projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Do not use bio->bi_bdev after submission
[deliverable/linux.git] / fs / btrfs / volumes.c
CommitLineData
0b86a832
CM		 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>
5a0e3ad6 20#include <linux/slab.h>
8a4b83cc 21#include <linux/buffer_head.h>
f2d8d74d 22#include <linux/blkdev.h>
788f20eb 23#include <linux/random.h>
b765ead5 24#include <linux/iocontext.h>
6f88a440 25#include <linux/capability.h>
593060d7 26#include <asm/div64.h>
4b4e25f2 27#include "compat.h"
0b86a832
CM		 28#include "ctree.h"
29#include "extent_map.h"
30#include "disk-io.h"
31#include "transaction.h"
32#include "print-tree.h"
33#include "volumes.h"
8b712842 34#include "async-thread.h"
0b86a832 35
2b82032c
YZ		 36static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
8a4b83cc
CM		 41static DEFINE_MUTEX(uuid_mutex);
42static LIST_HEAD(fs_uuids);
43
7d9eb12c
CM		 44static void lock_chunks(struct btrfs_root *root)
45{
7d9eb12c
CM		 46 mutex_lock(&root->fs_info->chunk_mutex);
47}
48
49static void unlock_chunks(struct btrfs_root *root)
50{
7d9eb12c
CM		 51 mutex_unlock(&root->fs_info->chunk_mutex);
52}
53
e4404d6e
YZ		 54static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
55{
56 struct btrfs_device *device;
57 WARN_ON(fs_devices->opened);
58 while (!list_empty(&fs_devices->devices)) {
59 device = list_entry(fs_devices->devices.next,
60 struct btrfs_device, dev_list);
61 list_del(&device->dev_list);
62 kfree(device->name);
63 kfree(device);
64 }
65 kfree(fs_devices);
66}
67
8a4b83cc
CM		 68int btrfs_cleanup_fs_uuids(void)
69{
70 struct btrfs_fs_devices *fs_devices;
8a4b83cc 71
2b82032c
YZ		 72 while (!list_empty(&fs_uuids)) {
73 fs_devices = list_entry(fs_uuids.next,
74 struct btrfs_fs_devices, list);
75 list_del(&fs_devices->list);
e4404d6e 76 free_fs_devices(fs_devices);
8a4b83cc
CM		 77 }
78 return 0;
79}
80
a1b32a59
CM		 81static noinline struct btrfs_device *__find_device(struct list_head *head,
82 u64 devid, u8 *uuid)
8a4b83cc
CM		 83{
84 struct btrfs_device *dev;
8a4b83cc 85
c6e30871 86 list_for_each_entry(dev, head, dev_list) {
a443755f 87 if (dev->devid == devid &&
8f18cf13 88 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 89 return dev;
a443755f 90 }
8a4b83cc
CM		 91 }
92 return NULL;
93}
94
a1b32a59 95static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 96{
8a4b83cc
CM		 97 struct btrfs_fs_devices *fs_devices;
98
c6e30871 99 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 100 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
101 return fs_devices;
102 }
103 return NULL;
104}
105
ffbd517d
CM		 106static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 struct bio *head, struct bio *tail)
108{
109
110 struct bio *old_head;
111
112 old_head = pending_bios->head;
113 pending_bios->head = head;
114 if (pending_bios->tail)
115 tail->bi_next = old_head;
116 else
117 pending_bios->tail = tail;
118}
119
8b712842
CM		 120/*
121 * we try to collect pending bios for a device so we don't get a large
122 * number of procs sending bios down to the same device. This greatly
123 * improves the schedulers ability to collect and merge the bios.
124 *
125 * But, it also turns into a long list of bios to process and that is sure
126 * to eventually make the worker thread block. The solution here is to
127 * make some progress and then put this work struct back at the end of
128 * the list if the block device is congested. This way, multiple devices
129 * can make progress from a single worker thread.
130 */
d397712b 131static noinline int run_scheduled_bios(struct btrfs_device *device)
8b712842
CM		 132{
133 struct bio *pending;
134 struct backing_dev_info *bdi;
b64a2851 135 struct btrfs_fs_info *fs_info;
ffbd517d 136 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 137 struct bio *tail;
138 struct bio *cur;
139 int again = 0;
ffbd517d 140 unsigned long num_run;
d644d8a1 141 unsigned long batch_run = 0;
b64a2851 142 unsigned long limit;
b765ead5 143 unsigned long last_waited = 0;
d84275c9 144 int force_reg = 0;
0e588859 145 int sync_pending = 0;
211588ad
CM		 146 struct blk_plug plug;
147
148 /*
149 * this function runs all the bios we've collected for
150 * a particular device. We don't want to wander off to
151 * another device without first sending all of these down.
152 * So, setup a plug here and finish it off before we return
153 */
154 blk_start_plug(&plug);
8b712842 155
bedf762b 156 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 157 fs_info = device->dev_root->fs_info;
158 limit = btrfs_async_submit_limit(fs_info);
159 limit = limit * 2 / 3;
160
8b712842
CM		 161loop:
162 spin_lock(&device->io_lock);
163
a6837051 164loop_lock:
d84275c9 165 num_run = 0;
ffbd517d 166
8b712842
CM		 167 /* take all the bios off the list at once and process them
168 * later on (without the lock held). But, remember the
169 * tail and other pointers so the bios can be properly reinserted
170 * into the list if we hit congestion
171 */
d84275c9 172 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 173 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 174 force_reg = 1;
175 } else {
ffbd517d 176 pending_bios = &device->pending_bios;
d84275c9
CM		 177 force_reg = 0;
178 }
ffbd517d
CM		 179
180 pending = pending_bios->head;
181 tail = pending_bios->tail;
8b712842 182 WARN_ON(pending && !tail);
8b712842
CM		 183
184 /*
185 * if pending was null this time around, no bios need processing
186 * at all and we can stop. Otherwise it'll loop back up again
187 * and do an additional check so no bios are missed.
188 *
189 * device->running_pending is used to synchronize with the
190 * schedule_bio code.
191 */
ffbd517d
CM		 192 if (device->pending_sync_bios.head == NULL &&
193 device->pending_bios.head == NULL) {
8b712842
CM		 194 again = 0;
195 device->running_pending = 0;
ffbd517d
CM		 196 } else {
197 again = 1;
198 device->running_pending = 1;
8b712842 199 }
ffbd517d
CM		 200
201 pending_bios->head = NULL;
202 pending_bios->tail = NULL;
203
8b712842
CM		 204 spin_unlock(&device->io_lock);
205
d397712b 206 while (pending) {
ffbd517d
CM		 207
208 rmb();
d84275c9
CM		 209 /* we want to work on both lists, but do more bios on the
210 * sync list than the regular list
211 */
212 if ((num_run > 32 &&
213 pending_bios != &device->pending_sync_bios &&
214 device->pending_sync_bios.head) ||
215 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
216 device->pending_bios.head)) {
ffbd517d
CM		 217 spin_lock(&device->io_lock);
218 requeue_list(pending_bios, pending, tail);
219 goto loop_lock;
220 }
221
8b712842
CM		 222 cur = pending;
223 pending = pending->bi_next;
224 cur->bi_next = NULL;
b64a2851
CM		 225 atomic_dec(&fs_info->nr_async_bios);
226
227 if (atomic_read(&fs_info->nr_async_bios) < limit &&
228 waitqueue_active(&fs_info->async_submit_wait))
229 wake_up(&fs_info->async_submit_wait);
492bb6de
CM		 230
231 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
d644d8a1 232
2ab1ba68
CM		 233 /*
234 * if we're doing the sync list, record that our
235 * plug has some sync requests on it
236 *
237 * If we're doing the regular list and there are
238 * sync requests sitting around, unplug before
239 * we add more
240 */
241 if (pending_bios == &device->pending_sync_bios) {
242 sync_pending = 1;
243 } else if (sync_pending) {
244 blk_finish_plug(&plug);
245 blk_start_plug(&plug);
246 sync_pending = 0;
247 }
248
5ff7ba3a
CM		 249 submit_bio(cur->bi_rw, cur);
250 num_run++;
251 batch_run++;
7eaceacc 252 if (need_resched())
ffbd517d 253 cond_resched();
8b712842
CM		 254
255 /*
256 * we made progress, there is more work to do and the bdi
257 * is now congested. Back off and let other work structs
258 * run instead
259 */
57fd5a5f 260 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 261 fs_info->fs_devices->open_devices > 1) {
b765ead5 262 struct io_context *ioc;
8b712842 263
b765ead5
CM		 264 ioc = current->io_context;
265
266 /*
267 * the main goal here is that we don't want to
268 * block if we're going to be able to submit
269 * more requests without blocking.
270 *
271 * This code does two great things, it pokes into
272 * the elevator code from a filesystem _and_
273 * it makes assumptions about how batching works.
274 */
275 if (ioc && ioc->nr_batch_requests > 0 &&
276 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
277 (last_waited == 0 ||
278 ioc->last_waited == last_waited)) {
279 /*
280 * we want to go through our batch of
281 * requests and stop. So, we copy out
282 * the ioc->last_waited time and test
283 * against it before looping
284 */
285 last_waited = ioc->last_waited;
7eaceacc 286 if (need_resched())
ffbd517d 287 cond_resched();
b765ead5
CM		 288 continue;
289 }
8b712842 290 spin_lock(&device->io_lock);
ffbd517d 291 requeue_list(pending_bios, pending, tail);
a6837051 292 device->running_pending = 1;
8b712842
CM		 293
294 spin_unlock(&device->io_lock);
295 btrfs_requeue_work(&device->work);
296 goto done;
297 }
298 }
ffbd517d 299
51684082
CM		 300 cond_resched();
301 if (again)
302 goto loop;
303
304 spin_lock(&device->io_lock);
305 if (device->pending_bios.head || device->pending_sync_bios.head)
306 goto loop_lock;
307 spin_unlock(&device->io_lock);
308
8b712842 309done:
211588ad 310 blk_finish_plug(&plug);
8b712842
CM		 311 return 0;
312}
313
b2950863 314static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 315{
316 struct btrfs_device *device;
317
318 device = container_of(work, struct btrfs_device, work);
319 run_scheduled_bios(device);
320}
321
a1b32a59 322static noinline int device_list_add(const char *path,
8a4b83cc
CM		 323 struct btrfs_super_block *disk_super,
324 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
325{
326 struct btrfs_device *device;
327 struct btrfs_fs_devices *fs_devices;
328 u64 found_transid = btrfs_super_generation(disk_super);
3a0524dc 329 char *name;
8a4b83cc
CM		 330
331 fs_devices = find_fsid(disk_super->fsid);
332 if (!fs_devices) {
515dc322 333 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 334 if (!fs_devices)
335 return -ENOMEM;
336 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 337 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 338 list_add(&fs_devices->list, &fs_uuids);
339 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
340 fs_devices->latest_devid = devid;
341 fs_devices->latest_trans = found_transid;
e5e9a520 342 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 343 device = NULL;
344 } else {
a443755f
CM		 345 device = __find_device(&fs_devices->devices, devid,
346 disk_super->dev_item.uuid);
8a4b83cc
CM		 347 }
348 if (!device) {
2b82032c
YZ		 349 if (fs_devices->opened)
350 return -EBUSY;
351
8a4b83cc
CM		 352 device = kzalloc(sizeof(*device), GFP_NOFS);
353 if (!device) {
354 /* we can safely leave the fs_devices entry around */
355 return -ENOMEM;
356 }
357 device->devid = devid;
8b712842 358 device->work.func = pending_bios_fn;
a443755f
CM		 359 memcpy(device->uuid, disk_super->dev_item.uuid,
360 BTRFS_UUID_SIZE);
b248a415 361 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 362 device->name = kstrdup(path, GFP_NOFS);
363 if (!device->name) {
364 kfree(device);
365 return -ENOMEM;
366 }
2b82032c 367 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520
CM		 368
369 mutex_lock(&fs_devices->device_list_mutex);
1f78160c 370 list_add_rcu(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 371 mutex_unlock(&fs_devices->device_list_mutex);
372
2b82032c 373 device->fs_devices = fs_devices;
8a4b83cc 374 fs_devices->num_devices++;
cd02dca5 375 } else if (!device->name || strcmp(device->name, path)) {
3a0524dc
TH		 376 name = kstrdup(path, GFP_NOFS);
377 if (!name)
378 return -ENOMEM;
379 kfree(device->name);
380 device->name = name;
cd02dca5
CM		 381 if (device->missing) {
382 fs_devices->missing_devices--;
383 device->missing = 0;
384 }
8a4b83cc
CM		 385 }
386
387 if (found_transid > fs_devices->latest_trans) {
388 fs_devices->latest_devid = devid;
389 fs_devices->latest_trans = found_transid;
390 }
8a4b83cc
CM		 391 *fs_devices_ret = fs_devices;
392 return 0;
393}
394
e4404d6e
YZ		 395static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
396{
397 struct btrfs_fs_devices *fs_devices;
398 struct btrfs_device *device;
399 struct btrfs_device *orig_dev;
400
401 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
402 if (!fs_devices)
403 return ERR_PTR(-ENOMEM);
404
405 INIT_LIST_HEAD(&fs_devices->devices);
406 INIT_LIST_HEAD(&fs_devices->alloc_list);
407 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 408 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 409 fs_devices->latest_devid = orig->latest_devid;
410 fs_devices->latest_trans = orig->latest_trans;
411 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
412
46224705 413 /* We have held the volume lock, it is safe to get the devices. */
e4404d6e
YZ		 414 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
415 device = kzalloc(sizeof(*device), GFP_NOFS);
416 if (!device)
417 goto error;
418
419 device->name = kstrdup(orig_dev->name, GFP_NOFS);
fd2696f3
JL		 420 if (!device->name) {
421 kfree(device);
e4404d6e 422 goto error;
fd2696f3 423 }
e4404d6e
YZ		 424
425 device->devid = orig_dev->devid;
426 device->work.func = pending_bios_fn;
427 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
e4404d6e
YZ		 428 spin_lock_init(&device->io_lock);
429 INIT_LIST_HEAD(&device->dev_list);
430 INIT_LIST_HEAD(&device->dev_alloc_list);
431
432 list_add(&device->dev_list, &fs_devices->devices);
433 device->fs_devices = fs_devices;
434 fs_devices->num_devices++;
435 }
436 return fs_devices;
437error:
438 free_fs_devices(fs_devices);
439 return ERR_PTR(-ENOMEM);
440}
441
dfe25020
CM		 442int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
443{
c6e30871 444 struct btrfs_device *device, *next;
dfe25020
CM		 445
446 mutex_lock(&uuid_mutex);
447again:
46224705 448 /* This is the initialized path, it is safe to release the devices. */
c6e30871 449 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 450 if (device->in_fs_metadata)
451 continue;
452
453 if (device->bdev) {
d4d77629 454 blkdev_put(device->bdev, device->mode);
2b82032c
YZ		 455 device->bdev = NULL;
456 fs_devices->open_devices--;
457 }
458 if (device->writeable) {
459 list_del_init(&device->dev_alloc_list);
460 device->writeable = 0;
461 fs_devices->rw_devices--;
462 }
e4404d6e
YZ		 463 list_del_init(&device->dev_list);
464 fs_devices->num_devices--;
465 kfree(device->name);
466 kfree(device);
dfe25020 467 }
2b82032c
YZ		 468
469 if (fs_devices->seed) {
470 fs_devices = fs_devices->seed;
2b82032c
YZ		 471 goto again;
472 }
473
dfe25020
CM		 474 mutex_unlock(&uuid_mutex);
475 return 0;
476}
a0af469b 477
1f78160c
XG		 478static void __free_device(struct work_struct *work)
479{
480 struct btrfs_device *device;
481
482 device = container_of(work, struct btrfs_device, rcu_work);
483
484 if (device->bdev)
485 blkdev_put(device->bdev, device->mode);
486
487 kfree(device->name);
488 kfree(device);
489}
490
491static void free_device(struct rcu_head *head)
492{
493 struct btrfs_device *device;
494
495 device = container_of(head, struct btrfs_device, rcu);
496
497 INIT_WORK(&device->rcu_work, __free_device);
498 schedule_work(&device->rcu_work);
499}
500
2b82032c 501static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 502{
8a4b83cc 503 struct btrfs_device *device;
e4404d6e 504
2b82032c
YZ		 505 if (--fs_devices->opened > 0)
506 return 0;
8a4b83cc 507
c9513edb 508 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 509 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1f78160c
XG		 510 struct btrfs_device *new_device;
511
512 if (device->bdev)
a0af469b 513 fs_devices->open_devices--;
1f78160c 514
2b82032c
YZ		 515 if (device->writeable) {
516 list_del_init(&device->dev_alloc_list);
517 fs_devices->rw_devices--;
518 }
519
d5e2003c
JB		 520 if (device->can_discard)
521 fs_devices->num_can_discard--;
522
1f78160c
XG		 523 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
524 BUG_ON(!new_device);
525 memcpy(new_device, device, sizeof(*new_device));
526 new_device->name = kstrdup(device->name, GFP_NOFS);
5f3f302a 527 BUG_ON(device->name && !new_device->name);
1f78160c
XG		 528 new_device->bdev = NULL;
529 new_device->writeable = 0;
530 new_device->in_fs_metadata = 0;
d5e2003c 531 new_device->can_discard = 0;
1f78160c
XG		 532 list_replace_rcu(&device->dev_list, &new_device->dev_list);
533
534 call_rcu(&device->rcu, free_device);
8a4b83cc 535 }
c9513edb
XG		 536 mutex_unlock(&fs_devices->device_list_mutex);
537
e4404d6e
YZ		 538 WARN_ON(fs_devices->open_devices);
539 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 540 fs_devices->opened = 0;
541 fs_devices->seeding = 0;
2b82032c 542
8a4b83cc
CM		 543 return 0;
544}
545
2b82032c
YZ		 546int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
547{
e4404d6e 548 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 549 int ret;
550
551 mutex_lock(&uuid_mutex);
552 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 553 if (!fs_devices->opened) {
554 seed_devices = fs_devices->seed;
555 fs_devices->seed = NULL;
556 }
2b82032c 557 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 558
559 while (seed_devices) {
560 fs_devices = seed_devices;
561 seed_devices = fs_devices->seed;
562 __btrfs_close_devices(fs_devices);
563 free_fs_devices(fs_devices);
564 }
2b82032c
YZ		 565 return ret;
566}
567
e4404d6e
YZ		 568static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
569 fmode_t flags, void *holder)
8a4b83cc 570{
d5e2003c 571 struct request_queue *q;
8a4b83cc
CM		 572 struct block_device *bdev;
573 struct list_head *head = &fs_devices->devices;
8a4b83cc 574 struct btrfs_device *device;
a0af469b
CM		 575 struct block_device *latest_bdev = NULL;
576 struct buffer_head *bh;
577 struct btrfs_super_block *disk_super;
578 u64 latest_devid = 0;
579 u64 latest_transid = 0;
a0af469b 580 u64 devid;
2b82032c 581 int seeding = 1;
a0af469b 582 int ret = 0;
8a4b83cc 583
d4d77629
TH		 584 flags |= FMODE_EXCL;
585
c6e30871 586 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 587 if (device->bdev)
588 continue;
dfe25020
CM		 589 if (!device->name)
590 continue;
591
d4d77629 592 bdev = blkdev_get_by_path(device->name, flags, holder);
8a4b83cc 593 if (IS_ERR(bdev)) {
d397712b 594 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 595 goto error;
8a4b83cc 596 }
a061fc8d 597 set_blocksize(bdev, 4096);
a0af469b 598
a512bbf8 599 bh = btrfs_read_dev_super(bdev);
20b45077
DY		 600 if (!bh) {
601 ret = -EINVAL;
a0af469b 602 goto error_close;
20b45077 603 }
a0af469b
CM		 604
605 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 606 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 607 if (devid != device->devid)
608 goto error_brelse;
609
2b82032c
YZ		 610 if (memcmp(device->uuid, disk_super->dev_item.uuid,
611 BTRFS_UUID_SIZE))
612 goto error_brelse;
613
614 device->generation = btrfs_super_generation(disk_super);
615 if (!latest_transid || device->generation > latest_transid) {
a0af469b 616 latest_devid = devid;
2b82032c 617 latest_transid = device->generation;
a0af469b
CM		 618 latest_bdev = bdev;
619 }
620
2b82032c
YZ		 621 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
622 device->writeable = 0;
623 } else {
624 device->writeable = !bdev_read_only(bdev);
625 seeding = 0;
626 }
627
d5e2003c
JB		 628 q = bdev_get_queue(bdev);
629 if (blk_queue_discard(q)) {
630 device->can_discard = 1;
631 fs_devices->num_can_discard++;
632 }
633
8a4b83cc 634 device->bdev = bdev;
dfe25020 635 device->in_fs_metadata = 0;
15916de8
CM		 636 device->mode = flags;
637
c289811c
CM		 638 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
639 fs_devices->rotating = 1;
640
a0af469b 641 fs_devices->open_devices++;
2b82032c
YZ		 642 if (device->writeable) {
643 fs_devices->rw_devices++;
644 list_add(&device->dev_alloc_list,
645 &fs_devices->alloc_list);
646 }
4f6c9328 647 brelse(bh);
a0af469b 648 continue;
a061fc8d 649
a0af469b
CM		 650error_brelse:
651 brelse(bh);
652error_close:
d4d77629 653 blkdev_put(bdev, flags);
a0af469b
CM		 654error:
655 continue;
8a4b83cc 656 }
a0af469b
CM		 657 if (fs_devices->open_devices == 0) {
658 ret = -EIO;
659 goto out;
660 }
2b82032c
YZ		 661 fs_devices->seeding = seeding;
662 fs_devices->opened = 1;
a0af469b
CM		 663 fs_devices->latest_bdev = latest_bdev;
664 fs_devices->latest_devid = latest_devid;
665 fs_devices->latest_trans = latest_transid;
2b82032c 666 fs_devices->total_rw_bytes = 0;
a0af469b 667out:
2b82032c
YZ		 668 return ret;
669}
670
671int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 672 fmode_t flags, void *holder)
2b82032c
YZ		 673{
674 int ret;
675
676 mutex_lock(&uuid_mutex);
677 if (fs_devices->opened) {
e4404d6e
YZ		 678 fs_devices->opened++;
679 ret = 0;
2b82032c 680 } else {
15916de8 681 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 682 }
8a4b83cc 683 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 684 return ret;
685}
686
97288f2c 687int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 688 struct btrfs_fs_devices **fs_devices_ret)
689{
690 struct btrfs_super_block *disk_super;
691 struct block_device *bdev;
692 struct buffer_head *bh;
693 int ret;
694 u64 devid;
f2984462 695 u64 transid;
8a4b83cc
CM		 696
697 mutex_lock(&uuid_mutex);
698
d4d77629
TH		 699 flags |= FMODE_EXCL;
700 bdev = blkdev_get_by_path(path, flags, holder);
8a4b83cc
CM		 701
702 if (IS_ERR(bdev)) {
8a4b83cc
CM		 703 ret = PTR_ERR(bdev);
704 goto error;
705 }
706
707 ret = set_blocksize(bdev, 4096);
708 if (ret)
709 goto error_close;
a512bbf8 710 bh = btrfs_read_dev_super(bdev);
8a4b83cc 711 if (!bh) {
20b45077 712 ret = -EINVAL;
8a4b83cc
CM		 713 goto error_close;
714 }
715 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 716 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 717 transid = btrfs_super_generation(disk_super);
7ae9c09d 718 if (disk_super->label[0])
d397712b 719 printk(KERN_INFO "device label %s ", disk_super->label);
22b63a29
ID		 720 else
721 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
119e10cf 722 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 723 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 724 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
725
8a4b83cc
CM		 726 brelse(bh);
727error_close:
d4d77629 728 blkdev_put(bdev, flags);
8a4b83cc
CM		 729error:
730 mutex_unlock(&uuid_mutex);
731 return ret;
732}
0b86a832 733
6d07bcec
MX		 734/* helper to account the used device space in the range */
735int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
736 u64 end, u64 *length)
737{
738 struct btrfs_key key;
739 struct btrfs_root *root = device->dev_root;
740 struct btrfs_dev_extent *dev_extent;
741 struct btrfs_path *path;
742 u64 extent_end;
743 int ret;
744 int slot;
745 struct extent_buffer *l;
746
747 *length = 0;
748
749 if (start >= device->total_bytes)
750 return 0;
751
752 path = btrfs_alloc_path();
753 if (!path)
754 return -ENOMEM;
755 path->reada = 2;
756
757 key.objectid = device->devid;
758 key.offset = start;
759 key.type = BTRFS_DEV_EXTENT_KEY;
760
761 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
762 if (ret < 0)
763 goto out;
764 if (ret > 0) {
765 ret = btrfs_previous_item(root, path, key.objectid, key.type);
766 if (ret < 0)
767 goto out;
768 }
769
770 while (1) {
771 l = path->nodes[0];
772 slot = path->slots[0];
773 if (slot >= btrfs_header_nritems(l)) {
774 ret = btrfs_next_leaf(root, path);
775 if (ret == 0)
776 continue;
777 if (ret < 0)
778 goto out;
779
780 break;
781 }
782 btrfs_item_key_to_cpu(l, &key, slot);
783
784 if (key.objectid < device->devid)
785 goto next;
786
787 if (key.objectid > device->devid)
788 break;
789
790 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
791 goto next;
792
793 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
794 extent_end = key.offset + btrfs_dev_extent_length(l,
795 dev_extent);
796 if (key.offset <= start && extent_end > end) {
797 *length = end - start + 1;
798 break;
799 } else if (key.offset <= start && extent_end > start)
800 *length += extent_end - start;
801 else if (key.offset > start && extent_end <= end)
802 *length += extent_end - key.offset;
803 else if (key.offset > start && key.offset <= end) {
804 *length += end - key.offset + 1;
805 break;
806 } else if (key.offset > end)
807 break;
808
809next:
810 path->slots[0]++;
811 }
812 ret = 0;
813out:
814 btrfs_free_path(path);
815 return ret;
816}
817
0b86a832 818/*
7bfc837d
MX		 819 * find_free_dev_extent - find free space in the specified device
820 * @trans: transaction handler
821 * @device: the device which we search the free space in
822 * @num_bytes: the size of the free space that we need
823 * @start: store the start of the free space.
824 * @len: the size of the free space. that we find, or the size of the max
825 * free space if we don't find suitable free space
826 *
0b86a832
CM		 827 * this uses a pretty simple search, the expectation is that it is
828 * called very infrequently and that a given device has a small number
829 * of extents
7bfc837d
MX		 830 *
831 * @start is used to store the start of the free space if we find. But if we
832 * don't find suitable free space, it will be used to store the start position
833 * of the max free space.
834 *
835 * @len is used to store the size of the free space that we find.
836 * But if we don't find suitable free space, it is used to store the size of
837 * the max free space.
0b86a832 838 */
ba1bf481
JB		 839int find_free_dev_extent(struct btrfs_trans_handle *trans,
840 struct btrfs_device *device, u64 num_bytes,
7bfc837d 841 u64 *start, u64 *len)
0b86a832
CM		 842{
843 struct btrfs_key key;
844 struct btrfs_root *root = device->dev_root;
7bfc837d 845 struct btrfs_dev_extent *dev_extent;
2b82032c 846 struct btrfs_path *path;
7bfc837d
MX		 847 u64 hole_size;
848 u64 max_hole_start;
849 u64 max_hole_size;
850 u64 extent_end;
851 u64 search_start;
0b86a832
CM		 852 u64 search_end = device->total_bytes;
853 int ret;
7bfc837d 854 int slot;
0b86a832
CM		 855 struct extent_buffer *l;
856
0b86a832
CM		 857 /* FIXME use last free of some kind */
858
8a4b83cc
CM		 859 /* we don't want to overwrite the superblock on the drive,
860 * so we make sure to start at an offset of at least 1MB
861 */
a9c9bf68 862 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
8f18cf13 863
7bfc837d
MX		 864 max_hole_start = search_start;
865 max_hole_size = 0;
38c01b96 866 hole_size = 0;
7bfc837d
MX		 867
868 if (search_start >= search_end) {
869 ret = -ENOSPC;
870 goto error;
871 }
872
873 path = btrfs_alloc_path();
874 if (!path) {
875 ret = -ENOMEM;
876 goto error;
877 }
878 path->reada = 2;
879
0b86a832
CM		 880 key.objectid = device->devid;
881 key.offset = search_start;
882 key.type = BTRFS_DEV_EXTENT_KEY;
7bfc837d 883
0b86a832
CM		 884 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
885 if (ret < 0)
7bfc837d 886 goto out;
1fcbac58
YZ		 887 if (ret > 0) {
888 ret = btrfs_previous_item(root, path, key.objectid, key.type);
889 if (ret < 0)
7bfc837d 890 goto out;
1fcbac58 891 }
7bfc837d 892
0b86a832
CM		 893 while (1) {
894 l = path->nodes[0];
895 slot = path->slots[0];
896 if (slot >= btrfs_header_nritems(l)) {
897 ret = btrfs_next_leaf(root, path);
898 if (ret == 0)
899 continue;
900 if (ret < 0)
7bfc837d
MX		 901 goto out;
902
903 break;
0b86a832
CM		 904 }
905 btrfs_item_key_to_cpu(l, &key, slot);
906
907 if (key.objectid < device->devid)
908 goto next;
909
910 if (key.objectid > device->devid)
7bfc837d 911 break;
0b86a832 912
7bfc837d
MX		 913 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
914 goto next;
9779b72f 915
7bfc837d
MX		 916 if (key.offset > search_start) {
917 hole_size = key.offset - search_start;
9779b72f 918
7bfc837d
MX		 919 if (hole_size > max_hole_size) {
920 max_hole_start = search_start;
921 max_hole_size = hole_size;
922 }
9779b72f 923
7bfc837d
MX		 924 /*
925 * If this free space is greater than which we need,
926 * it must be the max free space that we have found
927 * until now, so max_hole_start must point to the start
928 * of this free space and the length of this free space
929 * is stored in max_hole_size. Thus, we return
930 * max_hole_start and max_hole_size and go back to the
931 * caller.
932 */
933 if (hole_size >= num_bytes) {
934 ret = 0;
935 goto out;
0b86a832
CM		 936 }
937 }
0b86a832 938
0b86a832 939 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
7bfc837d
MX		 940 extent_end = key.offset + btrfs_dev_extent_length(l,
941 dev_extent);
942 if (extent_end > search_start)
943 search_start = extent_end;
0b86a832
CM		 944next:
945 path->slots[0]++;
946 cond_resched();
947 }
0b86a832 948
38c01b96 949 /*
950 * At this point, search_start should be the end of
951 * allocated dev extents, and when shrinking the device,
952 * search_end may be smaller than search_start.
953 */
954 if (search_end > search_start)
955 hole_size = search_end - search_start;
956
7bfc837d
MX		 957 if (hole_size > max_hole_size) {
958 max_hole_start = search_start;
959 max_hole_size = hole_size;
0b86a832 960 }
0b86a832 961
7bfc837d
MX		 962 /* See above. */
963 if (hole_size < num_bytes)
964 ret = -ENOSPC;
965 else
966 ret = 0;
967
968out:
2b82032c 969 btrfs_free_path(path);
7bfc837d
MX		 970error:
971 *start = max_hole_start;
b2117a39 972 if (len)
7bfc837d 973 *len = max_hole_size;
0b86a832
CM		 974 return ret;
975}
976
b2950863 977static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 978 struct btrfs_device *device,
979 u64 start)
980{
981 int ret;
982 struct btrfs_path *path;
983 struct btrfs_root *root = device->dev_root;
984 struct btrfs_key key;
a061fc8d
CM		 985 struct btrfs_key found_key;
986 struct extent_buffer *leaf = NULL;
987 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 988
989 path = btrfs_alloc_path();
990 if (!path)
991 return -ENOMEM;
992
993 key.objectid = device->devid;
994 key.offset = start;
995 key.type = BTRFS_DEV_EXTENT_KEY;
996
997 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 998 if (ret > 0) {
999 ret = btrfs_previous_item(root, path, key.objectid,
1000 BTRFS_DEV_EXTENT_KEY);
b0b802d7
TI		 1001 if (ret)
1002 goto out;
a061fc8d
CM		 1003 leaf = path->nodes[0];
1004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1005 extent = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_dev_extent);
1007 BUG_ON(found_key.offset > start || found_key.offset +
1008 btrfs_dev_extent_length(leaf, extent) < start);
a061fc8d
CM		 1009 } else if (ret == 0) {
1010 leaf = path->nodes[0];
1011 extent = btrfs_item_ptr(leaf, path->slots[0],
1012 struct btrfs_dev_extent);
1013 }
8f18cf13
CM		 1014 BUG_ON(ret);
1015
dfe25020
CM		 1016 if (device->bytes_used > 0)
1017 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
8f18cf13 1018 ret = btrfs_del_item(trans, root, path);
8f18cf13 1019
b0b802d7 1020out:
8f18cf13
CM		 1021 btrfs_free_path(path);
1022 return ret;
1023}
1024
2b82032c 1025int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 1026 struct btrfs_device *device,
e17cade2 1027 u64 chunk_tree, u64 chunk_objectid,
2b82032c 1028 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 1029{
1030 int ret;
1031 struct btrfs_path *path;
1032 struct btrfs_root *root = device->dev_root;
1033 struct btrfs_dev_extent *extent;
1034 struct extent_buffer *leaf;
1035 struct btrfs_key key;
1036
dfe25020 1037 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 1038 path = btrfs_alloc_path();
1039 if (!path)
1040 return -ENOMEM;
1041
0b86a832 1042 key.objectid = device->devid;
2b82032c 1043 key.offset = start;
0b86a832
CM		 1044 key.type = BTRFS_DEV_EXTENT_KEY;
1045 ret = btrfs_insert_empty_item(trans, root, path, &key,
1046 sizeof(*extent));
1047 BUG_ON(ret);
1048
1049 leaf = path->nodes[0];
1050 extent = btrfs_item_ptr(leaf, path->slots[0],
1051 struct btrfs_dev_extent);
e17cade2
CM		 1052 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1053 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1054 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1055
1056 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1057 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1058 BTRFS_UUID_SIZE);
1059
0b86a832
CM		 1060 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1061 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 1062 btrfs_free_path(path);
1063 return ret;
1064}
1065
a1b32a59
CM		 1066static noinline int find_next_chunk(struct btrfs_root *root,
1067 u64 objectid, u64 *offset)
0b86a832
CM		 1068{
1069 struct btrfs_path *path;
1070 int ret;
1071 struct btrfs_key key;
e17cade2 1072 struct btrfs_chunk *chunk;
0b86a832
CM		 1073 struct btrfs_key found_key;
1074
1075 path = btrfs_alloc_path();
92b8e897
MF		 1076 if (!path)
1077 return -ENOMEM;
0b86a832 1078
e17cade2 1079 key.objectid = objectid;
0b86a832
CM		 1080 key.offset = (u64)-1;
1081 key.type = BTRFS_CHUNK_ITEM_KEY;
1082
1083 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1084 if (ret < 0)
1085 goto error;
1086
1087 BUG_ON(ret == 0);
1088
1089 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1090 if (ret) {
e17cade2 1091 *offset = 0;
0b86a832
CM		 1092 } else {
1093 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1094 path->slots[0]);
e17cade2
CM		 1095 if (found_key.objectid != objectid)
1096 *offset = 0;
1097 else {
1098 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1099 struct btrfs_chunk);
1100 *offset = found_key.offset +
1101 btrfs_chunk_length(path->nodes[0], chunk);
1102 }
0b86a832
CM		 1103 }
1104 ret = 0;
1105error:
1106 btrfs_free_path(path);
1107 return ret;
1108}
1109
2b82032c 1110static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 1111{
1112 int ret;
1113 struct btrfs_key key;
1114 struct btrfs_key found_key;
2b82032c
YZ		 1115 struct btrfs_path *path;
1116
1117 root = root->fs_info->chunk_root;
1118
1119 path = btrfs_alloc_path();
1120 if (!path)
1121 return -ENOMEM;
0b86a832
CM		 1122
1123 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1124 key.type = BTRFS_DEV_ITEM_KEY;
1125 key.offset = (u64)-1;
1126
1127 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1128 if (ret < 0)
1129 goto error;
1130
1131 BUG_ON(ret == 0);
1132
1133 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1134 BTRFS_DEV_ITEM_KEY);
1135 if (ret) {
1136 *objectid = 1;
1137 } else {
1138 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1139 path->slots[0]);
1140 *objectid = found_key.offset + 1;
1141 }
1142 ret = 0;
1143error:
2b82032c 1144 btrfs_free_path(path);
0b86a832
CM		 1145 return ret;
1146}
1147
1148/*
1149 * the device information is stored in the chunk root
1150 * the btrfs_device struct should be fully filled in
1151 */
1152int btrfs_add_device(struct btrfs_trans_handle *trans,
1153 struct btrfs_root *root,
1154 struct btrfs_device *device)
1155{
1156 int ret;
1157 struct btrfs_path *path;
1158 struct btrfs_dev_item *dev_item;
1159 struct extent_buffer *leaf;
1160 struct btrfs_key key;
1161 unsigned long ptr;
0b86a832
CM		 1162
1163 root = root->fs_info->chunk_root;
1164
1165 path = btrfs_alloc_path();
1166 if (!path)
1167 return -ENOMEM;
1168
0b86a832
CM		 1169 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1170 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1171 key.offset = device->devid;
0b86a832
CM		 1172
1173 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1174 sizeof(*dev_item));
0b86a832
CM		 1175 if (ret)
1176 goto out;
1177
1178 leaf = path->nodes[0];
1179 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1180
1181 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1182 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1183 btrfs_set_device_type(leaf, dev_item, device->type);
1184 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1185 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1186 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1187 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1188 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1189 btrfs_set_device_group(leaf, dev_item, 0);
1190 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1191 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1192 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1193
0b86a832 1194 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1195 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1196 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1197 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1198 btrfs_mark_buffer_dirty(leaf);
0b86a832 1199
2b82032c 1200 ret = 0;
0b86a832
CM		 1201out:
1202 btrfs_free_path(path);
1203 return ret;
1204}
8f18cf13 1205
a061fc8d
CM		 1206static int btrfs_rm_dev_item(struct btrfs_root *root,
1207 struct btrfs_device *device)
1208{
1209 int ret;
1210 struct btrfs_path *path;
a061fc8d 1211 struct btrfs_key key;
a061fc8d
CM		 1212 struct btrfs_trans_handle *trans;
1213
1214 root = root->fs_info->chunk_root;
1215
1216 path = btrfs_alloc_path();
1217 if (!path)
1218 return -ENOMEM;
1219
a22285a6 1220 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 1221 if (IS_ERR(trans)) {
1222 btrfs_free_path(path);
1223 return PTR_ERR(trans);
1224 }
a061fc8d
CM		 1225 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1226 key.type = BTRFS_DEV_ITEM_KEY;
1227 key.offset = device->devid;
7d9eb12c 1228 lock_chunks(root);
a061fc8d
CM		 1229
1230 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1231 if (ret < 0)
1232 goto out;
1233
1234 if (ret > 0) {
1235 ret = -ENOENT;
1236 goto out;
1237 }
1238
1239 ret = btrfs_del_item(trans, root, path);
1240 if (ret)
1241 goto out;
a061fc8d
CM		 1242out:
1243 btrfs_free_path(path);
7d9eb12c 1244 unlock_chunks(root);
a061fc8d
CM		 1245 btrfs_commit_transaction(trans, root);
1246 return ret;
1247}
1248
1249int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1250{
1251 struct btrfs_device *device;
2b82032c 1252 struct btrfs_device *next_device;
a061fc8d 1253 struct block_device *bdev;
dfe25020 1254 struct buffer_head *bh = NULL;
a061fc8d 1255 struct btrfs_super_block *disk_super;
1f78160c 1256 struct btrfs_fs_devices *cur_devices;
a061fc8d
CM		 1257 u64 all_avail;
1258 u64 devid;
2b82032c
YZ		 1259 u64 num_devices;
1260 u8 *dev_uuid;
a061fc8d 1261 int ret = 0;
1f78160c 1262 bool clear_super = false;
a061fc8d 1263
a061fc8d 1264 mutex_lock(&uuid_mutex);
7d9eb12c 1265 mutex_lock(&root->fs_info->volume_mutex);
a061fc8d
CM		 1266
1267 all_avail = root->fs_info->avail_data_alloc_bits |
1268 root->fs_info->avail_system_alloc_bits |
1269 root->fs_info->avail_metadata_alloc_bits;
1270
1271 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
035fe03a 1272 root->fs_info->fs_devices->num_devices <= 4) {
d397712b
CM		 1273 printk(KERN_ERR "btrfs: unable to go below four devices "
1274 "on raid10\n");
a061fc8d
CM		 1275 ret = -EINVAL;
1276 goto out;
1277 }
1278
1279 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
035fe03a 1280 root->fs_info->fs_devices->num_devices <= 2) {
d397712b
CM		 1281 printk(KERN_ERR "btrfs: unable to go below two "
1282 "devices on raid1\n");
a061fc8d
CM		 1283 ret = -EINVAL;
1284 goto out;
1285 }
1286
dfe25020 1287 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1288 struct list_head *devices;
1289 struct btrfs_device *tmp;
a061fc8d 1290
dfe25020
CM		 1291 device = NULL;
1292 devices = &root->fs_info->fs_devices->devices;
46224705
XG		 1293 /*
1294 * It is safe to read the devices since the volume_mutex
1295 * is held.
1296 */
c6e30871 1297 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1298 if (tmp->in_fs_metadata && !tmp->bdev) {
1299 device = tmp;
1300 break;
1301 }
1302 }
1303 bdev = NULL;
1304 bh = NULL;
1305 disk_super = NULL;
1306 if (!device) {
d397712b
CM		 1307 printk(KERN_ERR "btrfs: no missing devices found to "
1308 "remove\n");
dfe25020
CM		 1309 goto out;
1310 }
dfe25020 1311 } else {
d4d77629
TH		 1312 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1313 root->fs_info->bdev_holder);
dfe25020
CM		 1314 if (IS_ERR(bdev)) {
1315 ret = PTR_ERR(bdev);
1316 goto out;
1317 }
a061fc8d 1318
2b82032c 1319 set_blocksize(bdev, 4096);
a512bbf8 1320 bh = btrfs_read_dev_super(bdev);
dfe25020 1321 if (!bh) {
20b45077 1322 ret = -EINVAL;
dfe25020
CM		 1323 goto error_close;
1324 }
1325 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 1326 devid = btrfs_stack_device_id(&disk_super->dev_item);
2b82032c
YZ		 1327 dev_uuid = disk_super->dev_item.uuid;
1328 device = btrfs_find_device(root, devid, dev_uuid,
1329 disk_super->fsid);
dfe25020
CM		 1330 if (!device) {
1331 ret = -ENOENT;
1332 goto error_brelse;
1333 }
2b82032c 1334 }
dfe25020 1335
2b82032c 1336 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1337 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1338 "device\n");
2b82032c
YZ		 1339 ret = -EINVAL;
1340 goto error_brelse;
1341 }
1342
1343 if (device->writeable) {
0c1daee0 1344 lock_chunks(root);
2b82032c 1345 list_del_init(&device->dev_alloc_list);
0c1daee0 1346 unlock_chunks(root);
2b82032c 1347 root->fs_info->fs_devices->rw_devices--;
1f78160c 1348 clear_super = true;
dfe25020 1349 }
a061fc8d
CM		 1350
1351 ret = btrfs_shrink_device(device, 0);
1352 if (ret)
9b3517e9 1353 goto error_undo;
a061fc8d 1354
a061fc8d
CM		 1355 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1356 if (ret)
9b3517e9 1357 goto error_undo;
a061fc8d 1358
2b82032c 1359 device->in_fs_metadata = 0;
a2de733c 1360 btrfs_scrub_cancel_dev(root, device);
e5e9a520
CM		 1361
1362 /*
1363 * the device list mutex makes sure that we don't change
1364 * the device list while someone else is writing out all
1365 * the device supers.
1366 */
1f78160c
XG		 1367
1368 cur_devices = device->fs_devices;
e5e9a520 1369 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1370 list_del_rcu(&device->dev_list);
e5e9a520 1371
e4404d6e 1372 device->fs_devices->num_devices--;
2b82032c 1373
cd02dca5
CM		 1374 if (device->missing)
1375 root->fs_info->fs_devices->missing_devices--;
1376
2b82032c
YZ		 1377 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1378 struct btrfs_device, dev_list);
1379 if (device->bdev == root->fs_info->sb->s_bdev)
1380 root->fs_info->sb->s_bdev = next_device->bdev;
1381 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1382 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1383
1f78160c 1384 if (device->bdev)
e4404d6e 1385 device->fs_devices->open_devices--;
1f78160c
XG		 1386
1387 call_rcu(&device->rcu, free_device);
1388 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1389
2b82032c
YZ		 1390 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1391 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1392
1f78160c 1393 if (cur_devices->open_devices == 0) {
e4404d6e
YZ		 1394 struct btrfs_fs_devices *fs_devices;
1395 fs_devices = root->fs_info->fs_devices;
1396 while (fs_devices) {
1f78160c 1397 if (fs_devices->seed == cur_devices)
e4404d6e
YZ		 1398 break;
1399 fs_devices = fs_devices->seed;
2b82032c 1400 }
1f78160c
XG		 1401 fs_devices->seed = cur_devices->seed;
1402 cur_devices->seed = NULL;
0c1daee0 1403 lock_chunks(root);
1f78160c 1404 __btrfs_close_devices(cur_devices);
0c1daee0 1405 unlock_chunks(root);
1f78160c 1406 free_fs_devices(cur_devices);
2b82032c
YZ		 1407 }
1408
1409 /*
1410 * at this point, the device is zero sized. We want to
1411 * remove it from the devices list and zero out the old super
1412 */
1f78160c 1413 if (clear_super) {
dfe25020
CM		 1414 /* make sure this device isn't detected as part of
1415 * the FS anymore
1416 */
1417 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1418 set_buffer_dirty(bh);
1419 sync_dirty_buffer(bh);
dfe25020 1420 }
a061fc8d 1421
a061fc8d 1422 ret = 0;
a061fc8d
CM		 1423
1424error_brelse:
1425 brelse(bh);
1426error_close:
dfe25020 1427 if (bdev)
e525fd89 1428 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
a061fc8d 1429out:
7d9eb12c 1430 mutex_unlock(&root->fs_info->volume_mutex);
a061fc8d 1431 mutex_unlock(&uuid_mutex);
a061fc8d 1432 return ret;
9b3517e9
ID		 1433error_undo:
1434 if (device->writeable) {
0c1daee0 1435 lock_chunks(root);
9b3517e9
ID		 1436 list_add(&device->dev_alloc_list,
1437 &root->fs_info->fs_devices->alloc_list);
0c1daee0 1438 unlock_chunks(root);
9b3517e9
ID		 1439 root->fs_info->fs_devices->rw_devices++;
1440 }
1441 goto error_brelse;
a061fc8d
CM		 1442}
1443
2b82032c
YZ		 1444/*
1445 * does all the dirty work required for changing file system's UUID.
1446 */
1447static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1448 struct btrfs_root *root)
1449{
1450 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1451 struct btrfs_fs_devices *old_devices;
e4404d6e 1452 struct btrfs_fs_devices *seed_devices;
2b82032c
YZ		 1453 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1454 struct btrfs_device *device;
1455 u64 super_flags;
1456
1457 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1458 if (!fs_devices->seeding)
2b82032c
YZ		 1459 return -EINVAL;
1460
e4404d6e
YZ		 1461 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1462 if (!seed_devices)
2b82032c
YZ		 1463 return -ENOMEM;
1464
e4404d6e
YZ		 1465 old_devices = clone_fs_devices(fs_devices);
1466 if (IS_ERR(old_devices)) {
1467 kfree(seed_devices);
1468 return PTR_ERR(old_devices);
2b82032c 1469 }
e4404d6e 1470
2b82032c
YZ		 1471 list_add(&old_devices->list, &fs_uuids);
1472
e4404d6e
YZ		 1473 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1474 seed_devices->opened = 1;
1475 INIT_LIST_HEAD(&seed_devices->devices);
1476 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1477 mutex_init(&seed_devices->device_list_mutex);
c9513edb
XG		 1478
1479 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c
XG		 1480 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1481 synchronize_rcu);
c9513edb
XG		 1482 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1483
e4404d6e
YZ		 1484 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1485 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1486 device->fs_devices = seed_devices;
1487 }
1488
2b82032c
YZ		 1489 fs_devices->seeding = 0;
1490 fs_devices->num_devices = 0;
1491 fs_devices->open_devices = 0;
e4404d6e 1492 fs_devices->seed = seed_devices;
2b82032c
YZ		 1493
1494 generate_random_uuid(fs_devices->fsid);
1495 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1496 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1497 super_flags = btrfs_super_flags(disk_super) &
1498 ~BTRFS_SUPER_FLAG_SEEDING;
1499 btrfs_set_super_flags(disk_super, super_flags);
1500
1501 return 0;
1502}
1503
1504/*
1505 * strore the expected generation for seed devices in device items.
1506 */
1507static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1508 struct btrfs_root *root)
1509{
1510 struct btrfs_path *path;
1511 struct extent_buffer *leaf;
1512 struct btrfs_dev_item *dev_item;
1513 struct btrfs_device *device;
1514 struct btrfs_key key;
1515 u8 fs_uuid[BTRFS_UUID_SIZE];
1516 u8 dev_uuid[BTRFS_UUID_SIZE];
1517 u64 devid;
1518 int ret;
1519
1520 path = btrfs_alloc_path();
1521 if (!path)
1522 return -ENOMEM;
1523
1524 root = root->fs_info->chunk_root;
1525 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1526 key.offset = 0;
1527 key.type = BTRFS_DEV_ITEM_KEY;
1528
1529 while (1) {
1530 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1531 if (ret < 0)
1532 goto error;
1533
1534 leaf = path->nodes[0];
1535next_slot:
1536 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1537 ret = btrfs_next_leaf(root, path);
1538 if (ret > 0)
1539 break;
1540 if (ret < 0)
1541 goto error;
1542 leaf = path->nodes[0];
1543 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 1544 btrfs_release_path(path);
2b82032c
YZ		 1545 continue;
1546 }
1547
1548 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1549 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1550 key.type != BTRFS_DEV_ITEM_KEY)
1551 break;
1552
1553 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1554 struct btrfs_dev_item);
1555 devid = btrfs_device_id(leaf, dev_item);
1556 read_extent_buffer(leaf, dev_uuid,
1557 (unsigned long)btrfs_device_uuid(dev_item),
1558 BTRFS_UUID_SIZE);
1559 read_extent_buffer(leaf, fs_uuid,
1560 (unsigned long)btrfs_device_fsid(dev_item),
1561 BTRFS_UUID_SIZE);
1562 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1563 BUG_ON(!device);
1564
1565 if (device->fs_devices->seeding) {
1566 btrfs_set_device_generation(leaf, dev_item,
1567 device->generation);
1568 btrfs_mark_buffer_dirty(leaf);
1569 }
1570
1571 path->slots[0]++;
1572 goto next_slot;
1573 }
1574 ret = 0;
1575error:
1576 btrfs_free_path(path);
1577 return ret;
1578}
1579
788f20eb
CM		 1580int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1581{
d5e2003c 1582 struct request_queue *q;
788f20eb
CM		 1583 struct btrfs_trans_handle *trans;
1584 struct btrfs_device *device;
1585 struct block_device *bdev;
788f20eb 1586 struct list_head *devices;
2b82032c 1587 struct super_block *sb = root->fs_info->sb;
788f20eb 1588 u64 total_bytes;
2b82032c 1589 int seeding_dev = 0;
788f20eb
CM		 1590 int ret = 0;
1591
2b82032c
YZ		 1592 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1593 return -EINVAL;
788f20eb 1594
d4d77629
TH		 1595 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1596 root->fs_info->bdev_holder);
7f59203a
JB		 1597 if (IS_ERR(bdev))
1598 return PTR_ERR(bdev);
a2135011 1599
2b82032c
YZ		 1600 if (root->fs_info->fs_devices->seeding) {
1601 seeding_dev = 1;
1602 down_write(&sb->s_umount);
1603 mutex_lock(&uuid_mutex);
1604 }
1605
8c8bee1d 1606 filemap_write_and_wait(bdev->bd_inode->i_mapping);
7d9eb12c 1607 mutex_lock(&root->fs_info->volume_mutex);
a2135011 1608
788f20eb 1609 devices = &root->fs_info->fs_devices->devices;
e5e9a520
CM		 1610 /*
1611 * we have the volume lock, so we don't need the extra
1612 * device list mutex while reading the list here.
1613 */
c6e30871 1614 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1615 if (device->bdev == bdev) {
1616 ret = -EEXIST;
2b82032c 1617 goto error;
788f20eb
CM		 1618 }
1619 }
1620
1621 device = kzalloc(sizeof(*device), GFP_NOFS);
1622 if (!device) {
1623 /* we can safely leave the fs_devices entry around */
1624 ret = -ENOMEM;
2b82032c 1625 goto error;
788f20eb
CM		 1626 }
1627
788f20eb
CM		 1628 device->name = kstrdup(device_path, GFP_NOFS);
1629 if (!device->name) {
1630 kfree(device);
2b82032c
YZ		 1631 ret = -ENOMEM;
1632 goto error;
788f20eb 1633 }
2b82032c
YZ		 1634
1635 ret = find_next_devid(root, &device->devid);
1636 if (ret) {
67100f25 1637 kfree(device->name);
2b82032c
YZ		 1638 kfree(device);
1639 goto error;
1640 }
1641
a22285a6 1642 trans = btrfs_start_transaction(root, 0);
98d5dc13 1643 if (IS_ERR(trans)) {
67100f25 1644 kfree(device->name);
98d5dc13
TI		 1645 kfree(device);
1646 ret = PTR_ERR(trans);
1647 goto error;
1648 }
1649
2b82032c
YZ		 1650 lock_chunks(root);
1651
d5e2003c
JB		 1652 q = bdev_get_queue(bdev);
1653 if (blk_queue_discard(q))
1654 device->can_discard = 1;
2b82032c
YZ		 1655 device->writeable = 1;
1656 device->work.func = pending_bios_fn;
1657 generate_random_uuid(device->uuid);
1658 spin_lock_init(&device->io_lock);
1659 device->generation = trans->transid;
788f20eb
CM		 1660 device->io_width = root->sectorsize;
1661 device->io_align = root->sectorsize;
1662 device->sector_size = root->sectorsize;
1663 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1664 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1665 device->dev_root = root->fs_info->dev_root;
1666 device->bdev = bdev;
dfe25020 1667 device->in_fs_metadata = 1;
fb01aa85 1668 device->mode = FMODE_EXCL;
2b82032c 1669 set_blocksize(device->bdev, 4096);
788f20eb 1670
2b82032c
YZ		 1671 if (seeding_dev) {
1672 sb->s_flags &= ~MS_RDONLY;
1673 ret = btrfs_prepare_sprout(trans, root);
1674 BUG_ON(ret);
1675 }
788f20eb 1676
2b82032c 1677 device->fs_devices = root->fs_info->fs_devices;
e5e9a520
CM		 1678
1679 /*
1680 * we don't want write_supers to jump in here with our device
1681 * half setup
1682 */
1683 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1684 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2b82032c
YZ		 1685 list_add(&device->dev_alloc_list,
1686 &root->fs_info->fs_devices->alloc_list);
1687 root->fs_info->fs_devices->num_devices++;
1688 root->fs_info->fs_devices->open_devices++;
1689 root->fs_info->fs_devices->rw_devices++;
d5e2003c
JB		 1690 if (device->can_discard)
1691 root->fs_info->fs_devices->num_can_discard++;
2b82032c 1692 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1693
c289811c
CM		 1694 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1695 root->fs_info->fs_devices->rotating = 1;
1696
788f20eb
CM		 1697 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1698 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1699 total_bytes + device->total_bytes);
1700
1701 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1702 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1703 total_bytes + 1);
e5e9a520 1704 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 1705
2b82032c
YZ		 1706 if (seeding_dev) {
1707 ret = init_first_rw_device(trans, root, device);
1708 BUG_ON(ret);
1709 ret = btrfs_finish_sprout(trans, root);
1710 BUG_ON(ret);
1711 } else {
1712 ret = btrfs_add_device(trans, root, device);
1713 }
1714
913d952e
CM		 1715 /*
1716 * we've got more storage, clear any full flags on the space
1717 * infos
1718 */
1719 btrfs_clear_space_info_full(root->fs_info);
1720
7d9eb12c 1721 unlock_chunks(root);
2b82032c 1722 btrfs_commit_transaction(trans, root);
a2135011 1723
2b82032c
YZ		 1724 if (seeding_dev) {
1725 mutex_unlock(&uuid_mutex);
1726 up_write(&sb->s_umount);
788f20eb 1727
2b82032c
YZ		 1728 ret = btrfs_relocate_sys_chunks(root);
1729 BUG_ON(ret);
1730 }
1731out:
1732 mutex_unlock(&root->fs_info->volume_mutex);
1733 return ret;
1734error:
e525fd89 1735 blkdev_put(bdev, FMODE_EXCL);
2b82032c
YZ		 1736 if (seeding_dev) {
1737 mutex_unlock(&uuid_mutex);
1738 up_write(&sb->s_umount);
1739 }
788f20eb
CM		 1740 goto out;
1741}
1742
d397712b
CM		 1743static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1744 struct btrfs_device *device)
0b86a832
CM		 1745{
1746 int ret;
1747 struct btrfs_path *path;
1748 struct btrfs_root *root;
1749 struct btrfs_dev_item *dev_item;
1750 struct extent_buffer *leaf;
1751 struct btrfs_key key;
1752
1753 root = device->dev_root->fs_info->chunk_root;
1754
1755 path = btrfs_alloc_path();
1756 if (!path)
1757 return -ENOMEM;
1758
1759 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1760 key.type = BTRFS_DEV_ITEM_KEY;
1761 key.offset = device->devid;
1762
1763 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1764 if (ret < 0)
1765 goto out;
1766
1767 if (ret > 0) {
1768 ret = -ENOENT;
1769 goto out;
1770 }
1771
1772 leaf = path->nodes[0];
1773 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1774
1775 btrfs_set_device_id(leaf, dev_item, device->devid);
1776 btrfs_set_device_type(leaf, dev_item, device->type);
1777 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1778 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1779 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 1780 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 1781 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1782 btrfs_mark_buffer_dirty(leaf);
1783
1784out:
1785 btrfs_free_path(path);
1786 return ret;
1787}
1788
7d9eb12c 1789static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1790 struct btrfs_device *device, u64 new_size)
1791{
1792 struct btrfs_super_block *super_copy =
1793 &device->dev_root->fs_info->super_copy;
1794 u64 old_total = btrfs_super_total_bytes(super_copy);
1795 u64 diff = new_size - device->total_bytes;
1796
2b82032c
YZ		 1797 if (!device->writeable)
1798 return -EACCES;
1799 if (new_size <= device->total_bytes)
1800 return -EINVAL;
1801
8f18cf13 1802 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1803 device->fs_devices->total_rw_bytes += diff;
1804
1805 device->total_bytes = new_size;
9779b72f 1806 device->disk_total_bytes = new_size;
4184ea7f
CM		 1807 btrfs_clear_space_info_full(device->dev_root->fs_info);
1808
8f18cf13
CM		 1809 return btrfs_update_device(trans, device);
1810}
1811
7d9eb12c
CM		 1812int btrfs_grow_device(struct btrfs_trans_handle *trans,
1813 struct btrfs_device *device, u64 new_size)
1814{
1815 int ret;
1816 lock_chunks(device->dev_root);
1817 ret = __btrfs_grow_device(trans, device, new_size);
1818 unlock_chunks(device->dev_root);
1819 return ret;
1820}
1821
8f18cf13
CM		 1822static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root,
1824 u64 chunk_tree, u64 chunk_objectid,
1825 u64 chunk_offset)
1826{
1827 int ret;
1828 struct btrfs_path *path;
1829 struct btrfs_key key;
1830
1831 root = root->fs_info->chunk_root;
1832 path = btrfs_alloc_path();
1833 if (!path)
1834 return -ENOMEM;
1835
1836 key.objectid = chunk_objectid;
1837 key.offset = chunk_offset;
1838 key.type = BTRFS_CHUNK_ITEM_KEY;
1839
1840 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1841 BUG_ON(ret);
1842
1843 ret = btrfs_del_item(trans, root, path);
8f18cf13
CM		 1844
1845 btrfs_free_path(path);
65a246c5 1846 return ret;
8f18cf13
CM		 1847}
1848
b2950863 1849static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1850 chunk_offset)
1851{
1852 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1853 struct btrfs_disk_key *disk_key;
1854 struct btrfs_chunk *chunk;
1855 u8 *ptr;
1856 int ret = 0;
1857 u32 num_stripes;
1858 u32 array_size;
1859 u32 len = 0;
1860 u32 cur;
1861 struct btrfs_key key;
1862
1863 array_size = btrfs_super_sys_array_size(super_copy);
1864
1865 ptr = super_copy->sys_chunk_array;
1866 cur = 0;
1867
1868 while (cur < array_size) {
1869 disk_key = (struct btrfs_disk_key *)ptr;
1870 btrfs_disk_key_to_cpu(&key, disk_key);
1871
1872 len = sizeof(*disk_key);
1873
1874 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1875 chunk = (struct btrfs_chunk *)(ptr + len);
1876 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1877 len += btrfs_chunk_item_size(num_stripes);
1878 } else {
1879 ret = -EIO;
1880 break;
1881 }
1882 if (key.objectid == chunk_objectid &&
1883 key.offset == chunk_offset) {
1884 memmove(ptr, ptr + len, array_size - (cur + len));
1885 array_size -= len;
1886 btrfs_set_super_sys_array_size(super_copy, array_size);
1887 } else {
1888 ptr += len;
1889 cur += len;
1890 }
1891 }
1892 return ret;
1893}
1894
b2950863 1895static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1896 u64 chunk_tree, u64 chunk_objectid,
1897 u64 chunk_offset)
1898{
1899 struct extent_map_tree *em_tree;
1900 struct btrfs_root *extent_root;
1901 struct btrfs_trans_handle *trans;
1902 struct extent_map *em;
1903 struct map_lookup *map;
1904 int ret;
1905 int i;
1906
1907 root = root->fs_info->chunk_root;
1908 extent_root = root->fs_info->extent_root;
1909 em_tree = &root->fs_info->mapping_tree.map_tree;
1910
ba1bf481
JB		 1911 ret = btrfs_can_relocate(extent_root, chunk_offset);
1912 if (ret)
1913 return -ENOSPC;
1914
8f18cf13 1915 /* step one, relocate all the extents inside this chunk */
1a40e23b 1916 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
a22285a6
YZ		 1917 if (ret)
1918 return ret;
8f18cf13 1919
a22285a6 1920 trans = btrfs_start_transaction(root, 0);
98d5dc13 1921 BUG_ON(IS_ERR(trans));
8f18cf13 1922
7d9eb12c
CM		 1923 lock_chunks(root);
1924
8f18cf13
CM		 1925 /*
1926 * step two, delete the device extents and the
1927 * chunk tree entries
1928 */
890871be 1929 read_lock(&em_tree->lock);
8f18cf13 1930 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 1931 read_unlock(&em_tree->lock);
8f18cf13 1932
a061fc8d
CM		 1933 BUG_ON(em->start > chunk_offset ||
1934 em->start + em->len < chunk_offset);
8f18cf13
CM		 1935 map = (struct map_lookup *)em->bdev;
1936
1937 for (i = 0; i < map->num_stripes; i++) {
1938 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1939 map->stripes[i].physical);
1940 BUG_ON(ret);
a061fc8d 1941
dfe25020
CM		 1942 if (map->stripes[i].dev) {
1943 ret = btrfs_update_device(trans, map->stripes[i].dev);
1944 BUG_ON(ret);
1945 }
8f18cf13
CM		 1946 }
1947 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1948 chunk_offset);
1949
1950 BUG_ON(ret);
1951
1abe9b8a 1952 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1953
8f18cf13
CM		 1954 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1955 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1956 BUG_ON(ret);
8f18cf13
CM		 1957 }
1958
2b82032c
YZ		 1959 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1960 BUG_ON(ret);
1961
890871be 1962 write_lock(&em_tree->lock);
2b82032c 1963 remove_extent_mapping(em_tree, em);
890871be 1964 write_unlock(&em_tree->lock);
2b82032c
YZ		 1965
1966 kfree(map);
1967 em->bdev = NULL;
1968
1969 /* once for the tree */
1970 free_extent_map(em);
1971 /* once for us */
1972 free_extent_map(em);
1973
1974 unlock_chunks(root);
1975 btrfs_end_transaction(trans, root);
1976 return 0;
1977}
1978
1979static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1980{
1981 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1982 struct btrfs_path *path;
1983 struct extent_buffer *leaf;
1984 struct btrfs_chunk *chunk;
1985 struct btrfs_key key;
1986 struct btrfs_key found_key;
1987 u64 chunk_tree = chunk_root->root_key.objectid;
1988 u64 chunk_type;
ba1bf481
JB		 1989 bool retried = false;
1990 int failed = 0;
2b82032c
YZ		 1991 int ret;
1992
1993 path = btrfs_alloc_path();
1994 if (!path)
1995 return -ENOMEM;
1996
ba1bf481 1997again:
2b82032c
YZ		 1998 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1999 key.offset = (u64)-1;
2000 key.type = BTRFS_CHUNK_ITEM_KEY;
2001
2002 while (1) {
2003 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2004 if (ret < 0)
2005 goto error;
2006 BUG_ON(ret == 0);
2007
2008 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2009 key.type);
2010 if (ret < 0)
2011 goto error;
2012 if (ret > 0)
2013 break;
1a40e23b 2014
2b82032c
YZ		 2015 leaf = path->nodes[0];
2016 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 2017
2b82032c
YZ		 2018 chunk = btrfs_item_ptr(leaf, path->slots[0],
2019 struct btrfs_chunk);
2020 chunk_type = btrfs_chunk_type(leaf, chunk);
b3b4aa74 2021 btrfs_release_path(path);
8f18cf13 2022
2b82032c
YZ		 2023 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2024 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2025 found_key.objectid,
2026 found_key.offset);
ba1bf481
JB		 2027 if (ret == -ENOSPC)
2028 failed++;
2029 else if (ret)
2030 BUG();
2b82032c 2031 }
8f18cf13 2032
2b82032c
YZ		 2033 if (found_key.offset == 0)
2034 break;
2035 key.offset = found_key.offset - 1;
2036 }
2037 ret = 0;
ba1bf481
JB		 2038 if (failed && !retried) {
2039 failed = 0;
2040 retried = true;
2041 goto again;
2042 } else if (failed && retried) {
2043 WARN_ON(1);
2044 ret = -ENOSPC;
2045 }
2b82032c
YZ		 2046error:
2047 btrfs_free_path(path);
2048 return ret;
8f18cf13
CM		 2049}
2050
ec44a35c
CM		 2051static u64 div_factor(u64 num, int factor)
2052{
2053 if (factor == 10)
2054 return num;
2055 num *= factor;
2056 do_div(num, 10);
2057 return num;
2058}
2059
ec44a35c
CM		 2060int btrfs_balance(struct btrfs_root *dev_root)
2061{
2062 int ret;
ec44a35c
CM		 2063 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2064 struct btrfs_device *device;
2065 u64 old_size;
2066 u64 size_to_free;
2067 struct btrfs_path *path;
2068 struct btrfs_key key;
ec44a35c
CM		 2069 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2070 struct btrfs_trans_handle *trans;
2071 struct btrfs_key found_key;
2072
2b82032c
YZ		 2073 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2074 return -EROFS;
ec44a35c 2075
6f88a440
BH		 2076 if (!capable(CAP_SYS_ADMIN))
2077 return -EPERM;
2078
7d9eb12c 2079 mutex_lock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 2080 dev_root = dev_root->fs_info->dev_root;
2081
ec44a35c 2082 /* step one make some room on all the devices */
c6e30871 2083 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 2084 old_size = device->total_bytes;
2085 size_to_free = div_factor(old_size, 1);
2086 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 2087 if (!device->writeable ||
2088 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 2089 continue;
2090
2091 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 2092 if (ret == -ENOSPC)
2093 break;
ec44a35c
CM		 2094 BUG_ON(ret);
2095
a22285a6 2096 trans = btrfs_start_transaction(dev_root, 0);
98d5dc13 2097 BUG_ON(IS_ERR(trans));
ec44a35c
CM		 2098
2099 ret = btrfs_grow_device(trans, device, old_size);
2100 BUG_ON(ret);
2101
2102 btrfs_end_transaction(trans, dev_root);
2103 }
2104
2105 /* step two, relocate all the chunks */
2106 path = btrfs_alloc_path();
17e9f796
MF		 2107 if (!path) {
2108 ret = -ENOMEM;
2109 goto error;
2110 }
ec44a35c
CM		 2111 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2112 key.offset = (u64)-1;
2113 key.type = BTRFS_CHUNK_ITEM_KEY;
2114
d397712b 2115 while (1) {
ec44a35c
CM		 2116 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2117 if (ret < 0)
2118 goto error;
2119
2120 /*
2121 * this shouldn't happen, it means the last relocate
2122 * failed
2123 */
2124 if (ret == 0)
2125 break;
2126
2127 ret = btrfs_previous_item(chunk_root, path, 0,
2128 BTRFS_CHUNK_ITEM_KEY);
7d9eb12c 2129 if (ret)
ec44a35c 2130 break;
7d9eb12c 2131
ec44a35c
CM		 2132 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2133 path->slots[0]);
2134 if (found_key.objectid != key.objectid)
2135 break;
7d9eb12c 2136
ec44a35c 2137 /* chunk zero is special */
ba1bf481 2138 if (found_key.offset == 0)
ec44a35c
CM		 2139 break;
2140
b3b4aa74 2141 btrfs_release_path(path);
ec44a35c
CM		 2142 ret = btrfs_relocate_chunk(chunk_root,
2143 chunk_root->root_key.objectid,
2144 found_key.objectid,
2145 found_key.offset);
508794eb
JB		 2146 if (ret && ret != -ENOSPC)
2147 goto error;
ba1bf481 2148 key.offset = found_key.offset - 1;
ec44a35c
CM		 2149 }
2150 ret = 0;
2151error:
2152 btrfs_free_path(path);
7d9eb12c 2153 mutex_unlock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 2154 return ret;
2155}
2156
8f18cf13
CM		 2157/*
2158 * shrinking a device means finding all of the device extents past
2159 * the new size, and then following the back refs to the chunks.
2160 * The chunk relocation code actually frees the device extent
2161 */
2162int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2163{
2164 struct btrfs_trans_handle *trans;
2165 struct btrfs_root *root = device->dev_root;
2166 struct btrfs_dev_extent *dev_extent = NULL;
2167 struct btrfs_path *path;
2168 u64 length;
2169 u64 chunk_tree;
2170 u64 chunk_objectid;
2171 u64 chunk_offset;
2172 int ret;
2173 int slot;
ba1bf481
JB		 2174 int failed = 0;
2175 bool retried = false;
8f18cf13
CM		 2176 struct extent_buffer *l;
2177 struct btrfs_key key;
2178 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2179 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 2180 u64 old_size = device->total_bytes;
8f18cf13
CM		 2181 u64 diff = device->total_bytes - new_size;
2182
2b82032c
YZ		 2183 if (new_size >= device->total_bytes)
2184 return -EINVAL;
8f18cf13
CM		 2185
2186 path = btrfs_alloc_path();
2187 if (!path)
2188 return -ENOMEM;
2189
8f18cf13
CM		 2190 path->reada = 2;
2191
7d9eb12c
CM		 2192 lock_chunks(root);
2193
8f18cf13 2194 device->total_bytes = new_size;
2b82032c
YZ		 2195 if (device->writeable)
2196 device->fs_devices->total_rw_bytes -= diff;
7d9eb12c 2197 unlock_chunks(root);
8f18cf13 2198
ba1bf481 2199again:
8f18cf13
CM		 2200 key.objectid = device->devid;
2201 key.offset = (u64)-1;
2202 key.type = BTRFS_DEV_EXTENT_KEY;
2203
2204 while (1) {
2205 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2206 if (ret < 0)
2207 goto done;
2208
2209 ret = btrfs_previous_item(root, path, 0, key.type);
2210 if (ret < 0)
2211 goto done;
2212 if (ret) {
2213 ret = 0;
b3b4aa74 2214 btrfs_release_path(path);
bf1fb512 2215 break;
8f18cf13
CM		 2216 }
2217
2218 l = path->nodes[0];
2219 slot = path->slots[0];
2220 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2221
ba1bf481 2222 if (key.objectid != device->devid) {
b3b4aa74 2223 btrfs_release_path(path);
bf1fb512 2224 break;
ba1bf481 2225 }
8f18cf13
CM		 2226
2227 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2228 length = btrfs_dev_extent_length(l, dev_extent);
2229
ba1bf481 2230 if (key.offset + length <= new_size) {
b3b4aa74 2231 btrfs_release_path(path);
d6397bae 2232 break;
ba1bf481 2233 }
8f18cf13
CM		 2234
2235 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2236 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2237 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
b3b4aa74 2238 btrfs_release_path(path);
8f18cf13
CM		 2239
2240 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2241 chunk_offset);
ba1bf481 2242 if (ret && ret != -ENOSPC)
8f18cf13 2243 goto done;
ba1bf481
JB		 2244 if (ret == -ENOSPC)
2245 failed++;
2246 key.offset -= 1;
2247 }
2248
2249 if (failed && !retried) {
2250 failed = 0;
2251 retried = true;
2252 goto again;
2253 } else if (failed && retried) {
2254 ret = -ENOSPC;
2255 lock_chunks(root);
2256
2257 device->total_bytes = old_size;
2258 if (device->writeable)
2259 device->fs_devices->total_rw_bytes += diff;
2260 unlock_chunks(root);
2261 goto done;
8f18cf13
CM		 2262 }
2263
d6397bae 2264 /* Shrinking succeeded, else we would be at "done". */
a22285a6 2265 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 2266 if (IS_ERR(trans)) {
2267 ret = PTR_ERR(trans);
2268 goto done;
2269 }
2270
d6397bae
CB		 2271 lock_chunks(root);
2272
2273 device->disk_total_bytes = new_size;
2274 /* Now btrfs_update_device() will change the on-disk size. */
2275 ret = btrfs_update_device(trans, device);
2276 if (ret) {
2277 unlock_chunks(root);
2278 btrfs_end_transaction(trans, root);
2279 goto done;
2280 }
2281 WARN_ON(diff > old_total);
2282 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2283 unlock_chunks(root);
2284 btrfs_end_transaction(trans, root);
8f18cf13
CM		 2285done:
2286 btrfs_free_path(path);
2287 return ret;
2288}
2289
b2950863 2290static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2291 struct btrfs_root *root,
2292 struct btrfs_key *key,
2293 struct btrfs_chunk *chunk, int item_size)
2294{
2295 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2296 struct btrfs_disk_key disk_key;
2297 u32 array_size;
2298 u8 *ptr;
2299
2300 array_size = btrfs_super_sys_array_size(super_copy);
2301 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2302 return -EFBIG;
2303
2304 ptr = super_copy->sys_chunk_array + array_size;
2305 btrfs_cpu_key_to_disk(&disk_key, key);
2306 memcpy(ptr, &disk_key, sizeof(disk_key));
2307 ptr += sizeof(disk_key);
2308 memcpy(ptr, chunk, item_size);
2309 item_size += sizeof(disk_key);
2310 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2311 return 0;
2312}
2313
73c5de00
AJ		 2314/*
2315 * sort the devices in descending order by max_avail, total_avail
2316 */
2317static int btrfs_cmp_device_info(const void *a, const void *b)
9b3f68b9 2318{
73c5de00
AJ		 2319 const struct btrfs_device_info *di_a = a;
2320 const struct btrfs_device_info *di_b = b;
9b3f68b9 2321
73c5de00 2322 if (di_a->max_avail > di_b->max_avail)
b2117a39 2323 return -1;
73c5de00 2324 if (di_a->max_avail < di_b->max_avail)
b2117a39 2325 return 1;
73c5de00
AJ		 2326 if (di_a->total_avail > di_b->total_avail)
2327 return -1;
2328 if (di_a->total_avail < di_b->total_avail)
2329 return 1;
2330 return 0;
b2117a39 2331}
0b86a832 2332
73c5de00
AJ		 2333static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *extent_root,
2335 struct map_lookup **map_ret,
2336 u64 *num_bytes_out, u64 *stripe_size_out,
2337 u64 start, u64 type)
b2117a39 2338{
73c5de00
AJ		 2339 struct btrfs_fs_info *info = extent_root->fs_info;
2340 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2341 struct list_head *cur;
2342 struct map_lookup *map = NULL;
2343 struct extent_map_tree *em_tree;
2344 struct extent_map *em;
2345 struct btrfs_device_info *devices_info = NULL;
2346 u64 total_avail;
2347 int num_stripes; /* total number of stripes to allocate */
2348 int sub_stripes; /* sub_stripes info for map */
2349 int dev_stripes; /* stripes per dev */
2350 int devs_max; /* max devs to use */
2351 int devs_min; /* min devs needed */
2352 int devs_increment; /* ndevs has to be a multiple of this */
2353 int ncopies; /* how many copies to data has */
2354 int ret;
2355 u64 max_stripe_size;
2356 u64 max_chunk_size;
2357 u64 stripe_size;
2358 u64 num_bytes;
2359 int ndevs;
2360 int i;
2361 int j;
593060d7 2362
73c5de00
AJ		 2363 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2364 (type & BTRFS_BLOCK_GROUP_DUP)) {
2365 WARN_ON(1);
2366 type &= ~BTRFS_BLOCK_GROUP_DUP;
321aecc6 2367 }
9b3f68b9 2368
73c5de00
AJ		 2369 if (list_empty(&fs_devices->alloc_list))
2370 return -ENOSPC;
b2117a39 2371
73c5de00
AJ		 2372 sub_stripes = 1;
2373 dev_stripes = 1;
2374 devs_increment = 1;
2375 ncopies = 1;
2376 devs_max = 0; /* 0 == as many as possible */
2377 devs_min = 1;
2378
2379 /*
2380 * define the properties of each RAID type.
2381 * FIXME: move this to a global table and use it in all RAID
2382 * calculation code
2383 */
2384 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2385 dev_stripes = 2;
b2117a39 2386 ncopies = 2;
73c5de00
AJ		 2387 devs_max = 1;
2388 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2389 devs_min = 2;
2390 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2391 devs_increment = 2;
b2117a39 2392 ncopies = 2;
73c5de00
AJ		 2393 devs_max = 2;
2394 devs_min = 2;
2395 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2396 sub_stripes = 2;
2397 devs_increment = 2;
2398 ncopies = 2;
2399 devs_min = 4;
2400 } else {
2401 devs_max = 1;
2402 }
b2117a39 2403
9b3f68b9 2404 if (type & BTRFS_BLOCK_GROUP_DATA) {
73c5de00
AJ		 2405 max_stripe_size = 1024 * 1024 * 1024;
2406 max_chunk_size = 10 * max_stripe_size;
9b3f68b9 2407 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
73c5de00
AJ		 2408 max_stripe_size = 256 * 1024 * 1024;
2409 max_chunk_size = max_stripe_size;
a40a90a0 2410 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
73c5de00
AJ		 2411 max_stripe_size = 8 * 1024 * 1024;
2412 max_chunk_size = 2 * max_stripe_size;
2413 } else {
2414 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2415 type);
2416 BUG_ON(1);
9b3f68b9
CM		 2417 }
2418
2b82032c
YZ		 2419 /* we don't want a chunk larger than 10% of writeable space */
2420 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2421 max_chunk_size);
9b3f68b9 2422
73c5de00
AJ		 2423 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2424 GFP_NOFS);
2425 if (!devices_info)
2426 return -ENOMEM;
0cad8a11 2427
73c5de00 2428 cur = fs_devices->alloc_list.next;
9b3f68b9 2429
9f680ce0 2430 /*
73c5de00
AJ		 2431 * in the first pass through the devices list, we gather information
2432 * about the available holes on each device.
9f680ce0 2433 */
73c5de00
AJ		 2434 ndevs = 0;
2435 while (cur != &fs_devices->alloc_list) {
2436 struct btrfs_device *device;
2437 u64 max_avail;
2438 u64 dev_offset;
b2117a39 2439
73c5de00 2440 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
9f680ce0 2441
73c5de00 2442 cur = cur->next;
b2117a39 2443
73c5de00
AJ		 2444 if (!device->writeable) {
2445 printk(KERN_ERR
2446 "btrfs: read-only device in alloc_list\n");
2447 WARN_ON(1);
2448 continue;
2449 }
b2117a39 2450
73c5de00
AJ		 2451 if (!device->in_fs_metadata)
2452 continue;
b2117a39 2453
73c5de00
AJ		 2454 if (device->total_bytes > device->bytes_used)
2455 total_avail = device->total_bytes - device->bytes_used;
2456 else
2457 total_avail = 0;
38c01b96 2458
2459 /* If there is no space on this device, skip it. */
2460 if (total_avail == 0)
2461 continue;
b2117a39 2462
73c5de00
AJ		 2463 ret = find_free_dev_extent(trans, device,
2464 max_stripe_size * dev_stripes,
2465 &dev_offset, &max_avail);
2466 if (ret && ret != -ENOSPC)
2467 goto error;
b2117a39 2468
73c5de00
AJ		 2469 if (ret == 0)
2470 max_avail = max_stripe_size * dev_stripes;
b2117a39 2471
73c5de00
AJ		 2472 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2473 continue;
b2117a39 2474
73c5de00
AJ		 2475 devices_info[ndevs].dev_offset = dev_offset;
2476 devices_info[ndevs].max_avail = max_avail;
2477 devices_info[ndevs].total_avail = total_avail;
2478 devices_info[ndevs].dev = device;
2479 ++ndevs;
2480 }
b2117a39 2481
73c5de00
AJ		 2482 /*
2483 * now sort the devices by hole size / available space
2484 */
2485 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2486 btrfs_cmp_device_info, NULL);
b2117a39 2487
73c5de00
AJ		 2488 /* round down to number of usable stripes */
2489 ndevs -= ndevs % devs_increment;
b2117a39 2490
73c5de00
AJ		 2491 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2492 ret = -ENOSPC;
2493 goto error;
b2117a39 2494 }
9f680ce0 2495
73c5de00
AJ		 2496 if (devs_max && ndevs > devs_max)
2497 ndevs = devs_max;
2498 /*
2499 * the primary goal is to maximize the number of stripes, so use as many
2500 * devices as possible, even if the stripes are not maximum sized.
2501 */
2502 stripe_size = devices_info[ndevs-1].max_avail;
2503 num_stripes = ndevs * dev_stripes;
b2117a39 2504
73c5de00
AJ		 2505 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2506 stripe_size = max_chunk_size * ncopies;
2507 do_div(stripe_size, num_stripes);
b2117a39 2508 }
b2117a39 2509
73c5de00
AJ		 2510 do_div(stripe_size, dev_stripes);
2511 do_div(stripe_size, BTRFS_STRIPE_LEN);
2512 stripe_size *= BTRFS_STRIPE_LEN;
b2117a39
MX		 2513
2514 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2515 if (!map) {
2516 ret = -ENOMEM;
2517 goto error;
2518 }
2519 map->num_stripes = num_stripes;
9b3f68b9 2520
73c5de00
AJ		 2521 for (i = 0; i < ndevs; ++i) {
2522 for (j = 0; j < dev_stripes; ++j) {
2523 int s = i * dev_stripes + j;
2524 map->stripes[s].dev = devices_info[i].dev;
2525 map->stripes[s].physical = devices_info[i].dev_offset +
2526 j * stripe_size;
6324fbf3 2527 }
6324fbf3 2528 }
2b82032c 2529 map->sector_size = extent_root->sectorsize;
b2117a39
MX		 2530 map->stripe_len = BTRFS_STRIPE_LEN;
2531 map->io_align = BTRFS_STRIPE_LEN;
2532 map->io_width = BTRFS_STRIPE_LEN;
2b82032c 2533 map->type = type;
2b82032c 2534 map->sub_stripes = sub_stripes;
0b86a832 2535
2b82032c 2536 *map_ret = map;
73c5de00 2537 num_bytes = stripe_size * (num_stripes / ncopies);
0b86a832 2538
73c5de00
AJ		 2539 *stripe_size_out = stripe_size;
2540 *num_bytes_out = num_bytes;
0b86a832 2541
73c5de00 2542 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
1abe9b8a 2543
172ddd60 2544 em = alloc_extent_map();
2b82032c 2545 if (!em) {
b2117a39
MX		 2546 ret = -ENOMEM;
2547 goto error;
593060d7 2548 }
2b82032c
YZ		 2549 em->bdev = (struct block_device *)map;
2550 em->start = start;
73c5de00 2551 em->len = num_bytes;
2b82032c
YZ		 2552 em->block_start = 0;
2553 em->block_len = em->len;
593060d7 2554
2b82032c 2555 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 2556 write_lock(&em_tree->lock);
2b82032c 2557 ret = add_extent_mapping(em_tree, em);
890871be 2558 write_unlock(&em_tree->lock);
2b82032c
YZ		 2559 BUG_ON(ret);
2560 free_extent_map(em);
0b86a832 2561
2b82032c
YZ		 2562 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2563 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
73c5de00 2564 start, num_bytes);
2b82032c 2565 BUG_ON(ret);
611f0e00 2566
73c5de00
AJ		 2567 for (i = 0; i < map->num_stripes; ++i) {
2568 struct btrfs_device *device;
2569 u64 dev_offset;
2570
2571 device = map->stripes[i].dev;
2572 dev_offset = map->stripes[i].physical;
0b86a832
CM		 2573
2574 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2575 info->chunk_root->root_key.objectid,
2576 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
73c5de00 2577 start, dev_offset, stripe_size);
0b86a832 2578 BUG_ON(ret);
2b82032c
YZ		 2579 }
2580
b2117a39 2581 kfree(devices_info);
2b82032c 2582 return 0;
b2117a39
MX		 2583
2584error:
2585 kfree(map);
2586 kfree(devices_info);
2587 return ret;
2b82032c
YZ		 2588}
2589
2590static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2591 struct btrfs_root *extent_root,
2592 struct map_lookup *map, u64 chunk_offset,
2593 u64 chunk_size, u64 stripe_size)
2594{
2595 u64 dev_offset;
2596 struct btrfs_key key;
2597 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2598 struct btrfs_device *device;
2599 struct btrfs_chunk *chunk;
2600 struct btrfs_stripe *stripe;
2601 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2602 int index = 0;
2603 int ret;
2604
2605 chunk = kzalloc(item_size, GFP_NOFS);
2606 if (!chunk)
2607 return -ENOMEM;
2608
2609 index = 0;
2610 while (index < map->num_stripes) {
2611 device = map->stripes[index].dev;
2612 device->bytes_used += stripe_size;
0b86a832
CM		 2613 ret = btrfs_update_device(trans, device);
2614 BUG_ON(ret);
2b82032c
YZ		 2615 index++;
2616 }
2617
2618 index = 0;
2619 stripe = &chunk->stripe;
2620 while (index < map->num_stripes) {
2621 device = map->stripes[index].dev;
2622 dev_offset = map->stripes[index].physical;
0b86a832 2623
e17cade2
CM		 2624 btrfs_set_stack_stripe_devid(stripe, device->devid);
2625 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2626 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2627 stripe++;
0b86a832
CM		 2628 index++;
2629 }
2630
2b82032c 2631 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2632 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2633 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2634 btrfs_set_stack_chunk_type(chunk, map->type);
2635 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2636 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2637 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2638 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2639 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2640
2b82032c
YZ		 2641 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2642 key.type = BTRFS_CHUNK_ITEM_KEY;
2643 key.offset = chunk_offset;
0b86a832 2644
2b82032c
YZ		 2645 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2646 BUG_ON(ret);
0b86a832 2647
2b82032c
YZ		 2648 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2649 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2650 item_size);
8f18cf13
CM		 2651 BUG_ON(ret);
2652 }
1abe9b8a 2653
0b86a832 2654 kfree(chunk);
2b82032c
YZ		 2655 return 0;
2656}
0b86a832 2657
2b82032c
YZ		 2658/*
2659 * Chunk allocation falls into two parts. The first part does works
2660 * that make the new allocated chunk useable, but not do any operation
2661 * that modifies the chunk tree. The second part does the works that
2662 * require modifying the chunk tree. This division is important for the
2663 * bootstrap process of adding storage to a seed btrfs.
2664 */
2665int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *extent_root, u64 type)
2667{
2668 u64 chunk_offset;
2669 u64 chunk_size;
2670 u64 stripe_size;
2671 struct map_lookup *map;
2672 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2673 int ret;
2674
2675 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2676 &chunk_offset);
2677 if (ret)
2678 return ret;
2679
2680 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2681 &stripe_size, chunk_offset, type);
2682 if (ret)
2683 return ret;
2684
2685 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2686 chunk_size, stripe_size);
2687 BUG_ON(ret);
2688 return 0;
2689}
2690
d397712b 2691static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2692 struct btrfs_root *root,
2693 struct btrfs_device *device)
2694{
2695 u64 chunk_offset;
2696 u64 sys_chunk_offset;
2697 u64 chunk_size;
2698 u64 sys_chunk_size;
2699 u64 stripe_size;
2700 u64 sys_stripe_size;
2701 u64 alloc_profile;
2702 struct map_lookup *map;
2703 struct map_lookup *sys_map;
2704 struct btrfs_fs_info *fs_info = root->fs_info;
2705 struct btrfs_root *extent_root = fs_info->extent_root;
2706 int ret;
2707
2708 ret = find_next_chunk(fs_info->chunk_root,
2709 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
92b8e897
MF		 2710 if (ret)
2711 return ret;
2b82032c
YZ		 2712
2713 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2714 (fs_info->metadata_alloc_profile &
2715 fs_info->avail_metadata_alloc_bits);
2716 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2717
2718 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2719 &stripe_size, chunk_offset, alloc_profile);
2720 BUG_ON(ret);
2721
2722 sys_chunk_offset = chunk_offset + chunk_size;
2723
2724 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2725 (fs_info->system_alloc_profile &
2726 fs_info->avail_system_alloc_bits);
2727 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2728
2729 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2730 &sys_chunk_size, &sys_stripe_size,
2731 sys_chunk_offset, alloc_profile);
2732 BUG_ON(ret);
2733
2734 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2735 BUG_ON(ret);
2736
2737 /*
2738 * Modifying chunk tree needs allocating new blocks from both
2739 * system block group and metadata block group. So we only can
2740 * do operations require modifying the chunk tree after both
2741 * block groups were created.
2742 */
2743 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2744 chunk_size, stripe_size);
2745 BUG_ON(ret);
2746
2747 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2748 sys_chunk_offset, sys_chunk_size,
2749 sys_stripe_size);
b248a415 2750 BUG_ON(ret);
2b82032c
YZ		 2751 return 0;
2752}
2753
2754int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2755{
2756 struct extent_map *em;
2757 struct map_lookup *map;
2758 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2759 int readonly = 0;
2760 int i;
2761
890871be 2762 read_lock(&map_tree->map_tree.lock);
2b82032c 2763 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 2764 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 2765 if (!em)
2766 return 1;
2767
f48b9075
JB		 2768 if (btrfs_test_opt(root, DEGRADED)) {
2769 free_extent_map(em);
2770 return 0;
2771 }
2772
2b82032c
YZ		 2773 map = (struct map_lookup *)em->bdev;
2774 for (i = 0; i < map->num_stripes; i++) {
2775 if (!map->stripes[i].dev->writeable) {
2776 readonly = 1;
2777 break;
2778 }
2779 }
0b86a832 2780 free_extent_map(em);
2b82032c 2781 return readonly;
0b86a832
CM		 2782}
2783
2784void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2785{
a8067e02 2786 extent_map_tree_init(&tree->map_tree);
0b86a832
CM		 2787}
2788
2789void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2790{
2791 struct extent_map *em;
2792
d397712b 2793 while (1) {
890871be 2794 write_lock(&tree->map_tree.lock);
0b86a832
CM		 2795 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2796 if (em)
2797 remove_extent_mapping(&tree->map_tree, em);
890871be 2798 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 2799 if (!em)
2800 break;
2801 kfree(em->bdev);
2802 /* once for us */
2803 free_extent_map(em);
2804 /* once for the tree */
2805 free_extent_map(em);
2806 }
2807}
2808
f188591e
CM		 2809int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2810{
2811 struct extent_map *em;
2812 struct map_lookup *map;
2813 struct extent_map_tree *em_tree = &map_tree->map_tree;
2814 int ret;
2815
890871be 2816 read_lock(&em_tree->lock);
f188591e 2817 em = lookup_extent_mapping(em_tree, logical, len);
890871be 2818 read_unlock(&em_tree->lock);
f188591e
CM		 2819 BUG_ON(!em);
2820
2821 BUG_ON(em->start > logical || em->start + em->len < logical);
2822 map = (struct map_lookup *)em->bdev;
2823 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2824 ret = map->num_stripes;
321aecc6
CM		 2825 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2826 ret = map->sub_stripes;
f188591e
CM		 2827 else
2828 ret = 1;
2829 free_extent_map(em);
f188591e
CM		 2830 return ret;
2831}
2832
dfe25020
CM		 2833static int find_live_mirror(struct map_lookup *map, int first, int num,
2834 int optimal)
2835{
2836 int i;
2837 if (map->stripes[optimal].dev->bdev)
2838 return optimal;
2839 for (i = first; i < first + num; i++) {
2840 if (map->stripes[i].dev->bdev)
2841 return i;
2842 }
2843 /* we couldn't find one that doesn't fail. Just return something
2844 * and the io error handling code will clean up eventually
2845 */
2846 return optimal;
2847}
2848
f2d8d74d
CM		 2849static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2850 u64 logical, u64 *length,
a1d3c478 2851 struct btrfs_bio **bbio_ret,
7eaceacc 2852 int mirror_num)
0b86a832
CM		 2853{
2854 struct extent_map *em;
2855 struct map_lookup *map;
2856 struct extent_map_tree *em_tree = &map_tree->map_tree;
2857 u64 offset;
593060d7 2858 u64 stripe_offset;
fce3bb9a 2859 u64 stripe_end_offset;
593060d7 2860 u64 stripe_nr;
fce3bb9a
LD		 2861 u64 stripe_nr_orig;
2862 u64 stripe_nr_end;
cea9e445 2863 int stripes_allocated = 8;
321aecc6 2864 int stripes_required = 1;
593060d7 2865 int stripe_index;
cea9e445 2866 int i;
f2d8d74d 2867 int num_stripes;
a236aed1 2868 int max_errors = 0;
a1d3c478 2869 struct btrfs_bio *bbio = NULL;
0b86a832 2870
a1d3c478 2871 if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
cea9e445 2872 stripes_allocated = 1;
cea9e445 2873again:
a1d3c478
JS		 2874 if (bbio_ret) {
2875 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
cea9e445 2876 GFP_NOFS);
a1d3c478 2877 if (!bbio)
cea9e445 2878 return -ENOMEM;
a236aed1 2879
a1d3c478 2880 atomic_set(&bbio->error, 0);
cea9e445 2881 }
0b86a832 2882
890871be 2883 read_lock(&em_tree->lock);
0b86a832 2884 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 2885 read_unlock(&em_tree->lock);
f2d8d74d 2886
3b951516 2887 if (!em) {
d397712b
CM		 2888 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2889 (unsigned long long)logical,
2890 (unsigned long long)*length);
f2d8d74d 2891 BUG();
3b951516 2892 }
0b86a832
CM		 2893
2894 BUG_ON(em->start > logical || em->start + em->len < logical);
2895 map = (struct map_lookup *)em->bdev;
2896 offset = logical - em->start;
593060d7 2897
f188591e
CM		 2898 if (mirror_num > map->num_stripes)
2899 mirror_num = 0;
2900
a1d3c478 2901 /* if our btrfs_bio struct is too small, back off and try again */
7b6d91da 2902 if (rw & REQ_WRITE) {
321aecc6
CM		 2903 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2904 BTRFS_BLOCK_GROUP_DUP)) {
2905 stripes_required = map->num_stripes;
a236aed1 2906 max_errors = 1;
321aecc6
CM		 2907 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2908 stripes_required = map->sub_stripes;
a236aed1 2909 max_errors = 1;
321aecc6
CM		 2910 }
2911 }
fce3bb9a
LD		 2912 if (rw & REQ_DISCARD) {
2913 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2914 BTRFS_BLOCK_GROUP_RAID1 |
2915 BTRFS_BLOCK_GROUP_DUP |
2916 BTRFS_BLOCK_GROUP_RAID10)) {
2917 stripes_required = map->num_stripes;
2918 }
2919 }
a1d3c478 2920 if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
321aecc6 2921 stripes_allocated < stripes_required) {
cea9e445 2922 stripes_allocated = map->num_stripes;
cea9e445 2923 free_extent_map(em);
a1d3c478 2924 kfree(bbio);
cea9e445
CM		 2925 goto again;
2926 }
593060d7
CM		 2927 stripe_nr = offset;
2928 /*
2929 * stripe_nr counts the total number of stripes we have to stride
2930 * to get to this block
2931 */
2932 do_div(stripe_nr, map->stripe_len);
2933
2934 stripe_offset = stripe_nr * map->stripe_len;
2935 BUG_ON(offset < stripe_offset);
2936
2937 /* stripe_offset is the offset of this block in its stripe*/
2938 stripe_offset = offset - stripe_offset;
2939
fce3bb9a
LD		 2940 if (rw & REQ_DISCARD)
2941 *length = min_t(u64, em->len - offset, *length);
2942 else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2943 BTRFS_BLOCK_GROUP_RAID1 |
2944 BTRFS_BLOCK_GROUP_RAID10 |
2945 BTRFS_BLOCK_GROUP_DUP)) {
cea9e445
CM		 2946 /* we limit the length of each bio to what fits in a stripe */
2947 *length = min_t(u64, em->len - offset,
fce3bb9a 2948 map->stripe_len - stripe_offset);
cea9e445
CM		 2949 } else {
2950 *length = em->len - offset;
2951 }
f2d8d74d 2952
a1d3c478 2953 if (!bbio_ret)
cea9e445
CM		 2954 goto out;
2955
f2d8d74d 2956 num_stripes = 1;
cea9e445 2957 stripe_index = 0;
fce3bb9a
LD		 2958 stripe_nr_orig = stripe_nr;
2959 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2960 (~(map->stripe_len - 1));
2961 do_div(stripe_nr_end, map->stripe_len);
2962 stripe_end_offset = stripe_nr_end * map->stripe_len -
2963 (offset + *length);
2964 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2965 if (rw & REQ_DISCARD)
2966 num_stripes = min_t(u64, map->num_stripes,
2967 stripe_nr_end - stripe_nr_orig);
2968 stripe_index = do_div(stripe_nr, map->num_stripes);
2969 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
212a17ab 2970 if (rw & (REQ_WRITE | REQ_DISCARD))
f2d8d74d 2971 num_stripes = map->num_stripes;
2fff734f 2972 else if (mirror_num)
f188591e 2973 stripe_index = mirror_num - 1;
dfe25020
CM		 2974 else {
2975 stripe_index = find_live_mirror(map, 0,
2976 map->num_stripes,
2977 current->pid % map->num_stripes);
a1d3c478 2978 mirror_num = stripe_index + 1;
dfe25020 2979 }
2fff734f 2980
611f0e00 2981 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
a1d3c478 2982 if (rw & (REQ_WRITE | REQ_DISCARD)) {
f2d8d74d 2983 num_stripes = map->num_stripes;
a1d3c478 2984 } else if (mirror_num) {
f188591e 2985 stripe_index = mirror_num - 1;
a1d3c478
JS		 2986 } else {
2987 mirror_num = 1;
2988 }
2fff734f 2989
321aecc6
CM		 2990 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2991 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 2992
2993 stripe_index = do_div(stripe_nr, factor);
2994 stripe_index *= map->sub_stripes;
2995
7eaceacc 2996 if (rw & REQ_WRITE)
f2d8d74d 2997 num_stripes = map->sub_stripes;
fce3bb9a
LD		 2998 else if (rw & REQ_DISCARD)
2999 num_stripes = min_t(u64, map->sub_stripes *
3000 (stripe_nr_end - stripe_nr_orig),
3001 map->num_stripes);
321aecc6
CM		 3002 else if (mirror_num)
3003 stripe_index += mirror_num - 1;
dfe25020
CM		 3004 else {
3005 stripe_index = find_live_mirror(map, stripe_index,
3006 map->sub_stripes, stripe_index +
3007 current->pid % map->sub_stripes);
a1d3c478 3008 mirror_num = stripe_index + 1;
dfe25020 3009 }
8790d502
CM		 3010 } else {
3011 /*
3012 * after this do_div call, stripe_nr is the number of stripes
3013 * on this device we have to walk to find the data, and
3014 * stripe_index is the number of our device in the stripe array
3015 */
3016 stripe_index = do_div(stripe_nr, map->num_stripes);
a1d3c478 3017 mirror_num = stripe_index + 1;
8790d502 3018 }
593060d7 3019 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 3020
fce3bb9a
LD		 3021 if (rw & REQ_DISCARD) {
3022 for (i = 0; i < num_stripes; i++) {
a1d3c478 3023 bbio->stripes[i].physical =
f2d8d74d
CM		 3024 map->stripes[stripe_index].physical +
3025 stripe_offset + stripe_nr * map->stripe_len;
a1d3c478 3026 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
fce3bb9a
LD		 3027
3028 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3029 u64 stripes;
d9d04879 3030 u32 last_stripe = 0;
fce3bb9a
LD		 3031 int j;
3032
d9d04879
CM		 3033 div_u64_rem(stripe_nr_end - 1,
3034 map->num_stripes,
3035 &last_stripe);
3036
fce3bb9a 3037 for (j = 0; j < map->num_stripes; j++) {
d9d04879
CM		 3038 u32 test;
3039
3040 div_u64_rem(stripe_nr_end - 1 - j,
3041 map->num_stripes, &test);
3042 if (test == stripe_index)
fce3bb9a
LD		 3043 break;
3044 }
3045 stripes = stripe_nr_end - 1 - j;
3046 do_div(stripes, map->num_stripes);
a1d3c478 3047 bbio->stripes[i].length = map->stripe_len *
fce3bb9a
LD		 3048 (stripes - stripe_nr + 1);
3049
3050 if (i == 0) {
a1d3c478 3051 bbio->stripes[i].length -=
fce3bb9a
LD		 3052 stripe_offset;
3053 stripe_offset = 0;
3054 }
3055 if (stripe_index == last_stripe)
a1d3c478 3056 bbio->stripes[i].length -=
fce3bb9a
LD		 3057 stripe_end_offset;
3058 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3059 u64 stripes;
3060 int j;
3061 int factor = map->num_stripes /
3062 map->sub_stripes;
d9d04879
CM		 3063 u32 last_stripe = 0;
3064
3065 div_u64_rem(stripe_nr_end - 1,
3066 factor, &last_stripe);
fce3bb9a
LD		 3067 last_stripe *= map->sub_stripes;
3068
3069 for (j = 0; j < factor; j++) {
d9d04879
CM		 3070 u32 test;
3071
3072 div_u64_rem(stripe_nr_end - 1 - j,
3073 factor, &test);
3074
3075 if (test ==
fce3bb9a
LD		 3076 stripe_index / map->sub_stripes)
3077 break;
3078 }
3079 stripes = stripe_nr_end - 1 - j;
3080 do_div(stripes, factor);
a1d3c478 3081 bbio->stripes[i].length = map->stripe_len *
fce3bb9a
LD		 3082 (stripes - stripe_nr + 1);
3083
3084 if (i < map->sub_stripes) {
a1d3c478 3085 bbio->stripes[i].length -=
fce3bb9a
LD		 3086 stripe_offset;
3087 if (i == map->sub_stripes - 1)
3088 stripe_offset = 0;
3089 }
3090 if (stripe_index >= last_stripe &&
3091 stripe_index <= (last_stripe +
3092 map->sub_stripes - 1)) {
a1d3c478 3093 bbio->stripes[i].length -=
fce3bb9a
LD		 3094 stripe_end_offset;
3095 }
3096 } else
a1d3c478 3097 bbio->stripes[i].length = *length;
fce3bb9a
LD		 3098
3099 stripe_index++;
3100 if (stripe_index == map->num_stripes) {
3101 /* This could only happen for RAID0/10 */
3102 stripe_index = 0;
3103 stripe_nr++;
3104 }
3105 }
3106 } else {
3107 for (i = 0; i < num_stripes; i++) {
a1d3c478 3108 bbio->stripes[i].physical =
212a17ab
LT		 3109 map->stripes[stripe_index].physical +
3110 stripe_offset +
3111 stripe_nr * map->stripe_len;
a1d3c478 3112 bbio->stripes[i].dev =
212a17ab 3113 map->stripes[stripe_index].dev;
fce3bb9a 3114 stripe_index++;
f2d8d74d 3115 }
593060d7 3116 }
a1d3c478
JS		 3117 if (bbio_ret) {
3118 *bbio_ret = bbio;
3119 bbio->num_stripes = num_stripes;
3120 bbio->max_errors = max_errors;
3121 bbio->mirror_num = mirror_num;
f2d8d74d 3122 }
cea9e445 3123out:
0b86a832 3124 free_extent_map(em);
0b86a832
CM		 3125 return 0;
3126}
3127
f2d8d74d
CM		 3128int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3129 u64 logical, u64 *length,
a1d3c478 3130 struct btrfs_bio **bbio_ret, int mirror_num)
f2d8d74d 3131{
a1d3c478 3132 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
7eaceacc 3133 mirror_num);
f2d8d74d
CM		 3134}
3135
a512bbf8
YZ		 3136int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3137 u64 chunk_start, u64 physical, u64 devid,
3138 u64 **logical, int *naddrs, int *stripe_len)
3139{
3140 struct extent_map_tree *em_tree = &map_tree->map_tree;
3141 struct extent_map *em;
3142 struct map_lookup *map;
3143 u64 *buf;
3144 u64 bytenr;
3145 u64 length;
3146 u64 stripe_nr;
3147 int i, j, nr = 0;
3148
890871be 3149 read_lock(&em_tree->lock);
a512bbf8 3150 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 3151 read_unlock(&em_tree->lock);
a512bbf8
YZ		 3152
3153 BUG_ON(!em || em->start != chunk_start);
3154 map = (struct map_lookup *)em->bdev;
3155
3156 length = em->len;
3157 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3158 do_div(length, map->num_stripes / map->sub_stripes);
3159 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3160 do_div(length, map->num_stripes);
3161
3162 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3163 BUG_ON(!buf);
3164
3165 for (i = 0; i < map->num_stripes; i++) {
3166 if (devid && map->stripes[i].dev->devid != devid)
3167 continue;
3168 if (map->stripes[i].physical > physical ||
3169 map->stripes[i].physical + length <= physical)
3170 continue;
3171
3172 stripe_nr = physical - map->stripes[i].physical;
3173 do_div(stripe_nr, map->stripe_len);
3174
3175 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3176 stripe_nr = stripe_nr * map->num_stripes + i;
3177 do_div(stripe_nr, map->sub_stripes);
3178 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3179 stripe_nr = stripe_nr * map->num_stripes + i;
3180 }
3181 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 3182 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 3183 for (j = 0; j < nr; j++) {
3184 if (buf[j] == bytenr)
3185 break;
3186 }
934d375b
CM		 3187 if (j == nr) {
3188 WARN_ON(nr >= map->num_stripes);
a512bbf8 3189 buf[nr++] = bytenr;
934d375b 3190 }
a512bbf8
YZ		 3191 }
3192
a512bbf8
YZ		 3193 *logical = buf;
3194 *naddrs = nr;
3195 *stripe_len = map->stripe_len;
3196
3197 free_extent_map(em);
3198 return 0;
f2d8d74d
CM		 3199}
3200
a1d3c478 3201static void btrfs_end_bio(struct bio *bio, int err)
8790d502 3202{
a1d3c478 3203 struct btrfs_bio *bbio = bio->bi_private;
7d2b4daa 3204 int is_orig_bio = 0;
8790d502 3205
8790d502 3206 if (err)
a1d3c478 3207 atomic_inc(&bbio->error);
8790d502 3208
a1d3c478 3209 if (bio == bbio->orig_bio)
7d2b4daa
CM		 3210 is_orig_bio = 1;
3211
a1d3c478 3212 if (atomic_dec_and_test(&bbio->stripes_pending)) {
7d2b4daa
CM		 3213 if (!is_orig_bio) {
3214 bio_put(bio);
a1d3c478 3215 bio = bbio->orig_bio;
7d2b4daa 3216 }
a1d3c478
JS		 3217 bio->bi_private = bbio->private;
3218 bio->bi_end_io = bbio->end_io;
a236aed1
CM		 3219 /* only send an error to the higher layers if it is
3220 * beyond the tolerance of the multi-bio
3221 */
a1d3c478 3222 if (atomic_read(&bbio->error) > bbio->max_errors) {
a236aed1 3223 err = -EIO;
1259ab75
CM		 3224 } else if (err) {
3225 /*
3226 * this bio is actually up to date, we didn't
3227 * go over the max number of errors
3228 */
3229 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 3230 err = 0;
1259ab75 3231 }
a1d3c478 3232 kfree(bbio);
8790d502
CM		 3233
3234 bio_endio(bio, err);
7d2b4daa 3235 } else if (!is_orig_bio) {
8790d502
CM		 3236 bio_put(bio);
3237 }
8790d502
CM		 3238}
3239
8b712842
CM		 3240struct async_sched {
3241 struct bio *bio;
3242 int rw;
3243 struct btrfs_fs_info *info;
3244 struct btrfs_work work;
3245};
3246
3247/*
3248 * see run_scheduled_bios for a description of why bios are collected for
3249 * async submit.
3250 *
3251 * This will add one bio to the pending list for a device and make sure
3252 * the work struct is scheduled.
3253 */
d397712b 3254static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 3255 struct btrfs_device *device,
3256 int rw, struct bio *bio)
8b712842
CM		 3257{
3258 int should_queue = 1;
ffbd517d 3259 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 3260
3261 /* don't bother with additional async steps for reads, right now */
7b6d91da 3262 if (!(rw & REQ_WRITE)) {
492bb6de 3263 bio_get(bio);
8b712842 3264 submit_bio(rw, bio);
492bb6de 3265 bio_put(bio);
8b712842
CM		 3266 return 0;
3267 }
3268
3269 /*
0986fe9e 3270 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 3271 * higher layers. Otherwise, the async bio makes it appear we have
3272 * made progress against dirty pages when we've really just put it
3273 * on a queue for later
3274 */
0986fe9e 3275 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 3276 WARN_ON(bio->bi_next);
8b712842
CM		 3277 bio->bi_next = NULL;
3278 bio->bi_rw |= rw;
3279
3280 spin_lock(&device->io_lock);
7b6d91da 3281 if (bio->bi_rw & REQ_SYNC)
ffbd517d
CM		 3282 pending_bios = &device->pending_sync_bios;
3283 else
3284 pending_bios = &device->pending_bios;
8b712842 3285
ffbd517d
CM		 3286 if (pending_bios->tail)
3287 pending_bios->tail->bi_next = bio;
8b712842 3288
ffbd517d
CM		 3289 pending_bios->tail = bio;
3290 if (!pending_bios->head)
3291 pending_bios->head = bio;
8b712842
CM		 3292 if (device->running_pending)
3293 should_queue = 0;
3294
3295 spin_unlock(&device->io_lock);
3296
3297 if (should_queue)
1cc127b5
CM		 3298 btrfs_queue_worker(&root->fs_info->submit_workers,
3299 &device->work);
8b712842
CM		 3300 return 0;
3301}
3302
f188591e 3303int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 3304 int mirror_num, int async_submit)
0b86a832
CM		 3305{
3306 struct btrfs_mapping_tree *map_tree;
3307 struct btrfs_device *dev;
8790d502 3308 struct bio *first_bio = bio;
a62b9401 3309 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 3310 u64 length = 0;
3311 u64 map_length;
0b86a832 3312 int ret;
8790d502
CM		 3313 int dev_nr = 0;
3314 int total_devs = 1;
a1d3c478 3315 struct btrfs_bio *bbio = NULL;
0b86a832 3316
f2d8d74d 3317 length = bio->bi_size;
0b86a832
CM		 3318 map_tree = &root->fs_info->mapping_tree;
3319 map_length = length;
cea9e445 3320
a1d3c478 3321 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
f188591e 3322 mirror_num);
cea9e445
CM		 3323 BUG_ON(ret);
3324
a1d3c478 3325 total_devs = bbio->num_stripes;
cea9e445 3326 if (map_length < length) {
d397712b
CM		 3327 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3328 "len %llu\n", (unsigned long long)logical,
3329 (unsigned long long)length,
3330 (unsigned long long)map_length);
cea9e445
CM		 3331 BUG();
3332 }
a1d3c478
JS		 3333
3334 bbio->orig_bio = first_bio;
3335 bbio->private = first_bio->bi_private;
3336 bbio->end_io = first_bio->bi_end_io;
3337 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
cea9e445 3338
d397712b 3339 while (dev_nr < total_devs) {
a1d3c478
JS		 3340 if (dev_nr < total_devs - 1) {
3341 bio = bio_clone(first_bio, GFP_NOFS);
3342 BUG_ON(!bio);
3343 } else {
3344 bio = first_bio;
8790d502 3345 }
a1d3c478
JS		 3346 bio->bi_private = bbio;
3347 bio->bi_end_io = btrfs_end_bio;
3348 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3349 dev = bbio->stripes[dev_nr].dev;
18e503d6 3350 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
a1d3c478
JS		 3351 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3352 "(%s id %llu), size=%u\n", rw,
3353 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3354 dev->name, dev->devid, bio->bi_size);
dfe25020 3355 bio->bi_bdev = dev->bdev;
8b712842
CM		 3356 if (async_submit)
3357 schedule_bio(root, dev, rw, bio);
3358 else
3359 submit_bio(rw, bio);
dfe25020
CM		 3360 } else {
3361 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3362 bio->bi_sector = logical >> 9;
dfe25020 3363 bio_endio(bio, -EIO);
dfe25020 3364 }
8790d502
CM		 3365 dev_nr++;
3366 }
0b86a832
CM		 3367 return 0;
3368}
3369
a443755f 3370struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 3371 u8 *uuid, u8 *fsid)
0b86a832 3372{
2b82032c
YZ		 3373 struct btrfs_device *device;
3374 struct btrfs_fs_devices *cur_devices;
3375
3376 cur_devices = root->fs_info->fs_devices;
3377 while (cur_devices) {
3378 if (!fsid ||
3379 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3380 device = __find_device(&cur_devices->devices,
3381 devid, uuid);
3382 if (device)
3383 return device;
3384 }
3385 cur_devices = cur_devices->seed;
3386 }
3387 return NULL;
0b86a832
CM		 3388}
3389
dfe25020
CM		 3390static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3391 u64 devid, u8 *dev_uuid)
3392{
3393 struct btrfs_device *device;
3394 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3395
3396 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 3397 if (!device)
3398 return NULL;
dfe25020
CM		 3399 list_add(&device->dev_list,
3400 &fs_devices->devices);
dfe25020
CM		 3401 device->dev_root = root->fs_info->dev_root;
3402 device->devid = devid;
8b712842 3403 device->work.func = pending_bios_fn;
e4404d6e 3404 device->fs_devices = fs_devices;
cd02dca5 3405 device->missing = 1;
dfe25020 3406 fs_devices->num_devices++;
cd02dca5 3407 fs_devices->missing_devices++;
dfe25020 3408 spin_lock_init(&device->io_lock);
d20f7043 3409 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 3410 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3411 return device;
3412}
3413
0b86a832
CM		 3414static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3415 struct extent_buffer *leaf,
3416 struct btrfs_chunk *chunk)
3417{
3418 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3419 struct map_lookup *map;
3420 struct extent_map *em;
3421 u64 logical;
3422 u64 length;
3423 u64 devid;
a443755f 3424 u8 uuid[BTRFS_UUID_SIZE];
593060d7 3425 int num_stripes;
0b86a832 3426 int ret;
593060d7 3427 int i;
0b86a832 3428
e17cade2
CM		 3429 logical = key->offset;
3430 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3431
890871be 3432 read_lock(&map_tree->map_tree.lock);
0b86a832 3433 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 3434 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3435
3436 /* already mapped? */
3437 if (em && em->start <= logical && em->start + em->len > logical) {
3438 free_extent_map(em);
0b86a832
CM		 3439 return 0;
3440 } else if (em) {
3441 free_extent_map(em);
3442 }
0b86a832 3443
172ddd60 3444 em = alloc_extent_map();
0b86a832
CM		 3445 if (!em)
3446 return -ENOMEM;
593060d7
CM		 3447 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3448 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3449 if (!map) {
3450 free_extent_map(em);
3451 return -ENOMEM;
3452 }
3453
3454 em->bdev = (struct block_device *)map;
3455 em->start = logical;
3456 em->len = length;
3457 em->block_start = 0;
c8b97818 3458 em->block_len = em->len;
0b86a832 3459
593060d7
CM		 3460 map->num_stripes = num_stripes;
3461 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3462 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3463 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3464 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3465 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3466 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3467 for (i = 0; i < num_stripes; i++) {
3468 map->stripes[i].physical =
3469 btrfs_stripe_offset_nr(leaf, chunk, i);
3470 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3471 read_extent_buffer(leaf, uuid, (unsigned long)
3472 btrfs_stripe_dev_uuid_nr(chunk, i),
3473 BTRFS_UUID_SIZE);
2b82032c
YZ		 3474 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3475 NULL);
dfe25020 3476 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3477 kfree(map);
3478 free_extent_map(em);
3479 return -EIO;
3480 }
dfe25020
CM		 3481 if (!map->stripes[i].dev) {
3482 map->stripes[i].dev =
3483 add_missing_dev(root, devid, uuid);
3484 if (!map->stripes[i].dev) {
3485 kfree(map);
3486 free_extent_map(em);
3487 return -EIO;
3488 }
3489 }
3490 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3491 }
3492
890871be 3493 write_lock(&map_tree->map_tree.lock);
0b86a832 3494 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 3495 write_unlock(&map_tree->map_tree.lock);
b248a415 3496 BUG_ON(ret);
0b86a832
CM		 3497 free_extent_map(em);
3498
3499 return 0;
3500}
3501
3502static int fill_device_from_item(struct extent_buffer *leaf,
3503 struct btrfs_dev_item *dev_item,
3504 struct btrfs_device *device)
3505{
3506 unsigned long ptr;
0b86a832
CM		 3507
3508 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 3509 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3510 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 3511 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3512 device->type = btrfs_device_type(leaf, dev_item);
3513 device->io_align = btrfs_device_io_align(leaf, dev_item);
3514 device->io_width = btrfs_device_io_width(leaf, dev_item);
3515 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3516
3517 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3518 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3519
0b86a832
CM		 3520 return 0;
3521}
3522
2b82032c
YZ		 3523static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3524{
3525 struct btrfs_fs_devices *fs_devices;
3526 int ret;
3527
3528 mutex_lock(&uuid_mutex);
3529
3530 fs_devices = root->fs_info->fs_devices->seed;
3531 while (fs_devices) {
3532 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3533 ret = 0;
3534 goto out;
3535 }
3536 fs_devices = fs_devices->seed;
3537 }
3538
3539 fs_devices = find_fsid(fsid);
3540 if (!fs_devices) {
3541 ret = -ENOENT;
3542 goto out;
3543 }
e4404d6e
YZ		 3544
3545 fs_devices = clone_fs_devices(fs_devices);
3546 if (IS_ERR(fs_devices)) {
3547 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3548 goto out;
3549 }
3550
97288f2c 3551 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3552 root->fs_info->bdev_holder);
2b82032c
YZ		 3553 if (ret)
3554 goto out;
3555
3556 if (!fs_devices->seeding) {
3557 __btrfs_close_devices(fs_devices);
e4404d6e 3558 free_fs_devices(fs_devices);
2b82032c
YZ		 3559 ret = -EINVAL;
3560 goto out;
3561 }
3562
3563 fs_devices->seed = root->fs_info->fs_devices->seed;
3564 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3565out:
3566 mutex_unlock(&uuid_mutex);
3567 return ret;
3568}
3569
0d81ba5d 3570static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3571 struct extent_buffer *leaf,
3572 struct btrfs_dev_item *dev_item)
3573{
3574 struct btrfs_device *device;
3575 u64 devid;
3576 int ret;
2b82032c 3577 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3578 u8 dev_uuid[BTRFS_UUID_SIZE];
3579
0b86a832 3580 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3581 read_extent_buffer(leaf, dev_uuid,
3582 (unsigned long)btrfs_device_uuid(dev_item),
3583 BTRFS_UUID_SIZE);
2b82032c
YZ		 3584 read_extent_buffer(leaf, fs_uuid,
3585 (unsigned long)btrfs_device_fsid(dev_item),
3586 BTRFS_UUID_SIZE);
3587
3588 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3589 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3590 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3591 return ret;
2b82032c
YZ		 3592 }
3593
3594 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3595 if (!device || !device->bdev) {
e4404d6e 3596 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3597 return -EIO;
3598
3599 if (!device) {
d397712b
CM		 3600 printk(KERN_WARNING "warning devid %llu missing\n",
3601 (unsigned long long)devid);
2b82032c
YZ		 3602 device = add_missing_dev(root, devid, dev_uuid);
3603 if (!device)
3604 return -ENOMEM;
cd02dca5
CM		 3605 } else if (!device->missing) {
3606 /*
3607 * this happens when a device that was properly setup
3608 * in the device info lists suddenly goes bad.
3609 * device->bdev is NULL, and so we have to set
3610 * device->missing to one here
3611 */
3612 root->fs_info->fs_devices->missing_devices++;
3613 device->missing = 1;
2b82032c
YZ		 3614 }
3615 }
3616
3617 if (device->fs_devices != root->fs_info->fs_devices) {
3618 BUG_ON(device->writeable);
3619 if (device->generation !=
3620 btrfs_device_generation(leaf, dev_item))
3621 return -EINVAL;
6324fbf3 3622 }
0b86a832
CM		 3623
3624 fill_device_from_item(leaf, dev_item, device);
3625 device->dev_root = root->fs_info->dev_root;
dfe25020 3626 device->in_fs_metadata = 1;
2b82032c
YZ		 3627 if (device->writeable)
3628 device->fs_devices->total_rw_bytes += device->total_bytes;
0b86a832 3629 ret = 0;
0b86a832
CM		 3630 return ret;
3631}
3632
e4404d6e 3633int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832
CM		 3634{
3635 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
a061fc8d 3636 struct extent_buffer *sb;
0b86a832 3637 struct btrfs_disk_key *disk_key;
0b86a832 3638 struct btrfs_chunk *chunk;
84eed90f
CM		 3639 u8 *ptr;
3640 unsigned long sb_ptr;
3641 int ret = 0;
0b86a832
CM		 3642 u32 num_stripes;
3643 u32 array_size;
3644 u32 len = 0;
0b86a832 3645 u32 cur;
84eed90f 3646 struct btrfs_key key;
0b86a832 3647
e4404d6e 3648 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3649 BTRFS_SUPER_INFO_SIZE);
3650 if (!sb)
3651 return -ENOMEM;
3652 btrfs_set_buffer_uptodate(sb);
85d4e461 3653 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4008c04a 3654
a061fc8d 3655 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3656 array_size = btrfs_super_sys_array_size(super_copy);
3657
0b86a832
CM		 3658 ptr = super_copy->sys_chunk_array;
3659 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3660 cur = 0;
3661
3662 while (cur < array_size) {
3663 disk_key = (struct btrfs_disk_key *)ptr;
3664 btrfs_disk_key_to_cpu(&key, disk_key);
3665
a061fc8d 3666 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3667 sb_ptr += len;
3668 cur += len;
3669
0d81ba5d 3670 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3671 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3672 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3673 if (ret)
3674 break;
0b86a832
CM		 3675 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3676 len = btrfs_chunk_item_size(num_stripes);
3677 } else {
84eed90f
CM		 3678 ret = -EIO;
3679 break;
0b86a832
CM		 3680 }
3681 ptr += len;
3682 sb_ptr += len;
3683 cur += len;
3684 }
a061fc8d 3685 free_extent_buffer(sb);
84eed90f 3686 return ret;
0b86a832
CM		 3687}
3688
3689int btrfs_read_chunk_tree(struct btrfs_root *root)
3690{
3691 struct btrfs_path *path;
3692 struct extent_buffer *leaf;
3693 struct btrfs_key key;
3694 struct btrfs_key found_key;
3695 int ret;
3696 int slot;
3697
3698 root = root->fs_info->chunk_root;
3699
3700 path = btrfs_alloc_path();
3701 if (!path)
3702 return -ENOMEM;
3703
3704 /* first we search for all of the device items, and then we
3705 * read in all of the chunk items. This way we can create chunk
3706 * mappings that reference all of the devices that are afound
3707 */
3708 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3709 key.offset = 0;
3710 key.type = 0;
3711again:
3712 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 3713 if (ret < 0)
3714 goto error;
d397712b 3715 while (1) {
0b86a832
CM		 3716 leaf = path->nodes[0];
3717 slot = path->slots[0];
3718 if (slot >= btrfs_header_nritems(leaf)) {
3719 ret = btrfs_next_leaf(root, path);
3720 if (ret == 0)
3721 continue;
3722 if (ret < 0)
3723 goto error;
3724 break;
3725 }
3726 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3727 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3728 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3729 break;
3730 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3731 struct btrfs_dev_item *dev_item;
3732 dev_item = btrfs_item_ptr(leaf, slot,
3733 struct btrfs_dev_item);
0d81ba5d 3734 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3735 if (ret)
3736 goto error;
0b86a832
CM		 3737 }
3738 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3739 struct btrfs_chunk *chunk;
3740 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3741 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3742 if (ret)
3743 goto error;
0b86a832
CM		 3744 }
3745 path->slots[0]++;
3746 }
3747 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3748 key.objectid = 0;
b3b4aa74 3749 btrfs_release_path(path);
0b86a832
CM		 3750 goto again;
3751 }
0b86a832
CM		 3752 ret = 0;
3753error:
2b82032c 3754 btrfs_free_path(path);
0b86a832
CM		 3755 return ret;
3756}
Linux kernel - Deliverables
This page took 0.42181 seconds and 5 git commands to generate.