46960983891333163eaddaadcbabccf5cd8f5e20
[linux-3.10.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include "compat.h"
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
38 #include "math.h"
39 #include "dev-replace.h"
40
41 static int init_first_rw_device(struct btrfs_trans_handle *trans,
42                                 struct btrfs_root *root,
43                                 struct btrfs_device *device);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
45 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
47
48 static DEFINE_MUTEX(uuid_mutex);
49 static LIST_HEAD(fs_uuids);
50
51 static void lock_chunks(struct btrfs_root *root)
52 {
53         mutex_lock(&root->fs_info->chunk_mutex);
54 }
55
56 static void unlock_chunks(struct btrfs_root *root)
57 {
58         mutex_unlock(&root->fs_info->chunk_mutex);
59 }
60
61 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62 {
63         struct btrfs_device *device;
64         WARN_ON(fs_devices->opened);
65         while (!list_empty(&fs_devices->devices)) {
66                 device = list_entry(fs_devices->devices.next,
67                                     struct btrfs_device, dev_list);
68                 list_del(&device->dev_list);
69                 rcu_string_free(device->name);
70                 kfree(device);
71         }
72         kfree(fs_devices);
73 }
74
75 static void btrfs_kobject_uevent(struct block_device *bdev,
76                                  enum kobject_action action)
77 {
78         int ret;
79
80         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
81         if (ret)
82                 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
83                         action,
84                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
85                         &disk_to_dev(bdev->bd_disk)->kobj);
86 }
87
88 void btrfs_cleanup_fs_uuids(void)
89 {
90         struct btrfs_fs_devices *fs_devices;
91
92         while (!list_empty(&fs_uuids)) {
93                 fs_devices = list_entry(fs_uuids.next,
94                                         struct btrfs_fs_devices, list);
95                 list_del(&fs_devices->list);
96                 free_fs_devices(fs_devices);
97         }
98 }
99
100 static noinline struct btrfs_device *__find_device(struct list_head *head,
101                                                    u64 devid, u8 *uuid)
102 {
103         struct btrfs_device *dev;
104
105         list_for_each_entry(dev, head, dev_list) {
106                 if (dev->devid == devid &&
107                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
108                         return dev;
109                 }
110         }
111         return NULL;
112 }
113
114 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 {
116         struct btrfs_fs_devices *fs_devices;
117
118         list_for_each_entry(fs_devices, &fs_uuids, list) {
119                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
120                         return fs_devices;
121         }
122         return NULL;
123 }
124
125 static int
126 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
127                       int flush, struct block_device **bdev,
128                       struct buffer_head **bh)
129 {
130         int ret;
131
132         *bdev = blkdev_get_by_path(device_path, flags, holder);
133
134         if (IS_ERR(*bdev)) {
135                 ret = PTR_ERR(*bdev);
136                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
137                 goto error;
138         }
139
140         if (flush)
141                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
142         ret = set_blocksize(*bdev, 4096);
143         if (ret) {
144                 blkdev_put(*bdev, flags);
145                 goto error;
146         }
147         invalidate_bdev(*bdev);
148         *bh = btrfs_read_dev_super(*bdev);
149         if (!*bh) {
150                 ret = -EINVAL;
151                 blkdev_put(*bdev, flags);
152                 goto error;
153         }
154
155         return 0;
156
157 error:
158         *bdev = NULL;
159         *bh = NULL;
160         return ret;
161 }
162
163 static void requeue_list(struct btrfs_pending_bios *pending_bios,
164                         struct bio *head, struct bio *tail)
165 {
166
167         struct bio *old_head;
168
169         old_head = pending_bios->head;
170         pending_bios->head = head;
171         if (pending_bios->tail)
172                 tail->bi_next = old_head;
173         else
174                 pending_bios->tail = tail;
175 }
176
177 /*
178  * we try to collect pending bios for a device so we don't get a large
179  * number of procs sending bios down to the same device.  This greatly
180  * improves the schedulers ability to collect and merge the bios.
181  *
182  * But, it also turns into a long list of bios to process and that is sure
183  * to eventually make the worker thread block.  The solution here is to
184  * make some progress and then put this work struct back at the end of
185  * the list if the block device is congested.  This way, multiple devices
186  * can make progress from a single worker thread.
187  */
188 static noinline void run_scheduled_bios(struct btrfs_device *device)
189 {
190         struct bio *pending;
191         struct backing_dev_info *bdi;
192         struct btrfs_fs_info *fs_info;
193         struct btrfs_pending_bios *pending_bios;
194         struct bio *tail;
195         struct bio *cur;
196         int again = 0;
197         unsigned long num_run;
198         unsigned long batch_run = 0;
199         unsigned long limit;
200         unsigned long last_waited = 0;
201         int force_reg = 0;
202         int sync_pending = 0;
203         struct blk_plug plug;
204
205         /*
206          * this function runs all the bios we've collected for
207          * a particular device.  We don't want to wander off to
208          * another device without first sending all of these down.
209          * So, setup a plug here and finish it off before we return
210          */
211         blk_start_plug(&plug);
212
213         bdi = blk_get_backing_dev_info(device->bdev);
214         fs_info = device->dev_root->fs_info;
215         limit = btrfs_async_submit_limit(fs_info);
216         limit = limit * 2 / 3;
217
218 loop:
219         spin_lock(&device->io_lock);
220
221 loop_lock:
222         num_run = 0;
223
224         /* take all the bios off the list at once and process them
225          * later on (without the lock held).  But, remember the
226          * tail and other pointers so the bios can be properly reinserted
227          * into the list if we hit congestion
228          */
229         if (!force_reg && device->pending_sync_bios.head) {
230                 pending_bios = &device->pending_sync_bios;
231                 force_reg = 1;
232         } else {
233                 pending_bios = &device->pending_bios;
234                 force_reg = 0;
235         }
236
237         pending = pending_bios->head;
238         tail = pending_bios->tail;
239         WARN_ON(pending && !tail);
240
241         /*
242          * if pending was null this time around, no bios need processing
243          * at all and we can stop.  Otherwise it'll loop back up again
244          * and do an additional check so no bios are missed.
245          *
246          * device->running_pending is used to synchronize with the
247          * schedule_bio code.
248          */
249         if (device->pending_sync_bios.head == NULL &&
250             device->pending_bios.head == NULL) {
251                 again = 0;
252                 device->running_pending = 0;
253         } else {
254                 again = 1;
255                 device->running_pending = 1;
256         }
257
258         pending_bios->head = NULL;
259         pending_bios->tail = NULL;
260
261         spin_unlock(&device->io_lock);
262
263         while (pending) {
264
265                 rmb();
266                 /* we want to work on both lists, but do more bios on the
267                  * sync list than the regular list
268                  */
269                 if ((num_run > 32 &&
270                     pending_bios != &device->pending_sync_bios &&
271                     device->pending_sync_bios.head) ||
272                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
273                     device->pending_bios.head)) {
274                         spin_lock(&device->io_lock);
275                         requeue_list(pending_bios, pending, tail);
276                         goto loop_lock;
277                 }
278
279                 cur = pending;
280                 pending = pending->bi_next;
281                 cur->bi_next = NULL;
282
283                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
284                     waitqueue_active(&fs_info->async_submit_wait))
285                         wake_up(&fs_info->async_submit_wait);
286
287                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
288
289                 /*
290                  * if we're doing the sync list, record that our
291                  * plug has some sync requests on it
292                  *
293                  * If we're doing the regular list and there are
294                  * sync requests sitting around, unplug before
295                  * we add more
296                  */
297                 if (pending_bios == &device->pending_sync_bios) {
298                         sync_pending = 1;
299                 } else if (sync_pending) {
300                         blk_finish_plug(&plug);
301                         blk_start_plug(&plug);
302                         sync_pending = 0;
303                 }
304
305                 btrfsic_submit_bio(cur->bi_rw, cur);
306                 num_run++;
307                 batch_run++;
308                 if (need_resched())
309                         cond_resched();
310
311                 /*
312                  * we made progress, there is more work to do and the bdi
313                  * is now congested.  Back off and let other work structs
314                  * run instead
315                  */
316                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
317                     fs_info->fs_devices->open_devices > 1) {
318                         struct io_context *ioc;
319
320                         ioc = current->io_context;
321
322                         /*
323                          * the main goal here is that we don't want to
324                          * block if we're going to be able to submit
325                          * more requests without blocking.
326                          *
327                          * This code does two great things, it pokes into
328                          * the elevator code from a filesystem _and_
329                          * it makes assumptions about how batching works.
330                          */
331                         if (ioc && ioc->nr_batch_requests > 0 &&
332                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
333                             (last_waited == 0 ||
334                              ioc->last_waited == last_waited)) {
335                                 /*
336                                  * we want to go through our batch of
337                                  * requests and stop.  So, we copy out
338                                  * the ioc->last_waited time and test
339                                  * against it before looping
340                                  */
341                                 last_waited = ioc->last_waited;
342                                 if (need_resched())
343                                         cond_resched();
344                                 continue;
345                         }
346                         spin_lock(&device->io_lock);
347                         requeue_list(pending_bios, pending, tail);
348                         device->running_pending = 1;
349
350                         spin_unlock(&device->io_lock);
351                         btrfs_requeue_work(&device->work);
352                         goto done;
353                 }
354                 /* unplug every 64 requests just for good measure */
355                 if (batch_run % 64 == 0) {
356                         blk_finish_plug(&plug);
357                         blk_start_plug(&plug);
358                         sync_pending = 0;
359                 }
360         }
361
362         cond_resched();
363         if (again)
364                 goto loop;
365
366         spin_lock(&device->io_lock);
367         if (device->pending_bios.head || device->pending_sync_bios.head)
368                 goto loop_lock;
369         spin_unlock(&device->io_lock);
370
371 done:
372         blk_finish_plug(&plug);
373 }
374
375 static void pending_bios_fn(struct btrfs_work *work)
376 {
377         struct btrfs_device *device;
378
379         device = container_of(work, struct btrfs_device, work);
380         run_scheduled_bios(device);
381 }
382
383 static noinline int device_list_add(const char *path,
384                            struct btrfs_super_block *disk_super,
385                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
386 {
387         struct btrfs_device *device;
388         struct btrfs_fs_devices *fs_devices;
389         struct rcu_string *name;
390         u64 found_transid = btrfs_super_generation(disk_super);
391
392         fs_devices = find_fsid(disk_super->fsid);
393         if (!fs_devices) {
394                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
395                 if (!fs_devices)
396                         return -ENOMEM;
397                 INIT_LIST_HEAD(&fs_devices->devices);
398                 INIT_LIST_HEAD(&fs_devices->alloc_list);
399                 list_add(&fs_devices->list, &fs_uuids);
400                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
401                 fs_devices->latest_devid = devid;
402                 fs_devices->latest_trans = found_transid;
403                 mutex_init(&fs_devices->device_list_mutex);
404                 device = NULL;
405         } else {
406                 device = __find_device(&fs_devices->devices, devid,
407                                        disk_super->dev_item.uuid);
408         }
409         if (!device) {
410                 if (fs_devices->opened)
411                         return -EBUSY;
412
413                 device = kzalloc(sizeof(*device), GFP_NOFS);
414                 if (!device) {
415                         /* we can safely leave the fs_devices entry around */
416                         return -ENOMEM;
417                 }
418                 device->devid = devid;
419                 device->dev_stats_valid = 0;
420                 device->work.func = pending_bios_fn;
421                 memcpy(device->uuid, disk_super->dev_item.uuid,
422                        BTRFS_UUID_SIZE);
423                 spin_lock_init(&device->io_lock);
424
425                 name = rcu_string_strdup(path, GFP_NOFS);
426                 if (!name) {
427                         kfree(device);
428                         return -ENOMEM;
429                 }
430                 rcu_assign_pointer(device->name, name);
431                 INIT_LIST_HEAD(&device->dev_alloc_list);
432
433                 /* init readahead state */
434                 spin_lock_init(&device->reada_lock);
435                 device->reada_curr_zone = NULL;
436                 atomic_set(&device->reada_in_flight, 0);
437                 device->reada_next = 0;
438                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
439                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
440
441                 mutex_lock(&fs_devices->device_list_mutex);
442                 list_add_rcu(&device->dev_list, &fs_devices->devices);
443                 mutex_unlock(&fs_devices->device_list_mutex);
444
445                 device->fs_devices = fs_devices;
446                 fs_devices->num_devices++;
447         } else if (!device->name || strcmp(device->name->str, path)) {
448                 name = rcu_string_strdup(path, GFP_NOFS);
449                 if (!name)
450                         return -ENOMEM;
451                 rcu_string_free(device->name);
452                 rcu_assign_pointer(device->name, name);
453                 if (device->missing) {
454                         fs_devices->missing_devices--;
455                         device->missing = 0;
456                 }
457         }
458
459         if (found_transid > fs_devices->latest_trans) {
460                 fs_devices->latest_devid = devid;
461                 fs_devices->latest_trans = found_transid;
462         }
463         *fs_devices_ret = fs_devices;
464         return 0;
465 }
466
467 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
468 {
469         struct btrfs_fs_devices *fs_devices;
470         struct btrfs_device *device;
471         struct btrfs_device *orig_dev;
472
473         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
474         if (!fs_devices)
475                 return ERR_PTR(-ENOMEM);
476
477         INIT_LIST_HEAD(&fs_devices->devices);
478         INIT_LIST_HEAD(&fs_devices->alloc_list);
479         INIT_LIST_HEAD(&fs_devices->list);
480         mutex_init(&fs_devices->device_list_mutex);
481         fs_devices->latest_devid = orig->latest_devid;
482         fs_devices->latest_trans = orig->latest_trans;
483         fs_devices->total_devices = orig->total_devices;
484         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
485
486         /* We have held the volume lock, it is safe to get the devices. */
487         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
488                 struct rcu_string *name;
489
490                 device = kzalloc(sizeof(*device), GFP_NOFS);
491                 if (!device)
492                         goto error;
493
494                 /*
495                  * This is ok to do without rcu read locked because we hold the
496                  * uuid mutex so nothing we touch in here is going to disappear.
497                  */
498                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
499                 if (!name) {
500                         kfree(device);
501                         goto error;
502                 }
503                 rcu_assign_pointer(device->name, name);
504
505                 device->devid = orig_dev->devid;
506                 device->work.func = pending_bios_fn;
507                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
508                 spin_lock_init(&device->io_lock);
509                 INIT_LIST_HEAD(&device->dev_list);
510                 INIT_LIST_HEAD(&device->dev_alloc_list);
511
512                 list_add(&device->dev_list, &fs_devices->devices);
513                 device->fs_devices = fs_devices;
514                 fs_devices->num_devices++;
515         }
516         return fs_devices;
517 error:
518         free_fs_devices(fs_devices);
519         return ERR_PTR(-ENOMEM);
520 }
521
522 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
523                                struct btrfs_fs_devices *fs_devices, int step)
524 {
525         struct btrfs_device *device, *next;
526
527         struct block_device *latest_bdev = NULL;
528         u64 latest_devid = 0;
529         u64 latest_transid = 0;
530
531         mutex_lock(&uuid_mutex);
532 again:
533         /* This is the initialized path, it is safe to release the devices. */
534         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
535                 if (device->in_fs_metadata) {
536                         if (!device->is_tgtdev_for_dev_replace &&
537                             (!latest_transid ||
538                              device->generation > latest_transid)) {
539                                 latest_devid = device->devid;
540                                 latest_transid = device->generation;
541                                 latest_bdev = device->bdev;
542                         }
543                         continue;
544                 }
545
546                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
547                         /*
548                          * In the first step, keep the device which has
549                          * the correct fsid and the devid that is used
550                          * for the dev_replace procedure.
551                          * In the second step, the dev_replace state is
552                          * read from the device tree and it is known
553                          * whether the procedure is really active or
554                          * not, which means whether this device is
555                          * used or whether it should be removed.
556                          */
557                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
558                                 continue;
559                         }
560                 }
561                 if (device->bdev) {
562                         blkdev_put(device->bdev, device->mode);
563                         device->bdev = NULL;
564                         fs_devices->open_devices--;
565                 }
566                 if (device->writeable) {
567                         list_del_init(&device->dev_alloc_list);
568                         device->writeable = 0;
569                         if (!device->is_tgtdev_for_dev_replace)
570                                 fs_devices->rw_devices--;
571                 }
572                 list_del_init(&device->dev_list);
573                 fs_devices->num_devices--;
574                 rcu_string_free(device->name);
575                 kfree(device);
576         }
577
578         if (fs_devices->seed) {
579                 fs_devices = fs_devices->seed;
580                 goto again;
581         }
582
583         fs_devices->latest_bdev = latest_bdev;
584         fs_devices->latest_devid = latest_devid;
585         fs_devices->latest_trans = latest_transid;
586
587         mutex_unlock(&uuid_mutex);
588 }
589
590 static void __free_device(struct work_struct *work)
591 {
592         struct btrfs_device *device;
593
594         device = container_of(work, struct btrfs_device, rcu_work);
595
596         if (device->bdev)
597                 blkdev_put(device->bdev, device->mode);
598
599         rcu_string_free(device->name);
600         kfree(device);
601 }
602
603 static void free_device(struct rcu_head *head)
604 {
605         struct btrfs_device *device;
606
607         device = container_of(head, struct btrfs_device, rcu);
608
609         INIT_WORK(&device->rcu_work, __free_device);
610         schedule_work(&device->rcu_work);
611 }
612
613 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
614 {
615         struct btrfs_device *device;
616
617         if (--fs_devices->opened > 0)
618                 return 0;
619
620         mutex_lock(&fs_devices->device_list_mutex);
621         list_for_each_entry(device, &fs_devices->devices, dev_list) {
622                 struct btrfs_device *new_device;
623                 struct rcu_string *name;
624
625                 if (device->bdev)
626                         fs_devices->open_devices--;
627
628                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
629                         list_del_init(&device->dev_alloc_list);
630                         fs_devices->rw_devices--;
631                 }
632
633                 if (device->can_discard)
634                         fs_devices->num_can_discard--;
635
636                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
637                 BUG_ON(!new_device); /* -ENOMEM */
638                 memcpy(new_device, device, sizeof(*new_device));
639
640                 /* Safe because we are under uuid_mutex */
641                 if (device->name) {
642                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
643                         BUG_ON(device->name && !name); /* -ENOMEM */
644                         rcu_assign_pointer(new_device->name, name);
645                 }
646                 new_device->bdev = NULL;
647                 new_device->writeable = 0;
648                 new_device->in_fs_metadata = 0;
649                 new_device->can_discard = 0;
650                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
651
652                 call_rcu(&device->rcu, free_device);
653         }
654         mutex_unlock(&fs_devices->device_list_mutex);
655
656         WARN_ON(fs_devices->open_devices);
657         WARN_ON(fs_devices->rw_devices);
658         fs_devices->opened = 0;
659         fs_devices->seeding = 0;
660
661         return 0;
662 }
663
664 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
665 {
666         struct btrfs_fs_devices *seed_devices = NULL;
667         int ret;
668
669         mutex_lock(&uuid_mutex);
670         ret = __btrfs_close_devices(fs_devices);
671         if (!fs_devices->opened) {
672                 seed_devices = fs_devices->seed;
673                 fs_devices->seed = NULL;
674         }
675         mutex_unlock(&uuid_mutex);
676
677         while (seed_devices) {
678                 fs_devices = seed_devices;
679                 seed_devices = fs_devices->seed;
680                 __btrfs_close_devices(fs_devices);
681                 free_fs_devices(fs_devices);
682         }
683         return ret;
684 }
685
686 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
687                                 fmode_t flags, void *holder)
688 {
689         struct request_queue *q;
690         struct block_device *bdev;
691         struct list_head *head = &fs_devices->devices;
692         struct btrfs_device *device;
693         struct block_device *latest_bdev = NULL;
694         struct buffer_head *bh;
695         struct btrfs_super_block *disk_super;
696         u64 latest_devid = 0;
697         u64 latest_transid = 0;
698         u64 devid;
699         int seeding = 1;
700         int ret = 0;
701
702         flags |= FMODE_EXCL;
703
704         list_for_each_entry(device, head, dev_list) {
705                 if (device->bdev)
706                         continue;
707                 if (!device->name)
708                         continue;
709
710                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
711                                             &bdev, &bh);
712                 if (ret)
713                         continue;
714
715                 disk_super = (struct btrfs_super_block *)bh->b_data;
716                 devid = btrfs_stack_device_id(&disk_super->dev_item);
717                 if (devid != device->devid)
718                         goto error_brelse;
719
720                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
721                            BTRFS_UUID_SIZE))
722                         goto error_brelse;
723
724                 device->generation = btrfs_super_generation(disk_super);
725                 if (!latest_transid || device->generation > latest_transid) {
726                         latest_devid = devid;
727                         latest_transid = device->generation;
728                         latest_bdev = bdev;
729                 }
730
731                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
732                         device->writeable = 0;
733                 } else {
734                         device->writeable = !bdev_read_only(bdev);
735                         seeding = 0;
736                 }
737
738                 q = bdev_get_queue(bdev);
739                 if (blk_queue_discard(q)) {
740                         device->can_discard = 1;
741                         fs_devices->num_can_discard++;
742                 }
743
744                 device->bdev = bdev;
745                 device->in_fs_metadata = 0;
746                 device->mode = flags;
747
748                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
749                         fs_devices->rotating = 1;
750
751                 fs_devices->open_devices++;
752                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
753                         fs_devices->rw_devices++;
754                         list_add(&device->dev_alloc_list,
755                                  &fs_devices->alloc_list);
756                 }
757                 brelse(bh);
758                 continue;
759
760 error_brelse:
761                 brelse(bh);
762                 blkdev_put(bdev, flags);
763                 continue;
764         }
765         if (fs_devices->open_devices == 0) {
766                 ret = -EINVAL;
767                 goto out;
768         }
769         fs_devices->seeding = seeding;
770         fs_devices->opened = 1;
771         fs_devices->latest_bdev = latest_bdev;
772         fs_devices->latest_devid = latest_devid;
773         fs_devices->latest_trans = latest_transid;
774         fs_devices->total_rw_bytes = 0;
775 out:
776         return ret;
777 }
778
779 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
780                        fmode_t flags, void *holder)
781 {
782         int ret;
783
784         mutex_lock(&uuid_mutex);
785         if (fs_devices->opened) {
786                 fs_devices->opened++;
787                 ret = 0;
788         } else {
789                 ret = __btrfs_open_devices(fs_devices, flags, holder);
790         }
791         mutex_unlock(&uuid_mutex);
792         return ret;
793 }
794
795 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
796                           struct btrfs_fs_devices **fs_devices_ret)
797 {
798         struct btrfs_super_block *disk_super;
799         struct block_device *bdev;
800         struct buffer_head *bh;
801         int ret;
802         u64 devid;
803         u64 transid;
804         u64 total_devices;
805
806         flags |= FMODE_EXCL;
807         mutex_lock(&uuid_mutex);
808         ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh);
809         if (ret)
810                 goto error;
811         disk_super = (struct btrfs_super_block *)bh->b_data;
812         devid = btrfs_stack_device_id(&disk_super->dev_item);
813         transid = btrfs_super_generation(disk_super);
814         total_devices = btrfs_super_num_devices(disk_super);
815         if (disk_super->label[0]) {
816                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
817                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
818                 printk(KERN_INFO "device label %s ", disk_super->label);
819         } else {
820                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
821         }
822         printk(KERN_CONT "devid %llu transid %llu %s\n",
823                (unsigned long long)devid, (unsigned long long)transid, path);
824         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
825         if (!ret && fs_devices_ret)
826                 (*fs_devices_ret)->total_devices = total_devices;
827         brelse(bh);
828         blkdev_put(bdev, flags);
829 error:
830         mutex_unlock(&uuid_mutex);
831         return ret;
832 }
833
834 /* helper to account the used device space in the range */
835 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
836                                    u64 end, u64 *length)
837 {
838         struct btrfs_key key;
839         struct btrfs_root *root = device->dev_root;
840         struct btrfs_dev_extent *dev_extent;
841         struct btrfs_path *path;
842         u64 extent_end;
843         int ret;
844         int slot;
845         struct extent_buffer *l;
846
847         *length = 0;
848
849         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
850                 return 0;
851
852         path = btrfs_alloc_path();
853         if (!path)
854                 return -ENOMEM;
855         path->reada = 2;
856
857         key.objectid = device->devid;
858         key.offset = start;
859         key.type = BTRFS_DEV_EXTENT_KEY;
860
861         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
862         if (ret < 0)
863                 goto out;
864         if (ret > 0) {
865                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
866                 if (ret < 0)
867                         goto out;
868         }
869
870         while (1) {
871                 l = path->nodes[0];
872                 slot = path->slots[0];
873                 if (slot >= btrfs_header_nritems(l)) {
874                         ret = btrfs_next_leaf(root, path);
875                         if (ret == 0)
876                                 continue;
877                         if (ret < 0)
878                                 goto out;
879
880                         break;
881                 }
882                 btrfs_item_key_to_cpu(l, &key, slot);
883
884                 if (key.objectid < device->devid)
885                         goto next;
886
887                 if (key.objectid > device->devid)
888                         break;
889
890                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
891                         goto next;
892
893                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
894                 extent_end = key.offset + btrfs_dev_extent_length(l,
895                                                                   dev_extent);
896                 if (key.offset <= start && extent_end > end) {
897                         *length = end - start + 1;
898                         break;
899                 } else if (key.offset <= start && extent_end > start)
900                         *length += extent_end - start;
901                 else if (key.offset > start && extent_end <= end)
902                         *length += extent_end - key.offset;
903                 else if (key.offset > start && key.offset <= end) {
904                         *length += end - key.offset + 1;
905                         break;
906                 } else if (key.offset > end)
907                         break;
908
909 next:
910                 path->slots[0]++;
911         }
912         ret = 0;
913 out:
914         btrfs_free_path(path);
915         return ret;
916 }
917
918 /*
919  * find_free_dev_extent - find free space in the specified device
920  * @device:     the device which we search the free space in
921  * @num_bytes:  the size of the free space that we need
922  * @start:      store the start of the free space.
923  * @len:        the size of the free space. that we find, or the size of the max
924  *              free space if we don't find suitable free space
925  *
926  * this uses a pretty simple search, the expectation is that it is
927  * called very infrequently and that a given device has a small number
928  * of extents
929  *
930  * @start is used to store the start of the free space if we find. But if we
931  * don't find suitable free space, it will be used to store the start position
932  * of the max free space.
933  *
934  * @len is used to store the size of the free space that we find.
935  * But if we don't find suitable free space, it is used to store the size of
936  * the max free space.
937  */
938 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
939                          u64 *start, u64 *len)
940 {
941         struct btrfs_key key;
942         struct btrfs_root *root = device->dev_root;
943         struct btrfs_dev_extent *dev_extent;
944         struct btrfs_path *path;
945         u64 hole_size;
946         u64 max_hole_start;
947         u64 max_hole_size;
948         u64 extent_end;
949         u64 search_start;
950         u64 search_end = device->total_bytes;
951         int ret;
952         int slot;
953         struct extent_buffer *l;
954
955         /* FIXME use last free of some kind */
956
957         /* we don't want to overwrite the superblock on the drive,
958          * so we make sure to start at an offset of at least 1MB
959          */
960         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
961
962         max_hole_start = search_start;
963         max_hole_size = 0;
964         hole_size = 0;
965
966         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
967                 ret = -ENOSPC;
968                 goto error;
969         }
970
971         path = btrfs_alloc_path();
972         if (!path) {
973                 ret = -ENOMEM;
974                 goto error;
975         }
976         path->reada = 2;
977
978         key.objectid = device->devid;
979         key.offset = search_start;
980         key.type = BTRFS_DEV_EXTENT_KEY;
981
982         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
983         if (ret < 0)
984                 goto out;
985         if (ret > 0) {
986                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
987                 if (ret < 0)
988                         goto out;
989         }
990
991         while (1) {
992                 l = path->nodes[0];
993                 slot = path->slots[0];
994                 if (slot >= btrfs_header_nritems(l)) {
995                         ret = btrfs_next_leaf(root, path);
996                         if (ret == 0)
997                                 continue;
998                         if (ret < 0)
999                                 goto out;
1000
1001                         break;
1002                 }
1003                 btrfs_item_key_to_cpu(l, &key, slot);
1004
1005                 if (key.objectid < device->devid)
1006                         goto next;
1007
1008                 if (key.objectid > device->devid)
1009                         break;
1010
1011                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1012                         goto next;
1013
1014                 if (key.offset > search_start) {
1015                         hole_size = key.offset - search_start;
1016
1017                         if (hole_size > max_hole_size) {
1018                                 max_hole_start = search_start;
1019                                 max_hole_size = hole_size;
1020                         }
1021
1022                         /*
1023                          * If this free space is greater than which we need,
1024                          * it must be the max free space that we have found
1025                          * until now, so max_hole_start must point to the start
1026                          * of this free space and the length of this free space
1027                          * is stored in max_hole_size. Thus, we return
1028                          * max_hole_start and max_hole_size and go back to the
1029                          * caller.
1030                          */
1031                         if (hole_size >= num_bytes) {
1032                                 ret = 0;
1033                                 goto out;
1034                         }
1035                 }
1036
1037                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1038                 extent_end = key.offset + btrfs_dev_extent_length(l,
1039                                                                   dev_extent);
1040                 if (extent_end > search_start)
1041                         search_start = extent_end;
1042 next:
1043                 path->slots[0]++;
1044                 cond_resched();
1045         }
1046
1047         /*
1048          * At this point, search_start should be the end of
1049          * allocated dev extents, and when shrinking the device,
1050          * search_end may be smaller than search_start.
1051          */
1052         if (search_end > search_start)
1053                 hole_size = search_end - search_start;
1054
1055         if (hole_size > max_hole_size) {
1056                 max_hole_start = search_start;
1057                 max_hole_size = hole_size;
1058         }
1059
1060         /* See above. */
1061         if (hole_size < num_bytes)
1062                 ret = -ENOSPC;
1063         else
1064                 ret = 0;
1065
1066 out:
1067         btrfs_free_path(path);
1068 error:
1069         *start = max_hole_start;
1070         if (len)
1071                 *len = max_hole_size;
1072         return ret;
1073 }
1074
1075 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1076                           struct btrfs_device *device,
1077                           u64 start)
1078 {
1079         int ret;
1080         struct btrfs_path *path;
1081         struct btrfs_root *root = device->dev_root;
1082         struct btrfs_key key;
1083         struct btrfs_key found_key;
1084         struct extent_buffer *leaf = NULL;
1085         struct btrfs_dev_extent *extent = NULL;
1086
1087         path = btrfs_alloc_path();
1088         if (!path)
1089                 return -ENOMEM;
1090
1091         key.objectid = device->devid;
1092         key.offset = start;
1093         key.type = BTRFS_DEV_EXTENT_KEY;
1094 again:
1095         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1096         if (ret > 0) {
1097                 ret = btrfs_previous_item(root, path, key.objectid,
1098                                           BTRFS_DEV_EXTENT_KEY);
1099                 if (ret)
1100                         goto out;
1101                 leaf = path->nodes[0];
1102                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1103                 extent = btrfs_item_ptr(leaf, path->slots[0],
1104                                         struct btrfs_dev_extent);
1105                 BUG_ON(found_key.offset > start || found_key.offset +
1106                        btrfs_dev_extent_length(leaf, extent) < start);
1107                 key = found_key;
1108                 btrfs_release_path(path);
1109                 goto again;
1110         } else if (ret == 0) {
1111                 leaf = path->nodes[0];
1112                 extent = btrfs_item_ptr(leaf, path->slots[0],
1113                                         struct btrfs_dev_extent);
1114         } else {
1115                 btrfs_error(root->fs_info, ret, "Slot search failed");
1116                 goto out;
1117         }
1118
1119         if (device->bytes_used > 0) {
1120                 u64 len = btrfs_dev_extent_length(leaf, extent);
1121                 device->bytes_used -= len;
1122                 spin_lock(&root->fs_info->free_chunk_lock);
1123                 root->fs_info->free_chunk_space += len;
1124                 spin_unlock(&root->fs_info->free_chunk_lock);
1125         }
1126         ret = btrfs_del_item(trans, root, path);
1127         if (ret) {
1128                 btrfs_error(root->fs_info, ret,
1129                             "Failed to remove dev extent item");
1130         }
1131 out:
1132         btrfs_free_path(path);
1133         return ret;
1134 }
1135
1136 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1137                            struct btrfs_device *device,
1138                            u64 chunk_tree, u64 chunk_objectid,
1139                            u64 chunk_offset, u64 start, u64 num_bytes)
1140 {
1141         int ret;
1142         struct btrfs_path *path;
1143         struct btrfs_root *root = device->dev_root;
1144         struct btrfs_dev_extent *extent;
1145         struct extent_buffer *leaf;
1146         struct btrfs_key key;
1147
1148         WARN_ON(!device->in_fs_metadata);
1149         WARN_ON(device->is_tgtdev_for_dev_replace);
1150         path = btrfs_alloc_path();
1151         if (!path)
1152                 return -ENOMEM;
1153
1154         key.objectid = device->devid;
1155         key.offset = start;
1156         key.type = BTRFS_DEV_EXTENT_KEY;
1157         ret = btrfs_insert_empty_item(trans, root, path, &key,
1158                                       sizeof(*extent));
1159         if (ret)
1160                 goto out;
1161
1162         leaf = path->nodes[0];
1163         extent = btrfs_item_ptr(leaf, path->slots[0],
1164                                 struct btrfs_dev_extent);
1165         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1166         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1167         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1168
1169         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1170                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1171                     BTRFS_UUID_SIZE);
1172
1173         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1174         btrfs_mark_buffer_dirty(leaf);
1175 out:
1176         btrfs_free_path(path);
1177         return ret;
1178 }
1179
1180 static noinline int find_next_chunk(struct btrfs_root *root,
1181                                     u64 objectid, u64 *offset)
1182 {
1183         struct btrfs_path *path;
1184         int ret;
1185         struct btrfs_key key;
1186         struct btrfs_chunk *chunk;
1187         struct btrfs_key found_key;
1188
1189         path = btrfs_alloc_path();
1190         if (!path)
1191                 return -ENOMEM;
1192
1193         key.objectid = objectid;
1194         key.offset = (u64)-1;
1195         key.type = BTRFS_CHUNK_ITEM_KEY;
1196
1197         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1198         if (ret < 0)
1199                 goto error;
1200
1201         BUG_ON(ret == 0); /* Corruption */
1202
1203         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1204         if (ret) {
1205                 *offset = 0;
1206         } else {
1207                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1208                                       path->slots[0]);
1209                 if (found_key.objectid != objectid)
1210                         *offset = 0;
1211                 else {
1212                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1213                                                struct btrfs_chunk);
1214                         *offset = found_key.offset +
1215                                 btrfs_chunk_length(path->nodes[0], chunk);
1216                 }
1217         }
1218         ret = 0;
1219 error:
1220         btrfs_free_path(path);
1221         return ret;
1222 }
1223
1224 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1225 {
1226         int ret;
1227         struct btrfs_key key;
1228         struct btrfs_key found_key;
1229         struct btrfs_path *path;
1230
1231         root = root->fs_info->chunk_root;
1232
1233         path = btrfs_alloc_path();
1234         if (!path)
1235                 return -ENOMEM;
1236
1237         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1238         key.type = BTRFS_DEV_ITEM_KEY;
1239         key.offset = (u64)-1;
1240
1241         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1242         if (ret < 0)
1243                 goto error;
1244
1245         BUG_ON(ret == 0); /* Corruption */
1246
1247         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1248                                   BTRFS_DEV_ITEM_KEY);
1249         if (ret) {
1250                 *objectid = 1;
1251         } else {
1252                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1253                                       path->slots[0]);
1254                 *objectid = found_key.offset + 1;
1255         }
1256         ret = 0;
1257 error:
1258         btrfs_free_path(path);
1259         return ret;
1260 }
1261
1262 /*
1263  * the device information is stored in the chunk root
1264  * the btrfs_device struct should be fully filled in
1265  */
1266 int btrfs_add_device(struct btrfs_trans_handle *trans,
1267                      struct btrfs_root *root,
1268                      struct btrfs_device *device)
1269 {
1270         int ret;
1271         struct btrfs_path *path;
1272         struct btrfs_dev_item *dev_item;
1273         struct extent_buffer *leaf;
1274         struct btrfs_key key;
1275         unsigned long ptr;
1276
1277         root = root->fs_info->chunk_root;
1278
1279         path = btrfs_alloc_path();
1280         if (!path)
1281                 return -ENOMEM;
1282
1283         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1284         key.type = BTRFS_DEV_ITEM_KEY;
1285         key.offset = device->devid;
1286
1287         ret = btrfs_insert_empty_item(trans, root, path, &key,
1288                                       sizeof(*dev_item));
1289         if (ret)
1290                 goto out;
1291
1292         leaf = path->nodes[0];
1293         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1294
1295         btrfs_set_device_id(leaf, dev_item, device->devid);
1296         btrfs_set_device_generation(leaf, dev_item, 0);
1297         btrfs_set_device_type(leaf, dev_item, device->type);
1298         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1299         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1300         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1301         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1302         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1303         btrfs_set_device_group(leaf, dev_item, 0);
1304         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1305         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1306         btrfs_set_device_start_offset(leaf, dev_item, 0);
1307
1308         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1309         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1310         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1311         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1312         btrfs_mark_buffer_dirty(leaf);
1313
1314         ret = 0;
1315 out:
1316         btrfs_free_path(path);
1317         return ret;
1318 }
1319
1320 static int btrfs_rm_dev_item(struct btrfs_root *root,
1321                              struct btrfs_device *device)
1322 {
1323         int ret;
1324         struct btrfs_path *path;
1325         struct btrfs_key key;
1326         struct btrfs_trans_handle *trans;
1327
1328         root = root->fs_info->chunk_root;
1329
1330         path = btrfs_alloc_path();
1331         if (!path)
1332                 return -ENOMEM;
1333
1334         trans = btrfs_start_transaction(root, 0);
1335         if (IS_ERR(trans)) {
1336                 btrfs_free_path(path);
1337                 return PTR_ERR(trans);
1338         }
1339         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1340         key.type = BTRFS_DEV_ITEM_KEY;
1341         key.offset = device->devid;
1342         lock_chunks(root);
1343
1344         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1345         if (ret < 0)
1346                 goto out;
1347
1348         if (ret > 0) {
1349                 ret = -ENOENT;
1350                 goto out;
1351         }
1352
1353         ret = btrfs_del_item(trans, root, path);
1354         if (ret)
1355                 goto out;
1356 out:
1357         btrfs_free_path(path);
1358         unlock_chunks(root);
1359         btrfs_commit_transaction(trans, root);
1360         return ret;
1361 }
1362
1363 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1364 {
1365         struct btrfs_device *device;
1366         struct btrfs_device *next_device;
1367         struct block_device *bdev;
1368         struct buffer_head *bh = NULL;
1369         struct btrfs_super_block *disk_super;
1370         struct btrfs_fs_devices *cur_devices;
1371         u64 all_avail;
1372         u64 devid;
1373         u64 num_devices;
1374         u8 *dev_uuid;
1375         int ret = 0;
1376         bool clear_super = false;
1377
1378         mutex_lock(&uuid_mutex);
1379
1380         all_avail = root->fs_info->avail_data_alloc_bits |
1381                 root->fs_info->avail_system_alloc_bits |
1382                 root->fs_info->avail_metadata_alloc_bits;
1383
1384         num_devices = root->fs_info->fs_devices->num_devices;
1385         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1386         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1387                 WARN_ON(num_devices < 1);
1388                 num_devices--;
1389         }
1390         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1391
1392         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1393                 printk(KERN_ERR "btrfs: unable to go below four devices "
1394                        "on raid10\n");
1395                 ret = -EINVAL;
1396                 goto out;
1397         }
1398
1399         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1400                 printk(KERN_ERR "btrfs: unable to go below two "
1401                        "devices on raid1\n");
1402                 ret = -EINVAL;
1403                 goto out;
1404         }
1405
1406         if (strcmp(device_path, "missing") == 0) {
1407                 struct list_head *devices;
1408                 struct btrfs_device *tmp;
1409
1410                 device = NULL;
1411                 devices = &root->fs_info->fs_devices->devices;
1412                 /*
1413                  * It is safe to read the devices since the volume_mutex
1414                  * is held.
1415                  */
1416                 list_for_each_entry(tmp, devices, dev_list) {
1417                         if (tmp->in_fs_metadata &&
1418                             !tmp->is_tgtdev_for_dev_replace &&
1419                             !tmp->bdev) {
1420                                 device = tmp;
1421                                 break;
1422                         }
1423                 }
1424                 bdev = NULL;
1425                 bh = NULL;
1426                 disk_super = NULL;
1427                 if (!device) {
1428                         printk(KERN_ERR "btrfs: no missing devices found to "
1429                                "remove\n");
1430                         goto out;
1431                 }
1432         } else {
1433                 ret = btrfs_get_bdev_and_sb(device_path,
1434                                             FMODE_WRITE | FMODE_EXCL,
1435                                             root->fs_info->bdev_holder, 0,
1436                                             &bdev, &bh);
1437                 if (ret)
1438                         goto out;
1439                 disk_super = (struct btrfs_super_block *)bh->b_data;
1440                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1441                 dev_uuid = disk_super->dev_item.uuid;
1442                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1443                                            disk_super->fsid);
1444                 if (!device) {
1445                         ret = -ENOENT;
1446                         goto error_brelse;
1447                 }
1448         }
1449
1450         if (device->is_tgtdev_for_dev_replace) {
1451                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1452                 ret = -EINVAL;
1453                 goto error_brelse;
1454         }
1455
1456         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1457                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1458                        "device\n");
1459                 ret = -EINVAL;
1460                 goto error_brelse;
1461         }
1462
1463         if (device->writeable) {
1464                 lock_chunks(root);
1465                 list_del_init(&device->dev_alloc_list);
1466                 unlock_chunks(root);
1467                 root->fs_info->fs_devices->rw_devices--;
1468                 clear_super = true;
1469         }
1470
1471         ret = btrfs_shrink_device(device, 0);
1472         if (ret)
1473                 goto error_undo;
1474
1475         /*
1476          * TODO: the superblock still includes this device in its num_devices
1477          * counter although write_all_supers() is not locked out. This
1478          * could give a filesystem state which requires a degraded mount.
1479          */
1480         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1481         if (ret)
1482                 goto error_undo;
1483
1484         spin_lock(&root->fs_info->free_chunk_lock);
1485         root->fs_info->free_chunk_space = device->total_bytes -
1486                 device->bytes_used;
1487         spin_unlock(&root->fs_info->free_chunk_lock);
1488
1489         device->in_fs_metadata = 0;
1490         btrfs_scrub_cancel_dev(root->fs_info, device);
1491
1492         /*
1493          * the device list mutex makes sure that we don't change
1494          * the device list while someone else is writing out all
1495          * the device supers.
1496          */
1497
1498         cur_devices = device->fs_devices;
1499         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1500         list_del_rcu(&device->dev_list);
1501
1502         device->fs_devices->num_devices--;
1503         device->fs_devices->total_devices--;
1504
1505         if (device->missing)
1506                 root->fs_info->fs_devices->missing_devices--;
1507
1508         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1509                                  struct btrfs_device, dev_list);
1510         if (device->bdev == root->fs_info->sb->s_bdev)
1511                 root->fs_info->sb->s_bdev = next_device->bdev;
1512         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1513                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1514
1515         if (device->bdev)
1516                 device->fs_devices->open_devices--;
1517
1518         call_rcu(&device->rcu, free_device);
1519         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1520
1521         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1522         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1523
1524         if (cur_devices->open_devices == 0) {
1525                 struct btrfs_fs_devices *fs_devices;
1526                 fs_devices = root->fs_info->fs_devices;
1527                 while (fs_devices) {
1528                         if (fs_devices->seed == cur_devices)
1529                                 break;
1530                         fs_devices = fs_devices->seed;
1531                 }
1532                 fs_devices->seed = cur_devices->seed;
1533                 cur_devices->seed = NULL;
1534                 lock_chunks(root);
1535                 __btrfs_close_devices(cur_devices);
1536                 unlock_chunks(root);
1537                 free_fs_devices(cur_devices);
1538         }
1539
1540         root->fs_info->num_tolerated_disk_barrier_failures =
1541                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1542
1543         /*
1544          * at this point, the device is zero sized.  We want to
1545          * remove it from the devices list and zero out the old super
1546          */
1547         if (clear_super && disk_super) {
1548                 /* make sure this device isn't detected as part of
1549                  * the FS anymore
1550                  */
1551                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1552                 set_buffer_dirty(bh);
1553                 sync_dirty_buffer(bh);
1554         }
1555
1556         ret = 0;
1557
1558         /* Notify udev that device has changed */
1559         btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1560
1561 error_brelse:
1562         brelse(bh);
1563         if (bdev)
1564                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1565 out:
1566         mutex_unlock(&uuid_mutex);
1567         return ret;
1568 error_undo:
1569         if (device->writeable) {
1570                 lock_chunks(root);
1571                 list_add(&device->dev_alloc_list,
1572                          &root->fs_info->fs_devices->alloc_list);
1573                 unlock_chunks(root);
1574                 root->fs_info->fs_devices->rw_devices++;
1575         }
1576         goto error_brelse;
1577 }
1578
1579 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1580                                  struct btrfs_device *srcdev)
1581 {
1582         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1583         list_del_rcu(&srcdev->dev_list);
1584         list_del_rcu(&srcdev->dev_alloc_list);
1585         fs_info->fs_devices->num_devices--;
1586         if (srcdev->missing) {
1587                 fs_info->fs_devices->missing_devices--;
1588                 fs_info->fs_devices->rw_devices++;
1589         }
1590         if (srcdev->can_discard)
1591                 fs_info->fs_devices->num_can_discard--;
1592         if (srcdev->bdev)
1593                 fs_info->fs_devices->open_devices--;
1594
1595         call_rcu(&srcdev->rcu, free_device);
1596 }
1597
1598 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1599                                       struct btrfs_device *tgtdev)
1600 {
1601         struct btrfs_device *next_device;
1602
1603         WARN_ON(!tgtdev);
1604         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1605         if (tgtdev->bdev) {
1606                 btrfs_scratch_superblock(tgtdev);
1607                 fs_info->fs_devices->open_devices--;
1608         }
1609         fs_info->fs_devices->num_devices--;
1610         if (tgtdev->can_discard)
1611                 fs_info->fs_devices->num_can_discard++;
1612
1613         next_device = list_entry(fs_info->fs_devices->devices.next,
1614                                  struct btrfs_device, dev_list);
1615         if (tgtdev->bdev == fs_info->sb->s_bdev)
1616                 fs_info->sb->s_bdev = next_device->bdev;
1617         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1618                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1619         list_del_rcu(&tgtdev->dev_list);
1620
1621         call_rcu(&tgtdev->rcu, free_device);
1622
1623         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1624 }
1625
1626 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1627                               struct btrfs_device **device)
1628 {
1629         int ret = 0;
1630         struct btrfs_super_block *disk_super;
1631         u64 devid;
1632         u8 *dev_uuid;
1633         struct block_device *bdev;
1634         struct buffer_head *bh;
1635
1636         *device = NULL;
1637         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1638                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1639         if (ret)
1640                 return ret;
1641         disk_super = (struct btrfs_super_block *)bh->b_data;
1642         devid = btrfs_stack_device_id(&disk_super->dev_item);
1643         dev_uuid = disk_super->dev_item.uuid;
1644         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1645                                     disk_super->fsid);
1646         brelse(bh);
1647         if (!*device)
1648                 ret = -ENOENT;
1649         blkdev_put(bdev, FMODE_READ);
1650         return ret;
1651 }
1652
1653 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1654                                          char *device_path,
1655                                          struct btrfs_device **device)
1656 {
1657         *device = NULL;
1658         if (strcmp(device_path, "missing") == 0) {
1659                 struct list_head *devices;
1660                 struct btrfs_device *tmp;
1661
1662                 devices = &root->fs_info->fs_devices->devices;
1663                 /*
1664                  * It is safe to read the devices since the volume_mutex
1665                  * is held by the caller.
1666                  */
1667                 list_for_each_entry(tmp, devices, dev_list) {
1668                         if (tmp->in_fs_metadata && !tmp->bdev) {
1669                                 *device = tmp;
1670                                 break;
1671                         }
1672                 }
1673
1674                 if (!*device) {
1675                         pr_err("btrfs: no missing device found\n");
1676                         return -ENOENT;
1677                 }
1678
1679                 return 0;
1680         } else {
1681                 return btrfs_find_device_by_path(root, device_path, device);
1682         }
1683 }
1684
1685 /*
1686  * does all the dirty work required for changing file system's UUID.
1687  */
1688 static int btrfs_prepare_sprout(struct btrfs_root *root)
1689 {
1690         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1691         struct btrfs_fs_devices *old_devices;
1692         struct btrfs_fs_devices *seed_devices;
1693         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1694         struct btrfs_device *device;
1695         u64 super_flags;
1696
1697         BUG_ON(!mutex_is_locked(&uuid_mutex));
1698         if (!fs_devices->seeding)
1699                 return -EINVAL;
1700
1701         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1702         if (!seed_devices)
1703                 return -ENOMEM;
1704
1705         old_devices = clone_fs_devices(fs_devices);
1706         if (IS_ERR(old_devices)) {
1707                 kfree(seed_devices);
1708                 return PTR_ERR(old_devices);
1709         }
1710
1711         list_add(&old_devices->list, &fs_uuids);
1712
1713         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1714         seed_devices->opened = 1;
1715         INIT_LIST_HEAD(&seed_devices->devices);
1716         INIT_LIST_HEAD(&seed_devices->alloc_list);
1717         mutex_init(&seed_devices->device_list_mutex);
1718
1719         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1720         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1721                               synchronize_rcu);
1722         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1723
1724         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1725         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1726                 device->fs_devices = seed_devices;
1727         }
1728
1729         fs_devices->seeding = 0;
1730         fs_devices->num_devices = 0;
1731         fs_devices->open_devices = 0;
1732         fs_devices->total_devices = 0;
1733         fs_devices->seed = seed_devices;
1734
1735         generate_random_uuid(fs_devices->fsid);
1736         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1737         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1738         super_flags = btrfs_super_flags(disk_super) &
1739                       ~BTRFS_SUPER_FLAG_SEEDING;
1740         btrfs_set_super_flags(disk_super, super_flags);
1741
1742         return 0;
1743 }
1744
1745 /*
1746  * strore the expected generation for seed devices in device items.
1747  */
1748 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1749                                struct btrfs_root *root)
1750 {
1751         struct btrfs_path *path;
1752         struct extent_buffer *leaf;
1753         struct btrfs_dev_item *dev_item;
1754         struct btrfs_device *device;
1755         struct btrfs_key key;
1756         u8 fs_uuid[BTRFS_UUID_SIZE];
1757         u8 dev_uuid[BTRFS_UUID_SIZE];
1758         u64 devid;
1759         int ret;
1760
1761         path = btrfs_alloc_path();
1762         if (!path)
1763                 return -ENOMEM;
1764
1765         root = root->fs_info->chunk_root;
1766         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1767         key.offset = 0;
1768         key.type = BTRFS_DEV_ITEM_KEY;
1769
1770         while (1) {
1771                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1772                 if (ret < 0)
1773                         goto error;
1774
1775                 leaf = path->nodes[0];
1776 next_slot:
1777                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1778                         ret = btrfs_next_leaf(root, path);
1779                         if (ret > 0)
1780                                 break;
1781                         if (ret < 0)
1782                                 goto error;
1783                         leaf = path->nodes[0];
1784                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1785                         btrfs_release_path(path);
1786                         continue;
1787                 }
1788
1789                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1790                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1791                     key.type != BTRFS_DEV_ITEM_KEY)
1792                         break;
1793
1794                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1795                                           struct btrfs_dev_item);
1796                 devid = btrfs_device_id(leaf, dev_item);
1797                 read_extent_buffer(leaf, dev_uuid,
1798                                    (unsigned long)btrfs_device_uuid(dev_item),
1799                                    BTRFS_UUID_SIZE);
1800                 read_extent_buffer(leaf, fs_uuid,
1801                                    (unsigned long)btrfs_device_fsid(dev_item),
1802                                    BTRFS_UUID_SIZE);
1803                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1804                                            fs_uuid);
1805                 BUG_ON(!device); /* Logic error */
1806
1807                 if (device->fs_devices->seeding) {
1808                         btrfs_set_device_generation(leaf, dev_item,
1809                                                     device->generation);
1810                         btrfs_mark_buffer_dirty(leaf);
1811                 }
1812
1813                 path->slots[0]++;
1814                 goto next_slot;
1815         }
1816         ret = 0;
1817 error:
1818         btrfs_free_path(path);
1819         return ret;
1820 }
1821
1822 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1823 {
1824         struct request_queue *q;
1825         struct btrfs_trans_handle *trans;
1826         struct btrfs_device *device;
1827         struct block_device *bdev;
1828         struct list_head *devices;
1829         struct super_block *sb = root->fs_info->sb;
1830         struct rcu_string *name;
1831         u64 total_bytes;
1832         int seeding_dev = 0;
1833         int ret = 0;
1834
1835         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1836                 return -EROFS;
1837
1838         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1839                                   root->fs_info->bdev_holder);
1840         if (IS_ERR(bdev))
1841                 return PTR_ERR(bdev);
1842
1843         if (root->fs_info->fs_devices->seeding) {
1844                 seeding_dev = 1;
1845                 down_write(&sb->s_umount);
1846                 mutex_lock(&uuid_mutex);
1847         }
1848
1849         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1850
1851         devices = &root->fs_info->fs_devices->devices;
1852
1853         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1854         list_for_each_entry(device, devices, dev_list) {
1855                 if (device->bdev == bdev) {
1856                         ret = -EEXIST;
1857                         mutex_unlock(
1858                                 &root->fs_info->fs_devices->device_list_mutex);
1859                         goto error;
1860                 }
1861         }
1862         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1863
1864         device = kzalloc(sizeof(*device), GFP_NOFS);
1865         if (!device) {
1866                 /* we can safely leave the fs_devices entry around */
1867                 ret = -ENOMEM;
1868                 goto error;
1869         }
1870
1871         name = rcu_string_strdup(device_path, GFP_NOFS);
1872         if (!name) {
1873                 kfree(device);
1874                 ret = -ENOMEM;
1875                 goto error;
1876         }
1877         rcu_assign_pointer(device->name, name);
1878
1879         ret = find_next_devid(root, &device->devid);
1880         if (ret) {
1881                 rcu_string_free(device->name);
1882                 kfree(device);
1883                 goto error;
1884         }
1885
1886         trans = btrfs_start_transaction(root, 0);
1887         if (IS_ERR(trans)) {
1888                 rcu_string_free(device->name);
1889                 kfree(device);
1890                 ret = PTR_ERR(trans);
1891                 goto error;
1892         }
1893
1894         lock_chunks(root);
1895
1896         q = bdev_get_queue(bdev);
1897         if (blk_queue_discard(q))
1898                 device->can_discard = 1;
1899         device->writeable = 1;
1900         device->work.func = pending_bios_fn;
1901         generate_random_uuid(device->uuid);
1902         spin_lock_init(&device->io_lock);
1903         device->generation = trans->transid;
1904         device->io_width = root->sectorsize;
1905         device->io_align = root->sectorsize;
1906         device->sector_size = root->sectorsize;
1907         device->total_bytes = i_size_read(bdev->bd_inode);
1908         device->disk_total_bytes = device->total_bytes;
1909         device->dev_root = root->fs_info->dev_root;
1910         device->bdev = bdev;
1911         device->in_fs_metadata = 1;
1912         device->is_tgtdev_for_dev_replace = 0;
1913         device->mode = FMODE_EXCL;
1914         set_blocksize(device->bdev, 4096);
1915
1916         if (seeding_dev) {
1917                 sb->s_flags &= ~MS_RDONLY;
1918                 ret = btrfs_prepare_sprout(root);
1919                 BUG_ON(ret); /* -ENOMEM */
1920         }
1921
1922         device->fs_devices = root->fs_info->fs_devices;
1923
1924         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1925         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1926         list_add(&device->dev_alloc_list,
1927                  &root->fs_info->fs_devices->alloc_list);
1928         root->fs_info->fs_devices->num_devices++;
1929         root->fs_info->fs_devices->open_devices++;
1930         root->fs_info->fs_devices->rw_devices++;
1931         root->fs_info->fs_devices->total_devices++;
1932         if (device->can_discard)
1933                 root->fs_info->fs_devices->num_can_discard++;
1934         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1935
1936         spin_lock(&root->fs_info->free_chunk_lock);
1937         root->fs_info->free_chunk_space += device->total_bytes;
1938         spin_unlock(&root->fs_info->free_chunk_lock);
1939
1940         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1941                 root->fs_info->fs_devices->rotating = 1;
1942
1943         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1944         btrfs_set_super_total_bytes(root->fs_info->super_copy,
1945                                     total_bytes + device->total_bytes);
1946
1947         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1948         btrfs_set_super_num_devices(root->fs_info->super_copy,
1949                                     total_bytes + 1);
1950         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1951
1952         if (seeding_dev) {
1953                 ret = init_first_rw_device(trans, root, device);
1954                 if (ret) {
1955                         btrfs_abort_transaction(trans, root, ret);
1956                         goto error_trans;
1957                 }
1958                 ret = btrfs_finish_sprout(trans, root);
1959                 if (ret) {
1960                         btrfs_abort_transaction(trans, root, ret);
1961                         goto error_trans;
1962                 }
1963         } else {
1964                 ret = btrfs_add_device(trans, root, device);
1965                 if (ret) {
1966                         btrfs_abort_transaction(trans, root, ret);
1967                         goto error_trans;
1968                 }
1969         }
1970
1971         /*
1972          * we've got more storage, clear any full flags on the space
1973          * infos
1974          */
1975         btrfs_clear_space_info_full(root->fs_info);
1976
1977         unlock_chunks(root);
1978         root->fs_info->num_tolerated_disk_barrier_failures =
1979                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1980         ret = btrfs_commit_transaction(trans, root);
1981
1982         if (seeding_dev) {
1983                 mutex_unlock(&uuid_mutex);
1984                 up_write(&sb->s_umount);
1985
1986                 if (ret) /* transaction commit */
1987                         return ret;
1988
1989                 ret = btrfs_relocate_sys_chunks(root);
1990                 if (ret < 0)
1991                         btrfs_error(root->fs_info, ret,
1992                                     "Failed to relocate sys chunks after "
1993                                     "device initialization. This can be fixed "
1994                                     "using the \"btrfs balance\" command.");
1995                 trans = btrfs_attach_transaction(root);
1996                 if (IS_ERR(trans)) {
1997                         if (PTR_ERR(trans) == -ENOENT)
1998                                 return 0;
1999                         return PTR_ERR(trans);
2000                 }
2001                 ret = btrfs_commit_transaction(trans, root);
2002         }
2003
2004         return ret;
2005
2006 error_trans:
2007         unlock_chunks(root);
2008         btrfs_end_transaction(trans, root);
2009         rcu_string_free(device->name);
2010         kfree(device);
2011 error:
2012         blkdev_put(bdev, FMODE_EXCL);
2013         if (seeding_dev) {
2014                 mutex_unlock(&uuid_mutex);
2015                 up_write(&sb->s_umount);
2016         }
2017         return ret;
2018 }
2019
2020 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2021                                   struct btrfs_device **device_out)
2022 {
2023         struct request_queue *q;
2024         struct btrfs_device *device;
2025         struct block_device *bdev;
2026         struct btrfs_fs_info *fs_info = root->fs_info;
2027         struct list_head *devices;
2028         struct rcu_string *name;
2029         int ret = 0;
2030
2031         *device_out = NULL;
2032         if (fs_info->fs_devices->seeding)
2033                 return -EINVAL;
2034
2035         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2036                                   fs_info->bdev_holder);
2037         if (IS_ERR(bdev))
2038                 return PTR_ERR(bdev);
2039
2040         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2041
2042         devices = &fs_info->fs_devices->devices;
2043         list_for_each_entry(device, devices, dev_list) {
2044                 if (device->bdev == bdev) {
2045                         ret = -EEXIST;
2046                         goto error;
2047                 }
2048         }
2049
2050         device = kzalloc(sizeof(*device), GFP_NOFS);
2051         if (!device) {
2052                 ret = -ENOMEM;
2053                 goto error;
2054         }
2055
2056         name = rcu_string_strdup(device_path, GFP_NOFS);
2057         if (!name) {
2058                 kfree(device);
2059                 ret = -ENOMEM;
2060                 goto error;
2061         }
2062         rcu_assign_pointer(device->name, name);
2063
2064         q = bdev_get_queue(bdev);
2065         if (blk_queue_discard(q))
2066                 device->can_discard = 1;
2067         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2068         device->writeable = 1;
2069         device->work.func = pending_bios_fn;
2070         generate_random_uuid(device->uuid);
2071         device->devid = BTRFS_DEV_REPLACE_DEVID;
2072         spin_lock_init(&device->io_lock);
2073         device->generation = 0;
2074         device->io_width = root->sectorsize;
2075         device->io_align = root->sectorsize;
2076         device->sector_size = root->sectorsize;
2077         device->total_bytes = i_size_read(bdev->bd_inode);
2078         device->disk_total_bytes = device->total_bytes;
2079         device->dev_root = fs_info->dev_root;
2080         device->bdev = bdev;
2081         device->in_fs_metadata = 1;
2082         device->is_tgtdev_for_dev_replace = 1;
2083         device->mode = FMODE_EXCL;
2084         set_blocksize(device->bdev, 4096);
2085         device->fs_devices = fs_info->fs_devices;
2086         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2087         fs_info->fs_devices->num_devices++;
2088         fs_info->fs_devices->open_devices++;
2089         if (device->can_discard)
2090                 fs_info->fs_devices->num_can_discard++;
2091         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2092
2093         *device_out = device;
2094         return ret;
2095
2096 error:
2097         blkdev_put(bdev, FMODE_EXCL);
2098         return ret;
2099 }
2100
2101 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2102                                               struct btrfs_device *tgtdev)
2103 {
2104         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2105         tgtdev->io_width = fs_info->dev_root->sectorsize;
2106         tgtdev->io_align = fs_info->dev_root->sectorsize;
2107         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2108         tgtdev->dev_root = fs_info->dev_root;
2109         tgtdev->in_fs_metadata = 1;
2110 }
2111
2112 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2113                                         struct btrfs_device *device)
2114 {
2115         int ret;
2116         struct btrfs_path *path;
2117         struct btrfs_root *root;
2118         struct btrfs_dev_item *dev_item;
2119         struct extent_buffer *leaf;
2120         struct btrfs_key key;
2121
2122         root = device->dev_root->fs_info->chunk_root;
2123
2124         path = btrfs_alloc_path();
2125         if (!path)
2126                 return -ENOMEM;
2127
2128         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2129         key.type = BTRFS_DEV_ITEM_KEY;
2130         key.offset = device->devid;
2131
2132         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2133         if (ret < 0)
2134                 goto out;
2135
2136         if (ret > 0) {
2137                 ret = -ENOENT;
2138                 goto out;
2139         }
2140
2141         leaf = path->nodes[0];
2142         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2143
2144         btrfs_set_device_id(leaf, dev_item, device->devid);
2145         btrfs_set_device_type(leaf, dev_item, device->type);
2146         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2147         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2148         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2149         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2150         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2151         btrfs_mark_buffer_dirty(leaf);
2152
2153 out:
2154         btrfs_free_path(path);
2155         return ret;
2156 }
2157
2158 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2159                       struct btrfs_device *device, u64 new_size)
2160 {
2161         struct btrfs_super_block *super_copy =
2162                 device->dev_root->fs_info->super_copy;
2163         u64 old_total = btrfs_super_total_bytes(super_copy);
2164         u64 diff = new_size - device->total_bytes;
2165
2166         if (!device->writeable)
2167                 return -EACCES;
2168         if (new_size <= device->total_bytes ||
2169             device->is_tgtdev_for_dev_replace)
2170                 return -EINVAL;
2171
2172         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2173         device->fs_devices->total_rw_bytes += diff;
2174
2175         device->total_bytes = new_size;
2176         device->disk_total_bytes = new_size;
2177         btrfs_clear_space_info_full(device->dev_root->fs_info);
2178
2179         return btrfs_update_device(trans, device);
2180 }
2181
2182 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2183                       struct btrfs_device *device, u64 new_size)
2184 {
2185         int ret;
2186         lock_chunks(device->dev_root);
2187         ret = __btrfs_grow_device(trans, device, new_size);
2188         unlock_chunks(device->dev_root);
2189         return ret;
2190 }
2191
2192 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2193                             struct btrfs_root *root,
2194                             u64 chunk_tree, u64 chunk_objectid,
2195                             u64 chunk_offset)
2196 {
2197         int ret;
2198         struct btrfs_path *path;
2199         struct btrfs_key key;
2200
2201         root = root->fs_info->chunk_root;
2202         path = btrfs_alloc_path();
2203         if (!path)
2204                 return -ENOMEM;
2205
2206         key.objectid = chunk_objectid;
2207         key.offset = chunk_offset;
2208         key.type = BTRFS_CHUNK_ITEM_KEY;
2209
2210         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2211         if (ret < 0)
2212                 goto out;
2213         else if (ret > 0) { /* Logic error or corruption */
2214                 btrfs_error(root->fs_info, -ENOENT,
2215                             "Failed lookup while freeing chunk.");
2216                 ret = -ENOENT;
2217                 goto out;
2218         }
2219
2220         ret = btrfs_del_item(trans, root, path);
2221         if (ret < 0)
2222                 btrfs_error(root->fs_info, ret,
2223                             "Failed to delete chunk item.");
2224 out:
2225         btrfs_free_path(path);
2226         return ret;
2227 }
2228
2229 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2230                         chunk_offset)
2231 {
2232         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2233         struct btrfs_disk_key *disk_key;
2234         struct btrfs_chunk *chunk;
2235         u8 *ptr;
2236         int ret = 0;
2237         u32 num_stripes;
2238         u32 array_size;
2239         u32 len = 0;
2240         u32 cur;
2241         struct btrfs_key key;
2242
2243         array_size = btrfs_super_sys_array_size(super_copy);
2244
2245         ptr = super_copy->sys_chunk_array;
2246         cur = 0;
2247
2248         while (cur < array_size) {
2249                 disk_key = (struct btrfs_disk_key *)ptr;
2250                 btrfs_disk_key_to_cpu(&key, disk_key);
2251
2252                 len = sizeof(*disk_key);
2253
2254                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2255                         chunk = (struct btrfs_chunk *)(ptr + len);
2256                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2257                         len += btrfs_chunk_item_size(num_stripes);
2258                 } else {
2259                         ret = -EIO;
2260                         break;
2261                 }
2262                 if (key.objectid == chunk_objectid &&
2263                     key.offset == chunk_offset) {
2264                         memmove(ptr, ptr + len, array_size - (cur + len));
2265                         array_size -= len;
2266                         btrfs_set_super_sys_array_size(super_copy, array_size);
2267                 } else {
2268                         ptr += len;
2269                         cur += len;
2270                 }
2271         }
2272         return ret;
2273 }
2274
2275 static int btrfs_relocate_chunk(struct btrfs_root *root,
2276                          u64 chunk_tree, u64 chunk_objectid,
2277                          u64 chunk_offset)
2278 {
2279         struct extent_map_tree *em_tree;
2280         struct btrfs_root *extent_root;
2281         struct btrfs_trans_handle *trans;
2282         struct extent_map *em;
2283         struct map_lookup *map;
2284         int ret;
2285         int i;
2286
2287         root = root->fs_info->chunk_root;
2288         extent_root = root->fs_info->extent_root;
2289         em_tree = &root->fs_info->mapping_tree.map_tree;
2290
2291         ret = btrfs_can_relocate(extent_root, chunk_offset);
2292         if (ret)
2293                 return -ENOSPC;
2294
2295         /* step one, relocate all the extents inside this chunk */
2296         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2297         if (ret)
2298                 return ret;
2299
2300         trans = btrfs_start_transaction(root, 0);
2301         BUG_ON(IS_ERR(trans));
2302
2303         lock_chunks(root);
2304
2305         /*
2306          * step two, delete the device extents and the
2307          * chunk tree entries
2308          */
2309         read_lock(&em_tree->lock);
2310         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2311         read_unlock(&em_tree->lock);
2312
2313         BUG_ON(!em || em->start > chunk_offset ||
2314                em->start + em->len < chunk_offset);
2315         map = (struct map_lookup *)em->bdev;
2316
2317         for (i = 0; i < map->num_stripes; i++) {
2318                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2319                                             map->stripes[i].physical);
2320                 BUG_ON(ret);
2321
2322                 if (map->stripes[i].dev) {
2323                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2324                         BUG_ON(ret);
2325                 }
2326         }
2327         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2328                                chunk_offset);
2329
2330         BUG_ON(ret);
2331
2332         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2333
2334         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2335                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2336                 BUG_ON(ret);
2337         }
2338
2339         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2340         BUG_ON(ret);
2341
2342         write_lock(&em_tree->lock);
2343         remove_extent_mapping(em_tree, em);
2344         write_unlock(&em_tree->lock);
2345
2346         kfree(map);
2347         em->bdev = NULL;
2348
2349         /* once for the tree */
2350         free_extent_map(em);
2351         /* once for us */
2352         free_extent_map(em);
2353
2354         unlock_chunks(root);
2355         btrfs_end_transaction(trans, root);
2356         return 0;
2357 }
2358
2359 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2360 {
2361         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2362         struct btrfs_path *path;
2363         struct extent_buffer *leaf;
2364         struct btrfs_chunk *chunk;
2365         struct btrfs_key key;
2366         struct btrfs_key found_key;
2367         u64 chunk_tree = chunk_root->root_key.objectid;
2368         u64 chunk_type;
2369         bool retried = false;
2370         int failed = 0;
2371         int ret;
2372
2373         path = btrfs_alloc_path();
2374         if (!path)
2375                 return -ENOMEM;
2376
2377 again:
2378         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2379         key.offset = (u64)-1;
2380         key.type = BTRFS_CHUNK_ITEM_KEY;
2381
2382         while (1) {
2383                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2384                 if (ret < 0)
2385                         goto error;
2386                 BUG_ON(ret == 0); /* Corruption */
2387
2388                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2389                                           key.type);
2390                 if (ret < 0)
2391                         goto error;
2392                 if (ret > 0)
2393                         break;
2394
2395                 leaf = path->nodes[0];
2396                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2397
2398                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2399                                        struct btrfs_chunk);
2400                 chunk_type = btrfs_chunk_type(leaf, chunk);
2401                 btrfs_release_path(path);
2402
2403                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2404                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2405                                                    found_key.objectid,
2406                                                    found_key.offset);
2407                         if (ret == -ENOSPC)
2408                                 failed++;
2409                         else if (ret)
2410                                 BUG();
2411                 }
2412
2413                 if (found_key.offset == 0)
2414                         break;
2415                 key.offset = found_key.offset - 1;
2416         }
2417         ret = 0;
2418         if (failed && !retried) {
2419                 failed = 0;
2420                 retried = true;
2421                 goto again;
2422         } else if (failed && retried) {
2423                 WARN_ON(1);
2424                 ret = -ENOSPC;
2425         }
2426 error:
2427         btrfs_free_path(path);
2428         return ret;
2429 }
2430
2431 static int insert_balance_item(struct btrfs_root *root,
2432                                struct btrfs_balance_control *bctl)
2433 {
2434         struct btrfs_trans_handle *trans;
2435         struct btrfs_balance_item *item;
2436         struct btrfs_disk_balance_args disk_bargs;
2437         struct btrfs_path *path;
2438         struct extent_buffer *leaf;
2439         struct btrfs_key key;
2440         int ret, err;
2441
2442         path = btrfs_alloc_path();
2443         if (!path)
2444                 return -ENOMEM;
2445
2446         trans = btrfs_start_transaction(root, 0);
2447         if (IS_ERR(trans)) {
2448                 btrfs_free_path(path);
2449                 return PTR_ERR(trans);
2450         }
2451
2452         key.objectid = BTRFS_BALANCE_OBJECTID;
2453         key.type = BTRFS_BALANCE_ITEM_KEY;
2454         key.offset = 0;
2455
2456         ret = btrfs_insert_empty_item(trans, root, path, &key,
2457                                       sizeof(*item));
2458         if (ret)
2459                 goto out;
2460
2461         leaf = path->nodes[0];
2462         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2463
2464         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2465
2466         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2467         btrfs_set_balance_data(leaf, item, &disk_bargs);
2468         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2469         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2470         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2471         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2472
2473         btrfs_set_balance_flags(leaf, item, bctl->flags);
2474
2475         btrfs_mark_buffer_dirty(leaf);
2476 out:
2477         btrfs_free_path(path);
2478         err = btrfs_commit_transaction(trans, root);
2479         if (err && !ret)
2480                 ret = err;
2481         return ret;
2482 }
2483
2484 static int del_balance_item(struct btrfs_root *root)
2485 {
2486         struct btrfs_trans_handle *trans;
2487         struct btrfs_path *path;
2488         struct btrfs_key key;
2489         int ret, err;
2490
2491         path = btrfs_alloc_path();
2492         if (!path)
2493                 return -ENOMEM;
2494
2495         trans = btrfs_start_transaction(root, 0);
2496         if (IS_ERR(trans)) {
2497                 btrfs_free_path(path);
2498                 return PTR_ERR(trans);
2499         }
2500
2501         key.objectid = BTRFS_BALANCE_OBJECTID;
2502         key.type = BTRFS_BALANCE_ITEM_KEY;
2503         key.offset = 0;
2504
2505         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2506         if (ret < 0)
2507                 goto out;
2508         if (ret > 0) {
2509                 ret = -ENOENT;
2510                 goto out;
2511         }
2512
2513         ret = btrfs_del_item(trans, root, path);
2514 out:
2515         btrfs_free_path(path);
2516         err = btrfs_commit_transaction(trans, root);
2517         if (err && !ret)
2518                 ret = err;
2519         return ret;
2520 }
2521
2522 /*
2523  * This is a heuristic used to reduce the number of chunks balanced on
2524  * resume after balance was interrupted.
2525  */
2526 static void update_balance_args(struct btrfs_balance_control *bctl)
2527 {
2528         /*
2529          * Turn on soft mode for chunk types that were being converted.
2530          */
2531         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2532                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2533         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2534                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2535         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2536                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2537
2538         /*
2539          * Turn on usage filter if is not already used.  The idea is
2540          * that chunks that we have already balanced should be
2541          * reasonably full.  Don't do it for chunks that are being
2542          * converted - that will keep us from relocating unconverted
2543          * (albeit full) chunks.
2544          */
2545         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2546             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2547                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2548                 bctl->data.usage = 90;
2549         }
2550         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2551             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2552                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2553                 bctl->sys.usage = 90;
2554         }
2555         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2556             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2557                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2558                 bctl->meta.usage = 90;
2559         }
2560 }
2561
2562 /*
2563  * Should be called with both balance and volume mutexes held to
2564  * serialize other volume operations (add_dev/rm_dev/resize) with
2565  * restriper.  Same goes for unset_balance_control.
2566  */
2567 static void set_balance_control(struct btrfs_balance_control *bctl)
2568 {
2569         struct btrfs_fs_info *fs_info = bctl->fs_info;
2570
2571         BUG_ON(fs_info->balance_ctl);
2572
2573         spin_lock(&fs_info->balance_lock);
2574         fs_info->balance_ctl = bctl;
2575         spin_unlock(&fs_info->balance_lock);
2576 }
2577
2578 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2579 {
2580         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2581
2582         BUG_ON(!fs_info->balance_ctl);
2583
2584         spin_lock(&fs_info->balance_lock);
2585         fs_info->balance_ctl = NULL;
2586         spin_unlock(&fs_info->balance_lock);
2587
2588         kfree(bctl);
2589 }
2590
2591 /*
2592  * Balance filters.  Return 1 if chunk should be filtered out
2593  * (should not be balanced).
2594  */
2595 static int chunk_profiles_filter(u64 chunk_type,
2596                                  struct btrfs_balance_args *bargs)
2597 {
2598         chunk_type = chunk_to_extended(chunk_type) &
2599                                 BTRFS_EXTENDED_PROFILE_MASK;
2600
2601         if (bargs->profiles & chunk_type)
2602                 return 0;
2603
2604         return 1;
2605 }
2606
2607 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2608                               struct btrfs_balance_args *bargs)
2609 {
2610         struct btrfs_block_group_cache *cache;
2611         u64 chunk_used, user_thresh;
2612         int ret = 1;
2613
2614         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2615         chunk_used = btrfs_block_group_used(&cache->item);
2616
2617         if (bargs->usage == 0)
2618                 user_thresh = 0;
2619         else if (bargs->usage > 100)
2620                 user_thresh = cache->key.offset;
2621         else
2622                 user_thresh = div_factor_fine(cache->key.offset,
2623                                               bargs->usage);
2624
2625         if (chunk_used < user_thresh)
2626                 ret = 0;
2627
2628         btrfs_put_block_group(cache);
2629         return ret;
2630 }
2631
2632 static int chunk_devid_filter(struct extent_buffer *leaf,
2633                               struct btrfs_chunk *chunk,
2634                               struct btrfs_balance_args *bargs)
2635 {
2636         struct btrfs_stripe *stripe;
2637         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2638         int i;
2639
2640         for (i = 0; i < num_stripes; i++) {
2641                 stripe = btrfs_stripe_nr(chunk, i);
2642                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2643                         return 0;
2644         }
2645
2646         return 1;
2647 }
2648
2649 /* [pstart, pend) */
2650 static int chunk_drange_filter(struct extent_buffer *leaf,
2651                                struct btrfs_chunk *chunk,
2652                                u64 chunk_offset,
2653                                struct btrfs_balance_args *bargs)
2654 {
2655         struct btrfs_stripe *stripe;
2656         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2657         u64 stripe_offset;
2658         u64 stripe_length;
2659         int factor;
2660         int i;
2661
2662         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2663                 return 0;
2664
2665         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2666              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2667                 factor = 2;
2668         else
2669                 factor = 1;
2670         factor = num_stripes / factor;
2671
2672         for (i = 0; i < num_stripes; i++) {
2673                 stripe = btrfs_stripe_nr(chunk, i);
2674                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2675                         continue;
2676
2677                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2678                 stripe_length = btrfs_chunk_length(leaf, chunk);
2679                 do_div(stripe_length, factor);
2680
2681                 if (stripe_offset < bargs->pend &&
2682                     stripe_offset + stripe_length > bargs->pstart)
2683                         return 0;
2684         }
2685
2686         return 1;
2687 }
2688
2689 /* [vstart, vend) */
2690 static int chunk_vrange_filter(struct extent_buffer *leaf,
2691                                struct btrfs_chunk *chunk,
2692                                u64 chunk_offset,
2693                                struct btrfs_balance_args *bargs)
2694 {
2695         if (chunk_offset < bargs->vend &&
2696             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2697                 /* at least part of the chunk is inside this vrange */
2698                 return 0;
2699
2700         return 1;
2701 }
2702
2703 static int chunk_soft_convert_filter(u64 chunk_type,
2704                                      struct btrfs_balance_args *bargs)
2705 {
2706         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2707                 return 0;
2708
2709         chunk_type = chunk_to_extended(chunk_type) &
2710                                 BTRFS_EXTENDED_PROFILE_MASK;
2711
2712         if (bargs->target == chunk_type)
2713                 return 1;
2714
2715         return 0;
2716 }
2717
2718 static int should_balance_chunk(struct btrfs_root *root,
2719                                 struct extent_buffer *leaf,
2720                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2721 {
2722         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2723         struct btrfs_balance_args *bargs = NULL;
2724         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2725
2726         /* type filter */
2727         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2728               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2729                 return 0;
2730         }
2731
2732         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2733                 bargs = &bctl->data;
2734         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2735                 bargs = &bctl->sys;
2736         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2737                 bargs = &bctl->meta;
2738
2739         /* profiles filter */
2740         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2741             chunk_profiles_filter(chunk_type, bargs)) {
2742                 return 0;
2743         }
2744
2745         /* usage filter */
2746         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2747             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2748                 return 0;
2749         }
2750
2751         /* devid filter */
2752         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2753             chunk_devid_filter(leaf, chunk, bargs)) {
2754                 return 0;
2755         }
2756
2757         /* drange filter, makes sense only with devid filter */
2758         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2759             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2760                 return 0;
2761         }
2762
2763         /* vrange filter */
2764         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2765             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2766                 return 0;
2767         }
2768
2769         /* soft profile changing mode */
2770         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2771             chunk_soft_convert_filter(chunk_type, bargs)) {
2772                 return 0;
2773         }
2774
2775         return 1;
2776 }
2777
2778 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2779 {
2780         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2781         struct btrfs_root *chunk_root = fs_info->chunk_root;
2782         struct btrfs_root *dev_root = fs_info->dev_root;
2783         struct list_head *devices;
2784         struct btrfs_device *device;
2785         u64 old_size;
2786         u64 size_to_free;
2787         struct btrfs_chunk *chunk;
2788         struct btrfs_path *path;
2789         struct btrfs_key key;
2790         struct btrfs_key found_key;
2791         struct btrfs_trans_handle *trans;
2792         struct extent_buffer *leaf;
2793         int slot;
2794         int ret;
2795         int enospc_errors = 0;
2796         bool counting = true;
2797
2798         /* step one make some room on all the devices */
2799         devices = &fs_info->fs_devices->devices;
2800         list_for_each_entry(device, devices, dev_list) {
2801                 old_size = device->total_bytes;
2802                 size_to_free = div_factor(old_size, 1);
2803                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2804                 if (!device->writeable ||
2805                     device->total_bytes - device->bytes_used > size_to_free ||
2806                     device->is_tgtdev_for_dev_replace)
2807                         continue;
2808
2809                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2810                 if (ret == -ENOSPC)
2811                         break;
2812                 BUG_ON(ret);
2813
2814                 trans = btrfs_start_transaction(dev_root, 0);
2815                 BUG_ON(IS_ERR(trans));
2816
2817                 ret = btrfs_grow_device(trans, device, old_size);
2818                 BUG_ON(ret);
2819
2820                 btrfs_end_transaction(trans, dev_root);
2821         }
2822
2823         /* step two, relocate all the chunks */
2824         path = btrfs_alloc_path();
2825         if (!path) {
2826                 ret = -ENOMEM;
2827                 goto error;
2828         }
2829
2830         /* zero out stat counters */
2831         spin_lock(&fs_info->balance_lock);
2832         memset(&bctl->stat, 0, sizeof(bctl->stat));
2833         spin_unlock(&fs_info->balance_lock);
2834 again:
2835         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2836         key.offset = (u64)-1;
2837         key.type = BTRFS_CHUNK_ITEM_KEY;
2838
2839         while (1) {
2840                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2841                     atomic_read(&fs_info->balance_cancel_req)) {
2842                         ret = -ECANCELED;
2843                         goto error;
2844                 }
2845
2846                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2847                 if (ret < 0)
2848                         goto error;
2849
2850                 /*
2851                  * this shouldn't happen, it means the last relocate
2852                  * failed
2853                  */
2854                 if (ret == 0)
2855                         BUG(); /* FIXME break ? */
2856
2857                 ret = btrfs_previous_item(chunk_root, path, 0,
2858                                           BTRFS_CHUNK_ITEM_KEY);
2859                 if (ret) {
2860                         ret = 0;
2861                         break;
2862                 }
2863
2864                 leaf = path->nodes[0];
2865                 slot = path->slots[0];
2866                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2867
2868                 if (found_key.objectid != key.objectid)
2869                         break;
2870
2871                 /* chunk zero is special */
2872                 if (found_key.offset == 0)
2873                         break;
2874
2875                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2876
2877                 if (!counting) {
2878                         spin_lock(&fs_info->balance_lock);
2879                         bctl->stat.considered++;
2880                         spin_unlock(&fs_info->balance_lock);
2881                 }
2882
2883                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2884                                            found_key.offset);
2885                 btrfs_release_path(path);
2886                 if (!ret)
2887                         goto loop;
2888
2889                 if (counting) {
2890                         spin_lock(&fs_info->balance_lock);
2891                         bctl->stat.expected++;
2892                         spin_unlock(&fs_info->balance_lock);
2893                         goto loop;
2894                 }
2895
2896                 ret = btrfs_relocate_chunk(chunk_root,
2897                                            chunk_root->root_key.objectid,
2898                                            found_key.objectid,
2899                                            found_key.offset);
2900                 if (ret && ret != -ENOSPC)
2901                         goto error;
2902                 if (ret == -ENOSPC) {
2903                         enospc_errors++;
2904                 } else {
2905                         spin_lock(&fs_info->balance_lock);
2906                         bctl->stat.completed++;
2907                         spin_unlock(&fs_info->balance_lock);
2908                 }
2909 loop:
2910                 key.offset = found_key.offset - 1;
2911         }
2912
2913         if (counting) {
2914                 btrfs_release_path(path);
2915                 counting = false;
2916                 goto again;
2917         }
2918 error:
2919         btrfs_free_path(path);
2920         if (enospc_errors) {
2921                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2922                        enospc_errors);
2923                 if (!ret)
2924                         ret = -ENOSPC;
2925         }
2926
2927         return ret;
2928 }
2929
2930 /**
2931  * alloc_profile_is_valid - see if a given profile is valid and reduced
2932  * @flags: profile to validate
2933  * @extended: if true @flags is treated as an extended profile
2934  */
2935 static int alloc_profile_is_valid(u64 flags, int extended)
2936 {
2937         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2938                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
2939
2940         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2941
2942         /* 1) check that all other bits are zeroed */
2943         if (flags & ~mask)
2944                 return 0;
2945
2946         /* 2) see if profile is reduced */
2947         if (flags == 0)
2948                 return !extended; /* "0" is valid for usual profiles */
2949
2950         /* true if exactly one bit set */
2951         return (flags & (flags - 1)) == 0;
2952 }
2953
2954 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2955 {
2956         /* cancel requested || normal exit path */
2957         return atomic_read(&fs_info->balance_cancel_req) ||
2958                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2959                  atomic_read(&fs_info->balance_cancel_req) == 0);
2960 }
2961
2962 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2963 {
2964         int ret;
2965
2966         unset_balance_control(fs_info);
2967         ret = del_balance_item(fs_info->tree_root);
2968         BUG_ON(ret);
2969
2970         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
2971 }
2972
2973 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2974                                struct btrfs_ioctl_balance_args *bargs);
2975
2976 /*
2977  * Should be called with both balance and volume mutexes held
2978  */
2979 int btrfs_balance(struct btrfs_balance_control *bctl,
2980                   struct btrfs_ioctl_balance_args *bargs)
2981 {
2982         struct btrfs_fs_info *fs_info = bctl->fs_info;
2983         u64 allowed;
2984         int mixed = 0;
2985         int ret;
2986         u64 num_devices;
2987
2988         if (btrfs_fs_closing(fs_info) ||
2989             atomic_read(&fs_info->balance_pause_req) ||
2990             atomic_read(&fs_info->balance_cancel_req)) {
2991                 ret = -EINVAL;
2992                 goto out;
2993         }
2994
2995         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2996         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2997                 mixed = 1;
2998
2999         /*
3000          * In case of mixed groups both data and meta should be picked,
3001          * and identical options should be given for both of them.
3002          */
3003         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3004         if (mixed && (bctl->flags & allowed)) {
3005                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3006                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3007                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3008                         printk(KERN_ERR "btrfs: with mixed groups data and "
3009                                "metadata balance options must be the same\n");
3010                         ret = -EINVAL;
3011                         goto out;
3012                 }
3013         }
3014
3015         num_devices = fs_info->fs_devices->num_devices;
3016         btrfs_dev_replace_lock(&fs_info->dev_replace);
3017         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3018                 BUG_ON(num_devices < 1);
3019                 num_devices--;
3020         }
3021         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3022         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3023         if (num_devices == 1)
3024                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3025         else if (num_devices < 4)
3026                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3027         else
3028                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3029                                 BTRFS_BLOCK_GROUP_RAID10);
3030
3031         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3032             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3033              (bctl->data.target & ~allowed))) {
3034                 printk(KERN_ERR "btrfs: unable to start balance with target "
3035                        "data profile %llu\n",
3036                        (unsigned long long)bctl->data.target);
3037                 ret = -EINVAL;
3038                 goto out;
3039         }
3040         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3041             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3042              (bctl->meta.target & ~allowed))) {
3043                 printk(KERN_ERR "btrfs: unable to start balance with target "
3044                        "metadata profile %llu\n",
3045                        (unsigned long long)bctl->meta.target);
3046                 ret = -EINVAL;
3047                 goto out;
3048         }
3049         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3050             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3051              (bctl->sys.target & ~allowed))) {
3052                 printk(KERN_ERR "btrfs: unable to start balance with target "
3053                        "system profile %llu\n",
3054                        (unsigned long long)bctl->sys.target);
3055                 ret = -EINVAL;
3056                 goto out;
3057         }
3058
3059         /* allow dup'ed data chunks only in mixed mode */
3060         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3061             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3062                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3063                 ret = -EINVAL;
3064                 goto out;
3065         }
3066
3067         /* allow to reduce meta or sys integrity only if force set */
3068         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3069                         BTRFS_BLOCK_GROUP_RAID10;
3070         if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3071              (fs_info->avail_system_alloc_bits & allowed) &&
3072              !(bctl->sys.target & allowed)) ||
3073             ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3074              (fs_info->avail_metadata_alloc_bits & allowed) &&
3075              !(bctl->meta.target & allowed))) {
3076                 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3077                         printk(KERN_INFO "btrfs: force reducing metadata "
3078                                "integrity\n");
3079                 } else {
3080                         printk(KERN_ERR "btrfs: balance will reduce metadata "
3081                                "integrity, use force if you want this\n");
3082                         ret = -EINVAL;
3083                         goto out;
3084                 }
3085         }
3086
3087         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3088                 int num_tolerated_disk_barrier_failures;
3089                 u64 target = bctl->sys.target;
3090
3091                 num_tolerated_disk_barrier_failures =
3092                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3093                 if (num_tolerated_disk_barrier_failures > 0 &&
3094                     (target &
3095                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3096                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3097                         num_tolerated_disk_barrier_failures = 0;
3098                 else if (num_tolerated_disk_barrier_failures > 1 &&
3099                          (target &
3100                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3101                         num_tolerated_disk_barrier_failures = 1;
3102
3103                 fs_info->num_tolerated_disk_barrier_failures =
3104                         num_tolerated_disk_barrier_failures;
3105         }
3106
3107         ret = insert_balance_item(fs_info->tree_root, bctl);
3108         if (ret && ret != -EEXIST)
3109                 goto out;
3110
3111         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3112                 BUG_ON(ret == -EEXIST);
3113                 set_balance_control(bctl);
3114         } else {
3115                 BUG_ON(ret != -EEXIST);
3116                 spin_lock(&fs_info->balance_lock);
3117                 update_balance_args(bctl);
3118                 spin_unlock(&fs_info->balance_lock);
3119         }
3120
3121         atomic_inc(&fs_info->balance_running);
3122         mutex_unlock(&fs_info->balance_mutex);
3123
3124         ret = __btrfs_balance(fs_info);
3125
3126         mutex_lock(&fs_info->balance_mutex);
3127         atomic_dec(&fs_info->balance_running);
3128
3129         if (bargs) {
3130                 memset(bargs, 0, sizeof(*bargs));
3131                 update_ioctl_balance_args(fs_info, 0, bargs);
3132         }
3133
3134         if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3135             balance_need_close(fs_info)) {
3136                 __cancel_balance(fs_info);
3137         }
3138
3139         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3140                 fs_info->num_tolerated_disk_barrier_failures =
3141                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3142         }
3143
3144         wake_up(&fs_info->balance_wait_q);
3145
3146         return ret;
3147 out:
3148         if (bctl->flags & BTRFS_BALANCE_RESUME)
3149                 __cancel_balance(fs_info);
3150         else {
3151                 kfree(bctl);
3152                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3153         }
3154         return ret;
3155 }
3156
3157 static int balance_kthread(void *data)
3158 {
3159         struct btrfs_fs_info *fs_info = data;
3160         int ret = 0;
3161
3162         mutex_lock(&fs_info->volume_mutex);
3163         mutex_lock(&fs_info->balance_mutex);
3164
3165         if (fs_info->balance_ctl) {
3166                 printk(KERN_INFO "btrfs: continuing balance\n");
3167                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3168         }
3169
3170         mutex_unlock(&fs_info->balance_mutex);
3171         mutex_unlock(&fs_info->volume_mutex);
3172
3173         return ret;
3174 }
3175
3176 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3177 {
3178         struct task_struct *tsk;
3179
3180         spin_lock(&fs_info->balance_lock);
3181         if (!fs_info->balance_ctl) {
3182                 spin_unlock(&fs_info->balance_lock);
3183                 return 0;
3184         }
3185         spin_unlock(&fs_info->balance_lock);
3186
3187         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3188                 printk(KERN_INFO "btrfs: force skipping balance\n");
3189                 return 0;
3190         }
3191
3192         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3193         if (IS_ERR(tsk))
3194                 return PTR_ERR(tsk);
3195
3196         return 0;
3197 }
3198
3199 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3200 {
3201         struct btrfs_balance_control *bctl;
3202         struct btrfs_balance_item *item;
3203         struct btrfs_disk_balance_args disk_bargs;
3204         struct btrfs_path *path;
3205         struct extent_buffer *leaf;
3206         struct btrfs_key key;
3207         int ret;
3208
3209         path = btrfs_alloc_path();
3210         if (!path)
3211                 return -ENOMEM;
3212
3213         key.objectid = BTRFS_BALANCE_OBJECTID;
3214         key.type = BTRFS_BALANCE_ITEM_KEY;
3215         key.offset = 0;
3216
3217         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3218         if (ret < 0)
3219                 goto out;
3220         if (ret > 0) { /* ret = -ENOENT; */
3221                 ret = 0;
3222                 goto out;
3223         }
3224
3225         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3226         if (!bctl) {
3227                 ret = -ENOMEM;
3228                 goto out;
3229         }
3230
3231         leaf = path->nodes[0];
3232         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3233
3234         bctl->fs_info = fs_info;
3235         bctl->flags = btrfs_balance_flags(leaf, item);
3236         bctl->flags |= BTRFS_BALANCE_RESUME;
3237
3238         btrfs_balance_data(leaf, item, &disk_bargs);
3239         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3240         btrfs_balance_meta(leaf, item, &disk_bargs);
3241         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3242         btrfs_balance_sys(leaf, item, &disk_bargs);
3243         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3244
3245         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3246
3247         mutex_lock(&fs_info->volume_mutex);
3248         mutex_lock(&fs_info->balance_mutex);
3249
3250         set_balance_control(bctl);
3251
3252         mutex_unlock(&fs_info->balance_mutex);
3253         mutex_unlock(&fs_info->volume_mutex);
3254 out:
3255         btrfs_free_path(path);
3256         return ret;
3257 }
3258
3259 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3260 {
3261         int ret = 0;
3262
3263         mutex_lock(&fs_info->balance_mutex);
3264         if (!fs_info->balance_ctl) {
3265                 mutex_unlock(&fs_info->balance_mutex);
3266                 return -ENOTCONN;
3267         }
3268
3269         if (atomic_read(&fs_info->balance_running)) {
3270                 atomic_inc(&fs_info->balance_pause_req);
3271                 mutex_unlock(&fs_info->balance_mutex);
3272
3273                 wait_event(fs_info->balance_wait_q,
3274                            atomic_read(&fs_info->balance_running) == 0);
3275
3276                 mutex_lock(&fs_info->balance_mutex);
3277                 /* we are good with balance_ctl ripped off from under us */
3278                 BUG_ON(atomic_read(&fs_info->balance_running));
3279                 atomic_dec(&fs_info->balance_pause_req);
3280         } else {
3281                 ret = -ENOTCONN;
3282         }
3283
3284         mutex_unlock(&fs_info->balance_mutex);
3285         return ret;
3286 }
3287
3288 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3289 {
3290         mutex_lock(&fs_info->balance_mutex);
3291         if (!fs_info->balance_ctl) {
3292                 mutex_unlock(&fs_info->balance_mutex);
3293                 return -ENOTCONN;
3294         }
3295
3296         atomic_inc(&fs_info->balance_cancel_req);
3297         /*
3298          * if we are running just wait and return, balance item is
3299          * deleted in btrfs_balance in this case
3300          */
3301         if (atomic_read(&fs_info->balance_running)) {
3302                 mutex_unlock(&fs_info->balance_mutex);
3303                 wait_event(fs_info->balance_wait_q,
3304                            atomic_read(&fs_info->balance_running) == 0);
3305                 mutex_lock(&fs_info->balance_mutex);
3306         } else {
3307                 /* __cancel_balance needs volume_mutex */
3308                 mutex_unlock(&fs_info->balance_mutex);
3309                 mutex_lock(&fs_info->volume_mutex);
3310                 mutex_lock(&fs_info->balance_mutex);
3311
3312                 if (fs_info->balance_ctl)
3313                         __cancel_balance(fs_info);
3314
3315                 mutex_unlock(&fs_info->volume_mutex);
3316         }
3317
3318         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3319         atomic_dec(&fs_info->balance_cancel_req);
3320         mutex_unlock(&fs_info->balance_mutex);
3321         return 0;
3322 }
3323
3324 /*
3325  * shrinking a device means finding all of the device extents past
3326  * the new size, and then following the back refs to the chunks.
3327  * The chunk relocation code actually frees the device extent
3328  */
3329 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3330 {
3331         struct btrfs_trans_handle *trans;
3332         struct btrfs_root *root = device->dev_root;
3333         struct btrfs_dev_extent *dev_extent = NULL;
3334         struct btrfs_path *path;
3335         u64 length;
3336         u64 chunk_tree;
3337         u64 chunk_objectid;
3338         u64 chunk_offset;
3339         int ret;
3340         int slot;
3341         int failed = 0;
3342         bool retried = false;
3343         struct extent_buffer *l;
3344         struct btrfs_key key;
3345         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3346         u64 old_total = btrfs_super_total_bytes(super_copy);
3347         u64 old_size = device->total_bytes;
3348         u64 diff = device->total_bytes - new_size;
3349
3350         if (device->is_tgtdev_for_dev_replace)
3351                 return -EINVAL;
3352
3353         path = btrfs_alloc_path();
3354         if (!path)
3355                 return -ENOMEM;
3356
3357         path->reada = 2;
3358
3359         lock_chunks(root);
3360
3361         device->total_bytes = new_size;
3362         if (device->writeable) {
3363                 device->fs_devices->total_rw_bytes -= diff;
3364                 spin_lock(&root->fs_info->free_chunk_lock);
3365                 root->fs_info->free_chunk_space -= diff;
3366                 spin_unlock(&root->fs_info->free_chunk_lock);
3367         }
3368         unlock_chunks(root);
3369
3370 again:
3371         key.objectid = device->devid;
3372         key.offset = (u64)-1;
3373         key.type = BTRFS_DEV_EXTENT_KEY;
3374
3375         do {
3376                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3377                 if (ret < 0)
3378                         goto done;
3379
3380                 ret = btrfs_previous_item(root, path, 0, key.type);
3381                 if (ret < 0)
3382                         goto done;
3383                 if (ret) {
3384                         ret = 0;
3385                         btrfs_release_path(path);
3386                         break;
3387                 }
3388
3389                 l = path->nodes[0];
3390                 slot = path->slots[0];
3391                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3392
3393                 if (key.objectid != device->devid) {
3394                         btrfs_release_path(path);
3395                         break;
3396                 }
3397
3398                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3399                 length = btrfs_dev_extent_length(l, dev_extent);
3400
3401                 if (key.offset + length <= new_size) {
3402                         btrfs_release_path(path);
3403                         break;
3404                 }
3405
3406                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3407                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3408                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3409                 btrfs_release_path(path);
3410
3411                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3412                                            chunk_offset);
3413                 if (ret && ret != -ENOSPC)
3414                         goto done;
3415                 if (ret == -ENOSPC)
3416                         failed++;
3417         } while (key.offset-- > 0);
3418
3419         if (failed && !retried) {
3420                 failed = 0;
3421                 retried = true;
3422                 goto again;
3423         } else if (failed && retried) {
3424                 ret = -ENOSPC;
3425                 lock_chunks(root);
3426
3427                 device->total_bytes = old_size;
3428                 if (device->writeable)
3429                         device->fs_devices->total_rw_bytes += diff;
3430                 spin_lock(&root->fs_info->free_chunk_lock);
3431                 root->fs_info->free_chunk_space += diff;
3432                 spin_unlock(&root->fs_info->free_chunk_lock);
3433                 unlock_chunks(root);
3434                 goto done;
3435         }
3436
3437         /* Shrinking succeeded, else we would be at "done". */
3438         trans = btrfs_start_transaction(root, 0);
3439         if (IS_ERR(trans)) {
3440                 ret = PTR_ERR(trans);
3441                 goto done;
3442         }
3443
3444         lock_chunks(root);
3445
3446         device->disk_total_bytes = new_size;
3447         /* Now btrfs_update_device() will change the on-disk size. */
3448         ret = btrfs_update_device(trans, device);
3449         if (ret) {
3450                 unlock_chunks(root);
3451                 btrfs_end_transaction(trans, root);
3452                 goto done;
3453         }
3454         WARN_ON(diff > old_total);
3455         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3456         unlock_chunks(root);
3457         btrfs_end_transaction(trans, root);
3458 done:
3459         btrfs_free_path(path);
3460         return ret;
3461 }
3462
3463 static int btrfs_add_system_chunk(struct btrfs_root *root,
3464                            struct btrfs_key *key,
3465                            struct btrfs_chunk *chunk, int item_size)
3466 {
3467         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3468         struct btrfs_disk_key disk_key;
3469         u32 array_size;
3470         u8 *ptr;
3471
3472         array_size = btrfs_super_sys_array_size(super_copy);
3473         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3474                 return -EFBIG;
3475
3476         ptr = super_copy->sys_chunk_array + array_size;
3477         btrfs_cpu_key_to_disk(&disk_key, key);
3478         memcpy(ptr, &disk_key, sizeof(disk_key));
3479         ptr += sizeof(disk_key);
3480         memcpy(ptr, chunk, item_size);
3481         item_size += sizeof(disk_key);
3482         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3483         return 0;
3484 }
3485
3486 /*
3487  * sort the devices in descending order by max_avail, total_avail
3488  */
3489 static int btrfs_cmp_device_info(const void *a, const void *b)
3490 {
3491         const struct btrfs_device_info *di_a = a;
3492         const struct btrfs_device_info *di_b = b;
3493
3494         if (di_a->max_avail > di_b->max_avail)
3495                 return -1;
3496         if (di_a->max_avail < di_b->max_avail)
3497                 return 1;
3498         if (di_a->total_avail > di_b->total_avail)
3499                 return -1;
3500         if (di_a->total_avail < di_b->total_avail)
3501                 return 1;
3502         return 0;
3503 }
3504
3505 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3506         { 2, 1, 0, 4, 2, 2 /* raid10 */ },
3507         { 1, 1, 2, 2, 2, 2 /* raid1 */ },
3508         { 1, 2, 1, 1, 1, 2 /* dup */ },
3509         { 1, 1, 0, 2, 1, 1 /* raid0 */ },
3510         { 1, 1, 0, 1, 1, 1 /* single */ },
3511 };
3512
3513 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3514                                struct btrfs_root *extent_root,
3515                                struct map_lookup **map_ret,
3516                                u64 *num_bytes_out, u64 *stripe_size_out,
3517                                u64 start, u64 type)
3518 {
3519         struct btrfs_fs_info *info = extent_root->fs_info;
3520         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3521         struct list_head *cur;
3522         struct map_lookup *map = NULL;
3523         struct extent_map_tree *em_tree;
3524         struct extent_map *em;
3525         struct btrfs_device_info *devices_info = NULL;
3526         u64 total_avail;
3527         int num_stripes;        /* total number of stripes to allocate */
3528         int sub_stripes;        /* sub_stripes info for map */
3529         int dev_stripes;        /* stripes per dev */
3530         int devs_max;           /* max devs to use */
3531         int devs_min;           /* min devs needed */
3532         int devs_increment;     /* ndevs has to be a multiple of this */
3533         int ncopies;            /* how many copies to data has */
3534         int ret;
3535         u64 max_stripe_size;
3536         u64 max_chunk_size;
3537         u64 stripe_size;
3538         u64 num_bytes;
3539         int ndevs;
3540         int i;
3541         int j;
3542         int index;
3543
3544         BUG_ON(!alloc_profile_is_valid(type, 0));
3545
3546         if (list_empty(&fs_devices->alloc_list))
3547                 return -ENOSPC;
3548
3549         index = __get_raid_index(type);
3550
3551         sub_stripes = btrfs_raid_array[index].sub_stripes;
3552         dev_stripes = btrfs_raid_array[index].dev_stripes;
3553         devs_max = btrfs_raid_array[index].devs_max;
3554         devs_min = btrfs_raid_array[index].devs_min;
3555         devs_increment = btrfs_raid_array[index].devs_increment;
3556         ncopies = btrfs_raid_array[index].ncopies;
3557
3558         if (type & BTRFS_BLOCK_GROUP_DATA) {
3559                 max_stripe_size = 1024 * 1024 * 1024;
3560                 max_chunk_size = 10 * max_stripe_size;
3561         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3562                 /* for larger filesystems, use larger metadata chunks */
3563                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3564                         max_stripe_size = 1024 * 1024 * 1024;
3565                 else
3566                         max_stripe_size = 256 * 1024 * 1024;
3567                 max_chunk_size = max_stripe_size;
3568         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3569                 max_stripe_size = 32 * 1024 * 1024;
3570                 max_chunk_size = 2 * max_stripe_size;
3571         } else {
3572                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3573                        type);
3574                 BUG_ON(1);
3575         }
3576
3577         /* we don't want a chunk larger than 10% of writeable space */
3578         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3579                              max_chunk_size);
3580
3581         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3582                                GFP_NOFS);
3583         if (!devices_info)
3584                 return -ENOMEM;
3585
3586         cur = fs_devices->alloc_list.next;
3587
3588         /*
3589          * in the first pass through the devices list, we gather information
3590          * about the available holes on each device.
3591          */
3592         ndevs = 0;
3593         while (cur != &fs_devices->alloc_list) {
3594                 struct btrfs_device *device;
3595                 u64 max_avail;
3596                 u64 dev_offset;
3597
3598                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3599
3600                 cur = cur->next;
3601
3602                 if (!device->writeable) {
3603                         WARN(1, KERN_ERR
3604                                "btrfs: read-only device in alloc_list\n");
3605                         continue;
3606                 }
3607
3608                 if (!device->in_fs_metadata ||
3609                     device->is_tgtdev_for_dev_replace)
3610                         continue;
3611
3612                 if (device->total_bytes > device->bytes_used)
3613                         total_avail = device->total_bytes - device->bytes_used;
3614                 else
3615                         total_avail = 0;
3616
3617                 /* If there is no space on this device, skip it. */
3618                 if (total_avail == 0)
3619                         continue;
3620
3621                 ret = find_free_dev_extent(device,
3622                                            max_stripe_size * dev_stripes,
3623                                            &dev_offset, &max_avail);
3624                 if (ret && ret != -ENOSPC)
3625                         goto error;
3626
3627                 if (ret == 0)
3628                         max_avail = max_stripe_size * dev_stripes;
3629
3630                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3631                         continue;
3632
3633                 devices_info[ndevs].dev_offset = dev_offset;
3634                 devices_info[ndevs].max_avail = max_avail;
3635                 devices_info[ndevs].total_avail = total_avail;
3636                 devices_info[ndevs].dev = device;
3637                 ++ndevs;
3638                 WARN_ON(ndevs > fs_devices->rw_devices);
3639         }
3640
3641         /*
3642          * now sort the devices by hole size / available space
3643          */
3644         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3645              btrfs_cmp_device_info, NULL);
3646
3647         /* round down to number of usable stripes */
3648         ndevs -= ndevs % devs_increment;
3649
3650         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3651                 ret = -ENOSPC;
3652                 goto error;
3653         }
3654
3655         if (devs_max && ndevs > devs_max)
3656                 ndevs = devs_max;
3657         /*
3658          * the primary goal is to maximize the number of stripes, so use as many
3659          * devices as possible, even if the stripes are not maximum sized.
3660          */
3661         stripe_size = devices_info[ndevs-1].max_avail;
3662         num_stripes = ndevs * dev_stripes;
3663
3664         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3665                 stripe_size = max_chunk_size * ncopies;
3666                 do_div(stripe_size, ndevs);
3667         }
3668
3669         do_div(stripe_size, dev_stripes);
3670
3671         /* align to BTRFS_STRIPE_LEN */
3672         do_div(stripe_size, BTRFS_STRIPE_LEN);
3673         stripe_size *= BTRFS_STRIPE_LEN;
3674
3675         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3676         if (!map) {
3677                 ret = -ENOMEM;
3678                 goto error;
3679         }
3680         map->num_stripes = num_stripes;
3681
3682         for (i = 0; i < ndevs; ++i) {
3683                 for (j = 0; j < dev_stripes; ++j) {
3684                         int s = i * dev_stripes + j;
3685                         map->stripes[s].dev = devices_info[i].dev;
3686                         map->stripes[s].physical = devices_info[i].dev_offset +
3687                                                    j * stripe_size;
3688                 }
3689         }
3690         map->sector_size = extent_root->sectorsize;
3691         map->stripe_len = BTRFS_STRIPE_LEN;
3692         map->io_align = BTRFS_STRIPE_LEN;
3693         map->io_width = BTRFS_STRIPE_LEN;
3694         map->type = type;
3695         map->sub_stripes = sub_stripes;
3696
3697         *map_ret = map;
3698         num_bytes = stripe_size * (num_stripes / ncopies);
3699
3700         *stripe_size_out = stripe_size;
3701         *num_bytes_out = num_bytes;
3702
3703         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3704
3705         em = alloc_extent_map();
3706         if (!em) {
3707                 ret = -ENOMEM;
3708                 goto error;
3709         }
3710         em->bdev = (struct block_device *)map;
3711         em->start = start;
3712         em->len = num_bytes;
3713         em->block_start = 0;
3714         em->block_len = em->len;
3715
3716         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3717         write_lock(&em_tree->lock);
3718         ret = add_extent_mapping(em_tree, em);
3719         write_unlock(&em_tree->lock);
3720         free_extent_map(em);
3721         if (ret)
3722                 goto error;
3723
3724         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3725                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3726                                      start, num_bytes);
3727         if (ret)
3728                 goto error;
3729
3730         for (i = 0; i < map->num_stripes; ++i) {
3731                 struct btrfs_device *device;
3732                 u64 dev_offset;
3733
3734                 device = map->stripes[i].dev;
3735                 dev_offset = map->stripes[i].physical;
3736
3737                 ret = btrfs_alloc_dev_extent(trans, device,
3738                                 info->chunk_root->root_key.objectid,
3739                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3740                                 start, dev_offset, stripe_size);
3741                 if (ret) {
3742                         btrfs_abort_transaction(trans, extent_root, ret);
3743                         goto error;
3744                 }
3745         }
3746
3747         kfree(devices_info);
3748         return 0;
3749
3750 error:
3751         kfree(map);
3752         kfree(devices_info);
3753         return ret;
3754 }
3755
3756 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3757                                 struct btrfs_root *extent_root,
3758                                 struct map_lookup *map, u64 chunk_offset,
3759                                 u64 chunk_size, u64 stripe_size)
3760 {
3761         u64 dev_offset;
3762         struct btrfs_key key;
3763         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3764         struct btrfs_device *device;
3765         struct btrfs_chunk *chunk;
3766         struct btrfs_stripe *stripe;
3767         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3768         int index = 0;
3769         int ret;
3770
3771         chunk = kzalloc(item_size, GFP_NOFS);
3772         if (!chunk)
3773                 return -ENOMEM;
3774
3775         index = 0;
3776         while (index < map->num_stripes) {
3777                 device = map->stripes[index].dev;
3778                 device->bytes_used += stripe_size;
3779                 ret = btrfs_update_device(trans, device);
3780                 if (ret)
3781                         goto out_free;
3782                 index++;
3783         }
3784
3785         spin_lock(&extent_root->fs_info->free_chunk_lock);
3786         extent_root->fs_info->free_chunk_space -= (stripe_size *
3787                                                    map->num_stripes);
3788         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3789
3790         index = 0;
3791         stripe = &chunk->stripe;
3792         while (index < map->num_stripes) {
3793                 device = map->stripes[index].dev;
3794                 dev_offset = map->stripes[index].physical;
3795
3796                 btrfs_set_stack_stripe_devid(stripe, device->devid);
3797                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3798                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3799                 stripe++;
3800                 index++;
3801         }
3802
3803         btrfs_set_stack_chunk_length(chunk, chunk_size);
3804         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3805         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3806         btrfs_set_stack_chunk_type(chunk, map->type);
3807         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3808         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3809         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3810         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3811         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3812
3813         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3814         key.type = BTRFS_CHUNK_ITEM_KEY;
3815         key.offset = chunk_offset;
3816
3817         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3818
3819         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3820                 /*
3821                  * TODO: Cleanup of inserted chunk root in case of
3822                  * failure.
3823                  */
3824                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3825                                              item_size);
3826         }
3827
3828 out_free:
3829         kfree(chunk);
3830         return ret;
3831 }
3832
3833 /*
3834  * Chunk allocation falls into two parts. The first part does works
3835  * that make the new allocated chunk useable, but not do any operation
3836  * that modifies the chunk tree. The second part does the works that
3837  * require modifying the chunk tree. This division is important for the
3838  * bootstrap process of adding storage to a seed btrfs.
3839  */
3840 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3841                       struct btrfs_root *extent_root, u64 type)
3842 {
3843         u64 chunk_offset;
3844         u64 chunk_size;
3845         u64 stripe_size;
3846         struct map_lookup *map;
3847         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3848         int ret;
3849
3850         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3851                               &chunk_offset);
3852         if (ret)
3853                 return ret;
3854
3855         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3856                                   &stripe_size, chunk_offset, type);
3857         if (ret)
3858                 return ret;
3859
3860         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3861                                    chunk_size, stripe_size);
3862         if (ret)
3863                 return ret;
3864         return 0;
3865 }
3866
3867 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3868                                          struct btrfs_root *root,
3869                                          struct btrfs_device *device)
3870 {
3871         u64 chunk_offset;
3872         u64 sys_chunk_offset;
3873         u64 chunk_size;
3874         u64 sys_chunk_size;
3875         u64 stripe_size;
3876         u64 sys_stripe_size;
3877         u64 alloc_profile;
3878         struct map_lookup *map;
3879         struct map_lookup *sys_map;
3880         struct btrfs_fs_info *fs_info = root->fs_info;
3881         struct btrfs_root *extent_root = fs_info->extent_root;
3882         int ret;
3883
3884         ret = find_next_chunk(fs_info->chunk_root,
3885                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3886         if (ret)
3887                 return ret;
3888
3889         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3890                                 fs_info->avail_metadata_alloc_bits;
3891         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3892
3893         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3894                                   &stripe_size, chunk_offset, alloc_profile);
3895         if (ret)
3896                 return ret;
3897
3898         sys_chunk_offset = chunk_offset + chunk_size;
3899
3900         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3901                                 fs_info->avail_system_alloc_bits;
3902         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3903
3904         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3905                                   &sys_chunk_size, &sys_stripe_size,
3906                                   sys_chunk_offset, alloc_profile);
3907         if (ret) {
3908                 btrfs_abort_transaction(trans, root, ret);
3909                 goto out;
3910         }
3911
3912         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3913         if (ret) {
3914                 btrfs_abort_transaction(trans, root, ret);
3915                 goto out;
3916         }
3917
3918         /*
3919          * Modifying chunk tree needs allocating new blocks from both
3920          * system block group and metadata block group. So we only can
3921          * do operations require modifying the chunk tree after both
3922          * block groups were created.
3923          */
3924         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3925                                    chunk_size, stripe_size);
3926         if (ret) {
3927                 btrfs_abort_transaction(trans, root, ret);
3928                 goto out;
3929         }
3930
3931         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3932                                    sys_chunk_offset, sys_chunk_size,
3933                                    sys_stripe_size);
3934         if (ret)
3935                 btrfs_abort_transaction(trans, root, ret);
3936
3937 out:
3938
3939         return ret;
3940 }
3941
3942 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3943 {
3944         struct extent_map *em;
3945         struct map_lookup *map;
3946         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3947         int readonly = 0;
3948         int i;
3949
3950         read_lock(&map_tree->map_tree.lock);
3951         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3952         read_unlock(&map_tree->map_tree.lock);
3953         if (!em)
3954                 return 1;
3955
3956         if (btrfs_test_opt(root, DEGRADED)) {
3957                 free_extent_map(em);
3958                 return 0;
3959         }
3960
3961         map = (struct map_lookup *)em->bdev;
3962         for (i = 0; i < map->num_stripes; i++) {
3963                 if (!map->stripes[i].dev->writeable) {
3964                         readonly = 1;
3965                         break;
3966                 }
3967         }
3968         free_extent_map(em);
3969         return readonly;
3970 }
3971
3972 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3973 {
3974         extent_map_tree_init(&tree->map_tree);
3975 }
3976
3977 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3978 {
3979         struct extent_map *em;
3980
3981         while (1) {
3982                 write_lock(&tree->map_tree.lock);
3983                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3984                 if (em)
3985                         remove_extent_mapping(&tree->map_tree, em);
3986                 write_unlock(&tree->map_tree.lock);
3987                 if (!em)
3988                         break;
3989                 kfree(em->bdev);
3990                 /* once for us */
3991                 free_extent_map(em);
3992                 /* once for the tree */
3993                 free_extent_map(em);
3994         }
3995 }
3996
3997 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
3998 {
3999         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4000         struct extent_map *em;
4001         struct map_lookup *map;
4002         struct extent_map_tree *em_tree = &map_tree->map_tree;
4003         int ret;
4004
4005         read_lock(&em_tree->lock);
4006         em = lookup_extent_mapping(em_tree, logical, len);
4007         read_unlock(&em_tree->lock);
4008         BUG_ON(!em);
4009
4010         BUG_ON(em->start > logical || em->start + em->len < logical);
4011         map = (struct map_lookup *)em->bdev;
4012         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4013                 ret = map->num_stripes;
4014         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4015                 ret = map->sub_stripes;
4016         else
4017                 ret = 1;
4018         free_extent_map(em);
4019
4020         btrfs_dev_replace_lock(&fs_info->dev_replace);
4021         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4022                 ret++;
4023         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4024
4025         return ret;
4026 }
4027
4028 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4029                             struct map_lookup *map, int first, int num,
4030                             int optimal, int dev_replace_is_ongoing)
4031 {
4032         int i;
4033         int tolerance;
4034         struct btrfs_device *srcdev;
4035
4036         if (dev_replace_is_ongoing &&
4037             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4038              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4039                 srcdev = fs_info->dev_replace.srcdev;
4040         else
4041                 srcdev = NULL;
4042
4043         /*
4044          * try to avoid the drive that is the source drive for a
4045          * dev-replace procedure, only choose it if no other non-missing
4046          * mirror is available
4047          */
4048         for (tolerance = 0; tolerance < 2; tolerance++) {
4049                 if (map->stripes[optimal].dev->bdev &&
4050                     (tolerance || map->stripes[optimal].dev != srcdev))
4051                         return optimal;
4052                 for (i = first; i < first + num; i++) {
4053                         if (map->stripes[i].dev->bdev &&
4054                             (tolerance || map->stripes[i].dev != srcdev))
4055                                 return i;
4056                 }
4057         }
4058
4059         /* we couldn't find one that doesn't fail.  Just return something
4060          * and the io error handling code will clean up eventually
4061          */
4062         return optimal;
4063 }
4064
4065 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4066                              u64 logical, u64 *length,
4067                              struct btrfs_bio **bbio_ret,
4068                              int mirror_num)
4069 {
4070         struct extent_map *em;
4071         struct map_lookup *map;
4072         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4073         struct extent_map_tree *em_tree = &map_tree->map_tree;
4074         u64 offset;
4075         u64 stripe_offset;
4076         u64 stripe_end_offset;
4077         u64 stripe_nr;
4078         u64 stripe_nr_orig;
4079         u64 stripe_nr_end;
4080         int stripe_index;
4081         int i;
4082         int ret = 0;
4083         int num_stripes;
4084         int max_errors = 0;
4085         struct btrfs_bio *bbio = NULL;
4086         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4087         int dev_replace_is_ongoing = 0;
4088         int num_alloc_stripes;
4089         int patch_the_first_stripe_for_dev_replace = 0;
4090         u64 physical_to_patch_in_first_stripe = 0;
4091
4092         read_lock(&em_tree->lock);
4093         em = lookup_extent_mapping(em_tree, logical, *length);
4094         read_unlock(&em_tree->lock);
4095
4096         if (!em) {
4097                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4098                        (unsigned long long)logical,
4099                        (unsigned long long)*length);
4100                 BUG();
4101         }
4102
4103         BUG_ON(em->start > logical || em->start + em->len < logical);
4104         map = (struct map_lookup *)em->bdev;
4105         offset = logical - em->start;
4106
4107         stripe_nr = offset;
4108         /*
4109          * stripe_nr counts the total number of stripes we have to stride
4110          * to get to this block
4111          */
4112         do_div(stripe_nr, map->stripe_len);
4113
4114         stripe_offset = stripe_nr * map->stripe_len;
4115         BUG_ON(offset < stripe_offset);
4116
4117         /* stripe_offset is the offset of this block in its stripe*/
4118         stripe_offset = offset - stripe_offset;
4119
4120         if (rw & REQ_DISCARD)
4121                 *length = min_t(u64, em->len - offset, *length);
4122         else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4123                 /* we limit the length of each bio to what fits in a stripe */
4124                 *length = min_t(u64, em->len - offset,
4125                                 map->stripe_len - stripe_offset);
4126         } else {
4127                 *length = em->len - offset;
4128         }
4129
4130         if (!bbio_ret)
4131                 goto out;
4132
4133         btrfs_dev_replace_lock(dev_replace);
4134         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4135         if (!dev_replace_is_ongoing)
4136                 btrfs_dev_replace_unlock(dev_replace);
4137
4138         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4139             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4140             dev_replace->tgtdev != NULL) {
4141                 /*
4142                  * in dev-replace case, for repair case (that's the only
4143                  * case where the mirror is selected explicitly when
4144                  * calling btrfs_map_block), blocks left of the left cursor
4145                  * can also be read from the target drive.
4146                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4147                  * the last one to the array of stripes. For READ, it also
4148                  * needs to be supported using the same mirror number.
4149                  * If the requested block is not left of the left cursor,
4150                  * EIO is returned. This can happen because btrfs_num_copies()
4151                  * returns one more in the dev-replace case.
4152                  */
4153                 u64 tmp_length = *length;
4154                 struct btrfs_bio *tmp_bbio = NULL;
4155                 int tmp_num_stripes;
4156                 u64 srcdev_devid = dev_replace->srcdev->devid;
4157                 int index_srcdev = 0;
4158                 int found = 0;
4159                 u64 physical_of_found = 0;
4160
4161                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4162                              logical, &tmp_length, &tmp_bbio, 0);
4163                 if (ret) {
4164                         WARN_ON(tmp_bbio != NULL);
4165                         goto out;
4166                 }
4167
4168                 tmp_num_stripes = tmp_bbio->num_stripes;
4169                 if (mirror_num > tmp_num_stripes) {
4170                         /*
4171                          * REQ_GET_READ_MIRRORS does not contain this
4172                          * mirror, that means that the requested area
4173                          * is not left of the left cursor
4174                          */
4175                         ret = -EIO;
4176                         kfree(tmp_bbio);
4177                         goto out;
4178                 }
4179
4180                 /*
4181                  * process the rest of the function using the mirror_num
4182                  * of the source drive. Therefore look it up first.
4183                  * At the end, patch the device pointer to the one of the
4184                  * target drive.
4185                  */
4186                 for (i = 0; i < tmp_num_stripes; i++) {
4187                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4188                                 /*
4189                                  * In case of DUP, in order to keep it
4190                                  * simple, only add the mirror with the
4191                                  * lowest physical address
4192                                  */
4193                                 if (found &&
4194                                     physical_of_found <=
4195                                      tmp_bbio->stripes[i].physical)
4196                                         continue;
4197                                 index_srcdev = i;
4198                                 found = 1;
4199                                 physical_of_found =
4200                                         tmp_bbio->stripes[i].physical;
4201                         }
4202                 }
4203
4204                 if (found) {
4205                         mirror_num = index_srcdev + 1;
4206                         patch_the_first_stripe_for_dev_replace = 1;
4207                         physical_to_patch_in_first_stripe = physical_of_found;
4208                 } else {
4209                         WARN_ON(1);
4210                         ret = -EIO;
4211                         kfree(tmp_bbio);
4212                         goto out;
4213                 }
4214
4215                 kfree(tmp_bbio);
4216         } else if (mirror_num > map->num_stripes) {
4217                 mirror_num = 0;
4218         }
4219
4220         num_stripes = 1;
4221         stripe_index = 0;
4222         stripe_nr_orig = stripe_nr;
4223         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4224                         (~(map->stripe_len - 1));
4225         do_div(stripe_nr_end, map->stripe_len);
4226         stripe_end_offset = stripe_nr_end * map->stripe_len -
4227                             (offset + *length);
4228         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4229                 if (rw & REQ_DISCARD)
4230                         num_stripes = min_t(u64, map->num_stripes,
4231                                             stripe_nr_end - stripe_nr_orig);
4232                 stripe_index = do_div(stripe_nr, map->num_stripes);
4233         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4234                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4235                         num_stripes = map->num_stripes;
4236                 else if (mirror_num)
4237                         stripe_index = mirror_num - 1;
4238                 else {
4239                         stripe_index = find_live_mirror(fs_info, map, 0,
4240                                             map->num_stripes,
4241                                             current->pid % map->num_stripes,
4242                                             dev_replace_is_ongoing);
4243                         mirror_num = stripe_index + 1;
4244                 }
4245
4246         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4247                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4248                         num_stripes = map->num_stripes;
4249                 } else if (mirror_num) {
4250                         stripe_index = mirror_num - 1;
4251                 } else {
4252                         mirror_num = 1;
4253                 }
4254
4255         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4256                 int factor = map->num_stripes / map->sub_stripes;
4257
4258                 stripe_index = do_div(stripe_nr, factor);
4259                 stripe_index *= map->sub_stripes;
4260
4261                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4262                         num_stripes = map->sub_stripes;
4263                 else if (rw & REQ_DISCARD)
4264                         num_stripes = min_t(u64, map->sub_stripes *
4265                                             (stripe_nr_end - stripe_nr_orig),
4266                                             map->num_stripes);
4267                 else if (mirror_num)
4268                         stripe_index += mirror_num - 1;
4269                 else {
4270                         int old_stripe_index = stripe_index;
4271                         stripe_index = find_live_mirror(fs_info, map,
4272                                               stripe_index,
4273                                               map->sub_stripes, stripe_index +
4274                                               current->pid % map->sub_stripes,
4275                                               dev_replace_is_ongoing);
4276                         mirror_num = stripe_index - old_stripe_index + 1;
4277                 }
4278         } else {
4279                 /*
4280                  * after this do_div call, stripe_nr is the number of stripes
4281                  * on this device we have to walk to find the data, and
4282                  * stripe_index is the number of our device in the stripe array
4283                  */
4284                 stripe_index = do_div(stripe_nr, map->num_stripes);
4285                 mirror_num = stripe_index + 1;
4286         }
4287         BUG_ON(stripe_index >= map->num_stripes);
4288
4289         num_alloc_stripes = num_stripes;
4290         if (dev_replace_is_ongoing) {
4291                 if (rw & (REQ_WRITE | REQ_DISCARD))
4292                         num_alloc_stripes <<= 1;
4293                 if (rw & REQ_GET_READ_MIRRORS)
4294                         num_alloc_stripes++;
4295         }
4296         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4297         if (!bbio) {
4298                 ret = -ENOMEM;
4299                 goto out;
4300         }
4301         atomic_set(&bbio->error, 0);
4302
4303         if (rw & REQ_DISCARD) {
4304                 int factor = 0;
4305                 int sub_stripes = 0;
4306                 u64 stripes_per_dev = 0;
4307                 u32 remaining_stripes = 0;
4308                 u32 last_stripe = 0;
4309
4310                 if (map->type &
4311                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4312                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4313                                 sub_stripes = 1;
4314                         else
4315                                 sub_stripes = map->sub_stripes;
4316
4317                         factor = map->num_stripes / sub_stripes;
4318                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4319                                                       stripe_nr_orig,
4320                                                       factor,
4321                                                       &remaining_stripes);
4322                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4323                         last_stripe *= sub_stripes;
4324                 }
4325
4326                 for (i = 0; i < num_stripes; i++) {
4327                         bbio->stripes[i].physical =
4328                                 map->stripes[stripe_index].physical +
4329                                 stripe_offset + stripe_nr * map->stripe_len;
4330                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4331
4332                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4333                                          BTRFS_BLOCK_GROUP_RAID10)) {
4334                                 bbio->stripes[i].length = stripes_per_dev *
4335                                                           map->stripe_len;
4336
4337                                 if (i / sub_stripes < remaining_stripes)
4338                                         bbio->stripes[i].length +=
4339                                                 map->stripe_len;
4340
4341                                 /*
4342                                  * Special for the first stripe and
4343                                  * the last stripe:
4344                                  *
4345                                  * |-------|...|-------|
4346                                  *     |----------|
4347                                  *    off     end_off
4348                                  */
4349                                 if (i < sub_stripes)
4350                                         bbio->stripes[i].length -=
4351                                                 stripe_offset;
4352
4353                                 if (stripe_index >= last_stripe &&
4354                                     stripe_index <= (last_stripe +
4355                                                      sub_stripes - 1))
4356                                         bbio->stripes[i].length -=
4357                                                 stripe_end_offset;
4358
4359                                 if (i == sub_stripes - 1)
4360                                         stripe_offset = 0;
4361                         } else
4362                                 bbio->stripes[i].length = *length;
4363
4364                         stripe_index++;
4365                         if (stripe_index == map->num_stripes) {
4366                                 /* This could only happen for RAID0/10 */
4367                                 stripe_index = 0;
4368                                 stripe_nr++;
4369                         }
4370                 }
4371         } else {
4372                 for (i = 0; i < num_stripes; i++) {
4373                         bbio->stripes[i].physical =
4374                                 map->stripes[stripe_index].physical +
4375                                 stripe_offset +
4376                                 stripe_nr * map->stripe_len;
4377                         bbio->stripes[i].dev =
4378                                 map->stripes[stripe_index].dev;
4379                         stripe_index++;
4380                 }
4381         }
4382
4383         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4384                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4385                                  BTRFS_BLOCK_GROUP_RAID10 |
4386                                  BTRFS_BLOCK_GROUP_DUP)) {
4387                         max_errors = 1;
4388                 }
4389         }
4390
4391         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4392             dev_replace->tgtdev != NULL) {
4393                 int index_where_to_add;
4394                 u64 srcdev_devid = dev_replace->srcdev->devid;
4395
4396                 /*
4397                  * duplicate the write operations while the dev replace
4398                  * procedure is running. Since the copying of the old disk
4399                  * to the new disk takes place at run time while the
4400                  * filesystem is mounted writable, the regular write
4401                  * operations to the old disk have to be duplicated to go
4402                  * to the new disk as well.
4403                  * Note that device->missing is handled by the caller, and
4404                  * that the write to the old disk is already set up in the
4405                  * stripes array.
4406                  */
4407                 index_where_to_add = num_stripes;
4408                 for (i = 0; i < num_stripes; i++) {
4409                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4410                                 /* write to new disk, too */
4411                                 struct btrfs_bio_stripe *new =
4412                                         bbio->stripes + index_where_to_add;
4413                                 struct btrfs_bio_stripe *old =
4414                                         bbio->stripes + i;
4415
4416                                 new->physical = old->physical;
4417                                 new->length = old->length;
4418                                 new->dev = dev_replace->tgtdev;
4419                                 index_where_to_add++;
4420                                 max_errors++;
4421                         }
4422                 }
4423                 num_stripes = index_where_to_add;
4424         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4425                    dev_replace->tgtdev != NULL) {
4426                 u64 srcdev_devid = dev_replace->srcdev->devid;
4427                 int index_srcdev = 0;
4428                 int found = 0;
4429                 u64 physical_of_found = 0;
4430
4431                 /*
4432                  * During the dev-replace procedure, the target drive can
4433                  * also be used to read data in case it is needed to repair
4434                  * a corrupt block elsewhere. This is possible if the
4435                  * requested area is left of the left cursor. In this area,
4436                  * the target drive is a full copy of the source drive.
4437                  */
4438                 for (i = 0; i < num_stripes; i++) {
4439                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4440                                 /*
4441                                  * In case of DUP, in order to keep it
4442                                  * simple, only add the mirror with the
4443                                  * lowest physical address
4444                                  */
4445                                 if (found &&
4446                                     physical_of_found <=
4447                                      bbio->stripes[i].physical)
4448                                         continue;
4449                                 index_srcdev = i;
4450                                 found = 1;
4451                                 physical_of_found = bbio->stripes[i].physical;
4452                         }
4453                 }
4454                 if (found) {
4455                         u64 length = map->stripe_len;
4456
4457                         if (physical_of_found + length <=
4458                             dev_replace->cursor_left) {
4459                                 struct btrfs_bio_stripe *tgtdev_stripe =
4460                                         bbio->stripes + num_stripes;
4461
4462                                 tgtdev_stripe->physical = physical_of_found;
4463                                 tgtdev_stripe->length =
4464                                         bbio->stripes[index_srcdev].length;
4465                                 tgtdev_stripe->dev = dev_replace->tgtdev;
4466
4467                                 num_stripes++;
4468                         }
4469                 }
4470         }
4471
4472         *bbio_ret = bbio;
4473         bbio->num_stripes = num_stripes;
4474         bbio->max_errors = max_errors;
4475         bbio->mirror_num = mirror_num;
4476
4477         /*
4478          * this is the case that REQ_READ && dev_replace_is_ongoing &&
4479          * mirror_num == num_stripes + 1 && dev_replace target drive is
4480          * available as a mirror
4481          */
4482         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4483                 WARN_ON(num_stripes > 1);
4484                 bbio->stripes[0].dev = dev_replace->tgtdev;
4485                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4486                 bbio->mirror_num = map->num_stripes + 1;
4487         }
4488 out:
4489         if (dev_replace_is_ongoing)
4490                 btrfs_dev_replace_unlock(dev_replace);
4491         free_extent_map(em);
4492         return ret;
4493 }
4494
4495 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4496                       u64 logical, u64 *length,
4497                       struct btrfs_bio **bbio_ret, int mirror_num)
4498 {
4499         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4500                                  mirror_num);
4501 }
4502
4503 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4504                      u64 chunk_start, u64 physical, u64 devid,
4505                      u64 **logical, int *naddrs, int *stripe_len)
4506 {
4507         struct extent_map_tree *em_tree = &map_tree->map_tree;
4508         struct extent_map *em;
4509         struct map_lookup *map;
4510         u64 *buf;
4511         u64 bytenr;
4512         u64 length;
4513         u64 stripe_nr;
4514         int i, j, nr = 0;
4515
4516         read_lock(&em_tree->lock);
4517         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4518         read_unlock(&em_tree->lock);
4519
4520         BUG_ON(!em || em->start != chunk_start);
4521         map = (struct map_lookup *)em->bdev;
4522
4523         length = em->len;
4524         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4525                 do_div(length, map->num_stripes / map->sub_stripes);
4526         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4527                 do_div(length, map->num_stripes);
4528
4529         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4530         BUG_ON(!buf); /* -ENOMEM */
4531
4532         for (i = 0; i < map->num_stripes; i++) {
4533                 if (devid && map->stripes[i].dev->devid != devid)
4534                         continue;
4535                 if (map->stripes[i].physical > physical ||
4536                     map->stripes[i].physical + length <= physical)
4537                         continue;
4538
4539                 stripe_nr = physical - map->stripes[i].physical;
4540                 do_div(stripe_nr, map->stripe_len);
4541
4542                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4543                         stripe_nr = stripe_nr * map->num_stripes + i;
4544                         do_div(stripe_nr, map->sub_stripes);
4545                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4546                         stripe_nr = stripe_nr * map->num_stripes + i;
4547                 }
4548                 bytenr = chunk_start + stripe_nr * map->stripe_len;
4549                 WARN_ON(nr >= map->num_stripes);
4550                 for (j = 0; j < nr; j++) {
4551                         if (buf[j] == bytenr)
4552                                 break;
4553                 }
4554                 if (j == nr) {
4555                         WARN_ON(nr >= map->num_stripes);
4556                         buf[nr++] = bytenr;
4557                 }
4558         }
4559
4560         *logical = buf;
4561         *naddrs = nr;
4562         *stripe_len = map->stripe_len;
4563
4564         free_extent_map(em);
4565         return 0;
4566 }
4567
4568 static void *merge_stripe_index_into_bio_private(void *bi_private,
4569                                                  unsigned int stripe_index)
4570 {
4571         /*
4572          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4573          * at most 1.
4574          * The alternative solution (instead of stealing bits from the
4575          * pointer) would be to allocate an intermediate structure
4576          * that contains the old private pointer plus the stripe_index.
4577          */
4578         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4579         BUG_ON(stripe_index > 3);
4580         return (void *)(((uintptr_t)bi_private) | stripe_index);
4581 }
4582
4583 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4584 {
4585         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4586 }
4587
4588 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4589 {
4590         return (unsigned int)((uintptr_t)bi_private) & 3;
4591 }
4592
4593 static void btrfs_end_bio(struct bio *bio, int err)
4594 {
4595         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4596         int is_orig_bio = 0;
4597
4598         if (err) {
4599                 atomic_inc(&bbio->error);
4600                 if (err == -EIO || err == -EREMOTEIO) {
4601                         unsigned int stripe_index =
4602                                 extract_stripe_index_from_bio_private(
4603                                         bio->bi_private);
4604                         struct btrfs_device *dev;
4605
4606                         BUG_ON(stripe_index >= bbio->num_stripes);
4607                         dev = bbio->stripes[stripe_index].dev;
4608                         if (dev->bdev) {
4609                                 if (bio->bi_rw & WRITE)
4610                                         btrfs_dev_stat_inc(dev,
4611                                                 BTRFS_DEV_STAT_WRITE_ERRS);
4612                                 else
4613                                         btrfs_dev_stat_inc(dev,
4614                                                 BTRFS_DEV_STAT_READ_ERRS);
4615                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4616                                         btrfs_dev_stat_inc(dev,
4617                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
4618                                 btrfs_dev_stat_print_on_error(dev);
4619                         }
4620                 }
4621         }
4622
4623         if (bio == bbio->orig_bio)
4624                 is_orig_bio = 1;
4625
4626         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4627                 if (!is_orig_bio) {
4628                         bio_put(bio);
4629                         bio = bbio->orig_bio;
4630                 }
4631                 bio->bi_private = bbio->private;
4632                 bio->bi_end_io = bbio->end_io;
4633                 bio->bi_bdev = (struct block_device *)
4634                                         (unsigned long)bbio->mirror_num;
4635                 /* only send an error to the higher layers if it is
4636                  * beyond the tolerance of the multi-bio
4637                  */
4638                 if (atomic_read(&bbio->error) > bbio->max_errors) {
4639                         err = -EIO;
4640                 } else {
4641                         /*
4642                          * this bio is actually up to date, we didn't
4643                          * go over the max number of errors
4644                          */
4645                         set_bit(BIO_UPTODATE, &bio->bi_flags);
4646                         err = 0;
4647                 }
4648                 kfree(bbio);
4649
4650                 bio_endio(bio, err);
4651         } else if (!is_orig_bio) {
4652                 bio_put(bio);
4653         }
4654 }
4655
4656 struct async_sched {
4657         struct bio *bio;
4658         int rw;
4659         struct btrfs_fs_info *info;
4660         struct btrfs_work work;
4661 };
4662
4663 /*
4664  * see run_scheduled_bios for a description of why bios are collected for
4665  * async submit.
4666  *
4667  * This will add one bio to the pending list for a device and make sure
4668  * the work struct is scheduled.
4669  */
4670 static noinline void schedule_bio(struct btrfs_root *root,
4671                                  struct btrfs_device *device,
4672                                  int rw, struct bio *bio)
4673 {
4674         int should_queue = 1;
4675         struct btrfs_pending_bios *pending_bios;
4676
4677         /* don't bother with additional async steps for reads, right now */
4678         if (!(rw & REQ_WRITE)) {
4679                 bio_get(bio);
4680                 btrfsic_submit_bio(rw, bio);
4681                 bio_put(bio);
4682                 return;
4683         }
4684
4685         /*
4686          * nr_async_bios allows us to reliably return congestion to the
4687          * higher layers.  Otherwise, the async bio makes it appear we have
4688          * made progress against dirty pages when we've really just put it
4689          * on a queue for later
4690          */
4691         atomic_inc(&root->fs_info->nr_async_bios);
4692         WARN_ON(bio->bi_next);
4693         bio->bi_next = NULL;
4694         bio->bi_rw |= rw;
4695
4696         spin_lock(&device->io_lock);
4697         if (bio->bi_rw & REQ_SYNC)
4698                 pending_bios = &device->pending_sync_bios;
4699         else
4700                 pending_bios = &device->pending_bios;
4701
4702         if (pending_bios->tail)
4703                 pending_bios->tail->bi_next = bio;
4704
4705         pending_bios->tail = bio;
4706         if (!pending_bios->head)
4707                 pending_bios->head = bio;
4708         if (device->running_pending)
4709                 should_queue = 0;
4710
4711         spin_unlock(&device->io_lock);
4712
4713         if (should_queue)
4714                 btrfs_queue_worker(&root->fs_info->submit_workers,
4715                                    &device->work);
4716 }
4717
4718 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4719                        sector_t sector)
4720 {
4721         struct bio_vec *prev;
4722         struct request_queue *q = bdev_get_queue(bdev);
4723         unsigned short max_sectors = queue_max_sectors(q);
4724         struct bvec_merge_data bvm = {
4725                 .bi_bdev = bdev,
4726                 .bi_sector = sector,
4727                 .bi_rw = bio->bi_rw,
4728         };
4729
4730         if (bio->bi_vcnt == 0) {
4731                 WARN_ON(1);
4732                 return 1;
4733         }
4734
4735         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4736         if ((bio->bi_size >> 9) > max_sectors)
4737                 return 0;
4738
4739         if (!q->merge_bvec_fn)
4740                 return 1;
4741
4742         bvm.bi_size = bio->bi_size - prev->bv_len;
4743         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4744                 return 0;
4745         return 1;
4746 }
4747
4748 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4749                               struct bio *bio, u64 physical, int dev_nr,
4750                               int rw, int async)
4751 {
4752         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4753
4754         bio->bi_private = bbio;
4755         bio->bi_private = merge_stripe_index_into_bio_private(
4756                         bio->bi_private, (unsigned int)dev_nr);
4757         bio->bi_end_io = btrfs_end_bio;
4758         bio->bi_sector = physical >> 9;
4759 #ifdef DEBUG
4760         {
4761                 struct rcu_string *name;
4762
4763                 rcu_read_lock();
4764                 name = rcu_dereference(dev->name);
4765                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4766                          "(%s id %llu), size=%u\n", rw,
4767                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4768                          name->str, dev->devid, bio->bi_size);
4769                 rcu_read_unlock();
4770         }
4771 #endif
4772         bio->bi_bdev = dev->bdev;
4773         if (async)
4774                 schedule_bio(root, dev, rw, bio);
4775         else
4776                 btrfsic_submit_bio(rw, bio);
4777 }
4778
4779 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4780                               struct bio *first_bio, struct btrfs_device *dev,
4781                               int dev_nr, int rw, int async)
4782 {
4783         struct bio_vec *bvec = first_bio->bi_io_vec;
4784         struct bio *bio;
4785         int nr_vecs = bio_get_nr_vecs(dev->bdev);
4786         u64 physical = bbio->stripes[dev_nr].physical;
4787
4788 again:
4789         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4790         if (!bio)
4791                 return -ENOMEM;
4792
4793         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4794                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4795                                  bvec->bv_offset) < bvec->bv_len) {
4796                         u64 len = bio->bi_size;
4797
4798                         atomic_inc(&bbio->stripes_pending);
4799                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4800                                           rw, async);
4801                         physical += len;
4802                         goto again;
4803                 }
4804                 bvec++;
4805         }
4806
4807         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4808         return 0;
4809 }
4810
4811 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4812 {
4813         atomic_inc(&bbio->error);
4814         if (atomic_dec_and_test(&bbio->stripes_pending)) {
4815                 bio->bi_private = bbio->private;
4816                 bio->bi_end_io = bbio->end_io;
4817                 bio->bi_bdev = (struct block_device *)
4818                         (unsigned long)bbio->mirror_num;
4819                 bio->bi_sector = logical >> 9;
4820                 kfree(bbio);
4821                 bio_endio(bio, -EIO);
4822         }
4823 }
4824
4825 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4826                   int mirror_num, int async_submit)
4827 {
4828         struct btrfs_device *dev;
4829         struct bio *first_bio = bio;
4830         u64 logical = (u64)bio->bi_sector << 9;
4831         u64 length = 0;
4832         u64 map_length;
4833         int ret;
4834         int dev_nr = 0;
4835         int total_devs = 1;
4836         struct btrfs_bio *bbio = NULL;
4837
4838         length = bio->bi_size;
4839         map_length = length;
4840
4841         ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
4842                               mirror_num);
4843         if (ret)
4844                 return ret;
4845
4846         total_devs = bbio->num_stripes;
4847         if (map_length < length) {
4848                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4849                        "len %llu\n", (unsigned long long)logical,
4850                        (unsigned long long)length,
4851                        (unsigned long long)map_length);
4852                 BUG();
4853         }
4854
4855         bbio->orig_bio = first_bio;
4856         bbio->private = first_bio->bi_private;
4857         bbio->end_io = first_bio->bi_end_io;
4858         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4859
4860         while (dev_nr < total_devs) {
4861                 dev = bbio->stripes[dev_nr].dev;
4862                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4863                         bbio_error(bbio, first_bio, logical);
4864                         dev_nr++;
4865                         continue;
4866                 }
4867
4868                 /*
4869                  * Check and see if we're ok with this bio based on it's size
4870                  * and offset with the given device.
4871                  */
4872                 if (!bio_size_ok(dev->bdev, first_bio,
4873                                  bbio->stripes[dev_nr].physical >> 9)) {
4874                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4875                                                  dev_nr, rw, async_submit);
4876                         BUG_ON(ret);
4877                         dev_nr++;
4878                         continue;
4879                 }
4880
4881                 if (dev_nr < total_devs - 1) {
4882                         bio = bio_clone(first_bio, GFP_NOFS);
4883                         BUG_ON(!bio); /* -ENOMEM */
4884                 } else {
4885                         bio = first_bio;
4886                 }
4887
4888                 submit_stripe_bio(root, bbio, bio,
4889                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
4890                                   async_submit);
4891                 dev_nr++;
4892         }
4893         return 0;
4894 }
4895
4896 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
4897                                        u8 *uuid, u8 *fsid)
4898 {
4899         struct btrfs_device *device;
4900         struct btrfs_fs_devices *cur_devices;
4901
4902         cur_devices = fs_info->fs_devices;
4903         while (cur_devices) {
4904                 if (!fsid ||
4905                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4906                         device = __find_device(&cur_devices->devices,
4907                                                devid, uuid);
4908                         if (device)
4909                                 return device;
4910                 }
4911                 cur_devices = cur_devices->seed;
4912         }
4913         return NULL;
4914 }
4915
4916 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4917                                             u64 devid, u8 *dev_uuid)
4918 {
4919         struct btrfs_device *device;
4920         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4921
4922         device = kzalloc(sizeof(*device), GFP_NOFS);
4923         if (!device)
4924                 return NULL;
4925         list_add(&device->dev_list,
4926                  &fs_devices->devices);
4927         device->dev_root = root->fs_info->dev_root;
4928         device->devid = devid;
4929         device->work.func = pending_bios_fn;
4930         device->fs_devices = fs_devices;
4931         device->missing = 1;
4932         fs_devices->num_devices++;
4933         fs_devices->missing_devices++;
4934         spin_lock_init(&device->io_lock);
4935         INIT_LIST_HEAD(&device->dev_alloc_list);
4936         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4937         return device;
4938 }
4939
4940 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4941                           struct extent_buffer *leaf,
4942                           struct btrfs_chunk *chunk)
4943 {
4944         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4945         struct map_lookup *map;
4946         struct extent_map *em;
4947         u64 logical;
4948         u64 length;
4949         u64 devid;
4950         u8 uuid[BTRFS_UUID_SIZE];
4951         int num_stripes;
4952         int ret;
4953         int i;
4954
4955         logical = key->offset;
4956         length = btrfs_chunk_length(leaf, chunk);
4957
4958         read_lock(&map_tree->map_tree.lock);
4959         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4960         read_unlock(&map_tree->map_tree.lock);
4961
4962         /* already mapped? */
4963         if (em && em->start <= logical && em->start + em->len > logical) {
4964                 free_extent_map(em);
4965                 return 0;
4966         } else if (em) {
4967                 free_extent_map(em);
4968         }
4969
4970         em = alloc_extent_map();
4971         if (!em)
4972                 return -ENOMEM;
4973         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4974         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4975         if (!map) {
4976                 free_extent_map(em);
4977                 return -ENOMEM;
4978         }
4979
4980         em->bdev = (struct block_device *)map;
4981         em->start = logical;
4982         em->len = length;
4983         em->orig_start = 0;
4984         em->block_start = 0;
4985         em->block_len = em->len;
4986
4987         map->num_stripes = num_stripes;
4988         map->io_width = btrfs_chunk_io_width(leaf, chunk);
4989         map->io_align = btrfs_chunk_io_align(leaf, chunk);
4990         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4991         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4992         map->type = btrfs_chunk_type(leaf, chunk);
4993         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4994         for (i = 0; i < num_stripes; i++) {
4995                 map->stripes[i].physical =
4996                         btrfs_stripe_offset_nr(leaf, chunk, i);
4997                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4998                 read_extent_buffer(leaf, uuid, (unsigned long)
4999                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5000                                    BTRFS_UUID_SIZE);
5001                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5002                                                         uuid, NULL);
5003                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5004                         kfree(map);
5005                         free_extent_map(em);
5006                         return -EIO;
5007                 }
5008                 if (!map->stripes[i].dev) {
5009                         map->stripes[i].dev =
5010                                 add_missing_dev(root, devid, uuid);
5011                         if (!map->stripes[i].dev) {
5012                                 kfree(map);
5013                                 free_extent_map(em);
5014                                 return -EIO;
5015                         }
5016                 }
5017                 map->stripes[i].dev->in_fs_metadata = 1;
5018         }
5019
5020         write_lock(&map_tree->map_tree.lock);
5021         ret = add_extent_mapping(&map_tree->map_tree, em);
5022         write_unlock(&map_tree->map_tree.lock);
5023         BUG_ON(ret); /* Tree corruption */
5024         free_extent_map(em);
5025
5026         return 0;
5027 }
5028
5029 static void fill_device_from_item(struct extent_buffer *leaf,
5030                                  struct btrfs_dev_item *dev_item,
5031                                  struct btrfs_device *device)
5032 {
5033         unsigned long ptr;
5034
5035         device->devid = btrfs_device_id(leaf, dev_item);
5036         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5037         device->total_bytes = device->disk_total_bytes;
5038         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5039         device->type = btrfs_device_type(leaf, dev_item);
5040         device->io_align = btrfs_device_io_align(leaf, dev_item);
5041         device->io_width = btrfs_device_io_width(leaf, dev_item);
5042         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5043         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5044         device->is_tgtdev_for_dev_replace = 0;
5045
5046         ptr = (unsigned long)btrfs_device_uuid(dev_item);
5047         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5048 }
5049
5050 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5051 {
5052         struct btrfs_fs_devices *fs_devices;
5053         int ret;
5054
5055         BUG_ON(!mutex_is_locked(&uuid_mutex));
5056
5057         fs_devices = root->fs_info->fs_devices->seed;
5058         while (fs_devices) {
5059                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5060                         ret = 0;
5061                         goto out;
5062                 }
5063                 fs_devices = fs_devices->seed;
5064         }
5065
5066         fs_devices = find_fsid(fsid);
5067         if (!fs_devices) {
5068                 ret = -ENOENT;
5069                 goto out;
5070         }
5071
5072         fs_devices = clone_fs_devices(fs_devices);
5073         if (IS_ERR(fs_devices)) {
5074                 ret = PTR_ERR(fs_devices);
5075                 goto out;
5076         }
5077
5078         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5079                                    root->fs_info->bdev_holder);
5080         if (ret) {
5081                 free_fs_devices(fs_devices);
5082                 goto out;
5083         }
5084
5085         if (!fs_devices->seeding) {
5086                 __btrfs_close_devices(fs_devices);
5087                 free_fs_devices(fs_devices);
5088                 ret = -EINVAL;
5089                 goto out;
5090         }
5091
5092         fs_devices->seed = root->fs_info->fs_devices->seed;
5093         root->fs_info->fs_devices->seed = fs_devices;
5094 out:
5095         return ret;
5096 }
5097
5098 static int read_one_dev(struct btrfs_root *root,
5099                         struct extent_buffer *leaf,
5100                         struct btrfs_dev_item *dev_item)
5101 {
5102         struct btrfs_device *device;
5103         u64 devid;
5104         int ret;
5105         u8 fs_uuid[BTRFS_UUID_SIZE];
5106         u8 dev_uuid[BTRFS_UUID_SIZE];
5107
5108         devid = btrfs_device_id(leaf, dev_item);
5109         read_extent_buffer(leaf, dev_uuid,
5110                            (unsigned long)btrfs_device_uuid(dev_item),
5111                            BTRFS_UUID_SIZE);
5112         read_extent_buffer(leaf, fs_uuid,
5113                            (unsigned long)btrfs_device_fsid(dev_item),
5114                            BTRFS_UUID_SIZE);
5115
5116         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5117                 ret = open_seed_devices(root, fs_uuid);
5118                 if (ret && !btrfs_test_opt(root, DEGRADED))
5119                         return ret;
5120         }
5121
5122         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5123         if (!device || !device->bdev) {
5124                 if (!btrfs_test_opt(root, DEGRADED))
5125                         return -EIO;
5126
5127                 if (!device) {
5128                         printk(KERN_WARNING "warning devid %llu missing\n",
5129                                (unsigned long long)devid);
5130                         device = add_missing_dev(root, devid, dev_uuid);
5131                         if (!device)
5132                                 return -ENOMEM;
5133                 } else if (!device->missing) {
5134                         /*
5135                          * this happens when a device that was properly setup
5136                          * in the device info lists suddenly goes bad.
5137                          * device->bdev is NULL, and so we have to set
5138                          * device->missing to one here
5139                          */
5140                         root->fs_info->fs_devices->missing_devices++;
5141                         device->missing = 1;
5142                 }
5143         }
5144
5145         if (device->fs_devices != root->fs_info->fs_devices) {
5146                 BUG_ON(device->writeable);
5147                 if (device->generation !=
5148                     btrfs_device_generation(leaf, dev_item))
5149                         return -EINVAL;
5150         }
5151
5152         fill_device_from_item(leaf, dev_item, device);
5153         device->dev_root = root->fs_info->dev_root;
5154         device->in_fs_metadata = 1;
5155         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5156                 device->fs_devices->total_rw_bytes += device->total_bytes;
5157                 spin_lock(&root->fs_info->free_chunk_lock);
5158                 root->fs_info->free_chunk_space += device->total_bytes -
5159                         device->bytes_used;
5160                 spin_unlock(&root->fs_info->free_chunk_lock);
5161         }
5162         ret = 0;
5163         return ret;
5164 }
5165
5166 int btrfs_read_sys_array(struct btrfs_root *root)
5167 {
5168         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5169         struct extent_buffer *sb;
5170         struct btrfs_disk_key *disk_key;
5171         struct btrfs_chunk *chunk;
5172         u8 *ptr;
5173         unsigned long sb_ptr;
5174         int ret = 0;
5175         u32 num_stripes;
5176         u32 array_size;
5177         u32 len = 0;
5178         u32 cur;
5179         struct btrfs_key key;
5180
5181         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5182                                           BTRFS_SUPER_INFO_SIZE);
5183         if (!sb)
5184                 return -ENOMEM;
5185         btrfs_set_buffer_uptodate(sb);
5186         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5187         /*
5188          * The sb extent buffer is artifical and just used to read the system array.
5189          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5190          * pages up-to-date when the page is larger: extent does not cover the
5191          * whole page and consequently check_page_uptodate does not find all
5192          * the page's extents up-to-date (the hole beyond sb),
5193          * write_extent_buffer then triggers a WARN_ON.
5194          *
5195          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5196          * but sb spans only this function. Add an explicit SetPageUptodate call
5197          * to silence the warning eg. on PowerPC 64.
5198          */
5199         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5200                 SetPageUptodate(sb->pages[0]);
5201
5202         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5203         array_size = btrfs_super_sys_array_size(super_copy);
5204
5205         ptr = super_copy->sys_chunk_array;
5206         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5207         cur = 0;
5208
5209         while (cur < array_size) {
5210                 disk_key = (struct btrfs_disk_key *)ptr;
5211                 btrfs_disk_key_to_cpu(&key, disk_key);
5212
5213                 len = sizeof(*disk_key); ptr += len;
5214                 sb_ptr += len;
5215                 cur += len;
5216
5217                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5218                         chunk = (struct btrfs_chunk *)sb_ptr;
5219                         ret = read_one_chunk(root, &key, sb, chunk);
5220                         if (ret)
5221                                 break;
5222                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5223                         len = btrfs_chunk_item_size(num_stripes);
5224                 } else {
5225                         ret = -EIO;
5226                         break;
5227                 }
5228                 ptr += len;
5229                 sb_ptr += len;
5230                 cur += len;
5231         }
5232         free_extent_buffer(sb);
5233         return ret;
5234 }
5235
5236 int btrfs_read_chunk_tree(struct btrfs_root *root)
5237 {
5238         struct btrfs_path *path;
5239         struct extent_buffer *leaf;
5240         struct btrfs_key key;
5241         struct btrfs_key found_key;
5242         int ret;
5243         int slot;
5244
5245         root = root->fs_info->chunk_root;
5246
5247         path = btrfs_alloc_path();
5248         if (!path)
5249                 return -ENOMEM;
5250
5251         mutex_lock(&uuid_mutex);
5252         lock_chunks(root);
5253
5254         /* first we search for all of the device items, and then we
5255          * read in all of the chunk items.  This way we can create chunk
5256          * mappings that reference all of the devices that are afound
5257          */
5258         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5259         key.offset = 0;
5260         key.type = 0;
5261 again:
5262         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5263         if (ret < 0)
5264                 goto error;
5265         while (1) {
5266                 leaf = path->nodes[0];
5267                 slot = path->slots[0];
5268                 if (slot >= btrfs_header_nritems(leaf)) {
5269                         ret = btrfs_next_leaf(root, path);
5270                         if (ret == 0)
5271                                 continue;
5272                         if (ret < 0)
5273                                 goto error;
5274                         break;
5275                 }
5276                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5277                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5278                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5279                                 break;
5280                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5281                                 struct btrfs_dev_item *dev_item;
5282                                 dev_item = btrfs_item_ptr(leaf, slot,
5283                                                   struct btrfs_dev_item);
5284                                 ret = read_one_dev(root, leaf, dev_item);
5285                                 if (ret)
5286                                         goto error;
5287                         }
5288                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5289                         struct btrfs_chunk *chunk;
5290                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5291                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5292                         if (ret)
5293                                 goto error;
5294                 }
5295                 path->slots[0]++;
5296         }
5297         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5298                 key.objectid = 0;
5299                 btrfs_release_path(path);
5300                 goto again;
5301         }
5302         ret = 0;
5303 error:
5304         unlock_chunks(root);
5305         mutex_unlock(&uuid_mutex);
5306
5307         btrfs_free_path(path);
5308         return ret;
5309 }
5310
5311 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5312 {
5313         int i;
5314
5315         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5316                 btrfs_dev_stat_reset(dev, i);
5317 }
5318
5319 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5320 {
5321         struct btrfs_key key;
5322         struct btrfs_key found_key;
5323         struct btrfs_root *dev_root = fs_info->dev_root;
5324         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5325         struct extent_buffer *eb;
5326         int slot;
5327         int ret = 0;
5328         struct btrfs_device *device;
5329         struct btrfs_path *path = NULL;
5330         int i;
5331
5332         path = btrfs_alloc_path();
5333         if (!path) {
5334                 ret = -ENOMEM;
5335                 goto out;
5336         }
5337
5338         mutex_lock(&fs_devices->device_list_mutex);
5339         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5340                 int item_size;
5341                 struct btrfs_dev_stats_item *ptr;
5342
5343                 key.objectid = 0;
5344                 key.type = BTRFS_DEV_STATS_KEY;
5345                 key.offset = device->devid;
5346                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5347                 if (ret) {
5348                         __btrfs_reset_dev_stats(device);
5349                         device->dev_stats_valid = 1;
5350                         btrfs_release_path(path);
5351                         continue;
5352                 }
5353                 slot = path->slots[0];
5354                 eb = path->nodes[0];
5355                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5356                 item_size = btrfs_item_size_nr(eb, slot);
5357
5358                 ptr = btrfs_item_ptr(eb, slot,
5359                                      struct btrfs_dev_stats_item);
5360
5361                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5362                         if (item_size >= (1 + i) * sizeof(__le64))
5363                                 btrfs_dev_stat_set(device, i,
5364                                         btrfs_dev_stats_value(eb, ptr, i));
5365                         else
5366                                 btrfs_dev_stat_reset(device, i);
5367                 }
5368
5369                 device->dev_stats_valid = 1;
5370                 btrfs_dev_stat_print_on_load(device);
5371                 btrfs_release_path(path);
5372         }
5373         mutex_unlock(&fs_devices->device_list_mutex);
5374
5375 out:
5376         btrfs_free_path(path);
5377         return ret < 0 ? ret : 0;
5378 }
5379
5380 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5381                                 struct btrfs_root *dev_root,
5382                                 struct btrfs_device *device)
5383 {
5384         struct btrfs_path *path;
5385         struct btrfs_key key;
5386         struct extent_buffer *eb;
5387         struct btrfs_dev_stats_item *ptr;
5388         int ret;
5389         int i;
5390
5391         key.objectid = 0;
5392         key.type = BTRFS_DEV_STATS_KEY;
5393         key.offset = device->devid;
5394
5395         path = btrfs_alloc_path();
5396         BUG_ON(!path);
5397         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5398         if (ret < 0) {
5399                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5400                               ret, rcu_str_deref(device->name));
5401                 goto out;
5402         }
5403
5404         if (ret == 0 &&
5405             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5406                 /* need to delete old one and insert a new one */
5407                 ret = btrfs_del_item(trans, dev_root, path);
5408                 if (ret != 0) {
5409                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5410                                       rcu_str_deref(device->name), ret);
5411                         goto out;
5412                 }
5413                 ret = 1;
5414         }
5415
5416         if (ret == 1) {
5417                 /* need to insert a new item */
5418                 btrfs_release_path(path);
5419                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5420                                               &key, sizeof(*ptr));
5421                 if (ret < 0) {
5422                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5423                                       rcu_str_deref(device->name), ret);
5424                         goto out;
5425                 }
5426         }
5427
5428         eb = path->nodes[0];
5429         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5430         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5431                 btrfs_set_dev_stats_value(eb, ptr, i,
5432                                           btrfs_dev_stat_read(device, i));
5433         btrfs_mark_buffer_dirty(eb);
5434
5435 out:
5436         btrfs_free_path(path);
5437         return ret;
5438 }
5439
5440 /*
5441  * called from commit_transaction. Writes all changed device stats to disk.
5442  */
5443 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5444                         struct btrfs_fs_info *fs_info)
5445 {
5446         struct btrfs_root *dev_root = fs_info->dev_root;
5447         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5448         struct btrfs_device *device;
5449         int ret = 0;
5450
5451         mutex_lock(&fs_devices->device_list_mutex);
5452         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5453                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5454                         continue;
5455
5456                 ret = update_dev_stat_item(trans, dev_root, device);
5457                 if (!ret)
5458                         device->dev_stats_dirty = 0;
5459         }
5460         mutex_unlock(&fs_devices->device_list_mutex);
5461
5462         return ret;
5463 }
5464
5465 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5466 {
5467         btrfs_dev_stat_inc(dev, index);
5468         btrfs_dev_stat_print_on_error(dev);
5469 }
5470
5471 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5472 {
5473         if (!dev->dev_stats_valid)
5474                 return;
5475         printk_ratelimited_in_rcu(KERN_ERR
5476                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5477                            rcu_str_deref(dev->name),
5478                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5479                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5480                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5481                            btrfs_dev_stat_read(dev,
5482                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5483                            btrfs_dev_stat_read(dev,
5484                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5485 }
5486
5487 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5488 {
5489         int i;
5490
5491         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5492                 if (btrfs_dev_stat_read(dev, i) != 0)
5493                         break;
5494         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5495                 return; /* all values == 0, suppress message */
5496
5497         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5498                rcu_str_deref(dev->name),
5499                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5500                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5501                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5502                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5503                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5504 }
5505
5506 int btrfs_get_dev_stats(struct btrfs_root *root,
5507                         struct btrfs_ioctl_get_dev_stats *stats)
5508 {
5509         struct btrfs_device *dev;
5510         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5511         int i;
5512
5513         mutex_lock(&fs_devices->device_list_mutex);
5514         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5515         mutex_unlock(&fs_devices->device_list_mutex);
5516
5517         if (!dev) {
5518                 printk(KERN_WARNING
5519                        "btrfs: get dev_stats failed, device not found\n");
5520                 return -ENODEV;
5521         } else if (!dev->dev_stats_valid) {
5522                 printk(KERN_WARNING
5523                        "btrfs: get dev_stats failed, not yet valid\n");
5524                 return -ENODEV;
5525         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5526                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5527                         if (stats->nr_items > i)
5528                                 stats->values[i] =
5529                                         btrfs_dev_stat_read_and_reset(dev, i);
5530                         else
5531                                 btrfs_dev_stat_reset(dev, i);
5532                 }
5533         } else {
5534                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5535                         if (stats->nr_items > i)
5536                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5537         }
5538         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5539                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5540         return 0;
5541 }
5542
5543 int btrfs_scratch_superblock(struct btrfs_device *device)
5544 {
5545         struct buffer_head *bh;
5546         struct btrfs_super_block *disk_super;
5547
5548         bh = btrfs_read_dev_super(device->bdev);
5549         if (!bh)
5550                 return -EINVAL;
5551         disk_super = (struct btrfs_super_block *)bh->b_data;
5552
5553         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5554         set_buffer_dirty(bh);
5555         sync_dirty_buffer(bh);
5556         brelse(bh);
5557
5558         return 0;
5559 }