]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/volumes.c
6b9884507837581ba6231bc6c4cfb85a036b82f6
[~shefty/rdma-dev.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <asm/div64.h>
26 #include "compat.h"
27 #include "ctree.h"
28 #include "extent_map.h"
29 #include "disk-io.h"
30 #include "transaction.h"
31 #include "print-tree.h"
32 #include "volumes.h"
33 #include "async-thread.h"
34
35 struct map_lookup {
36         u64 type;
37         int io_align;
38         int io_width;
39         int stripe_len;
40         int sector_size;
41         int num_stripes;
42         int sub_stripes;
43         struct btrfs_bio_stripe stripes[];
44 };
45
46 static int init_first_rw_device(struct btrfs_trans_handle *trans,
47                                 struct btrfs_root *root,
48                                 struct btrfs_device *device);
49 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50
51 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
52                             (sizeof(struct btrfs_bio_stripe) * (n)))
53
54 static DEFINE_MUTEX(uuid_mutex);
55 static LIST_HEAD(fs_uuids);
56
57 void btrfs_lock_volumes(void)
58 {
59         mutex_lock(&uuid_mutex);
60 }
61
62 void btrfs_unlock_volumes(void)
63 {
64         mutex_unlock(&uuid_mutex);
65 }
66
67 static void lock_chunks(struct btrfs_root *root)
68 {
69         mutex_lock(&root->fs_info->chunk_mutex);
70 }
71
72 static void unlock_chunks(struct btrfs_root *root)
73 {
74         mutex_unlock(&root->fs_info->chunk_mutex);
75 }
76
77 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78 {
79         struct btrfs_device *device;
80         WARN_ON(fs_devices->opened);
81         while (!list_empty(&fs_devices->devices)) {
82                 device = list_entry(fs_devices->devices.next,
83                                     struct btrfs_device, dev_list);
84                 list_del(&device->dev_list);
85                 kfree(device->name);
86                 kfree(device);
87         }
88         kfree(fs_devices);
89 }
90
91 int btrfs_cleanup_fs_uuids(void)
92 {
93         struct btrfs_fs_devices *fs_devices;
94
95         while (!list_empty(&fs_uuids)) {
96                 fs_devices = list_entry(fs_uuids.next,
97                                         struct btrfs_fs_devices, list);
98                 list_del(&fs_devices->list);
99                 free_fs_devices(fs_devices);
100         }
101         return 0;
102 }
103
104 static noinline struct btrfs_device *__find_device(struct list_head *head,
105                                                    u64 devid, u8 *uuid)
106 {
107         struct btrfs_device *dev;
108
109         list_for_each_entry(dev, head, dev_list) {
110                 if (dev->devid == devid &&
111                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
112                         return dev;
113                 }
114         }
115         return NULL;
116 }
117
118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 {
120         struct btrfs_fs_devices *fs_devices;
121
122         list_for_each_entry(fs_devices, &fs_uuids, list) {
123                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
124                         return fs_devices;
125         }
126         return NULL;
127 }
128
129 static void requeue_list(struct btrfs_pending_bios *pending_bios,
130                         struct bio *head, struct bio *tail)
131 {
132
133         struct bio *old_head;
134
135         old_head = pending_bios->head;
136         pending_bios->head = head;
137         if (pending_bios->tail)
138                 tail->bi_next = old_head;
139         else
140                 pending_bios->tail = tail;
141 }
142
143 /*
144  * we try to collect pending bios for a device so we don't get a large
145  * number of procs sending bios down to the same device.  This greatly
146  * improves the schedulers ability to collect and merge the bios.
147  *
148  * But, it also turns into a long list of bios to process and that is sure
149  * to eventually make the worker thread block.  The solution here is to
150  * make some progress and then put this work struct back at the end of
151  * the list if the block device is congested.  This way, multiple devices
152  * can make progress from a single worker thread.
153  */
154 static noinline int run_scheduled_bios(struct btrfs_device *device)
155 {
156         struct bio *pending;
157         struct backing_dev_info *bdi;
158         struct btrfs_fs_info *fs_info;
159         struct btrfs_pending_bios *pending_bios;
160         struct bio *tail;
161         struct bio *cur;
162         int again = 0;
163         unsigned long num_run;
164         unsigned long num_sync_run;
165         unsigned long batch_run = 0;
166         unsigned long limit;
167         unsigned long last_waited = 0;
168         int force_reg = 0;
169
170         bdi = blk_get_backing_dev_info(device->bdev);
171         fs_info = device->dev_root->fs_info;
172         limit = btrfs_async_submit_limit(fs_info);
173         limit = limit * 2 / 3;
174
175         /* we want to make sure that every time we switch from the sync
176          * list to the normal list, we unplug
177          */
178         num_sync_run = 0;
179
180 loop:
181         spin_lock(&device->io_lock);
182
183 loop_lock:
184         num_run = 0;
185
186         /* take all the bios off the list at once and process them
187          * later on (without the lock held).  But, remember the
188          * tail and other pointers so the bios can be properly reinserted
189          * into the list if we hit congestion
190          */
191         if (!force_reg && device->pending_sync_bios.head) {
192                 pending_bios = &device->pending_sync_bios;
193                 force_reg = 1;
194         } else {
195                 pending_bios = &device->pending_bios;
196                 force_reg = 0;
197         }
198
199         pending = pending_bios->head;
200         tail = pending_bios->tail;
201         WARN_ON(pending && !tail);
202
203         /*
204          * if pending was null this time around, no bios need processing
205          * at all and we can stop.  Otherwise it'll loop back up again
206          * and do an additional check so no bios are missed.
207          *
208          * device->running_pending is used to synchronize with the
209          * schedule_bio code.
210          */
211         if (device->pending_sync_bios.head == NULL &&
212             device->pending_bios.head == NULL) {
213                 again = 0;
214                 device->running_pending = 0;
215         } else {
216                 again = 1;
217                 device->running_pending = 1;
218         }
219
220         pending_bios->head = NULL;
221         pending_bios->tail = NULL;
222
223         spin_unlock(&device->io_lock);
224
225         /*
226          * if we're doing the regular priority list, make sure we unplug
227          * for any high prio bios we've sent down
228          */
229         if (pending_bios == &device->pending_bios && num_sync_run > 0) {
230                 num_sync_run = 0;
231                 blk_run_backing_dev(bdi, NULL);
232         }
233
234         while (pending) {
235
236                 rmb();
237                 /* we want to work on both lists, but do more bios on the
238                  * sync list than the regular list
239                  */
240                 if ((num_run > 32 &&
241                     pending_bios != &device->pending_sync_bios &&
242                     device->pending_sync_bios.head) ||
243                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
244                     device->pending_bios.head)) {
245                         spin_lock(&device->io_lock);
246                         requeue_list(pending_bios, pending, tail);
247                         goto loop_lock;
248                 }
249
250                 cur = pending;
251                 pending = pending->bi_next;
252                 cur->bi_next = NULL;
253                 atomic_dec(&fs_info->nr_async_bios);
254
255                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
256                     waitqueue_active(&fs_info->async_submit_wait))
257                         wake_up(&fs_info->async_submit_wait);
258
259                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
260
261                 if (cur->bi_rw & REQ_SYNC)
262                         num_sync_run++;
263
264                 submit_bio(cur->bi_rw, cur);
265                 num_run++;
266                 batch_run++;
267                 if (need_resched()) {
268                         if (num_sync_run) {
269                                 blk_run_backing_dev(bdi, NULL);
270                                 num_sync_run = 0;
271                         }
272                         cond_resched();
273                 }
274
275                 /*
276                  * we made progress, there is more work to do and the bdi
277                  * is now congested.  Back off and let other work structs
278                  * run instead
279                  */
280                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
281                     fs_info->fs_devices->open_devices > 1) {
282                         struct io_context *ioc;
283
284                         ioc = current->io_context;
285
286                         /*
287                          * the main goal here is that we don't want to
288                          * block if we're going to be able to submit
289                          * more requests without blocking.
290                          *
291                          * This code does two great things, it pokes into
292                          * the elevator code from a filesystem _and_
293                          * it makes assumptions about how batching works.
294                          */
295                         if (ioc && ioc->nr_batch_requests > 0 &&
296                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
297                             (last_waited == 0 ||
298                              ioc->last_waited == last_waited)) {
299                                 /*
300                                  * we want to go through our batch of
301                                  * requests and stop.  So, we copy out
302                                  * the ioc->last_waited time and test
303                                  * against it before looping
304                                  */
305                                 last_waited = ioc->last_waited;
306                                 if (need_resched()) {
307                                         if (num_sync_run) {
308                                                 blk_run_backing_dev(bdi, NULL);
309                                                 num_sync_run = 0;
310                                         }
311                                         cond_resched();
312                                 }
313                                 continue;
314                         }
315                         spin_lock(&device->io_lock);
316                         requeue_list(pending_bios, pending, tail);
317                         device->running_pending = 1;
318
319                         spin_unlock(&device->io_lock);
320                         btrfs_requeue_work(&device->work);
321                         goto done;
322                 }
323         }
324
325         if (num_sync_run) {
326                 num_sync_run = 0;
327                 blk_run_backing_dev(bdi, NULL);
328         }
329         /*
330          * IO has already been through a long path to get here.  Checksumming,
331          * async helper threads, perhaps compression.  We've done a pretty
332          * good job of collecting a batch of IO and should just unplug
333          * the device right away.
334          *
335          * This will help anyone who is waiting on the IO, they might have
336          * already unplugged, but managed to do so before the bio they
337          * cared about found its way down here.
338          */
339         blk_run_backing_dev(bdi, NULL);
340
341         cond_resched();
342         if (again)
343                 goto loop;
344
345         spin_lock(&device->io_lock);
346         if (device->pending_bios.head || device->pending_sync_bios.head)
347                 goto loop_lock;
348         spin_unlock(&device->io_lock);
349
350 done:
351         return 0;
352 }
353
354 static void pending_bios_fn(struct btrfs_work *work)
355 {
356         struct btrfs_device *device;
357
358         device = container_of(work, struct btrfs_device, work);
359         run_scheduled_bios(device);
360 }
361
362 static noinline int device_list_add(const char *path,
363                            struct btrfs_super_block *disk_super,
364                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
365 {
366         struct btrfs_device *device;
367         struct btrfs_fs_devices *fs_devices;
368         u64 found_transid = btrfs_super_generation(disk_super);
369         char *name;
370
371         fs_devices = find_fsid(disk_super->fsid);
372         if (!fs_devices) {
373                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
374                 if (!fs_devices)
375                         return -ENOMEM;
376                 INIT_LIST_HEAD(&fs_devices->devices);
377                 INIT_LIST_HEAD(&fs_devices->alloc_list);
378                 list_add(&fs_devices->list, &fs_uuids);
379                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
380                 fs_devices->latest_devid = devid;
381                 fs_devices->latest_trans = found_transid;
382                 mutex_init(&fs_devices->device_list_mutex);
383                 device = NULL;
384         } else {
385                 device = __find_device(&fs_devices->devices, devid,
386                                        disk_super->dev_item.uuid);
387         }
388         if (!device) {
389                 if (fs_devices->opened)
390                         return -EBUSY;
391
392                 device = kzalloc(sizeof(*device), GFP_NOFS);
393                 if (!device) {
394                         /* we can safely leave the fs_devices entry around */
395                         return -ENOMEM;
396                 }
397                 device->devid = devid;
398                 device->work.func = pending_bios_fn;
399                 memcpy(device->uuid, disk_super->dev_item.uuid,
400                        BTRFS_UUID_SIZE);
401                 spin_lock_init(&device->io_lock);
402                 device->name = kstrdup(path, GFP_NOFS);
403                 if (!device->name) {
404                         kfree(device);
405                         return -ENOMEM;
406                 }
407                 INIT_LIST_HEAD(&device->dev_alloc_list);
408
409                 mutex_lock(&fs_devices->device_list_mutex);
410                 list_add(&device->dev_list, &fs_devices->devices);
411                 mutex_unlock(&fs_devices->device_list_mutex);
412
413                 device->fs_devices = fs_devices;
414                 fs_devices->num_devices++;
415         } else if (!device->name || strcmp(device->name, path)) {
416                 name = kstrdup(path, GFP_NOFS);
417                 if (!name)
418                         return -ENOMEM;
419                 kfree(device->name);
420                 device->name = name;
421                 if (device->missing) {
422                         fs_devices->missing_devices--;
423                         device->missing = 0;
424                 }
425         }
426
427         if (found_transid > fs_devices->latest_trans) {
428                 fs_devices->latest_devid = devid;
429                 fs_devices->latest_trans = found_transid;
430         }
431         *fs_devices_ret = fs_devices;
432         return 0;
433 }
434
435 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
436 {
437         struct btrfs_fs_devices *fs_devices;
438         struct btrfs_device *device;
439         struct btrfs_device *orig_dev;
440
441         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
442         if (!fs_devices)
443                 return ERR_PTR(-ENOMEM);
444
445         INIT_LIST_HEAD(&fs_devices->devices);
446         INIT_LIST_HEAD(&fs_devices->alloc_list);
447         INIT_LIST_HEAD(&fs_devices->list);
448         mutex_init(&fs_devices->device_list_mutex);
449         fs_devices->latest_devid = orig->latest_devid;
450         fs_devices->latest_trans = orig->latest_trans;
451         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
452
453         mutex_lock(&orig->device_list_mutex);
454         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
455                 device = kzalloc(sizeof(*device), GFP_NOFS);
456                 if (!device)
457                         goto error;
458
459                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
460                 if (!device->name) {
461                         kfree(device);
462                         goto error;
463                 }
464
465                 device->devid = orig_dev->devid;
466                 device->work.func = pending_bios_fn;
467                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
468                 spin_lock_init(&device->io_lock);
469                 INIT_LIST_HEAD(&device->dev_list);
470                 INIT_LIST_HEAD(&device->dev_alloc_list);
471
472                 list_add(&device->dev_list, &fs_devices->devices);
473                 device->fs_devices = fs_devices;
474                 fs_devices->num_devices++;
475         }
476         mutex_unlock(&orig->device_list_mutex);
477         return fs_devices;
478 error:
479         mutex_unlock(&orig->device_list_mutex);
480         free_fs_devices(fs_devices);
481         return ERR_PTR(-ENOMEM);
482 }
483
484 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
485 {
486         struct btrfs_device *device, *next;
487
488         mutex_lock(&uuid_mutex);
489 again:
490         mutex_lock(&fs_devices->device_list_mutex);
491         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
492                 if (device->in_fs_metadata)
493                         continue;
494
495                 if (device->bdev) {
496                         close_bdev_exclusive(device->bdev, device->mode);
497                         device->bdev = NULL;
498                         fs_devices->open_devices--;
499                 }
500                 if (device->writeable) {
501                         list_del_init(&device->dev_alloc_list);
502                         device->writeable = 0;
503                         fs_devices->rw_devices--;
504                 }
505                 list_del_init(&device->dev_list);
506                 fs_devices->num_devices--;
507                 kfree(device->name);
508                 kfree(device);
509         }
510         mutex_unlock(&fs_devices->device_list_mutex);
511
512         if (fs_devices->seed) {
513                 fs_devices = fs_devices->seed;
514                 goto again;
515         }
516
517         mutex_unlock(&uuid_mutex);
518         return 0;
519 }
520
521 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
522 {
523         struct btrfs_device *device;
524
525         if (--fs_devices->opened > 0)
526                 return 0;
527
528         list_for_each_entry(device, &fs_devices->devices, dev_list) {
529                 if (device->bdev) {
530                         close_bdev_exclusive(device->bdev, device->mode);
531                         fs_devices->open_devices--;
532                 }
533                 if (device->writeable) {
534                         list_del_init(&device->dev_alloc_list);
535                         fs_devices->rw_devices--;
536                 }
537
538                 device->bdev = NULL;
539                 device->writeable = 0;
540                 device->in_fs_metadata = 0;
541         }
542         WARN_ON(fs_devices->open_devices);
543         WARN_ON(fs_devices->rw_devices);
544         fs_devices->opened = 0;
545         fs_devices->seeding = 0;
546
547         return 0;
548 }
549
550 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
551 {
552         struct btrfs_fs_devices *seed_devices = NULL;
553         int ret;
554
555         mutex_lock(&uuid_mutex);
556         ret = __btrfs_close_devices(fs_devices);
557         if (!fs_devices->opened) {
558                 seed_devices = fs_devices->seed;
559                 fs_devices->seed = NULL;
560         }
561         mutex_unlock(&uuid_mutex);
562
563         while (seed_devices) {
564                 fs_devices = seed_devices;
565                 seed_devices = fs_devices->seed;
566                 __btrfs_close_devices(fs_devices);
567                 free_fs_devices(fs_devices);
568         }
569         return ret;
570 }
571
572 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
573                                 fmode_t flags, void *holder)
574 {
575         struct block_device *bdev;
576         struct list_head *head = &fs_devices->devices;
577         struct btrfs_device *device;
578         struct block_device *latest_bdev = NULL;
579         struct buffer_head *bh;
580         struct btrfs_super_block *disk_super;
581         u64 latest_devid = 0;
582         u64 latest_transid = 0;
583         u64 devid;
584         int seeding = 1;
585         int ret = 0;
586
587         list_for_each_entry(device, head, dev_list) {
588                 if (device->bdev)
589                         continue;
590                 if (!device->name)
591                         continue;
592
593                 bdev = open_bdev_exclusive(device->name, flags, holder);
594                 if (IS_ERR(bdev)) {
595                         printk(KERN_INFO "open %s failed\n", device->name);
596                         goto error;
597                 }
598                 set_blocksize(bdev, 4096);
599
600                 bh = btrfs_read_dev_super(bdev);
601                 if (!bh)
602                         goto error_close;
603
604                 disk_super = (struct btrfs_super_block *)bh->b_data;
605                 devid = btrfs_stack_device_id(&disk_super->dev_item);
606                 if (devid != device->devid)
607                         goto error_brelse;
608
609                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
610                            BTRFS_UUID_SIZE))
611                         goto error_brelse;
612
613                 device->generation = btrfs_super_generation(disk_super);
614                 if (!latest_transid || device->generation > latest_transid) {
615                         latest_devid = devid;
616                         latest_transid = device->generation;
617                         latest_bdev = bdev;
618                 }
619
620                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
621                         device->writeable = 0;
622                 } else {
623                         device->writeable = !bdev_read_only(bdev);
624                         seeding = 0;
625                 }
626
627                 device->bdev = bdev;
628                 device->in_fs_metadata = 0;
629                 device->mode = flags;
630
631                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
632                         fs_devices->rotating = 1;
633
634                 fs_devices->open_devices++;
635                 if (device->writeable) {
636                         fs_devices->rw_devices++;
637                         list_add(&device->dev_alloc_list,
638                                  &fs_devices->alloc_list);
639                 }
640                 continue;
641
642 error_brelse:
643                 brelse(bh);
644 error_close:
645                 close_bdev_exclusive(bdev, FMODE_READ);
646 error:
647                 continue;
648         }
649         if (fs_devices->open_devices == 0) {
650                 ret = -EIO;
651                 goto out;
652         }
653         fs_devices->seeding = seeding;
654         fs_devices->opened = 1;
655         fs_devices->latest_bdev = latest_bdev;
656         fs_devices->latest_devid = latest_devid;
657         fs_devices->latest_trans = latest_transid;
658         fs_devices->total_rw_bytes = 0;
659 out:
660         return ret;
661 }
662
663 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
664                        fmode_t flags, void *holder)
665 {
666         int ret;
667
668         mutex_lock(&uuid_mutex);
669         if (fs_devices->opened) {
670                 fs_devices->opened++;
671                 ret = 0;
672         } else {
673                 ret = __btrfs_open_devices(fs_devices, flags, holder);
674         }
675         mutex_unlock(&uuid_mutex);
676         return ret;
677 }
678
679 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
680                           struct btrfs_fs_devices **fs_devices_ret)
681 {
682         struct btrfs_super_block *disk_super;
683         struct block_device *bdev;
684         struct buffer_head *bh;
685         int ret;
686         u64 devid;
687         u64 transid;
688
689         mutex_lock(&uuid_mutex);
690
691         bdev = open_bdev_exclusive(path, flags, holder);
692
693         if (IS_ERR(bdev)) {
694                 ret = PTR_ERR(bdev);
695                 goto error;
696         }
697
698         ret = set_blocksize(bdev, 4096);
699         if (ret)
700                 goto error_close;
701         bh = btrfs_read_dev_super(bdev);
702         if (!bh) {
703                 ret = -EIO;
704                 goto error_close;
705         }
706         disk_super = (struct btrfs_super_block *)bh->b_data;
707         devid = btrfs_stack_device_id(&disk_super->dev_item);
708         transid = btrfs_super_generation(disk_super);
709         if (disk_super->label[0])
710                 printk(KERN_INFO "device label %s ", disk_super->label);
711         else {
712                 /* FIXME, make a readl uuid parser */
713                 printk(KERN_INFO "device fsid %llx-%llx ",
714                        *(unsigned long long *)disk_super->fsid,
715                        *(unsigned long long *)(disk_super->fsid + 8));
716         }
717         printk(KERN_CONT "devid %llu transid %llu %s\n",
718                (unsigned long long)devid, (unsigned long long)transid, path);
719         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
720
721         brelse(bh);
722 error_close:
723         close_bdev_exclusive(bdev, flags);
724 error:
725         mutex_unlock(&uuid_mutex);
726         return ret;
727 }
728
729 /*
730  * this uses a pretty simple search, the expectation is that it is
731  * called very infrequently and that a given device has a small number
732  * of extents
733  */
734 int find_free_dev_extent(struct btrfs_trans_handle *trans,
735                          struct btrfs_device *device, u64 num_bytes,
736                          u64 *start, u64 *max_avail)
737 {
738         struct btrfs_key key;
739         struct btrfs_root *root = device->dev_root;
740         struct btrfs_dev_extent *dev_extent = NULL;
741         struct btrfs_path *path;
742         u64 hole_size = 0;
743         u64 last_byte = 0;
744         u64 search_start = 0;
745         u64 search_end = device->total_bytes;
746         int ret;
747         int slot = 0;
748         int start_found;
749         struct extent_buffer *l;
750
751         path = btrfs_alloc_path();
752         if (!path)
753                 return -ENOMEM;
754         path->reada = 2;
755         start_found = 0;
756
757         /* FIXME use last free of some kind */
758
759         /* we don't want to overwrite the superblock on the drive,
760          * so we make sure to start at an offset of at least 1MB
761          */
762         search_start = max((u64)1024 * 1024, search_start);
763
764         if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
765                 search_start = max(root->fs_info->alloc_start, search_start);
766
767         key.objectid = device->devid;
768         key.offset = search_start;
769         key.type = BTRFS_DEV_EXTENT_KEY;
770         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
771         if (ret < 0)
772                 goto error;
773         if (ret > 0) {
774                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
775                 if (ret < 0)
776                         goto error;
777                 if (ret > 0)
778                         start_found = 1;
779         }
780         l = path->nodes[0];
781         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
782         while (1) {
783                 l = path->nodes[0];
784                 slot = path->slots[0];
785                 if (slot >= btrfs_header_nritems(l)) {
786                         ret = btrfs_next_leaf(root, path);
787                         if (ret == 0)
788                                 continue;
789                         if (ret < 0)
790                                 goto error;
791 no_more_items:
792                         if (!start_found) {
793                                 if (search_start >= search_end) {
794                                         ret = -ENOSPC;
795                                         goto error;
796                                 }
797                                 *start = search_start;
798                                 start_found = 1;
799                                 goto check_pending;
800                         }
801                         *start = last_byte > search_start ?
802                                 last_byte : search_start;
803                         if (search_end <= *start) {
804                                 ret = -ENOSPC;
805                                 goto error;
806                         }
807                         goto check_pending;
808                 }
809                 btrfs_item_key_to_cpu(l, &key, slot);
810
811                 if (key.objectid < device->devid)
812                         goto next;
813
814                 if (key.objectid > device->devid)
815                         goto no_more_items;
816
817                 if (key.offset >= search_start && key.offset > last_byte &&
818                     start_found) {
819                         if (last_byte < search_start)
820                                 last_byte = search_start;
821                         hole_size = key.offset - last_byte;
822
823                         if (hole_size > *max_avail)
824                                 *max_avail = hole_size;
825
826                         if (key.offset > last_byte &&
827                             hole_size >= num_bytes) {
828                                 *start = last_byte;
829                                 goto check_pending;
830                         }
831                 }
832                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
833                         goto next;
834
835                 start_found = 1;
836                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
837                 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
838 next:
839                 path->slots[0]++;
840                 cond_resched();
841         }
842 check_pending:
843         /* we have to make sure we didn't find an extent that has already
844          * been allocated by the map tree or the original allocation
845          */
846         BUG_ON(*start < search_start);
847
848         if (*start + num_bytes > search_end) {
849                 ret = -ENOSPC;
850                 goto error;
851         }
852         /* check for pending inserts here */
853         ret = 0;
854
855 error:
856         btrfs_free_path(path);
857         return ret;
858 }
859
860 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
861                           struct btrfs_device *device,
862                           u64 start)
863 {
864         int ret;
865         struct btrfs_path *path;
866         struct btrfs_root *root = device->dev_root;
867         struct btrfs_key key;
868         struct btrfs_key found_key;
869         struct extent_buffer *leaf = NULL;
870         struct btrfs_dev_extent *extent = NULL;
871
872         path = btrfs_alloc_path();
873         if (!path)
874                 return -ENOMEM;
875
876         key.objectid = device->devid;
877         key.offset = start;
878         key.type = BTRFS_DEV_EXTENT_KEY;
879
880         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
881         if (ret > 0) {
882                 ret = btrfs_previous_item(root, path, key.objectid,
883                                           BTRFS_DEV_EXTENT_KEY);
884                 BUG_ON(ret);
885                 leaf = path->nodes[0];
886                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
887                 extent = btrfs_item_ptr(leaf, path->slots[0],
888                                         struct btrfs_dev_extent);
889                 BUG_ON(found_key.offset > start || found_key.offset +
890                        btrfs_dev_extent_length(leaf, extent) < start);
891                 ret = 0;
892         } else if (ret == 0) {
893                 leaf = path->nodes[0];
894                 extent = btrfs_item_ptr(leaf, path->slots[0],
895                                         struct btrfs_dev_extent);
896         }
897         BUG_ON(ret);
898
899         if (device->bytes_used > 0)
900                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
901         ret = btrfs_del_item(trans, root, path);
902         BUG_ON(ret);
903
904         btrfs_free_path(path);
905         return ret;
906 }
907
908 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
909                            struct btrfs_device *device,
910                            u64 chunk_tree, u64 chunk_objectid,
911                            u64 chunk_offset, u64 start, u64 num_bytes)
912 {
913         int ret;
914         struct btrfs_path *path;
915         struct btrfs_root *root = device->dev_root;
916         struct btrfs_dev_extent *extent;
917         struct extent_buffer *leaf;
918         struct btrfs_key key;
919
920         WARN_ON(!device->in_fs_metadata);
921         path = btrfs_alloc_path();
922         if (!path)
923                 return -ENOMEM;
924
925         key.objectid = device->devid;
926         key.offset = start;
927         key.type = BTRFS_DEV_EXTENT_KEY;
928         ret = btrfs_insert_empty_item(trans, root, path, &key,
929                                       sizeof(*extent));
930         BUG_ON(ret);
931
932         leaf = path->nodes[0];
933         extent = btrfs_item_ptr(leaf, path->slots[0],
934                                 struct btrfs_dev_extent);
935         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
936         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
937         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
938
939         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
940                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
941                     BTRFS_UUID_SIZE);
942
943         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
944         btrfs_mark_buffer_dirty(leaf);
945         btrfs_free_path(path);
946         return ret;
947 }
948
949 static noinline int find_next_chunk(struct btrfs_root *root,
950                                     u64 objectid, u64 *offset)
951 {
952         struct btrfs_path *path;
953         int ret;
954         struct btrfs_key key;
955         struct btrfs_chunk *chunk;
956         struct btrfs_key found_key;
957
958         path = btrfs_alloc_path();
959         BUG_ON(!path);
960
961         key.objectid = objectid;
962         key.offset = (u64)-1;
963         key.type = BTRFS_CHUNK_ITEM_KEY;
964
965         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
966         if (ret < 0)
967                 goto error;
968
969         BUG_ON(ret == 0);
970
971         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
972         if (ret) {
973                 *offset = 0;
974         } else {
975                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
976                                       path->slots[0]);
977                 if (found_key.objectid != objectid)
978                         *offset = 0;
979                 else {
980                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
981                                                struct btrfs_chunk);
982                         *offset = found_key.offset +
983                                 btrfs_chunk_length(path->nodes[0], chunk);
984                 }
985         }
986         ret = 0;
987 error:
988         btrfs_free_path(path);
989         return ret;
990 }
991
992 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
993 {
994         int ret;
995         struct btrfs_key key;
996         struct btrfs_key found_key;
997         struct btrfs_path *path;
998
999         root = root->fs_info->chunk_root;
1000
1001         path = btrfs_alloc_path();
1002         if (!path)
1003                 return -ENOMEM;
1004
1005         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1006         key.type = BTRFS_DEV_ITEM_KEY;
1007         key.offset = (u64)-1;
1008
1009         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1010         if (ret < 0)
1011                 goto error;
1012
1013         BUG_ON(ret == 0);
1014
1015         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1016                                   BTRFS_DEV_ITEM_KEY);
1017         if (ret) {
1018                 *objectid = 1;
1019         } else {
1020                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1021                                       path->slots[0]);
1022                 *objectid = found_key.offset + 1;
1023         }
1024         ret = 0;
1025 error:
1026         btrfs_free_path(path);
1027         return ret;
1028 }
1029
1030 /*
1031  * the device information is stored in the chunk root
1032  * the btrfs_device struct should be fully filled in
1033  */
1034 int btrfs_add_device(struct btrfs_trans_handle *trans,
1035                      struct btrfs_root *root,
1036                      struct btrfs_device *device)
1037 {
1038         int ret;
1039         struct btrfs_path *path;
1040         struct btrfs_dev_item *dev_item;
1041         struct extent_buffer *leaf;
1042         struct btrfs_key key;
1043         unsigned long ptr;
1044
1045         root = root->fs_info->chunk_root;
1046
1047         path = btrfs_alloc_path();
1048         if (!path)
1049                 return -ENOMEM;
1050
1051         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1052         key.type = BTRFS_DEV_ITEM_KEY;
1053         key.offset = device->devid;
1054
1055         ret = btrfs_insert_empty_item(trans, root, path, &key,
1056                                       sizeof(*dev_item));
1057         if (ret)
1058                 goto out;
1059
1060         leaf = path->nodes[0];
1061         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1062
1063         btrfs_set_device_id(leaf, dev_item, device->devid);
1064         btrfs_set_device_generation(leaf, dev_item, 0);
1065         btrfs_set_device_type(leaf, dev_item, device->type);
1066         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1067         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1068         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1069         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1070         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1071         btrfs_set_device_group(leaf, dev_item, 0);
1072         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1073         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1074         btrfs_set_device_start_offset(leaf, dev_item, 0);
1075
1076         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1077         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1078         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1079         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1080         btrfs_mark_buffer_dirty(leaf);
1081
1082         ret = 0;
1083 out:
1084         btrfs_free_path(path);
1085         return ret;
1086 }
1087
1088 static int btrfs_rm_dev_item(struct btrfs_root *root,
1089                              struct btrfs_device *device)
1090 {
1091         int ret;
1092         struct btrfs_path *path;
1093         struct btrfs_key key;
1094         struct btrfs_trans_handle *trans;
1095
1096         root = root->fs_info->chunk_root;
1097
1098         path = btrfs_alloc_path();
1099         if (!path)
1100                 return -ENOMEM;
1101
1102         trans = btrfs_start_transaction(root, 0);
1103         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1104         key.type = BTRFS_DEV_ITEM_KEY;
1105         key.offset = device->devid;
1106         lock_chunks(root);
1107
1108         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1109         if (ret < 0)
1110                 goto out;
1111
1112         if (ret > 0) {
1113                 ret = -ENOENT;
1114                 goto out;
1115         }
1116
1117         ret = btrfs_del_item(trans, root, path);
1118         if (ret)
1119                 goto out;
1120 out:
1121         btrfs_free_path(path);
1122         unlock_chunks(root);
1123         btrfs_commit_transaction(trans, root);
1124         return ret;
1125 }
1126
1127 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1128 {
1129         struct btrfs_device *device;
1130         struct btrfs_device *next_device;
1131         struct block_device *bdev;
1132         struct buffer_head *bh = NULL;
1133         struct btrfs_super_block *disk_super;
1134         u64 all_avail;
1135         u64 devid;
1136         u64 num_devices;
1137         u8 *dev_uuid;
1138         int ret = 0;
1139
1140         mutex_lock(&uuid_mutex);
1141         mutex_lock(&root->fs_info->volume_mutex);
1142
1143         all_avail = root->fs_info->avail_data_alloc_bits |
1144                 root->fs_info->avail_system_alloc_bits |
1145                 root->fs_info->avail_metadata_alloc_bits;
1146
1147         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1148             root->fs_info->fs_devices->num_devices <= 4) {
1149                 printk(KERN_ERR "btrfs: unable to go below four devices "
1150                        "on raid10\n");
1151                 ret = -EINVAL;
1152                 goto out;
1153         }
1154
1155         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1156             root->fs_info->fs_devices->num_devices <= 2) {
1157                 printk(KERN_ERR "btrfs: unable to go below two "
1158                        "devices on raid1\n");
1159                 ret = -EINVAL;
1160                 goto out;
1161         }
1162
1163         if (strcmp(device_path, "missing") == 0) {
1164                 struct list_head *devices;
1165                 struct btrfs_device *tmp;
1166
1167                 device = NULL;
1168                 devices = &root->fs_info->fs_devices->devices;
1169                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1170                 list_for_each_entry(tmp, devices, dev_list) {
1171                         if (tmp->in_fs_metadata && !tmp->bdev) {
1172                                 device = tmp;
1173                                 break;
1174                         }
1175                 }
1176                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1177                 bdev = NULL;
1178                 bh = NULL;
1179                 disk_super = NULL;
1180                 if (!device) {
1181                         printk(KERN_ERR "btrfs: no missing devices found to "
1182                                "remove\n");
1183                         goto out;
1184                 }
1185         } else {
1186                 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1187                                       root->fs_info->bdev_holder);
1188                 if (IS_ERR(bdev)) {
1189                         ret = PTR_ERR(bdev);
1190                         goto out;
1191                 }
1192
1193                 set_blocksize(bdev, 4096);
1194                 bh = btrfs_read_dev_super(bdev);
1195                 if (!bh) {
1196                         ret = -EIO;
1197                         goto error_close;
1198                 }
1199                 disk_super = (struct btrfs_super_block *)bh->b_data;
1200                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1201                 dev_uuid = disk_super->dev_item.uuid;
1202                 device = btrfs_find_device(root, devid, dev_uuid,
1203                                            disk_super->fsid);
1204                 if (!device) {
1205                         ret = -ENOENT;
1206                         goto error_brelse;
1207                 }
1208         }
1209
1210         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1211                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1212                        "device\n");
1213                 ret = -EINVAL;
1214                 goto error_brelse;
1215         }
1216
1217         if (device->writeable) {
1218                 list_del_init(&device->dev_alloc_list);
1219                 root->fs_info->fs_devices->rw_devices--;
1220         }
1221
1222         ret = btrfs_shrink_device(device, 0);
1223         if (ret)
1224                 goto error_brelse;
1225
1226         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1227         if (ret)
1228                 goto error_brelse;
1229
1230         device->in_fs_metadata = 0;
1231
1232         /*
1233          * the device list mutex makes sure that we don't change
1234          * the device list while someone else is writing out all
1235          * the device supers.
1236          */
1237         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1238         list_del_init(&device->dev_list);
1239         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1240
1241         device->fs_devices->num_devices--;
1242
1243         if (device->missing)
1244                 root->fs_info->fs_devices->missing_devices--;
1245
1246         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1247                                  struct btrfs_device, dev_list);
1248         if (device->bdev == root->fs_info->sb->s_bdev)
1249                 root->fs_info->sb->s_bdev = next_device->bdev;
1250         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1251                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1252
1253         if (device->bdev) {
1254                 close_bdev_exclusive(device->bdev, device->mode);
1255                 device->bdev = NULL;
1256                 device->fs_devices->open_devices--;
1257         }
1258
1259         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1260         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1261
1262         if (device->fs_devices->open_devices == 0) {
1263                 struct btrfs_fs_devices *fs_devices;
1264                 fs_devices = root->fs_info->fs_devices;
1265                 while (fs_devices) {
1266                         if (fs_devices->seed == device->fs_devices)
1267                                 break;
1268                         fs_devices = fs_devices->seed;
1269                 }
1270                 fs_devices->seed = device->fs_devices->seed;
1271                 device->fs_devices->seed = NULL;
1272                 __btrfs_close_devices(device->fs_devices);
1273                 free_fs_devices(device->fs_devices);
1274         }
1275
1276         /*
1277          * at this point, the device is zero sized.  We want to
1278          * remove it from the devices list and zero out the old super
1279          */
1280         if (device->writeable) {
1281                 /* make sure this device isn't detected as part of
1282                  * the FS anymore
1283                  */
1284                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1285                 set_buffer_dirty(bh);
1286                 sync_dirty_buffer(bh);
1287         }
1288
1289         kfree(device->name);
1290         kfree(device);
1291         ret = 0;
1292
1293 error_brelse:
1294         brelse(bh);
1295 error_close:
1296         if (bdev)
1297                 close_bdev_exclusive(bdev, FMODE_READ);
1298 out:
1299         mutex_unlock(&root->fs_info->volume_mutex);
1300         mutex_unlock(&uuid_mutex);
1301         return ret;
1302 }
1303
1304 /*
1305  * does all the dirty work required for changing file system's UUID.
1306  */
1307 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1308                                 struct btrfs_root *root)
1309 {
1310         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1311         struct btrfs_fs_devices *old_devices;
1312         struct btrfs_fs_devices *seed_devices;
1313         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1314         struct btrfs_device *device;
1315         u64 super_flags;
1316
1317         BUG_ON(!mutex_is_locked(&uuid_mutex));
1318         if (!fs_devices->seeding)
1319                 return -EINVAL;
1320
1321         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1322         if (!seed_devices)
1323                 return -ENOMEM;
1324
1325         old_devices = clone_fs_devices(fs_devices);
1326         if (IS_ERR(old_devices)) {
1327                 kfree(seed_devices);
1328                 return PTR_ERR(old_devices);
1329         }
1330
1331         list_add(&old_devices->list, &fs_uuids);
1332
1333         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1334         seed_devices->opened = 1;
1335         INIT_LIST_HEAD(&seed_devices->devices);
1336         INIT_LIST_HEAD(&seed_devices->alloc_list);
1337         mutex_init(&seed_devices->device_list_mutex);
1338         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1339         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1340         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1341                 device->fs_devices = seed_devices;
1342         }
1343
1344         fs_devices->seeding = 0;
1345         fs_devices->num_devices = 0;
1346         fs_devices->open_devices = 0;
1347         fs_devices->seed = seed_devices;
1348
1349         generate_random_uuid(fs_devices->fsid);
1350         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1351         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1352         super_flags = btrfs_super_flags(disk_super) &
1353                       ~BTRFS_SUPER_FLAG_SEEDING;
1354         btrfs_set_super_flags(disk_super, super_flags);
1355
1356         return 0;
1357 }
1358
1359 /*
1360  * strore the expected generation for seed devices in device items.
1361  */
1362 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1363                                struct btrfs_root *root)
1364 {
1365         struct btrfs_path *path;
1366         struct extent_buffer *leaf;
1367         struct btrfs_dev_item *dev_item;
1368         struct btrfs_device *device;
1369         struct btrfs_key key;
1370         u8 fs_uuid[BTRFS_UUID_SIZE];
1371         u8 dev_uuid[BTRFS_UUID_SIZE];
1372         u64 devid;
1373         int ret;
1374
1375         path = btrfs_alloc_path();
1376         if (!path)
1377                 return -ENOMEM;
1378
1379         root = root->fs_info->chunk_root;
1380         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1381         key.offset = 0;
1382         key.type = BTRFS_DEV_ITEM_KEY;
1383
1384         while (1) {
1385                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1386                 if (ret < 0)
1387                         goto error;
1388
1389                 leaf = path->nodes[0];
1390 next_slot:
1391                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1392                         ret = btrfs_next_leaf(root, path);
1393                         if (ret > 0)
1394                                 break;
1395                         if (ret < 0)
1396                                 goto error;
1397                         leaf = path->nodes[0];
1398                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1399                         btrfs_release_path(root, path);
1400                         continue;
1401                 }
1402
1403                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1404                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1405                     key.type != BTRFS_DEV_ITEM_KEY)
1406                         break;
1407
1408                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1409                                           struct btrfs_dev_item);
1410                 devid = btrfs_device_id(leaf, dev_item);
1411                 read_extent_buffer(leaf, dev_uuid,
1412                                    (unsigned long)btrfs_device_uuid(dev_item),
1413                                    BTRFS_UUID_SIZE);
1414                 read_extent_buffer(leaf, fs_uuid,
1415                                    (unsigned long)btrfs_device_fsid(dev_item),
1416                                    BTRFS_UUID_SIZE);
1417                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1418                 BUG_ON(!device);
1419
1420                 if (device->fs_devices->seeding) {
1421                         btrfs_set_device_generation(leaf, dev_item,
1422                                                     device->generation);
1423                         btrfs_mark_buffer_dirty(leaf);
1424                 }
1425
1426                 path->slots[0]++;
1427                 goto next_slot;
1428         }
1429         ret = 0;
1430 error:
1431         btrfs_free_path(path);
1432         return ret;
1433 }
1434
1435 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1436 {
1437         struct btrfs_trans_handle *trans;
1438         struct btrfs_device *device;
1439         struct block_device *bdev;
1440         struct list_head *devices;
1441         struct super_block *sb = root->fs_info->sb;
1442         u64 total_bytes;
1443         int seeding_dev = 0;
1444         int ret = 0;
1445
1446         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1447                 return -EINVAL;
1448
1449         bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1450         if (IS_ERR(bdev))
1451                 return PTR_ERR(bdev);
1452
1453         if (root->fs_info->fs_devices->seeding) {
1454                 seeding_dev = 1;
1455                 down_write(&sb->s_umount);
1456                 mutex_lock(&uuid_mutex);
1457         }
1458
1459         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1460         mutex_lock(&root->fs_info->volume_mutex);
1461
1462         devices = &root->fs_info->fs_devices->devices;
1463         /*
1464          * we have the volume lock, so we don't need the extra
1465          * device list mutex while reading the list here.
1466          */
1467         list_for_each_entry(device, devices, dev_list) {
1468                 if (device->bdev == bdev) {
1469                         ret = -EEXIST;
1470                         goto error;
1471                 }
1472         }
1473
1474         device = kzalloc(sizeof(*device), GFP_NOFS);
1475         if (!device) {
1476                 /* we can safely leave the fs_devices entry around */
1477                 ret = -ENOMEM;
1478                 goto error;
1479         }
1480
1481         device->name = kstrdup(device_path, GFP_NOFS);
1482         if (!device->name) {
1483                 kfree(device);
1484                 ret = -ENOMEM;
1485                 goto error;
1486         }
1487
1488         ret = find_next_devid(root, &device->devid);
1489         if (ret) {
1490                 kfree(device);
1491                 goto error;
1492         }
1493
1494         trans = btrfs_start_transaction(root, 0);
1495         lock_chunks(root);
1496
1497         device->writeable = 1;
1498         device->work.func = pending_bios_fn;
1499         generate_random_uuid(device->uuid);
1500         spin_lock_init(&device->io_lock);
1501         device->generation = trans->transid;
1502         device->io_width = root->sectorsize;
1503         device->io_align = root->sectorsize;
1504         device->sector_size = root->sectorsize;
1505         device->total_bytes = i_size_read(bdev->bd_inode);
1506         device->disk_total_bytes = device->total_bytes;
1507         device->dev_root = root->fs_info->dev_root;
1508         device->bdev = bdev;
1509         device->in_fs_metadata = 1;
1510         device->mode = 0;
1511         set_blocksize(device->bdev, 4096);
1512
1513         if (seeding_dev) {
1514                 sb->s_flags &= ~MS_RDONLY;
1515                 ret = btrfs_prepare_sprout(trans, root);
1516                 BUG_ON(ret);
1517         }
1518
1519         device->fs_devices = root->fs_info->fs_devices;
1520
1521         /*
1522          * we don't want write_supers to jump in here with our device
1523          * half setup
1524          */
1525         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1526         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1527         list_add(&device->dev_alloc_list,
1528                  &root->fs_info->fs_devices->alloc_list);
1529         root->fs_info->fs_devices->num_devices++;
1530         root->fs_info->fs_devices->open_devices++;
1531         root->fs_info->fs_devices->rw_devices++;
1532         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1533
1534         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1535                 root->fs_info->fs_devices->rotating = 1;
1536
1537         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1538         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1539                                     total_bytes + device->total_bytes);
1540
1541         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1542         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1543                                     total_bytes + 1);
1544         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1545
1546         if (seeding_dev) {
1547                 ret = init_first_rw_device(trans, root, device);
1548                 BUG_ON(ret);
1549                 ret = btrfs_finish_sprout(trans, root);
1550                 BUG_ON(ret);
1551         } else {
1552                 ret = btrfs_add_device(trans, root, device);
1553         }
1554
1555         /*
1556          * we've got more storage, clear any full flags on the space
1557          * infos
1558          */
1559         btrfs_clear_space_info_full(root->fs_info);
1560
1561         unlock_chunks(root);
1562         btrfs_commit_transaction(trans, root);
1563
1564         if (seeding_dev) {
1565                 mutex_unlock(&uuid_mutex);
1566                 up_write(&sb->s_umount);
1567
1568                 ret = btrfs_relocate_sys_chunks(root);
1569                 BUG_ON(ret);
1570         }
1571 out:
1572         mutex_unlock(&root->fs_info->volume_mutex);
1573         return ret;
1574 error:
1575         close_bdev_exclusive(bdev, 0);
1576         if (seeding_dev) {
1577                 mutex_unlock(&uuid_mutex);
1578                 up_write(&sb->s_umount);
1579         }
1580         goto out;
1581 }
1582
1583 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1584                                         struct btrfs_device *device)
1585 {
1586         int ret;
1587         struct btrfs_path *path;
1588         struct btrfs_root *root;
1589         struct btrfs_dev_item *dev_item;
1590         struct extent_buffer *leaf;
1591         struct btrfs_key key;
1592
1593         root = device->dev_root->fs_info->chunk_root;
1594
1595         path = btrfs_alloc_path();
1596         if (!path)
1597                 return -ENOMEM;
1598
1599         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1600         key.type = BTRFS_DEV_ITEM_KEY;
1601         key.offset = device->devid;
1602
1603         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1604         if (ret < 0)
1605                 goto out;
1606
1607         if (ret > 0) {
1608                 ret = -ENOENT;
1609                 goto out;
1610         }
1611
1612         leaf = path->nodes[0];
1613         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1614
1615         btrfs_set_device_id(leaf, dev_item, device->devid);
1616         btrfs_set_device_type(leaf, dev_item, device->type);
1617         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1618         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1619         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1620         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1621         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1622         btrfs_mark_buffer_dirty(leaf);
1623
1624 out:
1625         btrfs_free_path(path);
1626         return ret;
1627 }
1628
1629 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1630                       struct btrfs_device *device, u64 new_size)
1631 {
1632         struct btrfs_super_block *super_copy =
1633                 &device->dev_root->fs_info->super_copy;
1634         u64 old_total = btrfs_super_total_bytes(super_copy);
1635         u64 diff = new_size - device->total_bytes;
1636
1637         if (!device->writeable)
1638                 return -EACCES;
1639         if (new_size <= device->total_bytes)
1640                 return -EINVAL;
1641
1642         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1643         device->fs_devices->total_rw_bytes += diff;
1644
1645         device->total_bytes = new_size;
1646         device->disk_total_bytes = new_size;
1647         btrfs_clear_space_info_full(device->dev_root->fs_info);
1648
1649         return btrfs_update_device(trans, device);
1650 }
1651
1652 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1653                       struct btrfs_device *device, u64 new_size)
1654 {
1655         int ret;
1656         lock_chunks(device->dev_root);
1657         ret = __btrfs_grow_device(trans, device, new_size);
1658         unlock_chunks(device->dev_root);
1659         return ret;
1660 }
1661
1662 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1663                             struct btrfs_root *root,
1664                             u64 chunk_tree, u64 chunk_objectid,
1665                             u64 chunk_offset)
1666 {
1667         int ret;
1668         struct btrfs_path *path;
1669         struct btrfs_key key;
1670
1671         root = root->fs_info->chunk_root;
1672         path = btrfs_alloc_path();
1673         if (!path)
1674                 return -ENOMEM;
1675
1676         key.objectid = chunk_objectid;
1677         key.offset = chunk_offset;
1678         key.type = BTRFS_CHUNK_ITEM_KEY;
1679
1680         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1681         BUG_ON(ret);
1682
1683         ret = btrfs_del_item(trans, root, path);
1684         BUG_ON(ret);
1685
1686         btrfs_free_path(path);
1687         return 0;
1688 }
1689
1690 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1691                         chunk_offset)
1692 {
1693         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1694         struct btrfs_disk_key *disk_key;
1695         struct btrfs_chunk *chunk;
1696         u8 *ptr;
1697         int ret = 0;
1698         u32 num_stripes;
1699         u32 array_size;
1700         u32 len = 0;
1701         u32 cur;
1702         struct btrfs_key key;
1703
1704         array_size = btrfs_super_sys_array_size(super_copy);
1705
1706         ptr = super_copy->sys_chunk_array;
1707         cur = 0;
1708
1709         while (cur < array_size) {
1710                 disk_key = (struct btrfs_disk_key *)ptr;
1711                 btrfs_disk_key_to_cpu(&key, disk_key);
1712
1713                 len = sizeof(*disk_key);
1714
1715                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1716                         chunk = (struct btrfs_chunk *)(ptr + len);
1717                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1718                         len += btrfs_chunk_item_size(num_stripes);
1719                 } else {
1720                         ret = -EIO;
1721                         break;
1722                 }
1723                 if (key.objectid == chunk_objectid &&
1724                     key.offset == chunk_offset) {
1725                         memmove(ptr, ptr + len, array_size - (cur + len));
1726                         array_size -= len;
1727                         btrfs_set_super_sys_array_size(super_copy, array_size);
1728                 } else {
1729                         ptr += len;
1730                         cur += len;
1731                 }
1732         }
1733         return ret;
1734 }
1735
1736 static int btrfs_relocate_chunk(struct btrfs_root *root,
1737                          u64 chunk_tree, u64 chunk_objectid,
1738                          u64 chunk_offset)
1739 {
1740         struct extent_map_tree *em_tree;
1741         struct btrfs_root *extent_root;
1742         struct btrfs_trans_handle *trans;
1743         struct extent_map *em;
1744         struct map_lookup *map;
1745         int ret;
1746         int i;
1747
1748         root = root->fs_info->chunk_root;
1749         extent_root = root->fs_info->extent_root;
1750         em_tree = &root->fs_info->mapping_tree.map_tree;
1751
1752         ret = btrfs_can_relocate(extent_root, chunk_offset);
1753         if (ret)
1754                 return -ENOSPC;
1755
1756         /* step one, relocate all the extents inside this chunk */
1757         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1758         if (ret)
1759                 return ret;
1760
1761         trans = btrfs_start_transaction(root, 0);
1762         BUG_ON(!trans);
1763
1764         lock_chunks(root);
1765
1766         /*
1767          * step two, delete the device extents and the
1768          * chunk tree entries
1769          */
1770         read_lock(&em_tree->lock);
1771         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1772         read_unlock(&em_tree->lock);
1773
1774         BUG_ON(em->start > chunk_offset ||
1775                em->start + em->len < chunk_offset);
1776         map = (struct map_lookup *)em->bdev;
1777
1778         for (i = 0; i < map->num_stripes; i++) {
1779                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1780                                             map->stripes[i].physical);
1781                 BUG_ON(ret);
1782
1783                 if (map->stripes[i].dev) {
1784                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1785                         BUG_ON(ret);
1786                 }
1787         }
1788         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1789                                chunk_offset);
1790
1791         BUG_ON(ret);
1792
1793         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1794                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1795                 BUG_ON(ret);
1796         }
1797
1798         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1799         BUG_ON(ret);
1800
1801         write_lock(&em_tree->lock);
1802         remove_extent_mapping(em_tree, em);
1803         write_unlock(&em_tree->lock);
1804
1805         kfree(map);
1806         em->bdev = NULL;
1807
1808         /* once for the tree */
1809         free_extent_map(em);
1810         /* once for us */
1811         free_extent_map(em);
1812
1813         unlock_chunks(root);
1814         btrfs_end_transaction(trans, root);
1815         return 0;
1816 }
1817
1818 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1819 {
1820         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1821         struct btrfs_path *path;
1822         struct extent_buffer *leaf;
1823         struct btrfs_chunk *chunk;
1824         struct btrfs_key key;
1825         struct btrfs_key found_key;
1826         u64 chunk_tree = chunk_root->root_key.objectid;
1827         u64 chunk_type;
1828         bool retried = false;
1829         int failed = 0;
1830         int ret;
1831
1832         path = btrfs_alloc_path();
1833         if (!path)
1834                 return -ENOMEM;
1835
1836 again:
1837         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1838         key.offset = (u64)-1;
1839         key.type = BTRFS_CHUNK_ITEM_KEY;
1840
1841         while (1) {
1842                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1843                 if (ret < 0)
1844                         goto error;
1845                 BUG_ON(ret == 0);
1846
1847                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1848                                           key.type);
1849                 if (ret < 0)
1850                         goto error;
1851                 if (ret > 0)
1852                         break;
1853
1854                 leaf = path->nodes[0];
1855                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1856
1857                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1858                                        struct btrfs_chunk);
1859                 chunk_type = btrfs_chunk_type(leaf, chunk);
1860                 btrfs_release_path(chunk_root, path);
1861
1862                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1863                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1864                                                    found_key.objectid,
1865                                                    found_key.offset);
1866                         if (ret == -ENOSPC)
1867                                 failed++;
1868                         else if (ret)
1869                                 BUG();
1870                 }
1871
1872                 if (found_key.offset == 0)
1873                         break;
1874                 key.offset = found_key.offset - 1;
1875         }
1876         ret = 0;
1877         if (failed && !retried) {
1878                 failed = 0;
1879                 retried = true;
1880                 goto again;
1881         } else if (failed && retried) {
1882                 WARN_ON(1);
1883                 ret = -ENOSPC;
1884         }
1885 error:
1886         btrfs_free_path(path);
1887         return ret;
1888 }
1889
1890 static u64 div_factor(u64 num, int factor)
1891 {
1892         if (factor == 10)
1893                 return num;
1894         num *= factor;
1895         do_div(num, 10);
1896         return num;
1897 }
1898
1899 int btrfs_balance(struct btrfs_root *dev_root)
1900 {
1901         int ret;
1902         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1903         struct btrfs_device *device;
1904         u64 old_size;
1905         u64 size_to_free;
1906         struct btrfs_path *path;
1907         struct btrfs_key key;
1908         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1909         struct btrfs_trans_handle *trans;
1910         struct btrfs_key found_key;
1911
1912         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1913                 return -EROFS;
1914
1915         mutex_lock(&dev_root->fs_info->volume_mutex);
1916         dev_root = dev_root->fs_info->dev_root;
1917
1918         /* step one make some room on all the devices */
1919         list_for_each_entry(device, devices, dev_list) {
1920                 old_size = device->total_bytes;
1921                 size_to_free = div_factor(old_size, 1);
1922                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1923                 if (!device->writeable ||
1924                     device->total_bytes - device->bytes_used > size_to_free)
1925                         continue;
1926
1927                 ret = btrfs_shrink_device(device, old_size - size_to_free);
1928                 if (ret == -ENOSPC)
1929                         break;
1930                 BUG_ON(ret);
1931
1932                 trans = btrfs_start_transaction(dev_root, 0);
1933                 BUG_ON(!trans);
1934
1935                 ret = btrfs_grow_device(trans, device, old_size);
1936                 BUG_ON(ret);
1937
1938                 btrfs_end_transaction(trans, dev_root);
1939         }
1940
1941         /* step two, relocate all the chunks */
1942         path = btrfs_alloc_path();
1943         BUG_ON(!path);
1944
1945         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1946         key.offset = (u64)-1;
1947         key.type = BTRFS_CHUNK_ITEM_KEY;
1948
1949         while (1) {
1950                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1951                 if (ret < 0)
1952                         goto error;
1953
1954                 /*
1955                  * this shouldn't happen, it means the last relocate
1956                  * failed
1957                  */
1958                 if (ret == 0)
1959                         break;
1960
1961                 ret = btrfs_previous_item(chunk_root, path, 0,
1962                                           BTRFS_CHUNK_ITEM_KEY);
1963                 if (ret)
1964                         break;
1965
1966                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1967                                       path->slots[0]);
1968                 if (found_key.objectid != key.objectid)
1969                         break;
1970
1971                 /* chunk zero is special */
1972                 if (found_key.offset == 0)
1973                         break;
1974
1975                 btrfs_release_path(chunk_root, path);
1976                 ret = btrfs_relocate_chunk(chunk_root,
1977                                            chunk_root->root_key.objectid,
1978                                            found_key.objectid,
1979                                            found_key.offset);
1980                 BUG_ON(ret && ret != -ENOSPC);
1981                 key.offset = found_key.offset - 1;
1982         }
1983         ret = 0;
1984 error:
1985         btrfs_free_path(path);
1986         mutex_unlock(&dev_root->fs_info->volume_mutex);
1987         return ret;
1988 }
1989
1990 /*
1991  * shrinking a device means finding all of the device extents past
1992  * the new size, and then following the back refs to the chunks.
1993  * The chunk relocation code actually frees the device extent
1994  */
1995 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1996 {
1997         struct btrfs_trans_handle *trans;
1998         struct btrfs_root *root = device->dev_root;
1999         struct btrfs_dev_extent *dev_extent = NULL;
2000         struct btrfs_path *path;
2001         u64 length;
2002         u64 chunk_tree;
2003         u64 chunk_objectid;
2004         u64 chunk_offset;
2005         int ret;
2006         int slot;
2007         int failed = 0;
2008         bool retried = false;
2009         struct extent_buffer *l;
2010         struct btrfs_key key;
2011         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2012         u64 old_total = btrfs_super_total_bytes(super_copy);
2013         u64 old_size = device->total_bytes;
2014         u64 diff = device->total_bytes - new_size;
2015
2016         if (new_size >= device->total_bytes)
2017                 return -EINVAL;
2018
2019         path = btrfs_alloc_path();
2020         if (!path)
2021                 return -ENOMEM;
2022
2023         path->reada = 2;
2024
2025         lock_chunks(root);
2026
2027         device->total_bytes = new_size;
2028         if (device->writeable)
2029                 device->fs_devices->total_rw_bytes -= diff;
2030         unlock_chunks(root);
2031
2032 again:
2033         key.objectid = device->devid;
2034         key.offset = (u64)-1;
2035         key.type = BTRFS_DEV_EXTENT_KEY;
2036
2037         while (1) {
2038                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2039                 if (ret < 0)
2040                         goto done;
2041
2042                 ret = btrfs_previous_item(root, path, 0, key.type);
2043                 if (ret < 0)
2044                         goto done;
2045                 if (ret) {
2046                         ret = 0;
2047                         btrfs_release_path(root, path);
2048                         break;
2049                 }
2050
2051                 l = path->nodes[0];
2052                 slot = path->slots[0];
2053                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2054
2055                 if (key.objectid != device->devid) {
2056                         btrfs_release_path(root, path);
2057                         break;
2058                 }
2059
2060                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2061                 length = btrfs_dev_extent_length(l, dev_extent);
2062
2063                 if (key.offset + length <= new_size) {
2064                         btrfs_release_path(root, path);
2065                         break;
2066                 }
2067
2068                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2069                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2070                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2071                 btrfs_release_path(root, path);
2072
2073                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2074                                            chunk_offset);
2075                 if (ret && ret != -ENOSPC)
2076                         goto done;
2077                 if (ret == -ENOSPC)
2078                         failed++;
2079                 key.offset -= 1;
2080         }
2081
2082         if (failed && !retried) {
2083                 failed = 0;
2084                 retried = true;
2085                 goto again;
2086         } else if (failed && retried) {
2087                 ret = -ENOSPC;
2088                 lock_chunks(root);
2089
2090                 device->total_bytes = old_size;
2091                 if (device->writeable)
2092                         device->fs_devices->total_rw_bytes += diff;
2093                 unlock_chunks(root);
2094                 goto done;
2095         }
2096
2097         /* Shrinking succeeded, else we would be at "done". */
2098         trans = btrfs_start_transaction(root, 0);
2099         lock_chunks(root);
2100
2101         device->disk_total_bytes = new_size;
2102         /* Now btrfs_update_device() will change the on-disk size. */
2103         ret = btrfs_update_device(trans, device);
2104         if (ret) {
2105                 unlock_chunks(root);
2106                 btrfs_end_transaction(trans, root);
2107                 goto done;
2108         }
2109         WARN_ON(diff > old_total);
2110         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2111         unlock_chunks(root);
2112         btrfs_end_transaction(trans, root);
2113 done:
2114         btrfs_free_path(path);
2115         return ret;
2116 }
2117
2118 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2119                            struct btrfs_root *root,
2120                            struct btrfs_key *key,
2121                            struct btrfs_chunk *chunk, int item_size)
2122 {
2123         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2124         struct btrfs_disk_key disk_key;
2125         u32 array_size;
2126         u8 *ptr;
2127
2128         array_size = btrfs_super_sys_array_size(super_copy);
2129         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2130                 return -EFBIG;
2131
2132         ptr = super_copy->sys_chunk_array + array_size;
2133         btrfs_cpu_key_to_disk(&disk_key, key);
2134         memcpy(ptr, &disk_key, sizeof(disk_key));
2135         ptr += sizeof(disk_key);
2136         memcpy(ptr, chunk, item_size);
2137         item_size += sizeof(disk_key);
2138         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2139         return 0;
2140 }
2141
2142 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2143                                         int num_stripes, int sub_stripes)
2144 {
2145         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2146                 return calc_size;
2147         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2148                 return calc_size * (num_stripes / sub_stripes);
2149         else
2150                 return calc_size * num_stripes;
2151 }
2152
2153 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2154                                struct btrfs_root *extent_root,
2155                                struct map_lookup **map_ret,
2156                                u64 *num_bytes, u64 *stripe_size,
2157                                u64 start, u64 type)
2158 {
2159         struct btrfs_fs_info *info = extent_root->fs_info;
2160         struct btrfs_device *device = NULL;
2161         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2162         struct list_head *cur;
2163         struct map_lookup *map = NULL;
2164         struct extent_map_tree *em_tree;
2165         struct extent_map *em;
2166         struct list_head private_devs;
2167         int min_stripe_size = 1 * 1024 * 1024;
2168         u64 calc_size = 1024 * 1024 * 1024;
2169         u64 max_chunk_size = calc_size;
2170         u64 min_free;
2171         u64 avail;
2172         u64 max_avail = 0;
2173         u64 dev_offset;
2174         int num_stripes = 1;
2175         int min_stripes = 1;
2176         int sub_stripes = 0;
2177         int looped = 0;
2178         int ret;
2179         int index;
2180         int stripe_len = 64 * 1024;
2181
2182         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2183             (type & BTRFS_BLOCK_GROUP_DUP)) {
2184                 WARN_ON(1);
2185                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2186         }
2187         if (list_empty(&fs_devices->alloc_list))
2188                 return -ENOSPC;
2189
2190         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2191                 num_stripes = fs_devices->rw_devices;
2192                 min_stripes = 2;
2193         }
2194         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2195                 num_stripes = 2;
2196                 min_stripes = 2;
2197         }
2198         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2199                 if (fs_devices->rw_devices < 2)
2200                         return -ENOSPC;
2201                 num_stripes = 2;
2202                 min_stripes = 2;
2203         }
2204         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2205                 num_stripes = fs_devices->rw_devices;
2206                 if (num_stripes < 4)
2207                         return -ENOSPC;
2208                 num_stripes &= ~(u32)1;
2209                 sub_stripes = 2;
2210                 min_stripes = 4;
2211         }
2212
2213         if (type & BTRFS_BLOCK_GROUP_DATA) {
2214                 max_chunk_size = 10 * calc_size;
2215                 min_stripe_size = 64 * 1024 * 1024;
2216         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2217                 max_chunk_size = 256 * 1024 * 1024;
2218                 min_stripe_size = 32 * 1024 * 1024;
2219         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2220                 calc_size = 8 * 1024 * 1024;
2221                 max_chunk_size = calc_size * 2;
2222                 min_stripe_size = 1 * 1024 * 1024;
2223         }
2224
2225         /* we don't want a chunk larger than 10% of writeable space */
2226         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2227                              max_chunk_size);
2228
2229 again:
2230         max_avail = 0;
2231         if (!map || map->num_stripes != num_stripes) {
2232                 kfree(map);
2233                 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2234                 if (!map)
2235                         return -ENOMEM;
2236                 map->num_stripes = num_stripes;
2237         }
2238
2239         if (calc_size * num_stripes > max_chunk_size) {
2240                 calc_size = max_chunk_size;
2241                 do_div(calc_size, num_stripes);
2242                 do_div(calc_size, stripe_len);
2243                 calc_size *= stripe_len;
2244         }
2245
2246         /* we don't want tiny stripes */
2247         if (!looped)
2248                 calc_size = max_t(u64, min_stripe_size, calc_size);
2249
2250         /*
2251          * we're about to do_div by the stripe_len so lets make sure
2252          * we end up with something bigger than a stripe
2253          */
2254         calc_size = max_t(u64, calc_size, stripe_len * 4);
2255
2256         do_div(calc_size, stripe_len);
2257         calc_size *= stripe_len;
2258
2259         cur = fs_devices->alloc_list.next;
2260         index = 0;
2261
2262         if (type & BTRFS_BLOCK_GROUP_DUP)
2263                 min_free = calc_size * 2;
2264         else
2265                 min_free = calc_size;
2266
2267         /*
2268          * we add 1MB because we never use the first 1MB of the device, unless
2269          * we've looped, then we are likely allocating the maximum amount of
2270          * space left already
2271          */
2272         if (!looped)
2273                 min_free += 1024 * 1024;
2274
2275         INIT_LIST_HEAD(&private_devs);
2276         while (index < num_stripes) {
2277                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2278                 BUG_ON(!device->writeable);
2279                 if (device->total_bytes > device->bytes_used)
2280                         avail = device->total_bytes - device->bytes_used;
2281                 else
2282                         avail = 0;
2283                 cur = cur->next;
2284
2285                 if (device->in_fs_metadata && avail >= min_free) {
2286                         ret = find_free_dev_extent(trans, device,
2287                                                    min_free, &dev_offset,
2288                                                    &max_avail);
2289                         if (ret == 0) {
2290                                 list_move_tail(&device->dev_alloc_list,
2291                                                &private_devs);
2292                                 map->stripes[index].dev = device;
2293                                 map->stripes[index].physical = dev_offset;
2294                                 index++;
2295                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2296                                         map->stripes[index].dev = device;
2297                                         map->stripes[index].physical =
2298                                                 dev_offset + calc_size;
2299                                         index++;
2300                                 }
2301                         }
2302                 } else if (device->in_fs_metadata && avail > max_avail)
2303                         max_avail = avail;
2304                 if (cur == &fs_devices->alloc_list)
2305                         break;
2306         }
2307         list_splice(&private_devs, &fs_devices->alloc_list);
2308         if (index < num_stripes) {
2309                 if (index >= min_stripes) {
2310                         num_stripes = index;
2311                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2312                                 num_stripes /= sub_stripes;
2313                                 num_stripes *= sub_stripes;
2314                         }
2315                         looped = 1;
2316                         goto again;
2317                 }
2318                 if (!looped && max_avail > 0) {
2319                         looped = 1;
2320                         calc_size = max_avail;
2321                         goto again;
2322                 }
2323                 kfree(map);
2324                 return -ENOSPC;
2325         }
2326         map->sector_size = extent_root->sectorsize;
2327         map->stripe_len = stripe_len;
2328         map->io_align = stripe_len;
2329         map->io_width = stripe_len;
2330         map->type = type;
2331         map->num_stripes = num_stripes;
2332         map->sub_stripes = sub_stripes;
2333
2334         *map_ret = map;
2335         *stripe_size = calc_size;
2336         *num_bytes = chunk_bytes_by_type(type, calc_size,
2337                                          num_stripes, sub_stripes);
2338
2339         em = alloc_extent_map(GFP_NOFS);
2340         if (!em) {
2341                 kfree(map);
2342                 return -ENOMEM;
2343         }
2344         em->bdev = (struct block_device *)map;
2345         em->start = start;
2346         em->len = *num_bytes;
2347         em->block_start = 0;
2348         em->block_len = em->len;
2349
2350         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2351         write_lock(&em_tree->lock);
2352         ret = add_extent_mapping(em_tree, em);
2353         write_unlock(&em_tree->lock);
2354         BUG_ON(ret);
2355         free_extent_map(em);
2356
2357         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2358                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2359                                      start, *num_bytes);
2360         BUG_ON(ret);
2361
2362         index = 0;
2363         while (index < map->num_stripes) {
2364                 device = map->stripes[index].dev;
2365                 dev_offset = map->stripes[index].physical;
2366
2367                 ret = btrfs_alloc_dev_extent(trans, device,
2368                                 info->chunk_root->root_key.objectid,
2369                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2370                                 start, dev_offset, calc_size);
2371                 BUG_ON(ret);
2372                 index++;
2373         }
2374
2375         return 0;
2376 }
2377
2378 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2379                                 struct btrfs_root *extent_root,
2380                                 struct map_lookup *map, u64 chunk_offset,
2381                                 u64 chunk_size, u64 stripe_size)
2382 {
2383         u64 dev_offset;
2384         struct btrfs_key key;
2385         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2386         struct btrfs_device *device;
2387         struct btrfs_chunk *chunk;
2388         struct btrfs_stripe *stripe;
2389         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2390         int index = 0;
2391         int ret;
2392
2393         chunk = kzalloc(item_size, GFP_NOFS);
2394         if (!chunk)
2395                 return -ENOMEM;
2396
2397         index = 0;
2398         while (index < map->num_stripes) {
2399                 device = map->stripes[index].dev;
2400                 device->bytes_used += stripe_size;
2401                 ret = btrfs_update_device(trans, device);
2402                 BUG_ON(ret);
2403                 index++;
2404         }
2405
2406         index = 0;
2407         stripe = &chunk->stripe;
2408         while (index < map->num_stripes) {
2409                 device = map->stripes[index].dev;
2410                 dev_offset = map->stripes[index].physical;
2411
2412                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2413                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2414                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2415                 stripe++;
2416                 index++;
2417         }
2418
2419         btrfs_set_stack_chunk_length(chunk, chunk_size);
2420         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2421         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2422         btrfs_set_stack_chunk_type(chunk, map->type);
2423         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2424         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2425         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2426         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2427         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2428
2429         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2430         key.type = BTRFS_CHUNK_ITEM_KEY;
2431         key.offset = chunk_offset;
2432
2433         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2434         BUG_ON(ret);
2435
2436         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2437                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2438                                              item_size);
2439                 BUG_ON(ret);
2440         }
2441         kfree(chunk);
2442         return 0;
2443 }
2444
2445 /*
2446  * Chunk allocation falls into two parts. The first part does works
2447  * that make the new allocated chunk useable, but not do any operation
2448  * that modifies the chunk tree. The second part does the works that
2449  * require modifying the chunk tree. This division is important for the
2450  * bootstrap process of adding storage to a seed btrfs.
2451  */
2452 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2453                       struct btrfs_root *extent_root, u64 type)
2454 {
2455         u64 chunk_offset;
2456         u64 chunk_size;
2457         u64 stripe_size;
2458         struct map_lookup *map;
2459         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2460         int ret;
2461
2462         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2463                               &chunk_offset);
2464         if (ret)
2465                 return ret;
2466
2467         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2468                                   &stripe_size, chunk_offset, type);
2469         if (ret)
2470                 return ret;
2471
2472         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2473                                    chunk_size, stripe_size);
2474         BUG_ON(ret);
2475         return 0;
2476 }
2477
2478 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2479                                          struct btrfs_root *root,
2480                                          struct btrfs_device *device)
2481 {
2482         u64 chunk_offset;
2483         u64 sys_chunk_offset;
2484         u64 chunk_size;
2485         u64 sys_chunk_size;
2486         u64 stripe_size;
2487         u64 sys_stripe_size;
2488         u64 alloc_profile;
2489         struct map_lookup *map;
2490         struct map_lookup *sys_map;
2491         struct btrfs_fs_info *fs_info = root->fs_info;
2492         struct btrfs_root *extent_root = fs_info->extent_root;
2493         int ret;
2494
2495         ret = find_next_chunk(fs_info->chunk_root,
2496                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2497         BUG_ON(ret);
2498
2499         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2500                         (fs_info->metadata_alloc_profile &
2501                          fs_info->avail_metadata_alloc_bits);
2502         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2503
2504         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2505                                   &stripe_size, chunk_offset, alloc_profile);
2506         BUG_ON(ret);
2507
2508         sys_chunk_offset = chunk_offset + chunk_size;
2509
2510         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2511                         (fs_info->system_alloc_profile &
2512                          fs_info->avail_system_alloc_bits);
2513         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2514
2515         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2516                                   &sys_chunk_size, &sys_stripe_size,
2517                                   sys_chunk_offset, alloc_profile);
2518         BUG_ON(ret);
2519
2520         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2521         BUG_ON(ret);
2522
2523         /*
2524          * Modifying chunk tree needs allocating new blocks from both
2525          * system block group and metadata block group. So we only can
2526          * do operations require modifying the chunk tree after both
2527          * block groups were created.
2528          */
2529         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2530                                    chunk_size, stripe_size);
2531         BUG_ON(ret);
2532
2533         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2534                                    sys_chunk_offset, sys_chunk_size,
2535                                    sys_stripe_size);
2536         BUG_ON(ret);
2537         return 0;
2538 }
2539
2540 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2541 {
2542         struct extent_map *em;
2543         struct map_lookup *map;
2544         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2545         int readonly = 0;
2546         int i;
2547
2548         read_lock(&map_tree->map_tree.lock);
2549         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2550         read_unlock(&map_tree->map_tree.lock);
2551         if (!em)
2552                 return 1;
2553
2554         if (btrfs_test_opt(root, DEGRADED)) {
2555                 free_extent_map(em);
2556                 return 0;
2557         }
2558
2559         map = (struct map_lookup *)em->bdev;
2560         for (i = 0; i < map->num_stripes; i++) {
2561                 if (!map->stripes[i].dev->writeable) {
2562                         readonly = 1;
2563                         break;
2564                 }
2565         }
2566         free_extent_map(em);
2567         return readonly;
2568 }
2569
2570 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2571 {
2572         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2573 }
2574
2575 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2576 {
2577         struct extent_map *em;
2578
2579         while (1) {
2580                 write_lock(&tree->map_tree.lock);
2581                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2582                 if (em)
2583                         remove_extent_mapping(&tree->map_tree, em);
2584                 write_unlock(&tree->map_tree.lock);
2585                 if (!em)
2586                         break;
2587                 kfree(em->bdev);
2588                 /* once for us */
2589                 free_extent_map(em);
2590                 /* once for the tree */
2591                 free_extent_map(em);
2592         }
2593 }
2594
2595 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2596 {
2597         struct extent_map *em;
2598         struct map_lookup *map;
2599         struct extent_map_tree *em_tree = &map_tree->map_tree;
2600         int ret;
2601
2602         read_lock(&em_tree->lock);
2603         em = lookup_extent_mapping(em_tree, logical, len);
2604         read_unlock(&em_tree->lock);
2605         BUG_ON(!em);
2606
2607         BUG_ON(em->start > logical || em->start + em->len < logical);
2608         map = (struct map_lookup *)em->bdev;
2609         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2610                 ret = map->num_stripes;
2611         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2612                 ret = map->sub_stripes;
2613         else
2614                 ret = 1;
2615         free_extent_map(em);
2616         return ret;
2617 }
2618
2619 static int find_live_mirror(struct map_lookup *map, int first, int num,
2620                             int optimal)
2621 {
2622         int i;
2623         if (map->stripes[optimal].dev->bdev)
2624                 return optimal;
2625         for (i = first; i < first + num; i++) {
2626                 if (map->stripes[i].dev->bdev)
2627                         return i;
2628         }
2629         /* we couldn't find one that doesn't fail.  Just return something
2630          * and the io error handling code will clean up eventually
2631          */
2632         return optimal;
2633 }
2634
2635 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2636                              u64 logical, u64 *length,
2637                              struct btrfs_multi_bio **multi_ret,
2638                              int mirror_num, struct page *unplug_page)
2639 {
2640         struct extent_map *em;
2641         struct map_lookup *map;
2642         struct extent_map_tree *em_tree = &map_tree->map_tree;
2643         u64 offset;
2644         u64 stripe_offset;
2645         u64 stripe_nr;
2646         int stripes_allocated = 8;
2647         int stripes_required = 1;
2648         int stripe_index;
2649         int i;
2650         int num_stripes;
2651         int max_errors = 0;
2652         struct btrfs_multi_bio *multi = NULL;
2653
2654         if (multi_ret && !(rw & REQ_WRITE))
2655                 stripes_allocated = 1;
2656 again:
2657         if (multi_ret) {
2658                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2659                                 GFP_NOFS);
2660                 if (!multi)
2661                         return -ENOMEM;
2662
2663                 atomic_set(&multi->error, 0);
2664         }
2665
2666         read_lock(&em_tree->lock);
2667         em = lookup_extent_mapping(em_tree, logical, *length);
2668         read_unlock(&em_tree->lock);
2669
2670         if (!em && unplug_page) {
2671                 kfree(multi);
2672                 return 0;
2673         }
2674
2675         if (!em) {
2676                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2677                        (unsigned long long)logical,
2678                        (unsigned long long)*length);
2679                 BUG();
2680         }
2681
2682         BUG_ON(em->start > logical || em->start + em->len < logical);
2683         map = (struct map_lookup *)em->bdev;
2684         offset = logical - em->start;
2685
2686         if (mirror_num > map->num_stripes)
2687                 mirror_num = 0;
2688
2689         /* if our multi bio struct is too small, back off and try again */
2690         if (rw & REQ_WRITE) {
2691                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2692                                  BTRFS_BLOCK_GROUP_DUP)) {
2693                         stripes_required = map->num_stripes;
2694                         max_errors = 1;
2695                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2696                         stripes_required = map->sub_stripes;
2697                         max_errors = 1;
2698                 }
2699         }
2700         if (multi_ret && (rw & REQ_WRITE) &&
2701             stripes_allocated < stripes_required) {
2702                 stripes_allocated = map->num_stripes;
2703                 free_extent_map(em);
2704                 kfree(multi);
2705                 goto again;
2706         }
2707         stripe_nr = offset;
2708         /*
2709          * stripe_nr counts the total number of stripes we have to stride
2710          * to get to this block
2711          */
2712         do_div(stripe_nr, map->stripe_len);
2713
2714         stripe_offset = stripe_nr * map->stripe_len;
2715         BUG_ON(offset < stripe_offset);
2716
2717         /* stripe_offset is the offset of this block in its stripe*/
2718         stripe_offset = offset - stripe_offset;
2719
2720         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2721                          BTRFS_BLOCK_GROUP_RAID10 |
2722                          BTRFS_BLOCK_GROUP_DUP)) {
2723                 /* we limit the length of each bio to what fits in a stripe */
2724                 *length = min_t(u64, em->len - offset,
2725                               map->stripe_len - stripe_offset);
2726         } else {
2727                 *length = em->len - offset;
2728         }
2729
2730         if (!multi_ret && !unplug_page)
2731                 goto out;
2732
2733         num_stripes = 1;
2734         stripe_index = 0;
2735         if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2736                 if (unplug_page || (rw & REQ_WRITE))
2737                         num_stripes = map->num_stripes;
2738                 else if (mirror_num)
2739                         stripe_index = mirror_num - 1;
2740                 else {
2741                         stripe_index = find_live_mirror(map, 0,
2742                                             map->num_stripes,
2743                                             current->pid % map->num_stripes);
2744                 }
2745
2746         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2747                 if (rw & REQ_WRITE)
2748                         num_stripes = map->num_stripes;
2749                 else if (mirror_num)
2750                         stripe_index = mirror_num - 1;
2751
2752         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2753                 int factor = map->num_stripes / map->sub_stripes;
2754
2755                 stripe_index = do_div(stripe_nr, factor);
2756                 stripe_index *= map->sub_stripes;
2757
2758                 if (unplug_page || (rw & REQ_WRITE))
2759                         num_stripes = map->sub_stripes;
2760                 else if (mirror_num)
2761                         stripe_index += mirror_num - 1;
2762                 else {
2763                         stripe_index = find_live_mirror(map, stripe_index,
2764                                               map->sub_stripes, stripe_index +
2765                                               current->pid % map->sub_stripes);
2766                 }
2767         } else {
2768                 /*
2769                  * after this do_div call, stripe_nr is the number of stripes
2770                  * on this device we have to walk to find the data, and
2771                  * stripe_index is the number of our device in the stripe array
2772                  */
2773                 stripe_index = do_div(stripe_nr, map->num_stripes);
2774         }
2775         BUG_ON(stripe_index >= map->num_stripes);
2776
2777         for (i = 0; i < num_stripes; i++) {
2778                 if (unplug_page) {
2779                         struct btrfs_device *device;
2780                         struct backing_dev_info *bdi;
2781
2782                         device = map->stripes[stripe_index].dev;
2783                         if (device->bdev) {
2784                                 bdi = blk_get_backing_dev_info(device->bdev);
2785                                 if (bdi->unplug_io_fn)
2786                                         bdi->unplug_io_fn(bdi, unplug_page);
2787                         }
2788                 } else {
2789                         multi->stripes[i].physical =
2790                                 map->stripes[stripe_index].physical +
2791                                 stripe_offset + stripe_nr * map->stripe_len;
2792                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
2793                 }
2794                 stripe_index++;
2795         }
2796         if (multi_ret) {
2797                 *multi_ret = multi;
2798                 multi->num_stripes = num_stripes;
2799                 multi->max_errors = max_errors;
2800         }
2801 out:
2802         free_extent_map(em);
2803         return 0;
2804 }
2805
2806 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2807                       u64 logical, u64 *length,
2808                       struct btrfs_multi_bio **multi_ret, int mirror_num)
2809 {
2810         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2811                                  mirror_num, NULL);
2812 }
2813
2814 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2815                      u64 chunk_start, u64 physical, u64 devid,
2816                      u64 **logical, int *naddrs, int *stripe_len)
2817 {
2818         struct extent_map_tree *em_tree = &map_tree->map_tree;
2819         struct extent_map *em;
2820         struct map_lookup *map;
2821         u64 *buf;
2822         u64 bytenr;
2823         u64 length;
2824         u64 stripe_nr;
2825         int i, j, nr = 0;
2826
2827         read_lock(&em_tree->lock);
2828         em = lookup_extent_mapping(em_tree, chunk_start, 1);
2829         read_unlock(&em_tree->lock);
2830
2831         BUG_ON(!em || em->start != chunk_start);
2832         map = (struct map_lookup *)em->bdev;
2833
2834         length = em->len;
2835         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2836                 do_div(length, map->num_stripes / map->sub_stripes);
2837         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2838                 do_div(length, map->num_stripes);
2839
2840         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2841         BUG_ON(!buf);
2842
2843         for (i = 0; i < map->num_stripes; i++) {
2844                 if (devid && map->stripes[i].dev->devid != devid)
2845                         continue;
2846                 if (map->stripes[i].physical > physical ||
2847                     map->stripes[i].physical + length <= physical)
2848                         continue;
2849
2850                 stripe_nr = physical - map->stripes[i].physical;
2851                 do_div(stripe_nr, map->stripe_len);
2852
2853                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2854                         stripe_nr = stripe_nr * map->num_stripes + i;
2855                         do_div(stripe_nr, map->sub_stripes);
2856                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2857                         stripe_nr = stripe_nr * map->num_stripes + i;
2858                 }
2859                 bytenr = chunk_start + stripe_nr * map->stripe_len;
2860                 WARN_ON(nr >= map->num_stripes);
2861                 for (j = 0; j < nr; j++) {
2862                         if (buf[j] == bytenr)
2863                                 break;
2864                 }
2865                 if (j == nr) {
2866                         WARN_ON(nr >= map->num_stripes);
2867                         buf[nr++] = bytenr;
2868                 }
2869         }
2870
2871         *logical = buf;
2872         *naddrs = nr;
2873         *stripe_len = map->stripe_len;
2874
2875         free_extent_map(em);
2876         return 0;
2877 }
2878
2879 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2880                       u64 logical, struct page *page)
2881 {
2882         u64 length = PAGE_CACHE_SIZE;
2883         return __btrfs_map_block(map_tree, READ, logical, &length,
2884                                  NULL, 0, page);
2885 }
2886
2887 static void end_bio_multi_stripe(struct bio *bio, int err)
2888 {
2889         struct btrfs_multi_bio *multi = bio->bi_private;
2890         int is_orig_bio = 0;
2891
2892         if (err)
2893                 atomic_inc(&multi->error);
2894
2895         if (bio == multi->orig_bio)
2896                 is_orig_bio = 1;
2897
2898         if (atomic_dec_and_test(&multi->stripes_pending)) {
2899                 if (!is_orig_bio) {
2900                         bio_put(bio);
2901                         bio = multi->orig_bio;
2902                 }
2903                 bio->bi_private = multi->private;
2904                 bio->bi_end_io = multi->end_io;
2905                 /* only send an error to the higher layers if it is
2906                  * beyond the tolerance of the multi-bio
2907                  */
2908                 if (atomic_read(&multi->error) > multi->max_errors) {
2909                         err = -EIO;
2910                 } else if (err) {
2911                         /*
2912                          * this bio is actually up to date, we didn't
2913                          * go over the max number of errors
2914                          */
2915                         set_bit(BIO_UPTODATE, &bio->bi_flags);
2916                         err = 0;
2917                 }
2918                 kfree(multi);
2919
2920                 bio_endio(bio, err);
2921         } else if (!is_orig_bio) {
2922                 bio_put(bio);
2923         }
2924 }
2925
2926 struct async_sched {
2927         struct bio *bio;
2928         int rw;
2929         struct btrfs_fs_info *info;
2930         struct btrfs_work work;
2931 };
2932
2933 /*
2934  * see run_scheduled_bios for a description of why bios are collected for
2935  * async submit.
2936  *
2937  * This will add one bio to the pending list for a device and make sure
2938  * the work struct is scheduled.
2939  */
2940 static noinline int schedule_bio(struct btrfs_root *root,
2941                                  struct btrfs_device *device,
2942                                  int rw, struct bio *bio)
2943 {
2944         int should_queue = 1;
2945         struct btrfs_pending_bios *pending_bios;
2946
2947         /* don't bother with additional async steps for reads, right now */
2948         if (!(rw & REQ_WRITE)) {
2949                 bio_get(bio);
2950                 submit_bio(rw, bio);
2951                 bio_put(bio);
2952                 return 0;
2953         }
2954
2955         /*
2956          * nr_async_bios allows us to reliably return congestion to the
2957          * higher layers.  Otherwise, the async bio makes it appear we have
2958          * made progress against dirty pages when we've really just put it
2959          * on a queue for later
2960          */
2961         atomic_inc(&root->fs_info->nr_async_bios);
2962         WARN_ON(bio->bi_next);
2963         bio->bi_next = NULL;
2964         bio->bi_rw |= rw;
2965
2966         spin_lock(&device->io_lock);
2967         if (bio->bi_rw & REQ_SYNC)
2968                 pending_bios = &device->pending_sync_bios;
2969         else
2970                 pending_bios = &device->pending_bios;
2971
2972         if (pending_bios->tail)
2973                 pending_bios->tail->bi_next = bio;
2974
2975         pending_bios->tail = bio;
2976         if (!pending_bios->head)
2977                 pending_bios->head = bio;
2978         if (device->running_pending)
2979                 should_queue = 0;
2980
2981         spin_unlock(&device->io_lock);
2982
2983         if (should_queue)
2984                 btrfs_queue_worker(&root->fs_info->submit_workers,
2985                                    &device->work);
2986         return 0;
2987 }
2988
2989 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2990                   int mirror_num, int async_submit)
2991 {
2992         struct btrfs_mapping_tree *map_tree;
2993         struct btrfs_device *dev;
2994         struct bio *first_bio = bio;
2995         u64 logical = (u64)bio->bi_sector << 9;
2996         u64 length = 0;
2997         u64 map_length;
2998         struct btrfs_multi_bio *multi = NULL;
2999         int ret;
3000         int dev_nr = 0;
3001         int total_devs = 1;
3002
3003         length = bio->bi_size;
3004         map_tree = &root->fs_info->mapping_tree;
3005         map_length = length;
3006
3007         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3008                               mirror_num);
3009         BUG_ON(ret);
3010
3011         total_devs = multi->num_stripes;
3012         if (map_length < length) {
3013                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3014                        "len %llu\n", (unsigned long long)logical,
3015                        (unsigned long long)length,
3016                        (unsigned long long)map_length);
3017                 BUG();
3018         }
3019         multi->end_io = first_bio->bi_end_io;
3020         multi->private = first_bio->bi_private;
3021         multi->orig_bio = first_bio;
3022         atomic_set(&multi->stripes_pending, multi->num_stripes);
3023
3024         while (dev_nr < total_devs) {
3025                 if (total_devs > 1) {
3026                         if (dev_nr < total_devs - 1) {
3027                                 bio = bio_clone(first_bio, GFP_NOFS);
3028                                 BUG_ON(!bio);
3029                         } else {
3030                                 bio = first_bio;
3031                         }
3032                         bio->bi_private = multi;
3033                         bio->bi_end_io = end_bio_multi_stripe;
3034                 }
3035                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3036                 dev = multi->stripes[dev_nr].dev;
3037                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3038                         bio->bi_bdev = dev->bdev;
3039                         if (async_submit)
3040                                 schedule_bio(root, dev, rw, bio);
3041                         else
3042                                 submit_bio(rw, bio);
3043                 } else {
3044                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3045                         bio->bi_sector = logical >> 9;
3046                         bio_endio(bio, -EIO);
3047                 }
3048                 dev_nr++;
3049         }
3050         if (total_devs == 1)
3051                 kfree(multi);
3052         return 0;
3053 }
3054
3055 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3056                                        u8 *uuid, u8 *fsid)
3057 {
3058         struct btrfs_device *device;
3059         struct btrfs_fs_devices *cur_devices;
3060
3061         cur_devices = root->fs_info->fs_devices;
3062         while (cur_devices) {
3063                 if (!fsid ||
3064                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3065                         device = __find_device(&cur_devices->devices,
3066                                                devid, uuid);
3067                         if (device)
3068                                 return device;
3069                 }
3070                 cur_devices = cur_devices->seed;
3071         }
3072         return NULL;
3073 }
3074
3075 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3076                                             u64 devid, u8 *dev_uuid)
3077 {
3078         struct btrfs_device *device;
3079         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3080
3081         device = kzalloc(sizeof(*device), GFP_NOFS);
3082         if (!device)
3083                 return NULL;
3084         list_add(&device->dev_list,
3085                  &fs_devices->devices);
3086         device->dev_root = root->fs_info->dev_root;
3087         device->devid = devid;
3088         device->work.func = pending_bios_fn;
3089         device->fs_devices = fs_devices;
3090         device->missing = 1;
3091         fs_devices->num_devices++;
3092         fs_devices->missing_devices++;
3093         spin_lock_init(&device->io_lock);
3094         INIT_LIST_HEAD(&device->dev_alloc_list);
3095         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3096         return device;
3097 }
3098
3099 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3100                           struct extent_buffer *leaf,
3101                           struct btrfs_chunk *chunk)
3102 {
3103         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3104         struct map_lookup *map;
3105         struct extent_map *em;
3106         u64 logical;
3107         u64 length;
3108         u64 devid;
3109         u8 uuid[BTRFS_UUID_SIZE];
3110         int num_stripes;
3111         int ret;
3112         int i;
3113
3114         logical = key->offset;
3115         length = btrfs_chunk_length(leaf, chunk);
3116
3117         read_lock(&map_tree->map_tree.lock);
3118         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3119         read_unlock(&map_tree->map_tree.lock);
3120
3121         /* already mapped? */
3122         if (em && em->start <= logical && em->start + em->len > logical) {
3123                 free_extent_map(em);
3124                 return 0;
3125         } else if (em) {
3126                 free_extent_map(em);
3127         }
3128
3129         em = alloc_extent_map(GFP_NOFS);
3130         if (!em)
3131                 return -ENOMEM;
3132         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3133         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3134         if (!map) {
3135                 free_extent_map(em);
3136                 return -ENOMEM;
3137         }
3138
3139         em->bdev = (struct block_device *)map;
3140         em->start = logical;
3141         em->len = length;
3142         em->block_start = 0;
3143         em->block_len = em->len;
3144
3145         map->num_stripes = num_stripes;
3146         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3147         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3148         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3149         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3150         map->type = btrfs_chunk_type(leaf, chunk);
3151         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3152         for (i = 0; i < num_stripes; i++) {
3153                 map->stripes[i].physical =
3154                         btrfs_stripe_offset_nr(leaf, chunk, i);
3155                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3156                 read_extent_buffer(leaf, uuid, (unsigned long)
3157                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3158                                    BTRFS_UUID_SIZE);
3159                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3160                                                         NULL);
3161                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3162                         kfree(map);
3163                         free_extent_map(em);
3164                         return -EIO;
3165                 }
3166                 if (!map->stripes[i].dev) {
3167                         map->stripes[i].dev =
3168                                 add_missing_dev(root, devid, uuid);
3169                         if (!map->stripes[i].dev) {
3170                                 kfree(map);
3171                                 free_extent_map(em);
3172                                 return -EIO;
3173                         }
3174                 }
3175                 map->stripes[i].dev->in_fs_metadata = 1;
3176         }
3177
3178         write_lock(&map_tree->map_tree.lock);
3179         ret = add_extent_mapping(&map_tree->map_tree, em);
3180         write_unlock(&map_tree->map_tree.lock);
3181         BUG_ON(ret);
3182         free_extent_map(em);
3183
3184         return 0;
3185 }
3186
3187 static int fill_device_from_item(struct extent_buffer *leaf,
3188                                  struct btrfs_dev_item *dev_item,
3189                                  struct btrfs_device *device)
3190 {
3191         unsigned long ptr;
3192
3193         device->devid = btrfs_device_id(leaf, dev_item);
3194         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3195         device->total_bytes = device->disk_total_bytes;
3196         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3197         device->type = btrfs_device_type(leaf, dev_item);
3198         device->io_align = btrfs_device_io_align(leaf, dev_item);
3199         device->io_width = btrfs_device_io_width(leaf, dev_item);
3200         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3201
3202         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3203         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3204
3205         return 0;
3206 }
3207
3208 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3209 {
3210         struct btrfs_fs_devices *fs_devices;
3211         int ret;
3212
3213         mutex_lock(&uuid_mutex);
3214
3215         fs_devices = root->fs_info->fs_devices->seed;
3216         while (fs_devices) {
3217                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3218                         ret = 0;
3219                         goto out;
3220                 }
3221                 fs_devices = fs_devices->seed;
3222         }
3223
3224         fs_devices = find_fsid(fsid);
3225         if (!fs_devices) {
3226                 ret = -ENOENT;
3227                 goto out;
3228         }
3229
3230         fs_devices = clone_fs_devices(fs_devices);
3231         if (IS_ERR(fs_devices)) {
3232                 ret = PTR_ERR(fs_devices);
3233                 goto out;
3234         }
3235
3236         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3237                                    root->fs_info->bdev_holder);
3238         if (ret)
3239                 goto out;
3240
3241         if (!fs_devices->seeding) {
3242                 __btrfs_close_devices(fs_devices);
3243                 free_fs_devices(fs_devices);
3244                 ret = -EINVAL;
3245                 goto out;
3246         }
3247
3248         fs_devices->seed = root->fs_info->fs_devices->seed;
3249         root->fs_info->fs_devices->seed = fs_devices;
3250 out:
3251         mutex_unlock(&uuid_mutex);
3252         return ret;
3253 }
3254
3255 static int read_one_dev(struct btrfs_root *root,
3256                         struct extent_buffer *leaf,
3257                         struct btrfs_dev_item *dev_item)
3258 {
3259         struct btrfs_device *device;
3260         u64 devid;
3261         int ret;
3262         u8 fs_uuid[BTRFS_UUID_SIZE];
3263         u8 dev_uuid[BTRFS_UUID_SIZE];
3264
3265         devid = btrfs_device_id(leaf, dev_item);
3266         read_extent_buffer(leaf, dev_uuid,
3267                            (unsigned long)btrfs_device_uuid(dev_item),
3268                            BTRFS_UUID_SIZE);
3269         read_extent_buffer(leaf, fs_uuid,
3270                            (unsigned long)btrfs_device_fsid(dev_item),
3271                            BTRFS_UUID_SIZE);
3272
3273         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3274                 ret = open_seed_devices(root, fs_uuid);
3275                 if (ret && !btrfs_test_opt(root, DEGRADED))
3276                         return ret;
3277         }
3278
3279         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3280         if (!device || !device->bdev) {
3281                 if (!btrfs_test_opt(root, DEGRADED))
3282                         return -EIO;
3283
3284                 if (!device) {
3285                         printk(KERN_WARNING "warning devid %llu missing\n",
3286                                (unsigned long long)devid);
3287                         device = add_missing_dev(root, devid, dev_uuid);
3288                         if (!device)
3289                                 return -ENOMEM;
3290                 } else if (!device->missing) {
3291                         /*
3292                          * this happens when a device that was properly setup
3293                          * in the device info lists suddenly goes bad.
3294                          * device->bdev is NULL, and so we have to set
3295                          * device->missing to one here
3296                          */
3297                         root->fs_info->fs_devices->missing_devices++;
3298                         device->missing = 1;
3299                 }
3300         }
3301
3302         if (device->fs_devices != root->fs_info->fs_devices) {
3303                 BUG_ON(device->writeable);
3304                 if (device->generation !=
3305                     btrfs_device_generation(leaf, dev_item))
3306                         return -EINVAL;
3307         }
3308
3309         fill_device_from_item(leaf, dev_item, device);
3310         device->dev_root = root->fs_info->dev_root;
3311         device->in_fs_metadata = 1;
3312         if (device->writeable)
3313                 device->fs_devices->total_rw_bytes += device->total_bytes;
3314         ret = 0;
3315         return ret;
3316 }
3317
3318 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3319 {
3320         struct btrfs_dev_item *dev_item;
3321
3322         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3323                                                      dev_item);
3324         return read_one_dev(root, buf, dev_item);
3325 }
3326
3327 int btrfs_read_sys_array(struct btrfs_root *root)
3328 {
3329         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3330         struct extent_buffer *sb;
3331         struct btrfs_disk_key *disk_key;
3332         struct btrfs_chunk *chunk;
3333         u8 *ptr;
3334         unsigned long sb_ptr;
3335         int ret = 0;