Btrfs: return error code to caller when btrfs_previous_item fails
[linux-2.6.git] / fs / btrfs / volumes.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <asm/div64.h>
27 #include "compat.h"
28 #include "ctree.h"
29 #include "extent_map.h"
30 #include "disk-io.h"
31 #include "transaction.h"
32 #include "print-tree.h"
33 #include "volumes.h"
34 #include "async-thread.h"
35
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37                                 struct btrfs_root *root,
38                                 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42                             (sizeof(struct btrfs_bio_stripe) * (n)))
43
44 static DEFINE_MUTEX(uuid_mutex);
45 static LIST_HEAD(fs_uuids);
46
47 void btrfs_lock_volumes(void)
48 {
49         mutex_lock(&uuid_mutex);
50 }
51
52 void btrfs_unlock_volumes(void)
53 {
54         mutex_unlock(&uuid_mutex);
55 }
56
57 static void lock_chunks(struct btrfs_root *root)
58 {
59         mutex_lock(&root->fs_info->chunk_mutex);
60 }
61
62 static void unlock_chunks(struct btrfs_root *root)
63 {
64         mutex_unlock(&root->fs_info->chunk_mutex);
65 }
66
67 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
68 {
69         struct btrfs_device *device;
70         WARN_ON(fs_devices->opened);
71         while (!list_empty(&fs_devices->devices)) {
72                 device = list_entry(fs_devices->devices.next,
73                                     struct btrfs_device, dev_list);
74                 list_del(&device->dev_list);
75                 kfree(device->name);
76                 kfree(device);
77         }
78         kfree(fs_devices);
79 }
80
81 int btrfs_cleanup_fs_uuids(void)
82 {
83         struct btrfs_fs_devices *fs_devices;
84
85         while (!list_empty(&fs_uuids)) {
86                 fs_devices = list_entry(fs_uuids.next,
87                                         struct btrfs_fs_devices, list);
88                 list_del(&fs_devices->list);
89                 free_fs_devices(fs_devices);
90         }
91         return 0;
92 }
93
94 static noinline struct btrfs_device *__find_device(struct list_head *head,
95                                                    u64 devid, u8 *uuid)
96 {
97         struct btrfs_device *dev;
98
99         list_for_each_entry(dev, head, dev_list) {
100                 if (dev->devid == devid &&
101                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
102                         return dev;
103                 }
104         }
105         return NULL;
106 }
107
108 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
109 {
110         struct btrfs_fs_devices *fs_devices;
111
112         list_for_each_entry(fs_devices, &fs_uuids, list) {
113                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
114                         return fs_devices;
115         }
116         return NULL;
117 }
118
119 static void requeue_list(struct btrfs_pending_bios *pending_bios,
120                         struct bio *head, struct bio *tail)
121 {
122
123         struct bio *old_head;
124
125         old_head = pending_bios->head;
126         pending_bios->head = head;
127         if (pending_bios->tail)
128                 tail->bi_next = old_head;
129         else
130                 pending_bios->tail = tail;
131 }
132
133 /*
134  * we try to collect pending bios for a device so we don't get a large
135  * number of procs sending bios down to the same device.  This greatly
136  * improves the schedulers ability to collect and merge the bios.
137  *
138  * But, it also turns into a long list of bios to process and that is sure
139  * to eventually make the worker thread block.  The solution here is to
140  * make some progress and then put this work struct back at the end of
141  * the list if the block device is congested.  This way, multiple devices
142  * can make progress from a single worker thread.
143  */
144 static noinline int run_scheduled_bios(struct btrfs_device *device)
145 {
146         struct bio *pending;
147         struct backing_dev_info *bdi;
148         struct btrfs_fs_info *fs_info;
149         struct btrfs_pending_bios *pending_bios;
150         struct bio *tail;
151         struct bio *cur;
152         int again = 0;
153         unsigned long num_run;
154         unsigned long batch_run = 0;
155         unsigned long limit;
156         unsigned long last_waited = 0;
157         int force_reg = 0;
158         struct blk_plug plug;
159
160         /*
161          * this function runs all the bios we've collected for
162          * a particular device.  We don't want to wander off to
163          * another device without first sending all of these down.
164          * So, setup a plug here and finish it off before we return
165          */
166         blk_start_plug(&plug);
167
168         bdi = blk_get_backing_dev_info(device->bdev);
169         fs_info = device->dev_root->fs_info;
170         limit = btrfs_async_submit_limit(fs_info);
171         limit = limit * 2 / 3;
172
173 loop:
174         spin_lock(&device->io_lock);
175
176 loop_lock:
177         num_run = 0;
178
179         /* take all the bios off the list at once and process them
180          * later on (without the lock held).  But, remember the
181          * tail and other pointers so the bios can be properly reinserted
182          * into the list if we hit congestion
183          */
184         if (!force_reg && device->pending_sync_bios.head) {
185                 pending_bios = &device->pending_sync_bios;
186                 force_reg = 1;
187         } else {
188                 pending_bios = &device->pending_bios;
189                 force_reg = 0;
190         }
191
192         pending = pending_bios->head;
193         tail = pending_bios->tail;
194         WARN_ON(pending && !tail);
195
196         /*
197          * if pending was null this time around, no bios need processing
198          * at all and we can stop.  Otherwise it'll loop back up again
199          * and do an additional check so no bios are missed.
200          *
201          * device->running_pending is used to synchronize with the
202          * schedule_bio code.
203          */
204         if (device->pending_sync_bios.head == NULL &&
205             device->pending_bios.head == NULL) {
206                 again = 0;
207                 device->running_pending = 0;
208         } else {
209                 again = 1;
210                 device->running_pending = 1;
211         }
212
213         pending_bios->head = NULL;
214         pending_bios->tail = NULL;
215
216         spin_unlock(&device->io_lock);
217
218         while (pending) {
219
220                 rmb();
221                 /* we want to work on both lists, but do more bios on the
222                  * sync list than the regular list
223                  */
224                 if ((num_run > 32 &&
225                     pending_bios != &device->pending_sync_bios &&
226                     device->pending_sync_bios.head) ||
227                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
228                     device->pending_bios.head)) {
229                         spin_lock(&device->io_lock);
230                         requeue_list(pending_bios, pending, tail);
231                         goto loop_lock;
232                 }
233
234                 cur = pending;
235                 pending = pending->bi_next;
236                 cur->bi_next = NULL;
237                 atomic_dec(&fs_info->nr_async_bios);
238
239                 if (atomic_read(&fs_info->nr_async_bios) < limit &&
240                     waitqueue_active(&fs_info->async_submit_wait))
241                         wake_up(&fs_info->async_submit_wait);
242
243                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
244
245                 submit_bio(cur->bi_rw, cur);
246                 num_run++;
247                 batch_run++;
248                 if (need_resched())
249                         cond_resched();
250
251                 /*
252                  * we made progress, there is more work to do and the bdi
253                  * is now congested.  Back off and let other work structs
254                  * run instead
255                  */
256                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
257                     fs_info->fs_devices->open_devices > 1) {
258                         struct io_context *ioc;
259
260                         ioc = current->io_context;
261
262                         /*
263                          * the main goal here is that we don't want to
264                          * block if we're going to be able to submit
265                          * more requests without blocking.
266                          *
267                          * This code does two great things, it pokes into
268                          * the elevator code from a filesystem _and_
269                          * it makes assumptions about how batching works.
270                          */
271                         if (ioc && ioc->nr_batch_requests > 0 &&
272                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
273                             (last_waited == 0 ||
274                              ioc->last_waited == last_waited)) {
275                                 /*
276                                  * we want to go through our batch of
277                                  * requests and stop.  So, we copy out
278                                  * the ioc->last_waited time and test
279                                  * against it before looping
280                                  */
281                                 last_waited = ioc->last_waited;
282                                 if (need_resched())
283                                         cond_resched();
284                                 continue;
285                         }
286                         spin_lock(&device->io_lock);
287                         requeue_list(pending_bios, pending, tail);
288                         device->running_pending = 1;
289
290                         spin_unlock(&device->io_lock);
291                         btrfs_requeue_work(&device->work);
292                         goto done;
293                 }
294         }
295
296         cond_resched();
297         if (again)
298                 goto loop;
299
300         spin_lock(&device->io_lock);
301         if (device->pending_bios.head || device->pending_sync_bios.head)
302                 goto loop_lock;
303         spin_unlock(&device->io_lock);
304
305 done:
306         blk_finish_plug(&plug);
307         return 0;
308 }
309
310 static void pending_bios_fn(struct btrfs_work *work)
311 {
312         struct btrfs_device *device;
313
314         device = container_of(work, struct btrfs_device, work);
315         run_scheduled_bios(device);
316 }
317
318 static noinline int device_list_add(const char *path,
319                            struct btrfs_super_block *disk_super,
320                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
321 {
322         struct btrfs_device *device;
323         struct btrfs_fs_devices *fs_devices;
324         u64 found_transid = btrfs_super_generation(disk_super);
325         char *name;
326
327         fs_devices = find_fsid(disk_super->fsid);
328         if (!fs_devices) {
329                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
330                 if (!fs_devices)
331                         return -ENOMEM;
332                 INIT_LIST_HEAD(&fs_devices->devices);
333                 INIT_LIST_HEAD(&fs_devices->alloc_list);
334                 list_add(&fs_devices->list, &fs_uuids);
335                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
336                 fs_devices->latest_devid = devid;
337                 fs_devices->latest_trans = found_transid;
338                 mutex_init(&fs_devices->device_list_mutex);
339                 device = NULL;
340         } else {
341                 device = __find_device(&fs_devices->devices, devid,
342                                        disk_super->dev_item.uuid);
343         }
344         if (!device) {
345                 if (fs_devices->opened)
346                         return -EBUSY;
347
348                 device = kzalloc(sizeof(*device), GFP_NOFS);
349                 if (!device) {
350                         /* we can safely leave the fs_devices entry around */
351                         return -ENOMEM;
352                 }
353                 device->devid = devid;
354                 device->work.func = pending_bios_fn;
355                 memcpy(device->uuid, disk_super->dev_item.uuid,
356                        BTRFS_UUID_SIZE);
357                 spin_lock_init(&device->io_lock);
358                 device->name = kstrdup(path, GFP_NOFS);
359                 if (!device->name) {
360                         kfree(device);
361                         return -ENOMEM;
362                 }
363                 INIT_LIST_HEAD(&device->dev_alloc_list);
364
365                 mutex_lock(&fs_devices->device_list_mutex);
366                 list_add(&device->dev_list, &fs_devices->devices);
367                 mutex_unlock(&fs_devices->device_list_mutex);
368
369                 device->fs_devices = fs_devices;
370                 fs_devices->num_devices++;
371         } else if (!device->name || strcmp(device->name, path)) {
372                 name = kstrdup(path, GFP_NOFS);
373                 if (!name)
374                         return -ENOMEM;
375                 kfree(device->name);
376                 device->name = name;
377                 if (device->missing) {
378                         fs_devices->missing_devices--;
379                         device->missing = 0;
380                 }
381         }
382
383         if (found_transid > fs_devices->latest_trans) {
384                 fs_devices->latest_devid = devid;
385                 fs_devices->latest_trans = found_transid;
386         }
387         *fs_devices_ret = fs_devices;
388         return 0;
389 }
390
391 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
392 {
393         struct btrfs_fs_devices *fs_devices;
394         struct btrfs_device *device;
395         struct btrfs_device *orig_dev;
396
397         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
398         if (!fs_devices)
399                 return ERR_PTR(-ENOMEM);
400
401         INIT_LIST_HEAD(&fs_devices->devices);
402         INIT_LIST_HEAD(&fs_devices->alloc_list);
403         INIT_LIST_HEAD(&fs_devices->list);
404         mutex_init(&fs_devices->device_list_mutex);
405         fs_devices->latest_devid = orig->latest_devid;
406         fs_devices->latest_trans = orig->latest_trans;
407         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
408
409         mutex_lock(&orig->device_list_mutex);
410         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
411                 device = kzalloc(sizeof(*device), GFP_NOFS);
412                 if (!device)
413                         goto error;
414
415                 device->name = kstrdup(orig_dev->name, GFP_NOFS);
416                 if (!device->name) {
417                         kfree(device);
418                         goto error;
419                 }
420
421                 device->devid = orig_dev->devid;
422                 device->work.func = pending_bios_fn;
423                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
424                 spin_lock_init(&device->io_lock);
425                 INIT_LIST_HEAD(&device->dev_list);
426                 INIT_LIST_HEAD(&device->dev_alloc_list);
427
428                 list_add(&device->dev_list, &fs_devices->devices);
429                 device->fs_devices = fs_devices;
430                 fs_devices->num_devices++;
431         }
432         mutex_unlock(&orig->device_list_mutex);
433         return fs_devices;
434 error:
435         mutex_unlock(&orig->device_list_mutex);
436         free_fs_devices(fs_devices);
437         return ERR_PTR(-ENOMEM);
438 }
439
440 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
441 {
442         struct btrfs_device *device, *next;
443
444         mutex_lock(&uuid_mutex);
445 again:
446         mutex_lock(&fs_devices->device_list_mutex);
447         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
448                 if (device->in_fs_metadata)
449                         continue;
450
451                 if (device->bdev) {
452                         blkdev_put(device->bdev, device->mode);
453                         device->bdev = NULL;
454                         fs_devices->open_devices--;
455                 }
456                 if (device->writeable) {
457                         list_del_init(&device->dev_alloc_list);
458                         device->writeable = 0;
459                         fs_devices->rw_devices--;
460                 }
461                 list_del_init(&device->dev_list);
462                 fs_devices->num_devices--;
463                 kfree(device->name);
464                 kfree(device);
465         }
466         mutex_unlock(&fs_devices->device_list_mutex);
467
468         if (fs_devices->seed) {
469                 fs_devices = fs_devices->seed;
470                 goto again;
471         }
472
473         mutex_unlock(&uuid_mutex);
474         return 0;
475 }
476
477 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
478 {
479         struct btrfs_device *device;
480
481         if (--fs_devices->opened > 0)
482                 return 0;
483
484         list_for_each_entry(device, &fs_devices->devices, dev_list) {
485                 if (device->bdev) {
486                         blkdev_put(device->bdev, device->mode);
487                         fs_devices->open_devices--;
488                 }
489                 if (device->writeable) {
490                         list_del_init(&device->dev_alloc_list);
491                         fs_devices->rw_devices--;
492                 }
493
494                 device->bdev = NULL;
495                 device->writeable = 0;
496                 device->in_fs_metadata = 0;
497         }
498         WARN_ON(fs_devices->open_devices);
499         WARN_ON(fs_devices->rw_devices);
500         fs_devices->opened = 0;
501         fs_devices->seeding = 0;
502
503         return 0;
504 }
505
506 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
507 {
508         struct btrfs_fs_devices *seed_devices = NULL;
509         int ret;
510
511         mutex_lock(&uuid_mutex);
512         ret = __btrfs_close_devices(fs_devices);
513         if (!fs_devices->opened) {
514                 seed_devices = fs_devices->seed;
515                 fs_devices->seed = NULL;
516         }
517         mutex_unlock(&uuid_mutex);
518
519         while (seed_devices) {
520                 fs_devices = seed_devices;
521                 seed_devices = fs_devices->seed;
522                 __btrfs_close_devices(fs_devices);
523                 free_fs_devices(fs_devices);
524         }
525         return ret;
526 }
527
528 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
529                                 fmode_t flags, void *holder)
530 {
531         struct block_device *bdev;
532         struct list_head *head = &fs_devices->devices;
533         struct btrfs_device *device;
534         struct block_device *latest_bdev = NULL;
535         struct buffer_head *bh;
536         struct btrfs_super_block *disk_super;
537         u64 latest_devid = 0;
538         u64 latest_transid = 0;
539         u64 devid;
540         int seeding = 1;
541         int ret = 0;
542
543         flags |= FMODE_EXCL;
544
545         list_for_each_entry(device, head, dev_list) {
546                 if (device->bdev)
547                         continue;
548                 if (!device->name)
549                         continue;
550
551                 bdev = blkdev_get_by_path(device->name, flags, holder);
552                 if (IS_ERR(bdev)) {
553                         printk(KERN_INFO "open %s failed\n", device->name);
554                         goto error;
555                 }
556                 set_blocksize(bdev, 4096);
557
558                 bh = btrfs_read_dev_super(bdev);
559                 if (!bh) {
560                         ret = -EINVAL;
561                         goto error_close;
562                 }
563
564                 disk_super = (struct btrfs_super_block *)bh->b_data;
565                 devid = btrfs_stack_device_id(&disk_super->dev_item);
566                 if (devid != device->devid)
567                         goto error_brelse;
568
569                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
570                            BTRFS_UUID_SIZE))
571                         goto error_brelse;
572
573                 device->generation = btrfs_super_generation(disk_super);
574                 if (!latest_transid || device->generation > latest_transid) {
575                         latest_devid = devid;
576                         latest_transid = device->generation;
577                         latest_bdev = bdev;
578                 }
579
580                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
581                         device->writeable = 0;
582                 } else {
583                         device->writeable = !bdev_read_only(bdev);
584                         seeding = 0;
585                 }
586
587                 device->bdev = bdev;
588                 device->in_fs_metadata = 0;
589                 device->mode = flags;
590
591                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
592                         fs_devices->rotating = 1;
593
594                 fs_devices->open_devices++;
595                 if (device->writeable) {
596                         fs_devices->rw_devices++;
597                         list_add(&device->dev_alloc_list,
598                                  &fs_devices->alloc_list);
599                 }
600                 continue;
601
602 error_brelse:
603                 brelse(bh);
604 error_close:
605                 blkdev_put(bdev, flags);
606 error:
607                 continue;
608         }
609         if (fs_devices->open_devices == 0) {
610                 ret = -EIO;
611                 goto out;
612         }
613         fs_devices->seeding = seeding;
614         fs_devices->opened = 1;
615         fs_devices->latest_bdev = latest_bdev;
616         fs_devices->latest_devid = latest_devid;
617         fs_devices->latest_trans = latest_transid;
618         fs_devices->total_rw_bytes = 0;
619 out:
620         return ret;
621 }
622
623 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
624                        fmode_t flags, void *holder)
625 {
626         int ret;
627
628         mutex_lock(&uuid_mutex);
629         if (fs_devices->opened) {
630                 fs_devices->opened++;
631                 ret = 0;
632         } else {
633                 ret = __btrfs_open_devices(fs_devices, flags, holder);
634         }
635         mutex_unlock(&uuid_mutex);
636         return ret;
637 }
638
639 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
640                           struct btrfs_fs_devices **fs_devices_ret)
641 {
642         struct btrfs_super_block *disk_super;
643         struct block_device *bdev;
644         struct buffer_head *bh;
645         int ret;
646         u64 devid;
647         u64 transid;
648
649         mutex_lock(&uuid_mutex);
650
651         flags |= FMODE_EXCL;
652         bdev = blkdev_get_by_path(path, flags, holder);
653
654         if (IS_ERR(bdev)) {
655                 ret = PTR_ERR(bdev);
656                 goto error;
657         }
658
659         ret = set_blocksize(bdev, 4096);
660         if (ret)
661                 goto error_close;
662         bh = btrfs_read_dev_super(bdev);
663         if (!bh) {
664                 ret = -EINVAL;
665                 goto error_close;
666         }
667         disk_super = (struct btrfs_super_block *)bh->b_data;
668         devid = btrfs_stack_device_id(&disk_super->dev_item);
669         transid = btrfs_super_generation(disk_super);
670         if (disk_super->label[0])
671                 printk(KERN_INFO "device label %s ", disk_super->label);
672         else {
673                 /* FIXME, make a readl uuid parser */
674                 printk(KERN_INFO "device fsid %llx-%llx ",
675                        *(unsigned long long *)disk_super->fsid,
676                        *(unsigned long long *)(disk_super->fsid + 8));
677         }
678         printk(KERN_CONT "devid %llu transid %llu %s\n",
679                (unsigned long long)devid, (unsigned long long)transid, path);
680         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
681
682         brelse(bh);
683 error_close:
684         blkdev_put(bdev, flags);
685 error:
686         mutex_unlock(&uuid_mutex);
687         return ret;
688 }
689
690 /* helper to account the used device space in the range */
691 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
692                                    u64 end, u64 *length)
693 {
694         struct btrfs_key key;
695         struct btrfs_root *root = device->dev_root;
696         struct btrfs_dev_extent *dev_extent;
697         struct btrfs_path *path;
698         u64 extent_end;
699         int ret;
700         int slot;
701         struct extent_buffer *l;
702
703         *length = 0;
704
705         if (start >= device->total_bytes)
706                 return 0;
707
708         path = btrfs_alloc_path();
709         if (!path)
710                 return -ENOMEM;
711         path->reada = 2;
712
713         key.objectid = device->devid;
714         key.offset = start;
715         key.type = BTRFS_DEV_EXTENT_KEY;
716
717         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
718         if (ret < 0)
719                 goto out;
720         if (ret > 0) {
721                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
722                 if (ret < 0)
723                         goto out;
724         }
725
726         while (1) {
727                 l = path->nodes[0];
728                 slot = path->slots[0];
729                 if (slot >= btrfs_header_nritems(l)) {
730                         ret = btrfs_next_leaf(root, path);
731                         if (ret == 0)
732                                 continue;
733                         if (ret < 0)
734                                 goto out;
735
736                         break;
737                 }
738                 btrfs_item_key_to_cpu(l, &key, slot);
739
740                 if (key.objectid < device->devid)
741                         goto next;
742
743                 if (key.objectid > device->devid)
744                         break;
745
746                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
747                         goto next;
748
749                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
750                 extent_end = key.offset + btrfs_dev_extent_length(l,
751                                                                   dev_extent);
752                 if (key.offset <= start && extent_end > end) {
753                         *length = end - start + 1;
754                         break;
755                 } else if (key.offset <= start && extent_end > start)
756                         *length += extent_end - start;
757                 else if (key.offset > start && extent_end <= end)
758                         *length += extent_end - key.offset;
759                 else if (key.offset > start && key.offset <= end) {
760                         *length += end - key.offset + 1;
761                         break;
762                 } else if (key.offset > end)
763                         break;
764
765 next:
766                 path->slots[0]++;
767         }
768         ret = 0;
769 out:
770         btrfs_free_path(path);
771         return ret;
772 }
773
774 /*
775  * find_free_dev_extent - find free space in the specified device
776  * @trans:      transaction handler
777  * @device:     the device which we search the free space in
778  * @num_bytes:  the size of the free space that we need
779  * @start:      store the start of the free space.
780  * @len:        the size of the free space. that we find, or the size of the max
781  *              free space if we don't find suitable free space
782  *
783  * this uses a pretty simple search, the expectation is that it is
784  * called very infrequently and that a given device has a small number
785  * of extents
786  *
787  * @start is used to store the start of the free space if we find. But if we
788  * don't find suitable free space, it will be used to store the start position
789  * of the max free space.
790  *
791  * @len is used to store the size of the free space that we find.
792  * But if we don't find suitable free space, it is used to store the size of
793  * the max free space.
794  */
795 int find_free_dev_extent(struct btrfs_trans_handle *trans,
796                          struct btrfs_device *device, u64 num_bytes,
797                          u64 *start, u64 *len)
798 {
799         struct btrfs_key key;
800         struct btrfs_root *root = device->dev_root;
801         struct btrfs_dev_extent *dev_extent;
802         struct btrfs_path *path;
803         u64 hole_size;
804         u64 max_hole_start;
805         u64 max_hole_size;
806         u64 extent_end;
807         u64 search_start;
808         u64 search_end = device->total_bytes;
809         int ret;
810         int slot;
811         struct extent_buffer *l;
812
813         /* FIXME use last free of some kind */
814
815         /* we don't want to overwrite the superblock on the drive,
816          * so we make sure to start at an offset of at least 1MB
817          */
818         search_start = 1024 * 1024;
819
820         if (root->fs_info->alloc_start + num_bytes <= search_end)
821                 search_start = max(root->fs_info->alloc_start, search_start);
822
823         max_hole_start = search_start;
824         max_hole_size = 0;
825
826         if (search_start >= search_end) {
827                 ret = -ENOSPC;
828                 goto error;
829         }
830
831         path = btrfs_alloc_path();
832         if (!path) {
833                 ret = -ENOMEM;
834                 goto error;
835         }
836         path->reada = 2;
837
838         key.objectid = device->devid;
839         key.offset = search_start;
840         key.type = BTRFS_DEV_EXTENT_KEY;
841
842         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
843         if (ret < 0)
844                 goto out;
845         if (ret > 0) {
846                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
847                 if (ret < 0)
848                         goto out;
849         }
850
851         while (1) {
852                 l = path->nodes[0];
853                 slot = path->slots[0];
854                 if (slot >= btrfs_header_nritems(l)) {
855                         ret = btrfs_next_leaf(root, path);
856                         if (ret == 0)
857                                 continue;
858                         if (ret < 0)
859                                 goto out;
860
861                         break;
862                 }
863                 btrfs_item_key_to_cpu(l, &key, slot);
864
865                 if (key.objectid < device->devid)
866                         goto next;
867
868                 if (key.objectid > device->devid)
869                         break;
870
871                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
872                         goto next;
873
874                 if (key.offset > search_start) {
875                         hole_size = key.offset - search_start;
876
877                         if (hole_size > max_hole_size) {
878                                 max_hole_start = search_start;
879                                 max_hole_size = hole_size;
880                         }
881
882                         /*
883                          * If this free space is greater than which we need,
884                          * it must be the max free space that we have found
885                          * until now, so max_hole_start must point to the start
886                          * of this free space and the length of this free space
887                          * is stored in max_hole_size. Thus, we return
888                          * max_hole_start and max_hole_size and go back to the
889                          * caller.
890                          */
891                         if (hole_size >= num_bytes) {
892                                 ret = 0;
893                                 goto out;
894                         }
895                 }
896
897                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
898                 extent_end = key.offset + btrfs_dev_extent_length(l,
899                                                                   dev_extent);
900                 if (extent_end > search_start)
901                         search_start = extent_end;
902 next:
903                 path->slots[0]++;
904                 cond_resched();
905         }
906
907         hole_size = search_end- search_start;
908         if (hole_size > max_hole_size) {
909                 max_hole_start = search_start;
910                 max_hole_size = hole_size;
911         }
912
913         /* See above. */
914         if (hole_size < num_bytes)
915                 ret = -ENOSPC;
916         else
917                 ret = 0;
918
919 out:
920         btrfs_free_path(path);
921 error:
922         *start = max_hole_start;
923         if (len)
924                 *len = max_hole_size;
925         return ret;
926 }
927
928 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
929                           struct btrfs_device *device,
930                           u64 start)
931 {
932         int ret;
933         struct btrfs_path *path;
934         struct btrfs_root *root = device->dev_root;
935         struct btrfs_key key;
936         struct btrfs_key found_key;
937         struct extent_buffer *leaf = NULL;
938         struct btrfs_dev_extent *extent = NULL;
939
940         path = btrfs_alloc_path();
941         if (!path)
942                 return -ENOMEM;
943
944         key.objectid = device->devid;
945         key.offset = start;
946         key.type = BTRFS_DEV_EXTENT_KEY;
947
948         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
949         if (ret > 0) {
950                 ret = btrfs_previous_item(root, path, key.objectid,
951                                           BTRFS_DEV_EXTENT_KEY);
952                 if (ret)
953                         goto out;
954                 leaf = path->nodes[0];
955                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
956                 extent = btrfs_item_ptr(leaf, path->slots[0],
957                                         struct btrfs_dev_extent);
958                 BUG_ON(found_key.offset > start || found_key.offset +
959                        btrfs_dev_extent_length(leaf, extent) < start);
960         } else if (ret == 0) {
961                 leaf = path->nodes[0];
962                 extent = btrfs_item_ptr(leaf, path->slots[0],
963                                         struct btrfs_dev_extent);
964         }
965         BUG_ON(ret);
966
967         if (device->bytes_used > 0)
968                 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
969         ret = btrfs_del_item(trans, root, path);
970         BUG_ON(ret);
971
972 out:
973         btrfs_free_path(path);
974         return ret;
975 }
976
977 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
978                            struct btrfs_device *device,
979                            u64 chunk_tree, u64 chunk_objectid,
980                            u64 chunk_offset, u64 start, u64 num_bytes)
981 {
982         int ret;
983         struct btrfs_path *path;
984         struct btrfs_root *root = device->dev_root;
985         struct btrfs_dev_extent *extent;
986         struct extent_buffer *leaf;
987         struct btrfs_key key;
988
989         WARN_ON(!device->in_fs_metadata);
990         path = btrfs_alloc_path();
991         if (!path)
992                 return -ENOMEM;
993
994         key.objectid = device->devid;
995         key.offset = start;
996         key.type = BTRFS_DEV_EXTENT_KEY;
997         ret = btrfs_insert_empty_item(trans, root, path, &key,
998                                       sizeof(*extent));
999         BUG_ON(ret);
1000
1001         leaf = path->nodes[0];
1002         extent = btrfs_item_ptr(leaf, path->slots[0],
1003                                 struct btrfs_dev_extent);
1004         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1005         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1006         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1007
1008         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1009                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1010                     BTRFS_UUID_SIZE);
1011
1012         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1013         btrfs_mark_buffer_dirty(leaf);
1014         btrfs_free_path(path);
1015         return ret;
1016 }
1017
1018 static noinline int find_next_chunk(struct btrfs_root *root,
1019                                     u64 objectid, u64 *offset)
1020 {
1021         struct btrfs_path *path;
1022         int ret;
1023         struct btrfs_key key;
1024         struct btrfs_chunk *chunk;
1025         struct btrfs_key found_key;
1026
1027         path = btrfs_alloc_path();
1028         BUG_ON(!path);
1029
1030         key.objectid = objectid;
1031         key.offset = (u64)-1;
1032         key.type = BTRFS_CHUNK_ITEM_KEY;
1033
1034         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1035         if (ret < 0)
1036                 goto error;
1037
1038         BUG_ON(ret == 0);
1039
1040         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1041         if (ret) {
1042                 *offset = 0;
1043         } else {
1044                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1045                                       path->slots[0]);
1046                 if (found_key.objectid != objectid)
1047                         *offset = 0;
1048                 else {
1049                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1050                                                struct btrfs_chunk);
1051                         *offset = found_key.offset +
1052                                 btrfs_chunk_length(path->nodes[0], chunk);
1053                 }
1054         }
1055         ret = 0;
1056 error:
1057         btrfs_free_path(path);
1058         return ret;
1059 }
1060
1061 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1062 {
1063         int ret;
1064         struct btrfs_key key;
1065         struct btrfs_key found_key;
1066         struct btrfs_path *path;
1067
1068         root = root->fs_info->chunk_root;
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1075         key.type = BTRFS_DEV_ITEM_KEY;
1076         key.offset = (u64)-1;
1077
1078         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1079         if (ret < 0)
1080                 goto error;
1081
1082         BUG_ON(ret == 0);
1083
1084         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1085                                   BTRFS_DEV_ITEM_KEY);
1086         if (ret) {
1087                 *objectid = 1;
1088         } else {
1089                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1090                                       path->slots[0]);
1091                 *objectid = found_key.offset + 1;
1092         }
1093         ret = 0;
1094 error:
1095         btrfs_free_path(path);
1096         return ret;
1097 }
1098
1099 /*
1100  * the device information is stored in the chunk root
1101  * the btrfs_device struct should be fully filled in
1102  */
1103 int btrfs_add_device(struct btrfs_trans_handle *trans,
1104                      struct btrfs_root *root,
1105                      struct btrfs_device *device)
1106 {
1107         int ret;
1108         struct btrfs_path *path;
1109         struct btrfs_dev_item *dev_item;
1110         struct extent_buffer *leaf;
1111         struct btrfs_key key;
1112         unsigned long ptr;
1113
1114         root = root->fs_info->chunk_root;
1115
1116         path = btrfs_alloc_path();
1117         if (!path)
1118                 return -ENOMEM;
1119
1120         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1121         key.type = BTRFS_DEV_ITEM_KEY;
1122         key.offset = device->devid;
1123
1124         ret = btrfs_insert_empty_item(trans, root, path, &key,
1125                                       sizeof(*dev_item));
1126         if (ret)
1127                 goto out;
1128
1129         leaf = path->nodes[0];
1130         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1131
1132         btrfs_set_device_id(leaf, dev_item, device->devid);
1133         btrfs_set_device_generation(leaf, dev_item, 0);
1134         btrfs_set_device_type(leaf, dev_item, device->type);
1135         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1136         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1137         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1138         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1139         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1140         btrfs_set_device_group(leaf, dev_item, 0);
1141         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1142         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1143         btrfs_set_device_start_offset(leaf, dev_item, 0);
1144
1145         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1146         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1147         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1148         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1149         btrfs_mark_buffer_dirty(leaf);
1150
1151         ret = 0;
1152 out:
1153         btrfs_free_path(path);
1154         return ret;
1155 }
1156
1157 static int btrfs_rm_dev_item(struct btrfs_root *root,
1158                              struct btrfs_device *device)
1159 {
1160         int ret;
1161         struct btrfs_path *path;
1162         struct btrfs_key key;
1163         struct btrfs_trans_handle *trans;
1164
1165         root = root->fs_info->chunk_root;
1166
1167         path = btrfs_alloc_path();
1168         if (!path)
1169                 return -ENOMEM;
1170
1171         trans = btrfs_start_transaction(root, 0);
1172         if (IS_ERR(trans)) {
1173                 btrfs_free_path(path);
1174                 return PTR_ERR(trans);
1175         }
1176         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1177         key.type = BTRFS_DEV_ITEM_KEY;
1178         key.offset = device->devid;
1179         lock_chunks(root);
1180
1181         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1182         if (ret < 0)
1183                 goto out;
1184
1185         if (ret > 0) {
1186                 ret = -ENOENT;
1187                 goto out;
1188         }
1189
1190         ret = btrfs_del_item(trans, root, path);
1191         if (ret)
1192                 goto out;
1193 out:
1194         btrfs_free_path(path);
1195         unlock_chunks(root);
1196         btrfs_commit_transaction(trans, root);
1197         return ret;
1198 }
1199
1200 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1201 {
1202         struct btrfs_device *device;
1203         struct btrfs_device *next_device;
1204         struct block_device *bdev;
1205         struct buffer_head *bh = NULL;
1206         struct btrfs_super_block *disk_super;
1207         u64 all_avail;
1208         u64 devid;
1209         u64 num_devices;
1210         u8 *dev_uuid;
1211         int ret = 0;
1212
1213         mutex_lock(&uuid_mutex);
1214         mutex_lock(&root->fs_info->volume_mutex);
1215
1216         all_avail = root->fs_info->avail_data_alloc_bits |
1217                 root->fs_info->avail_system_alloc_bits |
1218                 root->fs_info->avail_metadata_alloc_bits;
1219
1220         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1221             root->fs_info->fs_devices->num_devices <= 4) {
1222                 printk(KERN_ERR "btrfs: unable to go below four devices "
1223                        "on raid10\n");
1224                 ret = -EINVAL;
1225                 goto out;
1226         }
1227
1228         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1229             root->fs_info->fs_devices->num_devices <= 2) {
1230                 printk(KERN_ERR "btrfs: unable to go below two "
1231                        "devices on raid1\n");
1232                 ret = -EINVAL;
1233                 goto out;
1234         }
1235
1236         if (strcmp(device_path, "missing") == 0) {
1237                 struct list_head *devices;
1238                 struct btrfs_device *tmp;
1239
1240                 device = NULL;
1241                 devices = &root->fs_info->fs_devices->devices;
1242                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1243                 list_for_each_entry(tmp, devices, dev_list) {
1244                         if (tmp->in_fs_metadata && !tmp->bdev) {
1245                                 device = tmp;
1246                                 break;
1247                         }
1248                 }
1249                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1250                 bdev = NULL;
1251                 bh = NULL;
1252                 disk_super = NULL;
1253                 if (!device) {
1254                         printk(KERN_ERR "btrfs: no missing devices found to "
1255                                "remove\n");
1256                         goto out;
1257                 }
1258         } else {
1259                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1260                                           root->fs_info->bdev_holder);
1261                 if (IS_ERR(bdev)) {
1262                         ret = PTR_ERR(bdev);
1263                         goto out;
1264                 }
1265
1266                 set_blocksize(bdev, 4096);
1267                 bh = btrfs_read_dev_super(bdev);
1268                 if (!bh) {
1269                         ret = -EINVAL;
1270                         goto error_close;
1271                 }
1272                 disk_super = (struct btrfs_super_block *)bh->b_data;
1273                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1274                 dev_uuid = disk_super->dev_item.uuid;
1275                 device = btrfs_find_device(root, devid, dev_uuid,
1276                                            disk_super->fsid);
1277                 if (!device) {
1278                         ret = -ENOENT;
1279                         goto error_brelse;
1280                 }
1281         }
1282
1283         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1284                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1285                        "device\n");
1286                 ret = -EINVAL;
1287                 goto error_brelse;
1288         }
1289
1290         if (device->writeable) {
1291                 list_del_init(&device->dev_alloc_list);
1292                 root->fs_info->fs_devices->rw_devices--;
1293         }
1294
1295         ret = btrfs_shrink_device(device, 0);
1296         if (ret)
1297                 goto error_undo;
1298
1299         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1300         if (ret)
1301                 goto error_undo;
1302
1303         device->in_fs_metadata = 0;
1304
1305         /*
1306          * the device list mutex makes sure that we don't change
1307          * the device list while someone else is writing out all
1308          * the device supers.
1309          */
1310         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1311         list_del_init(&device->dev_list);
1312         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1313
1314         device->fs_devices->num_devices--;
1315
1316         if (device->missing)
1317                 root->fs_info->fs_devices->missing_devices--;
1318
1319         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1320                                  struct btrfs_device, dev_list);
1321         if (device->bdev == root->fs_info->sb->s_bdev)
1322                 root->fs_info->sb->s_bdev = next_device->bdev;
1323         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1324                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1325
1326         if (device->bdev) {
1327                 blkdev_put(device->bdev, device->mode);
1328                 device->bdev = NULL;
1329                 device->fs_devices->open_devices--;
1330         }
1331
1332         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1333         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1334
1335         if (device->fs_devices->open_devices == 0) {
1336                 struct btrfs_fs_devices *fs_devices;
1337                 fs_devices = root->fs_info->fs_devices;
1338                 while (fs_devices) {
1339                         if (fs_devices->seed == device->fs_devices)
1340                                 break;
1341                         fs_devices = fs_devices->seed;
1342                 }
1343                 fs_devices->seed = device->fs_devices->seed;
1344                 device->fs_devices->seed = NULL;
1345                 __btrfs_close_devices(device->fs_devices);
1346                 free_fs_devices(device->fs_devices);
1347         }
1348
1349         /*
1350          * at this point, the device is zero sized.  We want to
1351          * remove it from the devices list and zero out the old super
1352          */
1353         if (device->writeable) {
1354                 /* make sure this device isn't detected as part of
1355                  * the FS anymore
1356                  */
1357                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1358                 set_buffer_dirty(bh);
1359                 sync_dirty_buffer(bh);
1360         }
1361
1362         kfree(device->name);
1363         kfree(device);
1364         ret = 0;
1365
1366 error_brelse:
1367         brelse(bh);
1368 error_close:
1369         if (bdev)
1370                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1371 out:
1372         mutex_unlock(&root->fs_info->volume_mutex);
1373         mutex_unlock(&uuid_mutex);
1374         return ret;
1375 error_undo:
1376         if (device->writeable) {
1377                 list_add(&device->dev_alloc_list,
1378                          &root->fs_info->fs_devices->alloc_list);
1379                 root->fs_info->fs_devices->rw_devices++;
1380         }
1381         goto error_brelse;
1382 }
1383
1384 /*
1385  * does all the dirty work required for changing file system's UUID.
1386  */
1387 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1388                                 struct btrfs_root *root)
1389 {
1390         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1391         struct btrfs_fs_devices *old_devices;
1392         struct btrfs_fs_devices *seed_devices;
1393         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1394         struct btrfs_device *device;
1395         u64 super_flags;
1396
1397         BUG_ON(!mutex_is_locked(&uuid_mutex));
1398         if (!fs_devices->seeding)
1399                 return -EINVAL;
1400
1401         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1402         if (!seed_devices)
1403                 return -ENOMEM;
1404
1405         old_devices = clone_fs_devices(fs_devices);
1406         if (IS_ERR(old_devices)) {
1407                 kfree(seed_devices);
1408                 return PTR_ERR(old_devices);
1409         }
1410
1411         list_add(&old_devices->list, &fs_uuids);
1412
1413         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1414         seed_devices->opened = 1;
1415         INIT_LIST_HEAD(&seed_devices->devices);
1416         INIT_LIST_HEAD(&seed_devices->alloc_list);
1417         mutex_init(&seed_devices->device_list_mutex);
1418         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1419         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1420         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1421                 device->fs_devices = seed_devices;
1422         }
1423
1424         fs_devices->seeding = 0;
1425         fs_devices->num_devices = 0;
1426         fs_devices->open_devices = 0;
1427         fs_devices->seed = seed_devices;
1428
1429         generate_random_uuid(fs_devices->fsid);
1430         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1431         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1432         super_flags = btrfs_super_flags(disk_super) &
1433                       ~BTRFS_SUPER_FLAG_SEEDING;
1434         btrfs_set_super_flags(disk_super, super_flags);
1435
1436         return 0;
1437 }
1438
1439 /*
1440  * strore the expected generation for seed devices in device items.
1441  */
1442 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1443                                struct btrfs_root *root)
1444 {
1445         struct btrfs_path *path;
1446         struct extent_buffer *leaf;
1447         struct btrfs_dev_item *dev_item;
1448         struct btrfs_device *device;
1449         struct btrfs_key key;
1450         u8 fs_uuid[BTRFS_UUID_SIZE];
1451         u8 dev_uuid[BTRFS_UUID_SIZE];
1452         u64 devid;
1453         int ret;
1454
1455         path = btrfs_alloc_path();
1456         if (!path)
1457                 return -ENOMEM;
1458
1459         root = root->fs_info->chunk_root;
1460         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1461         key.offset = 0;
1462         key.type = BTRFS_DEV_ITEM_KEY;
1463
1464         while (1) {
1465                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1466                 if (ret < 0)
1467                         goto error;
1468
1469                 leaf = path->nodes[0];
1470 next_slot:
1471                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1472                         ret = btrfs_next_leaf(root, path);
1473                         if (ret > 0)
1474                                 break;
1475                         if (ret < 0)
1476                                 goto error;
1477                         leaf = path->nodes[0];
1478                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1479                         btrfs_release_path(root, path);
1480                         continue;
1481                 }
1482
1483                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1484                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1485                     key.type != BTRFS_DEV_ITEM_KEY)
1486                         break;
1487
1488                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1489                                           struct btrfs_dev_item);
1490                 devid = btrfs_device_id(leaf, dev_item);
1491                 read_extent_buffer(leaf, dev_uuid,
1492                                    (unsigned long)btrfs_device_uuid(dev_item),
1493                                    BTRFS_UUID_SIZE);
1494                 read_extent_buffer(leaf, fs_uuid,
1495                                    (unsigned long)btrfs_device_fsid(dev_item),
1496                                    BTRFS_UUID_SIZE);
1497                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1498                 BUG_ON(!device);
1499
1500                 if (device->fs_devices->seeding) {
1501                         btrfs_set_device_generation(leaf, dev_item,
1502                                                     device->generation);
1503                         btrfs_mark_buffer_dirty(leaf);
1504                 }
1505
1506                 path->slots[0]++;
1507                 goto next_slot;
1508         }
1509         ret = 0;
1510 error:
1511         btrfs_free_path(path);
1512         return ret;
1513 }
1514
1515 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1516 {
1517         struct btrfs_trans_handle *trans;
1518         struct btrfs_device *device;
1519         struct block_device *bdev;
1520         struct list_head *devices;
1521         struct super_block *sb = root->fs_info->sb;
1522         u64 total_bytes;
1523         int seeding_dev = 0;
1524         int ret = 0;
1525
1526         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1527                 return -EINVAL;
1528
1529         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1530                                   root->fs_info->bdev_holder);
1531         if (IS_ERR(bdev))
1532                 return PTR_ERR(bdev);
1533
1534         if (root->fs_info->fs_devices->seeding) {
1535                 seeding_dev = 1;
1536                 down_write(&sb->s_umount);
1537                 mutex_lock(&uuid_mutex);
1538         }
1539
1540         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1541         mutex_lock(&root->fs_info->volume_mutex);
1542
1543         devices = &root->fs_info->fs_devices->devices;
1544         /*
1545          * we have the volume lock, so we don't need the extra
1546          * device list mutex while reading the list here.
1547          */
1548         list_for_each_entry(device, devices, dev_list) {
1549                 if (device->bdev == bdev) {
1550                         ret = -EEXIST;
1551                         goto error;
1552                 }
1553         }
1554
1555         device = kzalloc(sizeof(*device), GFP_NOFS);
1556         if (!device) {
1557                 /* we can safely leave the fs_devices entry around */
1558                 ret = -ENOMEM;
1559                 goto error;
1560         }
1561
1562         device->name = kstrdup(device_path, GFP_NOFS);
1563         if (!device->name) {
1564                 kfree(device);
1565                 ret = -ENOMEM;
1566                 goto error;
1567         }
1568
1569         ret = find_next_devid(root, &device->devid);
1570         if (ret) {
1571                 kfree(device->name);
1572                 kfree(device);
1573                 goto error;
1574         }
1575
1576         trans = btrfs_start_transaction(root, 0);
1577         if (IS_ERR(trans)) {
1578                 kfree(device->name);
1579                 kfree(device);
1580                 ret = PTR_ERR(trans);
1581                 goto error;
1582         }
1583
1584         lock_chunks(root);
1585
1586         device->writeable = 1;
1587         device->work.func = pending_bios_fn;
1588         generate_random_uuid(device->uuid);
1589         spin_lock_init(&device->io_lock);
1590         device->generation = trans->transid;
1591         device->io_width = root->sectorsize;
1592         device->io_align = root->sectorsize;
1593         device->sector_size = root->sectorsize;
1594         device->total_bytes = i_size_read(bdev->bd_inode);
1595         device->disk_total_bytes = device->total_bytes;
1596         device->dev_root = root->fs_info->dev_root;
1597         device->bdev = bdev;
1598         device->in_fs_metadata = 1;
1599         device->mode = FMODE_EXCL;
1600         set_blocksize(device->bdev, 4096);
1601
1602         if (seeding_dev) {
1603                 sb->s_flags &= ~MS_RDONLY;
1604                 ret = btrfs_prepare_sprout(trans, root);
1605                 BUG_ON(ret);
1606         }
1607
1608         device->fs_devices = root->fs_info->fs_devices;
1609
1610         /*
1611          * we don't want write_supers to jump in here with our device
1612          * half setup
1613          */
1614         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1615         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1616         list_add(&device->dev_alloc_list,
1617                  &root->fs_info->fs_devices->alloc_list);
1618         root->fs_info->fs_devices->num_devices++;
1619         root->fs_info->fs_devices->open_devices++;
1620         root->fs_info->fs_devices->rw_devices++;
1621         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1622
1623         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1624                 root->fs_info->fs_devices->rotating = 1;
1625
1626         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1627         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1628                                     total_bytes + device->total_bytes);
1629
1630         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1631         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1632                                     total_bytes + 1);
1633         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1634
1635         if (seeding_dev) {
1636                 ret = init_first_rw_device(trans, root, device);
1637                 BUG_ON(ret);
1638                 ret = btrfs_finish_sprout(trans, root);
1639                 BUG_ON(ret);
1640         } else {
1641                 ret = btrfs_add_device(trans, root, device);
1642         }
1643
1644         /*
1645          * we've got more storage, clear any full flags on the space
1646          * infos
1647          */
1648         btrfs_clear_space_info_full(root->fs_info);
1649
1650         unlock_chunks(root);
1651         btrfs_commit_transaction(trans, root);
1652
1653         if (seeding_dev) {
1654                 mutex_unlock(&uuid_mutex);
1655                 up_write(&sb->s_umount);
1656
1657                 ret = btrfs_relocate_sys_chunks(root);
1658                 BUG_ON(ret);
1659         }
1660 out:
1661         mutex_unlock(&root->fs_info->volume_mutex);
1662         return ret;
1663 error:
1664         blkdev_put(bdev, FMODE_EXCL);
1665         if (seeding_dev) {
1666                 mutex_unlock(&uuid_mutex);
1667                 up_write(&sb->s_umount);
1668         }
1669         goto out;
1670 }
1671
1672 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1673                                         struct btrfs_device *device)
1674 {
1675         int ret;
1676         struct btrfs_path *path;
1677         struct btrfs_root *root;
1678         struct btrfs_dev_item *dev_item;
1679         struct extent_buffer *leaf;
1680         struct btrfs_key key;
1681
1682         root = device->dev_root->fs_info->chunk_root;
1683
1684         path = btrfs_alloc_path();
1685         if (!path)
1686                 return -ENOMEM;
1687
1688         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1689         key.type = BTRFS_DEV_ITEM_KEY;
1690         key.offset = device->devid;
1691
1692         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1693         if (ret < 0)
1694                 goto out;
1695
1696         if (ret > 0) {
1697                 ret = -ENOENT;
1698                 goto out;
1699         }
1700
1701         leaf = path->nodes[0];
1702         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1703
1704         btrfs_set_device_id(leaf, dev_item, device->devid);
1705         btrfs_set_device_type(leaf, dev_item, device->type);
1706         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1707         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1708         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1709         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1710         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1711         btrfs_mark_buffer_dirty(leaf);
1712
1713 out:
1714         btrfs_free_path(path);
1715         return ret;
1716 }
1717
1718 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1719                       struct btrfs_device *device, u64 new_size)
1720 {
1721         struct btrfs_super_block *super_copy =
1722                 &device->dev_root->fs_info->super_copy;
1723         u64 old_total = btrfs_super_total_bytes(super_copy);
1724         u64 diff = new_size - device->total_bytes;
1725
1726         if (!device->writeable)
1727                 return -EACCES;
1728         if (new_size <= device->total_bytes)
1729                 return -EINVAL;
1730
1731         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1732         device->fs_devices->total_rw_bytes += diff;
1733
1734         device->total_bytes = new_size;
1735         device->disk_total_bytes = new_size;
1736         btrfs_clear_space_info_full(device->dev_root->fs_info);
1737
1738         return btrfs_update_device(trans, device);
1739 }
1740
1741 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1742                       struct btrfs_device *device, u64 new_size)
1743 {
1744         int ret;
1745         lock_chunks(device->dev_root);
1746         ret = __btrfs_grow_device(trans, device, new_size);
1747         unlock_chunks(device->dev_root);
1748         return ret;
1749 }
1750
1751 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1752                             struct btrfs_root *root,
1753                             u64 chunk_tree, u64 chunk_objectid,
1754                             u64 chunk_offset)
1755 {
1756         int ret;
1757         struct btrfs_path *path;
1758         struct btrfs_key key;
1759
1760         root = root->fs_info->chunk_root;
1761         path = btrfs_alloc_path();
1762         if (!path)
1763                 return -ENOMEM;
1764
1765         key.objectid = chunk_objectid;
1766         key.offset = chunk_offset;
1767         key.type = BTRFS_CHUNK_ITEM_KEY;
1768
1769         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1770         BUG_ON(ret);
1771
1772         ret = btrfs_del_item(trans, root, path);
1773         BUG_ON(ret);
1774
1775         btrfs_free_path(path);
1776         return 0;
1777 }
1778
1779 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1780                         chunk_offset)
1781 {
1782         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1783         struct btrfs_disk_key *disk_key;
1784         struct btrfs_chunk *chunk;
1785         u8 *ptr;
1786         int ret = 0;
1787         u32 num_stripes;
1788         u32 array_size;
1789         u32 len = 0;
1790         u32 cur;
1791         struct btrfs_key key;
1792
1793         array_size = btrfs_super_sys_array_size(super_copy);
1794
1795         ptr = super_copy->sys_chunk_array;
1796         cur = 0;
1797
1798         while (cur < array_size) {
1799                 disk_key = (struct btrfs_disk_key *)ptr;
1800                 btrfs_disk_key_to_cpu(&key, disk_key);
1801
1802                 len = sizeof(*disk_key);
1803
1804                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1805                         chunk = (struct btrfs_chunk *)(ptr + len);
1806                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1807                         len += btrfs_chunk_item_size(num_stripes);
1808                 } else {
1809                         ret = -EIO;
1810                         break;
1811                 }
1812                 if (key.objectid == chunk_objectid &&
1813                     key.offset == chunk_offset) {
1814                         memmove(ptr, ptr + len, array_size - (cur + len));
1815                         array_size -= len;
1816                         btrfs_set_super_sys_array_size(super_copy, array_size);
1817                 } else {
1818                         ptr += len;
1819                         cur += len;
1820                 }
1821         }
1822         return ret;
1823 }
1824
1825 static int btrfs_relocate_chunk(struct btrfs_root *root,
1826                          u64 chunk_tree, u64 chunk_objectid,
1827                          u64 chunk_offset)
1828 {
1829         struct extent_map_tree *em_tree;
1830         struct btrfs_root *extent_root;
1831         struct btrfs_trans_handle *trans;
1832         struct extent_map *em;
1833         struct map_lookup *map;
1834         int ret;
1835         int i;
1836
1837         root = root->fs_info->chunk_root;
1838         extent_root = root->fs_info->extent_root;
1839         em_tree = &root->fs_info->mapping_tree.map_tree;
1840
1841         ret = btrfs_can_relocate(extent_root, chunk_offset);
1842         if (ret)
1843                 return -ENOSPC;
1844
1845         /* step one, relocate all the extents inside this chunk */
1846         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1847         if (ret)
1848                 return ret;
1849
1850         trans = btrfs_start_transaction(root, 0);
1851         BUG_ON(IS_ERR(trans));
1852
1853         lock_chunks(root);
1854
1855         /*
1856          * step two, delete the device extents and the
1857          * chunk tree entries
1858          */
1859         read_lock(&em_tree->lock);
1860         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1861         read_unlock(&em_tree->lock);
1862
1863         BUG_ON(em->start > chunk_offset ||
1864                em->start + em->len < chunk_offset);
1865         map = (struct map_lookup *)em->bdev;
1866
1867         for (i = 0; i < map->num_stripes; i++) {
1868                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1869                                             map->stripes[i].physical);
1870                 BUG_ON(ret);
1871
1872                 if (map->stripes[i].dev) {
1873                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1874                         BUG_ON(ret);
1875                 }
1876         }
1877         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1878                                chunk_offset);
1879
1880         BUG_ON(ret);
1881
1882         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1883
1884         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1885                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1886                 BUG_ON(ret);
1887         }
1888
1889         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1890         BUG_ON(ret);
1891
1892         write_lock(&em_tree->lock);
1893         remove_extent_mapping(em_tree, em);
1894         write_unlock(&em_tree->lock);
1895
1896         kfree(map);
1897         em->bdev = NULL;
1898
1899         /* once for the tree */
1900         free_extent_map(em);
1901         /* once for us */
1902         free_extent_map(em);
1903
1904         unlock_chunks(root);
1905         btrfs_end_transaction(trans, root);
1906         return 0;
1907 }
1908
1909 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1910 {
1911         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1912         struct btrfs_path *path;
1913         struct extent_buffer *leaf;
1914         struct btrfs_chunk *chunk;
1915         struct btrfs_key key;
1916         struct btrfs_key found_key;
1917         u64 chunk_tree = chunk_root->root_key.objectid;
1918         u64 chunk_type;
1919         bool retried = false;
1920         int failed = 0;
1921         int ret;
1922
1923         path = btrfs_alloc_path();
1924         if (!path)
1925                 return -ENOMEM;
1926
1927 again:
1928         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1929         key.offset = (u64)-1;
1930         key.type = BTRFS_CHUNK_ITEM_KEY;
1931
1932         while (1) {
1933                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1934                 if (ret < 0)
1935                         goto error;
1936                 BUG_ON(ret == 0);
1937
1938                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1939                                           key.type);
1940                 if (ret < 0)
1941                         goto error;
1942                 if (ret > 0)
1943                         break;
1944
1945                 leaf = path->nodes[0];
1946                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1947
1948                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1949                                        struct btrfs_chunk);
1950                 chunk_type = btrfs_chunk_type(leaf, chunk);
1951                 btrfs_release_path(chunk_root, path);
1952
1953                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1954                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1955                                                    found_key.objectid,
1956                                                    found_key.offset);
1957                         if (ret == -ENOSPC)
1958                                 failed++;
1959                         else if (ret)
1960                                 BUG();
1961                 }
1962
1963                 if (found_key.offset == 0)
1964                         break;
1965                 key.offset = found_key.offset - 1;
1966         }
1967         ret = 0;
1968         if (failed && !retried) {
1969                 failed = 0;
1970                 retried = true;
1971                 goto again;
1972         } else if (failed && retried) {
1973                 WARN_ON(1);
1974                 ret = -ENOSPC;
1975         }
1976 error:
1977         btrfs_free_path(path);
1978         return ret;
1979 }
1980
1981 static u64 div_factor(u64 num, int factor)
1982 {
1983         if (factor == 10)
1984                 return num;
1985         num *= factor;
1986         do_div(num, 10);
1987         return num;
1988 }
1989
1990 int btrfs_balance(struct btrfs_root *dev_root)
1991 {
1992         int ret;
1993         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1994         struct btrfs_device *device;
1995         u64 old_size;
1996         u64 size_to_free;
1997         struct btrfs_path *path;
1998         struct btrfs_key key;
1999         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2000         struct btrfs_trans_handle *trans;
2001         struct btrfs_key found_key;
2002
2003         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2004                 return -EROFS;
2005
2006         if (!capable(CAP_SYS_ADMIN))
2007                 return -EPERM;
2008
2009         mutex_lock(&dev_root->fs_info->volume_mutex);
2010         dev_root = dev_root->fs_info->dev_root;
2011
2012         /* step one make some room on all the devices */
2013         list_for_each_entry(device, devices, dev_list) {
2014                 old_size = device->total_bytes;
2015                 size_to_free = div_factor(old_size, 1);
2016                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2017                 if (!device->writeable ||
2018                     device->total_bytes - device->bytes_used > size_to_free)
2019                         continue;
2020
2021                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2022                 if (ret == -ENOSPC)
2023                         break;
2024                 BUG_ON(ret);
2025
2026                 trans = btrfs_start_transaction(dev_root, 0);
2027                 BUG_ON(IS_ERR(trans));
2028
2029                 ret = btrfs_grow_device(trans, device, old_size);
2030                 BUG_ON(ret);
2031
2032                 btrfs_end_transaction(trans, dev_root);
2033         }
2034
2035         /* step two, relocate all the chunks */
2036         path = btrfs_alloc_path();
2037         BUG_ON(!path);
2038
2039         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2040         key.offset = (u64)-1;
2041         key.type = BTRFS_CHUNK_ITEM_KEY;
2042
2043         while (1) {
2044                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2045                 if (ret < 0)
2046                         goto error;
2047
2048                 /*
2049                  * this shouldn't happen, it means the last relocate
2050                  * failed
2051                  */
2052                 if (ret == 0)
2053                         break;
2054
2055                 ret = btrfs_previous_item(chunk_root, path, 0,
2056                                           BTRFS_CHUNK_ITEM_KEY);
2057                 if (ret)
2058                         break;
2059
2060                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2061                                       path->slots[0]);
2062                 if (found_key.objectid != key.objectid)
2063                         break;
2064
2065                 /* chunk zero is special */
2066                 if (found_key.offset == 0)
2067                         break;
2068
2069                 btrfs_release_path(chunk_root, path);
2070                 ret = btrfs_relocate_chunk(chunk_root,
2071                                            chunk_root->root_key.objectid,
2072                                            found_key.objectid,
2073                                            found_key.offset);
2074                 BUG_ON(ret && ret != -ENOSPC);
2075                 key.offset = found_key.offset - 1;
2076         }
2077         ret = 0;
2078 error:
2079         btrfs_free_path(path);
2080         mutex_unlock(&dev_root->fs_info->volume_mutex);
2081         return ret;
2082 }
2083
2084 /*
2085  * shrinking a device means finding all of the device extents past
2086  * the new size, and then following the back refs to the chunks.
2087  * The chunk relocation code actually frees the device extent
2088  */
2089 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2090 {
2091         struct btrfs_trans_handle *trans;
2092         struct btrfs_root *root = device->dev_root;
2093         struct btrfs_dev_extent *dev_extent = NULL;
2094         struct btrfs_path *path;
2095         u64 length;
2096         u64 chunk_tree;
2097         u64 chunk_objectid;
2098         u64 chunk_offset;
2099         int ret;
2100         int slot;
2101         int failed = 0;
2102         bool retried = false;
2103         struct extent_buffer *l;
2104         struct btrfs_key key;
2105         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2106         u64 old_total = btrfs_super_total_bytes(super_copy);
2107         u64 old_size = device->total_bytes;
2108         u64 diff = device->total_bytes - new_size;
2109
2110         if (new_size >= device->total_bytes)
2111                 return -EINVAL;
2112
2113         path = btrfs_alloc_path();
2114         if (!path)
2115                 return -ENOMEM;
2116
2117         path->reada = 2;
2118
2119         lock_chunks(root);
2120
2121         device->total_bytes = new_size;
2122         if (device->writeable)
2123                 device->fs_devices->total_rw_bytes -= diff;
2124         unlock_chunks(root);
2125
2126 again:
2127         key.objectid = device->devid;
2128         key.offset = (u64)-1;
2129         key.type = BTRFS_DEV_EXTENT_KEY;
2130
2131         while (1) {
2132                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2133                 if (ret < 0)
2134                         goto done;
2135
2136                 ret = btrfs_previous_item(root, path, 0, key.type);
2137                 if (ret < 0)
2138                         goto done;
2139                 if (ret) {
2140                         ret = 0;
2141                         btrfs_release_path(root, path);
2142                         break;
2143                 }
2144
2145                 l = path->nodes[0];
2146                 slot = path->slots[0];
2147                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2148
2149                 if (key.objectid != device->devid) {
2150                         btrfs_release_path(root, path);
2151                         break;
2152                 }
2153
2154                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2155                 length = btrfs_dev_extent_length(l, dev_extent);
2156
2157                 if (key.offset + length <= new_size) {
2158                         btrfs_release_path(root, path);
2159                         break;
2160                 }
2161
2162                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2163                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2164                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2165                 btrfs_release_path(root, path);
2166
2167                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2168                                            chunk_offset);
2169                 if (ret && ret != -ENOSPC)
2170                         goto done;
2171                 if (ret == -ENOSPC)
2172                         failed++;
2173                 key.offset -= 1;
2174         }
2175
2176         if (failed && !retried) {
2177                 failed = 0;
2178                 retried = true;
2179                 goto again;
2180         } else if (failed && retried) {
2181                 ret = -ENOSPC;
2182                 lock_chunks(root);
2183
2184                 device->total_bytes = old_size;
2185                 if (device->writeable)
2186                         device->fs_devices->total_rw_bytes += diff;
2187                 unlock_chunks(root);
2188                 goto done;
2189         }
2190
2191         /* Shrinking succeeded, else we would be at "done". */
2192         trans = btrfs_start_transaction(root, 0);
2193         if (IS_ERR(trans)) {
2194                 ret = PTR_ERR(trans);
2195                 goto done;
2196         }
2197
2198         lock_chunks(root);
2199
2200         device->disk_total_bytes = new_size;
2201         /* Now btrfs_update_device() will change the on-disk size. */
2202         ret = btrfs_update_device(trans, device);
2203         if (ret) {
2204                 unlock_chunks(root);
2205                 btrfs_end_transaction(trans, root);
2206                 goto done;
2207         }
2208         WARN_ON(diff > old_total);
2209         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2210         unlock_chunks(root);
2211         btrfs_end_transaction(trans, root);
2212 done:
2213         btrfs_free_path(path);
2214         return ret;
2215 }
2216
2217 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2218                            struct btrfs_root *root,
2219                            struct btrfs_key *key,
2220                            struct btrfs_chunk *chunk, int item_size)
2221 {
2222         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2223         struct btrfs_disk_key disk_key;
2224         u32 array_size;
2225         u8 *ptr;
2226
2227         array_size = btrfs_super_sys_array_size(super_copy);
2228         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2229                 return -EFBIG;
2230
2231         ptr = super_copy->sys_chunk_array + array_size;
2232         btrfs_cpu_key_to_disk(&disk_key, key);
2233         memcpy(ptr, &disk_key, sizeof(disk_key));
2234         ptr += sizeof(disk_key);
2235         memcpy(ptr, chunk, item_size);
2236         item_size += sizeof(disk_key);
2237         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2238         return 0;
2239 }
2240
2241 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2242                                         int num_stripes, int sub_stripes)
2243 {
2244         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2245                 return calc_size;
2246         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2247                 return calc_size * (num_stripes / sub_stripes);
2248         else
2249                 return calc_size * num_stripes;
2250 }
2251
2252 /* Used to sort the devices by max_avail(descending sort) */
2253 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2254 {
2255         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2256             ((struct btrfs_device_info *)dev_info2)->max_avail)
2257                 return -1;
2258         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2259                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2260                 return 1;
2261         else
2262                 return 0;
2263 }
2264
2265 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2266                                  int *num_stripes, int *min_stripes,
2267                                  int *sub_stripes)
2268 {
2269         *num_stripes = 1;
2270         *min_stripes = 1;
2271         *sub_stripes = 0;
2272
2273         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2274                 *num_stripes = fs_devices->rw_devices;
2275                 *min_stripes = 2;
2276         }
2277         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2278                 *num_stripes = 2;
2279                 *min_stripes = 2;
2280         }
2281         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2282                 if (fs_devices->rw_devices < 2)
2283                         return -ENOSPC;
2284                 *num_stripes = 2;
2285                 *min_stripes = 2;
2286         }
2287         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2288                 *num_stripes = fs_devices->rw_devices;
2289                 if (*num_stripes < 4)
2290                         return -ENOSPC;
2291                 *num_stripes &= ~(u32)1;
2292                 *sub_stripes = 2;
2293                 *min_stripes = 4;
2294         }
2295
2296         return 0;
2297 }
2298
2299 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2300                                     u64 proposed_size, u64 type,
2301                                     int num_stripes, int small_stripe)
2302 {
2303         int min_stripe_size = 1 * 1024 * 1024;
2304         u64 calc_size = proposed_size;
2305         u64 max_chunk_size = calc_size;
2306         int ncopies = 1;
2307
2308         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2309                     BTRFS_BLOCK_GROUP_DUP |
2310                     BTRFS_BLOCK_GROUP_RAID10))
2311                 ncopies = 2;
2312
2313         if (type & BTRFS_BLOCK_GROUP_DATA) {
2314                 max_chunk_size = 10 * calc_size;
2315                 min_stripe_size = 64 * 1024 * 1024;
2316         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2317                 max_chunk_size = 256 * 1024 * 1024;
2318                 min_stripe_size = 32 * 1024 * 1024;
2319         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2320                 calc_size = 8 * 1024 * 1024;
2321                 max_chunk_size = calc_size * 2;
2322                 min_stripe_size = 1 * 1024 * 1024;
2323         }
2324
2325         /* we don't want a chunk larger than 10% of writeable space */
2326         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2327                              max_chunk_size);
2328
2329         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2330                 calc_size = max_chunk_size * ncopies;
2331                 do_div(calc_size, num_stripes);
2332                 do_div(calc_size, BTRFS_STRIPE_LEN);
2333                 calc_size *= BTRFS_STRIPE_LEN;
2334         }
2335
2336         /* we don't want tiny stripes */
2337         if (!small_stripe)
2338                 calc_size = max_t(u64, min_stripe_size, calc_size);
2339
2340         /*
2341          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2342          * we end up with something bigger than a stripe
2343          */
2344         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2345
2346         do_div(calc_size, BTRFS_STRIPE_LEN);
2347         calc_size *= BTRFS_STRIPE_LEN;
2348
2349         return calc_size;
2350 }
2351
2352 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2353                                                       int num_stripes)
2354 {
2355         struct map_lookup *new;
2356         size_t len = map_lookup_size(num_stripes);
2357
2358         BUG_ON(map->num_stripes < num_stripes);
2359
2360         if (map->num_stripes == num_stripes)
2361                 return map;
2362
2363         new = kmalloc(len, GFP_NOFS);
2364         if (!new) {
2365                 /* just change map->num_stripes */
2366                 map->num_stripes = num_stripes;
2367                 return map;
2368         }
2369
2370         memcpy(new, map, len);
2371         new->num_stripes = num_stripes;
2372         kfree(map);
2373         return new;
2374 }
2375
2376 /*
2377  * helper to allocate device space from btrfs_device_info, in which we stored
2378  * max free space information of every device. It is used when we can not
2379  * allocate chunks by default size.
2380  *
2381  * By this helper, we can allocate a new chunk as larger as possible.
2382  */
2383 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2384                                     struct btrfs_fs_devices *fs_devices,
2385                                     struct btrfs_device_info *devices,
2386                                     int nr_device, u64 type,
2387                                     struct map_lookup **map_lookup,
2388                                     int min_stripes, u64 *stripe_size)
2389 {
2390         int i, index, sort_again = 0;
2391         int min_devices = min_stripes;
2392         u64 max_avail, min_free;
2393         struct map_lookup *map = *map_lookup;
2394         int ret;
2395
2396         if (nr_device < min_stripes)
2397                 return -ENOSPC;
2398
2399         btrfs_descending_sort_devices(devices, nr_device);
2400
2401         max_avail = devices[0].max_avail;
2402         if (!max_avail)
2403                 return -ENOSPC;
2404
2405         for (i = 0; i < nr_device; i++) {
2406                 /*
2407                  * if dev_offset = 0, it means the free space of this device
2408                  * is less than what we need, and we didn't search max avail
2409                  * extent on this device, so do it now.
2410                  */
2411                 if (!devices[i].dev_offset) {
2412                         ret = find_free_dev_extent(trans, devices[i].dev,
2413                                                    max_avail,
2414                                                    &devices[i].dev_offset,
2415                                                    &devices[i].max_avail);
2416                         if (ret != 0 && ret != -ENOSPC)
2417                                 return ret;
2418                         sort_again = 1;
2419                 }
2420         }
2421
2422         /* we update the max avail free extent of each devices, sort again */
2423         if (sort_again)
2424                 btrfs_descending_sort_devices(devices, nr_device);
2425
2426         if (type & BTRFS_BLOCK_GROUP_DUP)
2427                 min_devices = 1;
2428
2429         if (!devices[min_devices - 1].max_avail)
2430                 return -ENOSPC;
2431
2432         max_avail = devices[min_devices - 1].max_avail;
2433         if (type & BTRFS_BLOCK_GROUP_DUP)
2434                 do_div(max_avail, 2);
2435
2436         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2437                                              min_stripes, 1);
2438         if (type & BTRFS_BLOCK_GROUP_DUP)
2439                 min_free = max_avail * 2;
2440         else
2441                 min_free = max_avail;
2442
2443         if (min_free > devices[min_devices - 1].max_avail)
2444                 return -ENOSPC;
2445
2446         map = __shrink_map_lookup_stripes(map, min_stripes);
2447         *stripe_size = max_avail;
2448
2449         index = 0;
2450         for (i = 0; i < min_stripes; i++) {
2451                 map->stripes[i].dev = devices[index].dev;
2452                 map->stripes[i].physical = devices[index].dev_offset;
2453                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2454                         i++;
2455                         map->stripes[i].dev = devices[index].dev;
2456                         map->stripes[i].physical = devices[index].dev_offset +
2457                                                    max_avail;
2458                 }
2459                 index++;
2460         }
2461         *map_lookup = map;
2462
2463         return 0;
2464 }
2465
2466 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2467                                struct btrfs_root *extent_root,
2468                                struct map_lookup **map_ret,
2469                                u64 *num_bytes, u64 *stripe_size,
2470                                u64 start, u64 type)
2471 {
2472         struct btrfs_fs_info *info = extent_root->fs_info;
2473         struct btrfs_device *device = NULL;
2474         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2475         struct list_head *cur;
2476         struct map_lookup *map;
2477         struct extent_map_tree *em_tree;
2478         struct extent_map *em;
2479         struct btrfs_device_info *devices_info;
2480         struct list_head private_devs;
2481         u64 calc_size = 1024 * 1024 * 1024;
2482         u64 min_free;
2483         u64 avail;
2484         u64 dev_offset;
2485         int num_stripes;
2486         int min_stripes;
2487         int sub_stripes;
2488         int min_devices;        /* the min number of devices we need */
2489         int i;
2490         int ret;
2491         int index;
2492
2493         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2494             (type & BTRFS_BLOCK_GROUP_DUP)) {
2495                 WARN_ON(1);
2496                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2497         }
2498         if (list_empty(&fs_devices->alloc_list))
2499                 return -ENOSPC;
2500
2501         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2502                                     &min_stripes, &sub_stripes);
2503         if (ret)
2504                 return ret;
2505
2506         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2507                                GFP_NOFS);
2508         if (!devices_info)
2509                 return -ENOMEM;
2510
2511         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2512         if (!map) {
2513                 ret = -ENOMEM;
2514                 goto error;
2515         }
2516         map->num_stripes = num_stripes;
2517
2518         cur = fs_devices->alloc_list.next;
2519         index = 0;
2520         i = 0;
2521
2522         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2523                                              num_stripes, 0);
2524
2525         if (type & BTRFS_BLOCK_GROUP_DUP) {
2526                 min_free = calc_size * 2;
2527                 min_devices = 1;
2528         } else {
2529                 min_free = calc_size;
2530                 min_devices = min_stripes;
2531         }
2532
2533         INIT_LIST_HEAD(&private_devs);
2534         while (index < num_stripes) {
2535                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2536                 BUG_ON(!device->writeable);
2537                 if (device->total_bytes > device->bytes_used)
2538                         avail = device->total_bytes - device->bytes_used;
2539                 else
2540                         avail = 0;
2541                 cur = cur->next;
2542
2543                 if (device->in_fs_metadata && avail >= min_free) {
2544                         ret = find_free_dev_extent(trans, device, min_free,
2545                                                    &devices_info[i].dev_offset,
2546                                                    &devices_info[i].max_avail);
2547                         if (ret == 0) {
2548                                 list_move_tail(&device->dev_alloc_list,
2549                                                &private_devs);
2550                                 map->stripes[index].dev = device;
2551                                 map->stripes[index].physical =
2552                                                 devices_info[i].dev_offset;
2553                                 index++;
2554                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2555                                         map->stripes[index].dev = device;
2556                                         map->stripes[index].physical =
2557                                                 devices_info[i].dev_offset +
2558                                                 calc_size;
2559                                         index++;
2560                                 }
2561                         } else if (ret != -ENOSPC)
2562                                 goto error;
2563
2564                         devices_info[i].dev = device;
2565                         i++;
2566                 } else if (device->in_fs_metadata &&
2567                            avail >= BTRFS_STRIPE_LEN) {
2568                         devices_info[i].dev = device;
2569                         devices_info[i].max_avail = avail;
2570                         i++;
2571                 }
2572
2573                 if (cur == &fs_devices->alloc_list)
2574                         break;
2575         }
2576
2577         list_splice(&private_devs, &fs_devices->alloc_list);
2578         if (index < num_stripes) {
2579                 if (index >= min_stripes) {
2580                         num_stripes = index;
2581                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2582                                 num_stripes /= sub_stripes;
2583                                 num_stripes *= sub_stripes;
2584                         }
2585
2586                         map = __shrink_map_lookup_stripes(map, num_stripes);
2587                 } else if (i >= min_devices) {
2588                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2589                                                        devices_info, i, type,
2590                                                        &map, min_stripes,
2591                                                        &calc_size);
2592                         if (ret)
2593                                 goto error;
2594                 } else {
2595                         ret = -ENOSPC;
2596                         goto error;
2597                 }
2598         }
2599         map->sector_size = extent_root->sectorsize;
2600         map->stripe_len = BTRFS_STRIPE_LEN;
2601         map->io_align = BTRFS_STRIPE_LEN;
2602         map->io_width = BTRFS_STRIPE_LEN;
2603         map->type = type;
2604         map->sub_stripes = sub_stripes;
2605
2606         *map_ret = map;
2607         *stripe_size = calc_size;
2608         *num_bytes = chunk_bytes_by_type(type, calc_size,
2609                                          map->num_stripes, sub_stripes);
2610
2611         trace_btrfs_chunk_alloc(info->chunk_root, map, start, *num_bytes);
2612
2613         em = alloc_extent_map(GFP_NOFS);
2614         if (!em) {
2615                 ret = -ENOMEM;
2616                 goto error;
2617         }
2618         em->bdev = (struct block_device *)map;
2619         em->start = start;
2620         em->len = *num_bytes;
2621         em->block_start = 0;
2622         em->block_len = em->len;
2623
2624         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2625         write_lock(&em_tree->lock);
2626         ret = add_extent_mapping(em_tree, em);
2627         write_unlock(&em_tree->lock);
2628         BUG_ON(ret);
2629         free_extent_map(em);
2630
2631         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2632                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2633                                      start, *num_bytes);
2634         BUG_ON(ret);
2635
2636         index = 0;
2637         while (index < map->num_stripes) {
2638                 device = map->stripes[index].dev;
2639                 dev_offset = map->stripes[index].physical;
2640
2641                 ret = btrfs_alloc_dev_extent(trans, device,
2642                                 info->chunk_root->root_key.objectid,
2643                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2644                                 start, dev_offset, calc_size);
2645                 BUG_ON(ret);
2646                 index++;
2647         }
2648
2649         kfree(devices_info);
2650         return 0;
2651
2652 error:
2653         kfree(map);
2654         kfree(devices_info);
2655         return ret;
2656 }
2657
2658 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2659                                 struct btrfs_root *extent_root,
2660                                 struct map_lookup *map, u64 chunk_offset,
2661                                 u64 chunk_size, u64 stripe_size)
2662 {
2663         u64 dev_offset;
2664         struct btrfs_key key;
2665         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2666         struct btrfs_device *device;
2667         struct btrfs_chunk *chunk;
2668         struct btrfs_stripe *stripe;
2669         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2670         int index = 0;
2671         int ret;
2672
2673         chunk = kzalloc(item_size, GFP_NOFS);
2674         if (!chunk)
2675                 return -ENOMEM;
2676
2677         index = 0;
2678         while (index < map->num_stripes) {
2679                 device = map->stripes[index].dev;
2680                 device->bytes_used += stripe_size;
2681                 ret = btrfs_update_device(trans, device);
2682                 BUG_ON(ret);
2683                 index++;
2684         }
2685
2686         index = 0;
2687         stripe = &chunk->stripe;
2688         while (index < map->num_stripes) {
2689                 device = map->stripes[index].dev;
2690                 dev_offset = map->stripes[index].physical;
2691
2692                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2693                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2694                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2695                 stripe++;
2696                 index++;
2697         }
2698
2699         btrfs_set_stack_chunk_length(chunk, chunk_size);
2700         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2701         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2702         btrfs_set_stack_chunk_type(chunk, map->type);
2703         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2704         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2705         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2706         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2707         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2708
2709         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2710         key.type = BTRFS_CHUNK_ITEM_KEY;
2711         key.offset = chunk_offset;
2712
2713         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2714         BUG_ON(ret);
2715
2716         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2717                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2718                                              item_size);
2719                 BUG_ON(ret);
2720         }
2721
2722         kfree(chunk);
2723         return 0;
2724 }
2725
2726 /*
2727  * Chunk allocation falls into two parts. The first part does works
2728  * that make the new allocated chunk useable, but not do any operation
2729  * that modifies the chunk tree. The second part does the works that
2730  * require modifying the chunk tree. This division is important for the
2731  * bootstrap process of adding storage to a seed btrfs.
2732  */
2733 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2734                       struct btrfs_root *extent_root, u64 type)
2735 {
2736         u64 chunk_offset;
2737         u64 chunk_size;
2738         u64 stripe_size;
2739         struct map_lookup *map;
2740         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2741         int ret;
2742
2743         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2744                               &chunk_offset);
2745         if (ret)
2746                 return ret;
2747
2748         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2749                                   &stripe_size, chunk_offset, type);
2750         if (ret)
2751                 return ret;
2752
2753         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2754                                    chunk_size, stripe_size);
2755         BUG_ON(ret);
2756         return 0;
2757 }
2758
2759 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2760                                          struct btrfs_root *root,
2761                                          struct btrfs_device *device)
2762 {
2763         u64 chunk_offset;
2764         u64 sys_chunk_offset;
2765         u64 chunk_size;
2766         u64 sys_chunk_size;
2767         u64 stripe_size;
2768         u64 sys_stripe_size;
2769         u64 alloc_profile;
2770         struct map_lookup *map;
2771         struct map_lookup *sys_map;
2772         struct btrfs_fs_info *fs_info = root->fs_info;
2773         struct btrfs_root *extent_root = fs_info->extent_root;
2774         int ret;
2775
2776         ret = find_next_chunk(fs_info->chunk_root,
2777                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2778         BUG_ON(ret);
2779
2780         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2781                         (fs_info->metadata_alloc_profile &
2782                          fs_info->avail_metadata_alloc_bits);
2783         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2784
2785         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2786                                   &stripe_size, chunk_offset, alloc_profile);
2787         BUG_ON(ret);
2788
2789         sys_chunk_offset = chunk_offset + chunk_size;
2790
2791         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2792                         (fs_info->system_alloc_profile &
2793                          fs_info->avail_system_alloc_bits);
2794         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2795
2796         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2797                                   &sys_chunk_size, &sys_stripe_size,
2798                                   sys_chunk_offset, alloc_profile);
2799         BUG_ON(ret);
2800
2801         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2802         BUG_ON(ret);
2803
2804         /*
2805          * Modifying chunk tree needs allocating new blocks from both
2806          * system block group and metadata block group. So we only can
2807          * do operations require modifying the chunk tree after both
2808          * block groups were created.
2809          */
2810         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2811                                    chunk_size, stripe_size);
2812         BUG_ON(ret);
2813
2814         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2815                                    sys_chunk_offset, sys_chunk_size,
2816                                    sys_stripe_size);
2817         BUG_ON(ret);
2818         return 0;
2819 }
2820
2821 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2822 {
2823         struct extent_map *em;
2824         struct map_lookup *map;
2825         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2826         int readonly = 0;
2827         int i;
2828
2829         read_lock(&map_tree->map_tree.lock);
2830         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2831         read_unlock(&map_tree->map_tree.lock);
2832         if (!em)
2833                 return 1;
2834
2835         if (btrfs_test_opt(root, DEGRADED)) {
2836                 free_extent_map(em);
2837                 return 0;
2838         }
2839
2840         map = (struct map_lookup *)em->bdev;
2841         for (i = 0; i < map->num_stripes; i++) {
2842                 if (!map->stripes[i].dev->writeable) {
2843                         readonly = 1;
2844                         break;
2845                 }
2846         }
2847         free_extent_map(em);
2848         return readonly;
2849 }
2850
2851 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2852 {
2853         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2854 }
2855
2856 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2857 {
2858         struct extent_map *em;
2859
2860         while (1) {
2861                 write_lock(&tree->map_tree.lock);
2862                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2863                 if (em)
2864                         remove_extent_mapping(&tree->map_tree, em);
2865                 write_unlock(&tree->map_tree.lock);
2866                 if (!em)
2867                         break;
2868                 kfree(em->bdev);
2869                 /* once for us */
2870                 free_extent_map(em);
2871                 /* once for the tree */
2872                 free_extent_map(em);
2873         }
2874 }
2875
2876 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2877 {
2878         struct extent_map *em;
2879         struct map_lookup *map;
2880         struct extent_map_tree *em_tree = &map_tree->map_tree;
2881         int ret;
2882
2883         read_lock(&em_tree->lock);
2884         em = lookup_extent_mapping(em_tree, logical, len);
2885         read_unlock(&em_tree->lock);
2886         BUG_ON(!em);
2887
2888         BUG_ON(em->start > logical || em->start + em->len < logical);
2889         map = (struct map_lookup *)em->bdev;
2890         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2891                 ret = map->num_stripes;
2892         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2893                 ret = map->sub_stripes;
2894         else
2895                 ret = 1;
2896         free_extent_map(em);
2897         return ret;
2898 }
2899
2900 static int find_live_mirror(struct map_lookup *map, int first, int num,
2901                             int optimal)
2902 {
2903         int i;
2904         if (map->stripes[optimal].dev->bdev)
2905                 return optimal;
2906         for (i = first; i < first + num; i++) {
2907                 if (map->stripes[i].dev->bdev)
2908                         return i;
2909         }
2910         /* we couldn't find one that doesn't fail.  Just return something
2911          * and the io error handling code will clean up eventually
2912          */
2913         return optimal;
2914 }
2915
2916 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2917                              u64 logical, u64 *length,
2918                              struct btrfs_multi_bio **multi_ret,
2919                              int mirror_num)
2920 {
2921         struct extent_map *em;
2922         struct map_lookup *map;
2923         struct extent_map_tree *em_tree = &map_tree->map_tree;
2924         u64 offset;
2925         u64 stripe_offset;
2926         u64 stripe_end_offset;
2927         u64 stripe_nr;
2928         u64 stripe_nr_orig;
2929         u64 stripe_nr_end;
2930         int stripes_allocated = 8;
2931         int stripes_required = 1;
2932         int stripe_index;
2933         int i;
2934         int num_stripes;
2935         int max_errors = 0;
2936         struct btrfs_multi_bio *multi = NULL;
2937
2938         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2939                 stripes_allocated = 1;
2940 again:
2941         if (multi_ret) {
2942                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2943                                 GFP_NOFS);
2944                 if (!multi)
2945                         return -ENOMEM;
2946
2947                 atomic_set(&multi->error, 0);
2948         }
2949
2950         read_lock(&em_tree->lock);
2951         em = lookup_extent_mapping(em_tree, logical, *length);
2952         read_unlock(&em_tree->lock);
2953
2954         if (!em) {
2955                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2956                        (unsigned long long)logical,
2957                        (unsigned long long)*length);
2958                 BUG();
2959         }
2960
2961         BUG_ON(em->start > logical || em->start + em->len < logical);
2962         map = (struct map_lookup *)em->bdev;
2963         offset = logical - em->start;
2964
2965         if (mirror_num > map->num_stripes)
2966                 mirror_num = 0;
2967
2968         /* if our multi bio struct is too small, back off and try again */
2969         if (rw & REQ_WRITE) {
2970                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2971                                  BTRFS_BLOCK_GROUP_DUP)) {
2972                         stripes_required = map->num_stripes;
2973                         max_errors = 1;
2974                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2975                         stripes_required = map->sub_stripes;
2976                         max_errors = 1;
2977                 }
2978         }
2979         if (rw & REQ_DISCARD) {
2980                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2981                                  BTRFS_BLOCK_GROUP_RAID1 |
2982                                  BTRFS_BLOCK_GROUP_DUP |
2983                                  BTRFS_BLOCK_GROUP_RAID10)) {
2984                         stripes_required = map->num_stripes;
2985                 }
2986         }
2987         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2988             stripes_allocated < stripes_required) {
2989                 stripes_allocated = map->num_stripes;
2990                 free_extent_map(em);
2991                 kfree(multi);
2992                 goto again;
2993         }
2994         stripe_nr = offset;
2995         /*
2996          * stripe_nr counts the total number of stripes we have to stride
2997          * to get to this block
2998          */
2999         do_div(stripe_nr, map->stripe_len);
3000
3001         stripe_offset = stripe_nr * map->stripe_len;
3002         BUG_ON(offset < stripe_offset);
3003
3004         /* stripe_offset is the offset of this block in its stripe*/
3005         stripe_offset = offset - stripe_offset;
3006
3007         if (rw & REQ_DISCARD)
3008                 *length = min_t(u64, em->len - offset, *length);
3009         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3010                               BTRFS_BLOCK_GROUP_RAID1 |
3011                               BTRFS_BLOCK_GROUP_RAID10 |
3012                               BTRFS_BLOCK_GROUP_DUP)) {
3013                 /* we limit the length of each bio to what fits in a stripe */
3014                 *length = min_t(u64, em->len - offset,
3015                                 map->stripe_len - stripe_offset);
3016         } else {
3017                 *length = em->len - offset;
3018         }
3019
3020         if (!multi_ret)
3021                 goto out;
3022
3023         num_stripes = 1;
3024         stripe_index = 0;
3025         stripe_nr_orig = stripe_nr;
3026         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3027                         (~(map->stripe_len - 1));
3028         do_div(stripe_nr_end, map->stripe_len);
3029         stripe_end_offset = stripe_nr_end * map->stripe_len -
3030                             (offset + *length);
3031         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3032                 if (rw & REQ_DISCARD)
3033                         num_stripes = min_t(u64, map->num_stripes,
3034                                             stripe_nr_end - stripe_nr_orig);
3035                 stripe_index = do_div(stripe_nr, map->num_stripes);
3036         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3037                 if (rw & (REQ_WRITE | REQ_DISCARD))
3038                         num_stripes = map->num_stripes;
3039                 else if (mirror_num)
3040                         stripe_index = mirror_num - 1;
3041                 else {
3042                         stripe_index = find_live_mirror(map, 0,
3043                                             map->num_stripes,
3044                                             current->pid % map->num_stripes);
3045                 }
3046
3047         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3048                 if (rw & (REQ_WRITE | REQ_DISCARD))
3049                         num_stripes = map->num_stripes;
3050                 else if (mirror_num)
3051                         stripe_index = mirror_num - 1;
3052
3053         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3054                 int factor = map->num_stripes / map->sub_stripes;
3055
3056                 stripe_index = do_div(stripe_nr, factor);
3057                 stripe_index *= map->sub_stripes;
3058
3059                 if (rw & REQ_WRITE)
3060                         num_stripes = map->sub_stripes;
3061                 else if (rw & REQ_DISCARD)
3062                         num_stripes = min_t(u64, map->sub_stripes *
3063                                             (stripe_nr_end - stripe_nr_orig),
3064                                             map->num_stripes);
3065                 else if (mirror_num)
3066                         stripe_index += mirror_num - 1;
3067                 else {
3068                         stripe_index = find_live_mirror(map, stripe_index,
3069                                               map->sub_stripes, stripe_index +
3070                                               current->pid % map->sub_stripes);
3071                 }
3072         } else {
3073                 /*
3074                  * after this do_div call, stripe_nr is the number of stripes
3075                  * on this device we have to walk to find the data, and
3076                  * stripe_index is the number of our device in the stripe array
3077                  */
3078                 stripe_index = do_div(stripe_nr, map->num_stripes);
3079         }
3080         BUG_ON(stripe_index >= map->num_stripes);
3081
3082         if (rw & REQ_DISCARD) {
3083                 for (i = 0; i < num_stripes; i++) {
3084                         multi->stripes[i].physical =
3085                                 map->stripes[stripe_index].physical +
3086                                 stripe_offset + stripe_nr * map->stripe_len;
3087                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3088
3089                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3090                                 u64 stripes;
3091                                 u32 last_stripe = 0;
3092                                 int j;
3093
3094                                 div_u64_rem(stripe_nr_end - 1,
3095                                             map->num_stripes,
3096                                             &last_stripe);
3097
3098                                 for (j = 0; j < map->num_stripes; j++) {
3099                                         u32 test;
3100
3101                                         div_u64_rem(stripe_nr_end - 1 - j,
3102                                                     map->num_stripes, &test);
3103                                         if (test == stripe_index)
3104                                                 break;
3105                                 }
3106                                 stripes = stripe_nr_end - 1 - j;
3107                                 do_div(stripes, map->num_stripes);
3108                                 multi->stripes[i].length = map->stripe_len *
3109                                         (stripes - stripe_nr + 1);
3110
3111                                 if (i == 0) {
3112                                         multi->stripes[i].length -=
3113                                                 stripe_offset;
3114                                         stripe_offset = 0;
3115                                 }
3116                                 if (stripe_index == last_stripe)
3117                                         multi->stripes[i].length -=
3118                                                 stripe_end_offset;
3119                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3120                                 u64 stripes;
3121                                 int j;
3122                                 int factor = map->num_stripes /
3123                                              map->sub_stripes;
3124                                 u32 last_stripe = 0;
3125
3126                                 div_u64_rem(stripe_nr_end - 1,
3127                                             factor, &last_stripe);
3128                                 last_stripe *= map->sub_stripes;
3129
3130                                 for (j = 0; j < factor; j++) {
3131                                         u32 test;
3132
3133                                         div_u64_rem(stripe_nr_end - 1 - j,
3134                                                     factor, &test);
3135
3136                                         if (test ==
3137                                             stripe_index / map->sub_stripes)
3138                                                 break;
3139                                 }
3140                                 stripes = stripe_nr_end - 1 - j;
3141                                 do_div(stripes, factor);
3142                                 multi->stripes[i].length = map->stripe_len *
3143                                         (stripes - stripe_nr + 1);
3144
3145                                 if (i < map->sub_stripes) {
3146                                         multi->stripes[i].length -=
3147                                                 stripe_offset;
3148                                         if (i == map->sub_stripes - 1)
3149                                                 stripe_offset = 0;
3150                                 }
3151                                 if (stripe_index >= last_stripe &&
3152                                     stripe_index <= (last_stripe +
3153                                                      map->sub_stripes - 1)) {
3154                                         multi->stripes[i].length -=
3155                                                 stripe_end_offset;
3156                                 }
3157                         } else
3158                                 multi->stripes[i].length = *length;
3159
3160                         stripe_index++;
3161                         if (stripe_index == map->num_stripes) {
3162                                 /* This could only happen for RAID0/10 */
3163                                 stripe_index = 0;
3164                                 stripe_nr++;
3165                         }
3166                 }
3167         } else {
3168                 for (i = 0; i < num_stripes; i++) {
3169                         multi->stripes[i].physical =
3170                                 map->stripes[stripe_index].physical +
3171                                 stripe_offset +
3172                                 stripe_nr * map->stripe_len;
3173                         multi->stripes[i].dev =
3174                                 map->stripes[stripe_index].dev;
3175                         stripe_index++;
3176                 }
3177         }
3178         if (multi_ret) {
3179                 *multi_ret = multi;
3180                 multi->num_stripes = num_stripes;
3181                 multi->max_errors = max_errors;
3182         }
3183 out:
3184         free_extent_map(em);
3185         return 0;
3186 }
3187
3188 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3189                       u64 logical, u64 *length,
3190                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3191 {
3192         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3193                                  mirror_num);
3194 }
3195
3196 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3197                      u64 chunk_start, u64 physical, u64 devid,
3198                      u64 **logical, int *naddrs, int *stripe_len)
3199 {
3200         struct extent_map_tree *em_tree = &map_tree->map_tree;
3201         struct extent_map *em;
3202         struct map_lookup *map;
3203         u64 *buf;
3204         u64 bytenr;
3205         u64 length;
3206         u64 stripe_nr;
3207         int i, j, nr = 0;
3208
3209         read_lock(&em_tree->lock);
3210         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3211         read_unlock(&em_tree->lock);
3212
3213         BUG_ON(!em || em->start != chunk_start);
3214         map = (struct map_lookup *)em->bdev;
3215
3216         length = em->len;
3217         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3218                 do_div(length, map->num_stripes / map->sub_stripes);
3219         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3220                 do_div(length, map->num_stripes);
3221
3222         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3223         BUG_ON(!buf);
3224
3225         for (i = 0; i < map->num_stripes; i++) {
3226                 if (devid && map->stripes[i].dev->devid != devid)
3227                         continue;
3228                 if (map->stripes[i].physical > physical ||
3229                     map->stripes[i].physical + length <= physical)
3230                         continue;
3231
3232                 stripe_nr = physical - map->stripes[i].physical;
3233                 do_div(stripe_nr, map->stripe_len);
3234
3235                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3236                         stripe_nr = stripe_nr * map->num_stripes + i;
3237                         do_div(stripe_nr, map->sub_stripes);
3238                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3239                         stripe_nr = stripe_nr * map->num_stripes + i;
3240                 }
3241                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3242                 WARN_ON(nr >= map->num_stripes);
3243                 for (j = 0; j < nr; j++) {
3244                         if (buf[j] == bytenr)
3245                                 break;
3246                 }
3247                 if (j == nr) {
3248                         WARN_ON(nr >= map->num_stripes);
3249                         buf[nr++] = bytenr;
3250                 }
3251         }
3252
3253         *logical = buf;
3254         *naddrs = nr;
3255         *stripe_len = map->stripe_len;
3256
3257         free_extent_map(em);
3258         return 0;
3259 }
3260
3261 static void end_bio_multi_stripe(struct bio *bio, int err)
3262 {
3263         struct btrfs_multi_bio *multi = bio->bi_private;
3264         int is_orig_bio = 0;
3265
3266         if (err)
3267                 atomic_inc(&multi->error);
3268
3269         if (bio == multi->orig_bio)
3270                 is_orig_bio = 1;
3271
3272         if (atomic_dec_and_test(&multi->stripes_pending)) {
3273                 if (!is_orig_bio) {
3274                         bio_put(bio);
3275                         bio = multi->orig_bio;
3276                 }
3277                 bio->bi_private = multi->private;
3278                 bio->bi_end_io = multi->end_io;
3279                 /* only send an error to the higher layers if it is
3280                  * beyond the tolerance of the multi-bio
3281                  */
3282                 if (atomic_read(&multi->error) > multi->max_errors) {
3283                         err = -EIO;
3284                 } else if (err) {
3285                         /*
3286                          * this bio is actually up to date, we didn't
3287                          * go over the max number of errors
3288                          */
3289                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3290                         err = 0;
3291                 }
3292                 kfree(multi);
3293
3294                 bio_endio(bio, err);
3295         } else if (!is_orig_bio) {
3296                 bio_put(bio);
3297         }
3298 }
3299
3300 struct async_sched {
3301         struct bio *bio;
3302         int rw;
3303         struct btrfs_fs_info *info;
3304         struct btrfs_work work;
3305 };
3306
3307 /*
3308  * see run_scheduled_bios for a description of why bios are collected for
3309  * async submit.
3310  *
3311  * This will add one bio to the pending list for a device and make sure
3312  * the work struct is scheduled.
3313  */
3314 static noinline int schedule_bio(struct btrfs_root *root,
3315                                  struct btrfs_device *device,
3316                                  int rw, struct bio *bio)
3317 {
3318         int should_queue = 1;
3319         struct btrfs_pending_bios *pending_bios;
3320
3321         /* don't bother with additional async steps for reads, right now */
3322         if (!(rw & REQ_WRITE)) {
3323                 bio_get(bio);
3324                 submit_bio(rw, bio);
3325                 bio_put(bio);
3326                 return 0;
3327         }
3328
3329         /*
3330          * nr_async_bios allows us to reliably return congestion to the
3331          * higher layers.  Otherwise, the async bio makes it appear we have
3332          * made progress against dirty pages when we've really just put it
3333          * on a queue for later
3334          */
3335         atomic_inc(&root->fs_info->nr_async_bios);
3336         WARN_ON(bio->bi_next);
3337         bio->bi_next = NULL;
3338         bio->bi_rw |= rw;
3339
3340         spin_lock(&device->io_lock);
3341         if (bio->bi_rw & REQ_SYNC)
3342                 pending_bios = &device->pending_sync_bios;
3343         else
3344                 pending_bios = &device->pending_bios;
3345
3346         if (pending_bios->tail)
3347                 pending_bios->tail->bi_next = bio;
3348
3349         pending_bios->tail = bio;
3350         if (!pending_bios->head)
3351                 pending_bios->head = bio;
3352         if (device->running_pending)
3353                 should_queue = 0;
3354
3355         spin_unlock(&device->io_lock);
3356
3357         if (should_queue)
3358                 btrfs_queue_worker(&root->fs_info->submit_workers,
3359                                    &device->work);
3360         return 0;
3361 }
3362
3363 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3364                   int mirror_num, int async_submit)
3365 {
3366         struct btrfs_mapping_tree *map_tree;
3367         struct btrfs_device *dev;
3368         struct bio *first_bio = bio;
3369         u64 logical = (u64)bio->bi_sector << 9;
3370         u64 length = 0;
3371         u64 map_length;
3372         struct btrfs_multi_bio *multi = NULL;
3373         int ret;
3374         int dev_nr = 0;
3375         int total_devs = 1;
3376
3377         length = bio->bi_size;
3378         map_tree = &root->fs_info->mapping_tree;
3379         map_length = length;
3380
3381         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3382                               mirror_num);
3383         BUG_ON(ret);
3384
3385         total_devs = multi->num_stripes;
3386         if (map_length < length) {
3387                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3388                        "len %llu\n", (unsigned long long)logical,
3389                        (unsigned long long)length,
3390                        (unsigned long long)map_length);
3391                 BUG();
3392         }
3393         multi->end_io = first_bio->bi_end_io;
3394         multi->private = first_bio->bi_private;
3395         multi->orig_bio = first_bio;
3396         atomic_set(&multi->stripes_pending, multi->num_stripes);
3397
3398         while (dev_nr < total_devs) {
3399                 if (total_devs > 1) {
3400                         if (dev_nr < total_devs - 1) {
3401                                 bio = bio_clone(first_bio, GFP_NOFS);
3402                                 BUG_ON(!bio);
3403                         } else {
3404                                 bio = first_bio;
3405                         }
3406                         bio->bi_private = multi;
3407                         bio->bi_end_io = end_bio_multi_stripe;
3408                 }
3409                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3410                 dev = multi->stripes[dev_nr].dev;
3411                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3412                         bio->bi_bdev = dev->bdev;
3413                         if (async_submit)
3414                                 schedule_bio(root, dev, rw, bio);
3415                         else
3416                                 submit_bio(rw, bio);
3417                 } else {
3418                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3419                         bio->bi_sector = logical >> 9;
3420                         bio_endio(bio, -EIO);
3421                 }
3422                 dev_nr++;
3423         }
3424         if (total_devs == 1)
3425                 kfree(multi);
3426         return 0;
3427 }
3428
3429 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3430                                        u8 *uuid, u8 *fsid)
3431 {
3432         struct btrfs_device *device;
3433         struct btrfs_fs_devices *cur_devices;
3434
3435         cur_devices = root->fs_info->fs_devices;
3436         while (cur_devices) {
3437                 if (!fsid ||
3438                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3439                         device = __find_device(&cur_devices->devices,
3440                                                devid, uuid);
3441                         if (device)
3442                                 return device;
3443                 }
3444                 cur_devices = cur_devices->seed;
3445         }
3446         return NULL;
3447 }
3448
3449 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3450                                             u64 devid, u8 *dev_uuid)
3451 {
3452         struct btrfs_device *device;
3453         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3454
3455         device = kzalloc(sizeof(*device), GFP_NOFS);
3456         if (!device)
3457                 return NULL;
3458         list_add(&device->dev_list,
3459                  &fs_devices->devices);
3460         device->dev_root = root->fs_info->dev_root;
3461         device->devid = devid;
3462         device->work.func = pending_bios_fn;
3463         device->fs_devices = fs_devices;
3464         device->missing = 1;
3465         fs_devices->num_devices++;
3466         fs_devices->missing_devices++;
3467         spin_lock_init(&device->io_lock);
3468         INIT_LIST_HEAD(&device->dev_alloc_list);
3469         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3470         return device;
3471 }
3472
3473 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3474                           struct extent_buffer *leaf,
3475                           struct btrfs_chunk *chunk)
3476 {
3477         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3478         struct map_lookup *map;
3479         struct extent_map *em;
3480         u64 logical;
3481         u64 length;
3482         u64 devid;
3483         u8 uuid[BTRFS_UUID_SIZE];
3484         int num_stripes;
3485         int ret;
3486         int i;
3487
3488         logical = key->offset;
3489         length = btrfs_chunk_length(leaf, chunk);
3490
3491         read_lock(&map_tree->map_tree.lock);
3492         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3493         read_unlock(&map_tree->map_tree.lock);
3494
3495         /* already mapped? */
3496         if (em && em->start <= logical && em->start + em->len > logical) {
3497                 free_extent_map(em);
3498                 return 0;
3499         } else if (em) {
3500                 free_extent_map(em);
3501         }
3502
3503         em = alloc_extent_map(GFP_NOFS);
3504         if (!em)
3505                 return -ENOMEM;
3506         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3507         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3508         if (!map) {
3509                 free_extent_map(em);
3510                 return -ENOMEM;
3511         }
3512
3513         em->bdev = (struct block_device *)map;
3514         em->start = logical;
3515         em->len = length;
3516         em->block_start = 0;
3517         em->block_len = em->len;
3518
3519         map->num_stripes = num_stripes;
3520         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3521         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3522         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3523         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3524         map->type = btrfs_chunk_type(leaf, chunk);
3525         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3526         for (i = 0; i < num_stripes; i++) {
3527                 map->stripes[i].physical =
3528                         btrfs_stripe_offset_nr(leaf, chunk, i);
3529                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3530                 read_extent_buffer(leaf, uuid, (unsigned long)
3531                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3532                                    BTRFS_UUID_SIZE);
3533                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3534                                                         NULL);
3535                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3536                         kfree(map);
3537                         free_extent_map(em);
3538                         return -EIO;
3539                 }
3540                 if (!map->stripes[i].dev) {
3541                         map->stripes[i].dev =
3542                                 add_missing_dev(root, devid, uuid);
3543                         if (!map->stripes[i].dev) {
3544                                 kfree(map);
3545                                 free_extent_map(em);
3546                                 return -EIO;
3547                         }
3548                 }
3549                 map->stripes[i].dev->in_fs_metadata = 1;
3550         }
3551
3552         write_lock(&map_tree->map_tree.lock);
3553         ret = add_extent_mapping(&map_tree->map_tree, em);
3554         write_unlock(&map_tree->map_tree.lock);
3555         BUG_ON(ret);
3556         free_extent_map(em);
3557
3558         return 0;
3559 }
3560
3561 static int fill_device_from_item(struct extent_buffer *leaf,
3562                                  struct btrfs_dev_item *dev_item,
3563                                  struct btrfs_device *device)
3564 {
3565         unsigned long ptr;
3566
3567         device->devid = btrfs_device_id(leaf, dev_item);
3568         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3569         device->total_bytes = device->disk_total_bytes;
3570         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3571         device->type = btrfs_device_type(leaf, dev_item);
3572         device->io_align = btrfs_device_io_align(leaf, dev_item);
3573         device->io_width = btrfs_device_io_width(leaf, dev_item);
3574         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3575
3576         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3577         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3578
3579         return 0;
3580 }
3581
3582 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3583 {
3584         struct btrfs_fs_devices *fs_devices;
3585         int ret;
3586
3587         mutex_lock(&uuid_mutex);
3588
3589         fs_devices = root->fs_info->fs_devices->seed;
3590         while (fs_devices) {
3591                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3592                         ret = 0;
3593                         goto out;
3594                 }
3595                 fs_devices = fs_devices->seed;
3596         }
3597
3598         fs_devices = find_fsid(fsid);
3599         if (!fs_devices) {
3600                 ret = -ENOENT;
3601                 goto out;
3602         }
3603
3604         fs_devices = clone_fs_devices(fs_devices);
3605         if (IS_ERR(fs_devices)) {
3606                 ret = PTR_ERR(fs_devices);
3607                 goto out;
3608         }
3609
3610         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3611                                    root->fs_info->bdev_holder);
3612         if (ret)
3613                 goto out;
3614
3615         if (!fs_devices->seeding) {
3616                 __btrfs_close_devices(fs_devices);
3617                 free_fs_devices(fs_devices);
3618                 ret = -EINVAL;
3619                 goto out;
3620         }
3621
3622         fs_devices->seed = root->fs_info->fs_devices->seed;
3623         root->fs_info->fs_devices->seed = fs_devices;
3624 out:
3625         mutex_unlock(&uuid_mutex);
3626         return ret;
3627 }
3628
3629 static int read_one_dev(struct btrfs_root *root,
3630                         struct extent_buffer *leaf,
3631                         struct btrfs_dev_item *dev_item)
3632 {
3633         struct btrfs_device *device;
3634         u64 devid;
3635         int ret;
3636         u8 fs_uuid[BTRFS_UUID_SIZE];
3637         u8 dev_uuid[BTRFS_UUID_SIZE];
3638
3639         devid = btrfs_device_id(leaf, dev_item);
3640         read_extent_buffer(leaf, dev_uuid,
3641                            (unsigned long)btrfs_device_uuid(dev_item),
3642                            BTRFS_UUID_SIZE);
3643         read_extent_buffer(leaf, fs_uuid,
3644                            (unsigned long)btrfs_device_fsid(dev_item),
3645                            BTRFS_UUID_SIZE);
3646
3647         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3648                 ret = open_seed_devices(root, fs_uuid);
3649                 if (ret && !btrfs_test_opt(root, DEGRADED))
3650                         return ret;
3651         }
3652
3653         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3654         if (!device || !device->bdev) {
3655                 if (!btrfs_test_opt(root, DEGRADED))
3656                         return -EIO;
3657
3658                 if (!device) {
3659                         printk(KERN_WARNING "warning devid %llu missing\n",
3660                                (unsigned long long)devid);
3661                         device = add_missing_dev(root, devid, dev_uuid);
3662                         if (!device)
3663                                 return -ENOMEM;
3664                 } else if (!device->missing) {
3665                         /*
3666                          * this happens when a device that was properly setup
3667                          * in the device info lists suddenly goes bad.
3668                          * device->bdev is NULL, and so we have to set
3669                          * device->missing to one here
3670                          */
3671                         root->fs_info->fs_devices->missing_devices++;
3672                         device->missing = 1;
3673                 }
3674         }
3675
3676         if (device->fs_devices != root->fs_info->fs_devices) {
3677                 BUG_ON(device->writeable);
3678                 if (device->generation !=
3679                     btrfs_device_generation(leaf, dev_item))
3680                         return -EINVAL;
3681         }
3682
3683         fill_device_from_item(leaf, dev_item, device);
3684         device->dev_root = root->fs_info->dev_root;
3685         device->in_fs_metadata = 1;
3686         if (device->writeable)
3687                 device->fs_devices->total_rw_bytes += device->total_bytes;
3688         ret = 0;
3689         return ret;
3690 }
3691
3692 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3693 {
3694         struct btrfs_dev_item *dev_item;
3695
3696         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3697                                                      dev_item);
3698         return read_one_dev(root, buf, dev_item);
3699 }
3700
3701 int btrfs_read_sys_array(struct btrfs_root *root)
3702 {
3703         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3704         struct extent_buffer *sb;
3705         struct btrfs_disk_key *disk_key;
3706         struct btrfs_chunk *chunk;
3707         u8 *ptr;
3708         unsigned long sb_ptr;
3709         int ret = 0;
3710         u32 num_stripes;
3711         u32 array_size;
3712         u32 len = 0;
3713         u32 cur;
3714         struct btrfs_key key;
3715
3716         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3717                                           BTRFS_SUPER_INFO_SIZE);
3718         if (!sb)
3719                 return -ENOMEM;
3720         btrfs_set_buffer_uptodate(sb);
3721         btrfs_set_buffer_lockdep_class(sb, 0);
3722
3723         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3724         array_size = btrfs_super_sys_array_size(super_copy);
3725
3726         ptr = super_copy->sys_chunk_array;
3727         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3728         cur = 0;
3729
3730         while (cur < array_size) {
3731                 disk_key = (struct btrfs_disk_key *)ptr;
3732                 btrfs_disk_key_to_cpu(&key, disk_key);
3733
3734                 len = sizeof(*disk_key); ptr += len;
3735                 sb_ptr += len;
3736                 cur += len;
3737
3738                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3739                         chunk = (struct btrfs_chunk *)sb_ptr;
3740                         ret = read_one_chunk(root, &key, sb, chunk);
3741                         if (ret)
3742                                 break;
3743                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3744                         len = btrfs_chunk_item_size(num_stripes);
3745                 } else {
3746                         ret = -EIO;
3747                         break;
3748                 }
3749                 ptr += len;
3750                 sb_ptr += len;
3751                 cur += len;
3752         }
3753         free_extent_buffer(sb);
3754         return ret;
3755 }
3756
3757 int btrfs_read_chunk_tree(struct btrfs_root *root)
3758 {
3759         struct btrfs_path *path;
3760         struct extent_buffer *leaf;
3761         struct btrfs_key key;
3762         struct btrfs_key found_key;
3763         int ret;
3764         int slot;
3765
3766         root = root->fs_info->chunk_root;
3767
3768         path = btrfs_alloc_path();
3769         if (!path)
3770                 return -ENOMEM;
3771
3772         /* first we search for all of the device items, and then we
3773          * read in all of the chunk items.  This way we can create chunk
3774          * mappings that reference all of the devices that are afound
3775          */
3776         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3777         key.offset = 0;
3778         key.type = 0;
3779 again:
3780         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3781         if (ret < 0)
3782                 goto error;
3783         while (1) {
3784                 leaf = path->nodes[0];
3785                 slot = path->slots[0];
3786                 if (slot >= btrfs_header_nritems(leaf)) {
3787                         ret = btrfs_next_leaf(root, path);
3788                         if (ret == 0)
3789                                 continue;
3790                         if (ret < 0)
3791                                 goto error;
3792                         break;
3793                 }
3794                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3795                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3796                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3797                                 break;
3798                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3799                                 struct btrfs_dev_item *dev_item;
3800                                 dev_item = btrfs_item_ptr(leaf, slot,
3801                                                   struct btrfs_dev_item);
3802                                 ret = read_one_dev(root, leaf, dev_item);
3803                                 if (ret)
3804                                         goto error;
3805                         }
3806                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3807                         struct btrfs_chunk *chunk;
3808                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3809                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3810                         if (ret)
3811                                 goto error;
3812                 }
3813                 path->slots[0]++;
3814         }
3815         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3816                 key.objectid = 0;
3817                 btrfs_release_path(root, path);
3818                 goto again;
3819         }
3820         ret = 0;
3821 error:
3822         btrfs_free_path(path);
3823         return ret;
3824 }