Root/fs/btrfs/volumes.c

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

Archive Download this file



interactive