btrfs: move btrfs_cmp_device_free_bytes to super.c
[linux-2.6.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 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37                                 struct btrfs_root *root,
38                                 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42                             (sizeof(struct btrfs_bio_stripe) * (n)))
43
44 static DEFINE_MUTEX(uuid_mutex);
45 static LIST_HEAD(fs_uuids);
46
47 void btrfs_lock_volumes(void)
48 {
49         mutex_lock(&uuid_mutex);
50 }
51
52 void btrfs_unlock_volumes(void)
53 {
54         mutex_unlock(&uuid_mutex);
55 }
56
57 static void lock_chunks(struct btrfs_root *root)
58 {
59         mutex_lock(&root->fs_info->chunk_mutex);
60 }
61
62 static void unlock_chunks(struct btrfs_root *root)
63 {
64         mutex_unlock(&root->fs_info->chunk_mutex);
65 }
66
67 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
68 {
69         struct btrfs_device *device;
70         WARN_ON(fs_devices->opened);
71         while (!list_empty(&fs_devices->devices)) {
72                 device = list_entry(fs_devices->devices.next,
73                                     struct btrfs_device, dev_list);
74                 list_del(&device->dev_list);
75                 kfree(device->name);
76                 kfree(device);
77         }
78         kfree(fs_devices);
79 }
80
81 int btrfs_cleanup_fs_uuids(void)
82 {
83         struct btrfs_fs_devices *fs_devices;
84
85         while (!list_empty(&fs_uuids)) {
86                 fs_devices = list_entry(fs_uuids.next,
87                                         struct btrfs_fs_devices, list);
88                 list_del(&fs_devices->list);
89                 free_fs_devices(fs_devices);
90         }
91         return 0;
92 }
93
94 static noinline struct btrfs_device *__find_device(struct list_head *head,
95                                                    u64 devid, u8 *uuid)
96 {
97         struct btrfs_device *dev;
98
99         list_for_each_entry(dev, head, dev_list) {
100                 if (dev->devid == devid &&
101                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
102                         return dev;
103                 }
104         }
105         return NULL;
106 }
107
108 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
109 {
110         struct btrfs_fs_devices *fs_devices;
111
112         list_for_each_entry(fs_devices, &fs_uuids, list) {
113                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
114                         return fs_devices;
115         }
116         return NULL;
117 }
118
119 static void requeue_list(struct btrfs_pending_bios *pending_bios,
120                         struct bio *head, struct bio *tail)
121 {
122
123         struct bio *old_head;
124
125         old_head = pending_bios->head;
126         pending_bios->head = head;
127         if (pending_bios->tail)
128                 tail->bi_next = old_head;
129         else
130                 pending_bios->tail = tail;
131 }
132
133 /*
134  * we try to collect pending bios for a device so we don't get a large
135  * number of procs sending bios down to the same device.  This greatly
136  * improves the schedulers ability to collect and merge the bios.
137  *
138  * But, it also turns into a long list of bios to process and that is sure
139  * to eventually make the worker thread block.  The solution here is to
140  * make some progress and then put this work struct back at the end of
141  * the list if the block device is congested.  This way, multiple devices
142  * can make progress from a single worker thread.
143  */
144 static noinline int run_scheduled_bios(struct btrfs_device *device)
145 {
146         struct bio *pending;
147         struct backing_dev_info *bdi;
148         struct btrfs_fs_info *fs_info;
149         struct btrfs_pending_bios *pending_bios;
150         struct bio *tail;
151         struct bio *cur;
152         int again = 0;
153         unsigned long num_run;
154         unsigned long num_sync_run;
155         unsigned long batch_run = 0;
156         unsigned long limit;
157         unsigned long last_waited = 0;
158         int force_reg = 0;
159
160         bdi = blk_get_backing_dev_info(device->bdev);
161         fs_info = device->dev_root->fs_info;
162         limit = btrfs_async_submit_limit(fs_info);
163         limit = limit * 2 / 3;
164
165         /* we want to make sure that every time we switch from the sync
166          * list to the normal list, we unplug
167          */
168         num_sync_run = 0;
169
170 loop:
171         spin_lock(&device->io_lock);
172
173 loop_lock:
174         num_run = 0;
175
176         /* take all the bios off the list at once and process them
177          * later on (without the lock held).  But, remember the
178          * tail and other pointers so the bios can be properly reinserted
179          * into the list if we hit congestion
180          */
181         if (!force_reg && device->pending_sync_bios.head) {
182                 pending_bios = &device->pending_sync_bios;
183                 force_reg = 1;
184         } else {
185                 pending_bios = &device->pending_bios;
186                 force_reg = 0;
187         }
188
189         pending = pending_bios->head;
190         tail = pending_bios->tail;
191         WARN_ON(pending && !tail);
192
193         /*
194          * if pending was null this time around, no bios need processing
195          * at all and we can stop.  Otherwise it'll loop back up again
196          * and do an additional check so no bios are missed.
197          *
198          * device->running_pending is used to synchronize with the
199          * schedule_bio code.
200          */
201         if (device->pending_sync_bios.head == NULL &&
202             device->pending_bios.head == NULL) {
203                 again = 0;
204                 device->running_pending = 0;
205         } else {
206                 again = 1;
207                 device->running_pending = 1;
208         }
209
210         pending_bios->head = NULL;
211         pending_bios->tail = NULL;
212
213         spin_unlock(&device->io_lock);
214
215         /*
216          * if we're doing the regular priority list, make sure we unplug
217          * for any high prio bios we've sent down
218          */
219         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
220                 num_sync_run = 0;
221                 blk_run_backing_dev(bdi, NULL);
222         }
223
224         while (pending) {
225
226                 rmb();
227                 /* we want to work on both lists, but do more bios on the
228                  * sync list than the regular list
229                  */
230                 if ((num_run > 32 &&
231                     pending_bios != &device->pending_sync_bios &&
232                     device->pending_sync_bios.head) ||
233                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
234                     device->pending_bios.head)) {
235                         spin_lock(&device->io_lock);
236                         requeue_list(pending_bios, pending, tail);
237                         goto loop_lock;
238                 }
239
240                 cur = pending;
241                 pending = pending->bi_next;
242                 cur->bi_next = NULL;
243                 atomic_dec(&fs_info->nr_async_bios);
244
245                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
246                     waitqueue_active(&fs_info->async_submit_wait))
247                         wake_up(&fs_info->async_submit_wait);
248
249                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
250
251                 if (cur->bi_rw & REQ_SYNC)
252                         num_sync_run++;
253
254                 submit_bio(cur->bi_rw, cur);
255                 num_run++;
256                 batch_run++;
257                 if (need_resched()) {
258                         if (num_sync_run) {
259                                 blk_run_backing_dev(bdi, NULL);
260                                 num_sync_run = 0;
261                         }
262                         cond_resched();
263                 }
264
265                 /*
266                  * we made progress, there is more work to do and the bdi
267                  * is now congested.  Back off and let other work structs
268                  * run instead
269                  */
270                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
271                     fs_info->fs_devices->open_devices > 1) {
272                         struct io_context *ioc;
273
274                         ioc = current->io_context;
275
276                         /*
277                          * the main goal here is that we don't want to
278                          * block if we're going to be able to submit
279                          * more requests without blocking.
280                          *
281                          * This code does two great things, it pokes into
282                          * the elevator code from a filesystem _and_
283                          * it makes assumptions about how batching works.
284                          */
285                         if (ioc && ioc->nr_batch_requests > 0 &&
286                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
287                             (last_waited == 0 ||
288                              ioc->last_waited == last_waited)) {
289                                 /*
290                                  * we want to go through our batch of
291                                  * requests and stop.  So, we copy out
292                                  * the ioc->last_waited time and test
293                                  * against it before looping
294                                  */
295                                 last_waited = ioc->last_waited;
296                                 if (need_resched()) {
297                                         if (num_sync_run) {
298                                                 blk_run_backing_dev(bdi, NULL);
299                                                 num_sync_run = 0;
300                                         }
301                                         cond_resched();
302                                 }
303                                 continue;
304                         }
305                         spin_lock(&device->io_lock);
306                         requeue_list(pending_bios, pending, tail);
307                         device->running_pending = 1;
308
309                         spin_unlock(&device->io_lock);
310                         btrfs_requeue_work(&device->work);
311                         goto done;
312                 }
313         }
314
315         if (num_sync_run) {
316                 num_sync_run = 0;
317                 blk_run_backing_dev(bdi, NULL);
318         }
319         /*
320          * IO has already been through a long path to get here.  Checksumming,
321          * async helper threads, perhaps compression.  We've done a pretty
322          * good job of collecting a batch of IO and should just unplug
323          * the device right away.
324          *
325          * This will help anyone who is waiting on the IO, they might have
326          * already unplugged, but managed to do so before the bio they
327          * cared about found its way down here.
328          */
329         blk_run_backing_dev(bdi, NULL);
330
331         cond_resched();
332         if (again)
333                 goto loop;
334
335         spin_lock(&device->io_lock);
336         if (device->pending_bios.head || device->pending_sync_bios.head)
337                 goto loop_lock;
338         spin_unlock(&device->io_lock);
339
340 done:
341         return 0;
342 }
343
344 static void pending_bios_fn(struct btrfs_work *work)
345 {
346         struct btrfs_device *device;
347
348         device = container_of(work, struct btrfs_device, work);
349         run_scheduled_bios(device);
350 }
351
352 static noinline int device_list_add(const char *path,
353                            struct btrfs_super_block *disk_super,
354                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
355 {
356         struct btrfs_device *device;
357         struct btrfs_fs_devices *fs_devices;
358         u64 found_transid = btrfs_super_generation(disk_super);
359         char *name;
360
361         fs_devices = find_fsid(disk_super->fsid);
362         if (!fs_devices) {
363                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
364                 if (!fs_devices)
365                         return -ENOMEM;
366                 INIT_LIST_HEAD(&fs_devices->devices);
367                 INIT_LIST_HEAD(&fs_devices->alloc_list);
368                 list_add(&fs_devices->list, &fs_uuids);
369                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
370                 fs_devices->latest_devid = devid;
371                 fs_devices->latest_trans = found_transid;
372                 mutex_init(&fs_devices->device_list_mutex);
373                 device = NULL;
374         } else {
375                 device = __find_device(&fs_devices->devices, devid,
376                                        disk_super->dev_item.uuid);
377         }
378         if (!device) {
379                 if (fs_devices->opened)
380                         return -EBUSY;
381
382                 device = kzalloc(sizeof(*device), GFP_NOFS);
383                 if (!device) {
384                         /* we can safely leave the fs_devices entry around */
385                         return -ENOMEM;
386                 }
387                 device->devid = devid;
388                 device->work.func = pending_bios_fn;
389                 memcpy(device->uuid, disk_super->dev_item.uuid,
390                        BTRFS_UUID_SIZE);
391                 spin_lock_init(&device->io_lock);
392                 device->name = kstrdup(path, GFP_NOFS);
393                 if (!device->name) {
394                         kfree(device);
395                         return -ENOMEM;
396                 }
397                 INIT_LIST_HEAD(&device->dev_alloc_list);
398
399                 mutex_lock(&fs_devices->device_list_mutex);
400                 list_add(&device->dev_list, &fs_devices->devices);
401                 mutex_unlock(&fs_devices->device_list_mutex);
402
403                 device->fs_devices = fs_devices;
404                 fs_devices->num_devices++;
405         } else if (!device->name || strcmp(device->name, path)) {
406                 name = kstrdup(path, GFP_NOFS);
407                 if (!name)
408                         return -ENOMEM;
409                 kfree(device->name);
410                 device->name = name;
411                 if (device->missing) {
412                         fs_devices->missing_devices--;
413                         device->missing = 0;
414                 }
415         }
416
417         if (found_transid > fs_devices->latest_trans) {
418                 fs_devices->latest_devid = devid;
419                 fs_devices->latest_trans = found_transid;
420         }
421         *fs_devices_ret = fs_devices;
422         return 0;
423 }
424
425 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
426 {
427         struct btrfs_fs_devices *fs_devices;
428         struct btrfs_device *device;
429         struct btrfs_device *orig_dev;
430
431         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
432         if (!fs_devices)
433                 return ERR_PTR(-ENOMEM);
434
435         INIT_LIST_HEAD(&fs_devices->devices);
436         INIT_LIST_HEAD(&fs_devices->alloc_list);
437         INIT_LIST_HEAD(&fs_devices->list);
438         mutex_init(&fs_devices->device_list_mutex);
439         fs_devices->latest_devid = orig->latest_devid;
440         fs_devices->latest_trans = orig->latest_trans;
441         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
442
443         mutex_lock(&orig->device_list_mutex);
444         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
445                 device = kzalloc(sizeof(*device), GFP_NOFS);
446                 if (!device)
447                         goto error;
448
449                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
450                 if (!device->name) {
451                         kfree(device);
452                         goto error;
453                 }
454
455                 device->devid = orig_dev->devid;
456                 device->work.func = pending_bios_fn;
457                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
458                 spin_lock_init(&device->io_lock);
459                 INIT_LIST_HEAD(&device->dev_list);
460                 INIT_LIST_HEAD(&device->dev_alloc_list);
461
462                 list_add(&device->dev_list, &fs_devices->devices);
463                 device->fs_devices = fs_devices;
464                 fs_devices->num_devices++;
465         }
466         mutex_unlock(&orig->device_list_mutex);
467         return fs_devices;
468 error:
469         mutex_unlock(&orig->device_list_mutex);
470         free_fs_devices(fs_devices);
471         return ERR_PTR(-ENOMEM);
472 }
473
474 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
475 {
476         struct btrfs_device *device, *next;
477
478         mutex_lock(&uuid_mutex);
479 again:
480         mutex_lock(&fs_devices->device_list_mutex);
481         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
482                 if (device->in_fs_metadata)
483                         continue;
484
485                 if (device->bdev) {
486                         blkdev_put(device->bdev, device->mode);
487                         device->bdev = NULL;
488                         fs_devices->open_devices--;
489                 }
490                 if (device->writeable) {
491                         list_del_init(&device->dev_alloc_list);
492                         device->writeable = 0;
493                         fs_devices->rw_devices--;
494                 }
495                 list_del_init(&device->dev_list);
496                 fs_devices->num_devices--;
497                 kfree(device->name);
498                 kfree(device);
499         }
500         mutex_unlock(&fs_devices->device_list_mutex);
501
502         if (fs_devices->seed) {
503                 fs_devices = fs_devices->seed;
504                 goto again;
505         }
506
507         mutex_unlock(&uuid_mutex);
508         return 0;
509 }
510
511 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
512 {
513         struct btrfs_device *device;
514
515         if (--fs_devices->opened > 0)
516                 return 0;
517
518         list_for_each_entry(device, &fs_devices->devices, dev_list) {
519                 if (device->bdev) {
520                         blkdev_put(device->bdev, device->mode);
521                         fs_devices->open_devices--;
522                 }
523                 if (device->writeable) {
524                         list_del_init(&device->dev_alloc_list);
525                         fs_devices->rw_devices--;
526                 }
527
528                 device->bdev = NULL;
529                 device->writeable = 0;
530                 device->in_fs_metadata = 0;
531         }
532         WARN_ON(fs_devices->open_devices);
533         WARN_ON(fs_devices->rw_devices);
534         fs_devices->opened = 0;
535         fs_devices->seeding = 0;
536
537         return 0;
538 }
539
540 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
541 {
542         struct btrfs_fs_devices *seed_devices = NULL;
543         int ret;
544
545         mutex_lock(&uuid_mutex);
546         ret = __btrfs_close_devices(fs_devices);
547         if (!fs_devices->opened) {
548                 seed_devices = fs_devices->seed;
549                 fs_devices->seed = NULL;
550         }
551         mutex_unlock(&uuid_mutex);
552
553         while (seed_devices) {
554                 fs_devices = seed_devices;
555                 seed_devices = fs_devices->seed;
556                 __btrfs_close_devices(fs_devices);
557                 free_fs_devices(fs_devices);
558         }
559         return ret;
560 }
561
562 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563                                 fmode_t flags, void *holder)
564 {
565         struct block_device *bdev;
566         struct list_head *head = &fs_devices->devices;
567         struct btrfs_device *device;
568         struct block_device *latest_bdev = NULL;
569         struct buffer_head *bh;
570         struct btrfs_super_block *disk_super;
571         u64 latest_devid = 0;
572         u64 latest_transid = 0;
573         u64 devid;
574         int seeding = 1;
575         int ret = 0;
576
577         flags |= FMODE_EXCL;
578
579         list_for_each_entry(device, head, dev_list) {
580                 if (device->bdev)
581                         continue;
582                 if (!device->name)
583                         continue;
584
585                 bdev = blkdev_get_by_path(device->name, flags, holder);
586                 if (IS_ERR(bdev)) {
587                         printk(KERN_INFO "open %s failed\n", device->name);
588                         goto error;
589                 }
590                 set_blocksize(bdev, 4096);
591
592                 bh = btrfs_read_dev_super(bdev);
593                 if (!bh) {
594                         ret = -EINVAL;
595                         goto error_close;
596                 }
597
598                 disk_super = (struct btrfs_super_block *)bh->b_data;
599                 devid = btrfs_stack_device_id(&disk_super->dev_item);
600                 if (devid != device->devid)
601                         goto error_brelse;
602
603                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
604                            BTRFS_UUID_SIZE))
605                         goto error_brelse;
606
607                 device->generation = btrfs_super_generation(disk_super);
608                 if (!latest_transid || device->generation > latest_transid) {
609                         latest_devid = devid;
610                         latest_transid = device->generation;
611                         latest_bdev = bdev;
612                 }
613
614                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
615                         device->writeable = 0;
616                 } else {
617                         device->writeable = !bdev_read_only(bdev);
618                         seeding = 0;
619                 }
620
621                 device->bdev = bdev;
622                 device->in_fs_metadata = 0;
623                 device->mode = flags;
624
625                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
626                         fs_devices->rotating = 1;
627
628                 fs_devices->open_devices++;
629                 if (device->writeable) {
630                         fs_devices->rw_devices++;
631                         list_add(&device->dev_alloc_list,
632                                  &fs_devices->alloc_list);
633                 }
634                 continue;
635
636 error_brelse:
637                 brelse(bh);
638 error_close:
639                 blkdev_put(bdev, flags);
640 error:
641                 continue;
642         }
643         if (fs_devices->open_devices == 0) {
644                 ret = -EIO;
645                 goto out;
646         }
647         fs_devices->seeding = seeding;
648         fs_devices->opened = 1;
649         fs_devices->latest_bdev = latest_bdev;
650         fs_devices->latest_devid = latest_devid;
651         fs_devices->latest_trans = latest_transid;
652         fs_devices->total_rw_bytes = 0;
653 out:
654         return ret;
655 }
656
657 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
658                        fmode_t flags, void *holder)
659 {
660         int ret;
661
662         mutex_lock(&uuid_mutex);
663         if (fs_devices->opened) {
664                 fs_devices->opened++;
665                 ret = 0;
666         } else {
667                 ret = __btrfs_open_devices(fs_devices, flags, holder);
668         }
669         mutex_unlock(&uuid_mutex);
670         return ret;
671 }
672
673 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
674                           struct btrfs_fs_devices **fs_devices_ret)
675 {
676         struct btrfs_super_block *disk_super;
677         struct block_device *bdev;
678         struct buffer_head *bh;
679         int ret;
680         u64 devid;
681         u64 transid;
682
683         mutex_lock(&uuid_mutex);
684
685         flags |= FMODE_EXCL;
686         bdev = blkdev_get_by_path(path, flags, holder);
687
688         if (IS_ERR(bdev)) {
689                 ret = PTR_ERR(bdev);
690                 goto error;
691         }
692
693         ret = set_blocksize(bdev, 4096);
694         if (ret)
695                 goto error_close;
696         bh = btrfs_read_dev_super(bdev);
697         if (!bh) {
698                 ret = -EINVAL;
699                 goto error_close;
700         }
701         disk_super = (struct btrfs_super_block *)bh->b_data;
702         devid = btrfs_stack_device_id(&disk_super->dev_item);
703         transid = btrfs_super_generation(disk_super);
704         if (disk_super->label[0])
705                 printk(KERN_INFO "device label %s ", disk_super->label);
706         else {
707                 /* FIXME, make a readl uuid parser */
708                 printk(KERN_INFO "device fsid %llx-%llx ",
709                        *(unsigned long long *)disk_super->fsid,
710                        *(unsigned long long *)(disk_super->fsid + 8));
711         }
712         printk(KERN_CONT "devid %llu transid %llu %s\n",
713                (unsigned long long)devid, (unsigned long long)transid, path);
714         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
715
716         brelse(bh);
717 error_close:
718         blkdev_put(bdev, flags);
719 error:
720         mutex_unlock(&uuid_mutex);
721         return ret;
722 }
723
724 /* helper to account the used device space in the range */
725 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
726                                    u64 end, u64 *length)
727 {
728         struct btrfs_key key;
729         struct btrfs_root *root = device->dev_root;
730         struct btrfs_dev_extent *dev_extent;
731         struct btrfs_path *path;
732         u64 extent_end;
733         int ret;
734         int slot;
735         struct extent_buffer *l;
736
737         *length = 0;
738
739         if (start >= device->total_bytes)
740                 return 0;
741
742         path = btrfs_alloc_path();
743         if (!path)
744                 return -ENOMEM;
745         path->reada = 2;
746
747         key.objectid = device->devid;
748         key.offset = start;
749         key.type = BTRFS_DEV_EXTENT_KEY;
750
751         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
752         if (ret < 0)
753                 goto out;
754         if (ret > 0) {
755                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
756                 if (ret < 0)
757                         goto out;
758         }
759
760         while (1) {
761                 l = path->nodes[0];
762                 slot = path->slots[0];
763                 if (slot >= btrfs_header_nritems(l)) {
764                         ret = btrfs_next_leaf(root, path);
765                         if (ret == 0)
766                                 continue;
767                         if (ret < 0)
768                                 goto out;
769
770                         break;
771                 }
772                 btrfs_item_key_to_cpu(l, &key, slot);
773
774                 if (key.objectid < device->devid)
775                         goto next;
776
777                 if (key.objectid > device->devid)
778                         break;
779
780                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
781                         goto next;
782
783                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
784                 extent_end = key.offset + btrfs_dev_extent_length(l,
785                                                                   dev_extent);
786                 if (key.offset <= start && extent_end > end) {
787                         *length = end - start + 1;
788                         break;
789                 } else if (key.offset <= start && extent_end > start)
790                         *length += extent_end - start;
791                 else if (key.offset > start && extent_end <= end)
792                         *length += extent_end - key.offset;
793                 else if (key.offset > start && key.offset <= end) {
794                         *length += end - key.offset + 1;
795                         break;
796                 } else if (key.offset > end)
797                         break;
798
799 next:
800                 path->slots[0]++;
801         }
802         ret = 0;
803 out:
804         btrfs_free_path(path);
805         return ret;
806 }
807
808 /*
809  * find_free_dev_extent - find free space in the specified device
810  * @trans:      transaction handler
811  * @device:     the device which we search the free space in
812  * @num_bytes:  the size of the free space that we need
813  * @start:      store the start of the free space.
814  * @len:        the size of the free space. that we find, or the size of the max
815  *              free space if we don't find suitable free space
816  *
817  * this uses a pretty simple search, the expectation is that it is
818  * called very infrequently and that a given device has a small number
819  * of extents
820  *
821  * @start is used to store the start of the free space if we find. But if we
822  * don't find suitable free space, it will be used to store the start position
823  * of the max free space.
824  *
825  * @len is used to store the size of the free space that we find.
826  * But if we don't find suitable free space, it is used to store the size of
827  * the max free space.
828  */
829 int find_free_dev_extent(struct btrfs_trans_handle *trans,
830                          struct btrfs_device *device, u64 num_bytes,
831                          u64 *start, u64 *len)
832 {
833         struct btrfs_key key;
834         struct btrfs_root *root = device->dev_root;
835         struct btrfs_dev_extent *dev_extent;
836         struct btrfs_path *path;
837         u64 hole_size;
838         u64 max_hole_start;
839         u64 max_hole_size;
840         u64 extent_end;
841         u64 search_start;
842         u64 search_end = device->total_bytes;
843         int ret;
844         int slot;
845         struct extent_buffer *l;
846
847         /* FIXME use last free of some kind */
848
849         /* we don't want to overwrite the superblock on the drive,
850          * so we make sure to start at an offset of at least 1MB
851          */
852         search_start = 1024 * 1024;
853
854         if (root->fs_info->alloc_start + num_bytes <= search_end)
855                 search_start = max(root->fs_info->alloc_start, search_start);
856
857         max_hole_start = search_start;
858         max_hole_size = 0;
859
860         if (search_start >= search_end) {
861                 ret = -ENOSPC;
862                 goto error;
863         }
864
865         path = btrfs_alloc_path();
866         if (!path) {
867                 ret = -ENOMEM;
868                 goto error;
869         }
870         path->reada = 2;
871
872         key.objectid = device->devid;
873         key.offset = search_start;
874         key.type = BTRFS_DEV_EXTENT_KEY;
875
876         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
877         if (ret < 0)
878                 goto out;
879         if (ret > 0) {
880                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
881                 if (ret < 0)
882                         goto out;
883         }
884
885         while (1) {
886                 l = path->nodes[0];
887                 slot = path->slots[0];
888                 if (slot >= btrfs_header_nritems(l)) {
889                         ret = btrfs_next_leaf(root, path);
890                         if (ret == 0)
891                                 continue;
892                         if (ret < 0)
893                                 goto out;
894
895                         break;
896                 }
897                 btrfs_item_key_to_cpu(l, &key, slot);
898
899                 if (key.objectid < device->devid)
900                         goto next;
901
902                 if (key.objectid > device->devid)
903                         break;
904
905                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
906                         goto next;
907
908                 if (key.offset > search_start) {
909                         hole_size = key.offset - search_start;
910
911                         if (hole_size > max_hole_size) {
912                                 max_hole_start = search_start;
913                                 max_hole_size = hole_size;
914                         }
915
916                         /*
917                          * If this free space is greater than which we need,
918                          * it must be the max free space that we have found
919                          * until now, so max_hole_start must point to the start
920                          * of this free space and the length of this free space
921                          * is stored in max_hole_size. Thus, we return
922                          * max_hole_start and max_hole_size and go back to the
923                          * caller.
924                          */
925                         if (hole_size >= num_bytes) {
926                                 ret = 0;
927                                 goto out;
928                         }
929                 }
930
931                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
932                 extent_end = key.offset + btrfs_dev_extent_length(l,
933                                                                   dev_extent);
934                 if (extent_end > search_start)
935                         search_start = extent_end;
936 next:
937                 path->slots[0]++;
938                 cond_resched();
939         }
940
941         hole_size = search_end- search_start;
942         if (hole_size > max_hole_size) {
943                 max_hole_start = search_start;
944                 max_hole_size = hole_size;
945         }
946
947         /* See above. */
948         if (hole_size < num_bytes)
949                 ret = -ENOSPC;
950         else
951                 ret = 0;
952
953 out:
954         btrfs_free_path(path);
955 error:
956         *start = max_hole_start;
957         if (len)
958                 *len = max_hole_size;
959         return ret;
960 }
961
962 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
963                           struct btrfs_device *device,
964                           u64 start)
965 {
966         int ret;
967         struct btrfs_path *path;
968         struct btrfs_root *root = device->dev_root;
969         struct btrfs_key key;
970         struct btrfs_key found_key;
971         struct extent_buffer *leaf = NULL;
972         struct btrfs_dev_extent *extent = NULL;
973
974         path = btrfs_alloc_path();
975         if (!path)
976                 return -ENOMEM;
977
978         key.objectid = device->devid;
979         key.offset = start;
980         key.type = BTRFS_DEV_EXTENT_KEY;
981
982         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
983         if (ret > 0) {
984                 ret = btrfs_previous_item(root, path, key.objectid,
985                                           BTRFS_DEV_EXTENT_KEY);
986                 BUG_ON(ret);
987                 leaf = path->nodes[0];
988                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
989                 extent = btrfs_item_ptr(leaf, path->slots[0],
990                                         struct btrfs_dev_extent);
991                 BUG_ON(found_key.offset > start || found_key.offset +
992                        btrfs_dev_extent_length(leaf, extent) < start);
993                 ret = 0;
994         } else if (ret == 0) {
995                 leaf = path->nodes[0];
996                 extent = btrfs_item_ptr(leaf, path->slots[0],
997                                         struct btrfs_dev_extent);
998         }
999         BUG_ON(ret);
1000
1001         if (device->bytes_used > 0)
1002                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1003         ret = btrfs_del_item(trans, root, path);
1004         BUG_ON(ret);
1005
1006         btrfs_free_path(path);
1007         return ret;
1008 }
1009
1010 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1011                            struct btrfs_device *device,
1012                            u64 chunk_tree, u64 chunk_objectid,
1013                            u64 chunk_offset, u64 start, u64 num_bytes)
1014 {
1015         int ret;
1016         struct btrfs_path *path;
1017         struct btrfs_root *root = device->dev_root;
1018         struct btrfs_dev_extent *extent;
1019         struct extent_buffer *leaf;
1020         struct btrfs_key key;
1021
1022         WARN_ON(!device->in_fs_metadata);
1023         path = btrfs_alloc_path();
1024         if (!path)
1025                 return -ENOMEM;
1026
1027         key.objectid = device->devid;
1028         key.offset = start;
1029         key.type = BTRFS_DEV_EXTENT_KEY;
1030         ret = btrfs_insert_empty_item(trans, root, path, &key,
1031                                       sizeof(*extent));
1032         BUG_ON(ret);
1033
1034         leaf = path->nodes[0];
1035         extent = btrfs_item_ptr(leaf, path->slots[0],
1036                                 struct btrfs_dev_extent);
1037         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1038         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1039         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1040
1041         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1042                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1043                     BTRFS_UUID_SIZE);
1044
1045         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1046         btrfs_mark_buffer_dirty(leaf);
1047         btrfs_free_path(path);
1048         return ret;
1049 }
1050
1051 static noinline int find_next_chunk(struct btrfs_root *root,
1052                                     u64 objectid, u64 *offset)
1053 {
1054         struct btrfs_path *path;
1055         int ret;
1056         struct btrfs_key key;
1057         struct btrfs_chunk *chunk;
1058         struct btrfs_key found_key;
1059
1060         path = btrfs_alloc_path();
1061         BUG_ON(!path);
1062
1063         key.objectid = objectid;
1064         key.offset = (u64)-1;
1065         key.type = BTRFS_CHUNK_ITEM_KEY;
1066
1067         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1068         if (ret < 0)
1069                 goto error;
1070
1071         BUG_ON(ret == 0);
1072
1073         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1074         if (ret) {
1075                 *offset = 0;
1076         } else {
1077                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1078                                       path->slots[0]);
1079                 if (found_key.objectid != objectid)
1080                         *offset = 0;
1081                 else {
1082                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1083                                                struct btrfs_chunk);
1084                         *offset = found_key.offset +
1085                                 btrfs_chunk_length(path->nodes[0], chunk);
1086                 }
1087         }
1088         ret = 0;
1089 error:
1090         btrfs_free_path(path);
1091         return ret;
1092 }
1093
1094 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1095 {
1096         int ret;
1097         struct btrfs_key key;
1098         struct btrfs_key found_key;
1099         struct btrfs_path *path;
1100
1101         root = root->fs_info->chunk_root;
1102
1103         path = btrfs_alloc_path();
1104         if (!path)
1105                 return -ENOMEM;
1106
1107         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1108         key.type = BTRFS_DEV_ITEM_KEY;
1109         key.offset = (u64)-1;
1110
1111         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1112         if (ret < 0)
1113                 goto error;
1114
1115         BUG_ON(ret == 0);
1116
1117         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1118                                   BTRFS_DEV_ITEM_KEY);
1119         if (ret) {
1120                 *objectid = 1;
1121         } else {
1122                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1123                                       path->slots[0]);
1124                 *objectid = found_key.offset + 1;
1125         }
1126         ret = 0;
1127 error:
1128         btrfs_free_path(path);
1129         return ret;
1130 }
1131
1132 /*
1133  * the device information is stored in the chunk root
1134  * the btrfs_device struct should be fully filled in
1135  */
1136 int btrfs_add_device(struct btrfs_trans_handle *trans,
1137                      struct btrfs_root *root,
1138                      struct btrfs_device *device)
1139 {
1140         int ret;
1141         struct btrfs_path *path;
1142         struct btrfs_dev_item *dev_item;
1143         struct extent_buffer *leaf;
1144         struct btrfs_key key;
1145         unsigned long ptr;
1146
1147         root = root->fs_info->chunk_root;
1148
1149         path = btrfs_alloc_path();
1150         if (!path)
1151                 return -ENOMEM;
1152
1153         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1154         key.type = BTRFS_DEV_ITEM_KEY;
1155         key.offset = device->devid;
1156
1157         ret = btrfs_insert_empty_item(trans, root, path, &key,
1158                                       sizeof(*dev_item));
1159         if (ret)
1160                 goto out;
1161
1162         leaf = path->nodes[0];
1163         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1164
1165         btrfs_set_device_id(leaf, dev_item, device->devid);
1166         btrfs_set_device_generation(leaf, dev_item, 0);
1167         btrfs_set_device_type(leaf, dev_item, device->type);
1168         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1169         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1170         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1171         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1172         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1173         btrfs_set_device_group(leaf, dev_item, 0);
1174         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1175         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1176         btrfs_set_device_start_offset(leaf, dev_item, 0);
1177
1178         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1179         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1180         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1181         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1182         btrfs_mark_buffer_dirty(leaf);
1183
1184         ret = 0;
1185 out:
1186         btrfs_free_path(path);
1187         return ret;
1188 }
1189
1190 static int btrfs_rm_dev_item(struct btrfs_root *root,
1191                              struct btrfs_device *device)
1192 {
1193         int ret;
1194         struct btrfs_path *path;
1195         struct btrfs_key key;
1196         struct btrfs_trans_handle *trans;
1197
1198         root = root->fs_info->chunk_root;
1199
1200         path = btrfs_alloc_path();
1201         if (!path)
1202                 return -ENOMEM;
1203
1204         trans = btrfs_start_transaction(root, 0);
1205         if (IS_ERR(trans)) {
1206                 btrfs_free_path(path);
1207                 return PTR_ERR(trans);
1208         }
1209         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1210         key.type = BTRFS_DEV_ITEM_KEY;
1211         key.offset = device->devid;
1212         lock_chunks(root);
1213
1214         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1215         if (ret < 0)
1216                 goto out;
1217
1218         if (ret > 0) {
1219                 ret = -ENOENT;
1220                 goto out;
1221         }
1222
1223         ret = btrfs_del_item(trans, root, path);
1224         if (ret)
1225                 goto out;
1226 out:
1227         btrfs_free_path(path);
1228         unlock_chunks(root);
1229         btrfs_commit_transaction(trans, root);
1230         return ret;
1231 }
1232
1233 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1234 {
1235         struct btrfs_device *device;
1236         struct btrfs_device *next_device;
1237         struct block_device *bdev;
1238         struct buffer_head *bh = NULL;
1239         struct btrfs_super_block *disk_super;
1240         u64 all_avail;
1241         u64 devid;
1242         u64 num_devices;
1243         u8 *dev_uuid;
1244         int ret = 0;
1245
1246         mutex_lock(&uuid_mutex);
1247         mutex_lock(&root->fs_info->volume_mutex);
1248
1249         all_avail = root->fs_info->avail_data_alloc_bits |
1250                 root->fs_info->avail_system_alloc_bits |
1251                 root->fs_info->avail_metadata_alloc_bits;
1252
1253         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1254             root->fs_info->fs_devices->num_devices <= 4) {
1255                 printk(KERN_ERR "btrfs: unable to go below four devices "
1256                        "on raid10\n");
1257                 ret = -EINVAL;
1258                 goto out;
1259         }
1260
1261         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1262             root->fs_info->fs_devices->num_devices <= 2) {
1263                 printk(KERN_ERR "btrfs: unable to go below two "
1264                        "devices on raid1\n");
1265                 ret = -EINVAL;
1266                 goto out;
1267         }
1268
1269         if (strcmp(device_path, "missing") == 0) {
1270                 struct list_head *devices;
1271                 struct btrfs_device *tmp;
1272
1273                 device = NULL;
1274                 devices = &root->fs_info->fs_devices->devices;
1275                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1276                 list_for_each_entry(tmp, devices, dev_list) {
1277                         if (tmp->in_fs_metadata && !tmp->bdev) {
1278                                 device = tmp;
1279                                 break;
1280                         }
1281                 }
1282                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1283                 bdev = NULL;
1284                 bh = NULL;
1285                 disk_super = NULL;
1286                 if (!device) {
1287                         printk(KERN_ERR "btrfs: no missing devices found to "
1288                                "remove\n");
1289                         goto out;
1290                 }
1291         } else {
1292                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1293                                           root->fs_info->bdev_holder);
1294                 if (IS_ERR(bdev)) {
1295                         ret = PTR_ERR(bdev);
1296                         goto out;
1297                 }
1298
1299                 set_blocksize(bdev, 4096);
1300                 bh = btrfs_read_dev_super(bdev);
1301                 if (!bh) {
1302                         ret = -EINVAL;
1303                         goto error_close;
1304                 }
1305                 disk_super = (struct btrfs_super_block *)bh->b_data;
1306                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1307                 dev_uuid = disk_super->dev_item.uuid;
1308                 device = btrfs_find_device(root, devid, dev_uuid,
1309                                            disk_super->fsid);
1310                 if (!device) {
1311                         ret = -ENOENT;
1312                         goto error_brelse;
1313                 }
1314         }
1315
1316         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1317                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1318                        "device\n");
1319                 ret = -EINVAL;
1320                 goto error_brelse;
1321         }
1322
1323         if (device->writeable) {
1324                 list_del_init(&device->dev_alloc_list);
1325                 root->fs_info->fs_devices->rw_devices--;
1326         }
1327
1328         ret = btrfs_shrink_device(device, 0);
1329         if (ret)
1330                 goto error_undo;
1331
1332         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1333         if (ret)
1334                 goto error_undo;
1335
1336         device->in_fs_metadata = 0;
1337
1338         /*
1339          * the device list mutex makes sure that we don't change
1340          * the device list while someone else is writing out all
1341          * the device supers.
1342          */
1343         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1344         list_del_init(&device->dev_list);
1345         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1346
1347         device->fs_devices->num_devices--;
1348
1349         if (device->missing)
1350                 root->fs_info->fs_devices->missing_devices--;
1351
1352         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1353                                  struct btrfs_device, dev_list);
1354         if (device->bdev == root->fs_info->sb->s_bdev)
1355                 root->fs_info->sb->s_bdev = next_device->bdev;
1356         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1357                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1358
1359         if (device->bdev) {
1360                 blkdev_put(device->bdev, device->mode);
1361                 device->bdev = NULL;
1362                 device->fs_devices->open_devices--;
1363         }
1364
1365         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1366         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1367
1368         if (device->fs_devices->open_devices == 0) {
1369                 struct btrfs_fs_devices *fs_devices;
1370                 fs_devices = root->fs_info->fs_devices;
1371                 while (fs_devices) {
1372                         if (fs_devices->seed == device->fs_devices)
1373                                 break;
1374                         fs_devices = fs_devices->seed;
1375                 }
1376                 fs_devices->seed = device->fs_devices->seed;
1377                 device->fs_devices->seed = NULL;
1378                 __btrfs_close_devices(device->fs_devices);
1379                 free_fs_devices(device->fs_devices);
1380         }
1381
1382         /*
1383          * at this point, the device is zero sized.  We want to
1384          * remove it from the devices list and zero out the old super
1385          */
1386         if (device->writeable) {
1387                 /* make sure this device isn't detected as part of
1388                  * the FS anymore
1389                  */
1390                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1391                 set_buffer_dirty(bh);
1392                 sync_dirty_buffer(bh);
1393         }
1394
1395         kfree(device->name);
1396         kfree(device);
1397         ret = 0;
1398
1399 error_brelse:
1400         brelse(bh);
1401 error_close:
1402         if (bdev)
1403                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1404 out:
1405         mutex_unlock(&root->fs_info->volume_mutex);
1406         mutex_unlock(&uuid_mutex);
1407         return ret;
1408 error_undo:
1409         if (device->writeable) {
1410                 list_add(&device->dev_alloc_list,
1411                          &root->fs_info->fs_devices->alloc_list);
1412                 root->fs_info->fs_devices->rw_devices++;
1413         }
1414         goto error_brelse;
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->name);
1605                 kfree(device);
1606                 goto error;
1607         }
1608
1609         trans = btrfs_start_transaction(root, 0);
1610         if (IS_ERR(trans)) {
1611                 kfree(device->name);
1612                 kfree(device);
1613                 ret = PTR_ERR(trans);
1614                 goto error;
1615         }
1616
1617         lock_chunks(root);
1618
1619         device->writeable = 1;
1620         device->work.func = pending_bios_fn;
1621         generate_random_uuid(device->uuid);
1622         spin_lock_init(&device->io_lock);
1623         device->generation = trans->transid;
1624         device->io_width = root->sectorsize;
1625         device->io_align = root->sectorsize;
1626         device->sector_size = root->sectorsize;
1627         device->total_bytes = i_size_read(bdev->bd_inode);
1628         device->disk_total_bytes = device->total_bytes;
1629         device->dev_root = root->fs_info->dev_root;
1630         device->bdev = bdev;
1631         device->in_fs_metadata = 1;
1632         device->mode = FMODE_EXCL;
1633         set_blocksize(device->bdev, 4096);
1634
1635         if (seeding_dev) {
1636                 sb->s_flags &= ~MS_RDONLY;
1637                 ret = btrfs_prepare_sprout(trans, root);
1638                 BUG_ON(ret);
1639         }
1640
1641         device->fs_devices = root->fs_info->fs_devices;
1642
1643         /*
1644          * we don't want write_supers to jump in here with our device
1645          * half setup
1646          */
1647         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1648         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1649         list_add(&device->dev_alloc_list,
1650                  &root->fs_info->fs_devices->alloc_list);
1651         root->fs_info->fs_devices->num_devices++;
1652         root->fs_info->fs_devices->open_devices++;
1653         root->fs_info->fs_devices->rw_devices++;
1654         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1655
1656         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1657                 root->fs_info->fs_devices->rotating = 1;
1658
1659         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1660         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1661                                     total_bytes + device->total_bytes);
1662
1663         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1664         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1665                                     total_bytes + 1);
1666         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1667
1668         if (seeding_dev) {
1669                 ret = init_first_rw_device(trans, root, device);
1670                 BUG_ON(ret);
1671                 ret = btrfs_finish_sprout(trans, root);
1672                 BUG_ON(ret);
1673         } else {
1674                 ret = btrfs_add_device(trans, root, device);
1675         }
1676
1677         /*
1678          * we've got more storage, clear any full flags on the space
1679          * infos
1680          */
1681         btrfs_clear_space_info_full(root->fs_info);
1682
1683         unlock_chunks(root);
1684         btrfs_commit_transaction(trans, root);
1685
1686         if (seeding_dev) {
1687                 mutex_unlock(&uuid_mutex);
1688                 up_write(&sb->s_umount);
1689
1690                 ret = btrfs_relocate_sys_chunks(root);
1691                 BUG_ON(ret);
1692         }
1693 out:
1694         mutex_unlock(&root->fs_info->volume_mutex);
1695         return ret;
1696 error:
1697         blkdev_put(bdev, FMODE_EXCL);
1698         if (seeding_dev) {
1699                 mutex_unlock(&uuid_mutex);
1700                 up_write(&sb->s_umount);
1701         }
1702         goto out;
1703 }
1704
1705 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1706                                         struct btrfs_device *device)
1707 {
1708         int ret;
1709         struct btrfs_path *path;
1710         struct btrfs_root *root;
1711         struct btrfs_dev_item *dev_item;
1712         struct extent_buffer *leaf;
1713         struct btrfs_key key;
1714
1715         root = device->dev_root->fs_info->chunk_root;
1716
1717         path = btrfs_alloc_path();
1718         if (!path)
1719                 return -ENOMEM;
1720
1721         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1722         key.type = BTRFS_DEV_ITEM_KEY;
1723         key.offset = device->devid;
1724
1725         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1726         if (ret < 0)
1727                 goto out;
1728
1729         if (ret > 0) {
1730                 ret = -ENOENT;
1731                 goto out;
1732         }
1733
1734         leaf = path->nodes[0];
1735         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1736
1737         btrfs_set_device_id(leaf, dev_item, device->devid);
1738         btrfs_set_device_type(leaf, dev_item, device->type);
1739         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1740         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1741         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1742         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1743         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1744         btrfs_mark_buffer_dirty(leaf);
1745
1746 out:
1747         btrfs_free_path(path);
1748         return ret;
1749 }
1750
1751 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1752                       struct btrfs_device *device, u64 new_size)
1753 {
1754         struct btrfs_super_block *super_copy =
1755                 &device->dev_root->fs_info->super_copy;
1756         u64 old_total = btrfs_super_total_bytes(super_copy);
1757         u64 diff = new_size - device->total_bytes;
1758
1759         if (!device->writeable)
1760                 return -EACCES;
1761         if (new_size <= device->total_bytes)
1762                 return -EINVAL;
1763
1764         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1765         device->fs_devices->total_rw_bytes += diff;
1766
1767         device->total_bytes = new_size;
1768         device->disk_total_bytes = new_size;
1769         btrfs_clear_space_info_full(device->dev_root->fs_info);
1770
1771         return btrfs_update_device(trans, device);
1772 }
1773
1774 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1775                       struct btrfs_device *device, u64 new_size)
1776 {
1777         int ret;
1778         lock_chunks(device->dev_root);
1779         ret = __btrfs_grow_device(trans, device, new_size);
1780         unlock_chunks(device->dev_root);
1781         return ret;
1782 }
1783
1784 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1785                             struct btrfs_root *root,
1786                             u64 chunk_tree, u64 chunk_objectid,
1787                             u64 chunk_offset)
1788 {
1789         int ret;
1790         struct btrfs_path *path;
1791         struct btrfs_key key;
1792
1793         root = root->fs_info->chunk_root;
1794         path = btrfs_alloc_path();
1795         if (!path)
1796                 return -ENOMEM;
1797
1798         key.objectid = chunk_objectid;
1799         key.offset = chunk_offset;
1800         key.type = BTRFS_CHUNK_ITEM_KEY;
1801
1802         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1803         BUG_ON(ret);
1804
1805         ret = btrfs_del_item(trans, root, path);
1806         BUG_ON(ret);
1807
1808         btrfs_free_path(path);
1809         return 0;
1810 }
1811
1812 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1813                         chunk_offset)
1814 {
1815         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1816         struct btrfs_disk_key *disk_key;
1817         struct btrfs_chunk *chunk;
1818         u8 *ptr;
1819         int ret = 0;
1820         u32 num_stripes;
1821         u32 array_size;
1822         u32 len = 0;
1823         u32 cur;
1824         struct btrfs_key key;
1825
1826         array_size = btrfs_super_sys_array_size(super_copy);
1827
1828         ptr = super_copy->sys_chunk_array;
1829         cur = 0;
1830
1831         while (cur < array_size) {
1832                 disk_key = (struct btrfs_disk_key *)ptr;
1833                 btrfs_disk_key_to_cpu(&key, disk_key);
1834
1835                 len = sizeof(*disk_key);
1836
1837                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1838                         chunk = (struct btrfs_chunk *)(ptr + len);
1839                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1840                         len += btrfs_chunk_item_size(num_stripes);
1841                 } else {
1842                         ret = -EIO;
1843                         break;
1844                 }
1845                 if (key.objectid == chunk_objectid &&
1846                     key.offset == chunk_offset) {
1847                         memmove(ptr, ptr + len, array_size - (cur + len));
1848                         array_size -= len;
1849                         btrfs_set_super_sys_array_size(super_copy, array_size);
1850                 } else {
1851                         ptr += len;
1852                         cur += len;
1853                 }
1854         }
1855         return ret;
1856 }
1857
1858 static int btrfs_relocate_chunk(struct btrfs_root *root,
1859                          u64 chunk_tree, u64 chunk_objectid,
1860                          u64 chunk_offset)
1861 {
1862         struct extent_map_tree *em_tree;
1863         struct btrfs_root *extent_root;
1864         struct btrfs_trans_handle *trans;
1865         struct extent_map *em;
1866         struct map_lookup *map;
1867         int ret;
1868         int i;
1869
1870         root = root->fs_info->chunk_root;
1871         extent_root = root->fs_info->extent_root;
1872         em_tree = &root->fs_info->mapping_tree.map_tree;
1873
1874         ret = btrfs_can_relocate(extent_root, chunk_offset);
1875         if (ret)
1876                 return -ENOSPC;
1877
1878         /* step one, relocate all the extents inside this chunk */
1879         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1880         if (ret)
1881                 return ret;
1882
1883         trans = btrfs_start_transaction(root, 0);
1884         BUG_ON(IS_ERR(trans));
1885
1886         lock_chunks(root);
1887
1888         /*
1889          * step two, delete the device extents and the
1890          * chunk tree entries
1891          */
1892         read_lock(&em_tree->lock);
1893         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1894         read_unlock(&em_tree->lock);
1895
1896         BUG_ON(em->start > chunk_offset ||
1897                em->start + em->len < chunk_offset);
1898         map = (struct map_lookup *)em->bdev;
1899
1900         for (i = 0; i < map->num_stripes; i++) {
1901                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1902                                             map->stripes[i].physical);
1903                 BUG_ON(ret);
1904
1905                 if (map->stripes[i].dev) {
1906                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1907                         BUG_ON(ret);
1908                 }
1909         }
1910         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1911                                chunk_offset);
1912
1913         BUG_ON(ret);
1914
1915         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1916
1917         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1918                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1919                 BUG_ON(ret);
1920         }
1921
1922         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1923         BUG_ON(ret);
1924
1925         write_lock(&em_tree->lock);
1926         remove_extent_mapping(em_tree, em);
1927         write_unlock(&em_tree->lock);
1928
1929         kfree(map);
1930         em->bdev = NULL;
1931
1932         /* once for the tree */
1933         free_extent_map(em);
1934         /* once for us */
1935         free_extent_map(em);
1936
1937         unlock_chunks(root);
1938         btrfs_end_transaction(trans, root);
1939         return 0;
1940 }
1941
1942 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1943 {
1944         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1945         struct btrfs_path *path;
1946         struct extent_buffer *leaf;
1947         struct btrfs_chunk *chunk;
1948         struct btrfs_key key;
1949         struct btrfs_key found_key;
1950         u64 chunk_tree = chunk_root->root_key.objectid;
1951         u64 chunk_type;
1952         bool retried = false;
1953         int failed = 0;
1954         int ret;
1955
1956         path = btrfs_alloc_path();
1957         if (!path)
1958                 return -ENOMEM;
1959
1960 again:
1961         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1962         key.offset = (u64)-1;
1963         key.type = BTRFS_CHUNK_ITEM_KEY;
1964
1965         while (1) {
1966                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1967                 if (ret < 0)
1968                         goto error;
1969                 BUG_ON(ret == 0);
1970
1971                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1972                                           key.type);
1973                 if (ret < 0)
1974                         goto error;
1975                 if (ret > 0)
1976                         break;
1977
1978                 leaf = path->nodes[0];
1979                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1980
1981                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1982                                        struct btrfs_chunk);
1983                 chunk_type = btrfs_chunk_type(leaf, chunk);
1984                 btrfs_release_path(chunk_root, path);
1985
1986                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1987                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1988                                                    found_key.objectid,
1989                                                    found_key.offset);
1990                         if (ret == -ENOSPC)
1991                                 failed++;
1992                         else if (ret)
1993                                 BUG();
1994                 }
1995
1996                 if (found_key.offset == 0)
1997                         break;
1998                 key.offset = found_key.offset - 1;
1999         }
2000         ret = 0;
2001         if (failed && !retried) {
2002                 failed = 0;
2003                 retried = true;
2004                 goto again;
2005         } else if (failed && retried) {
2006                 WARN_ON(1);
2007                 ret = -ENOSPC;
2008         }
2009 error:
2010         btrfs_free_path(path);
2011         return ret;
2012 }
2013
2014 static u64 div_factor(u64 num, int factor)
2015 {
2016         if (factor == 10)
2017                 return num;
2018         num *= factor;
2019         do_div(num, 10);
2020         return num;
2021 }
2022
2023 int btrfs_balance(struct btrfs_root *dev_root)
2024 {
2025         int ret;
2026         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2027         struct btrfs_device *device;
2028         u64 old_size;
2029         u64 size_to_free;
2030         struct btrfs_path *path;
2031         struct btrfs_key key;
2032         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2033         struct btrfs_trans_handle *trans;
2034         struct btrfs_key found_key;
2035
2036         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2037                 return -EROFS;
2038
2039         if (!capable(CAP_SYS_ADMIN))
2040                 return -EPERM;
2041
2042         mutex_lock(&dev_root->fs_info->volume_mutex);
2043         dev_root = dev_root->fs_info->dev_root;
2044
2045         /* step one make some room on all the devices */
2046         list_for_each_entry(device, devices, dev_list) {
2047                 old_size = device->total_bytes;
2048                 size_to_free = div_factor(old_size, 1);
2049                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2050                 if (!device->writeable ||
2051                     device->total_bytes - device->bytes_used > size_to_free)
2052                         continue;
2053
2054                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2055                 if (ret == -ENOSPC)
2056                         break;
2057                 BUG_ON(ret);
2058
2059                 trans = btrfs_start_transaction(dev_root, 0);
2060                 BUG_ON(IS_ERR(trans));
2061
2062                 ret = btrfs_grow_device(trans, device, old_size);
2063                 BUG_ON(ret);
2064
2065                 btrfs_end_transaction(trans, dev_root);
2066         }
2067
2068         /* step two, relocate all the chunks */
2069         path = btrfs_alloc_path();
2070         BUG_ON(!path);
2071
2072         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2073         key.offset = (u64)-1;
2074         key.type = BTRFS_CHUNK_ITEM_KEY;
2075
2076         while (1) {
2077                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2078                 if (ret < 0)
2079                         goto error;
2080
2081                 /*
2082                  * this shouldn't happen, it means the last relocate
2083                  * failed
2084                  */
2085                 if (ret == 0)
2086                         break;
2087
2088                 ret = btrfs_previous_item(chunk_root, path, 0,
2089                                           BTRFS_CHUNK_ITEM_KEY);
2090                 if (ret)
2091                         break;
2092
2093                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2094                                       path->slots[0]);
2095                 if (found_key.objectid != key.objectid)
2096                         break;
2097
2098                 /* chunk zero is special */
2099                 if (found_key.offset == 0)
2100                         break;
2101
2102                 btrfs_release_path(chunk_root, path);
2103                 ret = btrfs_relocate_chunk(chunk_root,
2104                                            chunk_root->root_key.objectid,
2105                                            found_key.objectid,
2106                                            found_key.offset);
2107                 BUG_ON(ret && ret != -ENOSPC);
2108                 key.offset = found_key.offset - 1;
2109         }
2110         ret = 0;
2111 error:
2112         btrfs_free_path(path);
2113         mutex_unlock(&dev_root->fs_info->volume_mutex);
2114         return ret;
2115 }
2116
2117 /*
2118  * shrinking a device means finding all of the device extents past
2119  * the new size, and then following the back refs to the chunks.
2120  * The chunk relocation code actually frees the device extent
2121  */
2122 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2123 {
2124         struct btrfs_trans_handle *trans;
2125         struct btrfs_root *root = device->dev_root;
2126         struct btrfs_dev_extent *dev_extent = NULL;
2127         struct btrfs_path *path;
2128         u64 length;
2129         u64 chunk_tree;
2130         u64 chunk_objectid;
2131         u64 chunk_offset;
2132         int ret;
2133         int slot;
2134         int failed = 0;
2135         bool retried = false;
2136         struct extent_buffer *l;
2137         struct btrfs_key key;
2138         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2139         u64 old_total = btrfs_super_total_bytes(super_copy);
2140         u64 old_size = device->total_bytes;
2141         u64 diff = device->total_bytes - new_size;
2142
2143         if (new_size >= device->total_bytes)
2144                 return -EINVAL;
2145
2146         path = btrfs_alloc_path();
2147         if (!path)
2148                 return -ENOMEM;
2149
2150         path->reada = 2;
2151
2152         lock_chunks(root);
2153
2154         device->total_bytes = new_size;
2155         if (device->writeable)
2156                 device->fs_devices->total_rw_bytes -= diff;
2157         unlock_chunks(root);
2158
2159 again:
2160         key.objectid = device->devid;
2161         key.offset = (u64)-1;
2162         key.type = BTRFS_DEV_EXTENT_KEY;
2163
2164         while (1) {
2165                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2166                 if (ret < 0)
2167                         goto done;
2168
2169                 ret = btrfs_previous_item(root, path, 0, key.type);
2170                 if (ret < 0)
2171                         goto done;
2172                 if (ret) {
2173                         ret = 0;
2174                         btrfs_release_path(root, path);
2175                         break;
2176                 }
2177
2178                 l = path->nodes[0];
2179                 slot = path->slots[0];
2180                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2181
2182                 if (key.objectid != device->devid) {
2183                         btrfs_release_path(root, path);
2184                         break;
2185                 }
2186
2187                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2188                 length = btrfs_dev_extent_length(l, dev_extent);
2189
2190                 if (key.offset + length <= new_size) {
2191                         btrfs_release_path(root, path);
2192                         break;
2193                 }
2194
2195                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2196                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2197                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2198                 btrfs_release_path(root, path);
2199
2200                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2201                                            chunk_offset);
2202                 if (ret && ret != -ENOSPC)
2203                         goto done;
2204                 if (ret == -ENOSPC)
2205                         failed++;
2206                 key.offset -= 1;
2207         }
2208
2209         if (failed && !retried) {
2210                 failed = 0;
2211                 retried = true;
2212                 goto again;
2213         } else if (failed && retried) {
2214                 ret = -ENOSPC;
2215                 lock_chunks(root);
2216
2217                 device->total_bytes = old_size;
2218                 if (device->writeable)
2219                         device->fs_devices->total_rw_bytes += diff;
2220                 unlock_chunks(root);
2221                 goto done;
2222         }
2223
2224         /* Shrinking succeeded, else we would be at "done". */
2225         trans = btrfs_start_transaction(root, 0);
2226         if (IS_ERR(trans)) {
2227                 ret = PTR_ERR(trans);
2228                 goto done;
2229         }
2230
2231         lock_chunks(root);
2232
2233         device->disk_total_bytes = new_size;
2234         /* Now btrfs_update_device() will change the on-disk size. */
2235         ret = btrfs_update_device(trans, device);
2236         if (ret) {
2237                 unlock_chunks(root);
2238                 btrfs_end_transaction(trans, root);
2239                 goto done;
2240         }
2241         WARN_ON(diff > old_total);
2242         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2243         unlock_chunks(root);
2244         btrfs_end_transaction(trans, root);
2245 done:
2246         btrfs_free_path(path);
2247         return ret;
2248 }
2249
2250 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2251                            struct btrfs_root *root,
2252                            struct btrfs_key *key,
2253                            struct btrfs_chunk *chunk, int item_size)
2254 {
2255         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2256         struct btrfs_disk_key disk_key;
2257         u32 array_size;
2258         u8 *ptr;
2259
2260         array_size = btrfs_super_sys_array_size(super_copy);
2261         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2262                 return -EFBIG;
2263
2264         ptr = super_copy->sys_chunk_array + array_size;
2265         btrfs_cpu_key_to_disk(&disk_key, key);
2266         memcpy(ptr, &disk_key, sizeof(disk_key));
2267         ptr += sizeof(disk_key);
2268         memcpy(ptr, chunk, item_size);
2269         item_size += sizeof(disk_key);
2270         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2271         return 0;
2272 }
2273
2274 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2275                                         int num_stripes, int sub_stripes)
2276 {
2277         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2278                 return calc_size;
2279         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2280                 return calc_size * (num_stripes / sub_stripes);
2281         else
2282                 return calc_size * num_stripes;
2283 }
2284
2285 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2286                                  int *num_stripes, int *min_stripes,
2287                                  int *sub_stripes)
2288 {
2289         *num_stripes = 1;
2290         *min_stripes = 1;
2291         *sub_stripes = 0;
2292
2293         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2294                 *num_stripes = fs_devices->rw_devices;
2295                 *min_stripes = 2;
2296         }
2297         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2298                 *num_stripes = 2;
2299                 *min_stripes = 2;
2300         }
2301         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2302                 if (fs_devices->rw_devices < 2)
2303                         return -ENOSPC;
2304                 *num_stripes = 2;
2305                 *min_stripes = 2;
2306         }
2307         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2308                 *num_stripes = fs_devices->rw_devices;
2309                 if (*num_stripes < 4)
2310                         return -ENOSPC;
2311                 *num_stripes &= ~(u32)1;
2312                 *sub_stripes = 2;
2313                 *min_stripes = 4;
2314         }
2315
2316         return 0;
2317 }
2318
2319 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2320                                     u64 proposed_size, u64 type,
2321                                     int num_stripes, int small_stripe)
2322 {
2323         int min_stripe_size = 1 * 1024 * 1024;
2324         u64 calc_size = proposed_size;
2325         u64 max_chunk_size = calc_size;
2326         int ncopies = 1;
2327
2328         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2329                     BTRFS_BLOCK_GROUP_DUP |
2330                     BTRFS_BLOCK_GROUP_RAID10))
2331                 ncopies = 2;
2332
2333         if (type & BTRFS_BLOCK_GROUP_DATA) {
2334                 max_chunk_size = 10 * calc_size;
2335                 min_stripe_size = 64 * 1024 * 1024;
2336         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2337                 max_chunk_size = 256 * 1024 * 1024;
2338                 min_stripe_size = 32 * 1024 * 1024;
2339         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2340                 calc_size = 8 * 1024 * 1024;
2341                 max_chunk_size = calc_size * 2;
2342                 min_stripe_size = 1 * 1024 * 1024;
2343         }
2344
2345         /* we don't want a chunk larger than 10% of writeable space */
2346         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2347                              max_chunk_size);
2348
2349         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2350                 calc_size = max_chunk_size * ncopies;
2351                 do_div(calc_size, num_stripes);
2352                 do_div(calc_size, BTRFS_STRIPE_LEN);
2353                 calc_size *= BTRFS_STRIPE_LEN;
2354         }
2355
2356         /* we don't want tiny stripes */
2357         if (!small_stripe)
2358                 calc_size = max_t(u64, min_stripe_size, calc_size);
2359
2360         /*
2361          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2362          * we end up with something bigger than a stripe
2363          */
2364         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2365
2366         do_div(calc_size, BTRFS_STRIPE_LEN);
2367         calc_size *= BTRFS_STRIPE_LEN;
2368
2369         return calc_size;
2370 }
2371
2372 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2373                                                       int num_stripes)
2374 {
2375         struct map_lookup *new;
2376         size_t len = map_lookup_size(num_stripes);
2377
2378         BUG_ON(map->num_stripes < num_stripes);
2379
2380         if (map->num_stripes == num_stripes)
2381                 return map;
2382
2383         new = kmalloc(len, GFP_NOFS);
2384         if (!new) {
2385                 /* just change map->num_stripes */
2386                 map->num_stripes = num_stripes;
2387                 return map;
2388         }
2389
2390         memcpy(new, map, len);
2391         new->num_stripes = num_stripes;
2392         kfree(map);
2393         return new;
2394 }
2395
2396 /*
2397  * helper to allocate device space from btrfs_device_info, in which we stored
2398  * max free space information of every device. It is used when we can not
2399  * allocate chunks by default size.
2400  *
2401  * By this helper, we can allocate a new chunk as larger as possible.
2402  */
2403 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2404                                     struct btrfs_fs_devices *fs_devices,
2405                                     struct btrfs_device_info *devices,
2406                                     int nr_device, u64 type,
2407                                     struct map_lookup **map_lookup,
2408                                     int min_stripes, u64 *stripe_size)
2409 {
2410         int i, index, sort_again = 0;
2411         int min_devices = min_stripes;
2412         u64 max_avail, min_free;
2413         struct map_lookup *map = *map_lookup;
2414         int ret;
2415
2416         if (nr_device < min_stripes)
2417                 return -ENOSPC;
2418
2419         btrfs_descending_sort_devices(devices, nr_device);
2420
2421         max_avail = devices[0].max_avail;
2422         if (!max_avail)
2423                 return -ENOSPC;
2424
2425         for (i = 0; i < nr_device; i++) {
2426                 /*
2427                  * if dev_offset = 0, it means the free space of this device
2428                  * is less than what we need, and we didn't search max avail
2429                  * extent on this device, so do it now.
2430                  */
2431                 if (!devices[i].dev_offset) {
2432                         ret = find_free_dev_extent(trans, devices[i].dev,
2433                                                    max_avail,
2434                                                    &devices[i].dev_offset,
2435                                                    &devices[i].max_avail);
2436                         if (ret != 0 && ret != -ENOSPC)
2437                                 return ret;
2438                         sort_again = 1;
2439                 }
2440         }
2441
2442         /* we update the max avail free extent of each devices, sort again */
2443         if (sort_again)
2444                 btrfs_descending_sort_devices(devices, nr_device);
2445
2446         if (type & BTRFS_BLOCK_GROUP_DUP)
2447                 min_devices = 1;
2448
2449         if (!devices[min_devices - 1].max_avail)
2450                 return -ENOSPC;
2451
2452         max_avail = devices[min_devices - 1].max_avail;
2453         if (type & BTRFS_BLOCK_GROUP_DUP)
2454                 do_div(max_avail, 2);
2455
2456         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2457                                              min_stripes, 1);
2458         if (type & BTRFS_BLOCK_GROUP_DUP)
2459                 min_free = max_avail * 2;
2460         else
2461                 min_free = max_avail;
2462
2463         if (min_free > devices[min_devices - 1].max_avail)
2464                 return -ENOSPC;
2465
2466         map = __shrink_map_lookup_stripes(map, min_stripes);
2467         *stripe_size = max_avail;
2468
2469         index = 0;
2470         for (i = 0; i < min_stripes; i++) {
2471                 map->stripes[i].dev = devices[index].dev;
2472                 map->stripes[i].physical = devices[index].dev_offset;
2473                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2474                         i++;
2475                         map->stripes[i].dev = devices[index].dev;
2476                         map->stripes[i].physical = devices[index].dev_offset +
2477                                                    max_avail;
2478                 }
2479                 index++;
2480         }
2481         *map_lookup = map;
2482
2483         return 0;
2484 }
2485
2486 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2487                                struct btrfs_root *extent_root,
2488                                struct map_lookup **map_ret,
2489                                u64 *num_bytes, u64 *stripe_size,
2490                                u64 start, u64 type)
2491 {
2492         struct btrfs_fs_info *info = extent_root->fs_info;
2493         struct btrfs_device *device = NULL;
2494         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2495         struct list_head *cur;
2496         struct map_lookup *map;
2497         struct extent_map_tree *em_tree;
2498         struct extent_map *em;
2499         struct btrfs_device_info *devices_info;
2500         struct list_head private_devs;
2501         u64 calc_size = 1024 * 1024 * 1024;
2502         u64 min_free;
2503         u64 avail;
2504         u64 dev_offset;
2505         int num_stripes;
2506         int min_stripes;
2507         int sub_stripes;
2508         int min_devices;        /* the min number of devices we need */
2509         int i;
2510         int ret;
2511         int index;
2512
2513         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2514             (type & BTRFS_BLOCK_GROUP_DUP)) {
2515                 WARN_ON(1);
2516                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2517         }
2518         if (list_empty(&fs_devices->alloc_list))
2519                 return -ENOSPC;
2520
2521         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2522                                     &min_stripes, &sub_stripes);
2523         if (ret)
2524                 return ret;
2525
2526         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2527                                GFP_NOFS);
2528         if (!devices_info)
2529                 return -ENOMEM;
2530
2531         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2532         if (!map) {
2533                 ret = -ENOMEM;
2534                 goto error;
2535         }
2536         map->num_stripes = num_stripes;
2537
2538         cur = fs_devices->alloc_list.next;
2539         index = 0;
2540         i = 0;
2541
2542         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2543                                              num_stripes, 0);
2544
2545         if (type & BTRFS_BLOCK_GROUP_DUP) {
2546                 min_free = calc_size * 2;
2547                 min_devices = 1;
2548         } else {
2549                 min_free = calc_size;
2550                 min_devices = min_stripes;
2551         }
2552
2553         INIT_LIST_HEAD(&private_devs);
2554         while (index < num_stripes) {
2555                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2556                 BUG_ON(!device->writeable);
2557                 if (device->total_bytes > device->bytes_used)
2558                         avail = device->total_bytes - device->bytes_used;
2559                 else
2560                         avail = 0;
2561                 cur = cur->next;
2562
2563                 if (device->in_fs_metadata && avail >= min_free) {
2564                         ret = find_free_dev_extent(trans, device, min_free,
2565                                                    &devices_info[i].dev_offset,
2566                                                    &devices_info[i].max_avail);
2567                         if (ret == 0) {
2568                                 list_move_tail(&device->dev_alloc_list,
2569                                                &private_devs);
2570                                 map->stripes[index].dev = device;
2571                                 map->stripes[index].physical =
2572                                                 devices_info[i].dev_offset;
2573                                 index++;
2574                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2575                                         map->stripes[index].dev = device;
2576                                         map->stripes[index].physical =
2577                                                 devices_info[i].dev_offset +
2578                                                 calc_size;
2579                                         index++;
2580                                 }
2581                         } else if (ret != -ENOSPC)
2582                                 goto error;
2583
2584                         devices_info[i].dev = device;
2585                         i++;
2586                 } else if (device->in_fs_metadata &&
2587                            avail >= BTRFS_STRIPE_LEN) {
2588                         devices_info[i].dev = device;
2589                         devices_info[i].max_avail = avail;
2590                         i++;
2591                 }
2592
2593                 if (cur == &fs_devices->alloc_list)
2594                         break;
2595         }
2596
2597         list_splice(&private_devs, &fs_devices->alloc_list);
2598         if (index < num_stripes) {
2599                 if (index >= min_stripes) {
2600                         num_stripes = index;
2601                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2602                                 num_stripes /= sub_stripes;
2603                                 num_stripes *= sub_stripes;
2604                         }
2605
2606                         map = __shrink_map_lookup_stripes(map, num_stripes);
2607                 } else if (i >= min_devices) {
2608                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2609                                                        devices_info, i, type,
2610                                                        &map, min_stripes,
2611                                                        &calc_size);
2612                         if (ret)
2613                                 goto error;
2614                 } else {
2615                         ret = -ENOSPC;
2616                         goto error;
2617                 }
2618         }
2619         map->sector_size = extent_root->sectorsize;
2620         map->stripe_len = BTRFS_STRIPE_LEN;
2621         map->io_align = BTRFS_STRIPE_LEN;
2622         map->io_width = BTRFS_STRIPE_LEN;
2623         map->type = type;
2624         map->sub_stripes = sub_stripes;
2625
2626         *map_ret = map;
2627         *stripe_size = calc_size;
2628         *num_bytes = chunk_bytes_by_type(type, calc_size,
2629                                          map->num_stripes, sub_stripes);
2630
2631         trace_btrfs_chunk_alloc(info->chunk_root, map, start, *num_bytes);
2632
2633         em = alloc_extent_map(GFP_NOFS);
2634         if (!em) {
2635                 ret = -ENOMEM;
2636                 goto error;
2637         }
2638         em->bdev = (struct block_device *)map;
2639         em->start = start;
2640         em->len = *num_bytes;
2641         em->block_start = 0;
2642         em->block_len = em->len;
2643
2644         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2645         write_lock(&em_tree->lock);
2646         ret = add_extent_mapping(em_tree, em);
2647         write_unlock(&em_tree->lock);
2648         BUG_ON(ret);
2649         free_extent_map(em);
2650
2651         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2652                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2653                                      start, *num_bytes);
2654         BUG_ON(ret);
2655
2656         index = 0;
2657         while (index < map->num_stripes) {
2658                 device = map->stripes[index].dev;
2659                 dev_offset = map->stripes[index].physical;
2660
2661                 ret = btrfs_alloc_dev_extent(trans, device,
2662                                 info->chunk_root->root_key.objectid,
2663                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2664                                 start, dev_offset, calc_size);
2665                 BUG_ON(ret);
2666                 index++;
2667         }
2668
2669         kfree(devices_info);
2670         return 0;
2671
2672 error:
2673         kfree(map);
2674         kfree(devices_info);
2675         return ret;
2676 }
2677
2678 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2679                                 struct btrfs_root *extent_root,
2680                                 struct map_lookup *map, u64 chunk_offset,
2681                                 u64 chunk_size, u64 stripe_size)
2682 {
2683         u64 dev_offset;
2684         struct btrfs_key key;
2685         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2686         struct btrfs_device *device;
2687         struct btrfs_chunk *chunk;
2688         struct btrfs_stripe *stripe;
2689         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2690         int index = 0;
2691         int ret;
2692
2693         chunk = kzalloc(item_size, GFP_NOFS);
2694         if (!chunk)
2695                 return -ENOMEM;
2696
2697         index = 0;
2698         while (index < map->num_stripes) {
2699                 device = map->stripes[index].dev;
2700                 device->bytes_used += stripe_size;
2701                 ret = btrfs_update_device(trans, device);
2702                 BUG_ON(ret);
2703                 index++;
2704         }
2705
2706         index = 0;
2707         stripe = &chunk->stripe;
2708         while (index < map->num_stripes) {
2709                 device = map->stripes[index].dev;
2710                 dev_offset = map->stripes[index].physical;
2711
2712                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2713                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2714                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2715                 stripe++;
2716                 index++;
2717         }
2718
2719         btrfs_set_stack_chunk_length(chunk, chunk_size);
2720         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2721         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2722         btrfs_set_stack_chunk_type(chunk, map->type);
2723         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2724         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2725         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2726         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2727         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2728
2729         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2730         key.type = BTRFS_CHUNK_ITEM_KEY;
2731         key.offset = chunk_offset;
2732
2733         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2734         BUG_ON(ret);
2735
2736         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2737                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2738                                              item_size);
2739                 BUG_ON(ret);
2740         }
2741
2742         kfree(chunk);
2743         return 0;
2744 }
2745
2746 /*
2747  * Chunk allocation falls into two parts. The first part does works
2748  * that make the new allocated chunk useable, but not do any operation
2749  * that modifies the chunk tree. The second part does the works that
2750  * require modifying the chunk tree. This division is important for the
2751  * bootstrap process of adding storage to a seed btrfs.
2752  */
2753 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2754                       struct btrfs_root *extent_root, u64 type)
2755 {
2756         u64 chunk_offset;
2757         u64 chunk_size;
2758         u64 stripe_size;
2759         struct map_lookup *map;
2760         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2761         int ret;
2762
2763         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2764                               &chunk_offset);
2765         if (ret)
2766                 return ret;
2767
2768         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2769                                   &stripe_size, chunk_offset, type);
2770         if (ret)
2771                 return ret;
2772
2773         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2774                                    chunk_size, stripe_size);
2775         BUG_ON(ret);
2776         return 0;
2777 }
2778
2779 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2780                                          struct btrfs_root *root,
2781                                          struct btrfs_device *device)
2782 {
2783         u64 chunk_offset;
2784         u64 sys_chunk_offset;
2785         u64 chunk_size;
2786         u64 sys_chunk_size;
2787         u64 stripe_size;
2788         u64 sys_stripe_size;
2789         u64 alloc_profile;
2790         struct map_lookup *map;
2791         struct map_lookup *sys_map;
2792         struct btrfs_fs_info *fs_info = root->fs_info;
2793         struct btrfs_root *extent_root = fs_info->extent_root;
2794         int ret;
2795
2796         ret = find_next_chunk(fs_info->chunk_root,
2797                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2798         BUG_ON(ret);
2799
2800         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2801                         (fs_info->metadata_alloc_profile &
2802                          fs_info->avail_metadata_alloc_bits);
2803         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2804
2805         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2806                                   &stripe_size, chunk_offset, alloc_profile);
2807         BUG_ON(ret);
2808
2809         sys_chunk_offset = chunk_offset + chunk_size;
2810
2811         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2812                         (fs_info->system_alloc_profile &
2813                          fs_info->avail_system_alloc_bits);
2814         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2815
2816         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2817                                   &sys_chunk_size, &sys_stripe_size,
2818                                   sys_chunk_offset, alloc_profile);
2819         BUG_ON(ret);
2820
2821         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2822         BUG_ON(ret);
2823
2824         /*
2825          * Modifying chunk tree needs allocating new blocks from both
2826          * system block group and metadata block group. So we only can
2827          * do operations require modifying the chunk tree after both
2828          * block groups were created.
2829          */
2830         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2831                                    chunk_size, stripe_size);
2832         BUG_ON(ret);
2833
2834         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2835                                    sys_chunk_offset, sys_chunk_size,
2836                                    sys_stripe_size);
2837         BUG_ON(ret);
2838         return 0;
2839 }
2840
2841 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2842 {
2843         struct extent_map *em;
2844         struct map_lookup *map;
2845         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2846         int readonly = 0;
2847         int i;
2848
2849         read_lock(&map_tree->map_tree.lock);
2850         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2851         read_unlock(&map_tree->map_tree.lock);
2852         if (!em)
2853                 return 1;
2854
2855         if (btrfs_test_opt(root, DEGRADED)) {
2856                 free_extent_map(em);
2857                 return 0;
2858         }
2859
2860         map = (struct map_lookup *)em->bdev;
2861         for (i = 0; i < map->num_stripes; i++) {
2862                 if (!map->stripes[i].dev->writeable) {
2863                         readonly = 1;
2864                         break;
2865                 }
2866         }
2867         free_extent_map(em);
2868         return readonly;
2869 }
2870
2871 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2872 {
2873         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2874 }
2875
2876 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2877 {
2878         struct extent_map *em;
2879
2880         while (1) {
2881                 write_lock(&tree->map_tree.lock);
2882                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2883                 if (em)
2884                         remove_extent_mapping(&tree->map_tree, em);
2885                 write_unlock(&tree->map_tree.lock);
2886                 if (!em)
2887                         break;
2888                 kfree(em->bdev);
2889                 /* once for us */
2890                 free_extent_map(em);
2891                 /* once for the tree */
2892                 free_extent_map(em);
2893         }
2894 }
2895
2896 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2897 {
2898         struct extent_map *em;
2899         struct map_lookup *map;
2900         struct extent_map_tree *em_tree = &map_tree->map_tree;
2901         int ret;
2902
2903         read_lock(&em_tree->lock);
2904         em = lookup_extent_mapping(em_tree, logical, len);
2905         read_unlock(&em_tree->lock);
2906         BUG_ON(!em);
2907
2908         BUG_ON(em->start > logical || em->start + em->len < logical);
2909         map = (struct map_lookup *)em->bdev;
2910         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2911                 ret = map->num_stripes;
2912         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2913                 ret = map->sub_stripes;
2914         else
2915                 ret = 1;
2916         free_extent_map(em);
2917         return ret;
2918 }
2919
2920 static int find_live_mirror(struct map_lookup *map, int first, int num,
2921                             int optimal)
2922 {
2923         int i;
2924         if (map->stripes[optimal].dev->bdev)
2925                 return optimal;
2926         for (i = first; i < first + num; i++) {
2927                 if (map->stripes[i].dev->bdev)
2928                         return i;
2929         }
2930         /* we couldn't find one that doesn't fail.  Just return something
2931          * and the io error handling code will clean up eventually
2932          */
2933         return optimal;
2934 }
2935
2936 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2937                              u64 logical, u64 *length,
2938                              struct btrfs_multi_bio **multi_ret,
2939                              int mirror_num, struct page *unplug_page)
2940 {
2941         struct extent_map *em;
2942         struct map_lookup *map;
2943         struct extent_map_tree *em_tree = &map_tree->map_tree;
2944         u64 offset;
2945         u64 stripe_offset;
2946         u64 stripe_end_offset;
2947         u64 stripe_nr;
2948         u64 stripe_nr_orig;
2949         u64 stripe_nr_end;
2950         int stripes_allocated = 8;
2951         int stripes_required = 1;
2952         int stripe_index;
2953         int i;
2954         int num_stripes;
2955         int max_errors = 0;
2956         struct btrfs_multi_bio *multi = NULL;
2957
2958         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2959                 stripes_allocated = 1;
2960 again:
2961         if (multi_ret) {
2962                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2963                                 GFP_NOFS);
2964                 if (!multi)
2965                         return -ENOMEM;
2966
2967                 atomic_set(&multi->error, 0);
2968         }
2969
2970         read_lock(&em_tree->lock);
2971         em = lookup_extent_mapping(em_tree, logical, *length);
2972         read_unlock(&em_tree->lock);
2973
2974         if (!em && unplug_page) {
2975                 kfree(multi);
2976                 return 0;
2977         }
2978
2979         if (!em) {
2980                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2981                        (unsigned long long)logical,
2982                        (unsigned long long)*length);
2983                 BUG();
2984         }
2985
2986         BUG_ON(em->start > logical || em->start + em->len < logical);
2987         map = (struct map_lookup *)em->bdev;
2988         offset = logical - em->start;
2989
2990         if (mirror_num > map->num_stripes)
2991                 mirror_num = 0;
2992
2993         /* if our multi bio struct is too small, back off and try again */
2994         if (rw & REQ_WRITE) {
2995                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2996                                  BTRFS_BLOCK_GROUP_DUP)) {
2997                         stripes_required = map->num_stripes;
2998                         max_errors = 1;
2999                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3000                         stripes_required = map->sub_stripes;
3001                         max_errors = 1;
3002                 }
3003         }
3004         if (rw & REQ_DISCARD) {
3005                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3006                                  BTRFS_BLOCK_GROUP_RAID1 |
3007                                  BTRFS_BLOCK_GROUP_DUP |
3008                                  BTRFS_BLOCK_GROUP_RAID10)) {
3009                         stripes_required = map->num_stripes;
3010                 }
3011         }
3012         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
3013             stripes_allocated < stripes_required) {
3014                 stripes_allocated = map->num_stripes;
3015                 free_extent_map(em);
3016                 kfree(multi);
3017                 goto again;
3018         }
3019         stripe_nr = offset;
3020         /*
3021          * stripe_nr counts the total number of stripes we have to stride
3022          * to get to this block
3023          */
3024         do_div(stripe_nr, map->stripe_len);
3025
3026         stripe_offset = stripe_nr * map->stripe_len;
3027         BUG_ON(offset < stripe_offset);
3028
3029         /* stripe_offset is the offset of this block in its stripe*/
3030         stripe_offset = offset - stripe_offset;
3031
3032         if (rw & REQ_DISCARD)
3033                 *length = min_t(u64, em->len - offset, *length);
3034         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3035                               BTRFS_BLOCK_GROUP_RAID1 |
3036                               BTRFS_BLOCK_GROUP_RAID10 |
3037                               BTRFS_BLOCK_GROUP_DUP)) {
3038                 /* we limit the length of each bio to what fits in a stripe */
3039                 *length = min_t(u64, em->len - offset,
3040                                 map->stripe_len - stripe_offset);
3041         } else {
3042                 *length = em->len - offset;
3043         }
3044
3045         if (!multi_ret && !unplug_page)
3046                 goto out;
3047
3048         num_stripes = 1;
3049         stripe_index = 0;
3050         stripe_nr_orig = stripe_nr;
3051         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3052                         (~(map->stripe_len - 1));
3053         do_div(stripe_nr_end, map->stripe_len);
3054         stripe_end_offset = stripe_nr_end * map->stripe_len -
3055                             (offset + *length);
3056         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3057                 if (rw & REQ_DISCARD)
3058                         num_stripes = min_t(u64, map->num_stripes,
3059                                             stripe_nr_end - stripe_nr_orig);
3060                 stripe_index = do_div(stripe_nr, map->num_stripes);
3061         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3062                 if (unplug_page || (rw & (REQ_WRITE | REQ_DISCARD)))
3063                         num_stripes = map->num_stripes;
3064                 else if (mirror_num)
3065                         stripe_index = mirror_num - 1;
3066                 else {
3067                         stripe_index = find_live_mirror(map, 0,
3068                                             map->num_stripes,
3069                                             current->pid % map->num_stripes);
3070                 }
3071
3072         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3073                 if (rw & (REQ_WRITE | REQ_DISCARD))
3074                         num_stripes = map->num_stripes;
3075                 else if (mirror_num)
3076                         stripe_index = mirror_num - 1;
3077
3078         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3079                 int factor = map->num_stripes / map->sub_stripes;
3080
3081                 stripe_index = do_div(stripe_nr, factor);
3082                 stripe_index *= map->sub_stripes;
3083
3084                 if (unplug_page || (rw & REQ_WRITE))
3085                         num_stripes = map->sub_stripes;
3086                 else if (rw & REQ_DISCARD)
3087                         num_stripes = min_t(u64, map->sub_stripes *
3088                                             (stripe_nr_end - stripe_nr_orig),
3089                                             map->num_stripes);
3090                 else if (mirror_num)
3091                         stripe_index += mirror_num - 1;
3092                 else {
3093                         stripe_index = find_live_mirror(map, stripe_index,
3094                                               map->sub_stripes, stripe_index +
3095                                               current->pid % map->sub_stripes);
3096                 }
3097         } else {
3098                 /*
3099                  * after this do_div call, stripe_nr is the number of stripes
3100                  * on this device we have to walk to find the data, and
3101                  * stripe_index is the number of our device in the stripe array
3102                  */
3103                 stripe_index = do_div(stripe_nr, map->num_stripes);
3104         }
3105         BUG_ON(stripe_index >= map->num_stripes);
3106
3107         if (rw & REQ_DISCARD) {
3108                 for (i = 0; i < num_stripes; i++) {
3109                         multi->stripes[i].physical =
3110                                 map->stripes[stripe_index].physical +
3111                                 stripe_offset + stripe_nr * map->stripe_len;
3112                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3113
3114                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3115                                 u64 stripes;
3116                                 u32 last_stripe = 0;
3117                                 int j;
3118
3119                                 div_u64_rem(stripe_nr_end - 1,
3120                                             map->num_stripes,
3121                                             &last_stripe);
3122
3123                                 for (j = 0; j < map->num_stripes; j++) {
3124                                         u32 test;
3125
3126                                         div_u64_rem(stripe_nr_end - 1 - j,
3127                                                     map->num_stripes, &test);
3128                                         if (test == stripe_index)
3129                                                 break;
3130                                 }
3131                                 stripes = stripe_nr_end - 1 - j;
3132                                 do_div(stripes, map->num_stripes);
3133                                 multi->stripes[i].length = map->stripe_len *
3134                                         (stripes - stripe_nr + 1);
3135
3136                                 if (i == 0) {
3137                                         multi->stripes[i].length -=
3138                                                 stripe_offset;
3139                                         stripe_offset = 0;
3140                                 }
3141                                 if (stripe_index == last_stripe)
3142                                         multi->stripes[i].length -=
3143                                                 stripe_end_offset;
3144                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3145                                 u64 stripes;
3146                                 int j;
3147                                 int factor = map->num_stripes /
3148                                              map->sub_stripes;
3149                                 u32 last_stripe = 0;
3150
3151                                 div_u64_rem(stripe_nr_end - 1,
3152                                             factor, &last_stripe);
3153                                 last_stripe *= map->sub_stripes;
3154
3155                                 for (j = 0; j < factor; j++) {
3156                                         u32 test;
3157
3158                                         div_u64_rem(stripe_nr_end - 1 - j,
3159                                                     factor, &test);
3160
3161                                         if (test ==
3162                                             stripe_index / map->sub_stripes)
3163                                                 break;
3164                                 }
3165                                 stripes = stripe_nr_end - 1 - j;
3166                                 do_div(stripes, factor);
3167                                 multi->stripes[i].length = map->stripe_len *
3168                                         (stripes - stripe_nr + 1);
3169
3170                                 if (i < map->sub_stripes) {
3171                                         multi->stripes[i].length -=
3172                                                 stripe_offset;
3173                                         if (i == map->sub_stripes - 1)
3174                                                 stripe_offset = 0;
3175                                 }
3176                                 if (stripe_index >= last_stripe &&
3177                                     stripe_index <= (last_stripe +
3178                                                      map->sub_stripes - 1)) {
3179                                         multi->stripes[i].length -=
3180                                                 stripe_end_offset;
3181                                 }
3182                         } else
3183                                 multi->stripes[i].length = *length;
3184
3185                         stripe_index++;
3186                         if (stripe_index == map->num_stripes) {
3187                                 /* This could only happen for RAID0/10 */
3188                                 stripe_index = 0;
3189                                 stripe_nr++;
3190                         }
3191                 }
3192         } else {
3193                 for (i = 0; i < num_stripes; i++) {
3194                         if (unplug_page) {
3195                                 struct btrfs_device *device;
3196                                 struct backing_dev_info *bdi;
3197
3198                                 device = map->stripes[stripe_index].dev;
3199                                 if (device->bdev) {
3200                                         bdi = blk_get_backing_dev_info(device->
3201                                                                        bdev);
3202                                         if (bdi->unplug_io_fn)
3203                                                 bdi->unplug_io_fn(bdi,
3204                                                                   unplug_page);
3205                                 }
3206                         } else {
3207                                 multi->stripes[i].physical =
3208                                         map->stripes[stripe_index].physical +
3209                                         stripe_offset +
3210                                         stripe_nr * map->stripe_len;
3211                                 multi->stripes[i].dev =
3212                                         map->stripes[stripe_index].dev;
3213                         }
3214                         stripe_index++;
3215                 }
3216         }
3217         if (multi_ret) {
3218                 *multi_ret = multi;
3219                 multi->num_stripes = num_stripes;
3220                 multi->max_errors = max_errors;
3221         }
3222 out:
3223         free_extent_map(em);
3224         return 0;
3225 }
3226
3227 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3228                       u64 logical, u64 *length,
3229                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3230 {
3231         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3232                                  mirror_num, NULL);
3233 }
3234
3235 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3236                      u64 chunk_start, u64 physical, u64 devid,
3237                      u64 **logical, int *naddrs, int *stripe_len)
3238 {
3239         struct extent_map_tree *em_tree = &map_tree->map_tree;
3240         struct extent_map *em;
3241         struct map_lookup *map;
3242         u64 *buf;
3243         u64 bytenr;
3244         u64 length;
3245         u64 stripe_nr;
3246         int i, j, nr = 0;
3247
3248         read_lock(&em_tree->lock);
3249         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3250         read_unlock(&em_tree->lock);
3251
3252         BUG_ON(!em || em->start != chunk_start);
3253         map = (struct map_lookup *)em->bdev;
3254
3255         length = em->len;
3256         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3257                 do_div(length, map->num_stripes / map->sub_stripes);
3258         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3259                 do_div(length, map->num_stripes);
3260
3261         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3262         BUG_ON(!buf);
3263
3264         for (i = 0; i < map->num_stripes; i++) {
3265                 if (devid && map->stripes[i].dev->devid != devid)
3266                         continue;
3267                 if (map->stripes[i].physical > physical ||
3268                     map->stripes[i].physical + length <= physical)
3269                         continue;
3270
3271                 stripe_nr = physical - map->stripes[i].physical;
3272                 do_div(stripe_nr, map->stripe_len);
3273
3274                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3275                         stripe_nr = stripe_nr * map->num_stripes + i;
3276                         do_div(stripe_nr, map->sub_stripes);
3277                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3278                         stripe_nr = stripe_nr * map->num_stripes + i;
3279                 }
3280                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3281                 WARN_ON(nr >= map->num_stripes);
3282                 for (j = 0; j < nr; j++) {
3283                         if (buf[j] == bytenr)
3284                                 break;
3285                 }
3286                 if (j == nr) {
3287                         WARN_ON(nr >= map->num_stripes);
3288                         buf[nr++] = bytenr;
3289                 }
3290         }
3291
3292         *logical = buf;
3293         *naddrs = nr;
3294         *stripe_len = map->stripe_len;
3295
3296         free_extent_map(em);
3297         return 0;
3298 }
3299
3300 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3301                       u64 logical, struct page *page)
3302 {
3303         u64 length = PAGE_CACHE_SIZE;
3304         return __btrfs_map_block(map_tree, READ, logical, &length,
3305                                  NULL, 0, page);
3306 }
3307
3308 static void end_bio_multi_stripe(struct bio *bio, int err)
3309 {
3310         struct btrfs_multi_bio *multi = bio->bi_private;
3311         int is_orig_bio = 0;
3312
3313         if (err)
3314                 atomic_inc(&multi->error);
3315
3316         if (bio == multi->orig_bio)
3317                 is_orig_bio = 1;
3318
3319         if (atomic_dec_and_test(&multi->stripes_pending)) {
3320                 if (!is_orig_bio) {
3321                         bio_put(bio);
3322                         bio = multi->orig_bio;
3323                 }
3324                 bio->bi_private = multi->private;
3325                 bio->bi_end_io = multi->end_io;
3326                 /* only send an error to the higher layers if it is
3327                  * beyond the tolerance of the multi-bio
3328                  */
3329                 if (atomic_read(&multi->error) > multi->max_errors) {
3330                         err = -EIO;
3331                 } else if (err) {
3332                         /*
3333                          * this bio is actually up to date, we didn't
3334                          * go over the max number of errors
3335                          */
3336                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3337                         err = 0;
3338                 }
3339                 kfree(multi);
3340
3341                 bio_endio(bio, err);
3342         } else if (!is_orig_bio) {
3343                 bio_put(bio);
3344         }
3345 }
3346
3347 struct async_sched {
3348         struct bio *bio;
3349         int rw;
3350         struct btrfs_fs_info *info;
3351         struct btrfs_work work;
3352 };
3353
3354 /*
3355  * see run_scheduled_bios for a description of why bios are collected for
3356  * async submit.
3357  *
3358  * This will add one bio to the pending list for a device and make sure
3359  * the work struct is scheduled.
3360  */
3361 static noinline int schedule_bio(struct btrfs_root *root,
3362                                  struct btrfs_device *device,
3363                                  int rw, struct bio *bio)
3364 {
3365         int should_queue = 1;
3366         struct btrfs_pending_bios *pending_bios;
3367
3368         /* don't bother with additional async steps for reads, right now */
3369         if (!(rw & REQ_WRITE)) {
3370                 bio_get(bio);
3371                 submit_bio(rw, bio);
3372                 bio_put(bio);
3373                 return 0;
3374         }
3375
3376         /*
3377          * nr_async_bios allows us to reliably return congestion to the
3378          * higher layers.  Otherwise, the async bio makes it appear we have
3379          * made progress against dirty pages when we've really just put it
3380          * on a queue for later
3381          */
3382         atomic_inc(&root->fs_info->nr_async_bios);
3383         WARN_ON(bio->bi_next);
3384         bio->bi_next = NULL;
3385         bio->bi_rw |= rw;
3386
3387         spin_lock(&device->io_lock);
3388         if (bio->bi_rw & REQ_SYNC)
3389                 pending_bios = &device->pending_sync_bios;
3390         else
3391                 pending_bios = &device->pending_bios;
3392
3393         if (pending_bios->tail)
3394                 pending_bios->tail->bi_next = bio;
3395
3396         pending_bios->tail = bio;
3397         if (!pending_bios->head)
3398                 pending_bios->head = bio;
3399         if (device->running_pending)
3400                 should_queue = 0;
3401
3402         spin_unlock(&device->io_lock);
3403
3404         if (should_queue)
3405                 btrfs_queue_worker(&root->fs_info->submit_workers,
3406                                    &device->work);
3407         return 0;
3408 }
3409
3410 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3411                   int mirror_num, int async_submit)
3412 {
3413         struct btrfs_mapping_tree *map_tree;
3414         struct btrfs_device *dev;
3415         struct bio *first_bio = bio;
3416         u64 logical = (u64)bio->bi_sector << 9;
3417         u64 length = 0;
3418         u64 map_length;
3419         struct btrfs_multi_bio *multi = NULL;
3420         int ret;
3421         int dev_nr = 0;
3422         int total_devs = 1;
3423
3424         length = bio->bi_size;
3425         map_tree = &root->fs_info->mapping_tree;
3426         map_length = length;
3427
3428         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3429                               mirror_num);
3430         BUG_ON(ret);
3431
3432         total_devs = multi->num_stripes;
3433         if (map_length < length) {
3434                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3435                        "len %llu\n", (unsigned long long)logical,
3436                        (unsigned long long)length,
3437                        (unsigned long long)map_length);
3438                 BUG();
3439         }
3440         multi->end_io = first_bio->bi_end_io;
3441         multi->private = first_bio->bi_private;
3442         multi->orig_bio = first_bio;
3443         atomic_set(&multi->stripes_pending, multi->num_stripes);
3444
3445         while (dev_nr < total_devs) {
3446                 if (total_devs > 1) {
3447                         if (dev_nr < total_devs - 1) {
3448                                 bio = bio_clone(first_bio, GFP_NOFS);
3449                                 BUG_ON(!bio);
3450                         } else {
3451                                 bio = first_bio;
3452                         }
3453                         bio->bi_private = multi;
3454                         bio->bi_end_io = end_bio_multi_stripe;
3455                 }
3456                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3457                 dev = multi->stripes[dev_nr].dev;
3458                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3459                         bio->bi_bdev = dev->bdev;
3460                         if (async_submit)
3461                                 schedule_bio(root, dev, rw, bio);
3462                         else
3463                                 submit_bio(rw, bio);
3464                 } else {
3465                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3466                         bio->bi_sector = logical >> 9;
3467                         bio_endio(bio, -EIO);
3468                 }
3469                 dev_nr++;
3470         }
3471         if (total_devs == 1)
3472                 kfree(multi);
3473         return 0;
3474 }
3475
3476 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3477                                        u8 *uuid, u8 *fsid)
3478 {
3479         struct btrfs_device *device;
3480         struct btrfs_fs_devices *cur_devices;
3481
3482         cur_devices = root->fs_info->fs_devices;
3483         while (cur_devices) {
3484                 if (!fsid ||
3485                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3486                         device = __find_device(&cur_devices->devices,
3487                                                devid, uuid);
3488                         if (device)
3489                                 return device;
3490                 }
3491                 cur_devices = cur_devices->seed;
3492         }
3493         return NULL;
3494 }
3495
3496 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3497                                             u64 devid, u8 *dev_uuid)
3498 {
3499         struct btrfs_device *device;
3500         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3501
3502         device = kzalloc(sizeof(*device), GFP_NOFS);
3503         if (!device)
3504                 return NULL;
3505         list_add(&device->dev_list,
3506                  &fs_devices->devices);
3507         device->dev_root = root->fs_info->dev_root;
3508         device->devid = devid;
3509         device->work.func = pending_bios_fn;
3510         device->fs_devices = fs_devices;
3511         device->missing = 1;
3512         fs_devices->num_devices++;
3513         fs_devices->missing_devices++;
3514         spin_lock_init(&device->io_lock);
3515         INIT_LIST_HEAD(&device->dev_alloc_list);
3516         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3517         return device;
3518 }
3519
3520 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3521                           struct extent_buffer *leaf,
3522                           struct btrfs_chunk *chunk)
3523 {
3524         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3525         struct map_lookup *map;
3526         struct extent_map *em;
3527         u64 logical;
3528         u64 length;
3529         u64 devid;
3530         u8 uuid[BTRFS_UUID_SIZE];
3531         int num_stripes;
3532         int ret;
3533         int i;
3534
3535         logical = key->offset;
3536         length = btrfs_chunk_length(leaf, chunk);
3537
3538         read_lock(&map_tree->map_tree.lock);
3539         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3540         read_unlock(&map_tree->map_tree.lock);
3541
3542         /* already mapped? */
3543         if (em && em->start <= logical && em->start + em->len > logical) {
3544                 free_extent_map(em);
3545                 return 0;
3546         } else if (em) {
3547                 free_extent_map(em);
3548         }
3549
3550         em = alloc_extent_map(GFP_NOFS);
3551         if (!em)
3552                 return -ENOMEM;
3553         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3554         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3555         if (!map) {
3556                 free_extent_map(em);
3557                 return -ENOMEM;
3558         }
3559
3560         em->bdev = (struct block_device *)map;
3561         em->start = logical;
3562         em->len = length;
3563         em->block_start = 0;
3564         em->block_len = em->len;
3565
3566         map->num_stripes = num_stripes;
3567         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3568         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3569         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3570         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3571         map->type = btrfs_chunk_type(leaf, chunk);
3572         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3573         for (i = 0; i < num_stripes; i++) {
3574                 map->stripes[i].physical =
3575                         btrfs_stripe_offset_nr(leaf, chunk, i);
3576                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3577                 read_extent_buffer(leaf, uuid, (unsigned long)
3578                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3579                                    BTRFS_UUID_SIZE);
3580                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3581                                                         NULL);
3582                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3583                         kfree(map);
3584                         free_extent_map(em);
3585                         return -EIO;
3586                 }
3587                 if (!map->stripes[i].dev) {
3588                         map->stripes[i].dev =
3589                                 add_missing_dev(root, devid, uuid);
3590                         if (!map->stripes[i].dev) {
3591                                 kfree(map);
3592                                 free_extent_map(em);
3593                                 return -EIO;
3594                         }
3595                 }
3596                 map->stripes[i].dev->in_fs_metadata = 1;
3597         }
3598
3599         write_lock(&map_tree->map_tree.lock);
3600         ret = add_extent_mapping(&map_tree->map_tree, em);
3601         write_unlock(&map_tree->map_tree.lock);
3602         BUG_ON(ret);
3603         free_extent_map(em);
3604
3605         return 0;
3606 }
3607
3608 static int fill_device_from_item(struct extent_buffer *leaf,
3609                                  struct btrfs_dev_item *dev_item,
3610                                  struct btrfs_device *device)
3611 {
3612         unsigned long ptr;
3613
3614         device->devid = btrfs_device_id(leaf, dev_item);
3615         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3616         device->total_bytes = device->disk_total_bytes;
3617         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3618         device->type = btrfs_device_type(leaf, dev_item);
3619         device->io_align = btrfs_device_io_align(leaf, dev_item);
3620         device->io_width = btrfs_device_io_width(leaf, dev_item);
3621         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3622
3623         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3624         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3625
3626         return 0;
3627 }
3628
3629 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3630 {
3631         struct btrfs_fs_devices *fs_devices;
3632         int ret;
3633
3634         mutex_lock(&uuid_mutex);
3635
3636         fs_devices = root->fs_info->fs_devices->seed;
3637         while (fs_devices) {
3638                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3639                         ret = 0;
3640                         goto out;
3641                 }
3642                 fs_devices = fs_devices->seed;
3643         }
3644
3645         fs_devices = find_fsid(fsid);
3646         if (!fs_devices) {
3647                 ret = -ENOENT;
3648                 goto out;
3649         }
3650
3651         fs_devices = clone_fs_devices(fs_devices);
3652         if (IS_ERR(fs_devices)) {
3653                 ret = PTR_ERR(fs_devices);
3654                 goto out;
3655         }
3656
3657         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3658                                    root->fs_info->bdev_holder);
3659         if (ret)
3660                 goto out;
3661
3662         if (!fs_devices->seeding) {
3663                 __btrfs_close_devices(fs_devices);
3664                 free_fs_devices(fs_devices);
3665                 ret = -EINVAL;
3666                 goto out;
3667         }
3668
3669         fs_devices->seed = root->fs_info->fs_devices->seed;
3670         root->fs_info->fs_devices->seed = fs_devices;
3671 out:
3672         mutex_unlock(&uuid_mutex);
3673         return ret;
3674 }
3675
3676 static int read_one_dev(struct btrfs_root *root,
3677                         struct extent_buffer *leaf,
3678                         struct btrfs_dev_item *dev_item)
3679 {
3680         struct btrfs_device *device;
3681         u64 devid;
3682         int ret;
3683         u8 fs_uuid[BTRFS_UUID_SIZE];
3684         u8 dev_uuid[BTRFS_UUID_SIZE];
3685
3686         devid = btrfs_device_id(leaf, dev_item);
3687         read_extent_buffer(leaf, dev_uuid,
3688                            (unsigned long)btrfs_device_uuid(dev_item),
3689                            BTRFS_UUID_SIZE);
3690         read_extent_buffer(leaf, fs_uuid,
3691                            (unsigned long)btrfs_device_fsid(dev_item),
3692                            BTRFS_UUID_SIZE);
3693
3694         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3695                 ret = open_seed_devices(root, fs_uuid);
3696                 if (ret && !btrfs_test_opt(root, DEGRADED))
3697                         return ret;
3698         }
3699
3700         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3701         if (!device || !device->bdev) {
3702                 if (!btrfs_test_opt(root, DEGRADED))
3703                         return -EIO;
3704
3705                 if (!device) {
3706                         printk(KERN_WARNING "warning devid %llu missing\n",
3707                                (unsigned long long)devid);
3708                         device = add_missing_dev(root, devid, dev_uuid);
3709                         if (!device)
3710                                 return -ENOMEM;
3711                 } else if (!device->missing) {
3712                         /*
3713                          * this happens when a device that was properly setup
3714                          * in the device info lists suddenly goes bad.
3715                          * device->bdev is NULL, and so we have to set
3716                          * device->missing to one here
3717                          */
3718                         root->fs_info->fs_devices->missing_devices++;
3719                         device->missing = 1;
3720                 }
3721         }
3722
3723         if (device->fs_devices != root->fs_info->fs_devices) {
3724                 BUG_ON(device->writeable);
3725                 if (device->generation !=
3726                     btrfs_device_generation(leaf, dev_item))
3727                         return -EINVAL;
3728         }
3729
3730         fill_device_from_item(leaf, dev_item, device);
3731         device->dev_root = root->fs_info->dev_root;
3732         device->in_fs_metadata = 1;
3733         if (device->writeable)
3734                 device->fs_devices->total_rw_bytes += device->total_bytes;
3735         ret = 0;
3736         return ret;
3737 }
3738
3739 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3740 {
3741         struct btrfs_dev_item *dev_item;
3742
3743         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3744                                                      dev_item);
3745         return read_one_dev(root, buf, dev_item);
3746 }
3747
3748 int btrfs_read_sys_array(struct btrfs_root *root)
3749 {
3750         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3751         struct extent_buffer *sb;
3752         struct btrfs_disk_key *disk_key;
3753         struct btrfs_chunk *chunk;
3754         u8 *ptr;
3755         unsigned long sb_ptr;
3756         int ret = 0;
3757         u32 num_stripes;
3758         u32 array_size;
3759         u32 len = 0;
3760         u32 cur;
3761         struct btrfs_key key;
3762
3763         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3764                                           BTRFS_SUPER_INFO_SIZE);
3765         if (!sb)
3766                 return -ENOMEM;
3767         btrfs_set_buffer_uptodate(sb);
3768         btrfs_set_buffer_lockdep_class(sb, 0);
3769
3770         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3771         array_size = btrfs_super_sys_array_size(super_copy);
3772
3773         ptr = super_copy->sys_chunk_array;
3774         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3775         cur = 0;
3776
3777         while (cur < array_size) {
3778                 disk_key = (struct btrfs_disk_key *)ptr;
3779                 btrfs_disk_key_to_cpu(&key, disk_key);
3780
3781                 len = sizeof(*disk_key); ptr += len;
3782                 sb_ptr += len;
3783                 cur += len;
3784
3785                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3786                         chunk = (struct btrfs_chunk *)sb_ptr;
3787                         ret = read_one_chunk(root, &key, sb, chunk);
3788                         if (ret)
3789                                 break;
3790                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3791                         len = btrfs_chunk_item_size(num_stripes);
3792                 } else {
3793                         ret = -EIO;
3794                         break;
3795                 }
3796                 ptr += len;
3797                 sb_ptr += len;
3798                 cur += len;
3799         }
3800         free_extent_buffer(sb);
3801         return ret;
3802 }
3803
3804 int btrfs_read_chunk_tree(struct btrfs_root *root)
3805 {
3806         struct btrfs_path *path;
3807         struct extent_buffer *leaf;
3808         struct btrfs_key key;
3809         struct btrfs_key found_key;
3810         int ret;
3811         int slot;
3812
3813         root = root->fs_info->chunk_root;
3814
3815         path = btrfs_alloc_path();
3816         if (!path)
3817                 return -ENOMEM;
3818
3819         /* first we search for all of the device items, and then we
3820          * read in all of the chunk items.  This way we can create chunk
3821          * mappings that reference all of the devices that are afound
3822          */
3823         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3824         key.offset = 0;
3825         key.type = 0;
3826 again:
3827         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3828         if (ret < 0)
3829                 goto error;
3830         while (1) {
3831                 leaf = path->nodes[0];
3832                 slot = path->slots[0];
3833                 if (slot >= btrfs_header_nritems(leaf)) {
3834                         ret = btrfs_next_leaf(root, path);
3835                         if (ret == 0)
3836                                 continue;
3837                         if (ret < 0)
3838                                 goto error;
3839                         break;
3840                 }
3841                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3842                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3843                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3844                                 break;
3845                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3846                                 struct btrfs_dev_item *dev_item;
3847                                 dev_item = btrfs_item_ptr(leaf, slot,
3848                                                   struct btrfs_dev_item);
3849                                 ret = read_one_dev(root, leaf, dev_item);
3850                                 if (ret)
3851                                         goto error;
3852                         }
3853                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3854                         struct btrfs_chunk *chunk;
3855                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3856                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3857                         if (ret)
3858                                 goto error;
3859                 }
3860                 path->slots[0]++;
3861         }
3862         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3863                 key.objectid = 0;
3864                 btrfs_release_path(root, path);
3865                 goto again;
3866         }
3867         ret = 0;
3868 error:
3869         btrfs_free_path(path);
3870         return ret;
3871 }