Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc
[pandora-kernel.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         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1217         key.type = BTRFS_DEV_ITEM_KEY;
1218         key.offset = device->devid;
1219         lock_chunks(root);
1220
1221         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1222         if (ret < 0)
1223                 goto out;
1224
1225         if (ret > 0) {
1226                 ret = -ENOENT;
1227                 goto out;
1228         }
1229
1230         ret = btrfs_del_item(trans, root, path);
1231         if (ret)
1232                 goto out;
1233 out:
1234         btrfs_free_path(path);
1235         unlock_chunks(root);
1236         btrfs_commit_transaction(trans, root);
1237         return ret;
1238 }
1239
1240 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1241 {
1242         struct btrfs_device *device;
1243         struct btrfs_device *next_device;
1244         struct block_device *bdev;
1245         struct buffer_head *bh = NULL;
1246         struct btrfs_super_block *disk_super;
1247         u64 all_avail;
1248         u64 devid;
1249         u64 num_devices;
1250         u8 *dev_uuid;
1251         int ret = 0;
1252
1253         mutex_lock(&uuid_mutex);
1254         mutex_lock(&root->fs_info->volume_mutex);
1255
1256         all_avail = root->fs_info->avail_data_alloc_bits |
1257                 root->fs_info->avail_system_alloc_bits |
1258                 root->fs_info->avail_metadata_alloc_bits;
1259
1260         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1261             root->fs_info->fs_devices->num_devices <= 4) {
1262                 printk(KERN_ERR "btrfs: unable to go below four devices "
1263                        "on raid10\n");
1264                 ret = -EINVAL;
1265                 goto out;
1266         }
1267
1268         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1269             root->fs_info->fs_devices->num_devices <= 2) {
1270                 printk(KERN_ERR "btrfs: unable to go below two "
1271                        "devices on raid1\n");
1272                 ret = -EINVAL;
1273                 goto out;
1274         }
1275
1276         if (strcmp(device_path, "missing") == 0) {
1277                 struct list_head *devices;
1278                 struct btrfs_device *tmp;
1279
1280                 device = NULL;
1281                 devices = &root->fs_info->fs_devices->devices;
1282                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1283                 list_for_each_entry(tmp, devices, dev_list) {
1284                         if (tmp->in_fs_metadata && !tmp->bdev) {
1285                                 device = tmp;
1286                                 break;
1287                         }
1288                 }
1289                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1290                 bdev = NULL;
1291                 bh = NULL;
1292                 disk_super = NULL;
1293                 if (!device) {
1294                         printk(KERN_ERR "btrfs: no missing devices found to "
1295                                "remove\n");
1296                         goto out;
1297                 }
1298         } else {
1299                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1300                                           root->fs_info->bdev_holder);
1301                 if (IS_ERR(bdev)) {
1302                         ret = PTR_ERR(bdev);
1303                         goto out;
1304                 }
1305
1306                 set_blocksize(bdev, 4096);
1307                 bh = btrfs_read_dev_super(bdev);
1308                 if (!bh) {
1309                         ret = -EINVAL;
1310                         goto error_close;
1311                 }
1312                 disk_super = (struct btrfs_super_block *)bh->b_data;
1313                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1314                 dev_uuid = disk_super->dev_item.uuid;
1315                 device = btrfs_find_device(root, devid, dev_uuid,
1316                                            disk_super->fsid);
1317                 if (!device) {
1318                         ret = -ENOENT;
1319                         goto error_brelse;
1320                 }
1321         }
1322
1323         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1324                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1325                        "device\n");
1326                 ret = -EINVAL;
1327                 goto error_brelse;
1328         }
1329
1330         if (device->writeable) {
1331                 list_del_init(&device->dev_alloc_list);
1332                 root->fs_info->fs_devices->rw_devices--;
1333         }
1334
1335         ret = btrfs_shrink_device(device, 0);
1336         if (ret)
1337                 goto error_brelse;
1338
1339         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1340         if (ret)
1341                 goto error_brelse;
1342
1343         device->in_fs_metadata = 0;
1344
1345         /*
1346          * the device list mutex makes sure that we don't change
1347          * the device list while someone else is writing out all
1348          * the device supers.
1349          */
1350         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1351         list_del_init(&device->dev_list);
1352         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1353
1354         device->fs_devices->num_devices--;
1355
1356         if (device->missing)
1357                 root->fs_info->fs_devices->missing_devices--;
1358
1359         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1360                                  struct btrfs_device, dev_list);
1361         if (device->bdev == root->fs_info->sb->s_bdev)
1362                 root->fs_info->sb->s_bdev = next_device->bdev;
1363         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1364                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1365
1366         if (device->bdev) {
1367                 blkdev_put(device->bdev, device->mode);
1368                 device->bdev = NULL;
1369                 device->fs_devices->open_devices--;
1370         }
1371
1372         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1373         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1374
1375         if (device->fs_devices->open_devices == 0) {
1376                 struct btrfs_fs_devices *fs_devices;
1377                 fs_devices = root->fs_info->fs_devices;
1378                 while (fs_devices) {
1379                         if (fs_devices->seed == device->fs_devices)
1380                                 break;
1381                         fs_devices = fs_devices->seed;
1382                 }
1383                 fs_devices->seed = device->fs_devices->seed;
1384                 device->fs_devices->seed = NULL;
1385                 __btrfs_close_devices(device->fs_devices);
1386                 free_fs_devices(device->fs_devices);
1387         }
1388
1389         /*
1390          * at this point, the device is zero sized.  We want to
1391          * remove it from the devices list and zero out the old super
1392          */
1393         if (device->writeable) {
1394                 /* make sure this device isn't detected as part of
1395                  * the FS anymore
1396                  */
1397                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1398                 set_buffer_dirty(bh);
1399                 sync_dirty_buffer(bh);
1400         }
1401
1402         kfree(device->name);
1403         kfree(device);
1404         ret = 0;
1405
1406 error_brelse:
1407         brelse(bh);
1408 error_close:
1409         if (bdev)
1410                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1411 out:
1412         mutex_unlock(&root->fs_info->volume_mutex);
1413         mutex_unlock(&uuid_mutex);
1414         return ret;
1415 }
1416
1417 /*
1418  * does all the dirty work required for changing file system's UUID.
1419  */
1420 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1421                                 struct btrfs_root *root)
1422 {
1423         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1424         struct btrfs_fs_devices *old_devices;
1425         struct btrfs_fs_devices *seed_devices;
1426         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1427         struct btrfs_device *device;
1428         u64 super_flags;
1429
1430         BUG_ON(!mutex_is_locked(&uuid_mutex));
1431         if (!fs_devices->seeding)
1432                 return -EINVAL;
1433
1434         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1435         if (!seed_devices)
1436                 return -ENOMEM;
1437
1438         old_devices = clone_fs_devices(fs_devices);
1439         if (IS_ERR(old_devices)) {
1440                 kfree(seed_devices);
1441                 return PTR_ERR(old_devices);
1442         }
1443
1444         list_add(&old_devices->list, &fs_uuids);
1445
1446         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1447         seed_devices->opened = 1;
1448         INIT_LIST_HEAD(&seed_devices->devices);
1449         INIT_LIST_HEAD(&seed_devices->alloc_list);
1450         mutex_init(&seed_devices->device_list_mutex);
1451         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1452         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1453         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1454                 device->fs_devices = seed_devices;
1455         }
1456
1457         fs_devices->seeding = 0;
1458         fs_devices->num_devices = 0;
1459         fs_devices->open_devices = 0;
1460         fs_devices->seed = seed_devices;
1461
1462         generate_random_uuid(fs_devices->fsid);
1463         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1464         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1465         super_flags = btrfs_super_flags(disk_super) &
1466                       ~BTRFS_SUPER_FLAG_SEEDING;
1467         btrfs_set_super_flags(disk_super, super_flags);
1468
1469         return 0;
1470 }
1471
1472 /*
1473  * strore the expected generation for seed devices in device items.
1474  */
1475 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1476                                struct btrfs_root *root)
1477 {
1478         struct btrfs_path *path;
1479         struct extent_buffer *leaf;
1480         struct btrfs_dev_item *dev_item;
1481         struct btrfs_device *device;
1482         struct btrfs_key key;
1483         u8 fs_uuid[BTRFS_UUID_SIZE];
1484         u8 dev_uuid[BTRFS_UUID_SIZE];
1485         u64 devid;
1486         int ret;
1487
1488         path = btrfs_alloc_path();
1489         if (!path)
1490                 return -ENOMEM;
1491
1492         root = root->fs_info->chunk_root;
1493         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1494         key.offset = 0;
1495         key.type = BTRFS_DEV_ITEM_KEY;
1496
1497         while (1) {
1498                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1499                 if (ret < 0)
1500                         goto error;
1501
1502                 leaf = path->nodes[0];
1503 next_slot:
1504                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1505                         ret = btrfs_next_leaf(root, path);
1506                         if (ret > 0)
1507                                 break;
1508                         if (ret < 0)
1509                                 goto error;
1510                         leaf = path->nodes[0];
1511                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512                         btrfs_release_path(root, path);
1513                         continue;
1514                 }
1515
1516                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1517                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1518                     key.type != BTRFS_DEV_ITEM_KEY)
1519                         break;
1520
1521                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1522                                           struct btrfs_dev_item);
1523                 devid = btrfs_device_id(leaf, dev_item);
1524                 read_extent_buffer(leaf, dev_uuid,
1525                                    (unsigned long)btrfs_device_uuid(dev_item),
1526                                    BTRFS_UUID_SIZE);
1527                 read_extent_buffer(leaf, fs_uuid,
1528                                    (unsigned long)btrfs_device_fsid(dev_item),
1529                                    BTRFS_UUID_SIZE);
1530                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1531                 BUG_ON(!device);
1532
1533                 if (device->fs_devices->seeding) {
1534                         btrfs_set_device_generation(leaf, dev_item,
1535                                                     device->generation);
1536                         btrfs_mark_buffer_dirty(leaf);
1537                 }
1538
1539                 path->slots[0]++;
1540                 goto next_slot;
1541         }
1542         ret = 0;
1543 error:
1544         btrfs_free_path(path);
1545         return ret;
1546 }
1547
1548 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1549 {
1550         struct btrfs_trans_handle *trans;
1551         struct btrfs_device *device;
1552         struct block_device *bdev;
1553         struct list_head *devices;
1554         struct super_block *sb = root->fs_info->sb;
1555         u64 total_bytes;
1556         int seeding_dev = 0;
1557         int ret = 0;
1558
1559         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1560                 return -EINVAL;
1561
1562         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1563                                   root->fs_info->bdev_holder);
1564         if (IS_ERR(bdev))
1565                 return PTR_ERR(bdev);
1566
1567         if (root->fs_info->fs_devices->seeding) {
1568                 seeding_dev = 1;
1569                 down_write(&sb->s_umount);
1570                 mutex_lock(&uuid_mutex);
1571         }
1572
1573         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1574         mutex_lock(&root->fs_info->volume_mutex);
1575
1576         devices = &root->fs_info->fs_devices->devices;
1577         /*
1578          * we have the volume lock, so we don't need the extra
1579          * device list mutex while reading the list here.
1580          */
1581         list_for_each_entry(device, devices, dev_list) {
1582                 if (device->bdev == bdev) {
1583                         ret = -EEXIST;
1584                         goto error;
1585                 }
1586         }
1587
1588         device = kzalloc(sizeof(*device), GFP_NOFS);
1589         if (!device) {
1590                 /* we can safely leave the fs_devices entry around */
1591                 ret = -ENOMEM;
1592                 goto error;
1593         }
1594
1595         device->name = kstrdup(device_path, GFP_NOFS);
1596         if (!device->name) {
1597                 kfree(device);
1598                 ret = -ENOMEM;
1599                 goto error;
1600         }
1601
1602         ret = find_next_devid(root, &device->devid);
1603         if (ret) {
1604                 kfree(device);
1605                 goto error;
1606         }
1607
1608         trans = btrfs_start_transaction(root, 0);
1609         lock_chunks(root);
1610
1611         device->writeable = 1;
1612         device->work.func = pending_bios_fn;
1613         generate_random_uuid(device->uuid);
1614         spin_lock_init(&device->io_lock);
1615         device->generation = trans->transid;
1616         device->io_width = root->sectorsize;
1617         device->io_align = root->sectorsize;
1618         device->sector_size = root->sectorsize;
1619         device->total_bytes = i_size_read(bdev->bd_inode);
1620         device->disk_total_bytes = device->total_bytes;
1621         device->dev_root = root->fs_info->dev_root;
1622         device->bdev = bdev;
1623         device->in_fs_metadata = 1;
1624         device->mode = 0;
1625         set_blocksize(device->bdev, 4096);
1626
1627         if (seeding_dev) {
1628                 sb->s_flags &= ~MS_RDONLY;
1629                 ret = btrfs_prepare_sprout(trans, root);
1630                 BUG_ON(ret);
1631         }
1632
1633         device->fs_devices = root->fs_info->fs_devices;
1634
1635         /*
1636          * we don't want write_supers to jump in here with our device
1637          * half setup
1638          */
1639         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1640         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1641         list_add(&device->dev_alloc_list,
1642                  &root->fs_info->fs_devices->alloc_list);
1643         root->fs_info->fs_devices->num_devices++;
1644         root->fs_info->fs_devices->open_devices++;
1645         root->fs_info->fs_devices->rw_devices++;
1646         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1647
1648         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1649                 root->fs_info->fs_devices->rotating = 1;
1650
1651         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1652         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1653                                     total_bytes + device->total_bytes);
1654
1655         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1656         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1657                                     total_bytes + 1);
1658         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1659
1660         if (seeding_dev) {
1661                 ret = init_first_rw_device(trans, root, device);
1662                 BUG_ON(ret);
1663                 ret = btrfs_finish_sprout(trans, root);
1664                 BUG_ON(ret);
1665         } else {
1666                 ret = btrfs_add_device(trans, root, device);
1667         }
1668
1669         /*
1670          * we've got more storage, clear any full flags on the space
1671          * infos
1672          */
1673         btrfs_clear_space_info_full(root->fs_info);
1674
1675         unlock_chunks(root);
1676         btrfs_commit_transaction(trans, root);
1677
1678         if (seeding_dev) {
1679                 mutex_unlock(&uuid_mutex);
1680                 up_write(&sb->s_umount);
1681
1682                 ret = btrfs_relocate_sys_chunks(root);
1683                 BUG_ON(ret);
1684         }
1685 out:
1686         mutex_unlock(&root->fs_info->volume_mutex);
1687         return ret;
1688 error:
1689         blkdev_put(bdev, FMODE_EXCL);
1690         if (seeding_dev) {
1691                 mutex_unlock(&uuid_mutex);
1692                 up_write(&sb->s_umount);
1693         }
1694         goto out;
1695 }
1696
1697 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1698                                         struct btrfs_device *device)
1699 {
1700         int ret;
1701         struct btrfs_path *path;
1702         struct btrfs_root *root;
1703         struct btrfs_dev_item *dev_item;
1704         struct extent_buffer *leaf;
1705         struct btrfs_key key;
1706
1707         root = device->dev_root->fs_info->chunk_root;
1708
1709         path = btrfs_alloc_path();
1710         if (!path)
1711                 return -ENOMEM;
1712
1713         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1714         key.type = BTRFS_DEV_ITEM_KEY;
1715         key.offset = device->devid;
1716
1717         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1718         if (ret < 0)
1719                 goto out;
1720
1721         if (ret > 0) {
1722                 ret = -ENOENT;
1723                 goto out;
1724         }
1725
1726         leaf = path->nodes[0];
1727         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1728
1729         btrfs_set_device_id(leaf, dev_item, device->devid);
1730         btrfs_set_device_type(leaf, dev_item, device->type);
1731         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1732         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1733         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1734         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1735         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1736         btrfs_mark_buffer_dirty(leaf);
1737
1738 out:
1739         btrfs_free_path(path);
1740         return ret;
1741 }
1742
1743 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1744                       struct btrfs_device *device, u64 new_size)
1745 {
1746         struct btrfs_super_block *super_copy =
1747                 &device->dev_root->fs_info->super_copy;
1748         u64 old_total = btrfs_super_total_bytes(super_copy);
1749         u64 diff = new_size - device->total_bytes;
1750
1751         if (!device->writeable)
1752                 return -EACCES;
1753         if (new_size <= device->total_bytes)
1754                 return -EINVAL;
1755
1756         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1757         device->fs_devices->total_rw_bytes += diff;
1758
1759         device->total_bytes = new_size;
1760         device->disk_total_bytes = new_size;
1761         btrfs_clear_space_info_full(device->dev_root->fs_info);
1762
1763         return btrfs_update_device(trans, device);
1764 }
1765
1766 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1767                       struct btrfs_device *device, u64 new_size)
1768 {
1769         int ret;
1770         lock_chunks(device->dev_root);
1771         ret = __btrfs_grow_device(trans, device, new_size);
1772         unlock_chunks(device->dev_root);
1773         return ret;
1774 }
1775
1776 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1777                             struct btrfs_root *root,
1778                             u64 chunk_tree, u64 chunk_objectid,
1779                             u64 chunk_offset)
1780 {
1781         int ret;
1782         struct btrfs_path *path;
1783         struct btrfs_key key;
1784
1785         root = root->fs_info->chunk_root;
1786         path = btrfs_alloc_path();
1787         if (!path)
1788                 return -ENOMEM;
1789
1790         key.objectid = chunk_objectid;
1791         key.offset = chunk_offset;
1792         key.type = BTRFS_CHUNK_ITEM_KEY;
1793
1794         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1795         BUG_ON(ret);
1796
1797         ret = btrfs_del_item(trans, root, path);
1798         BUG_ON(ret);
1799
1800         btrfs_free_path(path);
1801         return 0;
1802 }
1803
1804 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1805                         chunk_offset)
1806 {
1807         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1808         struct btrfs_disk_key *disk_key;
1809         struct btrfs_chunk *chunk;
1810         u8 *ptr;
1811         int ret = 0;
1812         u32 num_stripes;
1813         u32 array_size;
1814         u32 len = 0;
1815         u32 cur;
1816         struct btrfs_key key;
1817
1818         array_size = btrfs_super_sys_array_size(super_copy);
1819
1820         ptr = super_copy->sys_chunk_array;
1821         cur = 0;
1822
1823         while (cur < array_size) {
1824                 disk_key = (struct btrfs_disk_key *)ptr;
1825                 btrfs_disk_key_to_cpu(&key, disk_key);
1826
1827                 len = sizeof(*disk_key);
1828
1829                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1830                         chunk = (struct btrfs_chunk *)(ptr + len);
1831                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1832                         len += btrfs_chunk_item_size(num_stripes);
1833                 } else {
1834                         ret = -EIO;
1835                         break;
1836                 }
1837                 if (key.objectid == chunk_objectid &&
1838                     key.offset == chunk_offset) {
1839                         memmove(ptr, ptr + len, array_size - (cur + len));
1840                         array_size -= len;
1841                         btrfs_set_super_sys_array_size(super_copy, array_size);
1842                 } else {
1843                         ptr += len;
1844                         cur += len;
1845                 }
1846         }
1847         return ret;
1848 }
1849
1850 static int btrfs_relocate_chunk(struct btrfs_root *root,
1851                          u64 chunk_tree, u64 chunk_objectid,
1852                          u64 chunk_offset)
1853 {
1854         struct extent_map_tree *em_tree;
1855         struct btrfs_root *extent_root;
1856         struct btrfs_trans_handle *trans;
1857         struct extent_map *em;
1858         struct map_lookup *map;
1859         int ret;
1860         int i;
1861
1862         root = root->fs_info->chunk_root;
1863         extent_root = root->fs_info->extent_root;
1864         em_tree = &root->fs_info->mapping_tree.map_tree;
1865
1866         ret = btrfs_can_relocate(extent_root, chunk_offset);
1867         if (ret)
1868                 return -ENOSPC;
1869
1870         /* step one, relocate all the extents inside this chunk */
1871         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1872         if (ret)
1873                 return ret;
1874
1875         trans = btrfs_start_transaction(root, 0);
1876         BUG_ON(!trans);
1877
1878         lock_chunks(root);
1879
1880         /*
1881          * step two, delete the device extents and the
1882          * chunk tree entries
1883          */
1884         read_lock(&em_tree->lock);
1885         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1886         read_unlock(&em_tree->lock);
1887
1888         BUG_ON(em->start > chunk_offset ||
1889                em->start + em->len < chunk_offset);
1890         map = (struct map_lookup *)em->bdev;
1891
1892         for (i = 0; i < map->num_stripes; i++) {
1893                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1894                                             map->stripes[i].physical);
1895                 BUG_ON(ret);
1896
1897                 if (map->stripes[i].dev) {
1898                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1899                         BUG_ON(ret);
1900                 }
1901         }
1902         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1903                                chunk_offset);
1904
1905         BUG_ON(ret);
1906
1907         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1908                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1909                 BUG_ON(ret);
1910         }
1911
1912         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1913         BUG_ON(ret);
1914
1915         write_lock(&em_tree->lock);
1916         remove_extent_mapping(em_tree, em);
1917         write_unlock(&em_tree->lock);
1918
1919         kfree(map);
1920         em->bdev = NULL;
1921
1922         /* once for the tree */
1923         free_extent_map(em);
1924         /* once for us */
1925         free_extent_map(em);
1926
1927         unlock_chunks(root);
1928         btrfs_end_transaction(trans, root);
1929         return 0;
1930 }
1931
1932 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1933 {
1934         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1935         struct btrfs_path *path;
1936         struct extent_buffer *leaf;
1937         struct btrfs_chunk *chunk;
1938         struct btrfs_key key;
1939         struct btrfs_key found_key;
1940         u64 chunk_tree = chunk_root->root_key.objectid;
1941         u64 chunk_type;
1942         bool retried = false;
1943         int failed = 0;
1944         int ret;
1945
1946         path = btrfs_alloc_path();
1947         if (!path)
1948                 return -ENOMEM;
1949
1950 again:
1951         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1952         key.offset = (u64)-1;
1953         key.type = BTRFS_CHUNK_ITEM_KEY;
1954
1955         while (1) {
1956                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1957                 if (ret < 0)
1958                         goto error;
1959                 BUG_ON(ret == 0);
1960
1961                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1962                                           key.type);
1963                 if (ret < 0)
1964                         goto error;
1965                 if (ret > 0)
1966                         break;
1967
1968                 leaf = path->nodes[0];
1969                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1970
1971                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1972                                        struct btrfs_chunk);
1973                 chunk_type = btrfs_chunk_type(leaf, chunk);
1974                 btrfs_release_path(chunk_root, path);
1975
1976                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1977                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1978                                                    found_key.objectid,
1979                                                    found_key.offset);
1980                         if (ret == -ENOSPC)
1981                                 failed++;
1982                         else if (ret)
1983                                 BUG();
1984                 }
1985
1986                 if (found_key.offset == 0)
1987                         break;
1988                 key.offset = found_key.offset - 1;
1989         }
1990         ret = 0;
1991         if (failed && !retried) {
1992                 failed = 0;
1993                 retried = true;
1994                 goto again;
1995         } else if (failed && retried) {
1996                 WARN_ON(1);
1997                 ret = -ENOSPC;
1998         }
1999 error:
2000         btrfs_free_path(path);
2001         return ret;
2002 }
2003
2004 static u64 div_factor(u64 num, int factor)
2005 {
2006         if (factor == 10)
2007                 return num;
2008         num *= factor;
2009         do_div(num, 10);
2010         return num;
2011 }
2012
2013 int btrfs_balance(struct btrfs_root *dev_root)
2014 {
2015         int ret;
2016         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2017         struct btrfs_device *device;
2018         u64 old_size;
2019         u64 size_to_free;
2020         struct btrfs_path *path;
2021         struct btrfs_key key;
2022         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2023         struct btrfs_trans_handle *trans;
2024         struct btrfs_key found_key;
2025
2026         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2027                 return -EROFS;
2028
2029         if (!capable(CAP_SYS_ADMIN))
2030                 return -EPERM;
2031
2032         mutex_lock(&dev_root->fs_info->volume_mutex);
2033         dev_root = dev_root->fs_info->dev_root;
2034
2035         /* step one make some room on all the devices */
2036         list_for_each_entry(device, devices, dev_list) {
2037                 old_size = device->total_bytes;
2038                 size_to_free = div_factor(old_size, 1);
2039                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2040                 if (!device->writeable ||
2041                     device->total_bytes - device->bytes_used > size_to_free)
2042                         continue;
2043
2044                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2045                 if (ret == -ENOSPC)
2046                         break;
2047                 BUG_ON(ret);
2048
2049                 trans = btrfs_start_transaction(dev_root, 0);
2050                 BUG_ON(!trans);
2051
2052                 ret = btrfs_grow_device(trans, device, old_size);
2053                 BUG_ON(ret);
2054
2055                 btrfs_end_transaction(trans, dev_root);
2056         }
2057
2058         /* step two, relocate all the chunks */
2059         path = btrfs_alloc_path();
2060         BUG_ON(!path);
2061
2062         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2063         key.offset = (u64)-1;
2064         key.type = BTRFS_CHUNK_ITEM_KEY;
2065
2066         while (1) {
2067                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2068                 if (ret < 0)
2069                         goto error;
2070
2071                 /*
2072                  * this shouldn't happen, it means the last relocate
2073                  * failed
2074                  */
2075                 if (ret == 0)
2076                         break;
2077
2078                 ret = btrfs_previous_item(chunk_root, path, 0,
2079                                           BTRFS_CHUNK_ITEM_KEY);
2080                 if (ret)
2081                         break;
2082
2083                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2084                                       path->slots[0]);
2085                 if (found_key.objectid != key.objectid)
2086                         break;
2087
2088                 /* chunk zero is special */
2089                 if (found_key.offset == 0)
2090                         break;
2091
2092                 btrfs_release_path(chunk_root, path);
2093                 ret = btrfs_relocate_chunk(chunk_root,
2094                                            chunk_root->root_key.objectid,
2095                                            found_key.objectid,
2096                                            found_key.offset);
2097                 BUG_ON(ret && ret != -ENOSPC);
2098                 key.offset = found_key.offset - 1;
2099         }
2100         ret = 0;
2101 error:
2102         btrfs_free_path(path);
2103         mutex_unlock(&dev_root->fs_info->volume_mutex);
2104         return ret;
2105 }
2106
2107 /*
2108  * shrinking a device means finding all of the device extents past
2109  * the new size, and then following the back refs to the chunks.
2110  * The chunk relocation code actually frees the device extent
2111  */
2112 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2113 {
2114         struct btrfs_trans_handle *trans;
2115         struct btrfs_root *root = device->dev_root;
2116         struct btrfs_dev_extent *dev_extent = NULL;
2117         struct btrfs_path *path;
2118         u64 length;
2119         u64 chunk_tree;
2120         u64 chunk_objectid;
2121         u64 chunk_offset;
2122         int ret;
2123         int slot;
2124         int failed = 0;
2125         bool retried = false;
2126         struct extent_buffer *l;
2127         struct btrfs_key key;
2128         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2129         u64 old_total = btrfs_super_total_bytes(super_copy);
2130         u64 old_size = device->total_bytes;
2131         u64 diff = device->total_bytes - new_size;
2132
2133         if (new_size >= device->total_bytes)
2134                 return -EINVAL;
2135
2136         path = btrfs_alloc_path();
2137         if (!path)
2138                 return -ENOMEM;
2139
2140         path->reada = 2;
2141
2142         lock_chunks(root);
2143
2144         device->total_bytes = new_size;
2145         if (device->writeable)
2146                 device->fs_devices->total_rw_bytes -= diff;
2147         unlock_chunks(root);
2148
2149 again:
2150         key.objectid = device->devid;
2151         key.offset = (u64)-1;
2152         key.type = BTRFS_DEV_EXTENT_KEY;
2153
2154         while (1) {
2155                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2156                 if (ret < 0)
2157                         goto done;
2158
2159                 ret = btrfs_previous_item(root, path, 0, key.type);
2160                 if (ret < 0)
2161                         goto done;
2162                 if (ret) {
2163                         ret = 0;
2164                         btrfs_release_path(root, path);
2165                         break;
2166                 }
2167
2168                 l = path->nodes[0];
2169                 slot = path->slots[0];
2170                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2171
2172                 if (key.objectid != device->devid) {
2173                         btrfs_release_path(root, path);
2174                         break;
2175                 }
2176
2177                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2178                 length = btrfs_dev_extent_length(l, dev_extent);
2179
2180                 if (key.offset + length <= new_size) {
2181                         btrfs_release_path(root, path);
2182                         break;
2183                 }
2184
2185                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2186                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2187                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2188                 btrfs_release_path(root, path);
2189
2190                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2191                                            chunk_offset);
2192                 if (ret && ret != -ENOSPC)
2193                         goto done;
2194                 if (ret == -ENOSPC)
2195                         failed++;
2196                 key.offset -= 1;
2197         }
2198
2199         if (failed && !retried) {
2200                 failed = 0;
2201                 retried = true;
2202                 goto again;
2203         } else if (failed && retried) {
2204                 ret = -ENOSPC;
2205                 lock_chunks(root);
2206
2207                 device->total_bytes = old_size;
2208                 if (device->writeable)
2209                         device->fs_devices->total_rw_bytes += diff;
2210                 unlock_chunks(root);
2211                 goto done;
2212         }
2213
2214         /* Shrinking succeeded, else we would be at "done". */
2215         trans = btrfs_start_transaction(root, 0);
2216         lock_chunks(root);
2217
2218         device->disk_total_bytes = new_size;
2219         /* Now btrfs_update_device() will change the on-disk size. */
2220         ret = btrfs_update_device(trans, device);
2221         if (ret) {
2222                 unlock_chunks(root);
2223                 btrfs_end_transaction(trans, root);
2224                 goto done;
2225         }
2226         WARN_ON(diff > old_total);
2227         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2228         unlock_chunks(root);
2229         btrfs_end_transaction(trans, root);
2230 done:
2231         btrfs_free_path(path);
2232         return ret;
2233 }
2234
2235 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2236                            struct btrfs_root *root,
2237                            struct btrfs_key *key,
2238                            struct btrfs_chunk *chunk, int item_size)
2239 {
2240         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2241         struct btrfs_disk_key disk_key;
2242         u32 array_size;
2243         u8 *ptr;
2244
2245         array_size = btrfs_super_sys_array_size(super_copy);
2246         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2247                 return -EFBIG;
2248
2249         ptr = super_copy->sys_chunk_array + array_size;
2250         btrfs_cpu_key_to_disk(&disk_key, key);
2251         memcpy(ptr, &disk_key, sizeof(disk_key));
2252         ptr += sizeof(disk_key);
2253         memcpy(ptr, chunk, item_size);
2254         item_size += sizeof(disk_key);
2255         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2256         return 0;
2257 }
2258
2259 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2260                                         int num_stripes, int sub_stripes)
2261 {
2262         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2263                 return calc_size;
2264         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2265                 return calc_size * (num_stripes / sub_stripes);
2266         else
2267                 return calc_size * num_stripes;
2268 }
2269
2270 /* Used to sort the devices by max_avail(descending sort) */
2271 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2272 {
2273         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2274             ((struct btrfs_device_info *)dev_info2)->max_avail)
2275                 return -1;
2276         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2277                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2278                 return 1;
2279         else
2280                 return 0;
2281 }
2282
2283 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2284                                  int *num_stripes, int *min_stripes,
2285                                  int *sub_stripes)
2286 {
2287         *num_stripes = 1;
2288         *min_stripes = 1;
2289         *sub_stripes = 0;
2290
2291         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2292                 *num_stripes = fs_devices->rw_devices;
2293                 *min_stripes = 2;
2294         }
2295         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2296                 *num_stripes = 2;
2297                 *min_stripes = 2;
2298         }
2299         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2300                 if (fs_devices->rw_devices < 2)
2301                         return -ENOSPC;
2302                 *num_stripes = 2;
2303                 *min_stripes = 2;
2304         }
2305         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2306                 *num_stripes = fs_devices->rw_devices;
2307                 if (*num_stripes < 4)
2308                         return -ENOSPC;
2309                 *num_stripes &= ~(u32)1;
2310                 *sub_stripes = 2;
2311                 *min_stripes = 4;
2312         }
2313
2314         return 0;
2315 }
2316
2317 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2318                                     u64 proposed_size, u64 type,
2319                                     int num_stripes, int small_stripe)
2320 {
2321         int min_stripe_size = 1 * 1024 * 1024;
2322         u64 calc_size = proposed_size;
2323         u64 max_chunk_size = calc_size;
2324         int ncopies = 1;
2325
2326         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2327                     BTRFS_BLOCK_GROUP_DUP |
2328                     BTRFS_BLOCK_GROUP_RAID10))
2329                 ncopies = 2;
2330
2331         if (type & BTRFS_BLOCK_GROUP_DATA) {
2332                 max_chunk_size = 10 * calc_size;
2333                 min_stripe_size = 64 * 1024 * 1024;
2334         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2335                 max_chunk_size = 256 * 1024 * 1024;
2336                 min_stripe_size = 32 * 1024 * 1024;
2337         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2338                 calc_size = 8 * 1024 * 1024;
2339                 max_chunk_size = calc_size * 2;
2340                 min_stripe_size = 1 * 1024 * 1024;
2341         }
2342
2343         /* we don't want a chunk larger than 10% of writeable space */
2344         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2345                              max_chunk_size);
2346
2347         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2348                 calc_size = max_chunk_size * ncopies;
2349                 do_div(calc_size, num_stripes);
2350                 do_div(calc_size, BTRFS_STRIPE_LEN);
2351                 calc_size *= BTRFS_STRIPE_LEN;
2352         }
2353
2354         /* we don't want tiny stripes */
2355         if (!small_stripe)
2356                 calc_size = max_t(u64, min_stripe_size, calc_size);
2357
2358         /*
2359          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2360          * we end up with something bigger than a stripe
2361          */
2362         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2363
2364         do_div(calc_size, BTRFS_STRIPE_LEN);
2365         calc_size *= BTRFS_STRIPE_LEN;
2366
2367         return calc_size;
2368 }
2369
2370 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2371                                                       int num_stripes)
2372 {
2373         struct map_lookup *new;
2374         size_t len = map_lookup_size(num_stripes);
2375
2376         BUG_ON(map->num_stripes < num_stripes);
2377
2378         if (map->num_stripes == num_stripes)
2379                 return map;
2380
2381         new = kmalloc(len, GFP_NOFS);
2382         if (!new) {
2383                 /* just change map->num_stripes */
2384                 map->num_stripes = num_stripes;
2385                 return map;
2386         }
2387
2388         memcpy(new, map, len);
2389         new->num_stripes = num_stripes;
2390         kfree(map);
2391         return new;
2392 }
2393
2394 /*
2395  * helper to allocate device space from btrfs_device_info, in which we stored
2396  * max free space information of every device. It is used when we can not
2397  * allocate chunks by default size.
2398  *
2399  * By this helper, we can allocate a new chunk as larger as possible.
2400  */
2401 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2402                                     struct btrfs_fs_devices *fs_devices,
2403                                     struct btrfs_device_info *devices,
2404                                     int nr_device, u64 type,
2405                                     struct map_lookup **map_lookup,
2406                                     int min_stripes, u64 *stripe_size)
2407 {
2408         int i, index, sort_again = 0;
2409         int min_devices = min_stripes;
2410         u64 max_avail, min_free;
2411         struct map_lookup *map = *map_lookup;
2412         int ret;
2413
2414         if (nr_device < min_stripes)
2415                 return -ENOSPC;
2416
2417         btrfs_descending_sort_devices(devices, nr_device);
2418
2419         max_avail = devices[0].max_avail;
2420         if (!max_avail)
2421                 return -ENOSPC;
2422
2423         for (i = 0; i < nr_device; i++) {
2424                 /*
2425                  * if dev_offset = 0, it means the free space of this device
2426                  * is less than what we need, and we didn't search max avail
2427                  * extent on this device, so do it now.
2428                  */
2429                 if (!devices[i].dev_offset) {
2430                         ret = find_free_dev_extent(trans, devices[i].dev,
2431                                                    max_avail,
2432                                                    &devices[i].dev_offset,
2433                                                    &devices[i].max_avail);
2434                         if (ret != 0 && ret != -ENOSPC)
2435                                 return ret;
2436                         sort_again = 1;
2437                 }
2438         }
2439
2440         /* we update the max avail free extent of each devices, sort again */
2441         if (sort_again)
2442                 btrfs_descending_sort_devices(devices, nr_device);
2443
2444         if (type & BTRFS_BLOCK_GROUP_DUP)
2445                 min_devices = 1;
2446
2447         if (!devices[min_devices - 1].max_avail)
2448                 return -ENOSPC;
2449
2450         max_avail = devices[min_devices - 1].max_avail;
2451         if (type & BTRFS_BLOCK_GROUP_DUP)
2452                 do_div(max_avail, 2);
2453
2454         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2455                                              min_stripes, 1);
2456         if (type & BTRFS_BLOCK_GROUP_DUP)
2457                 min_free = max_avail * 2;
2458         else
2459                 min_free = max_avail;
2460
2461         if (min_free > devices[min_devices - 1].max_avail)
2462                 return -ENOSPC;
2463
2464         map = __shrink_map_lookup_stripes(map, min_stripes);
2465         *stripe_size = max_avail;
2466
2467         index = 0;
2468         for (i = 0; i < min_stripes; i++) {
2469                 map->stripes[i].dev = devices[index].dev;
2470                 map->stripes[i].physical = devices[index].dev_offset;
2471                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2472                         i++;
2473                         map->stripes[i].dev = devices[index].dev;
2474                         map->stripes[i].physical = devices[index].dev_offset +
2475                                                    max_avail;
2476                 }
2477                 index++;
2478         }
2479         *map_lookup = map;
2480
2481         return 0;
2482 }
2483
2484 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2485                                struct btrfs_root *extent_root,
2486                                struct map_lookup **map_ret,
2487                                u64 *num_bytes, u64 *stripe_size,
2488                                u64 start, u64 type)
2489 {
2490         struct btrfs_fs_info *info = extent_root->fs_info;
2491         struct btrfs_device *device = NULL;
2492         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2493         struct list_head *cur;
2494         struct map_lookup *map;
2495         struct extent_map_tree *em_tree;
2496         struct extent_map *em;
2497         struct btrfs_device_info *devices_info;
2498         struct list_head private_devs;
2499         u64 calc_size = 1024 * 1024 * 1024;
2500         u64 min_free;
2501         u64 avail;
2502         u64 dev_offset;
2503         int num_stripes;
2504         int min_stripes;
2505         int sub_stripes;
2506         int min_devices;        /* the min number of devices we need */
2507         int i;
2508         int ret;
2509         int index;
2510
2511         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2512             (type & BTRFS_BLOCK_GROUP_DUP)) {
2513                 WARN_ON(1);
2514                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2515         }
2516         if (list_empty(&fs_devices->alloc_list))
2517                 return -ENOSPC;
2518
2519         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2520                                     &min_stripes, &sub_stripes);
2521         if (ret)
2522                 return ret;
2523
2524         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2525                                GFP_NOFS);
2526         if (!devices_info)
2527                 return -ENOMEM;
2528
2529         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2530         if (!map) {
2531                 ret = -ENOMEM;
2532                 goto error;
2533         }
2534         map->num_stripes = num_stripes;
2535
2536         cur = fs_devices->alloc_list.next;
2537         index = 0;
2538         i = 0;
2539
2540         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2541                                              num_stripes, 0);
2542
2543         if (type & BTRFS_BLOCK_GROUP_DUP) {
2544                 min_free = calc_size * 2;
2545                 min_devices = 1;
2546         } else {
2547                 min_free = calc_size;
2548                 min_devices = min_stripes;
2549         }
2550
2551         INIT_LIST_HEAD(&private_devs);
2552         while (index < num_stripes) {
2553                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2554                 BUG_ON(!device->writeable);
2555                 if (device->total_bytes > device->bytes_used)
2556                         avail = device->total_bytes - device->bytes_used;
2557                 else
2558                         avail = 0;
2559                 cur = cur->next;
2560
2561                 if (device->in_fs_metadata && avail >= min_free) {
2562                         ret = find_free_dev_extent(trans, device, min_free,
2563                                                    &devices_info[i].dev_offset,
2564                                                    &devices_info[i].max_avail);
2565                         if (ret == 0) {
2566                                 list_move_tail(&device->dev_alloc_list,
2567                                                &private_devs);
2568                                 map->stripes[index].dev = device;
2569                                 map->stripes[index].physical =
2570                                                 devices_info[i].dev_offset;
2571                                 index++;
2572                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2573                                         map->stripes[index].dev = device;
2574                                         map->stripes[index].physical =
2575                                                 devices_info[i].dev_offset +
2576                                                 calc_size;
2577                                         index++;
2578                                 }
2579                         } else if (ret != -ENOSPC)
2580                                 goto error;
2581
2582                         devices_info[i].dev = device;
2583                         i++;
2584                 } else if (device->in_fs_metadata &&
2585                            avail >= BTRFS_STRIPE_LEN) {
2586                         devices_info[i].dev = device;
2587                         devices_info[i].max_avail = avail;
2588                         i++;
2589                 }
2590
2591                 if (cur == &fs_devices->alloc_list)
2592                         break;
2593         }
2594
2595         list_splice(&private_devs, &fs_devices->alloc_list);
2596         if (index < num_stripes) {
2597                 if (index >= min_stripes) {
2598                         num_stripes = index;
2599                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2600                                 num_stripes /= sub_stripes;
2601                                 num_stripes *= sub_stripes;
2602                         }
2603
2604                         map = __shrink_map_lookup_stripes(map, num_stripes);
2605                 } else if (i >= min_devices) {
2606                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2607                                                        devices_info, i, type,
2608                                                        &map, min_stripes,
2609                                                        &calc_size);
2610                         if (ret)
2611                                 goto error;
2612                 } else {
2613                         ret = -ENOSPC;
2614                         goto error;
2615                 }
2616         }
2617         map->sector_size = extent_root->sectorsize;
2618         map->stripe_len = BTRFS_STRIPE_LEN;
2619         map->io_align = BTRFS_STRIPE_LEN;
2620         map->io_width = BTRFS_STRIPE_LEN;
2621         map->type = type;
2622         map->sub_stripes = sub_stripes;
2623
2624         *map_ret = map;
2625         *stripe_size = calc_size;
2626         *num_bytes = chunk_bytes_by_type(type, calc_size,
2627                                          map->num_stripes, sub_stripes);
2628
2629         em = alloc_extent_map(GFP_NOFS);
2630         if (!em) {
2631                 ret = -ENOMEM;
2632                 goto error;
2633         }
2634         em->bdev = (struct block_device *)map;
2635         em->start = start;
2636         em->len = *num_bytes;
2637         em->block_start = 0;
2638         em->block_len = em->len;
2639
2640         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2641         write_lock(&em_tree->lock);
2642         ret = add_extent_mapping(em_tree, em);
2643         write_unlock(&em_tree->lock);
2644         BUG_ON(ret);
2645         free_extent_map(em);
2646
2647         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2648                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2649                                      start, *num_bytes);
2650         BUG_ON(ret);
2651
2652         index = 0;
2653         while (index < map->num_stripes) {
2654                 device = map->stripes[index].dev;
2655                 dev_offset = map->stripes[index].physical;
2656
2657                 ret = btrfs_alloc_dev_extent(trans, device,
2658                                 info->chunk_root->root_key.objectid,
2659                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2660                                 start, dev_offset, calc_size);
2661                 BUG_ON(ret);
2662                 index++;
2663         }
2664
2665         kfree(devices_info);
2666         return 0;
2667
2668 error:
2669         kfree(map);
2670         kfree(devices_info);
2671         return ret;
2672 }
2673
2674 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2675                                 struct btrfs_root *extent_root,
2676                                 struct map_lookup *map, u64 chunk_offset,
2677                                 u64 chunk_size, u64 stripe_size)
2678 {
2679         u64 dev_offset;
2680         struct btrfs_key key;
2681         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2682         struct btrfs_device *device;
2683         struct btrfs_chunk *chunk;
2684         struct btrfs_stripe *stripe;
2685         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2686         int index = 0;
2687         int ret;
2688
2689         chunk = kzalloc(item_size, GFP_NOFS);
2690         if (!chunk)
2691                 return -ENOMEM;
2692
2693         index = 0;
2694         while (index < map->num_stripes) {
2695                 device = map->stripes[index].dev;
2696                 device->bytes_used += stripe_size;
2697                 ret = btrfs_update_device(trans, device);
2698                 BUG_ON(ret);
2699                 index++;
2700         }
2701
2702         index = 0;
2703         stripe = &chunk->stripe;
2704         while (index < map->num_stripes) {
2705                 device = map->stripes[index].dev;
2706                 dev_offset = map->stripes[index].physical;
2707
2708                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2709                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2710                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2711                 stripe++;
2712                 index++;
2713         }
2714
2715         btrfs_set_stack_chunk_length(chunk, chunk_size);
2716         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2717         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2718         btrfs_set_stack_chunk_type(chunk, map->type);
2719         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2720         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2721         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2722         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2723         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2724
2725         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2726         key.type = BTRFS_CHUNK_ITEM_KEY;
2727         key.offset = chunk_offset;
2728
2729         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2730         BUG_ON(ret);
2731
2732         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2733                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2734                                              item_size);
2735                 BUG_ON(ret);
2736         }
2737         kfree(chunk);
2738         return 0;
2739 }
2740
2741 /*
2742  * Chunk allocation falls into two parts. The first part does works
2743  * that make the new allocated chunk useable, but not do any operation
2744  * that modifies the chunk tree. The second part does the works that
2745  * require modifying the chunk tree. This division is important for the
2746  * bootstrap process of adding storage to a seed btrfs.
2747  */
2748 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2749                       struct btrfs_root *extent_root, u64 type)
2750 {
2751         u64 chunk_offset;
2752         u64 chunk_size;
2753         u64 stripe_size;
2754         struct map_lookup *map;
2755         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2756         int ret;
2757
2758         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2759                               &chunk_offset);
2760         if (ret)
2761                 return ret;
2762
2763         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2764                                   &stripe_size, chunk_offset, type);
2765         if (ret)
2766                 return ret;
2767
2768         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2769                                    chunk_size, stripe_size);
2770         BUG_ON(ret);
2771         return 0;
2772 }
2773
2774 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2775                                          struct btrfs_root *root,
2776                                          struct btrfs_device *device)
2777 {
2778         u64 chunk_offset;
2779         u64 sys_chunk_offset;
2780         u64 chunk_size;
2781         u64 sys_chunk_size;
2782         u64 stripe_size;
2783         u64 sys_stripe_size;
2784         u64 alloc_profile;
2785         struct map_lookup *map;
2786         struct map_lookup *sys_map;
2787         struct btrfs_fs_info *fs_info = root->fs_info;
2788         struct btrfs_root *extent_root = fs_info->extent_root;
2789         int ret;
2790
2791         ret = find_next_chunk(fs_info->chunk_root,
2792                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2793         BUG_ON(ret);
2794
2795         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2796                         (fs_info->metadata_alloc_profile &
2797                          fs_info->avail_metadata_alloc_bits);
2798         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2799
2800         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2801                                   &stripe_size, chunk_offset, alloc_profile);
2802         BUG_ON(ret);
2803
2804         sys_chunk_offset = chunk_offset + chunk_size;
2805
2806         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2807                         (fs_info->system_alloc_profile &
2808                          fs_info->avail_system_alloc_bits);
2809         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2810
2811         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2812                                   &sys_chunk_size, &sys_stripe_size,
2813                                   sys_chunk_offset, alloc_profile);
2814         BUG_ON(ret);
2815
2816         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2817         BUG_ON(ret);
2818
2819         /*
2820          * Modifying chunk tree needs allocating new blocks from both
2821          * system block group and metadata block group. So we only can
2822          * do operations require modifying the chunk tree after both
2823          * block groups were created.
2824          */
2825         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2826                                    chunk_size, stripe_size);
2827         BUG_ON(ret);
2828
2829         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2830                                    sys_chunk_offset, sys_chunk_size,
2831                                    sys_stripe_size);
2832         BUG_ON(ret);
2833         return 0;
2834 }
2835
2836 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2837 {
2838         struct extent_map *em;
2839         struct map_lookup *map;
2840         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2841         int readonly = 0;
2842         int i;
2843
2844         read_lock(&map_tree->map_tree.lock);
2845         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2846         read_unlock(&map_tree->map_tree.lock);
2847         if (!em)
2848                 return 1;
2849
2850         if (btrfs_test_opt(root, DEGRADED)) {
2851                 free_extent_map(em);
2852                 return 0;
2853         }
2854
2855         map = (struct map_lookup *)em->bdev;
2856         for (i = 0; i < map->num_stripes; i++) {
2857                 if (!map->stripes[i].dev->writeable) {
2858                         readonly = 1;
2859                         break;
2860                 }
2861         }
2862         free_extent_map(em);
2863         return readonly;
2864 }
2865
2866 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2867 {
2868         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2869 }
2870
2871 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2872 {
2873         struct extent_map *em;
2874
2875         while (1) {
2876                 write_lock(&tree->map_tree.lock);
2877                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2878                 if (em)
2879                         remove_extent_mapping(&tree->map_tree, em);
2880                 write_unlock(&tree->map_tree.lock);
2881                 if (!em)
2882                         break;
2883                 kfree(em->bdev);
2884                 /* once for us */
2885                 free_extent_map(em);
2886                 /* once for the tree */
2887                 free_extent_map(em);
2888         }
2889 }
2890
2891 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2892 {
2893         struct extent_map *em;
2894         struct map_lookup *map;
2895         struct extent_map_tree *em_tree = &map_tree->map_tree;
2896         int ret;
2897
2898         read_lock(&em_tree->lock);
2899         em = lookup_extent_mapping(em_tree, logical, len);
2900         read_unlock(&em_tree->lock);
2901         BUG_ON(!em);
2902
2903         BUG_ON(em->start > logical || em->start + em->len < logical);
2904         map = (struct map_lookup *)em->bdev;
2905         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2906                 ret = map->num_stripes;
2907         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2908                 ret = map->sub_stripes;
2909         else
2910                 ret = 1;
2911         free_extent_map(em);
2912         return ret;
2913 }
2914
2915 static int find_live_mirror(struct map_lookup *map, int first, int num,
2916                             int optimal)
2917 {
2918         int i;
2919         if (map->stripes[optimal].dev->bdev)
2920                 return optimal;
2921         for (i = first; i < first + num; i++) {
2922                 if (map->stripes[i].dev->bdev)
2923                         return i;
2924         }
2925         /* we couldn't find one that doesn't fail.  Just return something
2926          * and the io error handling code will clean up eventually
2927          */
2928         return optimal;
2929 }
2930
2931 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2932                              u64 logical, u64 *length,
2933                              struct btrfs_multi_bio **multi_ret,
2934                              int mirror_num, struct page *unplug_page)
2935 {
2936         struct extent_map *em;
2937         struct map_lookup *map;
2938         struct extent_map_tree *em_tree = &map_tree->map_tree;
2939         u64 offset;
2940         u64 stripe_offset;
2941         u64 stripe_nr;
2942         int stripes_allocated = 8;
2943         int stripes_required = 1;
2944         int stripe_index;
2945         int i;
2946         int num_stripes;
2947         int max_errors = 0;
2948         struct btrfs_multi_bio *multi = NULL;
2949
2950         if (multi_ret && !(rw & REQ_WRITE))
2951                 stripes_allocated = 1;
2952 again:
2953         if (multi_ret) {
2954                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2955                                 GFP_NOFS);
2956                 if (!multi)
2957                         return -ENOMEM;
2958
2959                 atomic_set(&multi->error, 0);
2960         }
2961
2962         read_lock(&em_tree->lock);
2963         em = lookup_extent_mapping(em_tree, logical, *length);
2964         read_unlock(&em_tree->lock);
2965
2966         if (!em && unplug_page) {
2967                 kfree(multi);
2968                 return 0;
2969         }
2970
2971         if (!em) {
2972                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2973                        (unsigned long long)logical,
2974                        (unsigned long long)*length);
2975                 BUG();
2976         }
2977
2978         BUG_ON(em->start > logical || em->start + em->len < logical);
2979         map = (struct map_lookup *)em->bdev;
2980         offset = logical - em->start;
2981
2982         if (mirror_num > map->num_stripes)
2983                 mirror_num = 0;
2984
2985         /* if our multi bio struct is too small, back off and try again */
2986         if (rw & REQ_WRITE) {
2987                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2988                                  BTRFS_BLOCK_GROUP_DUP)) {
2989                         stripes_required = map->num_stripes;
2990                         max_errors = 1;
2991                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2992                         stripes_required = map->sub_stripes;
2993                         max_errors = 1;
2994                 }
2995         }
2996         if (multi_ret && (rw & REQ_WRITE) &&
2997             stripes_allocated < stripes_required) {
2998                 stripes_allocated = map->num_stripes;
2999                 free_extent_map(em);
3000                 kfree(multi);
3001                 goto again;
3002         }
3003         stripe_nr = offset;
3004         /*
3005          * stripe_nr counts the total number of stripes we have to stride
3006          * to get to this block
3007          */
3008         do_div(stripe_nr, map->stripe_len);
3009
3010         stripe_offset = stripe_nr * map->stripe_len;
3011         BUG_ON(offset < stripe_offset);
3012
3013         /* stripe_offset is the offset of this block in its stripe*/
3014         stripe_offset = offset - stripe_offset;
3015
3016         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3017                          BTRFS_BLOCK_GROUP_RAID10 |
3018                          BTRFS_BLOCK_GROUP_DUP)) {
3019                 /* we limit the length of each bio to what fits in a stripe */
3020                 *length = min_t(u64, em->len - offset,
3021                               map->stripe_len - stripe_offset);
3022         } else {
3023                 *length = em->len - offset;
3024         }
3025
3026         if (!multi_ret && !unplug_page)
3027                 goto out;
3028
3029         num_stripes = 1;
3030         stripe_index = 0;
3031         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3032                 if (unplug_page || (rw & REQ_WRITE))
3033                         num_stripes = map->num_stripes;
3034                 else if (mirror_num)
3035                         stripe_index = mirror_num - 1;
3036                 else {
3037                         stripe_index = find_live_mirror(map, 0,
3038                                             map->num_stripes,
3039                                             current->pid % map->num_stripes);
3040                 }
3041
3042         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3043                 if (rw & REQ_WRITE)
3044                         num_stripes = map->num_stripes;
3045                 else if (mirror_num)
3046                         stripe_index = mirror_num - 1;
3047
3048         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3049                 int factor = map->num_stripes / map->sub_stripes;
3050
3051                 stripe_index = do_div(stripe_nr, factor);
3052                 stripe_index *= map->sub_stripes;
3053
3054                 if (unplug_page || (rw & REQ_WRITE))
3055                         num_stripes = map->sub_stripes;
3056                 else if (mirror_num)
3057                         stripe_index += mirror_num - 1;
3058                 else {
3059                         stripe_index = find_live_mirror(map, stripe_index,
3060                                               map->sub_stripes, stripe_index +
3061                                               current->pid % map->sub_stripes);
3062                 }
3063         } else {
3064                 /*
3065                  * after this do_div call, stripe_nr is the number of stripes
3066                  * on this device we have to walk to find the data, and
3067                  * stripe_index is the number of our device in the stripe array
3068                  */
3069                 stripe_index = do_div(stripe_nr, map->num_stripes);
3070         }
3071         BUG_ON(stripe_index >= map->num_stripes);
3072
3073         for (i = 0; i < num_stripes; i++) {
3074                 if (unplug_page) {
3075                         struct btrfs_device *device;
3076                         struct backing_dev_info *bdi;
3077
3078                         device = map->stripes[stripe_index].dev;
3079                         if (device->bdev) {
3080                                 bdi = blk_get_backing_dev_info(device->bdev);
3081                                 if (bdi->unplug_io_fn)
3082                                         bdi->unplug_io_fn(bdi, unplug_page);
3083                         }
3084                 } else {
3085                         multi->stripes[i].physical =
3086                                 map->stripes[stripe_index].physical +
3087                                 stripe_offset + stripe_nr * map->stripe_len;
3088                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3089                 }
3090                 stripe_index++;
3091         }
3092         if (multi_ret) {
3093                 *multi_ret = multi;
3094                 multi->num_stripes = num_stripes;
3095                 multi->max_errors = max_errors;
3096         }
3097 out:
3098         free_extent_map(em);
3099         return 0;
3100 }
3101
3102 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3103                       u64 logical, u64 *length,
3104                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3105 {
3106         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3107                                  mirror_num, NULL);
3108 }
3109
3110 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3111                      u64 chunk_start, u64 physical, u64 devid,
3112                      u64 **logical, int *naddrs, int *stripe_len)
3113 {
3114         struct extent_map_tree *em_tree = &map_tree->map_tree;
3115         struct extent_map *em;
3116         struct map_lookup *map;
3117         u64 *buf;
3118         u64 bytenr;
3119         u64 length;
3120         u64 stripe_nr;
3121         int i, j, nr = 0;
3122
3123         read_lock(&em_tree->lock);
3124         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3125         read_unlock(&em_tree->lock);
3126
3127         BUG_ON(!em || em->start != chunk_start);
3128         map = (struct map_lookup *)em->bdev;
3129
3130         length = em->len;
3131         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3132                 do_div(length, map->num_stripes / map->sub_stripes);
3133         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3134                 do_div(length, map->num_stripes);
3135
3136         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3137         BUG_ON(!buf);
3138
3139         for (i = 0; i < map->num_stripes; i++) {
3140                 if (devid && map->stripes[i].dev->devid != devid)
3141                         continue;
3142                 if (map->stripes[i].physical > physical ||
3143                     map->stripes[i].physical + length <= physical)
3144                         continue;
3145
3146                 stripe_nr = physical - map->stripes[i].physical;
3147                 do_div(stripe_nr, map->stripe_len);
3148
3149                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3150                         stripe_nr = stripe_nr * map->num_stripes + i;
3151                         do_div(stripe_nr, map->sub_stripes);
3152                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3153                         stripe_nr = stripe_nr * map->num_stripes + i;
3154                 }
3155                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3156                 WARN_ON(nr >= map->num_stripes);
3157                 for (j = 0; j < nr; j++) {
3158                         if (buf[j] == bytenr)
3159                                 break;
3160                 }
3161                 if (j == nr) {
3162                         WARN_ON(nr >= map->num_stripes);
3163                         buf[nr++] = bytenr;
3164                 }
3165         }
3166
3167         *logical = buf;
3168         *naddrs = nr;
3169         *stripe_len = map->stripe_len;