Btrfs: return error code to caller when btrfs_del_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
971 out:
972         btrfs_free_path(path);
973         return ret;
974 }
975
976 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
977                            struct btrfs_device *device,
978                            u64 chunk_tree, u64 chunk_objectid,
979                            u64 chunk_offset, u64 start, u64 num_bytes)
980 {
981         int ret;
982         struct btrfs_path *path;
983         struct btrfs_root *root = device->dev_root;
984         struct btrfs_dev_extent *extent;
985         struct extent_buffer *leaf;
986         struct btrfs_key key;
987
988         WARN_ON(!device->in_fs_metadata);
989         path = btrfs_alloc_path();
990         if (!path)
991                 return -ENOMEM;
992
993         key.objectid = device->devid;
994         key.offset = start;
995         key.type = BTRFS_DEV_EXTENT_KEY;
996         ret = btrfs_insert_empty_item(trans, root, path, &key,
997                                       sizeof(*extent));
998         BUG_ON(ret);
999
1000         leaf = path->nodes[0];
1001         extent = btrfs_item_ptr(leaf, path->slots[0],
1002                                 struct btrfs_dev_extent);
1003         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1004         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1005         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1006
1007         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1008                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1009                     BTRFS_UUID_SIZE);
1010
1011         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1012         btrfs_mark_buffer_dirty(leaf);
1013         btrfs_free_path(path);
1014         return ret;
1015 }
1016
1017 static noinline int find_next_chunk(struct btrfs_root *root,
1018                                     u64 objectid, u64 *offset)
1019 {
1020         struct btrfs_path *path;
1021         int ret;
1022         struct btrfs_key key;
1023         struct btrfs_chunk *chunk;
1024         struct btrfs_key found_key;
1025
1026         path = btrfs_alloc_path();
1027         BUG_ON(!path);
1028
1029         key.objectid = objectid;
1030         key.offset = (u64)-1;
1031         key.type = BTRFS_CHUNK_ITEM_KEY;
1032
1033         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1034         if (ret < 0)
1035                 goto error;
1036
1037         BUG_ON(ret == 0);
1038
1039         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1040         if (ret) {
1041                 *offset = 0;
1042         } else {
1043                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1044                                       path->slots[0]);
1045                 if (found_key.objectid != objectid)
1046                         *offset = 0;
1047                 else {
1048                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1049                                                struct btrfs_chunk);
1050                         *offset = found_key.offset +
1051                                 btrfs_chunk_length(path->nodes[0], chunk);
1052                 }
1053         }
1054         ret = 0;
1055 error:
1056         btrfs_free_path(path);
1057         return ret;
1058 }
1059
1060 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1061 {
1062         int ret;
1063         struct btrfs_key key;
1064         struct btrfs_key found_key;
1065         struct btrfs_path *path;
1066
1067         root = root->fs_info->chunk_root;
1068
1069         path = btrfs_alloc_path();
1070         if (!path)
1071                 return -ENOMEM;
1072
1073         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1074         key.type = BTRFS_DEV_ITEM_KEY;
1075         key.offset = (u64)-1;
1076
1077         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1078         if (ret < 0)
1079                 goto error;
1080
1081         BUG_ON(ret == 0);
1082
1083         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1084                                   BTRFS_DEV_ITEM_KEY);
1085         if (ret) {
1086                 *objectid = 1;
1087         } else {
1088                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1089                                       path->slots[0]);
1090                 *objectid = found_key.offset + 1;
1091         }
1092         ret = 0;
1093 error:
1094         btrfs_free_path(path);
1095         return ret;
1096 }
1097
1098 /*
1099  * the device information is stored in the chunk root
1100  * the btrfs_device struct should be fully filled in
1101  */
1102 int btrfs_add_device(struct btrfs_trans_handle *trans,
1103                      struct btrfs_root *root,
1104                      struct btrfs_device *device)
1105 {
1106         int ret;
1107         struct btrfs_path *path;
1108         struct btrfs_dev_item *dev_item;
1109         struct extent_buffer *leaf;
1110         struct btrfs_key key;
1111         unsigned long ptr;
1112
1113         root = root->fs_info->chunk_root;
1114
1115         path = btrfs_alloc_path();
1116         if (!path)
1117                 return -ENOMEM;
1118
1119         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1120         key.type = BTRFS_DEV_ITEM_KEY;
1121         key.offset = device->devid;
1122
1123         ret = btrfs_insert_empty_item(trans, root, path, &key,
1124                                       sizeof(*dev_item));
1125         if (ret)
1126                 goto out;
1127
1128         leaf = path->nodes[0];
1129         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1130
1131         btrfs_set_device_id(leaf, dev_item, device->devid);
1132         btrfs_set_device_generation(leaf, dev_item, 0);
1133         btrfs_set_device_type(leaf, dev_item, device->type);
1134         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1135         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1136         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1137         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1138         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1139         btrfs_set_device_group(leaf, dev_item, 0);
1140         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1141         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1142         btrfs_set_device_start_offset(leaf, dev_item, 0);
1143
1144         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1145         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1146         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1147         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1148         btrfs_mark_buffer_dirty(leaf);
1149
1150         ret = 0;
1151 out:
1152         btrfs_free_path(path);
1153         return ret;
1154 }
1155
1156 static int btrfs_rm_dev_item(struct btrfs_root *root,
1157                              struct btrfs_device *device)
1158 {
1159         int ret;
1160         struct btrfs_path *path;
1161         struct btrfs_key key;
1162         struct btrfs_trans_handle *trans;
1163
1164         root = root->fs_info->chunk_root;
1165
1166         path = btrfs_alloc_path();
1167         if (!path)
1168                 return -ENOMEM;
1169
1170         trans = btrfs_start_transaction(root, 0);
1171         if (IS_ERR(trans)) {
1172                 btrfs_free_path(path);
1173                 return PTR_ERR(trans);
1174         }
1175         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1176         key.type = BTRFS_DEV_ITEM_KEY;
1177         key.offset = device->devid;
1178         lock_chunks(root);
1179
1180         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1181         if (ret < 0)
1182                 goto out;
1183
1184         if (ret > 0) {
1185                 ret = -ENOENT;
1186                 goto out;
1187         }
1188
1189         ret = btrfs_del_item(trans, root, path);
1190         if (ret)
1191                 goto out;
1192 out:
1193         btrfs_free_path(path);
1194         unlock_chunks(root);
1195         btrfs_commit_transaction(trans, root);
1196         return ret;
1197 }
1198
1199 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1200 {
1201         struct btrfs_device *device;
1202         struct btrfs_device *next_device;
1203         struct block_device *bdev;
1204         struct buffer_head *bh = NULL;
1205         struct btrfs_super_block *disk_super;
1206         u64 all_avail;
1207         u64 devid;
1208         u64 num_devices;
1209         u8 *dev_uuid;
1210         int ret = 0;
1211
1212         mutex_lock(&uuid_mutex);
1213         mutex_lock(&root->fs_info->volume_mutex);
1214
1215         all_avail = root->fs_info->avail_data_alloc_bits |
1216                 root->fs_info->avail_system_alloc_bits |
1217                 root->fs_info->avail_metadata_alloc_bits;
1218
1219         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1220             root->fs_info->fs_devices->num_devices <= 4) {
1221                 printk(KERN_ERR "btrfs: unable to go below four devices "
1222                        "on raid10\n");
1223                 ret = -EINVAL;
1224                 goto out;
1225         }
1226
1227         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1228             root->fs_info->fs_devices->num_devices <= 2) {
1229                 printk(KERN_ERR "btrfs: unable to go below two "
1230                        "devices on raid1\n");
1231                 ret = -EINVAL;
1232                 goto out;
1233         }
1234
1235         if (strcmp(device_path, "missing") == 0) {
1236                 struct list_head *devices;
1237                 struct btrfs_device *tmp;
1238
1239                 device = NULL;
1240                 devices = &root->fs_info->fs_devices->devices;
1241                 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1242                 list_for_each_entry(tmp, devices, dev_list) {
1243                         if (tmp->in_fs_metadata && !tmp->bdev) {
1244                                 device = tmp;
1245                                 break;
1246                         }
1247                 }
1248                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1249                 bdev = NULL;
1250                 bh = NULL;
1251                 disk_super = NULL;
1252                 if (!device) {
1253                         printk(KERN_ERR "btrfs: no missing devices found to "
1254                                "remove\n");
1255                         goto out;
1256                 }
1257         } else {
1258                 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1259                                           root->fs_info->bdev_holder);
1260                 if (IS_ERR(bdev)) {
1261                         ret = PTR_ERR(bdev);
1262                         goto out;
1263                 }
1264
1265                 set_blocksize(bdev, 4096);
1266                 bh = btrfs_read_dev_super(bdev);
1267                 if (!bh) {
1268                         ret = -EINVAL;
1269                         goto error_close;
1270                 }
1271                 disk_super = (struct btrfs_super_block *)bh->b_data;
1272                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1273                 dev_uuid = disk_super->dev_item.uuid;
1274                 device = btrfs_find_device(root, devid, dev_uuid,
1275                                            disk_super->fsid);
1276                 if (!device) {
1277                         ret = -ENOENT;
1278                         goto error_brelse;
1279                 }
1280         }
1281
1282         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1283                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1284                        "device\n");
1285                 ret = -EINVAL;
1286                 goto error_brelse;
1287         }
1288
1289         if (device->writeable) {
1290                 list_del_init(&device->dev_alloc_list);
1291                 root->fs_info->fs_devices->rw_devices--;
1292         }
1293
1294         ret = btrfs_shrink_device(device, 0);
1295         if (ret)
1296                 goto error_undo;
1297
1298         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1299         if (ret)
1300                 goto error_undo;
1301
1302         device->in_fs_metadata = 0;
1303
1304         /*
1305          * the device list mutex makes sure that we don't change
1306          * the device list while someone else is writing out all
1307          * the device supers.
1308          */
1309         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1310         list_del_init(&device->dev_list);
1311         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1312
1313         device->fs_devices->num_devices--;
1314
1315         if (device->missing)
1316                 root->fs_info->fs_devices->missing_devices--;
1317
1318         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1319                                  struct btrfs_device, dev_list);
1320         if (device->bdev == root->fs_info->sb->s_bdev)
1321                 root->fs_info->sb->s_bdev = next_device->bdev;
1322         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1323                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1324
1325         if (device->bdev) {
1326                 blkdev_put(device->bdev, device->mode);
1327                 device->bdev = NULL;
1328                 device->fs_devices->open_devices--;
1329         }
1330
1331         num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1332         btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1333
1334         if (device->fs_devices->open_devices == 0) {
1335                 struct btrfs_fs_devices *fs_devices;
1336                 fs_devices = root->fs_info->fs_devices;
1337                 while (fs_devices) {
1338                         if (fs_devices->seed == device->fs_devices)
1339                                 break;
1340                         fs_devices = fs_devices->seed;
1341                 }
1342                 fs_devices->seed = device->fs_devices->seed;
1343                 device->fs_devices->seed = NULL;
1344                 __btrfs_close_devices(device->fs_devices);
1345                 free_fs_devices(device->fs_devices);
1346         }
1347
1348         /*
1349          * at this point, the device is zero sized.  We want to
1350          * remove it from the devices list and zero out the old super
1351          */
1352         if (device->writeable) {
1353                 /* make sure this device isn't detected as part of
1354                  * the FS anymore
1355                  */
1356                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1357                 set_buffer_dirty(bh);
1358                 sync_dirty_buffer(bh);
1359         }
1360
1361         kfree(device->name);
1362         kfree(device);
1363         ret = 0;
1364
1365 error_brelse:
1366         brelse(bh);
1367 error_close:
1368         if (bdev)
1369                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1370 out:
1371         mutex_unlock(&root->fs_info->volume_mutex);
1372         mutex_unlock(&uuid_mutex);
1373         return ret;
1374 error_undo:
1375         if (device->writeable) {
1376                 list_add(&device->dev_alloc_list,
1377                          &root->fs_info->fs_devices->alloc_list);
1378                 root->fs_info->fs_devices->rw_devices++;
1379         }
1380         goto error_brelse;
1381 }
1382
1383 /*
1384  * does all the dirty work required for changing file system's UUID.
1385  */
1386 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1387                                 struct btrfs_root *root)
1388 {
1389         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1390         struct btrfs_fs_devices *old_devices;
1391         struct btrfs_fs_devices *seed_devices;
1392         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1393         struct btrfs_device *device;
1394         u64 super_flags;
1395
1396         BUG_ON(!mutex_is_locked(&uuid_mutex));
1397         if (!fs_devices->seeding)
1398                 return -EINVAL;
1399
1400         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1401         if (!seed_devices)
1402                 return -ENOMEM;
1403
1404         old_devices = clone_fs_devices(fs_devices);
1405         if (IS_ERR(old_devices)) {
1406                 kfree(seed_devices);
1407                 return PTR_ERR(old_devices);
1408         }
1409
1410         list_add(&old_devices->list, &fs_uuids);
1411
1412         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1413         seed_devices->opened = 1;
1414         INIT_LIST_HEAD(&seed_devices->devices);
1415         INIT_LIST_HEAD(&seed_devices->alloc_list);
1416         mutex_init(&seed_devices->device_list_mutex);
1417         list_splice_init(&fs_devices->devices, &seed_devices->devices);
1418         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1419         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1420                 device->fs_devices = seed_devices;
1421         }
1422
1423         fs_devices->seeding = 0;
1424         fs_devices->num_devices = 0;
1425         fs_devices->open_devices = 0;
1426         fs_devices->seed = seed_devices;
1427
1428         generate_random_uuid(fs_devices->fsid);
1429         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1430         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1431         super_flags = btrfs_super_flags(disk_super) &
1432                       ~BTRFS_SUPER_FLAG_SEEDING;
1433         btrfs_set_super_flags(disk_super, super_flags);
1434
1435         return 0;
1436 }
1437
1438 /*
1439  * strore the expected generation for seed devices in device items.
1440  */
1441 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1442                                struct btrfs_root *root)
1443 {
1444         struct btrfs_path *path;
1445         struct extent_buffer *leaf;
1446         struct btrfs_dev_item *dev_item;
1447         struct btrfs_device *device;
1448         struct btrfs_key key;
1449         u8 fs_uuid[BTRFS_UUID_SIZE];
1450         u8 dev_uuid[BTRFS_UUID_SIZE];
1451         u64 devid;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         if (!path)
1456                 return -ENOMEM;
1457
1458         root = root->fs_info->chunk_root;
1459         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1460         key.offset = 0;
1461         key.type = BTRFS_DEV_ITEM_KEY;
1462
1463         while (1) {
1464                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1465                 if (ret < 0)
1466                         goto error;
1467
1468                 leaf = path->nodes[0];
1469 next_slot:
1470                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1471                         ret = btrfs_next_leaf(root, path);
1472                         if (ret > 0)
1473                                 break;
1474                         if (ret < 0)
1475                                 goto error;
1476                         leaf = path->nodes[0];
1477                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1478                         btrfs_release_path(root, path);
1479                         continue;
1480                 }
1481
1482                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1483                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1484                     key.type != BTRFS_DEV_ITEM_KEY)
1485                         break;
1486
1487                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1488                                           struct btrfs_dev_item);
1489                 devid = btrfs_device_id(leaf, dev_item);
1490                 read_extent_buffer(leaf, dev_uuid,
1491                                    (unsigned long)btrfs_device_uuid(dev_item),
1492                                    BTRFS_UUID_SIZE);
1493                 read_extent_buffer(leaf, fs_uuid,
1494                                    (unsigned long)btrfs_device_fsid(dev_item),
1495                                    BTRFS_UUID_SIZE);
1496                 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1497                 BUG_ON(!device);
1498
1499                 if (device->fs_devices->seeding) {
1500                         btrfs_set_device_generation(leaf, dev_item,
1501                                                     device->generation);
1502                         btrfs_mark_buffer_dirty(leaf);
1503                 }
1504
1505                 path->slots[0]++;
1506                 goto next_slot;
1507         }
1508         ret = 0;
1509 error:
1510         btrfs_free_path(path);
1511         return ret;
1512 }
1513
1514 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1515 {
1516         struct btrfs_trans_handle *trans;
1517         struct btrfs_device *device;
1518         struct block_device *bdev;
1519         struct list_head *devices;
1520         struct super_block *sb = root->fs_info->sb;
1521         u64 total_bytes;
1522         int seeding_dev = 0;
1523         int ret = 0;
1524
1525         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1526                 return -EINVAL;
1527
1528         bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1529                                   root->fs_info->bdev_holder);
1530         if (IS_ERR(bdev))
1531                 return PTR_ERR(bdev);
1532
1533         if (root->fs_info->fs_devices->seeding) {
1534                 seeding_dev = 1;
1535                 down_write(&sb->s_umount);
1536                 mutex_lock(&uuid_mutex);
1537         }
1538
1539         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1540         mutex_lock(&root->fs_info->volume_mutex);
1541
1542         devices = &root->fs_info->fs_devices->devices;
1543         /*
1544          * we have the volume lock, so we don't need the extra
1545          * device list mutex while reading the list here.
1546          */
1547         list_for_each_entry(device, devices, dev_list) {
1548                 if (device->bdev == bdev) {
1549                         ret = -EEXIST;
1550                         goto error;
1551                 }
1552         }
1553
1554         device = kzalloc(sizeof(*device), GFP_NOFS);
1555         if (!device) {
1556                 /* we can safely leave the fs_devices entry around */
1557                 ret = -ENOMEM;
1558                 goto error;
1559         }
1560
1561         device->name = kstrdup(device_path, GFP_NOFS);
1562         if (!device->name) {
1563                 kfree(device);
1564                 ret = -ENOMEM;
1565                 goto error;
1566         }
1567
1568         ret = find_next_devid(root, &device->devid);
1569         if (ret) {
1570                 kfree(device->name);
1571                 kfree(device);
1572                 goto error;
1573         }
1574
1575         trans = btrfs_start_transaction(root, 0);
1576         if (IS_ERR(trans)) {
1577                 kfree(device->name);
1578                 kfree(device);
1579                 ret = PTR_ERR(trans);
1580                 goto error;
1581         }
1582
1583         lock_chunks(root);
1584
1585         device->writeable = 1;
1586         device->work.func = pending_bios_fn;
1587         generate_random_uuid(device->uuid);
1588         spin_lock_init(&device->io_lock);
1589         device->generation = trans->transid;
1590         device->io_width = root->sectorsize;
1591         device->io_align = root->sectorsize;
1592         device->sector_size = root->sectorsize;
1593         device->total_bytes = i_size_read(bdev->bd_inode);
1594         device->disk_total_bytes = device->total_bytes;
1595         device->dev_root = root->fs_info->dev_root;
1596         device->bdev = bdev;
1597         device->in_fs_metadata = 1;
1598         device->mode = FMODE_EXCL;
1599         set_blocksize(device->bdev, 4096);
1600
1601         if (seeding_dev) {
1602                 sb->s_flags &= ~MS_RDONLY;
1603                 ret = btrfs_prepare_sprout(trans, root);
1604                 BUG_ON(ret);
1605         }
1606
1607         device->fs_devices = root->fs_info->fs_devices;
1608
1609         /*
1610          * we don't want write_supers to jump in here with our device
1611          * half setup
1612          */
1613         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1614         list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1615         list_add(&device->dev_alloc_list,
1616                  &root->fs_info->fs_devices->alloc_list);
1617         root->fs_info->fs_devices->num_devices++;
1618         root->fs_info->fs_devices->open_devices++;
1619         root->fs_info->fs_devices->rw_devices++;
1620         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1621
1622         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1623                 root->fs_info->fs_devices->rotating = 1;
1624
1625         total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1626         btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1627                                     total_bytes + device->total_bytes);
1628
1629         total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1630         btrfs_set_super_num_devices(&root->fs_info->super_copy,
1631                                     total_bytes + 1);
1632         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1633
1634         if (seeding_dev) {
1635                 ret = init_first_rw_device(trans, root, device);
1636                 BUG_ON(ret);
1637                 ret = btrfs_finish_sprout(trans, root);
1638                 BUG_ON(ret);
1639         } else {
1640                 ret = btrfs_add_device(trans, root, device);
1641         }
1642
1643         /*
1644          * we've got more storage, clear any full flags on the space
1645          * infos
1646          */
1647         btrfs_clear_space_info_full(root->fs_info);
1648
1649         unlock_chunks(root);
1650         btrfs_commit_transaction(trans, root);
1651
1652         if (seeding_dev) {
1653                 mutex_unlock(&uuid_mutex);
1654                 up_write(&sb->s_umount);
1655
1656                 ret = btrfs_relocate_sys_chunks(root);
1657                 BUG_ON(ret);
1658         }
1659 out:
1660         mutex_unlock(&root->fs_info->volume_mutex);
1661         return ret;
1662 error:
1663         blkdev_put(bdev, FMODE_EXCL);
1664         if (seeding_dev) {
1665                 mutex_unlock(&uuid_mutex);
1666                 up_write(&sb->s_umount);
1667         }
1668         goto out;
1669 }
1670
1671 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1672                                         struct btrfs_device *device)
1673 {
1674         int ret;
1675         struct btrfs_path *path;
1676         struct btrfs_root *root;
1677         struct btrfs_dev_item *dev_item;
1678         struct extent_buffer *leaf;
1679         struct btrfs_key key;
1680
1681         root = device->dev_root->fs_info->chunk_root;
1682
1683         path = btrfs_alloc_path();
1684         if (!path)
1685                 return -ENOMEM;
1686
1687         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1688         key.type = BTRFS_DEV_ITEM_KEY;
1689         key.offset = device->devid;
1690
1691         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1692         if (ret < 0)
1693                 goto out;
1694
1695         if (ret > 0) {
1696                 ret = -ENOENT;
1697                 goto out;
1698         }
1699
1700         leaf = path->nodes[0];
1701         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1702
1703         btrfs_set_device_id(leaf, dev_item, device->devid);
1704         btrfs_set_device_type(leaf, dev_item, device->type);
1705         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1706         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1707         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1708         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1709         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1710         btrfs_mark_buffer_dirty(leaf);
1711
1712 out:
1713         btrfs_free_path(path);
1714         return ret;
1715 }
1716
1717 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1718                       struct btrfs_device *device, u64 new_size)
1719 {
1720         struct btrfs_super_block *super_copy =
1721                 &device->dev_root->fs_info->super_copy;
1722         u64 old_total = btrfs_super_total_bytes(super_copy);
1723         u64 diff = new_size - device->total_bytes;
1724
1725         if (!device->writeable)
1726                 return -EACCES;
1727         if (new_size <= device->total_bytes)
1728                 return -EINVAL;
1729
1730         btrfs_set_super_total_bytes(super_copy, old_total + diff);
1731         device->fs_devices->total_rw_bytes += diff;
1732
1733         device->total_bytes = new_size;
1734         device->disk_total_bytes = new_size;
1735         btrfs_clear_space_info_full(device->dev_root->fs_info);
1736
1737         return btrfs_update_device(trans, device);
1738 }
1739
1740 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1741                       struct btrfs_device *device, u64 new_size)
1742 {
1743         int ret;
1744         lock_chunks(device->dev_root);
1745         ret = __btrfs_grow_device(trans, device, new_size);
1746         unlock_chunks(device->dev_root);
1747         return ret;
1748 }
1749
1750 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1751                             struct btrfs_root *root,
1752                             u64 chunk_tree, u64 chunk_objectid,
1753                             u64 chunk_offset)
1754 {
1755         int ret;
1756         struct btrfs_path *path;
1757         struct btrfs_key key;
1758
1759         root = root->fs_info->chunk_root;
1760         path = btrfs_alloc_path();
1761         if (!path)
1762                 return -ENOMEM;
1763
1764         key.objectid = chunk_objectid;
1765         key.offset = chunk_offset;
1766         key.type = BTRFS_CHUNK_ITEM_KEY;
1767
1768         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1769         BUG_ON(ret);
1770
1771         ret = btrfs_del_item(trans, root, path);
1772
1773         btrfs_free_path(path);
1774         return ret;
1775 }
1776
1777 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1778                         chunk_offset)
1779 {
1780         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1781         struct btrfs_disk_key *disk_key;
1782         struct btrfs_chunk *chunk;
1783         u8 *ptr;
1784         int ret = 0;
1785         u32 num_stripes;
1786         u32 array_size;
1787         u32 len = 0;
1788         u32 cur;
1789         struct btrfs_key key;
1790
1791         array_size = btrfs_super_sys_array_size(super_copy);
1792
1793         ptr = super_copy->sys_chunk_array;
1794         cur = 0;
1795
1796         while (cur < array_size) {
1797                 disk_key = (struct btrfs_disk_key *)ptr;
1798                 btrfs_disk_key_to_cpu(&key, disk_key);
1799
1800                 len = sizeof(*disk_key);
1801
1802                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1803                         chunk = (struct btrfs_chunk *)(ptr + len);
1804                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1805                         len += btrfs_chunk_item_size(num_stripes);
1806                 } else {
1807                         ret = -EIO;
1808                         break;
1809                 }
1810                 if (key.objectid == chunk_objectid &&
1811                     key.offset == chunk_offset) {
1812                         memmove(ptr, ptr + len, array_size - (cur + len));
1813                         array_size -= len;
1814                         btrfs_set_super_sys_array_size(super_copy, array_size);
1815                 } else {
1816                         ptr += len;
1817                         cur += len;
1818                 }
1819         }
1820         return ret;
1821 }
1822
1823 static int btrfs_relocate_chunk(struct btrfs_root *root,
1824                          u64 chunk_tree, u64 chunk_objectid,
1825                          u64 chunk_offset)
1826 {
1827         struct extent_map_tree *em_tree;
1828         struct btrfs_root *extent_root;
1829         struct btrfs_trans_handle *trans;
1830         struct extent_map *em;
1831         struct map_lookup *map;
1832         int ret;
1833         int i;
1834
1835         root = root->fs_info->chunk_root;
1836         extent_root = root->fs_info->extent_root;
1837         em_tree = &root->fs_info->mapping_tree.map_tree;
1838
1839         ret = btrfs_can_relocate(extent_root, chunk_offset);
1840         if (ret)
1841                 return -ENOSPC;
1842
1843         /* step one, relocate all the extents inside this chunk */
1844         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1845         if (ret)
1846                 return ret;
1847
1848         trans = btrfs_start_transaction(root, 0);
1849         BUG_ON(IS_ERR(trans));
1850
1851         lock_chunks(root);
1852
1853         /*
1854          * step two, delete the device extents and the
1855          * chunk tree entries
1856          */
1857         read_lock(&em_tree->lock);
1858         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1859         read_unlock(&em_tree->lock);
1860
1861         BUG_ON(em->start > chunk_offset ||
1862                em->start + em->len < chunk_offset);
1863         map = (struct map_lookup *)em->bdev;
1864
1865         for (i = 0; i < map->num_stripes; i++) {
1866                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1867                                             map->stripes[i].physical);
1868                 BUG_ON(ret);
1869
1870                 if (map->stripes[i].dev) {
1871                         ret = btrfs_update_device(trans, map->stripes[i].dev);
1872                         BUG_ON(ret);
1873                 }
1874         }
1875         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1876                                chunk_offset);
1877
1878         BUG_ON(ret);
1879
1880         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1881
1882         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1883                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1884                 BUG_ON(ret);
1885         }
1886
1887         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1888         BUG_ON(ret);
1889
1890         write_lock(&em_tree->lock);
1891         remove_extent_mapping(em_tree, em);
1892         write_unlock(&em_tree->lock);
1893
1894         kfree(map);
1895         em->bdev = NULL;
1896
1897         /* once for the tree */
1898         free_extent_map(em);
1899         /* once for us */
1900         free_extent_map(em);
1901
1902         unlock_chunks(root);
1903         btrfs_end_transaction(trans, root);
1904         return 0;
1905 }
1906
1907 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1908 {
1909         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1910         struct btrfs_path *path;
1911         struct extent_buffer *leaf;
1912         struct btrfs_chunk *chunk;
1913         struct btrfs_key key;
1914         struct btrfs_key found_key;
1915         u64 chunk_tree = chunk_root->root_key.objectid;
1916         u64 chunk_type;
1917         bool retried = false;
1918         int failed = 0;
1919         int ret;
1920
1921         path = btrfs_alloc_path();
1922         if (!path)
1923                 return -ENOMEM;
1924
1925 again:
1926         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1927         key.offset = (u64)-1;
1928         key.type = BTRFS_CHUNK_ITEM_KEY;
1929
1930         while (1) {
1931                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1932                 if (ret < 0)
1933                         goto error;
1934                 BUG_ON(ret == 0);
1935
1936                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1937                                           key.type);
1938                 if (ret < 0)
1939                         goto error;
1940                 if (ret > 0)
1941                         break;
1942
1943                 leaf = path->nodes[0];
1944                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1945
1946                 chunk = btrfs_item_ptr(leaf, path->slots[0],
1947                                        struct btrfs_chunk);
1948                 chunk_type = btrfs_chunk_type(leaf, chunk);
1949                 btrfs_release_path(chunk_root, path);
1950
1951                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1952                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1953                                                    found_key.objectid,
1954                                                    found_key.offset);
1955                         if (ret == -ENOSPC)
1956                                 failed++;
1957                         else if (ret)
1958                                 BUG();
1959                 }
1960
1961                 if (found_key.offset == 0)
1962                         break;
1963                 key.offset = found_key.offset - 1;
1964         }
1965         ret = 0;
1966         if (failed && !retried) {
1967                 failed = 0;
1968                 retried = true;
1969                 goto again;
1970         } else if (failed && retried) {
1971                 WARN_ON(1);
1972                 ret = -ENOSPC;
1973         }
1974 error:
1975         btrfs_free_path(path);
1976         return ret;
1977 }
1978
1979 static u64 div_factor(u64 num, int factor)
1980 {
1981         if (factor == 10)
1982                 return num;
1983         num *= factor;
1984         do_div(num, 10);
1985         return num;
1986 }
1987
1988 int btrfs_balance(struct btrfs_root *dev_root)
1989 {
1990         int ret;
1991         struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1992         struct btrfs_device *device;
1993         u64 old_size;
1994         u64 size_to_free;
1995         struct btrfs_path *path;
1996         struct btrfs_key key;
1997         struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1998         struct btrfs_trans_handle *trans;
1999         struct btrfs_key found_key;
2000
2001         if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2002                 return -EROFS;
2003
2004         if (!capable(CAP_SYS_ADMIN))
2005                 return -EPERM;
2006
2007         mutex_lock(&dev_root->fs_info->volume_mutex);
2008         dev_root = dev_root->fs_info->dev_root;
2009
2010         /* step one make some room on all the devices */
2011         list_for_each_entry(device, devices, dev_list) {
2012                 old_size = device->total_bytes;
2013                 size_to_free = div_factor(old_size, 1);
2014                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2015                 if (!device->writeable ||
2016                     device->total_bytes - device->bytes_used > size_to_free)
2017                         continue;
2018
2019                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2020                 if (ret == -ENOSPC)
2021                         break;
2022                 BUG_ON(ret);
2023
2024                 trans = btrfs_start_transaction(dev_root, 0);
2025                 BUG_ON(IS_ERR(trans));
2026
2027                 ret = btrfs_grow_device(trans, device, old_size);
2028                 BUG_ON(ret);
2029
2030                 btrfs_end_transaction(trans, dev_root);
2031         }
2032
2033         /* step two, relocate all the chunks */
2034         path = btrfs_alloc_path();
2035         BUG_ON(!path);
2036
2037         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2038         key.offset = (u64)-1;
2039         key.type = BTRFS_CHUNK_ITEM_KEY;
2040
2041         while (1) {
2042                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2043                 if (ret < 0)
2044                         goto error;
2045
2046                 /*
2047                  * this shouldn't happen, it means the last relocate
2048                  * failed
2049                  */
2050                 if (ret == 0)
2051                         break;
2052
2053                 ret = btrfs_previous_item(chunk_root, path, 0,
2054                                           BTRFS_CHUNK_ITEM_KEY);
2055                 if (ret)
2056                         break;
2057
2058                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2059                                       path->slots[0]);
2060                 if (found_key.objectid != key.objectid)
2061                         break;
2062
2063                 /* chunk zero is special */
2064                 if (found_key.offset == 0)
2065                         break;
2066
2067                 btrfs_release_path(chunk_root, path);
2068                 ret = btrfs_relocate_chunk(chunk_root,
2069                                            chunk_root->root_key.objectid,
2070                                            found_key.objectid,
2071                                            found_key.offset);
2072                 BUG_ON(ret && ret != -ENOSPC);
2073                 key.offset = found_key.offset - 1;
2074         }
2075         ret = 0;
2076 error:
2077         btrfs_free_path(path);
2078         mutex_unlock(&dev_root->fs_info->volume_mutex);
2079         return ret;
2080 }
2081
2082 /*
2083  * shrinking a device means finding all of the device extents past
2084  * the new size, and then following the back refs to the chunks.
2085  * The chunk relocation code actually frees the device extent
2086  */
2087 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2088 {
2089         struct btrfs_trans_handle *trans;
2090         struct btrfs_root *root = device->dev_root;
2091         struct btrfs_dev_extent *dev_extent = NULL;
2092         struct btrfs_path *path;
2093         u64 length;
2094         u64 chunk_tree;
2095         u64 chunk_objectid;
2096         u64 chunk_offset;
2097         int ret;
2098         int slot;
2099         int failed = 0;
2100         bool retried = false;
2101         struct extent_buffer *l;
2102         struct btrfs_key key;
2103         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2104         u64 old_total = btrfs_super_total_bytes(super_copy);
2105         u64 old_size = device->total_bytes;
2106         u64 diff = device->total_bytes - new_size;
2107
2108         if (new_size >= device->total_bytes)
2109                 return -EINVAL;
2110
2111         path = btrfs_alloc_path();
2112         if (!path)
2113                 return -ENOMEM;
2114
2115         path->reada = 2;
2116
2117         lock_chunks(root);
2118
2119         device->total_bytes = new_size;
2120         if (device->writeable)
2121                 device->fs_devices->total_rw_bytes -= diff;
2122         unlock_chunks(root);
2123
2124 again:
2125         key.objectid = device->devid;
2126         key.offset = (u64)-1;
2127         key.type = BTRFS_DEV_EXTENT_KEY;
2128
2129         while (1) {
2130                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2131                 if (ret < 0)
2132                         goto done;
2133
2134                 ret = btrfs_previous_item(root, path, 0, key.type);
2135                 if (ret < 0)
2136                         goto done;
2137                 if (ret) {
2138                         ret = 0;
2139                         btrfs_release_path(root, path);
2140                         break;
2141                 }
2142
2143                 l = path->nodes[0];
2144                 slot = path->slots[0];
2145                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2146
2147                 if (key.objectid != device->devid) {
2148                         btrfs_release_path(root, path);
2149                         break;
2150                 }
2151
2152                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2153                 length = btrfs_dev_extent_length(l, dev_extent);
2154
2155                 if (key.offset + length <= new_size) {
2156                         btrfs_release_path(root, path);
2157                         break;
2158                 }
2159
2160                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2161                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2162                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2163                 btrfs_release_path(root, path);
2164
2165                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2166                                            chunk_offset);
2167                 if (ret && ret != -ENOSPC)
2168                         goto done;
2169                 if (ret == -ENOSPC)
2170                         failed++;
2171                 key.offset -= 1;
2172         }
2173
2174         if (failed && !retried) {
2175                 failed = 0;
2176                 retried = true;
2177                 goto again;
2178         } else if (failed && retried) {
2179                 ret = -ENOSPC;
2180                 lock_chunks(root);
2181
2182                 device->total_bytes = old_size;
2183                 if (device->writeable)
2184                         device->fs_devices->total_rw_bytes += diff;
2185                 unlock_chunks(root);
2186                 goto done;
2187         }
2188
2189         /* Shrinking succeeded, else we would be at "done". */
2190         trans = btrfs_start_transaction(root, 0);
2191         if (IS_ERR(trans)) {
2192                 ret = PTR_ERR(trans);
2193                 goto done;
2194         }
2195
2196         lock_chunks(root);
2197
2198         device->disk_total_bytes = new_size;
2199         /* Now btrfs_update_device() will change the on-disk size. */
2200         ret = btrfs_update_device(trans, device);
2201         if (ret) {
2202                 unlock_chunks(root);
2203                 btrfs_end_transaction(trans, root);
2204                 goto done;
2205         }
2206         WARN_ON(diff > old_total);
2207         btrfs_set_super_total_bytes(super_copy, old_total - diff);
2208         unlock_chunks(root);
2209         btrfs_end_transaction(trans, root);
2210 done:
2211         btrfs_free_path(path);
2212         return ret;
2213 }
2214
2215 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2216                            struct btrfs_root *root,
2217                            struct btrfs_key *key,
2218                            struct btrfs_chunk *chunk, int item_size)
2219 {
2220         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2221         struct btrfs_disk_key disk_key;
2222         u32 array_size;
2223         u8 *ptr;
2224
2225         array_size = btrfs_super_sys_array_size(super_copy);
2226         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2227                 return -EFBIG;
2228
2229         ptr = super_copy->sys_chunk_array + array_size;
2230         btrfs_cpu_key_to_disk(&disk_key, key);
2231         memcpy(ptr, &disk_key, sizeof(disk_key));
2232         ptr += sizeof(disk_key);
2233         memcpy(ptr, chunk, item_size);
2234         item_size += sizeof(disk_key);
2235         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2236         return 0;
2237 }
2238
2239 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2240                                         int num_stripes, int sub_stripes)
2241 {
2242         if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2243                 return calc_size;
2244         else if (type & BTRFS_BLOCK_GROUP_RAID10)
2245                 return calc_size * (num_stripes / sub_stripes);
2246         else
2247                 return calc_size * num_stripes;
2248 }
2249
2250 /* Used to sort the devices by max_avail(descending sort) */
2251 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2252 {
2253         if (((struct btrfs_device_info *)dev_info1)->max_avail >
2254             ((struct btrfs_device_info *)dev_info2)->max_avail)
2255                 return -1;
2256         else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2257                  ((struct btrfs_device_info *)dev_info2)->max_avail)
2258                 return 1;
2259         else
2260                 return 0;
2261 }
2262
2263 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2264                                  int *num_stripes, int *min_stripes,
2265                                  int *sub_stripes)
2266 {
2267         *num_stripes = 1;
2268         *min_stripes = 1;
2269         *sub_stripes = 0;
2270
2271         if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2272                 *num_stripes = fs_devices->rw_devices;
2273                 *min_stripes = 2;
2274         }
2275         if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2276                 *num_stripes = 2;
2277                 *min_stripes = 2;
2278         }
2279         if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2280                 if (fs_devices->rw_devices < 2)
2281                         return -ENOSPC;
2282                 *num_stripes = 2;
2283                 *min_stripes = 2;
2284         }
2285         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2286                 *num_stripes = fs_devices->rw_devices;
2287                 if (*num_stripes < 4)
2288                         return -ENOSPC;
2289                 *num_stripes &= ~(u32)1;
2290                 *sub_stripes = 2;
2291                 *min_stripes = 4;
2292         }
2293
2294         return 0;
2295 }
2296
2297 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2298                                     u64 proposed_size, u64 type,
2299                                     int num_stripes, int small_stripe)
2300 {
2301         int min_stripe_size = 1 * 1024 * 1024;
2302         u64 calc_size = proposed_size;
2303         u64 max_chunk_size = calc_size;
2304         int ncopies = 1;
2305
2306         if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2307                     BTRFS_BLOCK_GROUP_DUP |
2308                     BTRFS_BLOCK_GROUP_RAID10))
2309                 ncopies = 2;
2310
2311         if (type & BTRFS_BLOCK_GROUP_DATA) {
2312                 max_chunk_size = 10 * calc_size;
2313                 min_stripe_size = 64 * 1024 * 1024;
2314         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2315                 max_chunk_size = 256 * 1024 * 1024;
2316                 min_stripe_size = 32 * 1024 * 1024;
2317         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2318                 calc_size = 8 * 1024 * 1024;
2319                 max_chunk_size = calc_size * 2;
2320                 min_stripe_size = 1 * 1024 * 1024;
2321         }
2322
2323         /* we don't want a chunk larger than 10% of writeable space */
2324         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2325                              max_chunk_size);
2326
2327         if (calc_size * num_stripes > max_chunk_size * ncopies) {
2328                 calc_size = max_chunk_size * ncopies;
2329                 do_div(calc_size, num_stripes);
2330                 do_div(calc_size, BTRFS_STRIPE_LEN);
2331                 calc_size *= BTRFS_STRIPE_LEN;
2332         }
2333
2334         /* we don't want tiny stripes */
2335         if (!small_stripe)
2336                 calc_size = max_t(u64, min_stripe_size, calc_size);
2337
2338         /*
2339          * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2340          * we end up with something bigger than a stripe
2341          */
2342         calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2343
2344         do_div(calc_size, BTRFS_STRIPE_LEN);
2345         calc_size *= BTRFS_STRIPE_LEN;
2346
2347         return calc_size;
2348 }
2349
2350 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2351                                                       int num_stripes)
2352 {
2353         struct map_lookup *new;
2354         size_t len = map_lookup_size(num_stripes);
2355
2356         BUG_ON(map->num_stripes < num_stripes);
2357
2358         if (map->num_stripes == num_stripes)
2359                 return map;
2360
2361         new = kmalloc(len, GFP_NOFS);
2362         if (!new) {
2363                 /* just change map->num_stripes */
2364                 map->num_stripes = num_stripes;
2365                 return map;
2366         }
2367
2368         memcpy(new, map, len);
2369         new->num_stripes = num_stripes;
2370         kfree(map);
2371         return new;
2372 }
2373
2374 /*
2375  * helper to allocate device space from btrfs_device_info, in which we stored
2376  * max free space information of every device. It is used when we can not
2377  * allocate chunks by default size.
2378  *
2379  * By this helper, we can allocate a new chunk as larger as possible.
2380  */
2381 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2382                                     struct btrfs_fs_devices *fs_devices,
2383                                     struct btrfs_device_info *devices,
2384                                     int nr_device, u64 type,
2385                                     struct map_lookup **map_lookup,
2386                                     int min_stripes, u64 *stripe_size)
2387 {
2388         int i, index, sort_again = 0;
2389         int min_devices = min_stripes;
2390         u64 max_avail, min_free;
2391         struct map_lookup *map = *map_lookup;
2392         int ret;
2393
2394         if (nr_device < min_stripes)
2395                 return -ENOSPC;
2396
2397         btrfs_descending_sort_devices(devices, nr_device);
2398
2399         max_avail = devices[0].max_avail;
2400         if (!max_avail)
2401                 return -ENOSPC;
2402
2403         for (i = 0; i < nr_device; i++) {
2404                 /*
2405                  * if dev_offset = 0, it means the free space of this device
2406                  * is less than what we need, and we didn't search max avail
2407                  * extent on this device, so do it now.
2408                  */
2409                 if (!devices[i].dev_offset) {
2410                         ret = find_free_dev_extent(trans, devices[i].dev,
2411                                                    max_avail,
2412                                                    &devices[i].dev_offset,
2413                                                    &devices[i].max_avail);
2414                         if (ret != 0 && ret != -ENOSPC)
2415                                 return ret;
2416                         sort_again = 1;
2417                 }
2418         }
2419
2420         /* we update the max avail free extent of each devices, sort again */
2421         if (sort_again)
2422                 btrfs_descending_sort_devices(devices, nr_device);
2423
2424         if (type & BTRFS_BLOCK_GROUP_DUP)
2425                 min_devices = 1;
2426
2427         if (!devices[min_devices - 1].max_avail)
2428                 return -ENOSPC;
2429
2430         max_avail = devices[min_devices - 1].max_avail;
2431         if (type & BTRFS_BLOCK_GROUP_DUP)
2432                 do_div(max_avail, 2);
2433
2434         max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2435                                              min_stripes, 1);
2436         if (type & BTRFS_BLOCK_GROUP_DUP)
2437                 min_free = max_avail * 2;
2438         else
2439                 min_free = max_avail;
2440
2441         if (min_free > devices[min_devices - 1].max_avail)
2442                 return -ENOSPC;
2443
2444         map = __shrink_map_lookup_stripes(map, min_stripes);
2445         *stripe_size = max_avail;
2446
2447         index = 0;
2448         for (i = 0; i < min_stripes; i++) {
2449                 map->stripes[i].dev = devices[index].dev;
2450                 map->stripes[i].physical = devices[index].dev_offset;
2451                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2452                         i++;
2453                         map->stripes[i].dev = devices[index].dev;
2454                         map->stripes[i].physical = devices[index].dev_offset +
2455                                                    max_avail;
2456                 }
2457                 index++;
2458         }
2459         *map_lookup = map;
2460
2461         return 0;
2462 }
2463
2464 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2465                                struct btrfs_root *extent_root,
2466                                struct map_lookup **map_ret,
2467                                u64 *num_bytes, u64 *stripe_size,
2468                                u64 start, u64 type)
2469 {
2470         struct btrfs_fs_info *info = extent_root->fs_info;
2471         struct btrfs_device *device = NULL;
2472         struct btrfs_fs_devices *fs_devices = info->fs_devices;
2473         struct list_head *cur;
2474         struct map_lookup *map;
2475         struct extent_map_tree *em_tree;
2476         struct extent_map *em;
2477         struct btrfs_device_info *devices_info;
2478         struct list_head private_devs;
2479         u64 calc_size = 1024 * 1024 * 1024;
2480         u64 min_free;
2481         u64 avail;
2482         u64 dev_offset;
2483         int num_stripes;
2484         int min_stripes;
2485         int sub_stripes;
2486         int min_devices;        /* the min number of devices we need */
2487         int i;
2488         int ret;
2489         int index;
2490
2491         if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2492             (type & BTRFS_BLOCK_GROUP_DUP)) {
2493                 WARN_ON(1);
2494                 type &= ~BTRFS_BLOCK_GROUP_DUP;
2495         }
2496         if (list_empty(&fs_devices->alloc_list))
2497                 return -ENOSPC;
2498
2499         ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2500                                     &min_stripes, &sub_stripes);
2501         if (ret)
2502                 return ret;
2503
2504         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2505                                GFP_NOFS);
2506         if (!devices_info)
2507                 return -ENOMEM;
2508
2509         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2510         if (!map) {
2511                 ret = -ENOMEM;
2512                 goto error;
2513         }
2514         map->num_stripes = num_stripes;
2515
2516         cur = fs_devices->alloc_list.next;
2517         index = 0;
2518         i = 0;
2519
2520         calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2521                                              num_stripes, 0);
2522
2523         if (type & BTRFS_BLOCK_GROUP_DUP) {
2524                 min_free = calc_size * 2;
2525                 min_devices = 1;
2526         } else {
2527                 min_free = calc_size;
2528                 min_devices = min_stripes;
2529         }
2530
2531         INIT_LIST_HEAD(&private_devs);
2532         while (index < num_stripes) {
2533                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2534                 BUG_ON(!device->writeable);
2535                 if (device->total_bytes > device->bytes_used)
2536                         avail = device->total_bytes - device->bytes_used;
2537                 else
2538                         avail = 0;
2539                 cur = cur->next;
2540
2541                 if (device->in_fs_metadata && avail >= min_free) {
2542                         ret = find_free_dev_extent(trans, device, min_free,
2543                                                    &devices_info[i].dev_offset,
2544                                                    &devices_info[i].max_avail);
2545                         if (ret == 0) {
2546                                 list_move_tail(&device->dev_alloc_list,
2547                                                &private_devs);
2548                                 map->stripes[index].dev = device;
2549                                 map->stripes[index].physical =
2550                                                 devices_info[i].dev_offset;
2551                                 index++;
2552                                 if (type & BTRFS_BLOCK_GROUP_DUP) {
2553                                         map->stripes[index].dev = device;
2554                                         map->stripes[index].physical =
2555                                                 devices_info[i].dev_offset +
2556                                                 calc_size;
2557                                         index++;
2558                                 }
2559                         } else if (ret != -ENOSPC)
2560                                 goto error;
2561
2562                         devices_info[i].dev = device;
2563                         i++;
2564                 } else if (device->in_fs_metadata &&
2565                            avail >= BTRFS_STRIPE_LEN) {
2566                         devices_info[i].dev = device;
2567                         devices_info[i].max_avail = avail;
2568                         i++;
2569                 }
2570
2571                 if (cur == &fs_devices->alloc_list)
2572                         break;
2573         }
2574
2575         list_splice(&private_devs, &fs_devices->alloc_list);
2576         if (index < num_stripes) {
2577                 if (index >= min_stripes) {
2578                         num_stripes = index;
2579                         if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2580                                 num_stripes /= sub_stripes;
2581                                 num_stripes *= sub_stripes;
2582                         }
2583
2584                         map = __shrink_map_lookup_stripes(map, num_stripes);
2585                 } else if (i >= min_devices) {
2586                         ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2587                                                        devices_info, i, type,
2588                                                        &map, min_stripes,
2589                                                        &calc_size);
2590                         if (ret)
2591                                 goto error;
2592                 } else {
2593                         ret = -ENOSPC;
2594                         goto error;
2595                 }
2596         }
2597         map->sector_size = extent_root->sectorsize;
2598         map->stripe_len = BTRFS_STRIPE_LEN;
2599         map->io_align = BTRFS_STRIPE_LEN;
2600         map->io_width = BTRFS_STRIPE_LEN;
2601         map->type = type;
2602         map->sub_stripes = sub_stripes;
2603
2604         *map_ret = map;
2605         *stripe_size = calc_size;
2606         *num_bytes = chunk_bytes_by_type(type, calc_size,
2607                                          map->num_stripes, sub_stripes);
2608
2609         trace_btrfs_chunk_alloc(info->chunk_root, map, start, *num_bytes);
2610
2611         em = alloc_extent_map(GFP_NOFS);
2612         if (!em) {
2613                 ret = -ENOMEM;
2614                 goto error;
2615         }
2616         em->bdev = (struct block_device *)map;
2617         em->start = start;
2618         em->len = *num_bytes;
2619         em->block_start = 0;
2620         em->block_len = em->len;
2621
2622         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2623         write_lock(&em_tree->lock);
2624         ret = add_extent_mapping(em_tree, em);
2625         write_unlock(&em_tree->lock);
2626         BUG_ON(ret);
2627         free_extent_map(em);
2628
2629         ret = btrfs_make_block_group(trans, extent_root, 0, type,
2630                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2631                                      start, *num_bytes);
2632         BUG_ON(ret);
2633
2634         index = 0;
2635         while (index < map->num_stripes) {
2636                 device = map->stripes[index].dev;
2637                 dev_offset = map->stripes[index].physical;
2638
2639                 ret = btrfs_alloc_dev_extent(trans, device,
2640                                 info->chunk_root->root_key.objectid,
2641                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2642                                 start, dev_offset, calc_size);
2643                 BUG_ON(ret);
2644                 index++;
2645         }
2646
2647         kfree(devices_info);
2648         return 0;
2649
2650 error:
2651         kfree(map);
2652         kfree(devices_info);
2653         return ret;
2654 }
2655
2656 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2657                                 struct btrfs_root *extent_root,
2658                                 struct map_lookup *map, u64 chunk_offset,
2659                                 u64 chunk_size, u64 stripe_size)
2660 {
2661         u64 dev_offset;
2662         struct btrfs_key key;
2663         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2664         struct btrfs_device *device;
2665         struct btrfs_chunk *chunk;
2666         struct btrfs_stripe *stripe;
2667         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2668         int index = 0;
2669         int ret;
2670
2671         chunk = kzalloc(item_size, GFP_NOFS);
2672         if (!chunk)
2673                 return -ENOMEM;
2674
2675         index = 0;
2676         while (index < map->num_stripes) {
2677                 device = map->stripes[index].dev;
2678                 device->bytes_used += stripe_size;
2679                 ret = btrfs_update_device(trans, device);
2680                 BUG_ON(ret);
2681                 index++;
2682         }
2683
2684         index = 0;
2685         stripe = &chunk->stripe;
2686         while (index < map->num_stripes) {
2687                 device = map->stripes[index].dev;
2688                 dev_offset = map->stripes[index].physical;
2689
2690                 btrfs_set_stack_stripe_devid(stripe, device->devid);
2691                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2692                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2693                 stripe++;
2694                 index++;
2695         }
2696
2697         btrfs_set_stack_chunk_length(chunk, chunk_size);
2698         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2699         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2700         btrfs_set_stack_chunk_type(chunk, map->type);
2701         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2702         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2703         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2704         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2705         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2706
2707         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2708         key.type = BTRFS_CHUNK_ITEM_KEY;
2709         key.offset = chunk_offset;
2710
2711         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2712         BUG_ON(ret);
2713
2714         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2715                 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2716                                              item_size);
2717                 BUG_ON(ret);
2718         }
2719
2720         kfree(chunk);
2721         return 0;
2722 }
2723
2724 /*
2725  * Chunk allocation falls into two parts. The first part does works
2726  * that make the new allocated chunk useable, but not do any operation
2727  * that modifies the chunk tree. The second part does the works that
2728  * require modifying the chunk tree. This division is important for the
2729  * bootstrap process of adding storage to a seed btrfs.
2730  */
2731 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2732                       struct btrfs_root *extent_root, u64 type)
2733 {
2734         u64 chunk_offset;
2735         u64 chunk_size;
2736         u64 stripe_size;
2737         struct map_lookup *map;
2738         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2739         int ret;
2740
2741         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2742                               &chunk_offset);
2743         if (ret)
2744                 return ret;
2745
2746         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2747                                   &stripe_size, chunk_offset, type);
2748         if (ret)
2749                 return ret;
2750
2751         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2752                                    chunk_size, stripe_size);
2753         BUG_ON(ret);
2754         return 0;
2755 }
2756
2757 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2758                                          struct btrfs_root *root,
2759                                          struct btrfs_device *device)
2760 {
2761         u64 chunk_offset;
2762         u64 sys_chunk_offset;
2763         u64 chunk_size;
2764         u64 sys_chunk_size;
2765         u64 stripe_size;
2766         u64 sys_stripe_size;
2767         u64 alloc_profile;
2768         struct map_lookup *map;
2769         struct map_lookup *sys_map;
2770         struct btrfs_fs_info *fs_info = root->fs_info;
2771         struct btrfs_root *extent_root = fs_info->extent_root;
2772         int ret;
2773
2774         ret = find_next_chunk(fs_info->chunk_root,
2775                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2776         BUG_ON(ret);
2777
2778         alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2779                         (fs_info->metadata_alloc_profile &
2780                          fs_info->avail_metadata_alloc_bits);
2781         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2782
2783         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2784                                   &stripe_size, chunk_offset, alloc_profile);
2785         BUG_ON(ret);
2786
2787         sys_chunk_offset = chunk_offset + chunk_size;
2788
2789         alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2790                         (fs_info->system_alloc_profile &
2791                          fs_info->avail_system_alloc_bits);
2792         alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2793
2794         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2795                                   &sys_chunk_size, &sys_stripe_size,
2796                                   sys_chunk_offset, alloc_profile);
2797         BUG_ON(ret);
2798
2799         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2800         BUG_ON(ret);
2801
2802         /*
2803          * Modifying chunk tree needs allocating new blocks from both
2804          * system block group and metadata block group. So we only can
2805          * do operations require modifying the chunk tree after both
2806          * block groups were created.
2807          */
2808         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2809                                    chunk_size, stripe_size);
2810         BUG_ON(ret);
2811
2812         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2813                                    sys_chunk_offset, sys_chunk_size,
2814                                    sys_stripe_size);
2815         BUG_ON(ret);
2816         return 0;
2817 }
2818
2819 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2820 {
2821         struct extent_map *em;
2822         struct map_lookup *map;
2823         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2824         int readonly = 0;
2825         int i;
2826
2827         read_lock(&map_tree->map_tree.lock);
2828         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2829         read_unlock(&map_tree->map_tree.lock);
2830         if (!em)
2831                 return 1;
2832
2833         if (btrfs_test_opt(root, DEGRADED)) {
2834                 free_extent_map(em);
2835                 return 0;
2836         }
2837
2838         map = (struct map_lookup *)em->bdev;
2839         for (i = 0; i < map->num_stripes; i++) {
2840                 if (!map->stripes[i].dev->writeable) {
2841                         readonly = 1;
2842                         break;
2843                 }
2844         }
2845         free_extent_map(em);
2846         return readonly;
2847 }
2848
2849 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2850 {
2851         extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2852 }
2853
2854 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2855 {
2856         struct extent_map *em;
2857
2858         while (1) {
2859                 write_lock(&tree->map_tree.lock);
2860                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2861                 if (em)
2862                         remove_extent_mapping(&tree->map_tree, em);
2863                 write_unlock(&tree->map_tree.lock);
2864                 if (!em)
2865                         break;
2866                 kfree(em->bdev);
2867                 /* once for us */
2868                 free_extent_map(em);
2869                 /* once for the tree */
2870                 free_extent_map(em);
2871         }
2872 }
2873
2874 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2875 {
2876         struct extent_map *em;
2877         struct map_lookup *map;
2878         struct extent_map_tree *em_tree = &map_tree->map_tree;
2879         int ret;
2880
2881         read_lock(&em_tree->lock);
2882         em = lookup_extent_mapping(em_tree, logical, len);
2883         read_unlock(&em_tree->lock);
2884         BUG_ON(!em);
2885
2886         BUG_ON(em->start > logical || em->start + em->len < logical);
2887         map = (struct map_lookup *)em->bdev;
2888         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2889                 ret = map->num_stripes;
2890         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2891                 ret = map->sub_stripes;
2892         else
2893                 ret = 1;
2894         free_extent_map(em);
2895         return ret;
2896 }
2897
2898 static int find_live_mirror(struct map_lookup *map, int first, int num,
2899                             int optimal)
2900 {
2901         int i;
2902         if (map->stripes[optimal].dev->bdev)
2903                 return optimal;
2904         for (i = first; i < first + num; i++) {
2905                 if (map->stripes[i].dev->bdev)
2906                         return i;
2907         }
2908         /* we couldn't find one that doesn't fail.  Just return something
2909          * and the io error handling code will clean up eventually
2910          */
2911         return optimal;
2912 }
2913
2914 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2915                              u64 logical, u64 *length,
2916                              struct btrfs_multi_bio **multi_ret,
2917                              int mirror_num)
2918 {
2919         struct extent_map *em;
2920         struct map_lookup *map;
2921         struct extent_map_tree *em_tree = &map_tree->map_tree;
2922         u64 offset;
2923         u64 stripe_offset;
2924         u64 stripe_end_offset;
2925         u64 stripe_nr;
2926         u64 stripe_nr_orig;
2927         u64 stripe_nr_end;
2928         int stripes_allocated = 8;
2929         int stripes_required = 1;
2930         int stripe_index;
2931         int i;
2932         int num_stripes;
2933         int max_errors = 0;
2934         struct btrfs_multi_bio *multi = NULL;
2935
2936         if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2937                 stripes_allocated = 1;
2938 again:
2939         if (multi_ret) {
2940                 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2941                                 GFP_NOFS);
2942                 if (!multi)
2943                         return -ENOMEM;
2944
2945                 atomic_set(&multi->error, 0);
2946         }
2947
2948         read_lock(&em_tree->lock);
2949         em = lookup_extent_mapping(em_tree, logical, *length);
2950         read_unlock(&em_tree->lock);
2951
2952         if (!em) {
2953                 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2954                        (unsigned long long)logical,
2955                        (unsigned long long)*length);
2956                 BUG();
2957         }
2958
2959         BUG_ON(em->start > logical || em->start + em->len < logical);
2960         map = (struct map_lookup *)em->bdev;
2961         offset = logical - em->start;
2962
2963         if (mirror_num > map->num_stripes)
2964                 mirror_num = 0;
2965
2966         /* if our multi bio struct is too small, back off and try again */
2967         if (rw & REQ_WRITE) {
2968                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2969                                  BTRFS_BLOCK_GROUP_DUP)) {
2970                         stripes_required = map->num_stripes;
2971                         max_errors = 1;
2972                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2973                         stripes_required = map->sub_stripes;
2974                         max_errors = 1;
2975                 }
2976         }
2977         if (rw & REQ_DISCARD) {
2978                 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2979                                  BTRFS_BLOCK_GROUP_RAID1 |
2980                                  BTRFS_BLOCK_GROUP_DUP |
2981                                  BTRFS_BLOCK_GROUP_RAID10)) {
2982                         stripes_required = map->num_stripes;
2983                 }
2984         }
2985         if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2986             stripes_allocated < stripes_required) {
2987                 stripes_allocated = map->num_stripes;
2988                 free_extent_map(em);
2989                 kfree(multi);
2990                 goto again;
2991         }
2992         stripe_nr = offset;
2993         /*
2994          * stripe_nr counts the total number of stripes we have to stride
2995          * to get to this block
2996          */
2997         do_div(stripe_nr, map->stripe_len);
2998
2999         stripe_offset = stripe_nr * map->stripe_len;
3000         BUG_ON(offset < stripe_offset);
3001
3002         /* stripe_offset is the offset of this block in its stripe*/
3003         stripe_offset = offset - stripe_offset;
3004
3005         if (rw & REQ_DISCARD)
3006                 *length = min_t(u64, em->len - offset, *length);
3007         else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3008                               BTRFS_BLOCK_GROUP_RAID1 |
3009                               BTRFS_BLOCK_GROUP_RAID10 |
3010                               BTRFS_BLOCK_GROUP_DUP)) {
3011                 /* we limit the length of each bio to what fits in a stripe */
3012                 *length = min_t(u64, em->len - offset,
3013                                 map->stripe_len - stripe_offset);
3014         } else {
3015                 *length = em->len - offset;
3016         }
3017
3018         if (!multi_ret)
3019                 goto out;
3020
3021         num_stripes = 1;
3022         stripe_index = 0;
3023         stripe_nr_orig = stripe_nr;
3024         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3025                         (~(map->stripe_len - 1));
3026         do_div(stripe_nr_end, map->stripe_len);
3027         stripe_end_offset = stripe_nr_end * map->stripe_len -
3028                             (offset + *length);
3029         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3030                 if (rw & REQ_DISCARD)
3031                         num_stripes = min_t(u64, map->num_stripes,
3032                                             stripe_nr_end - stripe_nr_orig);
3033                 stripe_index = do_div(stripe_nr, map->num_stripes);
3034         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3035                 if (rw & (REQ_WRITE | REQ_DISCARD))
3036                         num_stripes = map->num_stripes;
3037                 else if (mirror_num)
3038                         stripe_index = mirror_num - 1;
3039                 else {
3040                         stripe_index = find_live_mirror(map, 0,
3041                                             map->num_stripes,
3042                                             current->pid % map->num_stripes);
3043                 }
3044
3045         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3046                 if (rw & (REQ_WRITE | REQ_DISCARD))
3047                         num_stripes = map->num_stripes;
3048                 else if (mirror_num)
3049                         stripe_index = mirror_num - 1;
3050
3051         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3052                 int factor = map->num_stripes / map->sub_stripes;
3053
3054                 stripe_index = do_div(stripe_nr, factor);
3055                 stripe_index *= map->sub_stripes;
3056
3057                 if (rw & REQ_WRITE)
3058                         num_stripes = map->sub_stripes;
3059                 else if (rw & REQ_DISCARD)
3060                         num_stripes = min_t(u64, map->sub_stripes *
3061                                             (stripe_nr_end - stripe_nr_orig),
3062                                             map->num_stripes);
3063                 else if (mirror_num)
3064                         stripe_index += mirror_num - 1;
3065                 else {
3066                         stripe_index = find_live_mirror(map, stripe_index,
3067                                               map->sub_stripes, stripe_index +
3068                                               current->pid % map->sub_stripes);
3069                 }
3070         } else {
3071                 /*
3072                  * after this do_div call, stripe_nr is the number of stripes
3073                  * on this device we have to walk to find the data, and
3074                  * stripe_index is the number of our device in the stripe array
3075                  */
3076                 stripe_index = do_div(stripe_nr, map->num_stripes);
3077         }
3078         BUG_ON(stripe_index >= map->num_stripes);
3079
3080         if (rw & REQ_DISCARD) {
3081                 for (i = 0; i < num_stripes; i++) {
3082                         multi->stripes[i].physical =
3083                                 map->stripes[stripe_index].physical +
3084                                 stripe_offset + stripe_nr * map->stripe_len;
3085                         multi->stripes[i].dev = map->stripes[stripe_index].dev;
3086
3087                         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3088                                 u64 stripes;
3089                                 u32 last_stripe = 0;
3090                                 int j;
3091
3092                                 div_u64_rem(stripe_nr_end - 1,
3093                                             map->num_stripes,
3094                                             &last_stripe);
3095
3096                                 for (j = 0; j < map->num_stripes; j++) {
3097                                         u32 test;
3098
3099                                         div_u64_rem(stripe_nr_end - 1 - j,
3100                                                     map->num_stripes, &test);
3101                                         if (test == stripe_index)
3102                                                 break;
3103                                 }
3104                                 stripes = stripe_nr_end - 1 - j;
3105                                 do_div(stripes, map->num_stripes);
3106                                 multi->stripes[i].length = map->stripe_len *
3107                                         (stripes - stripe_nr + 1);
3108
3109                                 if (i == 0) {
3110                                         multi->stripes[i].length -=
3111                                                 stripe_offset;
3112                                         stripe_offset = 0;
3113                                 }
3114                                 if (stripe_index == last_stripe)
3115                                         multi->stripes[i].length -=
3116                                                 stripe_end_offset;
3117                         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3118                                 u64 stripes;
3119                                 int j;
3120                                 int factor = map->num_stripes /
3121                                              map->sub_stripes;
3122                                 u32 last_stripe = 0;
3123
3124                                 div_u64_rem(stripe_nr_end - 1,
3125                                             factor, &last_stripe);
3126                                 last_stripe *= map->sub_stripes;
3127
3128                                 for (j = 0; j < factor; j++) {
3129                                         u32 test;
3130
3131                                         div_u64_rem(stripe_nr_end - 1 - j,
3132                                                     factor, &test);
3133
3134                                         if (test ==
3135                                             stripe_index / map->sub_stripes)
3136                                                 break;
3137                                 }
3138                                 stripes = stripe_nr_end - 1 - j;
3139                                 do_div(stripes, factor);
3140                                 multi->stripes[i].length = map->stripe_len *
3141                                         (stripes - stripe_nr + 1);
3142
3143                                 if (i < map->sub_stripes) {
3144                                         multi->stripes[i].length -=
3145                                                 stripe_offset;
3146                                         if (i == map->sub_stripes - 1)
3147                                                 stripe_offset = 0;
3148                                 }
3149                                 if (stripe_index >= last_stripe &&
3150                                     stripe_index <= (last_stripe +
3151                                                      map->sub_stripes - 1)) {
3152                                         multi->stripes[i].length -=
3153                                                 stripe_end_offset;
3154                                 }
3155                         } else
3156                                 multi->stripes[i].length = *length;
3157
3158                         stripe_index++;
3159                         if (stripe_index == map->num_stripes) {
3160                                 /* This could only happen for RAID0/10 */
3161                                 stripe_index = 0;
3162                                 stripe_nr++;
3163                         }
3164                 }
3165         } else {
3166                 for (i = 0; i < num_stripes; i++) {
3167                         multi->stripes[i].physical =
3168                                 map->stripes[stripe_index].physical +
3169                                 stripe_offset +
3170                                 stripe_nr * map->stripe_len;
3171                         multi->stripes[i].dev =
3172                                 map->stripes[stripe_index].dev;
3173                         stripe_index++;
3174                 }
3175         }
3176         if (multi_ret) {
3177                 *multi_ret = multi;
3178                 multi->num_stripes = num_stripes;
3179                 multi->max_errors = max_errors;
3180         }
3181 out:
3182         free_extent_map(em);
3183         return 0;
3184 }
3185
3186 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3187                       u64 logical, u64 *length,
3188                       struct btrfs_multi_bio **multi_ret, int mirror_num)
3189 {
3190         return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3191                                  mirror_num);
3192 }
3193
3194 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3195                      u64 chunk_start, u64 physical, u64 devid,
3196                      u64 **logical, int *naddrs, int *stripe_len)
3197 {
3198         struct extent_map_tree *em_tree = &map_tree->map_tree;
3199         struct extent_map *em;
3200         struct map_lookup *map;
3201         u64 *buf;
3202         u64 bytenr;
3203         u64 length;
3204         u64 stripe_nr;
3205         int i, j, nr = 0;
3206
3207         read_lock(&em_tree->lock);
3208         em = lookup_extent_mapping(em_tree, chunk_start, 1);
3209         read_unlock(&em_tree->lock);
3210
3211         BUG_ON(!em || em->start != chunk_start);
3212         map = (struct map_lookup *)em->bdev;
3213
3214         length = em->len;
3215         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3216                 do_div(length, map->num_stripes / map->sub_stripes);
3217         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3218                 do_div(length, map->num_stripes);
3219
3220         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3221         BUG_ON(!buf);
3222
3223         for (i = 0; i < map->num_stripes; i++) {
3224                 if (devid && map->stripes[i].dev->devid != devid)
3225                         continue;
3226                 if (map->stripes[i].physical > physical ||
3227                     map->stripes[i].physical + length <= physical)
3228                         continue;
3229
3230                 stripe_nr = physical - map->stripes[i].physical;
3231                 do_div(stripe_nr, map->stripe_len);
3232
3233                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3234                         stripe_nr = stripe_nr * map->num_stripes + i;
3235                         do_div(stripe_nr, map->sub_stripes);
3236                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3237                         stripe_nr = stripe_nr * map->num_stripes + i;
3238                 }
3239                 bytenr = chunk_start + stripe_nr * map->stripe_len;
3240                 WARN_ON(nr >= map->num_stripes);
3241                 for (j = 0; j < nr; j++) {
3242                         if (buf[j] == bytenr)
3243                                 break;
3244                 }
3245                 if (j == nr) {
3246                         WARN_ON(nr >= map->num_stripes);
3247                         buf[nr++] = bytenr;
3248                 }
3249         }
3250
3251         *logical = buf;
3252         *naddrs = nr;
3253         *stripe_len = map->stripe_len;
3254
3255         free_extent_map(em);
3256         return 0;
3257 }
3258
3259 static void end_bio_multi_stripe(struct bio *bio, int err)
3260 {
3261         struct btrfs_multi_bio *multi = bio->bi_private;
3262         int is_orig_bio = 0;
3263
3264         if (err)
3265                 atomic_inc(&multi->error);
3266
3267         if (bio == multi->orig_bio)
3268                 is_orig_bio = 1;
3269
3270         if (atomic_dec_and_test(&multi->stripes_pending)) {
3271                 if (!is_orig_bio) {
3272                         bio_put(bio);
3273                         bio = multi->orig_bio;
3274                 }
3275                 bio->bi_private = multi->private;
3276                 bio->bi_end_io = multi->end_io;
3277                 /* only send an error to the higher layers if it is
3278                  * beyond the tolerance of the multi-bio
3279                  */
3280                 if (atomic_read(&multi->error) > multi->max_errors) {
3281                         err = -EIO;
3282                 } else if (err) {
3283                         /*
3284                          * this bio is actually up to date, we didn't
3285                          * go over the max number of errors
3286                          */
3287                         set_bit(BIO_UPTODATE, &bio->bi_flags);
3288                         err = 0;
3289                 }
3290                 kfree(multi);
3291
3292                 bio_endio(bio, err);
3293         } else if (!is_orig_bio) {
3294                 bio_put(bio);
3295         }
3296 }
3297
3298 struct async_sched {
3299         struct bio *bio;
3300         int rw;
3301         struct btrfs_fs_info *info;
3302         struct btrfs_work work;
3303 };
3304
3305 /*
3306  * see run_scheduled_bios for a description of why bios are collected for
3307  * async submit.
3308  *
3309  * This will add one bio to the pending list for a device and make sure
3310  * the work struct is scheduled.
3311  */
3312 static noinline int schedule_bio(struct btrfs_root *root,
3313                                  struct btrfs_device *device,
3314                                  int rw, struct bio *bio)
3315 {
3316         int should_queue = 1;
3317         struct btrfs_pending_bios *pending_bios;
3318
3319         /* don't bother with additional async steps for reads, right now */
3320         if (!(rw & REQ_WRITE)) {
3321                 bio_get(bio);
3322                 submit_bio(rw, bio);
3323                 bio_put(bio);
3324                 return 0;
3325         }
3326
3327         /*
3328          * nr_async_bios allows us to reliably return congestion to the
3329          * higher layers.  Otherwise, the async bio makes it appear we have
3330          * made progress against dirty pages when we've really just put it
3331          * on a queue for later
3332          */
3333         atomic_inc(&root->fs_info->nr_async_bios);
3334         WARN_ON(bio->bi_next);
3335         bio->bi_next = NULL;
3336         bio->bi_rw |= rw;
3337
3338         spin_lock(&device->io_lock);
3339         if (bio->bi_rw & REQ_SYNC)
3340                 pending_bios = &device->pending_sync_bios;
3341         else
3342                 pending_bios = &device->pending_bios;
3343
3344         if (pending_bios->tail)
3345                 pending_bios->tail->bi_next = bio;
3346
3347         pending_bios->tail = bio;
3348         if (!pending_bios->head)
3349                 pending_bios->head = bio;
3350         if (device->running_pending)
3351                 should_queue = 0;
3352
3353         spin_unlock(&device->io_lock);
3354
3355         if (should_queue)
3356                 btrfs_queue_worker(&root->fs_info->submit_workers,
3357                                    &device->work);
3358         return 0;
3359 }
3360
3361 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3362                   int mirror_num, int async_submit)
3363 {
3364         struct btrfs_mapping_tree *map_tree;
3365         struct btrfs_device *dev;
3366         struct bio *first_bio = bio;
3367         u64 logical = (u64)bio->bi_sector << 9;
3368         u64 length = 0;
3369         u64 map_length;
3370         struct btrfs_multi_bio *multi = NULL;
3371         int ret;
3372         int dev_nr = 0;
3373         int total_devs = 1;
3374
3375         length = bio->bi_size;
3376         map_tree = &root->fs_info->mapping_tree;
3377         map_length = length;
3378
3379         ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3380                               mirror_num);
3381         BUG_ON(ret);
3382
3383         total_devs = multi->num_stripes;
3384         if (map_length < length) {
3385                 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3386                        "len %llu\n", (unsigned long long)logical,
3387                        (unsigned long long)length,
3388                        (unsigned long long)map_length);
3389                 BUG();
3390         }
3391         multi->end_io = first_bio->bi_end_io;
3392         multi->private = first_bio->bi_private;
3393         multi->orig_bio = first_bio;
3394         atomic_set(&multi->stripes_pending, multi->num_stripes);
3395
3396         while (dev_nr < total_devs) {
3397                 if (total_devs > 1) {
3398                         if (dev_nr < total_devs - 1) {
3399                                 bio = bio_clone(first_bio, GFP_NOFS);
3400                                 BUG_ON(!bio);
3401                         } else {
3402                                 bio = first_bio;
3403                         }
3404                         bio->bi_private = multi;
3405                         bio->bi_end_io = end_bio_multi_stripe;
3406                 }
3407                 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3408                 dev = multi->stripes[dev_nr].dev;
3409                 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3410                         bio->bi_bdev = dev->bdev;
3411                         if (async_submit)
3412                                 schedule_bio(root, dev, rw, bio);
3413                         else
3414                                 submit_bio(rw, bio);
3415                 } else {
3416                         bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3417                         bio->bi_sector = logical >> 9;
3418                         bio_endio(bio, -EIO);
3419                 }
3420                 dev_nr++;
3421         }
3422         if (total_devs == 1)
3423                 kfree(multi);
3424         return 0;
3425 }
3426
3427 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3428                                        u8 *uuid, u8 *fsid)
3429 {
3430         struct btrfs_device *device;
3431         struct btrfs_fs_devices *cur_devices;
3432
3433         cur_devices = root->fs_info->fs_devices;
3434         while (cur_devices) {
3435                 if (!fsid ||
3436                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3437                         device = __find_device(&cur_devices->devices,
3438                                                devid, uuid);
3439                         if (device)
3440                                 return device;
3441                 }
3442                 cur_devices = cur_devices->seed;
3443         }
3444         return NULL;
3445 }
3446
3447 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3448                                             u64 devid, u8 *dev_uuid)
3449 {
3450         struct btrfs_device *device;
3451         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3452
3453         device = kzalloc(sizeof(*device), GFP_NOFS);
3454         if (!device)
3455                 return NULL;
3456         list_add(&device->dev_list,
3457                  &fs_devices->devices);
3458         device->dev_root = root->fs_info->dev_root;
3459         device->devid = devid;
3460         device->work.func = pending_bios_fn;
3461         device->fs_devices = fs_devices;
3462         device->missing = 1;
3463         fs_devices->num_devices++;
3464         fs_devices->missing_devices++;
3465         spin_lock_init(&device->io_lock);
3466         INIT_LIST_HEAD(&device->dev_alloc_list);
3467         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3468         return device;
3469 }
3470
3471 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3472                           struct extent_buffer *leaf,
3473                           struct btrfs_chunk *chunk)
3474 {
3475         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3476         struct map_lookup *map;
3477         struct extent_map *em;
3478         u64 logical;
3479         u64 length;
3480         u64 devid;
3481         u8 uuid[BTRFS_UUID_SIZE];
3482         int num_stripes;
3483         int ret;
3484         int i;
3485
3486         logical = key->offset;
3487         length = btrfs_chunk_length(leaf, chunk);
3488
3489         read_lock(&map_tree->map_tree.lock);
3490         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3491         read_unlock(&map_tree->map_tree.lock);
3492
3493         /* already mapped? */
3494         if (em && em->start <= logical && em->start + em->len > logical) {
3495                 free_extent_map(em);
3496                 return 0;
3497         } else if (em) {
3498                 free_extent_map(em);
3499         }
3500
3501         em = alloc_extent_map(GFP_NOFS);
3502         if (!em)
3503                 return -ENOMEM;
3504         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3505         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3506         if (!map) {
3507                 free_extent_map(em);
3508                 return -ENOMEM;
3509         }
3510
3511         em->bdev = (struct block_device *)map;
3512         em->start = logical;
3513         em->len = length;
3514         em->block_start = 0;
3515         em->block_len = em->len;
3516
3517         map->num_stripes = num_stripes;
3518         map->io_width = btrfs_chunk_io_width(leaf, chunk);
3519         map->io_align = btrfs_chunk_io_align(leaf, chunk);
3520         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3521         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3522         map->type = btrfs_chunk_type(leaf, chunk);
3523         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3524         for (i = 0; i < num_stripes; i++) {
3525                 map->stripes[i].physical =
3526                         btrfs_stripe_offset_nr(leaf, chunk, i);
3527                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3528                 read_extent_buffer(leaf, uuid, (unsigned long)
3529                                    btrfs_stripe_dev_uuid_nr(chunk, i),
3530                                    BTRFS_UUID_SIZE);
3531                 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3532                                                         NULL);
3533                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3534                         kfree(map);
3535                         free_extent_map(em);
3536                         return -EIO;
3537                 }
3538                 if (!map->stripes[i].dev) {
3539                         map->stripes[i].dev =
3540                                 add_missing_dev(root, devid, uuid);
3541                         if (!map->stripes[i].dev) {
3542                                 kfree(map);
3543                                 free_extent_map(em);
3544                                 return -EIO;
3545                         }
3546                 }
3547                 map->stripes[i].dev->in_fs_metadata = 1;
3548         }
3549
3550         write_lock(&map_tree->map_tree.lock);
3551         ret = add_extent_mapping(&map_tree->map_tree, em);
3552         write_unlock(&map_tree->map_tree.lock);
3553         BUG_ON(ret);
3554         free_extent_map(em);
3555
3556         return 0;
3557 }
3558
3559 static int fill_device_from_item(struct extent_buffer *leaf,
3560                                  struct btrfs_dev_item *dev_item,
3561                                  struct btrfs_device *device)
3562 {
3563         unsigned long ptr;
3564
3565         device->devid = btrfs_device_id(leaf, dev_item);
3566         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3567         device->total_bytes = device->disk_total_bytes;
3568         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3569         device->type = btrfs_device_type(leaf, dev_item);
3570         device->io_align = btrfs_device_io_align(leaf, dev_item);
3571         device->io_width = btrfs_device_io_width(leaf, dev_item);
3572         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3573
3574         ptr = (unsigned long)btrfs_device_uuid(dev_item);
3575         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3576
3577         return 0;
3578 }
3579
3580 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3581 {
3582         struct btrfs_fs_devices *fs_devices;
3583         int ret;
3584
3585         mutex_lock(&uuid_mutex);
3586
3587         fs_devices = root->fs_info->fs_devices->seed;
3588         while (fs_devices) {
3589                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3590                         ret = 0;
3591                         goto out;
3592                 }
3593                 fs_devices = fs_devices->seed;
3594         }
3595
3596         fs_devices = find_fsid(fsid);
3597         if (!fs_devices) {
3598                 ret = -ENOENT;
3599                 goto out;
3600         }
3601
3602         fs_devices = clone_fs_devices(fs_devices);
3603         if (IS_ERR(fs_devices)) {
3604                 ret = PTR_ERR(fs_devices);
3605                 goto out;
3606         }
3607
3608         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3609                                    root->fs_info->bdev_holder);
3610         if (ret)
3611                 goto out;
3612
3613         if (!fs_devices->seeding) {
3614                 __btrfs_close_devices(fs_devices);
3615                 free_fs_devices(fs_devices);
3616                 ret = -EINVAL;
3617                 goto out;
3618         }
3619
3620         fs_devices->seed = root->fs_info->fs_devices->seed;
3621         root->fs_info->fs_devices->seed = fs_devices;
3622 out:
3623         mutex_unlock(&uuid_mutex);
3624         return ret;
3625 }
3626
3627 static int read_one_dev(struct btrfs_root *root,
3628                         struct extent_buffer *leaf,
3629                         struct btrfs_dev_item *dev_item)
3630 {
3631         struct btrfs_device *device;
3632         u64 devid;
3633         int ret;
3634         u8 fs_uuid[BTRFS_UUID_SIZE];
3635         u8 dev_uuid[BTRFS_UUID_SIZE];
3636
3637         devid = btrfs_device_id(leaf, dev_item);
3638         read_extent_buffer(leaf, dev_uuid,
3639                            (unsigned long)btrfs_device_uuid(dev_item),
3640                            BTRFS_UUID_SIZE);
3641         read_extent_buffer(leaf, fs_uuid,
3642                            (unsigned long)btrfs_device_fsid(dev_item),
3643                            BTRFS_UUID_SIZE);
3644
3645         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3646                 ret = open_seed_devices(root, fs_uuid);
3647                 if (ret && !btrfs_test_opt(root, DEGRADED))
3648                         return ret;
3649         }
3650
3651         device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3652         if (!device || !device->bdev) {
3653                 if (!btrfs_test_opt(root, DEGRADED))
3654                         return -EIO;
3655
3656                 if (!device) {
3657                         printk(KERN_WARNING "warning devid %llu missing\n",
3658                                (unsigned long long)devid);
3659                         device = add_missing_dev(root, devid, dev_uuid);
3660                         if (!device)
3661                                 return -ENOMEM;
3662                 } else if (!device->missing) {
3663                         /*
3664                          * this happens when a device that was properly setup
3665                          * in the device info lists suddenly goes bad.
3666                          * device->bdev is NULL, and so we have to set
3667                          * device->missing to one here
3668                          */
3669                         root->fs_info->fs_devices->missing_devices++;
3670                         device->missing = 1;
3671                 }
3672         }
3673
3674         if (device->fs_devices != root->fs_info->fs_devices) {
3675                 BUG_ON(device->writeable);
3676                 if (device->generation !=
3677                     btrfs_device_generation(leaf, dev_item))
3678                         return -EINVAL;
3679         }
3680
3681         fill_device_from_item(leaf, dev_item, device);
3682         device->dev_root = root->fs_info->dev_root;
3683         device->in_fs_metadata = 1;
3684         if (device->writeable)
3685                 device->fs_devices->total_rw_bytes += device->total_bytes;
3686         ret = 0;
3687         return ret;
3688 }
3689
3690 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3691 {
3692         struct btrfs_dev_item *dev_item;
3693
3694         dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3695                                                      dev_item);
3696         return read_one_dev(root, buf, dev_item);
3697 }
3698
3699 int btrfs_read_sys_array(struct btrfs_root *root)
3700 {
3701         struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3702         struct extent_buffer *sb;
3703         struct btrfs_disk_key *disk_key;
3704         struct btrfs_chunk *chunk;
3705         u8 *ptr;
3706         unsigned long sb_ptr;
3707         int ret = 0;
3708         u32 num_stripes;
3709         u32 array_size;
3710         u32 len = 0;
3711         u32 cur;
3712         struct btrfs_key key;
3713
3714         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3715                                           BTRFS_SUPER_INFO_SIZE);
3716         if (!sb)
3717                 return -ENOMEM;
3718         btrfs_set_buffer_uptodate(sb);
3719         btrfs_set_buffer_lockdep_class(sb, 0);
3720
3721         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3722         array_size = btrfs_super_sys_array_size(super_copy);
3723
3724         ptr = super_copy->sys_chunk_array;
3725         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3726         cur = 0;
3727
3728         while (cur < array_size) {
3729                 disk_key = (struct btrfs_disk_key *)ptr;
3730                 btrfs_disk_key_to_cpu(&key, disk_key);
3731
3732                 len = sizeof(*disk_key); ptr += len;
3733                 sb_ptr += len;
3734                 cur += len;
3735
3736                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3737                         chunk = (struct btrfs_chunk *)sb_ptr;
3738                         ret = read_one_chunk(root, &key, sb, chunk);
3739                         if (ret)
3740                                 break;
3741                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3742                         len = btrfs_chunk_item_size(num_stripes);
3743                 } else {
3744                         ret = -EIO;
3745                         break;
3746                 }
3747                 ptr += len;
3748                 sb_ptr += len;
3749                 cur += len;
3750         }
3751         free_extent_buffer(sb);
3752         return ret;
3753 }
3754
3755 int btrfs_read_chunk_tree(struct btrfs_root *root)
3756 {
3757         struct btrfs_path *path;
3758         struct extent_buffer *leaf;
3759         struct btrfs_key key;
3760         struct btrfs_key found_key;
3761         int ret;
3762         int slot;
3763
3764         root = root->fs_info->chunk_root;
3765
3766         path = btrfs_alloc_path();
3767         if (!path)
3768                 return -ENOMEM;
3769
3770         /* first we search for all of the device items, and then we
3771          * read in all of the chunk items.  This way we can create chunk
3772          * mappings that reference all of the devices that are afound
3773          */
3774         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3775         key.offset = 0;
3776         key.type = 0;
3777 again:
3778         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3779         if (ret < 0)
3780                 goto error;
3781         while (1) {
3782                 leaf = path->nodes[0];
3783                 slot = path->slots[0];
3784                 if (slot >= btrfs_header_nritems(leaf)) {
3785                         ret = btrfs_next_leaf(root, path);
3786                         if (ret == 0)
3787                                 continue;
3788                         if (ret < 0)
3789                                 goto error;
3790                         break;
3791                 }
3792                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3793                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3794                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3795                                 break;
3796                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3797                                 struct btrfs_dev_item *dev_item;
3798                                 dev_item = btrfs_item_ptr(leaf, slot,
3799                                                   struct btrfs_dev_item);
3800                                 ret = read_one_dev(root, leaf, dev_item);
3801                                 if (ret)
3802                                         goto error;
3803                         }
3804                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3805                         struct btrfs_chunk *chunk;
3806                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3807                         ret = read_one_chunk(root, &found_key, leaf, chunk);
3808                         if (ret)
3809                                 goto error;
3810                 }
3811                 path->slots[0]++;
3812         }
3813         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3814                 key.objectid = 0;
3815                 btrfs_release_path(root, path);
3816                 goto again;
3817         }
3818         ret = 0;
3819 error:
3820         btrfs_free_path(path);
3821         return ret;
3822 }