Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[~shefty/rdma-dev.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 <asm/div64.h>
27 #include "compat.h"
28 #include "ctree.h"
29 #include "extent_map.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "print-tree.h"
33 #include "volumes.h"
34 #include "async-thread.h"
35
36 struct map_lookup {
37         u64 type;
38         int io_align;
39         int io_width;
40         int stripe_len;
41         int sector_size;
42         int num_stripes;
43         int sub_stripes;
44         struct btrfs_bio_stripe stripes[];
45 };
46
47 static int init_first_rw_device(struct btrfs_trans_handle *trans,
48                                 struct btrfs_root *root,
49                                 struct btrfs_device *device);
50 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
51
52 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
53                             (sizeof(struct btrfs_bio_stripe) * (n)))
54
55 static DEFINE_MUTEX(uuid_mutex);
56 static LIST_HEAD(fs_uuids);
57
58 void btrfs_lock_volumes(void)
59 {
60         mutex_lock(&uuid_mutex);
61 }
62
63 void btrfs_unlock_volumes(void)
64 {
65         mutex_unlock(&uuid_mutex);
66 }
67
68 static void lock_chunks(struct btrfs_root *root)
69 {
70         mutex_lock(&root->fs_info->chunk_mutex);
71 }
72
73 static void unlock_chunks(struct btrfs_root *root)
74 {
75         mutex_unlock(&root->fs_info->chunk_mutex);
76 }
77
78 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
79 {
80         struct btrfs_device *device;
81         WARN_ON(fs_devices->opened);
82         while (!list_empty(&fs_devices->devices)) {
83                 device = list_entry(fs_devices->devices.next,
84                                     struct btrfs_device, dev_list);
85                 list_del(&device->dev_list);
86                 kfree(device->name);
87                 kfree(device);
88         }
89         kfree(fs_devices);
90 }
91
92 int btrfs_cleanup_fs_uuids(void)
93 {
94         struct btrfs_fs_devices *fs_devices;
95
96         while (!list_empty(&fs_uuids)) {
97                 fs_devices = list_entry(fs_uuids.next,
98                                         struct btrfs_fs_devices, list);
99                 list_del(&fs_devices->list);
100                 free_fs_devices(fs_devices);
101         }
102         return 0;
103 }
104
105 static noinline struct btrfs_device *__find_device(struct list_head *head,
106                                                    u64 devid, u8 *uuid)
107 {
108         struct btrfs_device *dev;
109
110         list_for_each_entry(dev, head, dev_list) {
111                 if (dev->devid == devid &&
112                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
113                         return dev;
114                 }
115         }
116         return NULL;
117 }
118
119 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
120 {
121         struct btrfs_fs_devices *fs_devices;
122
123         list_for_each_entry(fs_devices, &fs_uuids, list) {
124                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
125                         return fs_devices;
126         }
127         return NULL;
128 }
129
130 static void requeue_list(struct btrfs_pending_bios *pending_bios,
131                         struct bio *head, struct bio *tail)
132 {
133
134         struct bio *old_head;
135
136         old_head = pending_bios->head;
137         pending_bios->head = head;
138         if (pending_bios->tail)
139                 tail->bi_next = old_head;
140         else
141                 pending_bios->tail = tail;
142 }
143
144 /*
145  * we try to collect pending bios for a device so we don't get a large
146  * number of procs sending bios down to the same device.  This greatly
147  * improves the schedulers ability to collect and merge the bios.
148  *
149  * But, it also turns into a long list of bios to process and that is sure
150  * to eventually make the worker thread block.  The solution here is to
151  * make some progress and then put this work struct back at the end of
152  * the list if the block device is congested.  This way, multiple devices
153  * can make progress from a single worker thread.
154  */
155 static noinline int run_scheduled_bios(struct btrfs_device *device)
156 {
157         struct bio *pending;
158         struct backing_dev_info *bdi;
159         struct btrfs_fs_info *fs_info;
160         struct btrfs_pending_bios *pending_bios;
161         struct bio *tail;
162         struct bio *cur;
163         int again = 0;
164         unsigned long num_run;
165         unsigned long num_sync_run;
166         unsigned long batch_run = 0;
167         unsigned long limit;
168         unsigned long last_waited = 0;
169         int force_reg = 0;
170
171         bdi = blk_get_backing_dev_info(device->bdev);
172         fs_info = device->dev_root->fs_info;
173         limit = btrfs_async_submit_limit(fs_info);
174         limit = limit * 2 / 3;
175
176         /* we want to make sure that every time we switch from the sync
177          * list to the normal list, we unplug
178          */
179         num_sync_run = 0;
180
181 loop:
182         spin_lock(&device->io_lock);
183
184 loop_lock:
185         num_run = 0;
186
187         /* take all the bios off the list at once and process them
188          * later on (without the lock held).  But, remember the
189          * tail and other pointers so the bios can be properly reinserted
190          * into the list if we hit congestion
191          */
192         if (!force_reg && device->pending_sync_bios.head) {
193                 pending_bios = &device->pending_sync_bios;
194                 force_reg = 1;
195         } else {
196                 pending_bios = &device->pending_bios;
197                 force_reg = 0;
198         }
199
200         pending = pending_bios->head;
201         tail = pending_bios->tail;
202         WARN_ON(pending && !tail);
203
204         /*
205          * if pending was null this time around, no bios need processing
206          * at all and we can stop.  Otherwise it'll loop back up again
207          * and do an additional check so no bios are missed.
208          *
209          * device->running_pending is used to synchronize with the
210          * schedule_bio code.
211          */
212         if (device->pending_sync_bios.head == NULL &&
213             device->pending_bios.head == NULL) {
214                 again = 0;
215                 device->running_pending = 0;
216         } else {
217                 again = 1;
218                 device->running_pending = 1;
219         }
220
221         pending_bios->head = NULL;
222         pending_bios->tail = NULL;
223
224         spin_unlock(&device->io_lock);
225
226         /*
227          * if we're doing the regular priority list, make sure we unplug
228          * for any high prio bios we've sent down
229          */
230         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
231                 num_sync_run = 0;
232                 blk_run_backing_dev(bdi, NULL);
233         }
234
235         while (pending) {
236
237                 rmb();
238                 /* we want to work on both lists, but do more bios on the
239                  * sync list than the regular list
240                  */
241                 if ((num_run > 32 &&
242                     pending_bios != &device->pending_sync_bios &&
243                     device->pending_sync_bios.head) ||
244                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
245                     device->pending_bios.head)) {
246                         spin_lock(&device->io_lock);
247                         requeue_list(pending_bios, pending, tail);
248                         goto loop_lock;
249                 }
250
251                 cur = pending;
252                 pending = pending->bi_next;
253                 cur->bi_next = NULL;
254                 atomic_dec(&fs_info->nr_async_bios);
255
256                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
257                     waitqueue_active(&fs_info->async_submit_wait))
258                         wake_up(&fs_info->async_submit_wait);
259
260                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
261
262                 if (cur->bi_rw & REQ_SYNC)
263                         num_sync_run++;
264
265                 submit_bio(cur->bi_rw, cur);
266                 num_run++;
267                 batch_run++;
268                 if (need_resched()) {
269                         if (num_sync_run) {
270                                 blk_run_backing_dev(bdi, NULL);
271                                 num_sync_run = 0;
272                         }
273                         cond_resched();
274                 }
275
276                 /*
277                  * we made progress, there is more work to do and the bdi
278                  * is now congested.  Back off and let other work structs
279                  * run instead
280                  */
281                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
282                     fs_info->fs_devices->open_devices > 1) {
283                         struct io_context *ioc;
284
285                         ioc = current->io_context;
286
287                         /*
288                          * the main goal here is that we don't want to
289                          * block if we're going to be able to submit
290                          * more requests without blocking.
291                          *
292                          * This code does two great things, it pokes into
293                          * the elevator code from a filesystem _and_
294                          * it makes assumptions about how batching works.
295                          */
296                         if (ioc && ioc->nr_batch_requests > 0 &&
297                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
298                             (last_waited == 0 ||
299                              ioc->last_waited == last_waited)) {
300                                 /*
301                                  * we want to go through our batch of
302                                  * requests and stop.  So, we copy out
303                                  * the ioc->last_waited time and test
304                                  * against it before looping
305                                  */
306                                 last_waited = ioc->last_waited;
307                                 if (need_resched()) {
308                                         if (num_sync_run) {
309                                                 blk_run_backing_dev(bdi, NULL);
310                                                 num_sync_run = 0;
311                                         }
312                                         cond_resched();
313                                 }
314                                 continue;
315                         }
316                         spin_lock(&device->io_lock);
317                         requeue_list(pending_bios, pending, tail);
318                         device->running_pending = 1;
319
320                         spin_unlock(&device->io_lock);
321                         btrfs_requeue_work(&device->work);
322                         goto done;
323                 }
324         }
325
326         if (num_sync_run) {
327                 num_sync_run = 0;
328                 blk_run_backing_dev(bdi, NULL);
329         }
330         /*
331          * IO has already been through a long path to get here.  Checksumming,
332          * async helper threads, perhaps compression.  We've done a pretty
333          * good job of collecting a batch of IO and should just unplug
334          * the device right away.
335          *
336          * This will help anyone who is waiting on the IO, they might have
337          * already unplugged, but managed to do so before the bio they
338          * cared about found its way down here.
339          */
340         blk_run_backing_dev(bdi, NULL);
341
342         cond_resched();
343         if (again)
344                 goto loop;
345
346         spin_lock(&device->io_lock);
347         if (device->pending_bios.head || device->pending_sync_bios.head)
348                 goto loop_lock;
349         spin_unlock(&device->io_lock);
350
351 done:
352         return 0;
353 }
354
355 static void pending_bios_fn(struct btrfs_work *work)
356 {
357         struct btrfs_device *device;
358
359         device = container_of(work, struct btrfs_device, work);
360         run_scheduled_bios(device);
361 }
362
363 static noinline int device_list_add(const char *path,
364                            struct btrfs_super_block *disk_super,
365                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
366 {
367         struct btrfs_device *device;
368         struct btrfs_fs_devices *fs_devices;
369         u64 found_transid = btrfs_super_generation(disk_super);
370         char *name;
371
372         fs_devices = find_fsid(disk_super->fsid);
373         if (!fs_devices) {
374                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
375                 if (!fs_devices)
376                         return -ENOMEM;
377                 INIT_LIST_HEAD(&fs_devices->devices);
378                 INIT_LIST_HEAD(&fs_devices->alloc_list);
379                 list_add(&fs_devices->list, &fs_uuids);
380                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
381                 fs_devices->latest_devid = devid;
382                 fs_devices->latest_trans = found_transid;
383                 mutex_init(&fs_devices->device_list_mutex);
384                 device = NULL;
385         } else {
386                 device = __find_device(&fs_devices->devices, devid,
387                                        disk_super->dev_item.uuid);
388         }
389         if (!device) {
390                 if (fs_devices->opened)
391                         return -EBUSY;
392
393                 device = kzalloc(sizeof(*device), GFP_NOFS);
394                 if (!device) {
395                         /* we can safely leave the fs_devices entry around */
396                         return -ENOMEM;
397                 }
398                 device->devid = devid;
399                 device->work.func = pending_bios_fn;
400                 memcpy(device->uuid, disk_super->dev_item.uuid,
401                        BTRFS_UUID_SIZE);
402                 spin_lock_init(&device->io_lock);
403                 device->name = kstrdup(path, GFP_NOFS);
404                 if (!device->name) {
405                         kfree(device);
406                         return -ENOMEM;
407                 }
408                 INIT_LIST_HEAD(&device->dev_alloc_list);
409
410                 mutex_lock(&fs_devices->device_list_mutex);
411                 list_add(&device->dev_list, &fs_devices->devices);
412                 mutex_unlock(&fs_devices->device_list_mutex);
413
414                 device->fs_devices = fs_devices;
415                 fs_devices->num_devices++;
416         } else if (!device->name || strcmp(device->name, path)) {
417                 name = kstrdup(path, GFP_NOFS);
418                 if (!name)
419                         return -ENOMEM;
420                 kfree(device->name);
421                 device->name = name;
422                 if (device->missing) {
423                         fs_devices->missing_devices--;
424                         device->missing = 0;
425                 }
426         }
427
428         if (found_transid > fs_devices->latest_trans) {
429                 fs_devices->latest_devid = devid;
430                 fs_devices->latest_trans = found_transid;
431         }
432         *fs_devices_ret = fs_devices;
433         return 0;
434 }
435
436 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
437 {
438         struct btrfs_fs_devices *fs_devices;
439         struct btrfs_device *device;
440         struct btrfs_device *orig_dev;
441
442         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
443         if (!fs_devices)
444                 return ERR_PTR(-ENOMEM);
445
446         INIT_LIST_HEAD(&fs_devices->devices);
447         INIT_LIST_HEAD(&fs_devices->alloc_list);
448         INIT_LIST_HEAD(&fs_devices->list);
449         mutex_init(&fs_devices->device_list_mutex);
450         fs_devices->latest_devid = orig->latest_devid;
451         fs_devices->latest_trans = orig->latest_trans;
452         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
453
454         mutex_lock(&orig->device_list_mutex);
455         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
456                 device = kzalloc(sizeof(*device), GFP_NOFS);
457                 if (!device)
458                         goto error;
459
460                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
461                 if (!device->name) {
462                         kfree(device);
463                         goto error;
464                 }
465
466                 device->devid = orig_dev->devid;
467                 device->work.func = pending_bios_fn;
468                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
469                 spin_lock_init(&device->io_lock);
470                 INIT_LIST_HEAD(&device->dev_list);
471                 INIT_LIST_HEAD(&device->dev_alloc_list);
472
473                 list_add(&device->dev_list, &fs_devices->devices);
474                 device->fs_devices = fs_devices;
475                 fs_devices->num_devices++;
476         }
477         mutex_unlock(&orig->device_list_mutex);
478         return fs_devices;
479 error:
480         mutex_unlock(&orig->device_list_mutex);
481         free_fs_devices(fs_devices);
482         return ERR_PTR(-ENOMEM);
483 }
484
485 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
486 {
487         struct btrfs_device *device, *next;
488
489         mutex_lock(&uuid_mutex);
490 again:
491         mutex_lock(&fs_devices->device_list_mutex);
492         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
493                 if (device->in_fs_metadata)
494                         continue;
495
496                 if (device->bdev) {
497                         blkdev_put(device->bdev, device->mode);
498                         device->bdev = NULL;
499                         fs_devices->open_devices--;
500                 }
501                 if (device->writeable) {
502                         list_del_init(&device->dev_alloc_list);
503                         device->writeable = 0;
504                         fs_devices->rw_devices--;
505                 }
506                 list_del_init(&device->dev_list);
507                 fs_devices->num_devices--;
508                 kfree(device->name);
509                 kfree(device);
510         }
511         mutex_unlock(&fs_devices->device_list_mutex);
512
513         if (fs_devices->seed) {
514                 fs_devices = fs_devices->seed;
515                 goto again;
516         }
517
518         mutex_unlock(&uuid_mutex);
519         return 0;
520 }
521
522 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
523 {
524         struct btrfs_device *device;
525
526         if (--fs_devices->opened > 0)
527                 return 0;
528
529         list_for_each_entry(device, &fs_devices->devices, dev_list) {
530                 if (device->bdev) {
531                         blkdev_put(device->bdev, device->mode);
532                         fs_devices->open_devices--;
533                 }
534                 if (device->writeable) {
535                         list_del_init(&device->dev_alloc_list);
536                         fs_devices->rw_devices--;
537                 }
538
539                 device->bdev = NULL;
540                 device->writeable = 0;
541                 device->in_fs_metadata = 0;
542         }
543         WARN_ON(fs_devices->open_devices);
544         WARN_ON(fs_devices->rw_devices);
545         fs_devices->opened = 0;
546         fs_devices->seeding = 0;
547
548         return 0;
549 }
550
551 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
552 {
553         struct btrfs_fs_devices *seed_devices = NULL;
554         int ret;
555
556         mutex_lock(&uuid_mutex);
557         ret = __btrfs_close_devices(fs_devices);
558         if (!fs_devices->opened) {
559                 seed_devices = fs_devices->seed;
560                 fs_devices->seed = NULL;
561         }
562         mutex_unlock(&uuid_mutex);
563
564         while (seed_devices) {
565                 fs_devices = seed_devices;
566                 seed_devices = fs_devices->seed;
567                 __btrfs_close_devices(fs_devices);
568                 free_fs_devices(fs_devices);
569         }
570         return ret;
571 }
572
573 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
574                                 fmode_t flags, void *holder)
575 {
576         struct block_device *bdev;
577         struct list_head *head = &fs_devices->devices;
578         struct btrfs_device *device;
579         struct block_device *latest_bdev = NULL;
580         struct buffer_head *bh;
581         struct btrfs_super_block *disk_super;
582         u64 latest_devid = 0;
583         u64 latest_transid = 0;
584         u64 devid;
585         int seeding = 1;
586         int ret = 0;
587
588         flags |= FMODE_EXCL;
589
590         list_for_each_entry(device, head, dev_list) {
591                 if (device->bdev)
592                         continue;
593                 if (!device->name)
594                         continue;
595
596                 bdev = blkdev_get_by_path(device->name, flags, holder);
597                 if (IS_ERR(bdev)) {
598                         printk(KERN_INFO "open %s failed\n", device->name);
599                         goto error;
600                 }
601                 set_blocksize(bdev, 4096);
602
603                 bh = btrfs_read_dev_super(bdev);
604                 if (!bh) {
605                         ret = -EINVAL;
606                         goto error_close;
607                 }
608
609                 disk_super = (struct btrfs_super_block *)bh->b_data;
610                 devid = btrfs_stack_device_id(&disk_super->dev_item);
611                 if (devid != device->devid)
612                         goto error_brelse;
613
614                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
615                            BTRFS_UUID_SIZE))
616                         goto error_brelse;
617
618                 device->generation = btrfs_super_generation(disk_super);
619                 if (!latest_transid || device->generation > latest_transid) {
620                         latest_devid = devid;
621                         latest_transid = device->generation;
622                         latest_bdev = bdev;
623                 }
624
625                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
626                         device->writeable = 0;
627                 } else {
628                         device->writeable = !bdev_read_only(bdev);
629                         seeding = 0;
630                 }
631
632                 device->bdev = bdev;
633                 device->in_fs_metadata = 0;
634                 device->mode = flags;
635
636                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
637                         fs_devices->rotating = 1;
638
639                 fs_devices->open_devices++;
640                 if (device->writeable) {
641                         fs_devices->rw_devices++;
642                         list_add(&device->dev_alloc_list,
643                                  &fs_devices->alloc_list);
644                 }
645                 continue;
646
647 error_brelse:
648                 brelse(bh);
649 error_close:
650                 blkdev_put(bdev, flags);
651 error:
652                 continue;
653         }
654         if (fs_devices->open_devices == 0) {
655                 ret = -EIO;
656                 goto out;
657         }
658         fs_devices->seeding = seeding;
659         fs_devices->opened = 1;
660         fs_devices->latest_bdev = latest_bdev;
661         fs_devices->latest_devid = latest_devid;
662         fs_devices->latest_trans = latest_transid;
663         fs_devices->total_rw_bytes = 0;
664 out:
665         return ret;
666 }
667
668 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
669                        fmode_t flags, void *holder)
670 {
671         int ret;
672
673         mutex_lock(&uuid_mutex);
674         if (fs_devices->opened) {
675                 fs_devices->opened++;
676                 ret = 0;
677         } else {
678                 ret = __btrfs_open_devices(fs_devices, flags, holder);
679         }
680         mutex_unlock(&uuid_mutex);
681         return ret;
682 }
683
684 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
685                           struct btrfs_fs_devices **fs_devices_ret)
686 {
687         struct btrfs_super_block *disk_super;
688         struct block_device *bdev;
689         struct buffer_head *bh;
690         int ret;
691         u64 devid;
692         u64 transid;
693
694         mutex_lock(&uuid_mutex);
695
696         flags |= FMODE_EXCL;
697         bdev = blkdev_get_by_path(path, flags, holder);
698
699         if (IS_ERR(bdev)) {
700                 ret = PTR_ERR(bdev);
701                 goto error;
702         }
703
704         ret = set_blocksize(bdev, 4096);
705         if (ret)
706                 goto error_close;
707         bh = btrfs_read_dev_super(bdev);
708         if (!bh) {
709                 ret = -EINVAL;
710                 goto error_close;
711         }
712         disk_super = (struct btrfs_super_block *)bh->b_data;
713         devid = btrfs_stack_device_id(&disk_super->dev_item);
714         transid = btrfs_super_generation(disk_super);
715         if (disk_super->label[0])
716                 printk(KERN_INFO "device label %s ", disk_super->label);
717         else {
718                 /* FIXME, make a readl uuid parser */
719                 printk(KERN_INFO "device fsid %llx-%llx ",
720                        *(unsigned long long *)disk_super->fsid,
721                        *(unsigned long long *)(disk_super->fsid + 8));
722         }
723         printk(KERN_CONT "devid %llu transid %llu %s\n",
724                (unsigned long long)devid, (unsigned long long)transid, path);
725         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
726
727         brelse(bh);
728 error_close:
729         blkdev_put(bdev, flags);
730 error:
731         mutex_unlock(&uuid_mutex);
732         return ret;
733 }
734
735 /* helper to account the used device space in the range */
736 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
737                                    u64 end, u64 *length)
738 {
739         struct btrfs_key key;
740         struct btrfs_root *root = device->dev_root;
741         struct btrfs_dev_extent *dev_extent;
742         struct btrfs_path *path;
743         u64 extent_end;
744         int ret;
745         int slot;
746         struct extent_buffer *l;
747
748         *length = 0;
749
750         if (start >= device->total_bytes)
751                 return 0;
752
753         path = btrfs_alloc_path();
754         if (!path)
755                 return -ENOMEM;
756         path->reada = 2;
757
758         key.objectid = device->devid;
759         key.offset = start;
760         key.type = BTRFS_DEV_EXTENT_KEY;
761
762         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
763         if (ret < 0)
764                 goto out;
765         if (ret > 0) {
766                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
767                 if (ret < 0)
768                         goto out;
769         }
770
771         while (1) {
772                 l = path->nodes[0];
773                 slot = path->slots[0];
774                 if (slot >= btrfs_header_nritems(l)) {
775                         ret = btrfs_next_leaf(root, path);
776                         if (ret == 0)
777                                 continue;
778                         if (ret < 0)
779                                 goto out;
780
781                         break;
782                 }
783                 btrfs_item_key_to_cpu(l, &key, slot);
784
785                 if (key.objectid < device->devid)
786                         goto next;
787
788                 if (key.objectid > device->devid)
789                         break;
790
791                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
792                         goto next;
793
794                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
795                 extent_end = key.offset + btrfs_dev_extent_length(l,
796                                                                   dev_extent);
797                 if (key.offset <= start && extent_end > end) {
798                         *length = end - start + 1;
799                         break;
800                 } else if (key.offset <= start && extent_end > start)
801                         *length += extent_end - start;
802                 else if (key.offset > start && extent_end <= end)
803                         *length += extent_end - key.offset;
804                 else if (key.offset > start && key.offset <= end) {
805                         *length += end - key.offset + 1;
806                         break;
807                 } else if (key.offset > end)
808                         break;
809
810 next:
811                 path->slots[0]++;
812         }
813         ret = 0;
814 out:
815         btrfs_free_path(path);
816         return ret;
817 }
818
819 /*
820  * find_free_dev_extent - find free space in the specified device
821  * @trans:      transaction handler
822  * @device:     the device which we search the free space in
823  * @num_bytes:  the size of the free space that we need
824  * @start:      store the start of the free space.
825  * @len:        the size of the free space. that we find, or the size of the max
826  *              free space if we don't find suitable free space
827  *
828  * this uses a pretty simple search, the expectation is that it is
829  * called very infrequently and that a given device has a small number
830  * of extents
831  *
832  * @start is used to store the start of the free space if we find. But if we
833  * don't find suitable free space, it will be used to store the start position
834  * of the max free space.
835  *
836  * @len is used to store the size of the free space that we find.
837  * But if we don't find suitable free space, it is used to store the size of
838  * the max free space.
839  */
840 int find_free_dev_extent(struct btrfs_trans_handle *trans,
841                          struct btrfs_device *device, u64 num_bytes,
842                          u64 *start, u64 *len)
843 {
844         struct btrfs_key key;
845         struct btrfs_root *root = device->dev_root;
846         struct btrfs_dev_extent *dev_extent;
847         struct btrfs_path *path;
848         u64 hole_size;
849         u64 max_hole_start;
850         u64 max_hole_size;
851         u64 extent_end;
852         u64 search_start;
853         u64 search_end = device->total_bytes;
854         int ret;
855         int slot;
856         struct extent_buffer *l;
857
858         /* FIXME use last free of some kind */
859
860         /* we don't want to overwrite the superblock on the drive,
861          * so we make sure to start at an offset of at least 1MB
862          */
863         search_start = 1024 * 1024;
864
865         if (root->fs_info->alloc_start + num_bytes <= search_end)
866                 search_start = max(root->fs_info->alloc_start, search_start);
867
868         max_hole_start = search_start;
869         max_hole_size = 0;
870
871         if (search_start >= search_end) {
872                 ret = -ENOSPC;
873                 goto error;
874         }
875
876         path = btrfs_alloc_path();
877         if (!path) {
878                 ret = -ENOMEM;
879                 goto error;
880         }
881         path->reada = 2;
882
883         key.objectid = device->devid;
884         key.offset = search_start;
885         key.type = BTRFS_DEV_EXTENT_KEY;
886
887         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
888         if (ret < 0)
889                 goto out;
890         if (ret > 0) {
891                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
892                 if (ret < 0)
893                         goto out;
894         }
895
896         while (1) {
897                 l = path->nodes[0];
898                 slot = path->slots[0];
899                 if (slot >= btrfs_header_nritems(l)) {
900                         ret = btrfs_next_leaf(root, path);
901                         if (ret == 0)
902                                 continue;
903                         if (ret < 0)
904                                 goto out;
905
906                         break;
907                 }
908                 btrfs_item_key_to_cpu(l, &key, slot);
909
910                 if (key.objectid < device->devid)
911                         goto next;
912
913                 if (key.objectid > device->devid)
914                         break;
915
916                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
917                         goto next;
918
919                 if (key.offset > search_start) {
920                         hole_size = key.offset - search_start;
921
922                         if (hole_size > max_hole_size) {
923                                 max_hole_start = search_start;
924                                 max_hole_size = hole_size;
925                         }
926
927                         /*
928                          * If this free space is greater than which we need,
929                          * it must be the max free space that we have found
930                          * until now, so max_hole_start must point to the start
931                          * of this free space and the length of this free space
932                          * is stored in max_hole_size. Thus, we return
933                          * max_hole_start and max_hole_size and go back to the
934                          * caller.
935                          */
936                         if (hole_size >= num_bytes) {
937                                 ret = 0;
938                                 goto out;
939                         }
940                 }
941
942                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
943                 extent_end = key.offset + btrfs_dev_extent_length(l,
944                                                                   dev_extent);
945                 if (extent_end > search_start)
946                         search_start = extent_end;
947 next:
948                 path->slots[0]++;
949                 cond_resched();
950         }
951
952         hole_size = search_end- search_start;
953         if (hole_size > max_hole_size) {
954                 max_hole_start = search_start;
955                 max_hole_size = hole_size;
956         }
957
958         /* See above. */
959         if (hole_size < num_bytes)
960                 ret = -ENOSPC;
961         else
962                 ret = 0;
963
964 out:
965         btrfs_free_path(path);
966 error:
967         *start = max_hole_start;
968         if (len)
969                 *len = max_hole_size;
970         return ret;
971 }
972
973 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
974                           struct btrfs_device *device,
975                           u64 start)
976 {
977         int ret;
978         struct btrfs_path *path;
979         struct btrfs_root *root = device->dev_root;
980         struct btrfs_key key;
981         struct btrfs_key found_key;
982         struct extent_buffer *leaf = NULL;
983         struct btrfs_dev_extent *extent = NULL;
984
985         path = btrfs_alloc_path();
986         if (!path)
987                 return -ENOMEM;
988
989         key.objectid = device->devid;
990         key.offset = start;
991         key.type = BTRFS_DEV_EXTENT_KEY;
992
993         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
994         if (ret > 0) {
995                 ret = btrfs_previous_item(root, path, key.objectid,
996                                           BTRFS_DEV_EXTENT_KEY);
997                 BUG_ON(ret);
998                 leaf = path->nodes[0];
999                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1000                 extent = btrfs_item_ptr(leaf, path->slots[0],
1001                                         struct btrfs_dev_extent);
1002                 BUG_ON(found_key.offset > start || found_key.offset +
1003                        btrfs_dev_extent_length(leaf, extent) < start);
1004                 ret = 0;
1005         } else if (ret == 0) {
1006                 leaf = path->nodes[0];
1007                 extent = btrfs_item_ptr(leaf, path->slots[0],
1008                                         struct btrfs_dev_extent);
1009         }
1010         BUG_ON(ret);
1011
1012         if (device->bytes_used > 0)
1013                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1014         ret = btrfs_del_item(trans, root, path);
1015         BUG_ON(ret);
1016
1017         btrfs_free_path(path);
1018         return ret;
1019 }
1020
1021 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1022                            struct btrfs_device *device,
1023                            u64 chunk_tree, u64 chunk_objectid,
1024                            u64 chunk_offset, u64 start, u64 num_bytes)
1025 {
1026         int ret;
1027         struct btrfs_path *path;
1028         struct btrfs_root *root = device->dev_root;
1029         struct btrfs_dev_extent *extent;
1030         struct extent_buffer *leaf;
1031         struct btrfs_key key;
1032
1033         WARN_ON(!device->in_fs_metadata);
1034         path = btrfs_alloc_path();
1035         if (!path)
1036                 return -ENOMEM;
1037
1038         key.objectid = device->devid;
1039         key.offset = start;
1040         key.type = BTRFS_DEV_EXTENT_KEY;
1041         ret = btrfs_insert_empty_item(trans, root, path, &key,
1042                                       sizeof(*extent));
1043         BUG_ON(ret);
1044
1045         leaf = path->nodes[0];
1046         extent = btrfs_item_ptr(leaf, path->slots[0],
1047                                 struct btrfs_dev_extent);
1048         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1049         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1050         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1051
1052         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1053                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1054                     BTRFS_UUID_SIZE);
1055
1056         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1057         btrfs_mark_buffer_dirty(leaf);
1058         btrfs_free_path(path);
1059         return ret;
1060 }
1061
1062 static noinline int find_next_chunk(struct btrfs_root *root,
1063                                     u64 objectid, u64 *offset)
1064 {
1065         struct btrfs_path *path;
1066         int ret;
1067         struct btrfs_key key;
1068         struct btrfs_chunk *chunk;
1069         struct btrfs_key found_key;
1070
1071         path = btrfs_alloc_path();
1072         BUG_ON(!path);
1073
1074         key.objectid = objectid;
1075         key.offset = (u64)-1;
1076         key.type = BTRFS_CHUNK_ITEM_KEY;
1077
1078         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1079         if (ret < 0)
1080                 goto error;
1081
1082         BUG_ON(ret == 0);
1083
1084         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1085         if (ret) {
1086                 *offset = 0;
1087         } else {
1088                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1089                                       path->slots[0]);
1090                 if (found_key.objectid != objectid)
1091                         *offset = 0;
1092                 else {
1093                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1094                                                struct btrfs_chunk);
1095                         *offset = found_key.offset +
1096                                 btrfs_chunk_length(path->nodes[0], chunk);
1097                 }
1098         }
1099         ret = 0;
1100 error:
1101         btrfs_free_path(path);
1102         return ret;
1103 }
1104
1105 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1106 {
1107         int ret;
1108         struct btrfs_key key;
1109         struct btrfs_key found_key;
1110         struct btrfs_path *path;
1111
1112         root = root->fs_info->chunk_root;
1113
1114         path = btrfs_alloc_path();
1115         if (!path)
1116                 return -ENOMEM;
1117
1118         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1119         key.type = BTRFS_DEV_ITEM_KEY;
1120         key.offset = (u64)-1;
1121
1122         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1123         if (ret < 0)
1124                 goto error;
1125
1126         BUG_ON(ret == 0);
1127
1128         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1129                                   BTRFS_DEV_ITEM_KEY);
1130         if (ret) {
1131                 *objectid = 1;
1132         } else {
1133                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1134                                       path->slots[0]);
1135                 *objectid = found_key.offset + 1;
1136         }
1137         ret = 0;
1138 error:
1139         btrfs_free_path(path);
1140         return ret;
1141 }
1142
1143 /*
1144  * the device information is stored in the chunk root
1145  * the btrfs_device struct should be fully filled in
1146  */
1147 int btrfs_add_device(struct btrfs_trans_handle *trans,
1148                      struct btrfs_root *root,
1149                      struct btrfs_device *device)
1150 {
1151         int ret;
1152         struct btrfs_path *path;
1153         struct btrfs_dev_item *dev_item;
1154         struct extent_buffer *leaf;
1155         struct btrfs_key key;
1156         unsigned long ptr;
1157
1158         root = root->fs_info->chunk_root;
1159
1160         path = btrfs_alloc_path();
1161         if (!path)
1162                 return -ENOMEM;
1163
1164         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1165         key.type = BTRFS_DEV_ITEM_KEY;
1166         key.offset = device->devid;
1167
1168         ret = btrfs_insert_empty_item(trans, root, path, &key,
1169                                       sizeof(*dev_item));
1170         if (ret)
1171                 goto out;
1172
1173         leaf = path->nodes[0];
1174         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1175
1176         btrfs_set_device_id(leaf, dev_item, device->devid);
1177         btrfs_set_device_generation(leaf, dev_item, 0);
1178         btrfs_set_device_type(leaf, dev_item, device->type);
1179         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1180         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1181         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1182         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1183         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1184         btrfs_set_device_group(leaf, dev_item, 0);
1185         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1186         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1187         btrfs_set_device_start_offset(leaf, dev_item, 0);
1188
1189         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1190         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1191         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1192         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1193         btrfs_mark_buffer_dirty(leaf);
1194
1195         ret = 0;
1196 out:
1197         btrfs_free_path(path);
1198         return ret;
1199 }
1200
1201 static int btrfs_rm_dev_item(struct btrfs_root *root,
1202                              struct btrfs_device *device)
1203 {
1204         int ret;
1205         struct btrfs_path *path;
1206         struct btrfs_key key;
1207         struct btrfs_trans_handle *trans;
1208
1209         root = root->fs_info->chunk_root;
1210
1211         path = btrfs_alloc_path();
1212         if (!path)
1213                 return -ENOMEM;
1214
1215         trans = btrfs_start_transaction(root, 0);
1216         if (IS_ERR(trans)) {
1217                 btrfs_free_path(path);
1218                 return PTR_ERR(trans);
1219         }
1220         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1221         key.type = BTRFS_DEV_ITEM_KEY;
1222         key.offset = device->devid;
1223         lock_chunks(root);
1224
1225         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1226         if (ret < 0)
1227                 goto out;
1228
1229         if (ret > 0) {
1230                 ret = -ENOENT;
1231                 goto out;
1232         }
1233
1234         ret = btrfs_del_item(trans, root, path);
1235         if (ret)
1236                 goto out;
1237 out:
1238         btrfs_free_path(path);
1239         unlock_chunks(root);
1240         btrfs_commit_transaction(trans, root);
1241         return ret;
1242 }
1243
1244 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1245 {
1246         struct btrfs_device *device;
1247         struct btrfs_device *next_device;
1248         struct block_device *bdev;
1249         struct buffer_head *bh = NULL;
1250         struct btrfs_super_block *disk_super;
1251         u64 all_avail;
1252         u64 devid;
1253         u64 num_devices;
1254         u8 *dev_uuid;
1255         int ret = 0;
1256
1257         mutex_lock(&uuid_mutex);
1258         mutex_lock(&root->fs_info->volume_mutex);
1259
1260         all_avail = root->fs_info->avail_data_alloc_bits |
1261                 root->fs_info->avail_system_alloc_bits |
1262                 root->fs_info->avail_metadata_alloc_bits;
1263
1264         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1265             root->fs_info->fs_devices->num_devices <= 4) {
1266                 printk(KERN_ERR "btrfs: unable to go below four devices "
1267                        "on raid10\n");
1268                 ret = -EINVAL;
1269                 goto out;
1270         }
1271
1272         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1273             root->fs_info->fs_devices->num_devices <= 2) {
1274                 printk(KERN_ERR "btrfs: unable to go below two "
1275                        "devices on raid1\n");
1276                 ret = -EINVAL;
1277                 goto out;
1278         }
1279
1280         if (strcmp(device_path, "missing") == 0) {
1281                 struct list_head *devices;
1282                 struct btrfs_device *tmp;
1283
1284                 device = NULL;
1285                 devices = &root->fs_info->fs_devices->devices;
1286                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1287                 list_for_each_entry(tmp, devices, dev_list) {
1288                         if (tmp->in_fs_metadata && !tmp->bdev) {
1289                                 device = tmp;
1290                                 break;
1291                         }
1292                 }
1293                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1294                 bdev = NULL;
1295                 bh = NULL;
1296                 disk_super = NULL;
1297                 if (!device) {
1298                         printk(KERN_ERR "btrfs: no missing devices found to "
1299                                "remove\n");
1300                         goto out;
1301                 }
1302         } else {
1303                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1304                                           root->fs_info->bdev_holder);
1305                 if (IS_ERR(bdev)) {
1306                         ret = PTR_ERR(bdev);
1307                         goto out;
1308                 }
1309
1310                 set_blocksize(bdev, 4096);
1311                 bh = btrfs_read_dev_super(bdev);
1312                 if (!bh) {
1313                         ret = -EINVAL;
1314                         goto error_close;
1315                 }
1316                 disk_super = (struct btrfs_super_block *)bh->b_data;
1317                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1318                 dev_uuid = disk_super->dev_item.uuid;
1319                 device = btrfs_find_device(root, devid, dev_uuid,
1320                                            disk_super->fsid);
1321                 if (!device) {
1322                         ret = -ENOENT;
1323                         goto error_brelse;
1324                 }
1325         }
1326
1327         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1328                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1329                        "device\n");
1330                 ret = -EINVAL;
1331                 goto error_brelse;
1332         }
1333
1334         if (device->writeable) {
1335                 list_del_init(&device->dev_alloc_list);
1336                 root->fs_info->fs_devices->rw_devices--;
1337         }
1338
1339         ret = btrfs_shrink_device(device, 0);
1340         if (ret)
1341                 goto error_brelse;
1342
1343         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1344         if (ret)
1345                 goto error_brelse;
1346
1347         device->in_fs_metadata = 0;
1348
1349         /*
1350          * the device list mutex makes sure that we don't change
1351          * the device list while someone else is writing out all
1352          * the device supers.
1353          */
1354         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1355         list_del_init(&device->dev_list);
1356         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1357
1358         device->fs_devices->num_devices--;
1359
1360         if (device->missing)
1361                 root->fs_info->fs_devices->missing_devices--;
1362
1363         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1364                                  struct btrfs_device, dev_list);
1365         if (device->bdev == root->fs_info->sb->s_bdev)
1366                 root->fs_info->sb->s_bdev = next_device->bdev;
1367         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1368                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1369
1370         if (device->bdev) {
1371                 blkdev_put(device->bdev, device->mode);
1372                 device->bdev = NULL;
1373                 device->fs_devices->open_devices--;
1374         }
1375
1376         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1377         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1378
1379         if (device->fs_devices->open_devices == 0) {
1380                 struct btrfs_fs_devices *fs_devices;
1381                 fs_devices = root->fs_info->fs_devices;
1382                 while (fs_devices) {
1383                         if (fs_devices->seed == device->fs_devices)
1384                                 break;
1385                         fs_devices = fs_devices->seed;
1386                 }
1387                 fs_devices->seed = device->fs_devices->seed;
1388                 device->fs_devices->seed = NULL;
1389                 __btrfs_close_devices(device->fs_devices);
1390                 free_fs_devices(device->fs_devices);
1391         }
1392
1393         /*
1394          * at this point, the device is zero sized.  We want to
1395          * remove it from the devices list and zero out the old super
1396          */
1397         if (device->writeable) {
1398                 /* make sure this device isn't detected as part of
1399                  * the FS anymore
1400                  */
1401                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1402                 set_buffer_dirty(bh);
1403                 sync_dirty_buffer(bh);
1404         }
1405
1406         kfree(device->name);
1407         kfree(device);
1408         ret = 0;
1409
1410 error_brelse:
1411         brelse(bh);
1412 error_close:
1413         if (bdev)
1414                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1415 out:
1416         mutex_unlock(&root->fs_info->volume_mutex);
1417         mutex_unlock(&uuid_mutex);
1418         return ret;
1419 }
1420
1421 /*
1422  * does all the dirty work required for changing file system's UUID.
1423  */
1424 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1425                                 struct btrfs_root *root)
1426 {
1427         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1428         struct btrfs_fs_devices *old_devices;
1429         struct btrfs_fs_devices *seed_devices;
1430         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1431         struct btrfs_device *device;
1432         u64 super_flags;
1433
1434         BUG_ON(!mutex_is_locked(&uuid_mutex));
1435         if (!fs_devices->seeding)
1436                 return -EINVAL;
1437
1438         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1439         if (!seed_devices)
1440                 return -ENOMEM;
1441
1442         old_devices = clone_fs_devices(fs_devices);
1443         if (IS_ERR(old_devices)) {
1444                 kfree(seed_devices);
1445                 return PTR_ERR(old_devices);
1446         }
1447
1448         list_add(&old_devices->list, &fs_uuids);
1449
1450         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1451         seed_devices->opened = 1;
1452         INIT_LIST_HEAD(&seed_devices->devices);
1453         INIT_LIST_HEAD(&seed_devices->alloc_list);
1454         mutex_init(&seed_devices->device_list_mutex);
1455         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1456         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1457         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1458                 device->fs_devices = seed_devices;
1459         }
1460
1461         fs_devices->seeding = 0;
1462         fs_devices->num_devices = 0;
1463         fs_devices->open_devices = 0;
1464         fs_devices->seed = seed_devices;
1465
1466         generate_random_uuid(fs_devices->fsid);
1467         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1468         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1469         super_flags = btrfs_super_flags(disk_super) &
1470                       ~BTRFS_SUPER_FLAG_SEEDING;
1471         btrfs_set_super_flags(disk_super, super_flags);
1472
1473         return 0;
1474 }
1475
1476 /*
1477  * strore the expected generation for seed devices in device items.
1478  */
1479 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1480                                struct btrfs_root *root)
1481 {
1482         struct btrfs_path *path;
1483         struct extent_buffer *leaf;
1484         struct btrfs_dev_item *dev_item;
1485         struct btrfs_device *device;
1486         struct btrfs_key key;
1487         u8 fs_uuid[BTRFS_UUID_SIZE];
1488         u8 dev_uuid[BTRFS_UUID_SIZE];
1489         u64 devid;
1490         int ret;
1491
1492         path = btrfs_alloc_path();
1493         if (!path)
1494                 return -ENOMEM;
1495
1496         root = root->fs_info->chunk_root;
1497         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1498         key.offset = 0;
1499         key.type = BTRFS_DEV_ITEM_KEY;
1500
1501         while (1) {
1502                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1503                 if (ret < 0)
1504                         goto error;
1505
1506                 leaf = path->nodes[0];
1507 next_slot:
1508                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1509                         ret = btrfs_next_leaf(root, path);
1510                         if (ret > 0)
1511                                 break;
1512                         if (ret < 0)
1513                                 goto error;
1514                         leaf = path->nodes[0];
1515                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1516                         btrfs_release_path(root, path);
1517                         continue;
1518                 }
1519
1520                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1521                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1522                     key.type != BTRFS_DEV_ITEM_KEY)
1523                         break;
1524
1525                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1526                                           struct btrfs_dev_item);
1527                 devid = btrfs_device_id(leaf, dev_item);
1528                 read_extent_buffer(leaf, dev_uuid,
1529                                    (unsigned long)btrfs_device_uuid(dev_item),
1530                                    BTRFS_UUID_SIZE);
1531                 read_extent_buffer(leaf, fs_uuid,
1532                                    (unsigned long)btrfs_device_fsid(dev_item),
1533                                    BTRFS_UUID_SIZE);
1534                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1535                 BUG_ON(!device);
1536
1537                 if (device->fs_devices->seeding) {
1538                         btrfs_set_device_generation(leaf, dev_item,
1539                                                     device->generation);
1540                         btrfs_mark_buffer_dirty(leaf);
1541                 }
1542
1543                 path->slots[0]++;
1544                 goto next_slot;
1545         }
1546         ret = 0;
1547 error:
1548         btrfs_free_path(path);
1549         return ret;
1550 }
1551
1552 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1553 {
1554         struct btrfs_trans_handle *trans;
1555         struct btrfs_device *device;
1556         struct block_device *bdev;
1557         struct list_head *devices;
1558         struct super_block *sb = root->fs_info->sb;
1559         u64 total_bytes;
1560         int seeding_dev = 0;
1561         int ret = 0;
1562
1563         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1564                 return -EINVAL;
1565
1566         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1567                                   root->fs_info->bdev_holder);
1568         if (IS_ERR(bdev))
1569                 return PTR_ERR(bdev);
1570
1571         if (root->fs_info->fs_devices->seeding) {
1572                 seeding_dev = 1;
1573                 down_write(&sb->s_umount);
1574                 mutex_lock(&uuid_mutex);
1575         }
1576
1577         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1578         mutex_lock(&root->fs_info->volume_mutex);
1579
1580         devices = &root->fs_info->fs_devices->devices;
1581         /*
1582          * we have the volume lock, so we don't need the extra
1583          * device list mutex while reading the list here.
1584          */
1585         list_for_each_entry(device, devices, dev_list) {
1586                 if (device->bdev == bdev) {
1587                         ret = -EEXIST;
1588                         goto error;
1589                 }
1590         }
1591
1592         device = kzalloc(sizeof(*device), GFP_NOFS);
1593         if (!device) {
1594                 /* we can safely leave the fs_devices entry around */
1595                 ret = -ENOMEM;
1596                 goto error;
1597         }
1598
1599         device->name = kstrdup(device_path, GFP_NOFS);
1600         if (!device->name) {
1601                 kfree(device);
1602                 ret = -ENOMEM;
1603                 goto error;
1604         }
1605
1606         ret = find_next_devid(root, &device->devid);
1607         if (ret) {
1608                 kfree(device);
1609                 goto error;
1610         }
1611
1612         trans = btrfs_start_transaction(root, 0);
1613         if (IS_ERR(trans)) {
1614                 kfree(device);
1615                 ret = PTR_ERR(trans);
1616                 goto error;
1617         }
1618
1619         lock_chunks(root);
1620
1621         device->writeable = 1;
1622         device->work.func = pending_bios_fn;
1623         generate_random_uuid(device->uuid);
1624         spin_lock_init(&device->io_lock);
1625         device->generation = trans->transid;
1626         device->io_width = root->sectorsize;
1627         device->io_align = root->sectorsize;
1628         device->sector_size = root->sectorsize;
1629         device->total_bytes = i_size_read(bdev->bd_inode);
1630         device->disk_total_bytes = device->total_bytes;
1631         device->dev_root = root->fs_info->dev_root;
1632         device->bdev = bdev;
1633         device->in_fs_metadata = 1;
1634         device->mode = 0;
1635         set_blocksize(device->bdev, 4096);
1636
1637         if (seeding_dev) {
1638                 sb->s_flags &= ~MS_RDONLY;
1639                 ret = btrfs_prepare_sprout(trans, root);
1640                 BUG_ON(ret);
1641         }
1642
1643         device->fs_devices = root->fs_info->fs_devices;
1644
1645         /*
1646          * we don't want write_supers to jump in here with our device
1647          * half setup
1648          */
1649         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1650         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1651         list_add(&device->dev_alloc_list,
1652                  &root->fs_info->fs_devices->alloc_list);
1653         root->fs_info->fs_devices->num_devices++;
1654         root->fs_info->fs_devices->open_devices++;
1655         root->fs_info->fs_devices->rw_devices++;
1656         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1657
1658         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1659                 root->fs_info->fs_devices->rotating = 1;
1660
1661         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1662         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1663                                     total_bytes + device->total_bytes);
1664
1665         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1666         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1667                                     total_bytes + 1);
1668         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1669
1670         if (seeding_dev) {
1671                 ret = init_first_rw_device(trans, root, device);
1672                 BUG_ON(ret);
1673                 ret = btrfs_finish_sprout(trans, root);
1674                 BUG_ON(ret);
1675         } else {
1676                 ret = btrfs_add_device(trans, root, device);
1677         }
1678
1679         /*
1680          * we've got more storage, clear any full flags on the space
1681          * infos
1682          */
1683         btrfs_clear_space_info_full(root->fs_info);
1684
1685         unlock_chunks(root);
1686         btrfs_commit_transaction(trans, root);
1687
1688         if (seeding_dev) {
1689                 mutex_unlock(&uuid_mutex);
1690                 up_write(&sb->s_umount);
1691
1692                 ret = btrfs_relocate_sys_chunks(root);
1693                 BUG_ON(ret);
1694         }
1695 out:
1696         mutex_unlock(&root->fs_info->volume_mutex);
1697         return ret;
1698 error:
1699         blkdev_put(bdev, FMODE_EXCL);
1700         if (seeding_dev) {
1701                 mutex_unlock(&uuid_mutex);
1702                 up_write(&sb->s_umount);
1703         }
1704         goto out;
1705 }
1706
1707 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1708                                         struct btrfs_device *device)
1709 {
1710         int ret;
1711         struct btrfs_path *path;
1712         struct btrfs_root *root;
1713         struct btrfs_dev_item *dev_item;
1714         struct extent_buffer *leaf;
1715         struct btrfs_key key;
1716
1717         root = device->dev_root->fs_info->chunk_root;
1718
1719         path = btrfs_alloc_path();
1720         if (!path)
1721                 return -ENOMEM;
1722
1723         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1724         key.type = BTRFS_DEV_ITEM_KEY;
1725         key.offset = device->devid;
1726
1727         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1728         if (ret < 0)
1729                 goto out;
1730
1731         if (ret > 0) {
1732                 ret = -ENOENT;
1733                 goto out;
1734         }
1735
1736         leaf = path->nodes[0];
1737         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1738
1739         btrfs_set_device_id(leaf, dev_item, device->devid);
1740         btrfs_set_device_type(leaf, dev_item, device->type);
1741         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1742         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1743         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1744         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1745         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1746         btrfs_mark_buffer_dirty(leaf);
1747
1748 out:
1749         btrfs_free_path(path);
1750         return ret;
1751 }
1752
1753 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1754                       struct btrfs_device *device, u64 new_size)
1755 {
1756         struct btrfs_super_block *super_copy =
1757                 &device->dev_root->fs_info->super_copy;
1758         u64 old_total = btrfs_super_total_bytes(super_copy);
1759         u64 diff = new_size - device->total_bytes;
1760
1761         if (!device->writeable)
1762                 return -EACCES;
1763         if (new_size <= device->total_bytes)
1764                 return -EINVAL;
1765
1766         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1767         device->fs_devices->total_rw_bytes += diff;
1768
1769         device->total_bytes = new_size;
1770         device->disk_total_bytes = new_size;
1771         btrfs_clear_space_info_full(device->dev_root->fs_info);
1772
1773         return btrfs_update_device(trans, device);
1774 }
1775
1776 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1777                       struct btrfs_device *device, u64 new_size)
1778 {
1779         int ret;
1780         lock_chunks(device->dev_root);
1781         ret = __btrfs_grow_device(trans, device, new_size);
1782         unlock_chunks(device->dev_root);
1783         return ret;
1784 }
1785
1786 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1787                             struct btrfs_root *root,
1788                             u64 chunk_tree, u64 chunk_objectid,
1789                             u64 chunk_offset)
1790 {
1791         int ret;
1792         struct btrfs_path *path;
1793         struct btrfs_key key;
1794
1795         root = root->fs_info->chunk_root;
1796         path = btrfs_alloc_path();
1797         if (!path)
1798                 return -ENOMEM;
1799
1800         key.objectid = chunk_objectid;
1801         key.offset = chunk_offset;
1802         key.type = BTRFS_CHUNK_ITEM_KEY;
1803
1804         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1805         BUG_ON(ret);
1806
1807         ret = btrfs_del_item(trans, root, path);
1808         BUG_ON(ret);
1809
1810         btrfs_free_path(path);
1811         return 0;
1812 }
1813
1814 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1815                         chunk_offset)
1816 {
1817         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1818         struct btrfs_disk_key *disk_key;
1819         struct btrfs_chunk *chunk;
1820         u8 *ptr;
1821         int ret = 0;
1822         u32 num_stripes;
1823         u32 array_size;
1824         u32 len = 0;
1825         u32 cur;
1826         struct btrfs_key key;
1827
1828         array_size = btrfs_super_sys_array_size(super_copy);
1829
1830         ptr = super_copy->sys_chunk_array;
1831         cur = 0;
1832
1833         while (cur < array_size) {
1834                 disk_key = (struct btrfs_disk_key *)ptr;
1835                 btrfs_disk_key_to_cpu(&key, disk_key);
1836
1837                 len = sizeof(*disk_key);
1838
1839                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1840                         chunk = (struct btrfs_chunk *)(ptr + len);
1841                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1842                         len += btrfs_chunk_item_size(num_stripes);
1843                 } else {
1844                         ret = -EIO;
1845                         break;
1846                 }
1847                 if (key.objectid == chunk_objectid &&
1848                     key.offset == chunk_offset) {
1849                         memmove(ptr, ptr + len, array_size - (cur + len));
1850                         array_size -= len;
1851                         btrfs_set_super_sys_array_size(super_copy, array_size);
1852                 } else {
1853                         ptr += len;
1854                         cur += len;
1855                 }
1856         }
1857         return ret;
1858 }
1859
1860 static int btrfs_relocate_chunk(struct btrfs_root *root,
1861                          u64 chunk_tree, u64 chunk_objectid,
1862                          u64 chunk_offset)
1863 {
1864         struct extent_map_tree *em_tree;
1865         struct btrfs_root *extent_root;
1866         struct btrfs_trans_handle *trans;
1867         struct extent_map *em;
1868         struct map_lookup *map;
1869         int ret;
1870         int i;
1871
1872         root = root->fs_info->chunk_root;
1873         extent_root = root->fs_info->extent_root;
1874         em_tree = &root->fs_info->mapping_tree.map_tree;
1875
1876         ret = btrfs_can_relocate(extent_root, chunk_offset);
1877         if (ret)
1878                 return -ENOSPC;
1879
1880         /* step one, relocate all the extents inside this chunk */
1881         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1882         if (ret)
1883                 return ret;
1884
1885         trans = btrfs_start_transaction(root, 0);
1886         BUG_ON(IS_ERR(trans));
1887
1888         lock_chunks(root);
1889
1890         /*
1891          * step two, delete the device extents and the
1892          * chunk tree entries
1893          */
1894         read_lock(&em_tree->lock);
1895         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1896         read_unlock(&em_tree->lock);
1897
1898         BUG_ON(em->start > chunk_offset ||
1899                em->start + em->len < chunk_offset);
1900         map = (struct map_lookup *)em->bdev;
1901
1902         for (i = 0; i < map->num_stripes; i++) {
1903                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1904                                             map->stripes[i].physical);
1905                 BUG_ON(ret);
1906
1907                 if (map->stripes[i].dev) {
1908                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1909                         BUG_ON(ret);
1910                 }
1911         }
1912         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1913                                chunk_offset);
1914
1915         BUG_ON(ret);
1916
1917         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1918                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1919                 BUG_ON(ret);
1920         }
1921
1922         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1923         BUG_ON(ret);
1924
1925         write_lock(&em_tree->lock);
1926         remove_extent_mapping(em_tree, em);
1927         write_unlock(&em_tree->lock);
1928
1929         kfree(map);
1930         em->bdev = NULL;
1931
1932         /* once for the tree */
1933         free_extent_map(em);
1934         /* once for us */
1935         free_extent_map(em);
1936
1937         unlock_chunks(root);
1938         btrfs_end_transaction(trans, root);
1939         return 0;
1940 }
1941
1942 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1943 {
1944         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1945         struct btrfs_path *path;
1946         struct extent_buffer *leaf;
1947         struct btrfs_chunk *chunk;
1948         struct btrfs_key key;
1949         struct btrfs_key found_key;
1950         u64 chunk_tree = chunk_root->root_key.objectid;
1951         u64 chunk_type;
1952         bool retried = false;
1953         int failed = 0;
1954         int ret;
1955
1956         path = btrfs_alloc_path();
1957         if (!path)
1958                 return -ENOMEM;
1959
1960 again:
1961         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1962         key.offset = (u64)-1;
1963         key.type = BTRFS_CHUNK_ITEM_KEY;
1964
1965         while (1) {
1966                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1967                 if (ret < 0)
1968                         goto error;
1969                 BUG_ON(ret == 0);
1970
1971                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1972                                           key.type);
1973                 if (ret < 0)
1974                         goto error;
1975                 if (ret > 0)
1976                         break;
1977
1978                 leaf = path->nodes[0];
1979                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1980
1981                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1982                                        struct btrfs_chunk);
1983                 chunk_type = btrfs_chunk_type(leaf, chunk);
1984                 btrfs_release_path(chunk_root, path);
1985
1986                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1987                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1988                                                    found_key.objectid,
1989                                                    found_key.offset);
1990                         if (ret == -ENOSPC)
1991                                 failed++;
1992                         else if (ret)
1993                                 BUG();
1994                 }
1995
1996                 if (found_key.offset == 0)
1997                         break;
1998                 key.offset = found_key.offset - 1;
1999         }
2000         ret = 0;
2001         if (failed && !retried) {
2002                 failed = 0;
2003                 retried = true;
2004                 goto again;
2005         } else if (failed && retried) {
2006                 WARN_ON(1);
2007                 ret = -ENOSPC;
2008         }
2009 error:
2010         btrfs_free_path(path);
2011         return ret;
2012 }
2013
2014 static u64 div_factor(u64 num, int factor)
2015 {
2016         if (factor == 10)
2017                 return num;
2018         num *= factor;
2019         do_div(num, 10);
2020         return num;
2021 }
2022
2023 int btrfs_balance(struct btrfs_root *dev_root)
2024 {
2025         int ret;
2026         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2027         struct btrfs_device *device;
2028         u64 old_size;
2029         u64 size_to_free;
2030         struct btrfs_path *path;
2031         struct btrfs_key key;
2032         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2033         struct btrfs_trans_handle *trans;
2034         struct btrfs_key found_key;
2035
2036         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2037                 return -EROFS;
2038
2039         if (!capable(CAP_SYS_ADMIN))
2040                 return -EPERM;
2041
2042         mutex_lock(&dev_root->fs_info->volume_mutex);
2043         dev_root = dev_root->fs_info->dev_root;
2044
2045         /* step one make some room on all the devices */
2046         list_for_each_entry(device, devices, dev_list) {
2047                 old_size = device->total_bytes;
2048                 size_to_free = div_factor(old_size, 1);
2049                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2050                 if (!device->writeable ||
2051                     device->total_bytes - device->bytes_used > size_to_free)
2052                         continue;
2053
2054                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2055                 if (ret == -ENOSPC)
2056                         break;
2057                 BUG_ON(ret);
2058
2059                 trans = btrfs_start_transaction(dev_root, 0);
2060                 BUG_ON(IS_ERR(trans));
2061
2062                 ret = btrfs_grow_device(trans, device, old_size);
2063                 BUG_ON(ret);
2064
2065                 btrfs_end_transaction(trans, dev_root);
2066         }
2067
2068         /* step two, relocate all the chunks */
2069         path = btrfs_alloc_path();
2070         BUG_ON(!path);
2071
2072         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2073         key.offset = (u64)-1;
2074         key.type = BTRFS_CHUNK_ITEM_KEY;
2075
2076         while (1) {
2077                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2078                 if (ret < 0)
2079                         goto error;
2080
2081                 /*
2082                  * this shouldn't happen, it means the last relocate
2083                  * failed
2084                  */
2085                 if (ret == 0)
2086                         break;
2087
2088                 ret = btrfs_previous_item(chunk_root, path, 0,
2089                                           BTRFS_CHUNK_ITEM_KEY);
2090                 if (ret)
2091                         break;
2092
2093                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2094                                       path->slots[0]);
2095                 if (found_key.objectid != key.objectid)
2096                         break;
2097
2098                 /* chunk zero is special */
2099                 if (found_key.offset == 0)
2100                         break;
2101
2102                 btrfs_release_path(chunk_root, path);
2103                 ret = btrfs_relocate_chunk(chunk_root,
2104                                            chunk_root->root_key.objectid,
2105                                            found_key.objectid,
2106                                            found_key.offset);
2107                 BUG_ON(ret && ret != -ENOSPC);
2108                 key.offset = found_key.offset - 1;
2109         }
2110         ret = 0;
2111 error:
2112         btrfs_free_path(path);
2113         mutex_unlock(&dev_root->fs_info->volume_mutex);
2114         return ret;
2115 }
2116
2117 /*
2118  * shrinking a device means finding all of the device extents past
2119  * the new size, and then following the back refs to the chunks.
2120  * The chunk relocation code actually frees the device extent
2121  */
2122 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2123 {
2124         struct btrfs_trans_handle *trans;
2125         struct btrfs_root *root = device->dev_root;
2126         struct btrfs_dev_extent *dev_extent = NULL;
2127         struct btrfs_path *path;
2128         u64 length;
2129         u64 chunk_tree;
2130         u64 chunk_objectid;
2131         u64 chunk_offset;
2132         int ret;
2133         int slot;
2134         int failed = 0;
2135         bool retried = false;
2136         struct extent_buffer *l;
2137         struct btrfs_key key;
2138         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2139         u64 old_total = btrfs_super_total_bytes(super_copy);
2140         u64 old_size = device->total_bytes;
2141         u64 diff = device->total_bytes - new_size;
2142
2143         if (new_size >= device->total_bytes)
2144                 return -EINVAL;
2145
2146         path = btrfs_alloc_path();
2147         if (!path)
2148                 return -ENOMEM;
2149
2150         path->reada = 2;
2151
2152         lock_chunks(root);
2153
2154         device->total_bytes = new_size;
2155         if (device->writeable)
2156                 device->fs_devices->total_rw_bytes -= diff;
2157         unlock_chunks(root);
2158
2159 again:
2160         key.objectid = device->devid;
2161         key.offset = (u64)-1;
2162         key.type = BTRFS_DEV_EXTENT_KEY;
2163
2164         while (1) {
2165                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2166                 if (ret < 0)
2167                         goto done;
2168
2169                 ret = btrfs_previous_item(root, path, 0, key.type);
2170                 if (ret < 0)
2171                         goto done;
2172                 if (ret) {
2173                         ret = 0;
2174                         btrfs_release_path(root, path);
2175                         break;
2176                 }
2177
2178                 l = path->nodes[0];
2179                 slot = path->slots[0];
2180                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2181
2182                 if (key.objectid != device->devid) {
2183                         btrfs_release_path(root, path);
2184                         break;
2185                 }
2186
2187                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2188                 length = btrfs_dev_extent_length(l, dev_extent);
2189
2190                 if (key.offset + length <= new_size) {
2191                         btrfs_release_path(root, path);
2192                         break;
2193                 }
2194
2195                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2196                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2197                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2198                 btrfs_release_path(root, path);
2199
2200                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2201                                            chunk_offset);
2202                 if (ret && ret != -ENOSPC)
2203                         goto done;
2204                 if (ret == -ENOSPC)
2205                         failed++;
2206                 key.offset -= 1;
2207         }
2208
2209         if (failed && !retried) {
2210                 failed = 0;
2211                 retried = true;
2212                 goto again;
2213         } else if (failed && retried) {
2214                 ret = -ENOSPC;
2215                 lock_chunks(root);
2216
2217                 device->total_bytes = old_size;
2218                 if (device->writeable)
2219                         device->fs_devices->total_rw_bytes += diff;
2220                 unlock_chunks(root);
2221                 goto done;
2222         }
2223
2224         /* Shrinking succeeded, else we would be at "done". */
2225         trans = btrfs_start_transaction(root, 0);
2226         if (IS_ERR(trans)) {
2227                 ret = PTR_ERR(trans);
2228                 goto done;
2229         }
2230
2231         lock_chunks(root);
2232
2233         device->disk_total_bytes = new_size;
2234         /* Now btrfs_update_device() will change the on-disk size. */
2235         ret = btrfs_update_device(trans, device);
2236         if (ret) {
2237                 unlock_chunks(root);
2238                 btrfs_end_transaction(trans, root);
2239                 goto done;
2240         }
2241         WARN_ON(diff > old_total);
2242         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2243         unlock_chunks(root);
2244         btrfs_end_transaction(trans, root);
2245 done:
2246         btrfs_free_path(path);
2247         return ret;
2248 }
2249
2250 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2251                            struct btrfs_root *root,
2252                            struct btrfs_key *key,
2253                            struct btrfs_chunk *chunk, int item_size)
2254 {
2255         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2256         struct btrfs_disk_key disk_key;
2257         u32 array_size;
2258         u8 *ptr;
2259
2260         array_size = btrfs_super_sys_array_size(super_copy);
2261         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2262                 return -EFBIG;
2263
2264         ptr = super_copy->sys_chunk_array + array_size;
2265         btrfs_cpu_key_to_disk(&disk_key, key);
2266         memcpy(ptr, &disk_key, sizeof(disk_key));
2267         ptr += sizeof(disk_key);
2268         memcpy(ptr, chunk, item_size);
2269         item_size += sizeof(disk_key);
2270         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2271         return 0;
2272 }
2273
2274 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2275                                         int num_stripes, int sub_stripes)
2276 {
2277         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2278                 return calc_size;
2279         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2280                 return calc_size * (num_stripes / sub_stripes);
2281         else
2282                 return calc_size * num_stripes;
2283 }
2284
2285 /* Used to sort the devices by max_avail(descending sort) */
2286 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2287 {
2288         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2289             ((struct btrfs_device_info *)dev_info2)->max_avail)
2290                 return -1;
2291         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2292                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2293                 return 1;
2294         else
2295                 return 0;
2296 }
2297
2298 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2299                                  int *num_stripes, int *min_stripes,
2300                                  int *sub_stripes)
2301 {
2302         *num_stripes = 1;
2303         *min_stripes = 1;
2304         *sub_stripes = 0;
2305
2306         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2307                 *num_stripes = fs_devices->rw_devices;
2308                 *min_stripes = 2;
2309         }
2310         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2311                 *num_stripes = 2;
2312                 *min_stripes = 2;
2313         }
2314         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2315                 if (fs_devices->rw_devices < 2)
2316                         return -ENOSPC;
2317                 *num_stripes = 2;
2318                 *min_stripes = 2;
2319         }
2320         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2321                 *num_stripes = fs_devices->rw_devices;
2322                 if (*num_stripes < 4)
2323                         return -ENOSPC;
2324                 *num_stripes &= ~(u32)1;
2325                 *sub_stripes = 2;
2326                 *min_stripes = 4;
2327         }
2328
2329         return 0;
2330 }
2331
2332 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2333                                     u64 proposed_size, u64 type,
2334                                     int num_stripes, int small_stripe)
2335 {
2336         int min_stripe_size = 1 * 1024 * 1024;
2337         u64 calc_size = proposed_size;
2338         u64 max_chunk_size = calc_size;
2339         int ncopies = 1;
2340
2341         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2342                     BTRFS_BLOCK_GROUP_DUP |
2343                     BTRFS_BLOCK_GROUP_RAID10))
2344                 ncopies = 2;
2345
2346         if (type & BTRFS_BLOCK_GROUP_DATA) {
2347                 max_chunk_size = 10 * calc_size;
2348                 min_stripe_size = 64 * 1024 * 1024;
2349         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2350                 max_chunk_size = 256 * 1024 * 1024;
2351                 min_stripe_size = 32 * 1024 * 1024;
2352         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2353                 calc_size = 8 * 1024 * 1024;
2354                 max_chunk_size = calc_size * 2;
2355                 min_stripe_size = 1 * 1024 * 1024;
2356         }
2357
2358         /* we don't want a chunk larger than 10% of writeable space */
2359         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2360                              max_chunk_size);
2361
2362         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2363                 calc_size = max_chunk_size * ncopies;
2364                 do_div(calc_size, num_stripes);
2365                 do_div(calc_size, BTRFS_STRIPE_LEN);
2366                 calc_size *= BTRFS_STRIPE_LEN;
2367         }
2368
2369         /* we don't want tiny stripes */
2370         if (!small_stripe)
2371                 calc_size = max_t(u64, min_stripe_size, calc_size);
2372
2373         /*
2374          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2375          * we end up with something bigger than a stripe
2376          */
2377         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2378
2379         do_div(calc_size, BTRFS_STRIPE_LEN);
2380         calc_size *= BTRFS_STRIPE_LEN;
2381
2382         return calc_size;
2383 }
2384
2385 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2386                                                       int num_stripes)
2387 {
2388         struct map_lookup *new;
2389         size_t len = map_lookup_size(num_stripes);
2390
2391         BUG_ON(map->num_stripes < num_stripes);
2392
2393         if (map->num_stripes == num_stripes)
2394                 return map;
2395
2396         new = kmalloc(len, GFP_NOFS);
2397         if (!new) {
2398                 /* just change map->num_stripes */
2399                 map->num_stripes = num_stripes;
2400                 return map;
2401         }
2402
2403         memcpy(new, map, len);
2404         new->num_stripes = num_stripes;
2405         kfree(map);
2406         return new;
2407 }
2408
2409 /*
2410  * helper to allocate device space from btrfs_device_info, in which we stored
2411  * max free space information of every device. It is used when we can not
2412  * allocate chunks by default size.
2413  *
2414  * By this helper, we can allocate a new chunk as larger as possible.
2415  */
2416 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2417                                     struct btrfs_fs_devices *fs_devices,
2418                                     struct btrfs_device_info *devices,
2419                                     int nr_device, u64 type,
2420                                     struct map_lookup **map_lookup,
2421                                     int min_stripes, u64 *stripe_size)
2422 {
2423         int i, index, sort_again = 0;
2424         int min_devices = min_stripes;
2425         u64 max_avail, min_free;
2426         struct map_lookup *map = *map_lookup;
2427         int ret;
2428
2429         if (nr_device < min_stripes)
2430                 return -ENOSPC;
2431
2432         btrfs_descending_sort_devices(devices, nr_device);
2433
2434         max_avail = devices[0].max_avail;
2435         if (!max_avail)
2436                 return -ENOSPC;
2437
2438         for (i = 0; i < nr_device; i++) {
2439                 /*
2440                  * if dev_offset = 0, it means the free space of this device
2441                  * is less than what we need, and we didn't search max avail
2442                  * extent on this device, so do it now.
2443                  */
2444                 if (!devices[i].dev_offset) {
2445                         ret = find_free_dev_extent(trans, devices[i].dev,
2446                                                    max_avail,
2447                                                    &devices[i].dev_offset,
2448                                                    &devices[i].max_avail);
2449                         if (ret != 0 && ret != -ENOSPC)
2450                                 return ret;
2451                         sort_again = 1;
2452                 }
2453         }
2454
2455         /* we update the max avail free extent of each devices, sort again */
2456         if (sort_again)
2457                 btrfs_descending_sort_devices(devices, nr_device);
2458
2459         if (type & BTRFS_BLOCK_GROUP_DUP)
2460                 min_devices = 1;
2461
2462         if (!devices[min_devices - 1].max_avail)
2463                 return -ENOSPC;
2464
2465         max_avail = devices[min_devices - 1].max_avail;
2466         if (type & BTRFS_BLOCK_GROUP_DUP)
2467                 do_div(max_avail, 2);
2468
2469         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2470                                              min_stripes, 1);
2471         if (type & BTRFS_BLOCK_GROUP_DUP)
2472                 min_free = max_avail * 2;
2473         else
2474                 min_free = max_avail;
2475
2476         if (min_free > devices[min_devices - 1].max_avail)
2477                 return -ENOSPC;
2478
2479         map = __shrink_map_lookup_stripes(map, min_stripes);
2480         *stripe_size = max_avail;
2481
2482         index = 0;
2483         for (i = 0; i < min_stripes; i++) {
2484                 map->stripes[i].dev = devices[index].dev;
2485                 map->stripes[i].physical = devices[index].dev_offset;
2486                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2487                         i++;
2488                         map->stripes[i].dev = devices[index].dev;
2489                         map->stripes[i].physical = devices[index].dev_offset +
2490                                                    max_avail;
2491                 }
2492                 index++;
2493         }
2494         *map_lookup = map;
2495
2496         return 0;
2497 }
2498
2499 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2500                                struct btrfs_root *extent_root,
2501                                struct map_lookup **map_ret,
2502                                u64 *num_bytes, u64 *stripe_size,
2503                                u64 start, u64 type)
2504 {
2505         struct btrfs_fs_info *info = extent_root->fs_info;
2506         struct btrfs_device *device = NULL;
2507         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2508         struct list_head *cur;
2509         struct map_lookup *map;
2510         struct extent_map_tree *em_tree;
2511         struct extent_map *em;
2512         struct btrfs_device_info *devices_info;
2513         struct list_head private_devs;
2514         u64 calc_size = 1024 * 1024 * 1024;
2515         u64 min_free;
2516         u64 avail;
2517         u64 dev_offset;
2518         int num_stripes;
2519         int min_stripes;
2520         int sub_stripes;
2521         int min_devices;        /* the min number of devices we need */
2522         int i;
2523         int ret;
2524         int index;
2525
2526         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2527             (type & BTRFS_BLOCK_GROUP_DUP)) {
2528                 WARN_ON(1);
2529                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2530         }
2531         if (list_empty(&fs_devices->alloc_list))
2532                 return -ENOSPC;
2533
2534         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2535                                     &min_stripes, &sub_stripes);
2536         if (ret)
2537                 return ret;
2538
2539         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2540                                GFP_NOFS);
2541         if (!devices_info)
2542                 return -ENOMEM;
2543
2544         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2545         if (!map) {
2546                 ret = -ENOMEM;
2547                 goto error;
2548         }
2549         map->num_stripes = num_stripes;
2550
2551         cur = fs_devices->alloc_list.next;
2552         index = 0;
2553         i = 0;
2554
2555         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2556                                              num_stripes, 0);
2557
2558         if (type & BTRFS_BLOCK_GROUP_DUP) {
2559                 min_free = calc_size * 2;
2560                 min_devices = 1;
2561         } else {
2562                 min_free = calc_size;
2563                 min_devices = min_stripes;
2564         }
2565
2566         INIT_LIST_HEAD(&private_devs);
2567         while (index < num_stripes) {
2568                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2569                 BUG_ON(!device->writeable);
2570                 if (device->total_bytes > device->bytes_used)
2571                         avail = device->total_bytes - device->bytes_used;
2572                 else
2573                         avail = 0;
2574                 cur = cur->next;
2575
2576                 if (device->in_fs_metadata && avail >= min_free) {
2577                         ret = find_free_dev_extent(trans, device, min_free,
2578                                                    &devices_info[i].dev_offset,
2579                                                    &devices_info[i].max_avail);
2580                         if (ret == 0) {
2581                                 list_move_tail(&device->dev_alloc_list,
2582                                                &private_devs);
2583                                 map->stripes[index].dev = device;
2584                                 map->stripes[index].physical =
2585                                                 devices_info[i].dev_offset;
2586                                 index++;
2587                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2588                                         map->stripes[index].dev = device;
2589                                         map->stripes[index].physical =
2590                                                 devices_info[i].dev_offset +
2591                                                 calc_size;
2592                                         index++;
2593                                 }
2594                         } else if (ret != -ENOSPC)
2595                                 goto error;
2596
2597                         devices_info[i].dev = device;
2598                         i++;
2599                 } else if (device->in_fs_metadata &&
2600                            avail >= BTRFS_STRIPE_LEN) {
2601                         devices_info[i].dev = device;
2602                         devices_info[i].max_avail = avail;
2603                         i++;
2604                 }
2605
2606                 if (cur == &fs_devices->alloc_list)
2607                         break;
2608         }
2609
2610         list_splice(&private_devs, &fs_devices->alloc_list);
2611         if (index < num_stripes) {
2612                 if (index >= min_stripes) {
2613                         num_stripes = index;
2614                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2615                                 num_stripes /= sub_stripes;
2616                                 num_stripes *= sub_stripes;
2617                         }
2618
2619                         map = __shrink_map_lookup_stripes(map, num_stripes);
2620                 } else if (i >= min_devices) {
2621                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2622                                                        devices_info, i, type,
2623                                                        &map, min_stripes,
2624                                                        &calc_size);
2625                         if (ret)
2626                                 goto error;
2627                 } else {
2628                         ret = -ENOSPC;
2629                         goto error;
2630                 }
2631         }
2632         map->sector_size = extent_root->sectorsize;
2633         map->stripe_len = BTRFS_STRIPE_LEN;
2634         map->io_align = BTRFS_STRIPE_LEN;
2635         map->io_width = BTRFS_STRIPE_LEN;
2636         map->type = type;
2637         map->sub_stripes = sub_stripes;
2638
2639         *map_ret = map;
2640         *stripe_size = calc_size;
2641         *num_bytes = chunk_bytes_by_type(type, calc_size,
2642                                          map->num_stripes, sub_stripes);
2643
2644         em = alloc_extent_map(GFP_NOFS);
2645         if (!em) {
2646                 ret = -ENOMEM;
2647                 goto error;
2648         }
2649         em->bdev = (struct block_device *)map;
2650         em->start = start;
2651         em->len = *num_bytes;
2652         em->block_start = 0;
2653         em->block_len = em->len;
2654
2655         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2656         write_lock(&em_tree->lock);
2657         ret = add_extent_mapping(em_tree, em);
2658         write_unlock(&em_tree->lock);
2659         BUG_ON(ret);
2660         free_extent_map(em);
2661
2662         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2663                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2664                                      start, *num_bytes);
2665         BUG_ON(ret);
2666
2667         index = 0;
2668         while (index < map->num_stripes) {
2669                 device = map->stripes[index].dev;
2670                 dev_offset = map->stripes[index].physical;
2671
2672                 ret = btrfs_alloc_dev_extent(trans, device,
2673                                 info->chunk_root->root_key.objectid,
2674                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2675                                 start, dev_offset, calc_size);
2676                 BUG_ON(ret);
2677                 index++;
2678         }
2679
2680         kfree(devices_info);
2681         return 0;
2682
2683 error:
2684         kfree(map);
2685         kfree(devices_info);
2686         return ret;
2687 }
2688
2689 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2690                                 struct btrfs_root *extent_root,
2691                                 struct map_lookup *map, u64 chunk_offset,
2692                                 u64 chunk_size, u64 stripe_size)
2693 {
2694         u64 dev_offset;
2695         struct btrfs_key key;
2696         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2697         struct btrfs_device *device;
2698         struct btrfs_chunk *chunk;
2699         struct btrfs_stripe *stripe;
2700         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2701         int index = 0;
2702         int ret;
2703
2704         chunk = kzalloc(item_size, GFP_NOFS);
2705         if (!chunk)
2706                 return -ENOMEM;
2707
2708         index = 0;
2709         while (index < map->num_stripes) {
2710                 device = map->stripes[index].dev;
2711                 device->bytes_used += stripe_size;
2712                 ret = btrfs_update_device(trans, device);
2713                 BUG_ON(ret);
2714                 index++;
2715         }
2716
2717         index = 0;
2718         stripe = &chunk->stripe;
2719         while (index < map->num_stripes) {
2720                 device = map->stripes[index].dev;
2721                 dev_offset = map->stripes[index].physical;
2722
2723                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2724                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2725                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2726                 stripe++;
2727                 index++;
2728         }
2729
2730         btrfs_set_stack_chunk_length(chunk, chunk_size);
2731         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2732         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2733         btrfs_set_stack_chunk_type(chunk, map->type);
2734         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2735         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2736         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2737         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2738         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2739
2740         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2741         key.type = BTRFS_CHUNK_ITEM_KEY;
2742         key.offset = chunk_offset;
2743
2744         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2745         BUG_ON(ret);
2746
2747         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2748                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2749                                              item_size);
2750                 BUG_ON(ret);
2751         }
2752         kfree(chunk);
2753         return 0;
2754 }
2755
2756 /*
2757  * Chunk allocation falls into two parts. The first part does works
2758  * that make the new allocated chunk useable, but not do any operation
2759  * that modifies the chunk tree. The second part does the works that
2760  * require modifying the chunk tree. This division is important for the
2761  * bootstrap process of adding storage to a seed btrfs.
2762  */
2763 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2764                       struct btrfs_root *extent_root, u64 type)
2765 {
2766         u64 chunk_offset;
2767         u64 chunk_size;
2768         u64 stripe_size;
2769         struct map_lookup *map;
2770         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2771         int ret;
2772
2773         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2774                               &chunk_offset);
2775         if (ret)
2776                 return ret;
2777
2778         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2779                                   &stripe_size, chunk_offset, type);
2780         if (ret)
2781                 return ret;
2782
2783         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2784                                    chunk_size, stripe_size);
2785         BUG_ON(ret);
2786         return 0;
2787 }
2788
2789 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2790                                          struct btrfs_root *root,
2791                                          struct btrfs_device *device)
2792 {
2793         u64 chunk_offset;
2794         u64 sys_chunk_offset;
2795         u64 chunk_size;
2796         u64 sys_chunk_size;
2797         u64 stripe_size;
2798         u64 sys_stripe_size;
2799         u64 alloc_profile;
2800         struct map_lookup *map;
2801         struct map_lookup *sys_map;
2802         struct btrfs_fs_info *fs_info = root->fs_info;
2803         struct btrfs_root *extent_root = fs_info->extent_root;
2804         int ret;
2805
2806         ret = find_next_chunk(fs_info->chunk_root,
2807                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2808         BUG_ON(ret);
2809
2810         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2811                         (fs_info->metadata_alloc_profile &
2812                          fs_info->avail_metadata_alloc_bits);
2813         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2814
2815         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2816                                   &stripe_size, chunk_offset, alloc_profile);
2817         BUG_ON(ret);
2818
2819         sys_chunk_offset = chunk_offset + chunk_size;
2820
2821         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2822                         (fs_info->system_alloc_profile &
2823                          fs_info->avail_system_alloc_bits);
2824         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2825
2826         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2827                                   &sys_chunk_size, &sys_stripe_size,
2828                                   sys_chunk_offset, alloc_profile);
2829         BUG_ON(ret);
2830
2831         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2832         BUG_ON(ret);
2833
2834         /*
2835          * Modifying chunk tree needs allocating new blocks from both
2836          * system block group and metadata block group. So we only can
2837          * do operations require modifying the chunk tree after both
2838          * block groups were created.
2839          */
2840         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2841                                    chunk_size, stripe_size);
2842         BUG_ON(ret);
2843
2844         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2845                                    sys_chunk_offset, sys_chunk_size,
2846                                    sys_stripe_size);
2847         BUG_ON(ret);
2848         return 0;
2849 }
2850
2851 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2852 {
2853         struct extent_map *em;
2854         struct map_lookup *map;
2855         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2856         int readonly = 0;
2857         int i;
2858
2859         read_lock(&map_tree->map_tree.lock);
2860         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2861         read_unlock(&map_tree->map_tree.lock);
2862         if (!em)
2863                 return 1;
2864
2865         if (btrfs_test_opt(root, DEGRADED)) {
2866                 free_extent_map(em);
2867                 return 0;
2868         }
2869
2870         map = (struct map_lookup *)em->bdev;
2871         for (i = 0; i < map->num_stripes; i++) {
2872                 if (!map->stripes[i].dev->writeable) {
2873                         readonly = 1;
2874                         break;
2875                 }
2876         }
2877         free_extent_map(em);
2878         return readonly;
2879 }
2880
2881 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2882 {
2883         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2884 }
2885
2886 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2887 {
2888         struct extent_map *em;
2889
2890         while (1) {
2891                 write_lock(&tree->map_tree.lock);
2892                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2893                 if (em)
2894                         remove_extent_mapping(&tree->map_tree, em);
2895                 write_unlock(&tree->map_tree.lock);
2896                 if (!em)
2897                         break;
2898                 kfree(em->bdev);
2899                 /* once for us */
2900                 free_extent_map(em);
2901                 /* once for the tree */
2902                 free_extent_map(em);
2903         }
2904 }
2905
2906 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2907 {
2908         struct extent_map *em;
2909         struct map_lookup *map;
2910         struct extent_map_tree *em_tree = &map_tree->map_tree;
2911         int ret;
2912
2913         read_lock(&em_tree->lock);
2914         em = lookup_extent_mapping(em_tree, logical, len);
2915         read_unlock(&em_tree->lock);
2916         BUG_ON(!em);
2917
2918         BUG_ON(em->start > logical || em->start + em->len < logical);
2919         map = (struct map_lookup *)em->bdev;
2920         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2921                 ret = map->num_stripes;
2922         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2923                 ret = map->sub_stripes;
2924         else
2925                 ret = 1;
2926         free_extent_map(em);
2927         return ret;
2928 }
2929
2930 static int find_live_mirror(struct map_lookup *map, int first, int num,
2931                             int optimal)
2932 {
2933         int i;
2934         if (map->stripes[optimal].dev->bdev)
2935                 return optimal;
2936         for (i = first; i < first + num; i++) {
2937                 if (map->stripes[i].dev->bdev)
2938                         return i;
2939         }
2940         /* we couldn't find one that doesn't fail.  Just return something
2941          * and the io error handling code will clean up eventually
2942          */
2943         return optimal;
2944 }
2945
2946 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2947                              u64 logical, u64 *length,
2948                              struct btrfs_multi_bio **multi_ret,
2949                              int mirror_num, struct page *unplug_page)
2950 {
2951         struct extent_map *em;
2952         struct map_lookup *map;
2953         struct extent_map_tree *em_tree = &map_tree->map_tree;
2954         u64 offset;
2955         u64 stripe_offset;
2956         u64 stripe_nr;
2957         int stripes_allocated = 8;
2958         int stripes_required = 1;
2959         int stripe_index;
2960         int i;
2961         int num_stripes;
2962         int max_errors = 0;
2963         struct btrfs_multi_bio *multi = NULL;
2964
2965         if (multi_ret && !(rw & REQ_WRITE))
2966                 stripes_allocated = 1;
2967 again:
2968         if (multi_ret) {
2969                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2970                                 GFP_NOFS);
2971                 if (!multi)
2972                         return -ENOMEM;
2973
2974                 atomic_set(&multi->error, 0);
2975         }
2976
2977         read_lock(&em_tree->lock);
2978         em = lookup_extent_mapping(em_tree, logical, *length);
2979         read_unlock(&em_tree->lock);
2980
2981         if (!em && unplug_page) {
2982                 kfree(multi);
2983                 return 0;
2984         }
2985
2986         if (!em) {
2987                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2988                        (unsigned long long)logical,
2989                        (unsigned long long)*length);
2990                 BUG();
2991         }
2992
2993         BUG_ON(em->start > logical || em->start + em->len < logical);
2994         map = (struct map_lookup *)em->bdev;
2995         offset = logical - em->start;
2996
2997         if (mirror_num > map->num_stripes)
2998                 mirror_num = 0;
2999
3000         /* if our multi bio struct is too small, back off and try again */
3001         if (rw & REQ_WRITE) {
3002                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3003                                  BTRFS_BLOCK_GROUP_DUP)) {
3004                         stripes_required = map->num_stripes;
3005                         max_errors = 1;
3006                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3007                         stripes_required = map->sub_stripes;
3008                         max_errors = 1;
3009                 }
3010         }
3011         if (multi_ret && (rw & REQ_WRITE) &&
3012             stripes_allocated < stripes_required) {
3013                 stripes_allocated = map->num_stripes;
3014                 free_extent_map(em);
3015                 kfree(multi);
3016                 goto again;
3017         }
3018         stripe_nr = offset;
3019         /*
3020          * stripe_nr counts the total number of stripes we have to stride
3021          * to get to this block
3022          */
3023         do_div(stripe_nr, map->stripe_len);
3024
3025         stripe_offset = stripe_nr * map->stripe_len;
3026         BUG_ON(offset < stripe_offset);
3027
3028         /* stripe_offset is the offset of this block in its stripe*/
3029         stripe_offset = offset - stripe_offset;
3030
3031         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3032                          BTRFS_BLOCK_GROUP_RAID10 |
3033                          BTRFS_BLOCK_GROUP_DUP)) {
3034                 /* we limit the length of each bio to what fits in a stripe */
3035                 *length = min_t(u64, em->len - offset,
3036                               map->stripe_len - stripe_offset);
3037         } else {
3038                 *length = em->len - offset;
3039         }
3040
3041         if (!multi_ret && !unplug_page)
3042                 goto out;
3043
3044         num_stripes = 1;
3045         stripe_index = 0;
3046         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3047                 if (unplug_page || (rw & REQ_WRITE))
3048                         num_stripes = map->num_stripes;
3049                 else if (mirror_num)
3050                         stripe_index = mirror_num - 1;
3051                 else {
3052                         stripe_index = find_live_mirror(map, 0,
3053                                             map->num_stripes,
3054                                             current->pid % map->num_stripes);
3055                 }
3056
3057         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3058                 if (rw & REQ_WRITE)
3059                         num_stripes = map->num_stripes;
3060                 else if (mirror_num)
3061                         stripe_index = mirror_num - 1;
3062
3063         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3064                 int factor = map->num_stripes / map->sub_stripes;
3065
3066                 stripe_index = do_div(stripe_nr, factor);
3067                 stripe_index *= map->sub_stripes;
3068
3069                 if (unplug_page || (rw & REQ_WRITE))
3070                         num_stripes = map->sub_stripes;
3071                 else if (mirror_num)
3072                         stripe_index += mirror_num - 1;
3073                 else {
3074                         stripe_index = find_live_mirror(map, stripe_index,
3075                                               map->sub_stripes, stripe_index +
3076                                               current->pid % map->sub_stripes);
3077                 }
3078         } else {
3079                 /*
3080                  * after this do_div call, stripe_nr is the number of stripes
3081                  * on this device we have to walk to find the data, and
3082                  * stripe_index is the number of our device in the stripe array
3083                  */
3084                 stripe_index = do_div(stripe_nr, map->num_stripes);
3085         }
3086         BUG_ON(stripe_index >= map->num_stripes);
3087
3088         for (i = 0; i < num_stripes; i++) {
3089                 if (unplug_page) {
3090                         struct btrfs_device *device;
3091                         struct backing_dev_info *bdi;
3092
3093                         device = map->stripes[stripe_index].dev;
3094                         if (device->bdev) {
3095                                 bdi = blk_get_backing_dev_info(device->bdev);
3096                                 if (bdi->unplug_io_fn)
3097                                         bdi->unplug_io_fn(bdi, unplug_page);
3098                         }
3099                 } else {
3100                         multi->stripes[i].physical =
3101                                 map->stripes[stripe_index].physical +
3102                                 stripe_offset + stripe_nr * map->stripe_len;
3103                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3104                 }
3105                 stripe_index++;
3106         }
3107         if (multi_ret) {
3108                 *multi_ret = multi;
3109                 multi->num_stripes = num_stripes;
3110                 multi->max_errors = max_errors;
3111         }
3112 out:
3113         free_extent_map(em);
3114         return 0;
3115 }
3116
3117 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3118                       u64 logical, u64 *length,
3119                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3120 {
3121         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3122                                  mirror_num, NULL);
3123 }
3124
3125 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3126                      u64 chunk_start, u64 physical, u64 devid,
3127                      u64 **logical, int *naddrs, int *stripe_len)
3128 {
3129         struct extent_map_tree *em_tree = &map_tree->map_tree;
3130         struct extent_map *em;
3131         struct map_lookup *map;
3132         u64 *buf;
3133         u64 bytenr;
3134         u64 length;
3135         u64 stripe_nr;
3136         int i, j, nr = 0;
3137
3138         read_lock(&em_tree->lock);
3139         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3140         read_unlock(&em_tree->lock);
3141
3142         BUG_ON(!em || em->start != chunk_start);
3143         map = (struct map_lookup *)em->bdev;
3144
3145         length = em->len;
3146         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3147                 do_div(length, map->num_stripes / map->sub_stripes);
3148         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3149                 do_div(length, map->num_stripes);
3150
3151         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3152         BUG_ON(!buf);
3153
3154         for (i = 0; i < map->num_stripes; i++) {
3155                 if (devid && map->stripes[i].dev->devid != devid)
3156                         continue;
3157                 if (map->stripes[i].physical > physical ||
3158                     map->stripes[i].physical + length <= physical)
3159                         continue;
3160
3161                 stripe_nr = physical - map->stripes[i].physical;
3162                 do_div(stripe_nr, map->stripe_len);
3163
3164                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3165                         stripe_nr = stripe_nr * map->num_stripes + i;
3166                         do_div(stripe_nr, map->sub_stripes);
3167                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3168                         stripe_nr = stripe_nr * map->num_stripes + i;
3169                 }
3170                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3171                 WARN_ON(nr >= map->num_stripes);
3172                 for (j = 0; j < nr; j++) {
3173                         if (buf[j] == bytenr)
3174                                 break;
3175                 }
3176                 if (j == nr) {
3177                         WARN_ON(nr >= map->num_stripes);
3178                         buf[nr++] = bytenr;
3179                 }
3180         }
3181
3182         *logical = buf;
3183         *naddrs = nr;
3184         *stripe_len = map->stripe_len;
3185
3186         free_extent_map(em);
3187         return 0;
3188 }
3189
3190 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3191                       u64 logical, struct page *page)
3192 {
3193         u64 length = PAGE_CACHE_SIZE;
3194         return __btrfs_map_block(map_tree, READ, logical, &length,
3195                                  NULL, 0, page);
3196 }
3197
3198 static void end_bio_multi_stripe(struct bio *bio, int err)
3199 {
3200         struct btrfs_multi_bio *multi = bio->bi_private;
3201         int is_orig_bio = 0;
3202
3203         if (err)
3204                 atomic_inc(&multi->error);
3205
3206         if (bio == multi->orig_bio)
3207                 is_orig_bio = 1;
3208
3209         if (atomic_dec_and_test(&multi->stripes_pending)) {
3210                 if (!is_orig_bio) {
3211                         bio_put(bio);
3212                         bio = multi->orig_bio;
3213                 }
3214                 bio->bi_private = multi->private;
3215                 bio->bi_end_io = multi->end_io;
3216                 /* only send an error to the higher layers if it is
3217                  * beyond the tolerance of the multi-bio
3218                  */
3219                 if (atomic_read(&multi->error) > multi->max_errors) {
3220                         err = -EIO;
3221                 } else if (err) {
3222                         /*
3223                          * this bio is actually up to date, we didn't
3224                          * go over the max number of errors
3225                          */
3226                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3227                         err = 0;
3228                 }
3229                 kfree(multi);
3230
3231                 bio_endio(bio, err);
3232         } else if (!is_orig_bio) {
3233                 bio_put(bio);
3234         }
3235 }
3236
3237 struct async_sched {
3238         struct bio *bio;
3239         int rw;
3240         struct btrfs_fs_info *info;
3241         struct btrfs_work work;
3242 };
3243
3244 /*
3245  * see run_scheduled_bios for a description of why bios are collected for
3246  * async submit.
3247  *
3248  * This will add one bio to the pending list for a device and make sure
3249  * the work struct is scheduled.
3250  */
3251 static noinline int schedule_bio(struct btrfs_root *root,
3252                                  struct btrfs_device *device,
3253                                  int rw, struct bio *bio)
3254 {
3255         int should_queue = 1;
3256         struct btrfs_pending_bios *pending_bios;
3257
3258         /* don't bother with additional async steps for reads, right now */
3259         if (!(rw & REQ_WRITE)) {
3260                 bio_get(bio);
3261                 submit_bio(rw, bio);
3262                 bio_put(bio);
3263                 return 0;
3264         }
3265
3266         /*
3267          * nr_async_bios allows us to reliably return congestion to the
3268          * higher layers.  Otherwise, the async bio makes it appear we have
3269          * made progress against dirty pages when we've really just put it
3270          * on a queue for later
3271          */
3272         atomic_inc(&root->fs_info->nr_async_bios);
3273         WARN_ON(bio->bi_next);
3274         bio->bi_next = NULL;
3275         bio->bi_rw |= rw;
3276
3277         spin_lock(&device->io_lock);
3278         if (bio->bi_rw & REQ_SYNC)
3279                 pending_bios = &device->pending_sync_bios;
3280         else
3281                 pending_bios = &device->pending_bios;
3282
3283         if (pending_bios->tail)
3284                 pending_bios->tail->bi_next = bio;
3285
3286         pending_bios->tail = bio;
3287         if (!pending_bios->head)
3288                 pending_bios->head = bio;
3289         if (device->running_pending)
3290                 should_queue = 0;
3291
3292         spin_unlock(&device->io_lock);
3293
3294         if (should_queue)
3295                 btrfs_queue_worker(&root->fs_info->submit_workers,
3296                                    &device->work);
3297         return 0;
3298 }
3299
3300 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3301                   int mirror_num, int async_submit)
3302 {
3303         struct btrfs_mapping_tree *map_tree;
3304         struct btrfs_device *dev;
3305         struct bio *first_bio = bio;
3306         u64 logical = (u64)bio->bi_sector << 9;
3307         u64 length = 0;
3308         u64 map_length;
3309         struct btrfs_multi_bio *multi = NULL;
3310         int ret;
3311         int dev_nr = 0;
3312         int total_devs = 1;
3313
3314         length = bio->bi_size;
3315         map_tree = &root->fs_info->mapping_tree;
3316         map_length = length;
3317
3318         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3319                               mirror_num);
3320         BUG_ON(ret);
3321
3322         total_devs = multi->num_stripes;
3323         if (map_length < length) {
3324                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3325                        "len %llu\n", (unsigned long long)logical,
3326                        (unsigned long long)length,
3327                        (unsigned long long)map_length);
3328                 BUG();
3329         }
3330         multi->end_io = first_bio->bi_end_io;
3331         multi->private = first_bio->bi_private;
3332         multi->orig_bio = first_bio;
3333         atomic_set(&multi->stripes_pending, multi->num_stripes);
3334
3335         while (dev_nr < total_devs) {
3336                 if (total_devs > 1) {