->io_lock);
404 
405         while (pending) {
406 
407                 rmb();
408                 /* we want to work on both lists, but do more bios on the
409                  * sync list than the regular list
410                  */
411                 if ((num_run > 32 &&
412                     pending_bios != &device->pending_sync_bios &&
413                     device->pending_sync_bios.head) ||
414                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
415                     device->pending_bios.head)) {
416                         spin_lock(&device->io_lock);
417                         requeue_list(pending_bios, pending, tail);
418                         goto loop_lock;
419                 }
420 
421                 cur = pending;
422                 pending = pending->bi_next;
423                 cur->bi_next = NULL;
424 
425                 /*
426                  * atomic_dec_return implies a barrier for waitqueue_active
427                  */
428                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
429                     waitqueue_active(&fs_info->async_submit_wait))
430                         wake_up(&fs_info->async_submit_wait);
431 
432                 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
433 
434                 /*
435                  * if we're doing the sync list, record that our
436                  * plug has some sync requests on it
437                  *
438                  * If we're doing the regular list and there are
439                  * sync requests sitting around, unplug before
440                  * we add more
441                  */
442                 if (pending_bios == &device->pending_sync_bios) {
443                         sync_pending = 1;
444                 } else if (sync_pending) {
445                         blk_finish_plug(&plug);
446                         blk_start_plug(&plug);
447                         sync_pending = 0;
448                 }
449 
450                 btrfsic_submit_bio(cur->bi_rw, cur);
451                 num_run++;
452                 batch_run++;
453 
454                 cond_resched();
455 
456                 /*
457                  * we made progress, there is more work to do and the bdi
458                  * is now congested.  Back off and let other work structs
459                  * run instead
460                  */
461                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
462                     fs_info->fs_devices->open_devices > 1) {
463                         struct io_context *ioc;
464 
465                         ioc = current->io_context;
466 
467                         /*
468                          * the main goal here is that we don't want to
469                          * block if we're going to be able to submit
470                          * more requests without blocking.
471                          *
472                          * This code does two great things, it pokes into
473                          * the elevator code from a filesystem _and_
474                          * it makes assumptions about how batching works.
475                          */
476                         if (ioc && ioc->nr_batch_requests > 0 &&
477                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
478                             (last_waited == 0 ||
479                              ioc->last_waited == last_waited)) {
480                                 /*
481                                  * we want to go through our batch of
482                                  * requests and stop.  So, we copy out
483                                  * the ioc->last_waited time and test
484                                  * against it before looping
485                                  */
486                                 last_waited = ioc->last_waited;
487                                 cond_resched();
488                                 continue;
489                         }
490                         spin_lock(&device->io_lock);
491                         requeue_list(pending_bios, pending, tail);
492                         device->running_pending = 1;
493 
494                         spin_unlock(&device->io_lock);
495                         btrfs_queue_work(fs_info->submit_workers,
496                                          &device->work);
497                         goto done;
498                 }
499                 /* unplug every 64 requests just for good measure */
500                 if (batch_run % 64 == 0) {
501                         blk_finish_plug(&plug);
502                         blk_start_plug(&plug);
503                         sync_pending = 0;
504                 }
505         }
506 
507         cond_resched();
508         if (again)
509                 goto loop;
510 
511         spin_lock(&device->io_lock);
512         if (device->pending_bios.head || device->pending_sync_bios.head)
513                 goto loop_lock;
514         spin_unlock(&device->io_lock);
515 
516 done:
517         blk_finish_plug(&plug);
518 }
519 
520 static void pending_bios_fn(struct btrfs_work *work)
521 {
522         struct btrfs_device *device;
523 
524         device = container_of(work, struct btrfs_device, work);
525         run_scheduled_bios(device);
526 }
527 
528 
529 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
530 {
531         struct btrfs_fs_devices *fs_devs;
532         struct btrfs_device *dev;
533 
534         if (!cur_dev->name)
535                 return;
536 
537         list_for_each_entry(fs_devs, &fs_uuids, list) {
538                 int del = 1;
539 
540                 if (fs_devs->opened)
541                         continue;
542                 if (fs_devs->seeding)
543                         continue;
544 
545                 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
546 
547                         if (dev == cur_dev)
548                                 continue;
549                         if (!dev->name)
550                                 continue;
551 
552                         /*
553                          * Todo: This won't be enough. What if the same device
554                          * comes back (with new uuid and) with its mapper path?
555                          * But for now, this does help as mostly an admin will
556                          * either use mapper or non mapper path throughout.
557                          */
558                         rcu_read_lock();
559                         del = strcmp(rcu_str_deref(dev->name),
560                                                 rcu_str_deref(cur_dev->name));
561                         rcu_read_unlock();
562                         if (!del)
563                                 break;
564                 }
565 
566                 if (!del) {
567                         /* delete the stale device */
568                         if (fs_devs->num_devices == 1) {
569                                 btrfs_sysfs_remove_fsid(fs_devs);
570                                 list_del(&fs_devs->list);
571                                 free_fs_devices(fs_devs);
572                         } else {
573                                 fs_devs->num_devices--;
574                                 list_del(&dev->dev_list);
575                                 rcu_string_free(dev->name);
576                                 kfree(dev);
577                         }
578                         break;
579                 }
580         }
581 }
582 
583 /*
584  * Add new device to list of registered devices
585  *
586  * Returns:
587  * 1   - first time device is seen
588  * 0   - device already known
589  * < 0 - error
590  */
591 static noinline int device_list_add(const char *path,
592                            struct btrfs_super_block *disk_super,
593                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
594 {
595         struct btrfs_device *device;
596         struct btrfs_fs_devices *fs_devices;
597         struct rcu_string *name;
598         int ret = 0;
599         u64 found_transid = btrfs_super_generation(disk_super);
600 
601         fs_devices = find_fsid(disk_super->fsid);
602         if (!fs_devices) {
603                 fs_devices = alloc_fs_devices(disk_super->fsid);
604                 if (IS_ERR(fs_devices))
605                         return PTR_ERR(fs_devices);
606 
607                 list_add(&fs_devices->list, &fs_uuids);
608 
609                 device = NULL;
610         } else {
611                 device = __find_device(&fs_devices->devices, devid,
612                                        disk_super->dev_item.uuid);
613         }
614 
615         if (!device) {
616                 if (fs_devices->opened)
617                         return -EBUSY;
618 
619                 device = btrfs_alloc_device(NULL, &devid,
620                                             disk_super->dev_item.uuid);
621                 if (IS_ERR(device)) {
622                         /* we can safely leave the fs_devices entry around */
623                         return PTR_ERR(device);
624                 }
625 
626                 name = rcu_string_strdup(path, GFP_NOFS);
627                 if (!name) {
628                         kfree(device);
629                         return -ENOMEM;
630                 }
631                 rcu_assign_pointer(device->name, name);
632 
633                 mutex_lock(&fs_devices->device_list_mutex);
634                 list_add_rcu(&device->dev_list, &fs_devices->devices);
635                 fs_devices->num_devices++;
636                 mutex_unlock(&fs_devices->device_list_mutex);
637 
638                 ret = 1;
639                 device->fs_devices = fs_devices;
640         } else if (!device->name || strcmp(device->name->str, path)) {
641                 /*
642                  * When FS is already mounted.
643                  * 1. If you are here and if the device->name is NULL that
644                  *    means this device was missing at time of FS mount.
645                  * 2. If you are here and if the device->name is different
646                  *    from 'path' that means either
647                  *      a. The same device disappeared and reappeared with
648                  *         different name. or
649                  *      b. The missing-disk-which-was-replaced, has
650                  *         reappeared now.
651                  *
652                  * We must allow 1 and 2a above. But 2b would be a spurious
653                  * and unintentional.
654                  *
655                  * Further in case of 1 and 2a above, the disk at 'path'
656                  * would have missed some transaction when it was away and
657                  * in case of 2a the stale bdev has to be updated as well.
658                  * 2b must not be allowed at all time.
659                  */
660 
661                 /*
662                  * For now, we do allow update to btrfs_fs_device through the
663                  * btrfs dev scan cli after FS has been mounted.  We're still
664                  * tracking a problem where systems fail mount by subvolume id
665                  * when we reject replacement on a mounted FS.
666                  */
667                 if (!fs_devices->opened && found_transid < device->generation) {
668                         /*
669                          * That is if the FS is _not_ mounted and if you
670                          * are here, that means there is more than one
671                          * disk with same uuid and devid.We keep the one
672                          * with larger generation number or the last-in if
673                          * generation are equal.
674                          */
675                         return -EEXIST;
676                 }
677 
678                 name = rcu_string_strdup(path, GFP_NOFS);
679                 if (!name)
680                         return -ENOMEM;
681                 rcu_string_free(device->name);
682                 rcu_assign_pointer(device->name, name);
683                 if (device->missing) {
684                         fs_devices->missing_devices--;
685                         device->missing = 0;
686                 }
687         }
688 
689         /*
690          * Unmount does not free the btrfs_device struct but would zero
691          * generation along with most of the other members. So just update
692          * it back. We need it to pick the disk with largest generation
693          * (as above).
694          */
695         if (!fs_devices->opened)
696                 device->generation = found_transid;
697 
698         /*
699          * if there is new btrfs on an already registered device,
700          * then remove the stale device entry.
701          */
702         btrfs_free_stale_device(device);
703 
704         *fs_devices_ret = fs_devices;
705 
706         return ret;
707 }
708 
709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
710 {
711         struct btrfs_fs_devices *fs_devices;
712         struct btrfs_device *device;
713         struct btrfs_device *orig_dev;
714 
715         fs_devices = alloc_fs_devices(orig->fsid);
716         if (IS_ERR(fs_devices))
717                 return fs_devices;
718 
719         mutex_lock(&orig->device_list_mutex);
720         fs_devices->total_devices = orig->total_devices;
721 
722         /* We have held the volume lock, it is safe to get the devices. */
723         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
724                 struct rcu_string *name;
725 
726                 device = btrfs_alloc_device(NULL, &orig_dev->devid,
727                                             orig_dev->uuid);
728                 if (IS_ERR(device))
729                         goto error;
730 
731                 /*
732                  * This is ok to do without rcu read locked because we hold the
733                  * uuid mutex so nothing we touch in here is going to disappear.
734                  */
735                 if (orig_dev->name) {
736                         name = rcu_string_strdup(orig_dev->name->str,
737                                         GFP_KERNEL);
738                         if (!name) {
739                                 kfree(device);
740                                 goto error;
741                         }
742                         rcu_assign_pointer(device->name, name);
743                 }
744 
745                 list_add(&device->dev_list, &fs_devices->devices);
746                 device->fs_devices = fs_devices;
747                 fs_devices->num_devices++;
748         }
749         mutex_unlock(&orig->device_list_mutex);
750         return fs_devices;
751 error:
752         mutex_unlock(&orig->device_list_mutex);
753         free_fs_devices(fs_devices);
754         return ERR_PTR(-ENOMEM);
755 }
756 
757 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
758 {
759         struct btrfs_device *device, *next;
760         struct btrfs_device *latest_dev = NULL;
761 
762         mutex_lock(&uuid_mutex);
763 again:
764         /* This is the initialized path, it is safe to release the devices. */
765         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
766                 if (device->in_fs_metadata) {
767                         if (!device->is_tgtdev_for_dev_replace &&
768                             (!latest_dev ||
769                              device->generation > latest_dev->generation)) {
770                                 latest_dev = device;
771                         }
772                         continue;
773                 }
774 
775                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
776                         /*
777                          * In the first step, keep the device which has
778                          * the correct fsid and the devid that is used
779                          * for the dev_replace procedure.
780                          * In the second step, the dev_replace state is
781                          * read from the device tree and it is known
782                          * whether the procedure is really active or
783                          * not, which means whether this device is
784                          * used or whether it should be removed.
785                          */
786                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
787                                 continue;
788                         }
789                 }
790                 if (device->bdev) {
791                         blkdev_put(device->bdev, device->mode);
792                         device->bdev = NULL;
793                         fs_devices->open_devices--;
794                 }
795                 if (device->writeable) {
796                         list_del_init(&device->dev_alloc_list);
797                         device->writeable = 0;
798                         if (!device->is_tgtdev_for_dev_replace)
799                                 fs_devices->rw_devices--;
800                 }
801                 list_del_init(&device->dev_list);
802                 fs_devices->num_devices--;
803                 rcu_string_free(device->name);
804                 kfree(device);
805         }
806 
807         if (fs_devices->seed) {
808                 fs_devices = fs_devices->seed;
809                 goto again;
810         }
811 
812         fs_devices->latest_bdev = latest_dev->bdev;
813 
814         mutex_unlock(&uuid_mutex);
815 }
816 
817 static void __free_device(struct work_struct *work)
818 {
819         struct btrfs_device *device;
820 
821         device = container_of(work, struct btrfs_device, rcu_work);
822 
823         if (device->bdev)
824                 blkdev_put(device->bdev, device->mode);
825 
826         rcu_string_free(device->name);
827         kfree(device);
828 }
829 
830 static void free_device(struct rcu_head *head)
831 {
832         struct btrfs_device *device;
833 
834         device = container_of(head, struct btrfs_device, rcu);
835 
836         INIT_WORK(&device->rcu_work, __free_device);
837         schedule_work(&device->rcu_work);
838 }
839 
840 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
841 {
842         struct btrfs_device *device, *tmp;
843 
844         if (--fs_devices->opened > 0)
845                 return 0;
846 
847         mutex_lock(&fs_devices->device_list_mutex);
848         list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
849                 btrfs_close_one_device(device);
850         }
851         mutex_unlock(&fs_devices->device_list_mutex);
852 
853         WARN_ON(fs_devices->open_devices);
854         WARN_ON(fs_devices->rw_devices);
855         fs_devices->opened = 0;
856         fs_devices->seeding = 0;
857 
858         return 0;
859 }
860 
861 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
862 {
863         struct btrfs_fs_devices *seed_devices = NULL;
864         int ret;
865 
866         mutex_lock(&uuid_mutex);
867         ret = __btrfs_close_devices(fs_devices);
868         if (!fs_devices->opened) {
869                 seed_devices = fs_devices->seed;
870                 fs_devices->seed = NULL;
871         }
872         mutex_unlock(&uuid_mutex);
873 
874         while (seed_devices) {
875                 fs_devices = seed_devices;
876                 seed_devices = fs_devices->seed;
877                 __btrfs_close_devices(fs_devices);
878                 free_fs_devices(fs_devices);
879         }
880         /*
881          * Wait for rcu kworkers under __btrfs_close_devices
882          * to finish all blkdev_puts so device is really
883          * free when umount is done.
884          */
885         rcu_barrier();
886         return ret;
887 }
888 
889 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
890                                 fmode_t flags, void *holder)
891 {
892         struct request_queue *q;
893         struct block_device *bdev;
894         struct list_head *head = &fs_devices->devices;
895         struct btrfs_device *device;
896         struct btrfs_device *latest_dev = NULL;
897         struct buffer_head *bh;
898         struct btrfs_super_block *disk_super;
899         u64 devid;
900         int seeding = 1;
901         int ret = 0;
902 
903         flags |= FMODE_EXCL;
904 
905         list_for_each_entry(device, head, dev_list) {
906                 if (device->bdev)
907                         continue;
908                 if (!device->name)
909                         continue;
910 
911                 /* Just open everything we can; ignore failures here */
912                 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
913                                             &bdev, &bh))
914                         continue;
915 
916                 disk_super = (struct btrfs_super_block *)bh->b_data;
917                 devid = btrfs_stack_device_id(&disk_super->dev_item);
918                 if (devid != device->devid)
919                         goto error_brelse;
920 
921                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
922                            BTRFS_UUID_SIZE))
923                         goto error_brelse;
924 
925                 device->generation = btrfs_super_generation(disk_super);
926                 if (!latest_dev ||
927                     device->generation > latest_dev->generation)
928                         latest_dev = device;
929 
930                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
931                         device->writeable = 0;
932                 } else {
933                         device->writeable = !bdev_read_only(bdev);
934                         seeding = 0;
935                 }
936 
937                 q = bdev_get_queue(bdev);
938                 if (blk_queue_discard(q))
939                         device->can_discard = 1;
940 
941                 device->bdev = bdev;
942                 device->in_fs_metadata = 0;
943                 device->mode = flags;
944 
945                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
946                         fs_devices->rotating = 1;
947 
948                 fs_devices->open_devices++;
949                 if (device->writeable &&
950                     device->devid != BTRFS_DEV_REPLACE_DEVID) {
951                         fs_devices->rw_devices++;
952                         list_add(&device->dev_alloc_list,
953                                  &fs_devices->alloc_list);
954                 }
955                 brelse(bh);
956                 continue;
957 
958 error_brelse:
959                 brelse(bh);
960                 blkdev_put(bdev, flags);
961                 continue;
962         }
963         if (fs_devices->open_devices == 0) {
964                 ret = -EINVAL;
965                 goto out;
966         }
967         fs_devices->seeding = seeding;
968         fs_devices->opened = 1;
969         fs_devices->latest_bdev = latest_dev->bdev;
970         fs_devices->total_rw_bytes = 0;
971 out:
972         return ret;
973 }
974 
975 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
976                        fmode_t flags, void *holder)
977 {
978         int ret;
979 
980         mutex_lock(&uuid_mutex);
981         if (fs_devices->opened) {
982                 fs_devices->opened++;
983                 ret = 0;
984         } else {
985                 ret = __btrfs_open_devices(fs_devices, flags, holder);
986         }
987         mutex_unlock(&uuid_mutex);
988         return ret;
989 }
990 
991 /*
992  * Look for a btrfs signature on a device. This may be called out of the mount path
993  * and we are not allowed to call set_blocksize during the scan. The superblock
994  * is read via pagecache
995  */
996 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
997                           struct btrfs_fs_devices **fs_devices_ret)
998 {
999         struct btrfs_super_block *disk_super;
1000         struct block_device *bdev;
1001         struct page *page;
1002         void *p;
1003         int ret = -EINVAL;
1004         u64 devid;
1005         u64 transid;
1006         u64 total_devices;
1007         u64 bytenr;
1008         pgoff_t index;
1009 
1010         /*
1011          * we would like to check all the supers, but that would make
1012          * a btrfs mount succeed after a mkfs from a different FS.
1013          * So, we need to add a special mount option to scan for
1014          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1015          */
1016         bytenr = btrfs_sb_offset(0);
1017         flags |= FMODE_EXCL;
1018         mutex_lock(&uuid_mutex);
1019 
1020         bdev = blkdev_get_by_path(path, flags, holder);
1021 
1022         if (IS_ERR(bdev)) {
1023                 ret = PTR_ERR(bdev);
1024                 goto error;
1025         }
1026 
1027         /* make sure our super fits in the device */
1028         if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1029                 goto error_bdev_put;
1030 
1031         /* make sure our super fits in the page */
1032         if (sizeof(*disk_super) > PAGE_SIZE)
1033                 goto error_bdev_put;
1034 
1035         /* make sure our super doesn't straddle pages on disk */
1036         index = bytenr >> PAGE_SHIFT;
1037         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1038                 goto error_bdev_put;
1039 
1040         /* pull in the page with our super */
1041         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1042                                    index, GFP_NOFS);
1043 
1044         if (IS_ERR_OR_NULL(page))
1045                 goto error_bdev_put;
1046 
1047         p = kmap(page);
1048 
1049         /* align our pointer to the offset of the super block */
1050         disk_super = p + (bytenr & ~PAGE_MASK);
1051 
1052         if (btrfs_super_bytenr(disk_super) != bytenr ||
1053             btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1054                 goto error_unmap;
1055 
1056         devid = btrfs_stack_device_id(&disk_super->dev_item);
1057         transid = btrfs_super_generation(disk_super);
1058         total_devices = btrfs_super_num_devices(disk_super);
1059 
1060         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1061         if (ret > 0) {
1062                 if (disk_super->label[0]) {
1063                         if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1064                                 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1065                         printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1066                 } else {
1067                         printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1068                 }
1069 
1070                 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1071                 ret = 0;
1072         }
1073         if (!ret && fs_devices_ret)
1074                 (*fs_devices_ret)->total_devices = total_devices;
1075 
1076 error_unmap:
1077         kunmap(page);
1078         put_page(page);
1079 
1080 error_bdev_put:
1081         blkdev_put(bdev, flags);
1082 error:
1083         mutex_unlock(&uuid_mutex);
1084         return ret;
1085 }
1086 
1087 /* helper to account the used device space in the range */
1088 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1089                                    u64 end, u64 *length)
1090 {
1091         struct btrfs_key key;
1092         struct btrfs_root *root = device->dev_root;
1093         struct btrfs_dev_extent *dev_extent;
1094         struct btrfs_path *path;
1095         u64 extent_end;
1096         int ret;
1097         int slot;
1098         struct extent_buffer *l;
1099 
1100         *length = 0;
1101 
1102         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1103                 return 0;
1104 
1105         path = btrfs_alloc_path();
1106         if (!path)
1107                 return -ENOMEM;
1108         path->reada = READA_FORWARD;
1109 
1110         key.objectid = device->devid;
1111         key.offset = start;
1112         key.type = BTRFS_DEV_EXTENT_KEY;
1113 
1114         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1115         if (ret < 0)
1116                 goto out;
1117         if (ret > 0) {
1118                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1119                 if (ret < 0)
1120                         goto out;
1121         }
1122 
1123         while (1) {
1124                 l = path->nodes[0];
1125                 slot = path->slots[0];
1126                 if (slot >= btrfs_header_nritems(l)) {
1127                         ret = btrfs_next_leaf(root, path);
1128                         if (ret == 0)
1129                                 continue;
1130                         if (ret < 0)
1131                                 goto out;
1132 
1133                         break;
1134                 }
1135                 btrfs_item_key_to_cpu(l, &key, slot);
1136 
1137                 if (key.objectid < device->devid)
1138                         goto next;
1139 
1140                 if (key.objectid > device->devid)
1141                         break;
1142 
1143                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1144                         goto next;
1145 
1146                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1147                 extent_end = key.offset + btrfs_dev_extent_length(l,
1148                                                                   dev_extent);
1149                 if (key.offset <= start && extent_end > end) {
1150                         *length = end - start + 1;
1151                         break;
1152                 } else if (key.offset <= start && extent_end > start)
1153                         *length += extent_end - start;
1154                 else if (key.offset > start && extent_end <= end)
1155                         *length += extent_end - key.offset;
1156                 else if (key.offset > start && key.offset <= end) {
1157                         *length += end - key.offset + 1;
1158                         break;
1159                 } else if (key.offset > end)
1160                         break;
1161 
1162 next:
1163                 path->slots[0]++;
1164         }
1165         ret = 0;
1166 out:
1167         btrfs_free_path(path);
1168         return ret;
1169 }
1170 
1171 static int contains_pending_extent(struct btrfs_transaction *transaction,
1172                                    struct btrfs_device *device,
1173                                    u64 *start, u64 len)
1174 {
1175         struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1176         struct extent_map *em;
1177         struct list_head *search_list = &fs_info->pinned_chunks;
1178         int ret = 0;
1179         u64 physical_start = *start;
1180 
1181         if (transaction)
1182                 search_list = &transaction->pending_chunks;
1183 again:
1184         list_for_each_entry(em, search_list, list) {
1185                 struct map_lookup *map;
1186                 int i;
1187 
1188                 map = em->map_lookup;
1189                 for (i = 0; i < map->num_stripes; i++) {
1190                         u64 end;
1191 
1192                         if (map->stripes[i].dev != device)
1193                                 continue;
1194                         if (map->stripes[i].physical >= physical_start + len ||
1195                             map->stripes[i].physical + em->orig_block_len <=
1196                             physical_start)
1197                                 continue;
1198                         /*
1199                          * Make sure that while processing the pinned list we do
1200                          * not override our *start with a lower value, because
1201                          * we can have pinned chunks that fall within this
1202                          * device hole and that have lower physical addresses
1203                          * than the pending chunks we processed before. If we
1204                          * do not take this special care we can end up getting
1205                          * 2 pending chunks that start at the same physical
1206                          * device offsets because the end offset of a pinned
1207                          * chunk can be equal to the start offset of some
1208                          * pending chunk.
1209                          */
1210                         end = map->stripes[i].physical + em->orig_block_len;
1211                         if (end > *start) {
1212                                 *start = end;
1213                                 ret = 1;
1214                         }
1215                 }
1216         }
1217         if (search_list != &fs_info->pinned_chunks) {
1218                 search_list = &fs_info->pinned_chunks;
1219                 goto again;
1220         }
1221 
1222         return ret;
1223 }
1224 
1225 
1226 /*
1227  * find_free_dev_extent_start - find free space in the specified device
1228  * @device:       the device which we search the free space in
1229  * @num_bytes:    the size of the free space that we need
1230  * @search_start: the position from which to begin the search
1231  * @start:        store the start of the free space.
1232  * @len:          the size of the free space. that we find, or the size
1233  *                of the max free space if we don't find suitable free space
1234  *
1235  * this uses a pretty simple search, the expectation is that it is
1236  * called very infrequently and that a given device has a small number
1237  * of extents
1238  *
1239  * @start is used to store the start of the free space if we find. But if we
1240  * don't find suitable free space, it will be used to store the start position
1241  * of the max free space.
1242  *
1243  * @len is used to store the size of the free space that we find.
1244  * But if we don't find suitable free space, it is used to store the size of
1245  * the max free space.
1246  */
1247 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1248                                struct btrfs_device *device, u64 num_bytes,
1249                                u64 search_start, u64 *start, u64 *len)
1250 {
1251         struct btrfs_key key;
1252         struct btrfs_root *root = device->dev_root;
1253         struct btrfs_dev_extent *dev_extent;
1254         struct btrfs_path *path;
1255         u64 hole_size;
1256         u64 max_hole_start;
1257         u64 max_hole_size;
1258         u64 extent_end;
1259         u64 search_end = device->total_bytes;
1260         int ret;
1261         int slot;
1262         struct extent_buffer *l;
1263         u64 min_search_start;
1264 
1265         /*
1266          * We don't want to overwrite the superblock on the drive nor any area
1267          * used by the boot loader (grub for example), so we make sure to start
1268          * at an offset of at least 1MB.
1269          */
1270         min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1271         search_start = max(search_start, min_search_start);
1272 
1273         path = btrfs_alloc_path();
1274         if (!path)
1275                 return -ENOMEM;
1276 
1277         max_hole_start = search_start;
1278         max_hole_size = 0;
1279 
1280 again:
1281         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1282                 ret = -ENOSPC;
1283                 goto out;
1284         }
1285 
1286         path->reada = READA_FORWARD;
1287         path->search_commit_root = 1;
1288         path->skip_locking = 1;
1289 
1290         key.objectid = device->devid;
1291         key.offset = search_start;
1292         key.type = BTRFS_DEV_EXTENT_KEY;
1293 
1294         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1295         if (ret < 0)
1296                 goto out;
1297         if (ret > 0) {
1298                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1299                 if (ret < 0)
1300                         goto out;
1301         }
1302 
1303         while (1) {
1304                 l = path->nodes[0];
1305                 slot = path->slots[0];
1306                 if (slot >= btrfs_header_nritems(l)) {
1307                         ret = btrfs_next_leaf(root, path);
1308                         if (ret == 0)
1309                                 continue;
1310                         if (ret < 0)
1311                                 goto out;
1312 
1313                         break;
1314                 }
1315                 btrfs_item_key_to_cpu(l, &key, slot);
1316 
1317                 if (key.objectid < device->devid)
1318                         goto next;
1319 
1320                 if (key.objectid > device->devid)
1321                         break;
1322 
1323                 if (key.type != BTRFS_DEV_EXTENT_KEY)
1324                         goto next;
1325 
1326                 if (key.offset > search_start) {
1327                         hole_size = key.offset - search_start;
1328 
1329                         /*
1330                          * Have to check before we set max_hole_start, otherwise
1331                          * we could end up sending back this offset anyway.
1332                          */
1333                         if (contains_pending_extent(transaction, device,
1334                                                     &search_start,
1335                                                     hole_size)) {
1336                                 if (key.offset >= search_start) {
1337                                         hole_size = key.offset - search_start;
1338                                 } else {
1339                                         WARN_ON_ONCE(1);
1340                                         hole_size = 0;
1341                                 }
1342                         }
1343 
1344                         if (hole_size > max_hole_size) {
1345                                 max_hole_start = search_start;
1346                                 max_hole_size = hole_size;
1347                         }
1348 
1349                         /*
1350                          * If this free space is greater than which we need,
1351                          * it must be the max free space that we have found
1352                          * until now, so max_hole_start must point to the start
1353                          * of this free space and the length of this free space
1354                          * is stored in max_hole_size. Thus, we return
1355                          * max_hole_start and max_hole_size and go back to the
1356                          * caller.
1357                          */
1358                         if (hole_size >= num_bytes) {
1359                                 ret = 0;
1360                                 goto out;
1361                         }
1362                 }
1363 
1364                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1365                 extent_end = key.offset + btrfs_dev_extent_length(l,
1366                                                                   dev_extent);
1367                 if (extent_end > search_start)
1368                         search_start = extent_end;
1369 next:
1370                 path->slots[0]++;
1371                 cond_resched();
1372         }
1373 
1374         /*
1375          * At this point, search_start should be the end of
1376          * allocated dev extents, and when shrinking the device,
1377          * search_end may be smaller than search_start.
1378          */
1379         if (search_end > search_start) {
1380                 hole_size = search_end - search_start;
1381 
1382                 if (contains_pending_extent(transaction, device, &search_start,
1383                                             hole_size)) {
1384                         btrfs_release_path(path);
1385                         goto again;
1386                 }
1387 
1388                 if (hole_size > max_hole_size) {
1389                         max_hole_start = search_start;
1390                         max_hole_size = hole_size;
1391                 }
1392         }
1393 
1394         /* See above. */
1395         if (max_hole_size < num_bytes)
1396                 ret = -ENOSPC;
1397         else
1398                 ret = 0;
1399 
1400 out:
1401         btrfs_free_path(path);
1402         *start = max_hole_start;
1403         if (len)
1404                 *len = max_hole_size;
1405         return ret;
1406 }
1407 
1408 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1409                          struct btrfs_device *device, u64 num_bytes,
1410                          u64 *start, u64 *len)
1411 {
1412         /* FIXME use last free of some kind */
1413         return find_free_dev_extent_start(trans->transaction, device,
1414                                           num_bytes, 0, start, len);
1415 }
1416 
1417 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1418                           struct btrfs_device *device,
1419                           u64 start, u64 *dev_extent_len)
1420 {
1421         int ret;
1422         struct btrfs_path *path;
1423         struct btrfs_root *root = device->dev_root;
1424         struct btrfs_key key;
1425         struct btrfs_key found_key;
1426         struct extent_buffer *leaf = NULL;
1427         struct btrfs_dev_extent *extent = NULL;
1428 
1429         path = btrfs_alloc_path();
1430         if (!path)
1431                 return -ENOMEM;
1432 
1433         key.objectid = device->devid;
1434         key.offset = start;
1435         key.type = BTRFS_DEV_EXTENT_KEY;
1436 again:
1437         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1438         if (ret > 0) {
1439                 ret = btrfs_previous_item(root, path, key.objectid,
1440                                           BTRFS_DEV_EXTENT_KEY);
1441                 if (ret)
1442                         goto out;
1443                 leaf = path->nodes[0];
1444                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1445                 extent = btrfs_item_ptr(leaf, path->slots[0],
1446                                         struct btrfs_dev_extent);
1447                 BUG_ON(found_key.offset > start || found_key.offset +
1448                        btrfs_dev_extent_length(leaf, extent) < start);
1449                 key = found_key;
1450                 btrfs_release_path(path);
1451                 goto again;
1452         } else if (ret == 0) {
1453                 leaf = path->nodes[0];
1454                 extent = btrfs_item_ptr(leaf, path->slots[0],
1455                                         struct btrfs_dev_extent);
1456         } else {
1457                 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1458                 goto out;
1459         }
1460 
1461         *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1462 
1463         ret = btrfs_del_item(trans, root, path);
1464         if (ret) {
1465                 btrfs_std_error(root->fs_info, ret,
1466                             "Failed to remove dev extent item");
1467         } else {
1468                 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1469         }
1470 out:
1471         btrfs_free_path(path);
1472         return ret;
1473 }
1474 
1475 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1476                                   struct btrfs_device *device,
1477                                   u64 chunk_tree, u64 chunk_objectid,
1478                                   u64 chunk_offset, u64 start, u64 num_bytes)
1479 {
1480         int ret;
1481         struct btrfs_path *path;
1482         struct btrfs_root *root = device->dev_root;
1483         struct btrfs_dev_extent *extent;
1484         struct extent_buffer *leaf;
1485         struct btrfs_key key;
1486 
1487         WARN_ON(!device->in_fs_metadata);
1488         WARN_ON(device->is_tgtdev_for_dev_replace);
1489         path = btrfs_alloc_path();
1490         if (!path)
1491                 return -ENOMEM;
1492 
1493         key.objectid = device->devid;
1494         key.offset = start;
1495         key.type = BTRFS_DEV_EXTENT_KEY;
1496         ret = btrfs_insert_empty_item(trans, root, path, &key,
1497                                       sizeof(*extent));
1498         if (ret)
1499                 goto out;
1500 
1501         leaf = path->nodes[0];
1502         extent = btrfs_item_ptr(leaf, path->slots[0],
1503                                 struct btrfs_dev_extent);
1504         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1505         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1506         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1507 
1508         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1509                     btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1510 
1511         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1512         btrfs_mark_buffer_dirty(leaf);
1513 out:
1514         btrfs_free_path(path);
1515         return ret;
1516 }
1517 
1518 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1519 {
1520         struct extent_map_tree *em_tree;
1521         struct extent_map *em;
1522         struct rb_node *n;
1523         u64 ret = 0;
1524 
1525         em_tree = &fs_info->mapping_tree.map_tree;
1526         read_lock(&em_tree->lock);
1527         n = rb_last(&em_tree->map);
1528         if (n) {
1529                 em = rb_entry(n, struct extent_map, rb_node);
1530                 ret = em->start + em->len;
1531         }
1532         read_unlock(&em_tree->lock);
1533 
1534         return ret;
1535 }
1536 
1537 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1538                                     u64 *devid_ret)
1539 {
1540         int ret;
1541         struct btrfs_key key;
1542         struct btrfs_key found_key;
1543         struct btrfs_path *path;
1544 
1545         path = btrfs_alloc_path();
1546         if (!path)
1547                 return -ENOMEM;
1548 
1549         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1550         key.type = BTRFS_DEV_ITEM_KEY;
1551         key.offset = (u64)-1;
1552 
1553         ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1554         if (ret < 0)
1555                 goto error;
1556 
1557         BUG_ON(ret == 0); /* Corruption */
1558 
1559         ret = btrfs_previous_item(fs_info->chunk_root, path,
1560                                   BTRFS_DEV_ITEMS_OBJECTID,
1561                                   BTRFS_DEV_ITEM_KEY);
1562         if (ret) {
1563                 *devid_ret = 1;
1564         } else {
1565                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1566                                       path->slots[0]);
1567                 *devid_ret = found_key.offset + 1;
1568         }
1569         ret = 0;
1570 error:
1571         btrfs_free_path(path);
1572         return ret;
1573 }
1574 
1575 /*
1576  * the device information is stored in the chunk root
1577  * the btrfs_device struct should be fully filled in
1578  */
1579 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1580                             struct btrfs_root *root,
1581                             struct btrfs_device *device)
1582 {
1583         int ret;
1584         struct btrfs_path *path;
1585         struct btrfs_dev_item *dev_item;
1586         struct extent_buffer *leaf;
1587         struct btrfs_key key;
1588         unsigned long ptr;
1589 
1590         root = root->fs_info->chunk_root;
1591 
1592         path = btrfs_alloc_path();
1593         if (!path)
1594                 return -ENOMEM;
1595 
1596         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1597         key.type = BTRFS_DEV_ITEM_KEY;
1598         key.offset = device->devid;
1599 
1600         ret = btrfs_insert_empty_item(trans, root, path, &key,
1601                                       sizeof(*dev_item));
1602         if (ret)
1603                 goto out;
1604 
1605         leaf = path->nodes[0];
1606         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1607 
1608         btrfs_set_device_id(leaf, dev_item, device->devid);
1609         btrfs_set_device_generation(leaf, dev_item, 0);
1610         btrfs_set_device_type(leaf, dev_item, device->type);
1611         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1612         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1613         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1614         btrfs_set_device_total_bytes(leaf, dev_item,
1615                                      btrfs_device_get_disk_total_bytes(device));
1616         btrfs_set_device_bytes_used(leaf, dev_item,
1617                                     btrfs_device_get_bytes_used(device));
1618         btrfs_set_device_group(leaf, dev_item, 0);
1619         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1620         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1621         btrfs_set_device_start_offset(leaf, dev_item, 0);
1622 
1623         ptr = btrfs_device_uuid(dev_item);
1624         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1625         ptr = btrfs_device_fsid(dev_item);
1626         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1627         btrfs_mark_buffer_dirty(leaf);
1628 
1629         ret = 0;
1630 out:
1631         btrfs_free_path(path);
1632         return ret;
1633 }
1634 
1635 /*
1636  * Function to update ctime/mtime for a given device path.
1637  * Mainly used for ctime/mtime based probe like libblkid.
1638  */
1639 static void update_dev_time(char *path_name)
1640 {
1641         struct file *filp;
1642 
1643         filp = filp_open(path_name, O_RDWR, 0);
1644         if (IS_ERR(filp))
1645                 return;
1646         file_update_time(filp);
1647         filp_close(filp, NULL);
1648 }
1649 
1650 static int btrfs_rm_dev_item(struct btrfs_root *root,
1651                              struct btrfs_device *device)
1652 {
1653         int ret;
1654         struct btrfs_path *path;
1655         struct btrfs_key key;
1656         struct btrfs_trans_handle *trans;
1657 
1658         root = root->fs_info->chunk_root;
1659 
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663 
1664         trans = btrfs_start_transaction(root, 0);
1665         if (IS_ERR(trans)) {
1666                 btrfs_free_path(path);
1667                 return PTR_ERR(trans);
1668         }
1669         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1670         key.type = BTRFS_DEV_ITEM_KEY;
1671         key.offset = device->devid;
1672 
1673         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1674         if (ret < 0)
1675                 goto out;
1676 
1677         if (ret > 0) {
1678                 ret = -ENOENT;
1679                 goto out;
1680         }
1681 
1682         ret = btrfs_del_item(trans, root, path);
1683         if (ret)
1684                 goto out;
1685 out:
1686         btrfs_free_path(path);
1687         btrfs_commit_transaction(trans, root);
1688         return ret;
1689 }
1690 
1691 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1692 {
1693         struct btrfs_device *device;
1694         struct btrfs_device *next_device;
1695         struct block_device *bdev;
1696         struct buffer_head *bh = NULL;
1697         struct btrfs_super_block *disk_super;
1698         struct btrfs_fs_devices *cur_devices;
1699         u64 all_avail;
1700         u64 devid;
1701         u64 num_devices;
1702         u8 *dev_uuid;
1703         unsigned seq;
1704         int ret = 0;
1705         bool clear_super = false;
1706 
1707         mutex_lock(&uuid_mutex);
1708 
1709         do {
1710                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1711 
1712                 all_avail = root->fs_info->avail_data_alloc_bits |
1713                             root->fs_info->avail_system_alloc_bits |
1714                             root->fs_info->avail_metadata_alloc_bits;
1715         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1716 
1717         num_devices = root->fs_info->fs_devices->num_devices;
1718         btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1719         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1720                 WARN_ON(num_devices < 1);
1721                 num_devices--;
1722         }
1723         btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1724 
1725         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1726                 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1727                 goto out;
1728         }
1729 
1730         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1731                 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1732                 goto out;
1733         }
1734 
1735         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1736             root->fs_info->fs_devices->rw_devices <= 2) {
1737                 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1738                 goto out;
1739         }
1740         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1741             root->fs_info->fs_devices->rw_devices <= 3) {
1742                 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1743                 goto out;
1744         }
1745 
1746         if (strcmp(device_path, "missing") == 0) {
1747                 struct list_head *devices;
1748                 struct btrfs_device *tmp;
1749 
1750                 device = NULL;
1751                 devices = &root->fs_info->fs_devices->devices;
1752                 /*
1753                  * It is safe to read the devices since the volume_mutex
1754                  * is held.
1755                  */
1756                 list_for_each_entry(tmp, devices, dev_list) {
1757                         if (tmp->in_fs_metadata &&
1758                             !tmp->is_tgtdev_for_dev_replace &&
1759                             !tmp->bdev) {
1760                                 device = tmp;
1761                                 break;
1762                         }
1763                 }
1764                 bdev = NULL;
1765                 bh = NULL;
1766                 disk_super = NULL;
1767                 if (!device) {
1768                         ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1769                         goto out;
1770                 }
1771         } else {
1772                 ret = btrfs_get_bdev_and_sb(device_path,
1773                                             FMODE_WRITE | FMODE_EXCL,
1774                                             root->fs_info->bdev_holder, 0,
1775                                             &bdev, &bh);
1776                 if (ret)
1777                         goto out;
1778                 disk_super = (struct btrfs_super_block *)bh->b_data;
1779                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1780                 dev_uuid = disk_super->dev_item.uuid;
1781                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1782                                            disk_super->fsid);
1783                 if (!device) {
1784                         ret = -ENOENT;
1785                         goto error_brelse;
1786                 }
1787         }
1788 
1789         if (device->is_tgtdev_for_dev_replace) {
1790                 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1791                 goto error_brelse;
1792         }
1793 
1794         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1795                 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1796                 goto error_brelse;
1797         }
1798 
1799         if (device->writeable) {
1800                 lock_chunks(root);
1801                 list_del_init(&device->dev_alloc_list);
1802                 device->fs_devices->rw_devices--;
1803                 unlock_chunks(root);
1804                 clear_super = true;
1805         }
1806 
1807         mutex_unlock(&uuid_mutex);
1808         ret = btrfs_shrink_device(device, 0);
1809         mutex_lock(&uuid_mutex);
1810         if (ret)
1811                 goto error_undo;
1812 
1813         /*
1814          * TODO: the superblock still includes this device in its num_devices
1815          * counter although write_all_supers() is not locked out. This
1816          * could give a filesystem state which requires a degraded mount.
1817          */
1818         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1819         if (ret)
1820                 goto error_undo;
1821 
1822         device->in_fs_metadata = 0;
1823         btrfs_scrub_cancel_dev(root->fs_info, device);
1824 
1825         /*
1826          * the device list mutex makes sure that we don't change
1827          * the device list while someone else is writing out all
1828          * the device supers. Whoever is writing all supers, should
1829          * lock the device list mutex before getting the number of
1830          * devices in the super block (super_copy). Conversely,
1831          * whoever updates the number of devices in the super block
1832          * (super_copy) should hold the device list mutex.
1833          */
1834 
1835         cur_devices = device->fs_devices;
1836         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1837         list_del_rcu(&device->dev_list);
1838 
1839         device->fs_devices->num_devices--;
1840         device->fs_devices->total_devices--;
1841 
1842         if (device->missing)
1843                 device->fs_devices->missing_devices--;
1844 
1845         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1846                                  struct btrfs_device, dev_list);
1847         if (device->bdev == root->fs_info->sb->s_bdev)
1848                 root->fs_info->sb->s_bdev = next_device->bdev;
1849         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1850                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1851 
1852         if (device->bdev) {
1853                 device->fs_devices->open_devices--;
1854                 /* remove sysfs entry */
1855                 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1856         }
1857 
1858         call_rcu(&device->rcu, free_device);
1859 
1860         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1861         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1862         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1863 
1864         if (cur_devices->open_devices == 0) {
1865                 struct btrfs_fs_devices *fs_devices;
1866                 fs_devices = root->fs_info->fs_devices;
1867                 while (fs_devices) {
1868                         if (fs_devices->seed == cur_devices) {
1869                                 fs_devices->seed = cur_devices->seed;
1870                                 break;
1871                         }
1872                         fs_devices = fs_devices->seed;
1873                 }
1874                 cur_devices->seed = NULL;
1875                 __btrfs_close_devices(cur_devices);
1876                 free_fs_devices(cur_devices);
1877         }
1878 
1879         root->fs_info->num_tolerated_disk_barrier_failures =
1880                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1881 
1882         /*
1883          * at this point, the device is zero sized.  We want to
1884          * remove it from the devices list and zero out the old super
1885          */
1886         if (clear_super && disk_super) {
1887                 u64 bytenr;
1888                 int i;
1889 
1890                 /* make sure this device isn't detected as part of
1891                  * the FS anymore
1892                  */
1893                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1894                 set_buffer_dirty(bh);
1895                 sync_dirty_buffer(bh);
1896 
1897                 /* clear the mirror copies of super block on the disk
1898                  * being removed, 0th copy is been taken care above and
1899                  * the below would take of the rest
1900                  */
1901                 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1902                         bytenr = btrfs_sb_offset(i);
1903                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1904                                         i_size_read(bdev->bd_inode))
1905                                 break;
1906 
1907                         brelse(bh);
1908                         bh = __bread(bdev, bytenr / 4096,
1909                                         BTRFS_SUPER_INFO_SIZE);
1910                         if (!bh)
1911                                 continue;
1912 
1913                         disk_super = (struct btrfs_super_block *)bh->b_data;
1914 
1915                         if (btrfs_super_bytenr(disk_super) != bytenr ||
1916                                 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1917                                 continue;
1918                         }
1919                         memset(&disk_super->magic, 0,
1920                                                 sizeof(disk_super->magic));
1921                         set_buffer_dirty(bh);
1922                         sync_dirty_buffer(bh);
1923                 }
1924         }
1925 
1926         ret = 0;
1927 
1928         if (bdev) {
1929                 /* Notify udev that device has changed */
1930                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1931 
1932                 /* Update ctime/mtime for device path for libblkid */
1933                 update_dev_time(device_path);
1934         }
1935 
1936 error_brelse:
1937         brelse(bh);
1938         if (bdev)
1939                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1940 out:
1941         mutex_unlock(&uuid_mutex);
1942         return ret;
1943 error_undo:
1944         if (device->writeable) {
1945                 lock_chunks(root);
1946                 list_add(&device->dev_alloc_list,
1947                          &root->fs_info->fs_devices->alloc_list);
1948                 device->fs_devices->rw_devices++;
1949                 unlock_chunks(root);
1950         }
1951         goto error_brelse;
1952 }
1953 
1954 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1955                                         struct btrfs_device *srcdev)
1956 {
1957         struct btrfs_fs_devices *fs_devices;
1958 
1959         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1960 
1961         /*
1962          * in case of fs with no seed, srcdev->fs_devices will point
1963          * to fs_devices of fs_info. However when the dev being replaced is
1964          * a seed dev it will point to the seed's local fs_devices. In short
1965          * srcdev will have its correct fs_devices in both the cases.
1966          */
1967         fs_devices = srcdev->fs_devices;
1968 
1969         list_del_rcu(&srcdev->dev_list);
1970         list_del_rcu(&srcdev->dev_alloc_list);
1971         fs_devices->num_devices--;
1972         if (srcdev->missing)
1973                 fs_devices->missing_devices--;
1974 
1975         if (srcdev->writeable) {
1976                 fs_devices->rw_devices--;
1977                 /* zero out the old super if it is writable */
1978                 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1979         }
1980 
1981         if (srcdev->bdev)
1982                 fs_devices->open_devices--;
1983 }
1984 
1985 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1986                                       struct btrfs_device *srcdev)
1987 {
1988         struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1989 
1990         call_rcu(&srcdev->rcu, free_device);
1991 
1992         /*
1993          * unless fs_devices is seed fs, num_devices shouldn't go
1994          * zero
1995          */
1996         BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1997 
1998         /* if this is no devs we rather delete the fs_devices */
1999         if (!fs_devices->num_devices) {
2000                 struct btrfs_fs_devices *tmp_fs_devices;
2001 
2002                 tmp_fs_devices = fs_info->fs_devices;
2003                 while (tmp_fs_devices) {
2004                         if (tmp_fs_devices->seed == fs_devices) {
2005                                 tmp_fs_devices->seed = fs_devices->seed;
2006                                 break;
2007                         }
2008                         tmp_fs_devices = tmp_fs_devices->seed;
2009                 }
2010                 fs_devices->seed = NULL;
2011                 __btrfs_close_devices(fs_devices);
2012                 free_fs_devices(fs_devices);
2013         }
2014 }
2015 
2016 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2017                                       struct btrfs_device *tgtdev)
2018 {
2019         struct btrfs_device *next_device;
2020 
2021         mutex_lock(&uuid_mutex);
2022         WARN_ON(!tgtdev);
2023         mutex_lock(&fs_info->fs_devices->device_list_mutex);
2024 
2025         btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2026 
2027         if (tgtdev->bdev) {
2028                 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2029                 fs_info->fs_devices->open_devices--;
2030         }
2031         fs_info->fs_devices->num_devices--;
2032 
2033         next_device = list_entry(fs_info->fs_devices->devices.next,
2034                                  struct btrfs_device, dev_list);
2035         if (tgtdev->bdev == fs_info->sb->s_bdev)
2036                 fs_info->sb->s_bdev = next_device->bdev;
2037         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038                 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039         list_del_rcu(&tgtdev->dev_list);
2040 
2041         call_rcu(&tgtdev->rcu, free_device);
2042 
2043         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044         mutex_unlock(&uuid_mutex);
2045 }
2046 
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048                                      struct btrfs_device **device)
2049 {
2050         int ret = 0;
2051         struct btrfs_super_block *disk_super;
2052         u64 devid;
2053         u8 *dev_uuid;
2054         struct block_device *bdev;
2055         struct buffer_head *bh;
2056 
2057         *device = NULL;
2058         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
2060         if (ret)
2061                 return ret;
2062         disk_super = (struct btrfs_super_block *)bh->b_data;
2063         devid = btrfs_stack_device_id(&disk_super->dev_item);
2064         dev_uuid = disk_super->dev_item.uuid;
2065         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066                                     disk_super->fsid);
2067         brelse(bh);
2068         if (!*device)
2069                 ret = -ENOENT;
2070         blkdev_put(bdev, FMODE_READ);
2071         return ret;
2072 }
2073 
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075                                          char *device_path,
2076                                          struct btrfs_device **device)
2077 {
2078         *device = NULL;
2079         if (strcmp(device_path, "missing") == 0) {
2080                 struct list_head *devices;
2081                 struct btrfs_device *tmp;
2082 
2083                 devices = &root->fs_info->fs_devices->devices;
2084                 /*
2085                  * It is safe to read the devices since the volume_mutex
2086                  * is held by the caller.
2087                  */
2088                 list_for_each_entry(tmp, devices, dev_list) {
2089                         if (tmp->in_fs_metadata && !tmp->bdev) {
2090                                 *device = tmp;
2091                                 break;
2092                         }
2093                 }
2094 
2095                 if (!*device)
2096                         return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2097 
2098                 return 0;
2099         } else {
2100                 return btrfs_find_device_by_path(root, device_path, device);
2101         }
2102 }
2103 
2104 /*
2105  * does all the dirty work required for changing file system's UUID.
2106  */
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2108 {
2109         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110         struct btrfs_fs_devices *old_devices;
2111         struct btrfs_fs_devices *seed_devices;
2112         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113         struct btrfs_device *device;
2114         u64 super_flags;
2115 
2116         BUG_ON(!mutex_is_locked(&uuid_mutex));
2117         if (!fs_devices->seeding)
2118                 return -EINVAL;
2119 
2120         seed_devices = __alloc_fs_devices();
2121         if (IS_ERR(seed_devices))
2122                 return PTR_ERR(seed_devices);
2123 
2124         old_devices = clone_fs_devices(fs_devices);
2125         if (IS_ERR(old_devices)) {
2126                 kfree(seed_devices);
2127                 return PTR_ERR(old_devices);
2128         }
2129 
2130         list_add(&old_devices->list, &fs_uuids);
2131 
2132         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133         seed_devices->opened = 1;
2134         INIT_LIST_HEAD(&seed_devices->devices);
2135         INIT_LIST_HEAD(&seed_devices->alloc_list);
2136         mutex_init(&seed_devices->device_list_mutex);
2137 
2138         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2140                               synchronize_rcu);
2141         list_for_each_entry(device, &seed_devices->devices, dev_list)
2142                 device->fs_devices = seed_devices;
2143 
2144         lock_chunks(root);
2145         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146         unlock_chunks(root);
2147 
2148         fs_devices->seeding = 0;
2149         fs_devices->num_devices = 0;
2150         fs_devices->open_devices = 0;
2151         fs_devices->missing_devices = 0;
2152         fs_devices->rotating = 0;
2153         fs_devices->seed = seed_devices;
2154 
2155         generate_random_uuid(fs_devices->fsid);
2156         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2159 
2160         super_flags = btrfs_super_flags(disk_super) &
2161                       ~BTRFS_SUPER_FLAG_SEEDING;
2162         btrfs_set_super_flags(disk_super, super_flags);
2163 
2164         return 0;
2165 }
2166 
2167 /*
2168  * strore the expected generation for seed devices in device items.
2169  */
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171                                struct btrfs_root *root)
2172 {
2173         struct btrfs_path *path;
2174         struct extent_buffer *leaf;
2175         struct btrfs_dev_item *dev_item;
2176         struct btrfs_device *device;
2177         struct btrfs_key key;
2178         u8 fs_uuid[BTRFS_UUID_SIZE];
2179         u8 dev_uuid[BTRFS_UUID_SIZE];
2180         u64 devid;
2181         int ret;
2182 
2183         path = btrfs_alloc_path();
2184         if (!path)
2185                 return -ENOMEM;
2186 
2187         root = root->fs_info->chunk_root;
2188         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2189         key.offset = 0;
2190         key.type = BTRFS_DEV_ITEM_KEY;
2191 
2192         while (1) {
2193                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2194                 if (ret < 0)
2195                         goto error;
2196 
2197                 leaf = path->nodes[0];
2198 next_slot:
2199                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200                         ret = btrfs_next_leaf(root, path);
2201                         if (ret > 0)
2202                                 break;
2203                         if (ret < 0)
2204                                 goto error;
2205                         leaf = path->nodes[0];
2206                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207                         btrfs_release_path(path);
2208                         continue;
2209                 }
2210 
2211                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213                     key.type != BTRFS_DEV_ITEM_KEY)
2214                         break;
2215 
2216                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217                                           struct btrfs_dev_item);
2218                 devid = btrfs_device_id(leaf, dev_item);
2219                 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2220                                    BTRFS_UUID_SIZE);
2221                 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2222                                    BTRFS_UUID_SIZE);
2223                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2224                                            fs_uuid);
2225                 BUG_ON(!device); /* Logic error */
2226 
2227                 if (device->fs_devices->seeding) {
2228                         btrfs_set_device_generation(leaf, dev_item,
2229                                                     device->generation);
2230                         btrfs_mark_buffer_dirty(leaf);
2231                 }
2232 
2233                 path->slots[0]++;
2234                 goto next_slot;
2235         }
2236         ret = 0;
2237 error:
2238         btrfs_free_path(path);
2239         return ret;
2240 }
2241 
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2243 {
2244         struct request_queue *q;
2245         struct btrfs_trans_handle *trans;
2246         struct btrfs_device *device;
2247         struct block_device *bdev;
2248         struct list_head *devices;
2249         struct super_block *sb = root->fs_info->sb;
2250         struct rcu_string *name;
2251         u64 tmp;
2252         int seeding_dev = 0;
2253         int ret = 0;
2254 
2255         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2256                 return -EROFS;
2257 
2258         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259                                   root->fs_info->bdev_holder);
2260         if (IS_ERR(bdev))
2261                 return PTR_ERR(bdev);
2262 
2263         if (root->fs_info->fs_devices->seeding) {
2264                 seeding_dev = 1;
2265                 down_write(&sb->s_umount);
2266                 mutex_lock(&uuid_mutex);
2267         }
2268 
2269         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2270 
2271         devices = &root->fs_info->fs_devices->devices;
2272 
2273         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274         list_for_each_entry(device, devices, dev_list) {
2275                 if (device->bdev == bdev) {
2276                         ret = -EEXIST;
2277                         mutex_unlock(
2278                                 &root->fs_info->fs_devices->device_list_mutex);
2279                         goto error;
2280                 }
2281         }
2282         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2283 
2284         device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285         if (IS_ERR(device)) {
2286                 /* we can safely leave the fs_devices entry around */
2287                 ret = PTR_ERR(device);
2288                 goto error;
2289         }
2290 
2291         name = rcu_string_strdup(device_path, GFP_KERNEL);
2292         if (!name) {
2293                 kfree(device);
2294                 ret = -ENOMEM;
2295                 goto error;
2296         }
2297         rcu_assign_pointer(device->name, name);
2298 
2299         trans = btrfs_start_transaction(root, 0);
2300         if (IS_ERR(trans)) {
2301                 rcu_string_free(device->name);
2302                 kfree(device);
2303                 ret = PTR_ERR(trans);
2304                 goto error;
2305         }
2306 
2307         q = bdev_get_queue(bdev);
2308         if (blk_queue_discard(q))
2309                 device->can_discard = 1;
2310         device->writeable = 1;
2311         device->generation = trans->transid;
2312         device->io_width = root->sectorsize;
2313         device->io_align = root->sectorsize;
2314         device->sector_size = root->sectorsize;
2315         device->total_bytes = i_size_read(bdev->bd_inode);
2316         device->disk_total_bytes = device->total_bytes;
2317         device->commit_total_bytes = device->total_bytes;
2318         device->dev_root = root->fs_info->dev_root;
2319         device->bdev = bdev;
2320         device->in_fs_metadata = 1;
2321         device->is_tgtdev_for_dev_replace = 0;
2322         device->mode = FMODE_EXCL;
2323         device->dev_stats_valid = 1;
2324         set_blocksize(device->bdev, 4096);
2325 
2326         if (seeding_dev) {
2327                 sb->s_flags &= ~MS_RDONLY;
2328                 ret = btrfs_prepare_sprout(root);
2329                 BUG_ON(ret); /* -ENOMEM */
2330         }
2331 
2332         device->fs_devices = root->fs_info->fs_devices;
2333 
2334         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2335         lock_chunks(root);
2336         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2337         list_add(&device->dev_alloc_list,
2338                  &root->fs_info->fs_devices->alloc_list);
2339         root->fs_info->fs_devices->num_devices++;
2340         root->fs_info->fs_devices->open_devices++;
2341         root->fs_info->fs_devices->rw_devices++;
2342         root->fs_info->fs_devices->total_devices++;
2343         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2344 
2345         spin_lock(&root->fs_info->free_chunk_lock);
2346         root->fs_info->free_chunk_space += device->total_bytes;
2347         spin_unlock(&root->fs_info->free_chunk_lock);
2348 
2349         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2350                 root->fs_info->fs_devices->rotating = 1;
2351 
2352         tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2353         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2354                                     tmp + device->total_bytes);
2355 
2356         tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2357         btrfs_set_super_num_devices(root->fs_info->super_copy,
2358                                     tmp + 1);
2359 
2360         /* add sysfs device entry */
2361         btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2362 
2363         /*
2364          * we've got more storage, clear any full flags on the space
2365          * infos
2366          */
2367         btrfs_clear_space_info_full(root->fs_info);
2368 
2369         unlock_chunks(root);
2370         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2371 
2372         if (seeding_dev) {
2373                 lock_chunks(root);
2374                 ret = init_first_rw_device(trans, root, device);
2375                 unlock_chunks(root);
2376                 if (ret) {
2377                         btrfs_abort_transaction(trans, root, ret);
2378                         goto error_trans;
2379                 }
2380         }
2381 
2382         ret = btrfs_add_device(trans, root, device);
2383         if (ret) {
2384                 btrfs_abort_transaction(trans, root, ret);
2385                 goto error_trans;
2386         }
2387 
2388         if (seeding_dev) {
2389                 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2390 
2391                 ret = btrfs_finish_sprout(trans, root);
2392                 if (ret) {
2393                         btrfs_abort_transaction(trans, root, ret);
2394                         goto error_trans;
2395                 }
2396 
2397                 /* Sprouting would change fsid of the mounted root,
2398                  * so rename the fsid on the sysfs
2399                  */
2400                 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2401                                                 root->fs_info->fsid);
2402                 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2403                                                                 fsid_buf))
2404                         btrfs_warn(root->fs_info,
2405                                 "sysfs: failed to create fsid for sprout");
2406         }
2407 
2408         root->fs_info->num_tolerated_disk_barrier_failures =
2409                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2410         ret = btrfs_commit_transaction(trans, root);
2411 
2412         if (seeding_dev) {
2413                 mutex_unlock(&uuid_mutex);
2414                 up_write(&sb->s_umount);
2415 
2416                 if (ret) /* transaction commit */
2417                         return ret;
2418 
2419                 ret = btrfs_relocate_sys_chunks(root);
2420                 if (ret < 0)
2421                         btrfs_std_error(root->fs_info, ret,
2422                                     "Failed to relocate sys chunks after "
2423                                     "device initialization. This can be fixed "
2424                                     "using the \"btrfs balance\" command.");
2425                 trans = btrfs_attach_transaction(root);
2426                 if (IS_ERR(trans)) {
2427                         if (PTR_ERR(trans) == -ENOENT)
2428                                 return 0;
2429                         return PTR_ERR(trans);
2430                 }
2431                 ret = btrfs_commit_transaction(trans, root);
2432         }
2433 
2434         /* Update ctime/mtime for libblkid */
2435         update_dev_time(device_path);
2436         return ret;
2437 
2438 error_trans:
2439         btrfs_end_transaction(trans, root);
2440         rcu_string_free(device->name);
2441         btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2442         kfree(device);
2443 error:
2444         blkdev_put(bdev, FMODE_EXCL);
2445         if (seeding_dev) {
2446                 mutex_unlock(&uuid_mutex);
2447                 up_write(&sb->s_umount);
2448         }
2449         return ret;
2450 }
2451 
2452 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2453                                   struct btrfs_device *srcdev,
2454                                   struct btrfs_device **device_out)
2455 {
2456         struct request_queue *q;
2457         struct btrfs_device *device;
2458         struct block_device *bdev;
2459         struct btrfs_fs_info *fs_info = root->fs_info;
2460         struct list_head *devices;
2461         struct rcu_string *name;
2462         u64 devid = BTRFS_DEV_REPLACE_DEVID;
2463         int ret = 0;
2464 
2465         *device_out = NULL;
2466         if (fs_info->fs_devices->seeding) {
2467                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2468                 return -EINVAL;
2469         }
2470 
2471         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2472                                   fs_info->bdev_holder);
2473         if (IS_ERR(bdev)) {
2474                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2475                 return PTR_ERR(bdev);
2476         }
2477 
2478         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2479 
2480         devices = &fs_info->fs_devices->devices;
2481         list_for_each_entry(device, devices, dev_list) {
2482                 if (device->bdev == bdev) {
2483                         btrfs_err(fs_info, "target device is in the filesystem!");
2484                         ret = -EEXIST;
2485                         goto error;
2486                 }
2487         }
2488 
2489 
2490         if (i_size_read(bdev->bd_inode) <
2491             btrfs_device_get_total_bytes(srcdev)) {
2492                 btrfs_err(fs_info, "target device is smaller than source device!");
2493                 ret = -EINVAL;
2494                 goto error;
2495         }
2496 
2497 
2498         device = btrfs_alloc_device(NULL, &devid, NULL);
2499         if (IS_ERR(device)) {
2500                 ret = PTR_ERR(device);
2501                 goto error;
2502         }
2503 
2504         name = rcu_string_strdup(device_path, GFP_NOFS);
2505         if (!name) {
2506                 kfree(device);
2507                 ret = -ENOMEM;
2508                 goto error;
2509         }
2510         rcu_assign_pointer(device->name, name);
2511 
2512         q = bdev_get_queue(bdev);
2513         if (blk_queue_discard(q))
2514                 device->can_discard = 1;
2515         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2516         device->writeable = 1;
2517         device->generation = 0;
2518         device->io_width = root->sectorsize;
2519         device->io_align = root->sectorsize;
2520         device->sector_size = root->sectorsize;
2521         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2522         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2523         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2524         ASSERT(list_empty(&srcdev->resized_list));
2525         device->commit_total_bytes = srcdev->commit_total_bytes;
2526         device->commit_bytes_used = device->bytes_used;
2527         device->dev_root = fs_info->dev_root;
2528         device->bdev = bdev;
2529         device->in_fs_metadata = 1;
2530         device->is_tgtdev_for_dev_replace = 1;
2531         device->mode = FMODE_EXCL;
2532         device->dev_stats_valid = 1;
2533         set_blocksize(device->bdev, 4096);
2534         device->fs_devices = fs_info->fs_devices;
2535         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2536         fs_info->fs_devices->num_devices++;
2537         fs_info->fs_devices->open_devices++;
2538         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2539 
2540         *device_out = device;
2541         return ret;
2542 
2543 error:
2544         blkdev_put(bdev, FMODE_EXCL);
2545         return ret;
2546 }
2547 
2548 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2549                                               struct btrfs_device *tgtdev)
2550 {
2551         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2552         tgtdev->io_width = fs_info->dev_root->sectorsize;
2553         tgtdev->io_align = fs_info->dev_root->sectorsize;
2554         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2555         tgtdev->dev_root = fs_info->dev_root;
2556         tgtdev->in_fs_metadata = 1;
2557 }
2558 
2559 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2560                                         struct btrfs_device *device)
2561 {
2562         int ret;
2563         struct btrfs_path *path;
2564         struct btrfs_root *root;
2565         struct btrfs_dev_item *dev_item;
2566         struct extent_buffer *leaf;
2567         struct btrfs_key key;
2568 
2569         root = device->dev_root->fs_info->chunk_root;
2570 
2571         path = btrfs_alloc_path();
2572         if (!path)
2573                 return -ENOMEM;
2574 
2575         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2576         key.type = BTRFS_DEV_ITEM_KEY;
2577         key.offset = device->devid;
2578 
2579         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2580         if (ret < 0)
2581                 goto out;
2582 
2583         if (ret > 0) {
2584                 ret = -ENOENT;
2585                 goto out;
2586         }
2587 
2588         leaf = path->nodes[0];
2589         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2590 
2591         btrfs_set_device_id(leaf, dev_item, device->devid);
2592         btrfs_set_device_type(leaf, dev_item, device->type);
2593         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2594         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2595         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2596         btrfs_set_device_total_bytes(leaf, dev_item,
2597                                      btrfs_device_get_disk_total_bytes(device));
2598         btrfs_set_device_bytes_used(leaf, dev_item,
2599                                     btrfs_device_get_bytes_used(device));
2600         btrfs_mark_buffer_dirty(leaf);
2601 
2602 out:
2603         btrfs_free_path(path);
2604         return ret;
2605 }
2606 
2607 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2608                       struct btrfs_device *device, u64 new_size)
2609 {
2610         struct btrfs_super_block *super_copy =
2611                 device->dev_root->fs_info->super_copy;
2612         struct btrfs_fs_devices *fs_devices;
2613         u64 old_total;
2614         u64 diff;
2615 
2616         if (!device->writeable)
2617                 return -EACCES;
2618 
2619         lock_chunks(device->dev_root);
2620         old_total = btrfs_super_total_bytes(super_copy);
2621         diff = new_size - device->total_bytes;
2622 
2623         if (new_size <= device->total_bytes ||
2624             device->is_tgtdev_for_dev_replace) {
2625                 unlock_chunks(device->dev_root);
2626                 return -EINVAL;
2627         }
2628 
2629         fs_devices = device->dev_root->fs_info->fs_devices;
2630 
2631         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2632         device->fs_devices->total_rw_bytes += diff;
2633 
2634         btrfs_device_set_total_bytes(device, new_size);
2635         btrfs_device_set_disk_total_bytes(device, new_size);
2636         btrfs_clear_space_info_full(device->dev_root->fs_info);
2637         if (list_empty(&device->resized_list))
2638                 list_add_tail(&device->resized_list,
2639                               &fs_devices->resized_devices);
2640         unlock_chunks(device->dev_root);
2641 
2642         return btrfs_update_device(trans, device);
2643 }
2644 
2645 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2646                             struct btrfs_root *root, u64 chunk_objectid,
2647                             u64 chunk_offset)
2648 {
2649         int ret;
2650         struct btrfs_path *path;
2651         struct btrfs_key key;
2652 
2653         root = root->fs_info->chunk_root;
2654         path = btrfs_alloc_path();
2655         if (!path)
2656                 return -ENOMEM;
2657 
2658         key.objectid = chunk_objectid;
2659         key.offset = chunk_offset;
2660         key.type = BTRFS_CHUNK_ITEM_KEY;
2661 
2662         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2663         if (ret < 0)
2664                 goto out;
2665         else if (ret > 0) { /* Logic error or corruption */
2666                 btrfs_std_error(root->fs_info, -ENOENT,
2667                             "Failed lookup while freeing chunk.");
2668                 ret = -ENOENT;
2669                 goto out;
2670         }
2671 
2672         ret = btrfs_del_item(trans, root, path);
2673         if (ret < 0)
2674                 btrfs_std_error(root->fs_info, ret,
2675                             "Failed to delete chunk item.");
2676 out:
2677         btrfs_free_path(path);
2678         return ret;
2679 }
2680 
2681 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2682                         chunk_offset)
2683 {
2684         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2685         struct btrfs_disk_key *disk_key;
2686         struct btrfs_chunk *chunk;
2687         u8 *ptr;
2688         int ret = 0;
2689         u32 num_stripes;
2690         u32 array_size;
2691         u32 len = 0;
2692         u32 cur;
2693         struct btrfs_key key;
2694 
2695         lock_chunks(root);
2696         array_size = btrfs_super_sys_array_size(super_copy);
2697 
2698         ptr = super_copy->sys_chunk_array;
2699         cur = 0;
2700 
2701         while (cur < array_size) {
2702                 disk_key = (struct btrfs_disk_key *)ptr;
2703                 btrfs_disk_key_to_cpu(&key, disk_key);
2704 
2705                 len = sizeof(*disk_key);
2706 
2707                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2708                         chunk = (struct btrfs_chunk *)(ptr + len);
2709                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2710                         len += btrfs_chunk_item_size(num_stripes);
2711                 } else {
2712                         ret = -EIO;
2713                         break;
2714                 }
2715                 if (key.objectid == chunk_objectid &&
2716                     key.offset == chunk_offset) {
2717                         memmove(ptr, ptr + len, array_size - (cur + len));
2718                         array_size -= len;
2719                         btrfs_set_super_sys_array_size(super_copy, array_size);
2720                 } else {
2721                         ptr += len;
2722                         cur += len;
2723                 }
2724         }
2725         unlock_chunks(root);
2726         return ret;
2727 }
2728 
2729 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2730                        struct btrfs_root *root, u64 chunk_offset)
2731 {
2732         struct extent_map_tree *em_tree;
2733         struct extent_map *em;
2734         struct btrfs_root *extent_root = root->fs_info->extent_root;
2735         struct map_lookup *map;
2736         u64 dev_extent_len = 0;
2737         u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2738         int i, ret = 0;
2739 
2740         /* Just in case */
2741         root = root->fs_info->chunk_root;
2742         em_tree = &root->fs_info->mapping_tree.map_tree;
2743 
2744         read_lock(&em_tree->lock);
2745         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2746         read_unlock(&em_tree->lock);
2747 
2748         if (!em || em->start > chunk_offset ||
2749             em->start + em->len < chunk_offset) {
2750                 /*
2751                  * This is a logic error, but we don't want to just rely on the
2752                  * user having built with ASSERT enabled, so if ASSERT doesn't
2753                  * do anything we still error out.
2754                  */
2755                 ASSERT(0);
2756                 if (em)
2757                         free_extent_map(em);
2758                 return -EINVAL;
2759         }
2760         map = em->map_lookup;
2761         lock_chunks(root->fs_info->chunk_root);
2762         check_system_chunk(trans, extent_root, map->type);
2763         unlock_chunks(root->fs_info->chunk_root);
2764 
2765         for (i = 0; i < map->num_stripes; i++) {
2766                 struct btrfs_device *device = map->stripes[i].dev;
2767                 ret = btrfs_free_dev_extent(trans, device,
2768                                             map->stripes[i].physical,
2769                                             &dev_extent_len);
2770                 if (ret) {
2771                         btrfs_abort_transaction(trans, root, ret);
2772                         goto out;
2773                 }
2774 
2775                 if (device->bytes_used > 0) {
2776                         lock_chunks(root);
2777                         btrfs_device_set_bytes_used(device,
2778                                         device->bytes_used - dev_extent_len);
2779                         spin_lock(&root->fs_info->free_chunk_lock);
2780                         root->fs_info->free_chunk_space += dev_extent_len;
2781                         spin_unlock(&root->fs_info->free_chunk_lock);
2782                         btrfs_clear_space_info_full(root->fs_info);
2783                         unlock_chunks(root);
2784                 }
2785 
2786                 if (map->stripes[i].dev) {
2787                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2788                         if (ret) {
2789                                 btrfs_abort_transaction(trans, root, ret);
2790                                 goto out;
2791                         }
2792                 }
2793         }
2794         ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2795         if (ret) {
2796                 btrfs_abort_transaction(trans, root, ret);
2797                 goto out;
2798         }
2799 
2800         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2801 
2802         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2803                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2804                 if (ret) {
2805                         btrfs_abort_transaction(trans, root, ret);
2806                         goto out;
2807                 }
2808         }
2809 
2810         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2811         if (ret) {
2812                 btrfs_abort_transaction(trans, extent_root, ret);
2813                 goto out;
2814         }
2815 
2816 out:
2817         /* once for us */
2818         free_extent_map(em);
2819         return ret;
2820 }
2821 
2822 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2823 {
2824         struct btrfs_root *extent_root;
2825         struct btrfs_trans_handle *trans;
2826         int ret;
2827 
2828         root = root->fs_info->chunk_root;
2829         extent_root = root->fs_info->extent_root;
2830 
2831         /*
2832          * Prevent races with automatic removal of unused block groups.
2833          * After we relocate and before we remove the chunk with offset
2834          * chunk_offset, automatic removal of the block group can kick in,
2835          * resulting in a failure when calling btrfs_remove_chunk() below.
2836          *
2837          * Make sure to acquire this mutex before doing a tree search (dev
2838          * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2839          * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2840          * we release the path used to search the chunk/dev tree and before
2841          * the current task acquires this mutex and calls us.
2842          */
2843         ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2844 
2845         ret = btrfs_can_relocate(extent_root, chunk_offset);
2846         if (ret)
2847                 return -ENOSPC;
2848 
2849         /* step one, relocate all the extents inside this chunk */
2850         btrfs_scrub_pause(root);
2851         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2852         btrfs_scrub_continue(root);
2853         if (ret)
2854                 return ret;
2855 
2856         trans = btrfs_start_trans_remove_block_group(root->fs_info,
2857                                                      chunk_offset);
2858         if (IS_ERR(trans)) {
2859                 ret = PTR_ERR(trans);
2860                 btrfs_std_error(root->fs_info, ret, NULL);
2861                 return ret;
2862         }
2863 
2864         /*
2865          * step two, delete the device extents and the
2866          * chunk tree entries
2867          */
2868         ret = btrfs_remove_chunk(trans, root, chunk_offset);
2869         btrfs_end_transaction(trans, root);
2870         return ret;
2871 }
2872 
2873 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2874 {
2875         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2876         struct btrfs_path *path;
2877         struct extent_buffer *leaf;
2878         struct btrfs_chunk *chunk;
2879         struct btrfs_key key;
2880         struct btrfs_key found_key;
2881         u64 chunk_type;
2882         bool retried = false;
2883         int failed = 0;
2884         int ret;
2885 
2886         path = btrfs_alloc_path();
2887         if (!path)
2888                 return -ENOMEM;
2889 
2890 again:
2891         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2892         key.offset = (u64)-1;
2893         key.type = BTRFS_CHUNK_ITEM_KEY;
2894 
2895         while (1) {
2896                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2897                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2898                 if (ret < 0) {
2899                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2900                         goto error;
2901                 }
2902                 BUG_ON(ret == 0); /* Corruption */
2903 
2904                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2905                                           key.type);
2906                 if (ret)
2907                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2908                 if (ret < 0)
2909                         goto error;
2910                 if (ret > 0)
2911                         break;
2912 
2913                 leaf = path->nodes[0];
2914                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2915 
2916                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2917                                        struct btrfs_chunk);
2918                 chunk_type = btrfs_chunk_type(leaf, chunk);
2919                 btrfs_release_path(path);
2920 
2921                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2922                         ret = btrfs_relocate_chunk(chunk_root,
2923                                                    found_key.offset);
2924                         if (ret == -ENOSPC)
2925                                 failed++;
2926                         else
2927                                 BUG_ON(ret);
2928                 }
2929                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2930 
2931                 if (found_key.offset == 0)
2932                         break;
2933                 key.offset = found_key.offset - 1;
2934         }
2935         ret = 0;
2936         if (failed && !retried) {
2937                 failed = 0;
2938                 retried = true;
2939                 goto again;
2940         } else if (WARN_ON(failed && retried)) {
2941                 ret = -ENOSPC;
2942         }
2943 error:
2944         btrfs_free_path(path);
2945         return ret;
2946 }
2947 
2948 static int insert_balance_item(struct btrfs_root *root,
2949                                struct btrfs_balance_control *bctl)
2950 {
2951         struct btrfs_trans_handle *trans;
2952         struct btrfs_balance_item *item;
2953         struct btrfs_disk_balance_args disk_bargs;
2954         struct btrfs_path *path;
2955         struct extent_buffer *leaf;
2956         struct btrfs_key key;
2957         int ret, err;
2958 
2959         path = btrfs_alloc_path();
2960         if (!path)
2961                 return -ENOMEM;
2962 
2963         trans = btrfs_start_transaction(root, 0);
2964         if (IS_ERR(trans)) {
2965                 btrfs_free_path(path);
2966                 return PTR_ERR(trans);
2967         }
2968 
2969         key.objectid = BTRFS_BALANCE_OBJECTID;
2970         key.type = BTRFS_TEMPORARY_ITEM_KEY;
2971         key.offset = 0;
2972 
2973         ret = btrfs_insert_empty_item(trans, root, path, &key,
2974                                       sizeof(*item));
2975         if (ret)
2976                 goto out;
2977 
2978         leaf = path->nodes[0];
2979         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2980 
2981         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2982 
2983         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2984         btrfs_set_balance_data(leaf, item, &disk_bargs);
2985         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2986         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2987         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2988         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2989 
2990         btrfs_set_balance_flags(leaf, item, bctl->flags);
2991 
2992         btrfs_mark_buffer_dirty(leaf);
2993 out:
2994         btrfs_free_path(path);
2995         err = btrfs_commit_transaction(trans, root);
2996         if (err && !ret)
2997                 ret = err;
2998         return ret;
2999 }
3000 
3001 static int del_balance_item(struct btrfs_root *root)
3002 {
3003         struct btrfs_trans_handle *trans;
3004         struct btrfs_path *path;
3005         struct btrfs_key key;
3006         int ret, err;
3007 
3008         path = btrfs_alloc_path();
3009         if (!path)
3010                 return -ENOMEM;
3011 
3012         trans = btrfs_start_transaction(root, 0);
3013         if (IS_ERR(trans)) {
3014                 btrfs_free_path(path);
3015                 return PTR_ERR(trans);
3016         }
3017 
3018         key.objectid = BTRFS_BALANCE_OBJECTID;
3019         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3020         key.offset = 0;
3021 
3022         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3023         if (ret < 0)
3024                 goto out;
3025         if (ret > 0) {
3026                 ret = -ENOENT;
3027                 goto out;
3028         }
3029 
3030         ret = btrfs_del_item(trans, root, path);
3031 out:
3032         btrfs_free_path(path);
3033         err = btrfs_commit_transaction(trans, root);
3034         if (err && !ret)
3035                 ret = err;
3036         return ret;
3037 }
3038 
3039 /*
3040  * This is a heuristic used to reduce the number of chunks balanced on
3041  * resume after balance was interrupted.
3042  */
3043 static void update_balance_args(struct btrfs_balance_control *bctl)
3044 {
3045         /*
3046          * Turn on soft mode for chunk types that were being converted.
3047          */
3048         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3052         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3053                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3054 
3055         /*
3056          * Turn on usage filter if is not already used.  The idea is
3057          * that chunks that we have already balanced should be
3058          * reasonably full.  Don't do it for chunks that are being
3059          * converted - that will keep us from relocating unconverted
3060          * (albeit full) chunks.
3061          */
3062         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3063             !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3064             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3065                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3066                 bctl->data.usage = 90;
3067         }
3068         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3069             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3070             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3071                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3072                 bctl->sys.usage = 90;
3073         }
3074         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3075             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3076             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3077                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3078                 bctl->meta.usage = 90;
3079         }
3080 }
3081 
3082 /*
3083  * Should be called with both balance and volume mutexes held to
3084  * serialize other volume operations (add_dev/rm_dev/resize) with
3085  * restriper.  Same goes for unset_balance_control.
3086  */
3087 static void set_balance_control(struct btrfs_balance_control *bctl)
3088 {
3089         struct btrfs_fs_info *fs_info = bctl->fs_info;
3090 
3091         BUG_ON(fs_info->balance_ctl);
3092 
3093         spin_lock(&fs_info->balance_lock);
3094         fs_info->balance_ctl = bctl;
3095         spin_unlock(&fs_info->balance_lock);
3096 }
3097 
3098 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3099 {
3100         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3101 
3102         BUG_ON(!fs_info->balance_ctl);
3103 
3104         spin_lock(&fs_info->balance_lock);
3105         fs_info->balance_ctl = NULL;
3106         spin_unlock(&fs_info->balance_lock);
3107 
3108         kfree(bctl);
3109 }
3110 
3111 /*
3112  * Balance filters.  Return 1 if chunk should be filtered out
3113  * (should not be balanced).
3114  */
3115 static int chunk_profiles_filter(u64 chunk_type,
3116                                  struct btrfs_balance_args *bargs)
3117 {
3118         chunk_type = chunk_to_extended(chunk_type) &
3119                                 BTRFS_EXTENDED_PROFILE_MASK;
3120 
3121         if (bargs->profiles & chunk_type)
3122                 return 0;
3123 
3124         return 1;
3125 }
3126 
3127 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3128                               struct btrfs_balance_args *bargs)
3129 {
3130         struct btrfs_block_group_cache *cache;
3131         u64 chunk_used;
3132         u64 user_thresh_min;
3133         u64 user_thresh_max;
3134         int ret = 1;
3135 
3136         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3137         chunk_used = btrfs_block_group_used(&cache->item);
3138 
3139         if (bargs->usage_min == 0)
3140                 user_thresh_min = 0;
3141         else
3142                 user_thresh_min = div_factor_fine(cache->key.offset,
3143                                         bargs->usage_min);
3144 
3145         if (bargs->usage_max == 0)
3146                 user_thresh_max = 1;
3147         else if (bargs->usage_max > 100)
3148                 user_thresh_max = cache->key.offset;
3149         else
3150                 user_thresh_max = div_factor_fine(cache->key.offset,
3151                                         bargs->usage_max);
3152 
3153         if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3154                 ret = 0;
3155 
3156         btrfs_put_block_group(cache);
3157         return ret;
3158 }
3159 
3160 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3161                 u64 chunk_offset, struct btrfs_balance_args *bargs)
3162 {
3163         struct btrfs_block_group_cache *cache;
3164         u64 chunk_used, user_thresh;
3165         int ret = 1;
3166 
3167         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3168         chunk_used = btrfs_block_group_used(&cache->item);
3169 
3170         if (bargs->usage_min == 0)
3171                 user_thresh = 1;
3172         else if (bargs->usage > 100)
3173                 user_thresh = cache->key.offset;
3174         else
3175                 user_thresh = div_factor_fine(cache->key.offset,
3176                                               bargs->usage);
3177 
3178         if (chunk_used < user_thresh)
3179                 ret = 0;
3180 
3181         btrfs_put_block_group(cache);
3182         return ret;
3183 }
3184 
3185 static int chunk_devid_filter(struct extent_buffer *leaf,
3186                               struct btrfs_chunk *chunk,
3187                               struct btrfs_balance_args *bargs)
3188 {
3189         struct btrfs_stripe *stripe;
3190         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3191         int i;
3192 
3193         for (i = 0; i < num_stripes; i++) {
3194                 stripe = btrfs_stripe_nr(chunk, i);
3195                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3196                         return 0;
3197         }
3198 
3199         return 1;
3200 }
3201 
3202 /* [pstart, pend) */
3203 static int chunk_drange_filter(struct extent_buffer *leaf,
3204                                struct btrfs_chunk *chunk,
3205                                u64 chunk_offset,
3206                                struct btrfs_balance_args *bargs)
3207 {
3208         struct btrfs_stripe *stripe;
3209         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3210         u64 stripe_offset;
3211         u64 stripe_length;
3212         int factor;
3213         int i;
3214 
3215         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3216                 return 0;
3217 
3218         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3219              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3220                 factor = num_stripes / 2;
3221         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3222                 factor = num_stripes - 1;
3223         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3224                 factor = num_stripes - 2;
3225         } else {
3226                 factor = num_stripes;
3227         }
3228 
3229         for (i = 0; i < num_stripes; i++) {
3230                 stripe = btrfs_stripe_nr(chunk, i);
3231                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3232                         continue;
3233 
3234                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3235                 stripe_length = btrfs_chunk_length(leaf, chunk);
3236                 stripe_length = div_u64(stripe_length, factor);
3237 
3238                 if (stripe_offset < bargs->pend &&
3239                     stripe_offset + stripe_length > bargs->pstart)
3240                         return 0;
3241         }
3242 
3243         return 1;
3244 }
3245 
3246 /* [vstart, vend) */
3247 static int chunk_vrange_filter(struct extent_buffer *leaf,
3248                                struct btrfs_chunk *chunk,
3249                                u64 chunk_offset,
3250                                struct btrfs_balance_args *bargs)
3251 {
3252         if (chunk_offset < bargs->vend &&
3253             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3254                 /* at least part of the chunk is inside this vrange */
3255                 return 0;
3256 
3257         return 1;
3258 }
3259 
3260 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3261                                struct btrfs_chunk *chunk,
3262                                struct btrfs_balance_args *bargs)
3263 {
3264         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3265 
3266         if (bargs->stripes_min <= num_stripes
3267                         && num_stripes <= bargs->stripes_max)
3268                 return 0;
3269 
3270         return 1;
3271 }
3272 
3273 static int chunk_soft_convert_filter(u64 chunk_type,
3274                                      struct btrfs_balance_args *bargs)
3275 {
3276         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3277                 return 0;
3278 
3279         chunk_type = chunk_to_extended(chunk_type) &
3280                                 BTRFS_EXTENDED_PROFILE_MASK;
3281 
3282         if (bargs->target == chunk_type)
3283                 return 1;
3284 
3285         return 0;
3286 }
3287 
3288 static int should_balance_chunk(struct btrfs_root *root,
3289                                 struct extent_buffer *leaf,
3290                                 struct btrfs_chunk *chunk, u64 chunk_offset)
3291 {
3292         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3293         struct btrfs_balance_args *bargs = NULL;
3294         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3295 
3296         /* type filter */
3297         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3298               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3299                 return 0;
3300         }
3301 
3302         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3303                 bargs = &bctl->data;
3304         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3305                 bargs = &bctl->sys;
3306         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3307                 bargs = &bctl->meta;
3308 
3309         /* profiles filter */
3310         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3311             chunk_profiles_filter(chunk_type, bargs)) {
3312                 return 0;
3313         }
3314 
3315         /* usage filter */
3316         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3317             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3318                 return 0;
3319         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3320             chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3321                 return 0;
3322         }
3323 
3324         /* devid filter */
3325         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3326             chunk_devid_filter(leaf, chunk, bargs)) {
3327                 return 0;
3328         }
3329 
3330         /* drange filter, makes sense only with devid filter */
3331         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3332             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3333                 return 0;
3334         }
3335 
3336         /* vrange filter */
3337         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3338             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3339                 return 0;
3340         }
3341 
3342         /* stripes filter */
3343         if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3344             chunk_stripes_range_filter(leaf, chunk, bargs)) {
3345                 return 0;
3346         }
3347 
3348         /* soft profile changing mode */
3349         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3350             chunk_soft_convert_filter(chunk_type, bargs)) {
3351                 return 0;
3352         }
3353 
3354         /*
3355          * limited by count, must be the last filter
3356          */
3357         if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3358                 if (bargs->limit == 0)
3359                         return 0;
3360                 else
3361                         bargs->limit--;
3362         } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3363                 /*
3364                  * Same logic as the 'limit' filter; the minimum cannot be
3365                  * determined here because we do not have the global informatoin
3366                  * about the count of all chunks that satisfy the filters.
3367                  */
3368                 if (bargs->limit_max == 0)
3369                         return 0;
3370                 else
3371                         bargs->limit_max--;
3372         }
3373 
3374         return 1;
3375 }
3376 
3377 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3378 {
3379         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3380         struct btrfs_root *chunk_root = fs_info->chunk_root;
3381         struct btrfs_root *dev_root = fs_info->dev_root;
3382         struct list_head *devices;
3383         struct btrfs_device *device;
3384         u64 old_size;
3385         u64 size_to_free;
3386         u64 chunk_type;
3387         struct btrfs_chunk *chunk;
3388         struct btrfs_path *path;
3389         struct btrfs_key key;
3390         struct btrfs_key found_key;
3391         struct btrfs_trans_handle *trans;
3392         struct extent_buffer *leaf;
3393         int slot;
3394         int ret;
3395         int enospc_errors = 0;
3396         bool counting = true;
3397         /* The single value limit and min/max limits use the same bytes in the */
3398         u64 limit_data = bctl->data.limit;
3399         u64 limit_meta = bctl->meta.limit;
3400         u64 limit_sys = bctl->sys.limit;
3401         u32 count_data = 0;
3402         u32 count_meta = 0;
3403         u32 count_sys = 0;
3404         int chunk_reserved = 0;
3405 
3406         /* step one make some room on all the devices */
3407         devices = &fs_info->fs_devices->devices;
3408         list_for_each_entry(device, devices, dev_list) {
3409                 old_size = btrfs_device_get_total_bytes(device);
3410                 size_to_free = div_factor(old_size, 1);
3411                 size_to_free = min_t(u64, size_to_free, SZ_1M);
3412                 if (!device->writeable ||
3413                     btrfs_device_get_total_bytes(device) -
3414                     btrfs_device_get_bytes_used(device) > size_to_free ||
3415                     device->is_tgtdev_for_dev_replace)
3416                         continue;
3417 
3418                 ret = btrfs_shrink_device(device, old_size - size_to_free);
3419                 if (ret == -ENOSPC)
3420                         break;
3421                 BUG_ON(ret);
3422 
3423                 trans = btrfs_start_transaction(dev_root, 0);
3424                 BUG_ON(IS_ERR(trans));
3425 
3426                 ret = btrfs_grow_device(trans, device, old_size);
3427                 BUG_ON(ret);
3428 
3429                 btrfs_end_transaction(trans, dev_root);
3430         }
3431 
3432         /* step two, relocate all the chunks */
3433         path = btrfs_alloc_path();
3434         if (!path) {
3435                 ret = -ENOMEM;
3436                 goto error;
3437         }
3438 
3439         /* zero out stat counters */
3440         spin_lock(&fs_info->balance_lock);
3441         memset(&bctl->stat, 0, sizeof(bctl->stat));
3442         spin_unlock(&fs_info->balance_lock);
3443 again:
3444         if (!counting) {
3445                 /*
3446                  * The single value limit and min/max limits use the same bytes
3447                  * in the
3448                  */
3449                 bctl->data.limit = limit_data;
3450                 bctl->meta.limit = limit_meta;
3451                 bctl->sys.limit = limit_sys;
3452         }
3453         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3454         key.offset = (u64)-1;
3455         key.type = BTRFS_CHUNK_ITEM_KEY;
3456 
3457         while (1) {
3458                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3459                     atomic_read(&fs_info->balance_cancel_req)) {
3460                         ret = -ECANCELED;
3461                         goto error;
3462                 }
3463 
3464                 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3465                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3466                 if (ret < 0) {
3467                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3468                         goto error;
3469                 }
3470 
3471                 /*
3472                  * this shouldn't happen, it means the last relocate
3473                  * failed
3474                  */
3475                 if (ret == 0)
3476                         BUG(); /* FIXME break ? */
3477 
3478                 ret = btrfs_previous_item(chunk_root, path, 0,
3479                                           BTRFS_CHUNK_ITEM_KEY);
3480                 if (ret) {
3481                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3482                         ret = 0;
3483                         break;
3484                 }
3485 
3486                 leaf = path->nodes[0];
3487                 slot = path->slots[0];
3488                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3489 
3490                 if (found_key.objectid != key.objectid) {
3491                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3492                         break;
3493                 }
3494 
3495                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3496                 chunk_type = btrfs_chunk_type(leaf, chunk);
3497 
3498                 if (!counting) {
3499                         spin_lock(&fs_info->balance_lock);
3500                         bctl->stat.considered++;
3501                         spin_unlock(&fs_info->balance_lock);
3502                 }
3503 
3504                 ret = should_balance_chunk(chunk_root, leaf, chunk,
3505                                            found_key.offset);
3506 
3507                 btrfs_release_path(path);
3508                 if (!ret) {
3509                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3510                         goto loop;
3511                 }
3512 
3513                 if (counting) {
3514                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3515                         spin_lock(&fs_info->balance_lock);
3516                         bctl->stat.expected++;
3517                         spin_unlock(&fs_info->balance_lock);
3518 
3519                         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3520                                 count_data++;
3521                         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3522                                 count_sys++;
3523                         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3524                                 count_meta++;
3525 
3526                         goto loop;
3527                 }
3528 
3529                 /*
3530                  * Apply limit_min filter, no need to check if the LIMITS
3531                  * filter is used, limit_min is 0 by default
3532                  */
3533                 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3534                                         count_data < bctl->data.limit_min)
3535                                 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3536                                         count_meta < bctl->meta.limit_min)
3537                                 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3538                                         count_sys < bctl->sys.limit_min)) {
3539                         mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3540                         goto loop;
3541                 }
3542 
3543                 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3544                         trans = btrfs_start_transaction(chunk_root, 0);
3545                         if (IS_ERR(trans)) {
3546                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3547                                 ret = PTR_ERR(trans);
3548                                 goto error;
3549                         }
3550 
3551                         ret = btrfs_force_chunk_alloc(trans, chunk_root,
3552                                                       BTRFS_BLOCK_GROUP_DATA);
3553                         btrfs_end_transaction(trans, chunk_root);
3554                         if (ret < 0) {
3555                                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3556                                 goto error;
3557                         }
3558                         chunk_reserved = 1;
3559                 }
3560 
3561                 ret = btrfs_relocate_chunk(chunk_root,
3562                                            found_key.offset);
3563                 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3564                 if (ret && ret != -ENOSPC)
3565                         goto error;
3566                 if (ret == -ENOSPC) {
3567                         enospc_errors++;
3568                 } else {
3569                         spin_lock(&fs_info->balance_lock);
3570                         bctl->stat.completed++;
3571                         spin_unlock(&fs_info->balance_lock);
3572                 }
3573 loop:
3574                 if (found_key.offset == 0)
3575                         break;
3576                 key.offset = found_key.offset - 1;
3577         }
3578 
3579         if (counting) {
3580                 btrfs_release_path(path);
3581                 counting = false;
3582                 goto again;
3583         }
3584 error:
3585         btrfs_free_path(path);
3586         if (enospc_errors) {
3587                 btrfs_info(fs_info, "%d enospc errors during balance",
3588                        enospc_errors);
3589                 if (!ret)
3590                         ret = -ENOSPC;
3591         }
3592 
3593         return ret;
3594 }
3595 
3596 /**
3597  * alloc_profile_is_valid - see if a given profile is valid and reduced
3598  * @flags: profile to validate
3599  * @extended: if true @flags is treated as an extended profile
3600  */
3601 static int alloc_profile_is_valid(u64 flags, int extended)
3602 {
3603         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3604                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3605 
3606         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3607 
3608         /* 1) check that all other bits are zeroed */
3609         if (flags & ~mask)
3610                 return 0;
3611 
3612         /* 2) see if profile is reduced */
3613         if (flags == 0)
3614                 return !extended; /* "" is valid for usual profiles */
3615 
3616         /* true if exactly one bit set */
3617         return (flags & (flags - 1)) == 0;
3618 }
3619 
3620 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3621 {
3622         /* cancel requested || normal exit path */
3623         return atomic_read(&fs_info->balance_cancel_req) ||
3624                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3625                  atomic_read(&fs_info->balance_cancel_req) == 0);
3626 }
3627 
3628 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3629 {
3630         int ret;
3631 
3632         unset_balance_control(fs_info);
3633         ret = del_balance_item(fs_info->tree_root);
3634         if (ret)
3635                 btrfs_std_error(fs_info, ret, NULL);
3636 
3637         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3638 }
3639 
3640 /* Non-zero return value signifies invalidity */
3641 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3642                 u64 allowed)
3643 {
3644         return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3645                 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3646                  (bctl_arg->target & ~allowed)));
3647 }
3648 
3649 /*
3650  * Should be called with both balance and volume mutexes held
3651  */
3652 int btrfs_balance(struct btrfs_balance_control *bctl,
3653                   struct btrfs_ioctl_balance_args *bargs)
3654 {
3655         struct btrfs_fs_info *fs_info = bctl->fs_info;
3656         u64 allowed;
3657         int mixed = 0;
3658         int ret;
3659         u64 num_devices;
3660         unsigned seq;
3661 
3662         if (btrfs_fs_closing(fs_info) ||
3663             atomic_read(&fs_info->balance_pause_req) ||
3664             atomic_read(&fs_info->balance_cancel_req)) {
3665                 ret = -EINVAL;
3666                 goto out;
3667         }
3668 
3669         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3670         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3671                 mixed = 1;
3672 
3673         /*
3674          * In case of mixed groups both data and meta should be picked,
3675          * and identical options should be given for both of them.
3676          */
3677         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3678         if (mixed && (bctl->flags & allowed)) {
3679                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3680                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3681                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3682                         btrfs_err(fs_info, "with mixed groups data and "
3683                                    "metadata balance options must be the same");
3684                         ret = -EINVAL;
3685                         goto out;
3686                 }
3687         }
3688 
3689         num_devices = fs_info->fs_devices->num_devices;
3690         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3691         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3692                 BUG_ON(num_devices < 1);
3693                 num_devices--;
3694         }
3695         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3696         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3697         if (num_devices == 1)
3698                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3699         else if (num_devices > 1)
3700                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3701         if (num_devices > 2)
3702                 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3703         if (num_devices > 3)
3704                 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3705                             BTRFS_BLOCK_GROUP_RAID6);
3706         if (validate_convert_profile(&bctl->data, allowed)) {
3707                 btrfs_err(fs_info, "unable to start balance with target "
3708                            "data profile %llu",
3709                        bctl->data.target);
3710                 ret = -EINVAL;
3711                 goto out;
3712         }
3713         if (validate_convert_profile(&bctl->meta, allowed)) {
3714                 btrfs_err(fs_info,
3715                            "unable to start balance with target metadata profile %llu",
3716                        bctl->meta.target);
3717                 ret = -EINVAL;
3718                 goto out;
3719         }
3720         if (validate_convert_profile(&bctl->sys, allowed)) {
3721                 btrfs_err(fs_info,
3722                            "unable to start balance with target system profile %llu",
3723                        bctl->sys.target);
3724                 ret = -EINVAL;
3725                 goto out;
3726         }
3727 
3728         /* allow to reduce meta or sys integrity only if force set */
3729         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3730                         BTRFS_BLOCK_GROUP_RAID10 |
3731                         BTRFS_BLOCK_GROUP_RAID5 |
3732                         BTRFS_BLOCK_GROUP_RAID6;
3733         do {
3734                 seq = read_seqbegin(&fs_info->profiles_lock);
3735 
3736                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3737                      (fs_info->avail_system_alloc_bits & allowed) &&
3738                      !(bctl->sys.target & allowed)) ||
3739                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3740                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3741                      !(bctl->meta.target & allowed))) {
3742                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3743                                 btrfs_info(fs_info, "force reducing metadata integrity");
3744                         } else {
3745                                 btrfs_err(fs_info, "balance will reduce metadata "
3746                                            "integrity, use force if you want this");
3747                                 ret = -EINVAL;
3748                                 goto out;
3749                         }
3750                 }
3751         } while (read_seqretry(&fs_info->profiles_lock, seq));
3752 
3753         if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3754                 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3755                 btrfs_warn(fs_info,
3756         "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3757                         bctl->meta.target, bctl->data.target);
3758         }
3759 
3760         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3761                 fs_info->num_tolerated_disk_barrier_failures = min(
3762                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3763                         btrfs_get_num_tolerated_disk_barrier_failures(
3764                                 bctl->sys.target));
3765         }
3766 
3767         ret = insert_balance_item(fs_info->tree_root, bctl);
3768         if (ret && ret != -EEXIST)
3769                 goto out;
3770 
3771         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3772                 BUG_ON(ret == -EEXIST);
3773                 set_balance_control(bctl);
3774         } else {
3775                 BUG_ON(ret != -EEXIST);
3776                 spin_lock(&fs_info->balance_lock);
3777                 update_balance_args(bctl);
3778                 spin_unlock(&fs_info->balance_lock);
3779         }
3780 
3781         atomic_inc(&fs_info->balance_running);
3782         mutex_unlock(&fs_info->balance_mutex);
3783 
3784         ret = __btrfs_balance(fs_info);
3785 
3786         mutex_lock(&fs_info->balance_mutex);
3787         atomic_dec(&fs_info->balance_running);
3788 
3789         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3790                 fs_info->num_tolerated_disk_barrier_failures =
3791                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3792         }
3793 
3794         if (bargs) {
3795                 memset(bargs, 0, sizeof(*bargs));
3796                 update_ioctl_balance_args(fs_info, 0, bargs);
3797         }
3798 
3799         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3800             balance_need_close(fs_info)) {
3801                 __cancel_balance(fs_info);
3802         }
3803 
3804         wake_up(&fs_info->balance_wait_q);
3805 
3806         return ret;
3807 out:
3808         if (bctl->flags & BTRFS_BALANCE_RESUME)
3809                 __cancel_balance(fs_info);
3810         else {
3811                 kfree(bctl);
3812                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3813         }
3814         return ret;
3815 }
3816 
3817 static int balance_kthread(void *data)
3818 {
3819         struct btrfs_fs_info *fs_info = data;
3820         int ret = 0;
3821 
3822         mutex_lock(&fs_info->volume_mutex);
3823         mutex_lock(&fs_info->balance_mutex);
3824 
3825         if (fs_info->balance_ctl) {
3826                 btrfs_info(fs_info, "continuing balance");
3827                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3828         }
3829 
3830         mutex_unlock(&fs_info->balance_mutex);
3831         mutex_unlock(&fs_info->volume_mutex);
3832 
3833         return ret;
3834 }
3835 
3836 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3837 {
3838         struct task_struct *tsk;
3839 
3840         spin_lock(&fs_info->balance_lock);
3841         if (!fs_info->balance_ctl) {
3842                 spin_unlock(&fs_info->balance_lock);
3843                 return 0;
3844         }
3845         spin_unlock(&fs_info->balance_lock);
3846 
3847         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3848                 btrfs_info(fs_info, "force skipping balance");
3849                 return 0;
3850         }
3851 
3852         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3853         return PTR_ERR_OR_ZERO(tsk);
3854 }
3855 
3856 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3857 {
3858         struct btrfs_balance_control *bctl;
3859         struct btrfs_balance_item *item;
3860         struct btrfs_disk_balance_args disk_bargs;
3861         struct btrfs_path *path;
3862         struct extent_buffer *leaf;
3863         struct btrfs_key key;
3864         int ret;
3865 
3866         path = btrfs_alloc_path();
3867         if (!path)
3868                 return -ENOMEM;
3869 
3870         key.objectid = BTRFS_BALANCE_OBJECTID;
3871         key.type = BTRFS_TEMPORARY_ITEM_KEY;
3872         key.offset = 0;
3873 
3874         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3875         if (ret < 0)
3876                 goto out;
3877         if (ret > 0) { /* ret = -ENOENT; */
3878                 ret = 0;
3879                 goto out;
3880         }
3881 
3882         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3883         if (!bctl) {
3884                 ret = -ENOMEM;
3885                 goto out;
3886         }
3887 
3888         leaf = path->nodes[0];
3889         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3890 
3891         bctl->fs_info = fs_info;
3892         bctl->flags = btrfs_balance_flags(leaf, item);
3893         bctl->flags |= BTRFS_BALANCE_RESUME;
3894 
3895         btrfs_balance_data(leaf, item, &disk_bargs);
3896         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3897         btrfs_balance_meta(leaf, item, &disk_bargs);
3898         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3899         btrfs_balance_sys(leaf, item, &disk_bargs);
3900         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3901 
3902         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3903 
3904         mutex_lock(&fs_info->volume_mutex);
3905         mutex_lock(&fs_info->balance_mutex);
3906 
3907         set_balance_control(bctl);
3908 
3909         mutex_unlock(&fs_info->balance_mutex);
3910         mutex_unlock(&fs_info->volume_mutex);
3911 out:
3912         btrfs_free_path(path);
3913         return ret;
3914 }
3915 
3916 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3917 {
3918         int ret = 0;
3919 
3920         mutex_lock(&fs_info->balance_mutex);
3921         if (!fs_info->balance_ctl) {
3922                 mutex_unlock(&fs_info->balance_mutex);
3923                 return -ENOTCONN;
3924         }
3925 
3926         if (atomic_read(&fs_info->balance_running)) {
3927                 atomic_inc(&fs_info->balance_pause_req);
3928                 mutex_unlock(&fs_info->balance_mutex);
3929 
3930                 wait_event(fs_info->balance_wait_q,
3931                            atomic_read(&fs_info->balance_running) == 0);
3932 
3933                 mutex_lock(&fs_info->balance_mutex);
3934                 /* we are good with balance_ctl ripped off from under us */
3935                 BUG_ON(atomic_read(&fs_info->balance_running));
3936                 atomic_dec(&fs_info->balance_pause_req);
3937         } else {
3938                 ret = -ENOTCONN;
3939         }
3940 
3941         mutex_unlock(&fs_info->balance_mutex);
3942         return ret;
3943 }
3944 
3945 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3946 {
3947         if (fs_info->sb->s_flags & MS_RDONLY)
3948                 return -EROFS;
3949 
3950         mutex_lock(&fs_info->balance_mutex);
3951         if (!fs_info->balance_ctl) {
3952                 mutex_unlock(&fs_info->balance_mutex);
3953                 return -ENOTCONN;
3954         }
3955 
3956         atomic_inc(&fs_info->balance_cancel_req);
3957         /*
3958          * if we are running just wait and return, balance item is
3959          * deleted in btrfs_balance in this case
3960          */
3961         if (atomic_read(&fs_info->balance_running)) {
3962                 mutex_unlock(&fs_info->balance_mutex);
3963                 wait_event(fs_info->balance_wait_q,
3964                            atomic_read(&fs_info->balance_running) == 0);
3965                 mutex_lock(&fs_info->balance_mutex);
3966         } else {
3967                 /* __cancel_balance needs volume_mutex */
3968                 mutex_unlock(&fs_info->balance_mutex);
3969                 mutex_lock(&fs_info->volume_mutex);
3970                 mutex_lock(&fs_info->balance_mutex);
3971 
3972                 if (fs_info->balance_ctl)
3973                         __cancel_balance(fs_info);
3974 
3975                 mutex_unlock(&fs_info->volume_mutex);
3976         }
3977 
3978         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3979         atomic_dec(&fs_info->balance_cancel_req);
3980         mutex_unlock(&fs_info->balance_mutex);
3981         return 0;
3982 }
3983 
3984 static int btrfs_uuid_scan_kthread(void *data)
3985 {
3986         struct btrfs_fs_info *fs_info = data;
3987         struct btrfs_root *root = fs_info->tree_root;
3988         struct btrfs_key key;
3989         struct btrfs_key max_key;
3990         struct btrfs_path *path = NULL;
3991         int ret = 0;
3992         struct extent_buffer *eb;
3993         int slot;
3994         struct btrfs_root_item root_item;
3995         u32 item_size;
3996         struct btrfs_trans_handle *trans = NULL;
3997 
3998         path = btrfs_alloc_path();
3999         if (!path) {
4000                 ret = -ENOMEM;
4001                 goto out;
4002         }
4003 
4004         key.objectid = 0;
4005         key.type = BTRFS_ROOT_ITEM_KEY;
4006         key.offset = 0;
4007 
4008         max_key.objectid = (u64)-1;
4009         max_key.type = BTRFS_ROOT_ITEM_KEY;
4010         max_key.offset = (u64)-1;
4011 
4012         while (1) {
4013                 ret = btrfs_search_forward(root, &key, path, 0);
4014                 if (ret) {
4015                         if (ret > 0)
4016                                 ret = 0;
4017                         break;
4018                 }
4019 
4020                 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4021                     (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4022                      key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4023                     key.objectid > BTRFS_LAST_FREE_OBJECTID)
4024                         goto skip;
4025 
4026                 eb = path->nodes[0];
4027                 slot = path->slots[0];
4028                 item_size = btrfs_item_size_nr(eb, slot);
4029                 if (item_size < sizeof(root_item))
4030                         goto skip;
4031 
4032                 read_extent_buffer(eb, &root_item,
4033                                    btrfs_item_ptr_offset(eb, slot),
4034                                    (int)sizeof(root_item));
4035                 if (btrfs_root_refs(&root_item) == 0)
4036                         goto skip;
4037 
4038                 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4039                     !btrfs_is_empty_uuid(root_item.received_uuid)) {
4040                         if (trans)
4041                                 goto update_tree;
4042 
4043                         btrfs_release_path(path);
4044                         /*
4045                          * 1 - subvol uuid item
4046                          * 1 - received_subvol uuid item
4047                          */
4048                         trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4049                         if (IS_ERR(trans)) {
4050                                 ret = PTR_ERR(trans);
4051                                 break;
4052                         }
4053                         continue;
4054                 } else {
4055                         goto skip;
4056                 }
4057 update_tree:
4058                 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4059                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4060                                                   root_item.uuid,
4061                                                   BTRFS_UUID_KEY_SUBVOL,
4062                                                   key.objectid);
4063                         if (ret < 0) {
4064                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4065                                         ret);
4066                                 break;
4067                         }
4068                 }
4069 
4070                 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4071                         ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4072                                                   root_item.received_uuid,
4073                                                  BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4074                                                   key.objectid);
4075                         if (ret < 0) {
4076                                 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4077                                         ret);
4078                                 break;
4079                         }
4080                 }
4081 
4082 skip:
4083                 if (trans) {
4084                         ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4085                         trans = NULL;
4086                         if (ret)
4087                                 break;
4088                 }
4089 
4090                 btrfs_release_path(path);
4091                 if (key.offset < (u64)-1) {
4092                         key.offset++;
4093                 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4094                         key.offset = 0;
4095                         key.type = BTRFS_ROOT_ITEM_KEY;
4096                 } else if (key.objectid < (u64)-1) {
4097                         key.offset = 0;
4098                         key.type = BTRFS_ROOT_ITEM_KEY;
4099                         key.objectid++;
4100                 } else {
4101                         break;
4102                 }
4103                 cond_resched();
4104         }
4105 
4106 out:
4107         btrfs_free_path(path);
4108         if (trans && !IS_ERR(trans))
4109                 btrfs_end_transaction(trans, fs_info->uuid_root);
4110         if (ret)
4111                 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4112         else
4113                 fs_info->update_uuid_tree_gen = 1;
4114         up(&fs_info->uuid_tree_rescan_sem);
4115         return 0;
4116 }
4117 
4118 /*
4119  * Callback for btrfs_uuid_tree_iterate().
4120  * returns:
4121  * 0    check succeeded, the entry is not outdated.
4122  * < 0  if an error occurred.
4123  * > 0  if the check failed, which means the caller shall remove the entry.
4124  */
4125 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4126                                        u8 *uuid, u8 type, u64 subid)
4127 {
4128         struct btrfs_key key;
4129         int ret = 0;
4130         struct btrfs_root *subvol_root;
4131 
4132         if (type != BTRFS_UUID_KEY_SUBVOL &&
4133             type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4134                 goto out;
4135 
4136         key.objectid = subid;
4137         key.type = BTRFS_ROOT_ITEM_KEY;
4138         key.offset = (u64)-1;
4139         subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4140         if (IS_ERR(subvol_root)) {
4141                 ret = PTR_ERR(subvol_root);
4142                 if (ret == -ENOENT)
4143                         ret = 1;
4144                 goto out;
4145         }
4146 
4147         switch (type) {
4148         case BTRFS_UUID_KEY_SUBVOL:
4149                 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4150                         ret = 1;
4151                 break;
4152         case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4153                 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4154                            BTRFS_UUID_SIZE))
4155                         ret = 1;
4156                 break;
4157         }
4158 
4159 out:
4160         return ret;
4161 }
4162 
4163 static int btrfs_uuid_rescan_kthread(void *data)
4164 {
4165         struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4166         int ret;
4167 
4168         /*
4169          * 1st step is to iterate through the existing UUID tree and
4170          * to delete all entries that contain outdated data.
4171          * 2nd step is to add all missing entries to the UUID tree.
4172          */
4173         ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4174         if (ret < 0) {
4175                 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4176                 up(&fs_info->uuid_tree_rescan_sem);
4177                 return ret;
4178         }
4179         return btrfs_uuid_scan_kthread(data);
4180 }
4181 
4182 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4183 {
4184         struct btrfs_trans_handle *trans;
4185         struct btrfs_root *tree_root = fs_info->tree_root;
4186         struct btrfs_root *uuid_root;
4187         struct task_struct *task;
4188         int ret;
4189 
4190         /*
4191          * 1 - root node
4192          * 1 - root item
4193          */
4194         trans = btrfs_start_transaction(tree_root, 2);
4195         if (IS_ERR(trans))
4196                 return PTR_ERR(trans);
4197 
4198         uuid_root = btrfs_create_tree(trans, fs_info,
4199                                       BTRFS_UUID_TREE_OBJECTID);
4200         if (IS_ERR(uuid_root)) {
4201                 ret = PTR_ERR(uuid_root);
4202                 btrfs_abort_transaction(trans, tree_root, ret);
4203                 return ret;
4204         }
4205 
4206         fs_info->uuid_root = uuid_root;
4207 
4208         ret = btrfs_commit_transaction(trans, tree_root);
4209         if (ret)
4210                 return ret;
4211 
4212         down(&fs_info->uuid_tree_rescan_sem);
4213         task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4214         if (IS_ERR(task)) {
4215                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4216                 btrfs_warn(fs_info, "failed to start uuid_scan task");
4217                 up(&fs_info->uuid_tree_rescan_sem);
4218                 return PTR_ERR(task);
4219         }
4220 
4221         return 0;
4222 }
4223 
4224 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4225 {
4226         struct task_struct *task;
4227 
4228         down(&fs_info->uuid_tree_rescan_sem);
4229         task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4230         if (IS_ERR(task)) {
4231                 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4232                 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4233                 up(&fs_info->uuid_tree_rescan_sem);
4234                 return PTR_ERR(task);
4235         }
4236 
4237         return 0;
4238 }
4239 
4240 /*
4241  * shrinking a device means finding all of the device extents past
4242  * the new size, and then following the back refs to the chunks.
4243  * The chunk relocation code actually frees the device extent
4244  */
4245 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4246 {
4247         struct btrfs_trans_handle *trans;
4248         struct btrfs_root *root = device->dev_root;
4249         struct btrfs_dev_extent *dev_extent = NULL;
4250         struct btrfs_path *path;
4251         u64 length;
4252         u64 chunk_offset;
4253         int ret;
4254         int slot;
4255         int failed = 0;
4256         bool retried = false;
4257         bool checked_pending_chunks = false;
4258         struct extent_buffer *l;
4259         struct btrfs_key key;
4260         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4261         u64 old_total = btrfs_super_total_bytes(super_copy);
4262         u64 old_size = btrfs_device_get_total_bytes(device);
4263         u64 diff = old_size - new_size;
4264 
4265         if (device->is_tgtdev_for_dev_replace)
4266                 return -EINVAL;
4267 
4268         path = btrfs_alloc_path();
4269         if (!path)
4270                 return -ENOMEM;
4271 
4272         path->reada = READA_FORWARD;
4273 
4274         lock_chunks(root);
4275 
4276         btrfs_device_set_total_bytes(device, new_size);
4277         if (device->writeable) {
4278                 device->fs_devices->total_rw_bytes -= diff;
4279                 spin_lock(&root->fs_info->free_chunk_lock);
4280                 root->fs_info->free_chunk_space -= diff;
4281                 spin_unlock(&root->fs_info->free_chunk_lock);
4282         }
4283         unlock_chunks(root);
4284 
4285 again:
4286         key.objectid = device->devid;
4287         key.offset = (u64)-1;
4288         key.type = BTRFS_DEV_EXTENT_KEY;
4289 
4290         do {
4291                 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4292                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4293                 if (ret < 0) {
4294                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4295                         goto done;
4296                 }
4297 
4298                 ret = btrfs_previous_item(root, path, 0, key.type);
4299                 if (ret)
4300                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4301                 if (ret < 0)
4302                         goto done;
4303                 if (ret) {
4304                         ret = 0;
4305                         btrfs_release_path(path);
4306                         break;
4307                 }
4308 
4309                 l = path->nodes[0];
4310                 slot = path->slots[0];
4311                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4312 
4313                 if (key.objectid != device->devid) {
4314                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4315                         btrfs_release_path(path);
4316                         break;
4317                 }
4318 
4319                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4320                 length = btrfs_dev_extent_length(l, dev_extent);
4321 
4322                 if (key.offset + length <= new_size) {
4323                         mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4324                         btrfs_release_path(path);
4325                         break;
4326                 }
4327 
4328                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4329                 btrfs_release_path(path);
4330 
4331                 ret = btrfs_relocate_chunk(root, chunk_offset);
4332                 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4333                 if (ret && ret != -ENOSPC)
4334                         goto done;
4335                 if (ret == -ENOSPC)
4336                         failed++;
4337         } while (key.offset-- > 0);
4338 
4339         if (failed && !retried) {
4340                 failed = 0;
4341                 retried = true;
4342                 goto again;
4343         } else if (failed && retried) {
4344                 ret = -ENOSPC;
4345                 goto done;
4346         }
4347 
4348         /* Shrinking succeeded, else we would be at "done". */
4349         trans = btrfs_start_transaction(root, 0);
4350         if (IS_ERR(trans)) {
4351                 ret = PTR_ERR(trans);
4352                 goto done;
4353         }
4354 
4355         lock_chunks(root);
4356 
4357         /*
4358          * We checked in the above loop all device extents that were already in
4359          * the device tree. However before we have updated the device's
4360          * total_bytes to the new size, we might have had chunk allocations that
4361          * have not complete yet (new block groups attached to transaction
4362          * handles), and therefore their device extents were not yet in the
4363          * device tree and we missed them in the loop above. So if we have any
4364          * pending chunk using a device extent that overlaps the device range
4365          * that we can not use anymore, commit the current transaction and
4366          * repeat the search on the device tree - this way we guarantee we will
4367          * not have chunks using device extents that end beyond 'new_size'.
4368          */
4369         if (!checked_pending_chunks) {
4370                 u64 start = new_size;
4371                 u64 len = old_size - new_size;
4372 
4373                 if (contains_pending_extent(trans->transaction, device,
4374                                             &start, len)) {
4375                         unlock_chunks(root);
4376                         checked_pending_chunks = true;
4377                         failed = 0;
4378                         retried = false;
4379                         ret = btrfs_commit_transaction(trans, root);
4380                         if (ret)
4381                                 goto done;
4382                         goto again;
4383                 }
4384         }
4385 
4386         btrfs_device_set_disk_total_bytes(device, new_size);
4387         if (list_empty(&device->resized_list))
4388                 list_add_tail(&device->resized_list,
4389                               &root->fs_info->fs_devices->resized_devices);
4390 
4391         WARN_ON(diff > old_total);
4392         btrfs_set_super_total_bytes(super_copy, old_total - diff);
4393         unlock_chunks(root);
4394 
4395         /* Now btrfs_update_device() will change the on-disk size. */
4396         ret = btrfs_update_device(trans, device);
4397         btrfs_end_transaction(trans, root);
4398 done:
4399         btrfs_free_path(path);
4400         if (ret) {
4401                 lock_chunks(root);
4402                 btrfs_device_set_total_bytes(device, old_size);
4403                 if (device->writeable)
4404                         device->fs_devices->total_rw_bytes += diff;
4405                 spin_lock(&root->fs_info->free_chunk_lock);
4406                 root->fs_info->free_chunk_space += diff;
4407                 spin_unlock(&root->fs_info->free_chunk_lock);
4408                 unlock_chunks(root);
4409         }
4410         return ret;
4411 }
4412 
4413 static int btrfs_add_system_chunk(struct btrfs_root *root,
4414                            struct btrfs_key *key,
4415                            struct btrfs_chunk *chunk, int item_size)
4416 {
4417         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4418         struct btrfs_disk_key disk_key;
4419         u32 array_size;
4420         u8 *ptr;
4421 
4422         lock_chunks(root);
4423         array_size = btrfs_super_sys_array_size(super_copy);
4424         if (array_size + item_size + sizeof(disk_key)
4425                         > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4426                 unlock_chunks(root);
4427                 return -EFBIG;
4428         }
4429 
4430         ptr = super_copy->sys_chunk_array + array_size;
4431         btrfs_cpu_key_to_disk(&disk_key, key);
4432         memcpy(ptr, &disk_key, sizeof(disk_key));
4433         ptr += sizeof(disk_key);
4434         memcpy(ptr, chunk, item_size);
4435         item_size += sizeof(disk_key);
4436         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4437         unlock_chunks(root);
4438 
4439         return 0;
4440 }
4441 
4442 /*
4443  * sort the devices in descending order by max_avail, total_avail
4444  */
4445 static int btrfs_cmp_device_info(const void *a, const void *b)
4446 {
4447         const struct btrfs_device_info *di_a = a;
4448         const struct btrfs_device_info *di_b = b;
4449 
4450         if (di_a->max_avail > di_b->max_avail)
4451                 return -1;
4452         if (di_a->max_avail < di_b->max_avail)
4453                 return 1;
4454         if (di_a->total_avail > di_b->total_avail)
4455                 return -1;
4456         if (di_a->total_avail < di_b->total_avail)
4457                 return 1;
4458         return 0;
4459 }
4460 
4461 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4462 {
4463         /* TODO allow them to set a preferred stripe size */
4464         return SZ_64K;
4465 }
4466 
4467 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4468 {
4469         if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4470                 return;
4471 
4472         btrfs_set_fs_incompat(info, RAID56);
4473 }
4474 
4475 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)             \
4476                         - sizeof(struct btrfs_item)             \
4477                         - sizeof(struct btrfs_chunk))           \
4478                         / sizeof(struct btrfs_stripe) + 1)
4479 
4480 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE        \
4481                                 - 2 * sizeof(struct btrfs_disk_key)     \
4482                                 - 2 * sizeof(struct btrfs_chunk))       \
4483                                 / sizeof(struct btrfs_stripe) + 1)
4484 
4485 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4486                                struct btrfs_root *extent_root, u64 start,
4487                                u64 type)
4488 {
4489         struct btrfs_fs_info *info = extent_root->fs_info;
4490         struct btrfs_fs_devices *fs_devices = info->fs_devices;
4491         struct list_head *cur;
4492         struct map_lookup *map = NULL;
4493         struct extent_map_tree *em_tree;
4494         struct extent_map *em;
4495         struct btrfs_device_info *devices_info = NULL;
4496         u64 total_avail;
4497         int num_stripes;        /* total number of stripes to allocate */
4498         int data_stripes;       /* number of stripes that count for
4499                                    block group size */
4500         int sub_stripes;        /* sub_stripes info for map */
4501         int dev_stripes;        /* stripes per dev */
4502         int devs_max;           /* max devs to use */
4503         int devs_min;           /* min devs needed */
4504         int devs_increment;     /* ndevs has to be a multiple of this */
4505         int ncopies;            /* how many copies to data has */
4506         int ret;
4507         u64 max_stripe_size;
4508         u64 max_chunk_size;
4509         u64 stripe_size;
4510         u64 num_bytes;
4511         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4512         int ndevs;
4513         int i;
4514         int j;
4515         int index;
4516 
4517         BUG_ON(!alloc_profile_is_valid(type, 0));
4518 
4519         if (list_empty(&fs_devices->alloc_list))
4520                 return -ENOSPC;
4521 
4522         index = __get_raid_index(type);
4523 
4524         sub_stripes = btrfs_raid_array[index].sub_stripes;
4525         dev_stripes = btrfs_raid_array[index].dev_stripes;
4526         devs_max = btrfs_raid_array[index].devs_max;
4527         devs_min = btrfs_raid_array[index].devs_min;
4528         devs_increment = btrfs_raid_array[index].devs_increment;
4529         ncopies = btrfs_raid_array[index].ncopies;
4530 
4531         if (type & BTRFS_BLOCK_GROUP_DATA) {
4532                 max_stripe_size = SZ_1G;
4533                 max_chunk_size = 10 * max_stripe_size;
4534                 if (!devs_max)
4535                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4536         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4537                 /* for larger filesystems, use larger metadata chunks */
4538                 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4539                         max_stripe_size = SZ_1G;
4540                 else
4541                         max_stripe_size = SZ_256M;
4542                 max_chunk_size = max_stripe_size;
4543                 if (!devs_max)
4544                         devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4545         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4546                 max_stripe_size = SZ_32M;
4547                 max_chunk_size = 2 * max_stripe_size;
4548                 if (!devs_max)
4549                         devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4550         } else {
4551                 btrfs_err(info, "invalid chunk type 0x%llx requested",
4552                        type);
4553                 BUG_ON(1);
4554         }
4555 
4556         /* we don't want a chunk larger than 10% of writeable space */
4557         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4558                              max_chunk_size);
4559 
4560         devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4561                                GFP_NOFS);
4562         if (!devices_info)
4563                 return -ENOMEM;
4564 
4565         cur = fs_devices->alloc_list.next;
4566 
4567         /*
4568          * in the first pass through the devices list, we gather information
4569          * about the available holes on each device.
4570          */
4571         ndevs = 0;
4572         while (cur != &fs_devices->alloc_list) {
4573                 struct btrfs_device *device;
4574                 u64 max_avail;
4575                 u64 dev_offset;
4576 
4577                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4578 
4579                 cur = cur->next;
4580 
4581                 if (!device->writeable) {
4582                         WARN(1, KERN_ERR
4583                                "BTRFS: read-only device in alloc_list\n");
4584                         continue;
4585                 }
4586 
4587                 if (!device->in_fs_metadata ||
4588                     device->is_tgtdev_for_dev_replace)
4589                         continue;
4590 
4591                 if (device->total_bytes > device->bytes_used)
4592                         total_avail = device->total_bytes - device->bytes_used;
4593                 else
4594                         total_avail = 0;
4595 
4596                 /* If there is no space on this device, skip it. */
4597                 if (total_avail == 0)
4598                         continue;
4599 
4600                 ret = find_free_dev_extent(trans, device,
4601                                            max_stripe_size * dev_stripes,
4602                                            &dev_offset, &max_avail);
4603                 if (ret && ret != -ENOSPC)
4604                         goto error;
4605 
4606                 if (ret == 0)
4607                         max_avail = max_stripe_size * dev_stripes;
4608 
4609                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4610                         continue;
4611 
4612                 if (ndevs == fs_devices->rw_devices) {
4613                         WARN(1, "%s: found more than %llu devices\n",
4614                              __func__, fs_devices->rw_devices);
4615                         break;
4616                 }
4617                 devices_info[ndevs].dev_offset = dev_offset;
4618                 devices_info[ndevs].max_avail = max_avail;
4619                 devices_info[ndevs].total_avail = total_avail;
4620                 devices_info[ndevs].dev = device;
4621                 ++ndevs;
4622         }
4623 
4624         /*
4625          * now sort the devices by hole size / available space
4626          */
4627         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4628              btrfs_cmp_device_info, NULL);
4629 
4630         /* round down to number of usable stripes */
4631         ndevs -= ndevs % devs_increment;
4632 
4633         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4634                 ret = -ENOSPC;
4635                 goto error;
4636         }
4637 
4638         if (devs_max && ndevs > devs_max)
4639                 ndevs = devs_max;
4640         /*
4641          * the primary goal is to maximize the number of stripes, so use as many
4642          * devices as possible, even if the stripes are not maximum sized.
4643          */
4644         stripe_size = devices_info[ndevs-1].max_avail;
4645         num_stripes = ndevs * dev_stripes;
4646 
4647         /*
4648          * this will have to be fixed for RAID1 and RAID10 over
4649          * more drives
4650          */
4651         data_stripes = num_stripes / ncopies;
4652 
4653         if (type & BTRFS_BLOCK_GROUP_RAID5) {
4654                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4655                                  btrfs_super_stripesize(info->super_copy));
4656                 data_stripes = num_stripes - 1;
4657         }
4658         if (type & BTRFS_BLOCK_GROUP_RAID6) {
4659                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4660                                  btrfs_super_stripesize(info->super_copy));
4661                 data_stripes = num_stripes - 2;
4662         }
4663 
4664         /*
4665          * Use the number of data stripes to figure out how big this chunk
4666          * is really going to be in terms of logical address space,
4667          * and compare that answer with the max chunk size
4668          */
4669         if (stripe_size * data_stripes > max_chunk_size) {
4670                 u64 mask = (1ULL << 24) - 1;
4671 
4672                 stripe_size = div_u64(max_chunk_size, data_stripes);
4673 
4674                 /* bump the answer up to a 16MB boundary */
4675                 stripe_size = (stripe_size + mask) & ~mask;
4676 
4677                 /* but don't go higher than the limits we found
4678                  * while searching for free extents
4679                  */
4680                 if (stripe_size > devices_info[ndevs-1].max_avail)
4681                         stripe_size = devices_info[ndevs-1].max_avail;
4682         }
4683 
4684         stripe_size = div_u64(stripe_size, dev_stripes);
4685 
4686         /* align to BTRFS_STRIPE_LEN */
4687         stripe_size = div_u64(stripe_size, raid_stripe_len);
4688         stripe_size *= raid_stripe_len;
4689 
4690         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4691         if (!map) {
4692                 ret = -ENOMEM;
4693                 goto error;
4694         }
4695         map->num_stripes = num_stripes;
4696 
4697         for (i = 0; i < ndevs; ++i) {
4698                 for (j = 0; j < dev_stripes; ++j) {
4699                         int s = i * dev_stripes + j;
4700                         map->stripes[s].dev = devices_info[i].dev;
4701                         map->stripes[s].physical = devices_info[i].dev_offset +
4702                                                    j * stripe_size;
4703                 }
4704         }
4705         map->sector_size = extent_root->sectorsize;
4706         map->stripe_len = raid_stripe_len;
4707         map->io_align = raid_stripe_len;
4708         map->io_width = raid_stripe_len;
4709         map->type = type;
4710         map->sub_stripes = sub_stripes;
4711 
4712         num_bytes = stripe_size * data_stripes;
4713 
4714         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4715 
4716         em = alloc_extent_map();
4717         if (!em) {
4718                 kfree(map);
4719                 ret = -ENOMEM;
4720                 goto error;
4721         }
4722         set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4723         em->map_lookup = map;
4724         em->start = start;
4725         em->len = num_bytes;
4726         em->block_start = 0;
4727         em->block_len = em->len;
4728         em->orig_block_len = stripe_size;
4729 
4730         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4731         write_lock(&em_tree->lock);
4732         ret = add_extent_mapping(em_tree, em, 0);
4733         if (!ret) {
4734                 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4735                 atomic_inc(&em->refs);
4736         }
4737         write_unlock(&em_tree->lock);
4738         if (ret) {
4739                 free_extent_map(em);
4740                 goto error;
4741         }
4742 
4743         ret = btrfs_make_block_group(trans, extent_root, 0, type,
4744                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4745                                      start, num_bytes);
4746         if (ret)
4747                 goto error_del_extent;
4748 
4749         for (i = 0; i < map->num_stripes; i++) {
4750                 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4751                 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4752         }
4753 
4754         spin_lock(&extent_root->fs_info->free_chunk_lock);
4755         extent_root->fs_info->free_chunk_space -= (stripe_size *
4756                                                    map->num_stripes);
4757         spin_unlock(&extent_root->fs_info->free_chunk_lock);
4758 
4759         free_extent_map(em);
4760         check_raid56_incompat_flag(extent_root->fs_info, type);
4761 
4762         kfree(devices_info);
4763         return 0;
4764 
4765 error_del_extent:
4766         write_lock(&em_tree->lock);
4767         remove_extent_mapping(em_tree, em);
4768         write_unlock(&em_tree->lock);
4769 
4770         /* One for our allocation */
4771         free_extent_map(em);
4772         /* One for the tree reference */
4773         free_extent_map(em);
4774         /* One for the pending_chunks list reference */
4775         free_extent_map(em);
4776 error:
4777         kfree(devices_info);
4778         return ret;
4779 }
4780 
4781 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4782                                 struct btrfs_root *extent_root,
4783                                 u64 chunk_offset, u64 chunk_size)
4784 {
4785         struct btrfs_key key;
4786         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4787         struct btrfs_device *device;
4788         struct btrfs_chunk *chunk;
4789         struct btrfs_stripe *stripe;
4790         struct extent_map_tree *em_tree;
4791         struct extent_map *em;
4792         struct map_lookup *map;
4793         size_t item_size;
4794         u64 dev_offset;
4795         u64 stripe_size;
4796         int i = 0;
4797         int ret = 0;
4798 
4799         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4800         read_lock(&em_tree->lock);
4801         em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4802         read_unlock(&em_tree->lock);
4803 
4804         if (!em) {
4805                 btrfs_crit(extent_root->fs_info, "unable to find logical "
4806                            "%Lu len %Lu", chunk_offset, chunk_size);
4807                 return -EINVAL;
4808         }
4809 
4810         if (em->start != chunk_offset || em->len != chunk_size) {
4811                 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4812                           " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4813                           chunk_size, em->start, em->len);
4814                 free_extent_map(em);
4815                 return -EINVAL;
4816         }
4817 
4818         map = em->map_lookup;
4819         item_size = btrfs_chunk_item_size(map->num_stripes);
4820         stripe_size = em->orig_block_len;
4821 
4822         chunk = kzalloc(item_size, GFP_NOFS);
4823         if (!chunk) {
4824                 ret = -ENOMEM;
4825                 goto out;
4826         }
4827 
4828         /*
4829          * Take the device list mutex to prevent races with the final phase of
4830          * a device replace operation that replaces the device object associated
4831          * with the map's stripes, because the device object's id can change
4832          * at any time during that final phase of the device replace operation
4833          * (dev-replace.c:btrfs_dev_replace_finishing()).
4834          */
4835         mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4836         for (i = 0; i < map->num_stripes; i++) {
4837                 device = map->stripes[i].dev;
4838                 dev_offset = map->stripes[i].physical;
4839 
4840                 ret = btrfs_update_device(trans, device);
4841                 if (ret)
4842                         break;
4843                 ret = btrfs_alloc_dev_extent(trans, device,
4844                                              chunk_root->root_key.objectid,
4845                                              BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4846                                              chunk_offset, dev_offset,
4847                                              stripe_size);
4848                 if (ret)
4849                         break;
4850         }
4851         if (ret) {
4852                 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4853                 goto out;
4854         }
4855 
4856         stripe = &chunk->stripe;
4857         for (i = 0; i < map->num_stripes; i++) {
4858                 device = map->stripes[i].dev;
4859                 dev_offset = map->stripes[i].physical;
4860 
4861                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4862                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4863                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4864                 stripe++;
4865         }
4866         mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4867 
4868         btrfs_set_stack_chunk_length(chunk, chunk_size);
4869         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4870         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4871         btrfs_set_stack_chunk_type(chunk, map->type);
4872         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4873         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4874         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4875         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4876         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4877 
4878         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4879         key.type = BTRFS_CHUNK_ITEM_KEY;
4880         key.offset = chunk_offset;
4881 
4882         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4883         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4884                 /*
4885                  * TODO: Cleanup of inserted chunk root in case of
4886                  * failure.
4887                  */
4888                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4889                                              item_size);
4890         }
4891 
4892 out:
4893         kfree(chunk);
4894         free_extent_map(em);
4895         return ret;
4896 }
4897 
4898 /*
4899  * Chunk allocation falls into two parts. The first part does works
4900  * that make the new allocated chunk useable, but not do any operation
4901  * that modifies the chunk tree. The second part does the works that
4902  * require modifying the chunk tree. This division is important for the
4903  * bootstrap process of adding storage to a seed btrfs.
4904  */
4905 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4906                       struct btrfs_root *extent_root, u64 type)
4907 {
4908         u64 chunk_offset;
4909 
4910         ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4911         chunk_offset = find_next_chunk(extent_root->fs_info);
4912         return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4913 }
4914 
4915 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4916                                          struct btrfs_root *root,
4917                                          struct btrfs_device *device)
4918 {
4919         u64 chunk_offset;
4920         u64 sys_chunk_offset;
4921         u64 alloc_profile;
4922         struct btrfs_fs_info *fs_info = root->fs_info;
4923         struct btrfs_root *extent_root = fs_info->extent_root;
4924         int ret;
4925 
4926         chunk_offset = find_next_chunk(fs_info);
4927         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4928         ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4929                                   alloc_profile);
4930         if (ret)
4931                 return ret;
4932 
4933         sys_chunk_offset = find_next_chunk(root->fs_info);
4934         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4935         ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4936                                   alloc_profile);
4937         return ret;
4938 }
4939 
4940 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4941 {
4942         int max_errors;
4943 
4944         if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4945                          BTRFS_BLOCK_GROUP_RAID10 |
4946                          BTRFS_BLOCK_GROUP_RAID5 |
4947                          BTRFS_BLOCK_GROUP_DUP)) {
4948                 max_errors = 1;
4949         } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4950                 max_errors = 2;
4951         } else {
4952                 max_errors = 0;
4953         }
4954 
4955         return max_errors;
4956 }
4957 
4958 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4959 {
4960         struct extent_map *em;
4961         struct map_lookup *map;
4962         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4963         int readonly = 0;
4964         int miss_ndevs = 0;
4965         int i;
4966 
4967         read_lock(&map_tree->map_tree.lock);
4968         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4969         read_unlock(&map_tree->map_tree.lock);
4970         if (!em)
4971                 return 1;
4972 
4973         map = em->map_lookup;
4974         for (i = 0; i < map->num_stripes; i++) {
4975                 if (map->stripes[i].dev->missing) {
4976                         miss_ndevs++;
4977                         continue;
4978                 }
4979 
4980                 if (!map->stripes[i].dev->writeable) {
4981                         readonly = 1;
4982                         goto end;
4983                 }
4984         }
4985 
4986         /*
4987          * If the number of missing devices is larger than max errors,
4988          * we can not write the data into that chunk successfully, so
4989          * set it readonly.
4990          */
4991         if (miss_ndevs > btrfs_chunk_max_errors(map))
4992                 readonly = 1;
4993 end:
4994         free_extent_map(em);
4995         return readonly;
4996 }
4997 
4998 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4999 {
5000         extent_map_tree_init(&tree->map_tree);
5001 }
5002 
5003 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5004 {
5005         struct extent_map *em;
5006 
5007         while (1) {
5008                 write_lock(&tree->map_tree.lock);
5009                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5010                 if (em)
5011                         remove_extent_mapping(&tree->map_tree, em);
5012                 write_unlock(&tree->map_tree.lock);
5013                 if (!em)
5014                         break;
5015                 /* once for us */
5016                 free_extent_map(em);
5017                 /* once for the tree */
5018                 free_extent_map(em);
5019         }
5020 }
5021 
5022 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5023 {
5024         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5025         struct extent_map *em;
5026         struct map_lookup *map;
5027         struct extent_map_tree *em_tree = &map_tree->map_tree;
5028         int ret;
5029 
5030         read_lock(&em_tree->lock);
5031         em = lookup_extent_mapping(em_tree, logical, len);
5032         read_unlock(&em_tree->lock);
5033 
5034         /*
5035          * We could return errors for these cases, but that could get ugly and
5036          * we'd probably do the same thing which is just not do anything else
5037          * and exit, so return 1 so the callers don't try to use other copies.
5038          */
5039         if (!em) {
5040                 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5041                             logical+len);
5042                 return 1;
5043         }
5044 
5045         if (em->start > logical || em->start + em->len < logical) {
5046                 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5047                             "%Lu-%Lu", logical, logical+len, em->start,
5048                             em->start + em->len);
5049                 free_extent_map(em);
5050                 return 1;
5051         }
5052 
5053         map = em->map_lookup;
5054         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5055                 ret = map->num_stripes;
5056         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5057                 ret = map->sub_stripes;
5058         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5059                 ret = 2;
5060         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5061                 ret = 3;
5062         else
5063                 ret = 1;
5064         free_extent_map(em);
5065 
5066         btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5067         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5068                 ret++;
5069         btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5070 
5071         return ret;
5072 }
5073 
5074 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5075                                     struct btrfs_mapping_tree *map_tree,
5076                                     u64 logical)
5077 {
5078         struct extent_map *em;
5079         struct map_lookup *map;
5080         struct extent_map_tree *em_tree = &map_tree->map_tree;
5081         unsigned long len = root->sectorsize;
5082 
5083         read_lock(&em_tree->lock);
5084         em = lookup_extent_mapping(em_tree, logical, len);
5085         read_unlock(&em_tree->lock);
5086         BUG_ON(!em);
5087 
5088         BUG_ON(em->start > logical || em->start + em->len < logical);
5089         map = em->map_lookup;
5090         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5091                 len = map->stripe_len * nr_data_stripes(map);
5092         free_extent_map(em);
5093         return len;
5094 }
5095 
5096 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5097                            u64 logical, u64 len, int mirror_num)
5098 {
5099         struct extent_map *em;
5100         struct map_lookup *map;
5101         struct extent_map_tree *em_tree = &map_tree->map_tree;
5102         int ret = 0;
5103 
5104         read_lock(&em_tree->lock);
5105         em = lookup_extent_mapping(em_tree, logical, len);
5106         read_unlock(&em_tree->lock);
5107         BUG_ON(!em);
5108 
5109         BUG_ON(em->start > logical || em->start + em->len < logical);
5110         map = em->map_lookup;
5111         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5112                 ret = 1;
5113         free_extent_map(em);
5114         return ret;
5115 }
5116 
5117 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5118                             struct map_lookup *map, int first, int num,
5119                             int optimal, int dev_replace_is_ongoing)
5120 {
5121         int i;
5122         int tolerance;
5123         struct btrfs_device *srcdev;
5124 
5125         if (dev_replace_is_ongoing &&
5126             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5127              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5128                 srcdev = fs_info->dev_replace.srcdev;
5129         else
5130                 srcdev = NULL;
5131 
5132         /*
5133          * try to avoid the drive that is the source drive for a
5134          * dev-replace procedure, only choose it if no other non-missing
5135          * mirror is available
5136          */
5137         for (tolerance = 0; tolerance < 2; tolerance++) {
5138                 if (map->stripes[optimal].dev->bdev &&
5139                     (tolerance || map->stripes[optimal].dev != srcdev))
5140                         return optimal;
5141                 for (i = first; i < first + num; i++) {
5142                         if (map->stripes[i].dev->bdev &&
5143                             (tolerance || map->stripes[i].dev != srcdev))
5144                                 return i;
5145                 }
5146         }
5147 
5148         /* we couldn't find one that doesn't fail.  Just return something
5149          * and the io error handling code will clean up eventually
5150          */
5151         return optimal;
5152 }
5153 
5154 static inline int parity_smaller(u64 a, u64 b)
5155 {
5156         return a > b;
5157 }
5158 
5159 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5160 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5161 {
5162         struct btrfs_bio_stripe s;
5163         int i;
5164         u64 l;
5165         int again = 1;
5166 
5167         while (again) {
5168                 again = 0;
5169                 for (i = 0; i < num_stripes - 1; i++) {
5170                         if (parity_smaller(bbio->raid_map[i],
5171                                            bbio->raid_map[i+1])) {
5172                                 s = bbio->stripes[i];
5173                                 l = bbio->raid_map[i];
5174                                 bbio->stripes[i] = bbio->stripes[i+1];
5175                                 bbio->raid_map[i] = bbio->raid_map[i+1];
5176                                 bbio->stripes[i+1] = s;
5177                                 bbio->raid_map[i+1] = l;
5178 
5179                                 again = 1;
5180                         }
5181                 }
5182         }
5183 }
5184 
5185 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5186 {
5187         struct btrfs_bio *bbio = kzalloc(
5188                  /* the size of the btrfs_bio */
5189                 sizeof(struct btrfs_bio) +
5190                 /* plus the variable array for the stripes */
5191                 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5192                 /* plus the variable array for the tgt dev */
5193                 sizeof(int) * (real_stripes) +
5194                 /*
5195                  * plus the raid_map, which includes both the tgt dev
5196                  * and the stripes
5197                  */
5198                 sizeof(u64) * (total_stripes),
5199                 GFP_NOFS|__GFP_NOFAIL);
5200 
5201         atomic_set(&bbio->error, 0);
5202         atomic_set(&bbio->refs, 1);
5203 
5204         return bbio;
5205 }
5206 
5207 void btrfs_get_bbio(struct btrfs_bio *bbio)
5208 {
5209         WARN_ON(!atomic_read(&bbio->refs));
5210         atomic_inc(&bbio->refs);
5211 }
5212 
5213 void btrfs_put_bbio(struct btrfs_bio *bbio)
5214 {
5215         if (!bbio)
5216                 return;
5217         if (atomic_dec_and_test(&bbio->refs))
5218                 kfree(bbio);
5219 }
5220 
5221 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5222                              u64 logical, u64 *length,
5223                              struct btrfs_bio **bbio_ret,
5224                              int mirror_num, int need_raid_map)
5225 {
5226         struct extent_map *em;
5227         struct map_lookup *map;
5228         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5229         struct extent_map_tree *em_tree = &map_tree->map_tree;
5230         u64 offset;
5231         u64 stripe_offset;
5232         u64 stripe_end_offset;
5233         u64 stripe_nr;
5234         u64 stripe_nr_orig;
5235         u64 stripe_nr_end;
5236         u64 stripe_len;
5237         u32 stripe_index;
5238         int i;
5239         int ret = 0;
5240         int num_stripes;
5241         int max_errors = 0;
5242         int tgtdev_indexes = 0;
5243         struct btrfs_bio *bbio = NULL;
5244         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5245         int dev_replace_is_ongoing = 0;
5246         int num_alloc_stripes;
5247         int patch_the_first_stripe_for_dev_replace = 0;
5248         u64 physical_to_patch_in_first_stripe = 0;
5249         u64 raid56_full_stripe_start = (u64)-1;
5250 
5251         read_lock(&em_tree->lock);
5252         em = lookup_extent_mapping(em_tree, logical, *length);
5253         read_unlock(&em_tree->lock);
5254 
5255         if (!em) {
5256                 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5257                         logical, *length);
5258                 return -EINVAL;
5259         }
5260 
5261         if (em->start > logical || em->start + em->len < logical) {
5262                 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5263                            "found %Lu-%Lu", logical, em->start,
5264                            em->start + em->len);
5265                 free_extent_map(em);
5266                 return -EINVAL;
5267         }
5268 
5269         map = em->map_lookup;
5270         offset = logical - em->start;
5271 
5272         stripe_len = map->stripe_len;
5273         stripe_nr = offset;
5274         /*
5275          * stripe_nr counts the total number of stripes we have to stride
5276          * to get to this block
5277          */
5278         stripe_nr = div64_u64(stripe_nr, stripe_len);
5279 
5280         stripe_offset = stripe_nr * stripe_len;
5281         BUG_ON(offset < stripe_offset);
5282 
5283         /* stripe_offset is the offset of this block in its stripe*/
5284         stripe_offset = offset - stripe_offset;
5285 
5286         /* if we're here for raid56, we need to know the stripe aligned start */
5287         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5288                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5289                 raid56_full_stripe_start = offset;
5290 
5291                 /* allow a write of a full stripe, but make sure we don't
5292                  * allow straddling of stripes
5293                  */
5294                 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5295                                 full_stripe_len);
5296                 raid56_full_stripe_start *= full_stripe_len;
5297         }
5298 
5299         if (rw & REQ_DISCARD) {
5300                 /* we don't discard raid56 yet */
5301                 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5302                         ret = -EOPNOTSUPP;
5303                         goto out;
5304                 }
5305                 *length = min_t(u64, em->len - offset, *length);
5306         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5307                 u64 max_len;
5308                 /* For writes to RAID[56], allow a full stripeset across all disks.
5309                    For other RAID types and for RAID[56] reads, just allow a single
5310                    stripe (on a single disk). */
5311                 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5312                     (rw & REQ_WRITE)) {
5313                         max_len = stripe_len * nr_data_stripes(map) -
5314                                 (offset - raid56_full_stripe_start);
5315                 } else {
5316                         /* we limit the length of each bio to what fits in a stripe */
5317                         max_len = stripe_len - stripe_offset;
5318                 }
5319                 *length = min_t(u64, em->len - offset, max_len);
5320         } else {
5321                 *length = em->len - offset;
5322         }
5323 
5324         /* This is for when we're called from btrfs_merge_bio_hook() and all
5325            it cares about is the length */
5326         if (!bbio_ret)
5327                 goto out;
5328 
5329         btrfs_dev_replace_lock(dev_replace, 0);
5330         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5331         if (!dev_replace_is_ongoing)
5332                 btrfs_dev_replace_unlock(dev_replace, 0);
5333         else
5334                 btrfs_dev_replace_set_lock_blocking(dev_replace);
5335 
5336         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5337             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5338             dev_replace->tgtdev != NULL) {
5339                 /*
5340                  * in dev-replace case, for repair case (that's the only
5341                  * case where the mirror is selected explicitly when
5342                  * calling btrfs_map_block), blocks left of the left cursor
5343                  * can also be read from the target drive.
5344                  * For REQ_GET_READ_MIRRORS, the target drive is added as
5345                  * the last one to the array of stripes. For READ, it also
5346                  * needs to be supported using the same mirror number.
5347                  * If the requested block is not left of the left cursor,
5348                  * EIO is returned. This can happen because btrfs_num_copies()
5349                  * returns one more in the dev-replace case.
5350                  */
5351                 u64 tmp_length = *length;
5352                 struct btrfs_bio *tmp_bbio = NULL;
5353                 int tmp_num_stripes;
5354                 u64 srcdev_devid = dev_replace->srcdev->devid;
5355                 int index_srcdev = 0;
5356                 int found = 0;
5357                 u64 physical_of_found = 0;
5358 
5359                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5360                              logical, &tmp_length, &tmp_bbio, 0, 0);
5361                 if (ret) {
5362                         WARN_ON(tmp_bbio != NULL);
5363                         goto out;
5364                 }
5365 
5366                 tmp_num_stripes = tmp_bbio->num_stripes;
5367                 if (mirror_num > tmp_num_stripes) {
5368                         /*
5369                          * REQ_GET_READ_MIRRORS does not contain this
5370                          * mirror, that means that the requested area
5371                          * is not left of the left cursor
5372                          */
5373                         ret = -EIO;
5374                         btrfs_put_bbio(tmp_bbio);
5375                         goto out;
5376                 }
5377 
5378                 /*
5379                  * process the rest of the function using the mirror_num
5380                  * of the source drive. Therefore look it up first.
5381                  * At the end, patch the device pointer to the one of the
5382                  * target drive.
5383                  */
5384                 for (i = 0; i < tmp_num_stripes; i++) {
5385                         if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5386                                 continue;
5387 
5388                         /*
5389                          * In case of DUP, in order to keep it simple, only add
5390                          * the mirror with the lowest physical address
5391                          */
5392                         if (found &&
5393                             physical_of_found <= tmp_bbio->stripes[i].physical)
5394                                 continue;
5395 
5396                         index_srcdev = i;
5397                         found = 1;
5398                         physical_of_found = tmp_bbio->stripes[i].physical;
5399                 }
5400 
5401                 btrfs_put_bbio(tmp_bbio);
5402 
5403                 if (!found) {
5404                         WARN_ON(1);
5405                         ret = -EIO;
5406                         goto out;
5407                 }
5408 
5409                 mirror_num = index_srcdev + 1;
5410                 patch_the_first_stripe_for_dev_replace = 1;
5411                 physical_to_patch_in_first_stripe = physical_of_found;
5412         } else if (mirror_num > map->num_stripes) {
5413                 mirror_num = 0;
5414         }
5415 
5416         num_stripes = 1;
5417         stripe_index = 0;
5418         stripe_nr_orig = stripe_nr;
5419         stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5420         stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5421         stripe_end_offset = stripe_nr_end * map->stripe_len -
5422                             (offset + *length);
5423 
5424         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5425                 if (rw & REQ_DISCARD)
5426                         num_stripes = min_t(u64, map->num_stripes,
5427                                             stripe_nr_end - stripe_nr_orig);
5428                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5429                                 &stripe_index);
5430                 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5431                         mirror_num = 1;
5432         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5433                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5434                         num_stripes = map->num_stripes;
5435                 else if (mirror_num)
5436                         stripe_index = mirror_num - 1;
5437                 else {
5438                         stripe_index = find_live_mirror(fs_info, map, 0,
5439                                             map->num_stripes,
5440                                             current->pid % map->num_stripes,
5441                                             dev_replace_is_ongoing);
5442                         mirror_num = stripe_index + 1;
5443                 }
5444 
5445         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5446                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5447                         num_stripes = map->num_stripes;
5448                 } else if (mirror_num) {
5449                         stripe_index = mirror_num - 1;
5450                 } else {
5451                         mirror_num = 1;
5452                 }
5453 
5454         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5455                 u32 factor = map->num_stripes / map->sub_stripes;
5456 
5457                 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5458                 stripe_index *= map->sub_stripes;
5459 
5460                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5461                         num_stripes = map->sub_stripes;
5462                 else if (rw & REQ_DISCARD)
5463                         num_stripes = min_t(u64, map->sub_stripes *
5464                                             (stripe_nr_end - stripe_nr_orig),
5465                                             map->num_stripes);
5466                 else if (mirror_num)
5467                         stripe_index += mirror_num - 1;
5468                 else {
5469                         int old_stripe_index = stripe_index;
5470                         stripe_index = find_live_mirror(fs_info, map,
5471                                               stripe_index,
5472                                               map->sub_stripes, stripe_index +
5473                                               current->pid % map->sub_stripes,
5474                                               dev_replace_is_ongoing);
5475                         mirror_num = stripe_index - old_stripe_index + 1;
5476                 }
5477 
5478         } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5479                 if (need_raid_map &&
5480                     ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5481                      mirror_num > 1)) {
5482                         /* push stripe_nr back to the start of the full stripe */
5483                         stripe_nr = div_u64(raid56_full_stripe_start,
5484                                         stripe_len * nr_data_stripes(map));
5485 
5486                         /* RAID[56] write or recovery. Return all stripes */
5487                         num_stripes = map->num_stripes;
5488                         max_errors = nr_parity_stripes(map);
5489 
5490                         *length = map->stripe_len;
5491                         stripe_index = 0;
5492                         stripe_offset = 0;
5493                 } else {
5494                         /*
5495                          * Mirror #0 or #1 means the original data block.
5496                          * Mirror #2 is RAID5 parity block.
5497                          * Mirror #3 is RAID6 Q block.
5498                          */
5499                         stripe_nr = div_u64_rem(stripe_nr,
5500                                         nr_data_stripes(map), &stripe_index);
5501                         if (mirror_num > 1)
5502                                 stripe_index = nr_data_stripes(map) +
5503                                                 mirror_num - 2;
5504 
5505                         /* We distribute the parity blocks across stripes */
5506                         div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5507                                         &stripe_index);
5508                         if (!(rw & (REQ_WRITE | REQ_DISCARD |
5509                                     REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5510                                 mirror_num = 1;
5511                 }
5512         } else {
5513                 /*
5514                  * after this, stripe_nr is the number of stripes on this
5515                  * device we have to walk to find the data, and stripe_index is
5516                  * the number of our device in the stripe array
5517                  */
5518                 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5519                                 &stripe_index);
5520                 mirror_num = stripe_index + 1;
5521         }
5522         BUG_ON(stripe_index >= map->num_stripes);
5523 
5524         num_alloc_stripes = num_stripes;
5525         if (dev_replace_is_ongoing) {
5526                 if (rw & (REQ_WRITE | REQ_DISCARD))
5527                         num_alloc_stripes <<= 1;
5528                 if (rw & REQ_GET_READ_MIRRORS)
5529                         num_alloc_stripes++;
5530                 tgtdev_indexes = num_stripes;
5531         }
5532 
5533         bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5534         if (!bbio) {
5535                 ret = -ENOMEM;
5536                 goto out;
5537         }
5538         if (dev_replace_is_ongoing)
5539                 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5540 
5541         /* build raid_map */
5542         if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5543             need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5544             mirror_num > 1)) {
5545                 u64 tmp;
5546                 unsigned rot;
5547 
5548                 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5549                                  sizeof(struct btrfs_bio_stripe) *
5550                                  num_alloc_stripes +
5551                                  sizeof(int) * tgtdev_indexes);
5552 
5553                 /* Work out the disk rotation on this stripe-set */
5554                 div_u64_rem(stripe_nr, num_stripes, &rot);
5555 
5556                 /* Fill in the logical address of each stripe */
5557                 tmp = stripe_nr * nr_data_stripes(map);
5558                 for (i = 0; i < nr_data_stripes(map); i++)
5559                         bbio->raid_map[(i+rot) % num_stripes] =
5560                                 em->start + (tmp + i) * map->stripe_len;
5561 
5562                 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5563                 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5564                         bbio->raid_map[(i+rot+1) % num_stripes] =
5565                                 RAID6_Q_STRIPE;
5566         }
5567 
5568         if (rw & REQ_DISCARD) {
5569                 u32 factor = 0;
5570                 u32 sub_stripes = 0;
5571                 u64 stripes_per_dev = 0;
5572                 u32 remaining_stripes = 0;
5573                 u32 last_stripe = 0;
5574 
5575                 if (map->type &
5576                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5577                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5578                                 sub_stripes = 1;
5579                         else
5580                                 sub_stripes = map->sub_stripes;
5581 
5582                         factor = map->num_stripes / sub_stripes;
5583                         stripes_per_dev = div_u64_rem(stripe_nr_end -
5584                                                       stripe_nr_orig,
5585                                                       factor,
5586                                                       &remaining_stripes);
5587                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5588                         last_stripe *= sub_stripes;
5589                 }
5590 
5591                 for (i = 0; i < num_stripes; i++) {
5592                         bbio->stripes[i].physical =
5593                                 map->stripes[stripe_index].physical +
5594                                 stripe_offset + stripe_nr * map->stripe_len;
5595                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5596 
5597                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5598                                          BTRFS_BLOCK_GROUP_RAID10)) {
5599                                 bbio->stripes[i].length = stripes_per_dev *
5600                                                           map->stripe_len;
5601 
5602                                 if (i / sub_stripes < remaining_stripes)
5603                                         bbio->stripes[i].length +=
5604                                                 map->stripe_len;
5605 
5606                                 /*
5607                                  * Special for the first stripe and
5608                                  * the last stripe:
5609                                  *
5610                                  * |-------|...|-------|
5611                                  *     |----------|
5612                                  *    off     end_off
5613                                  */
5614                                 if (i < sub_stripes)
5615                                         bbio->stripes[i].length -=
5616                                                 stripe_offset;
5617 
5618                                 if (stripe_index >= last_stripe &&
5619                                     stripe_index <= (last_stripe +
5620                                                      sub_stripes - 1))
5621                                         bbio->stripes[i].length -=
5622                                                 stripe_end_offset;
5623 
5624                                 if (i == sub_stripes - 1)
5625                                         stripe_offset = 0;
5626                         } else
5627                                 bbio->stripes[i].length = *length;
5628 
5629                         stripe_index++;
5630                         if (stripe_index == map->num_stripes) {
5631                                 /* This could only happen for RAID0/10 */
5632                                 stripe_index = 0;
5633                                 stripe_nr++;
5634                         }
5635                 }
5636         } else {
5637                 for (i = 0; i < num_stripes; i++) {
5638                         bbio->stripes[i].physical =
5639                                 map->stripes[stripe_index].physical +
5640                                 stripe_offset +
5641                                 stripe_nr * map->stripe_len;
5642                         bbio->stripes[i].dev =
5643                                 map->stripes[stripe_index].dev;
5644                         stripe_index++;
5645                 }
5646         }
5647 
5648         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5649                 max_errors = btrfs_chunk_max_errors(map);
5650 
5651         if (bbio->raid_map)
5652                 sort_parity_stripes(bbio, num_stripes);
5653 
5654         tgtdev_indexes = 0;
5655         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5656             dev_replace->tgtdev != NULL) {
5657                 int index_where_to_add;
5658                 u64 srcdev_devid = dev_replace->srcdev->devid;
5659 
5660                 /*
5661                  * duplicate the write operations while the dev replace
5662                  * procedure is running. Since the copying of the old disk
5663                  * to the new disk takes place at run time while the
5664                  * filesystem is mounted writable, the regular write
5665                  * operations to the old disk have to be duplicated to go
5666                  * to the new disk as well.
5667                  * Note that device->missing is handled by the caller, and
5668                  * that the write to the old disk is already set up in the
5669                  * stripes array.
5670                  */
5671                 index_where_to_add = num_stripes;
5672                 for (i = 0; i < num_stripes; i++) {
5673                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5674                                 /* write to new disk, too */
5675                                 struct btrfs_bio_stripe *new =
5676                                         bbio->stripes + index_where_to_add;
5677                                 struct btrfs_bio_stripe *old =
5678                                         bbio->stripes + i;
5679 
5680                                 new->physical = old->physical;
5681                                 new->length = old->length;
5682                                 new->dev = dev_replace->tgtdev;
5683                                 bbio->tgtdev_map[i] = index_where_to_add;
5684                                 index_where_to_add++;
5685                                 max_errors++;
5686                                 tgtdev_indexes++;
5687                         }
5688                 }
5689                 num_stripes = index_where_to_add;
5690         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5691                    dev_replace->tgtdev != NULL) {
5692                 u64 srcdev_devid = dev_replace->srcdev->devid;
5693                 int index_srcdev = 0;
5694                 int found = 0;
5695                 u64 physical_of_found = 0;
5696 
5697                 /*
5698                  * During the dev-replace procedure, the target drive can
5699                  * also be used to read data in case it is needed to repair
5700                  * a corrupt block elsewhere. This is possible if the
5701                  * requested area is left of the left cursor. In this area,
5702                  * the target drive is a full copy of the source drive.
5703                  */
5704                 for (i = 0; i < num_stripes; i++) {
5705                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
5706                                 /*
5707                                  * In case of DUP, in order to keep it
5708                                  * simple, only add the mirror with the
5709                                  * lowest physical address
5710                                  */
5711                                 if (found &&
5712                                     physical_of_found <=
5713                                      bbio->stripes[i].physical)
5714                                         continue;
5715                                 index_srcdev = i;
5716                                 found = 1;
5717                                 physical_of_found = bbio->stripes[i].physical;
5718                         }
5719                 }
5720                 if (found) {
5721                         if (physical_of_found + map->stripe_len <=
5722                             dev_replace->cursor_left) {
5723                                 struct btrfs_bio_stripe *tgtdev_stripe =
5724                                         bbio->stripes + num_stripes;
5725 
5726                                 tgtdev_stripe->physical = physical_of_found;
5727                                 tgtdev_stripe->length =
5728                                         bbio->stripes[index_srcdev].length;
5729                                 tgtdev_stripe->dev = dev_replace->tgtdev;
5730                                 bbio->tgtdev_map[index_srcdev] = num_stripes;
5731 
5732                                 tgtdev_indexes++;
5733                                 num_stripes++;
5734                         }
5735                 }
5736         }
5737 
5738         *bbio_ret = bbio;
5739         bbio->map_type = map->type;
5740         bbio->num_stripes = num_stripes;
5741         bbio->max_errors = max_errors;
5742         bbio->mirror_num = mirror_num;
5743         bbio->num_tgtdevs = tgtdev_indexes;
5744 
5745         /*
5746          * this is the case that REQ_READ && dev_replace_is_ongoing &&
5747          * mirror_num == num_stripes + 1 && dev_replace target drive is
5748          * available as a mirror
5749          */
5750         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5751                 WARN_ON(num_stripes > 1);
5752                 bbio->stripes[0].dev = dev_replace->tgtdev;
5753                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5754                 bbio->mirror_num = map->num_stripes + 1;
5755         }
5756 out:
5757         if (dev_replace_is_ongoing) {
5758                 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5759                 btrfs_dev_replace_unlock(dev_replace, 0);
5760         }
5761         free_extent_map(em);
5762         return ret;
5763 }
5764 
5765 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5766                       u64 logical, u64 *length,
5767                       struct btrfs_bio **bbio_ret, int mirror_num)
5768 {
5769         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5770                                  mirror_num, 0);
5771 }
5772 
5773 /* For Scrub/replace */
5774 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5775                      u64 logical, u64 *length,
5776                      struct btrfs_bio **bbio_ret, int mirror_num,
5777                      int need_raid_map)
5778 {
5779         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5780                                  mirror_num, need_raid_map);
5781 }
5782 
5783 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5784                      u64 chunk_start, u64 physical, u64 devid,
5785                      u64 **logical, int *naddrs, int *stripe_len)
5786 {
5787         struct extent_map_tree *em_tree = &map_tree->map_tree;
5788         struct extent_map *em;
5789         struct map_lookup *map;
5790         u64 *buf;
5791         u64 bytenr;
5792         u64 length;
5793         u64 stripe_nr;
5794         u64 rmap_len;
5795         int i, j, nr = 0;
5796 
5797         read_lock(&em_tree->lock);
5798         em = lookup_extent_mapping(em_tree, chunk_start, 1);
5799         read_unlock(&em_tree->lock);
5800 
5801         if (!em) {
5802                 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5803                        chunk_start);
5804                 return -EIO;
5805         }
5806 
5807         if (em->start != chunk_start) {
5808                 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5809                        em->start, chunk_start);
5810                 free_extent_map(em);
5811                 return -EIO;
5812         }
5813         map = em->map_lookup;
5814 
5815         length = em->len;
5816         rmap_len = map->stripe_len;
5817 
5818         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5819                 length = div_u64(length, map->num_stripes / map->sub_stripes);
5820         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5821                 length = div_u64(length, map->num_stripes);
5822         else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5823                 length = div_u64(length, nr_data_stripes(map));
5824                 rmap_len = map->stripe_len * nr_data_stripes(map);
5825         }
5826 
5827         buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5828         BUG_ON(!buf); /* -ENOMEM */
5829 
5830         for (i = 0; i < map->num_stripes; i++) {
5831                 if (devid && map->stripes[i].dev->devid != devid)
5832                         continue;
5833                 if (map->stripes[i].physical > physical ||
5834                     map->stripes[i].physical + length <= physical)
5835                         continue;
5836 
5837                 stripe_nr = physical - map->stripes[i].physical;
5838                 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5839 
5840                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5841                         stripe_nr = stripe_nr * map->num_stripes + i;
5842                         stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5843                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5844                         stripe_nr = stripe_nr * map->num_stripes + i;
5845                 } /* else if RAID[56], multiply by nr_data_stripes().
5846                    * Alternatively, just use rmap_len below instead of
5847                    * map->stripe_len */
5848 
5849                 bytenr = chunk_start + stripe_nr * rmap_len;
5850                 WARN_ON(nr >= map->num_stripes);
5851                 for (j = 0; j < nr; j++) {
5852                         if (buf[j] == bytenr)
5853                                 break;
5854                 }
5855                 if (j == nr) {
5856                         WARN_ON(nr >= map->num_stripes);
5857                         buf[nr++] = bytenr;
5858                 }
5859         }
5860 
5861         *logical = buf;
5862         *naddrs = nr;
5863         *stripe_len = rmap_len;
5864 
5865         free_extent_map(em);
5866         return 0;
5867 }
5868 
5869 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5870 {
5871         bio->bi_private = bbio->private;
5872         bio->bi_end_io = bbio->end_io;
5873         bio_endio(bio);
5874 
5875         btrfs_put_bbio(bbio);
5876 }
5877 
5878 static void btrfs_end_bio(struct bio *bio)
5879 {
5880         struct btrfs_bio *bbio = bio->bi_private;
5881         int is_orig_bio = 0;
5882 
5883         if (bio->bi_error) {
5884                 atomic_inc(&bbio->error);
5885                 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5886                         unsigned int stripe_index =
5887                                 btrfs_io_bio(bio)->stripe_index;
5888                         struct btrfs_device *dev;
5889 
5890                         BUG_ON(stripe_index >= bbio->num_stripes);
5891                         dev = bbio->stripes[stripe_index].dev;
5892                         if (dev->bdev) {
5893                                 if (bio->bi_rw & WRITE)
5894                                         btrfs_dev_stat_inc(dev,
5895                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5896                                 else
5897                                         btrfs_dev_stat_inc(dev,
5898                                                 BTRFS_DEV_STAT_READ_ERRS);
5899                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5900                                         btrfs_dev_stat_inc(dev,
5901                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5902                                 btrfs_dev_stat_print_on_error(dev);
5903                         }
5904                 }
5905         }
5906 
5907         if (bio == bbio->orig_bio)
5908                 is_orig_bio = 1;
5909 
5910         btrfs_bio_counter_dec(bbio->fs_info);
5911 
5912         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5913                 if (!is_orig_bio) {
5914                         bio_put(bio);
5915                         bio = bbio->orig_bio;
5916                 }
5917 
5918                 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5919                 /* only send an error to the higher layers if it is
5920                  * beyond the tolerance of the btrfs bio
5921                  */
5922                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5923                         bio->bi_error = -EIO;
5924                 } else {
5925                         /*
5926                          * this bio is actually up to date, we didn't
5927                          * go over the max number of errors
5928                          */
5929                         bio->bi_error = 0;
5930                 }
5931 
5932                 btrfs_end_bbio(bbio, bio);
5933         } else if (!is_orig_bio) {
5934                 bio_put(bio);
5935         }
5936 }
5937 
5938 /*
5939  * see run_scheduled_bios for a description of why bios are collected for
5940  * async submit.
5941  *
5942  * This will add one bio to the pending list for a device and make sure
5943  * the work struct is scheduled.
5944  */
5945 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5946                                         struct btrfs_device *device,
5947                                         int rw, struct bio *bio)
5948 {
5949         int should_queue = 1;
5950         struct btrfs_pending_bios *pending_bios;
5951 
5952         if (device->missing || !device->bdev) {
5953                 bio_io_error(bio);
5954                 return;
5955         }
5956 
5957         /* don't bother with additional async steps for reads, right now */
5958         if (!(rw & REQ_WRITE)) {
5959                 bio_get(bio);
5960                 btrfsic_submit_bio(rw, bio);
5961                 bio_put(bio);
5962                 return;
5963         }
5964 
5965         /*
5966          * nr_async_bios allows us to reliably return congestion to the
5967          * higher layers.  Otherwise, the async bio makes it appear we have
5968          * made progress against dirty pages when we've really just put it
5969          * on a queue for later
5970          */
5971         atomic_inc(&root->fs_info->nr_async_bios);
5972         WARN_ON(bio->bi_next);
5973         bio->bi_next = NULL;
5974         bio->bi_rw |= rw;
5975 
5976         spin_lock(&device->io_lock);
5977         if (bio->bi_rw & REQ_SYNC)
5978                 pending_bios = &device->pending_sync_bios;
5979         else
5980                 pending_bios = &device->pending_bios;
5981 
5982         if (pending_bios->tail)
5983                 pending_bios->tail->bi_next = bio;
5984 
5985         pending_bios->tail = bio;
5986         if (!pending_bios->head)
5987                 pending_bios->head = bio;
5988         if (device->running_pending)
5989                 should_queue = 0;
5990 
5991         spin_unlock(&device->io_lock);
5992 
5993         if (should_queue)
5994                 btrfs_queue_work(root->fs_info->submit_workers,
5995                                  &device->work);
5996 }
5997 
5998 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5999                               struct bio *bio, u64 physical, int dev_nr,
6000                               int rw, int async)
6001 {
6002         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6003 
6004         bio->bi_private = bbio;
6005         btrfs_io_bio(bio)->stripe_index = dev_nr;
6006         bio->bi_end_io = btrfs_end_bio;
6007         bio->bi_iter.bi_sector = physical >> 9;
6008 #ifdef DEBUG
6009         {
6010                 struct rcu_string *name;
6011 
6012                 rcu_read_lock();
6013                 name = rcu_dereference(dev->name);
6014                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6015                          "(%s id %llu), size=%u\n", rw,
6016                          (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6017                          name->str, dev->devid, bio->bi_iter.bi_size);
6018                 rcu_read_unlock();
6019         }
6020 #endif
6021         bio->bi_bdev = dev->bdev;
6022 
6023         btrfs_bio_counter_inc_noblocked(root->fs_info);
6024 
6025         if (async)
6026                 btrfs_schedule_bio(root, dev, rw, bio);
6027         else
6028                 btrfsic_submit_bio(rw, bio);
6029 }
6030 
6031 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6032 {
6033         atomic_inc(&bbio->error);
6034         if (atomic_dec_and_test(&bbio->stripes_pending)) {
6035                 /* Shoud be the original bio. */
6036                 WARN_ON(bio != bbio->orig_bio);
6037 
<