projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: devid filter
[GitHub/mt8127/android_kernel_alcatel_ttab.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 }
d85c8a6f
CM		 298 /* unplug every 64 requests just for good measure */
299 if (batch_run % 64 == 0) {
300 blk_finish_plug(&plug);
301 blk_start_plug(&plug);
302 sync_pending = 0;
303 }
8b712842 304 }
ffbd517d 305
51684082
CM		 306 cond_resched();
307 if (again)
308 goto loop;
309
310 spin_lock(&device->io_lock);
311 if (device->pending_bios.head || device->pending_sync_bios.head)
312 goto loop_lock;
313 spin_unlock(&device->io_lock);
314
8b712842 315done:
211588ad 316 blk_finish_plug(&plug);
8b712842
CM		 317 return 0;
318}
319
b2950863 320static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 321{
322 struct btrfs_device *device;
323
324 device = container_of(work, struct btrfs_device, work);
325 run_scheduled_bios(device);
326}
327
a1b32a59 328static noinline int device_list_add(const char *path,
8a4b83cc
CM		 329 struct btrfs_super_block *disk_super,
330 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
331{
332 struct btrfs_device *device;
333 struct btrfs_fs_devices *fs_devices;
334 u64 found_transid = btrfs_super_generation(disk_super);
3a0524dc 335 char *name;
8a4b83cc
CM		 336
337 fs_devices = find_fsid(disk_super->fsid);
338 if (!fs_devices) {
515dc322 339 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 340 if (!fs_devices)
341 return -ENOMEM;
342 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 343 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 344 list_add(&fs_devices->list, &fs_uuids);
345 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
346 fs_devices->latest_devid = devid;
347 fs_devices->latest_trans = found_transid;
e5e9a520 348 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 349 device = NULL;
350 } else {
a443755f
CM		 351 device = __find_device(&fs_devices->devices, devid,
352 disk_super->dev_item.uuid);
8a4b83cc
CM		 353 }
354 if (!device) {
2b82032c
YZ		 355 if (fs_devices->opened)
356 return -EBUSY;
357
8a4b83cc
CM		 358 device = kzalloc(sizeof(*device), GFP_NOFS);
359 if (!device) {
360 /* we can safely leave the fs_devices entry around */
361 return -ENOMEM;
362 }
363 device->devid = devid;
8b712842 364 device->work.func = pending_bios_fn;
a443755f
CM		 365 memcpy(device->uuid, disk_super->dev_item.uuid,
366 BTRFS_UUID_SIZE);
b248a415 367 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 368 device->name = kstrdup(path, GFP_NOFS);
369 if (!device->name) {
370 kfree(device);
371 return -ENOMEM;
372 }
2b82032c 373 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520 374
90519d66
AJ		 375 /* init readahead state */
376 spin_lock_init(&device->reada_lock);
377 device->reada_curr_zone = NULL;
378 atomic_set(&device->reada_in_flight, 0);
379 device->reada_next = 0;
380 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
381 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
382
e5e9a520 383 mutex_lock(&fs_devices->device_list_mutex);
1f78160c 384 list_add_rcu(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 385 mutex_unlock(&fs_devices->device_list_mutex);
386
2b82032c 387 device->fs_devices = fs_devices;
8a4b83cc 388 fs_devices->num_devices++;
cd02dca5 389 } else if (!device->name || strcmp(device->name, path)) {
3a0524dc
TH		 390 name = kstrdup(path, GFP_NOFS);
391 if (!name)
392 return -ENOMEM;
393 kfree(device->name);
394 device->name = name;
cd02dca5
CM		 395 if (device->missing) {
396 fs_devices->missing_devices--;
397 device->missing = 0;
398 }
8a4b83cc
CM		 399 }
400
401 if (found_transid > fs_devices->latest_trans) {
402 fs_devices->latest_devid = devid;
403 fs_devices->latest_trans = found_transid;
404 }
8a4b83cc
CM		 405 *fs_devices_ret = fs_devices;
406 return 0;
407}
408
e4404d6e
YZ		 409static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
410{
411 struct btrfs_fs_devices *fs_devices;
412 struct btrfs_device *device;
413 struct btrfs_device *orig_dev;
414
415 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
416 if (!fs_devices)
417 return ERR_PTR(-ENOMEM);
418
419 INIT_LIST_HEAD(&fs_devices->devices);
420 INIT_LIST_HEAD(&fs_devices->alloc_list);
421 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 422 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 423 fs_devices->latest_devid = orig->latest_devid;
424 fs_devices->latest_trans = orig->latest_trans;
425 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
426
46224705 427 /* We have held the volume lock, it is safe to get the devices. */
e4404d6e
YZ		 428 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
429 device = kzalloc(sizeof(*device), GFP_NOFS);
430 if (!device)
431 goto error;
432
433 device->name = kstrdup(orig_dev->name, GFP_NOFS);
fd2696f3
JL		 434 if (!device->name) {
435 kfree(device);
e4404d6e 436 goto error;
fd2696f3 437 }
e4404d6e
YZ		 438
439 device->devid = orig_dev->devid;
440 device->work.func = pending_bios_fn;
441 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
e4404d6e
YZ		 442 spin_lock_init(&device->io_lock);
443 INIT_LIST_HEAD(&device->dev_list);
444 INIT_LIST_HEAD(&device->dev_alloc_list);
445
446 list_add(&device->dev_list, &fs_devices->devices);
447 device->fs_devices = fs_devices;
448 fs_devices->num_devices++;
449 }
450 return fs_devices;
451error:
452 free_fs_devices(fs_devices);
453 return ERR_PTR(-ENOMEM);
454}
455
dfe25020
CM		 456int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
457{
c6e30871 458 struct btrfs_device *device, *next;
dfe25020
CM		 459
460 mutex_lock(&uuid_mutex);
461again:
46224705 462 /* This is the initialized path, it is safe to release the devices. */
c6e30871 463 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 464 if (device->in_fs_metadata)
465 continue;
466
467 if (device->bdev) {
d4d77629 468 blkdev_put(device->bdev, device->mode);
2b82032c
YZ		 469 device->bdev = NULL;
470 fs_devices->open_devices--;
471 }
472 if (device->writeable) {
473 list_del_init(&device->dev_alloc_list);
474 device->writeable = 0;
475 fs_devices->rw_devices--;
476 }
e4404d6e
YZ		 477 list_del_init(&device->dev_list);
478 fs_devices->num_devices--;
479 kfree(device->name);
480 kfree(device);
dfe25020 481 }
2b82032c
YZ		 482
483 if (fs_devices->seed) {
484 fs_devices = fs_devices->seed;
2b82032c
YZ		 485 goto again;
486 }
487
dfe25020
CM		 488 mutex_unlock(&uuid_mutex);
489 return 0;
490}
a0af469b 491
1f78160c
XG		 492static void __free_device(struct work_struct *work)
493{
494 struct btrfs_device *device;
495
496 device = container_of(work, struct btrfs_device, rcu_work);
497
498 if (device->bdev)
499 blkdev_put(device->bdev, device->mode);
500
501 kfree(device->name);
502 kfree(device);
503}
504
505static void free_device(struct rcu_head *head)
506{
507 struct btrfs_device *device;
508
509 device = container_of(head, struct btrfs_device, rcu);
510
511 INIT_WORK(&device->rcu_work, __free_device);
512 schedule_work(&device->rcu_work);
513}
514
2b82032c 515static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 516{
8a4b83cc 517 struct btrfs_device *device;
e4404d6e 518
2b82032c
YZ		 519 if (--fs_devices->opened > 0)
520 return 0;
8a4b83cc 521
c9513edb 522 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 523 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1f78160c
XG		 524 struct btrfs_device *new_device;
525
526 if (device->bdev)
a0af469b 527 fs_devices->open_devices--;
1f78160c 528
2b82032c
YZ		 529 if (device->writeable) {
530 list_del_init(&device->dev_alloc_list);
531 fs_devices->rw_devices--;
532 }
533
d5e2003c
JB		 534 if (device->can_discard)
535 fs_devices->num_can_discard--;
536
1f78160c
XG		 537 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
538 BUG_ON(!new_device);
539 memcpy(new_device, device, sizeof(*new_device));
540 new_device->name = kstrdup(device->name, GFP_NOFS);
5f3f302a 541 BUG_ON(device->name && !new_device->name);
1f78160c
XG		 542 new_device->bdev = NULL;
543 new_device->writeable = 0;
544 new_device->in_fs_metadata = 0;
d5e2003c 545 new_device->can_discard = 0;
1f78160c
XG		 546 list_replace_rcu(&device->dev_list, &new_device->dev_list);
547
548 call_rcu(&device->rcu, free_device);
8a4b83cc 549 }
c9513edb
XG		 550 mutex_unlock(&fs_devices->device_list_mutex);
551
e4404d6e
YZ		 552 WARN_ON(fs_devices->open_devices);
553 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 554 fs_devices->opened = 0;
555 fs_devices->seeding = 0;
2b82032c 556
8a4b83cc
CM		 557 return 0;
558}
559
2b82032c
YZ		 560int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
561{
e4404d6e 562 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 563 int ret;
564
565 mutex_lock(&uuid_mutex);
566 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 567 if (!fs_devices->opened) {
568 seed_devices = fs_devices->seed;
569 fs_devices->seed = NULL;
570 }
2b82032c 571 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 572
573 while (seed_devices) {
574 fs_devices = seed_devices;
575 seed_devices = fs_devices->seed;
576 __btrfs_close_devices(fs_devices);
577 free_fs_devices(fs_devices);
578 }
2b82032c
YZ		 579 return ret;
580}
581
e4404d6e
YZ		 582static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
583 fmode_t flags, void *holder)
8a4b83cc 584{
d5e2003c 585 struct request_queue *q;
8a4b83cc
CM		 586 struct block_device *bdev;
587 struct list_head *head = &fs_devices->devices;
8a4b83cc 588 struct btrfs_device *device;
a0af469b
CM		 589 struct block_device *latest_bdev = NULL;
590 struct buffer_head *bh;
591 struct btrfs_super_block *disk_super;
592 u64 latest_devid = 0;
593 u64 latest_transid = 0;
a0af469b 594 u64 devid;
2b82032c 595 int seeding = 1;
a0af469b 596 int ret = 0;
8a4b83cc 597
d4d77629
TH		 598 flags |= FMODE_EXCL;
599
c6e30871 600 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 601 if (device->bdev)
602 continue;
dfe25020
CM		 603 if (!device->name)
604 continue;
605
d4d77629 606 bdev = blkdev_get_by_path(device->name, flags, holder);
8a4b83cc 607 if (IS_ERR(bdev)) {
d397712b 608 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 609 goto error;
8a4b83cc 610 }
a061fc8d 611 set_blocksize(bdev, 4096);
a0af469b 612
a512bbf8 613 bh = btrfs_read_dev_super(bdev);
20bcd649 614 if (!bh)
a0af469b
CM		 615 goto error_close;
616
617 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 618 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 619 if (devid != device->devid)
620 goto error_brelse;
621
2b82032c
YZ		 622 if (memcmp(device->uuid, disk_super->dev_item.uuid,
623 BTRFS_UUID_SIZE))
624 goto error_brelse;
625
626 device->generation = btrfs_super_generation(disk_super);
627 if (!latest_transid || device->generation > latest_transid) {
a0af469b 628 latest_devid = devid;
2b82032c 629 latest_transid = device->generation;
a0af469b
CM		 630 latest_bdev = bdev;
631 }
632
2b82032c
YZ		 633 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
634 device->writeable = 0;
635 } else {
636 device->writeable = !bdev_read_only(bdev);
637 seeding = 0;
638 }
639
d5e2003c
JB		 640 q = bdev_get_queue(bdev);
641 if (blk_queue_discard(q)) {
642 device->can_discard = 1;
643 fs_devices->num_can_discard++;
644 }
645
8a4b83cc 646 device->bdev = bdev;
dfe25020 647 device->in_fs_metadata = 0;
15916de8
CM		 648 device->mode = flags;
649
c289811c
CM		 650 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
651 fs_devices->rotating = 1;
652
a0af469b 653 fs_devices->open_devices++;
2b82032c
YZ		 654 if (device->writeable) {
655 fs_devices->rw_devices++;
656 list_add(&device->dev_alloc_list,
657 &fs_devices->alloc_list);
658 }
4f6c9328 659 brelse(bh);
a0af469b 660 continue;
a061fc8d 661
a0af469b
CM		 662error_brelse:
663 brelse(bh);
664error_close:
d4d77629 665 blkdev_put(bdev, flags);
a0af469b
CM		 666error:
667 continue;
8a4b83cc 668 }
a0af469b 669 if (fs_devices->open_devices == 0) {
20bcd649 670 ret = -EINVAL;
a0af469b
CM		 671 goto out;
672 }
2b82032c
YZ		 673 fs_devices->seeding = seeding;
674 fs_devices->opened = 1;
a0af469b
CM		 675 fs_devices->latest_bdev = latest_bdev;
676 fs_devices->latest_devid = latest_devid;
677 fs_devices->latest_trans = latest_transid;
2b82032c 678 fs_devices->total_rw_bytes = 0;
a0af469b 679out:
2b82032c
YZ		 680 return ret;
681}
682
683int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 684 fmode_t flags, void *holder)
2b82032c
YZ		 685{
686 int ret;
687
688 mutex_lock(&uuid_mutex);
689 if (fs_devices->opened) {
e4404d6e
YZ		 690 fs_devices->opened++;
691 ret = 0;
2b82032c 692 } else {
15916de8 693 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 694 }
8a4b83cc 695 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 696 return ret;
697}
698
97288f2c 699int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 700 struct btrfs_fs_devices **fs_devices_ret)
701{
702 struct btrfs_super_block *disk_super;
703 struct block_device *bdev;
704 struct buffer_head *bh;
705 int ret;
706 u64 devid;
f2984462 707 u64 transid;
8a4b83cc
CM		 708
709 mutex_lock(&uuid_mutex);
710
d4d77629
TH		 711 flags |= FMODE_EXCL;
712 bdev = blkdev_get_by_path(path, flags, holder);
8a4b83cc
CM		 713
714 if (IS_ERR(bdev)) {
8a4b83cc
CM		 715 ret = PTR_ERR(bdev);
716 goto error;
717 }
718
719 ret = set_blocksize(bdev, 4096);
720 if (ret)
721 goto error_close;
a512bbf8 722 bh = btrfs_read_dev_super(bdev);
8a4b83cc 723 if (!bh) {
20b45077 724 ret = -EINVAL;
8a4b83cc
CM		 725 goto error_close;
726 }
727 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 728 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 729 transid = btrfs_super_generation(disk_super);
7ae9c09d 730 if (disk_super->label[0])
d397712b 731 printk(KERN_INFO "device label %s ", disk_super->label);
22b63a29
ID		 732 else
733 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
119e10cf 734 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 735 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 736 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
737
8a4b83cc
CM		 738 brelse(bh);
739error_close:
d4d77629 740 blkdev_put(bdev, flags);
8a4b83cc
CM		 741error:
742 mutex_unlock(&uuid_mutex);
743 return ret;
744}
0b86a832 745
6d07bcec
MX		 746/* helper to account the used device space in the range */
747int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
748 u64 end, u64 *length)
749{
750 struct btrfs_key key;
751 struct btrfs_root *root = device->dev_root;
752 struct btrfs_dev_extent *dev_extent;
753 struct btrfs_path *path;
754 u64 extent_end;
755 int ret;
756 int slot;
757 struct extent_buffer *l;
758
759 *length = 0;
760
761 if (start >= device->total_bytes)
762 return 0;
763
764 path = btrfs_alloc_path();
765 if (!path)
766 return -ENOMEM;
767 path->reada = 2;
768
769 key.objectid = device->devid;
770 key.offset = start;
771 key.type = BTRFS_DEV_EXTENT_KEY;
772
773 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
774 if (ret < 0)
775 goto out;
776 if (ret > 0) {
777 ret = btrfs_previous_item(root, path, key.objectid, key.type);
778 if (ret < 0)
779 goto out;
780 }
781
782 while (1) {
783 l = path->nodes[0];
784 slot = path->slots[0];
785 if (slot >= btrfs_header_nritems(l)) {
786 ret = btrfs_next_leaf(root, path);
787 if (ret == 0)
788 continue;
789 if (ret < 0)
790 goto out;
791
792 break;
793 }
794 btrfs_item_key_to_cpu(l, &key, slot);
795
796 if (key.objectid < device->devid)
797 goto next;
798
799 if (key.objectid > device->devid)
800 break;
801
802 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
803 goto next;
804
805 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
806 extent_end = key.offset + btrfs_dev_extent_length(l,
807 dev_extent);
808 if (key.offset <= start && extent_end > end) {
809 *length = end - start + 1;
810 break;
811 } else if (key.offset <= start && extent_end > start)
812 *length += extent_end - start;
813 else if (key.offset > start && extent_end <= end)
814 *length += extent_end - key.offset;
815 else if (key.offset > start && key.offset <= end) {
816 *length += end - key.offset + 1;
817 break;
818 } else if (key.offset > end)
819 break;
820
821next:
822 path->slots[0]++;
823 }
824 ret = 0;
825out:
826 btrfs_free_path(path);
827 return ret;
828}
829
0b86a832 830/*
7bfc837d
MX		 831 * find_free_dev_extent - find free space in the specified device
832 * @trans: transaction handler
833 * @device: the device which we search the free space in
834 * @num_bytes: the size of the free space that we need
835 * @start: store the start of the free space.
836 * @len: the size of the free space. that we find, or the size of the max
837 * free space if we don't find suitable free space
838 *
0b86a832
CM		 839 * this uses a pretty simple search, the expectation is that it is
840 * called very infrequently and that a given device has a small number
841 * of extents
7bfc837d
MX		 842 *
843 * @start is used to store the start of the free space if we find. But if we
844 * don't find suitable free space, it will be used to store the start position
845 * of the max free space.
846 *
847 * @len is used to store the size of the free space that we find.
848 * But if we don't find suitable free space, it is used to store the size of
849 * the max free space.
0b86a832 850 */
ba1bf481
JB		 851int find_free_dev_extent(struct btrfs_trans_handle *trans,
852 struct btrfs_device *device, u64 num_bytes,
7bfc837d 853 u64 *start, u64 *len)
0b86a832
CM		 854{
855 struct btrfs_key key;
856 struct btrfs_root *root = device->dev_root;
7bfc837d 857 struct btrfs_dev_extent *dev_extent;
2b82032c 858 struct btrfs_path *path;
7bfc837d
MX		 859 u64 hole_size;
860 u64 max_hole_start;
861 u64 max_hole_size;
862 u64 extent_end;
863 u64 search_start;
0b86a832
CM		 864 u64 search_end = device->total_bytes;
865 int ret;
7bfc837d 866 int slot;
0b86a832
CM		 867 struct extent_buffer *l;
868
0b86a832
CM		 869 /* FIXME use last free of some kind */
870
8a4b83cc
CM		 871 /* we don't want to overwrite the superblock on the drive,
872 * so we make sure to start at an offset of at least 1MB
873 */
a9c9bf68 874 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
8f18cf13 875
7bfc837d
MX		 876 max_hole_start = search_start;
877 max_hole_size = 0;
38c01b96 878 hole_size = 0;
7bfc837d
MX		 879
880 if (search_start >= search_end) {
881 ret = -ENOSPC;
882 goto error;
883 }
884
885 path = btrfs_alloc_path();
886 if (!path) {
887 ret = -ENOMEM;
888 goto error;
889 }
890 path->reada = 2;
891
0b86a832
CM		 892 key.objectid = device->devid;
893 key.offset = search_start;
894 key.type = BTRFS_DEV_EXTENT_KEY;
7bfc837d 895
0b86a832
CM		 896 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
897 if (ret < 0)
7bfc837d 898 goto out;
1fcbac58
YZ		 899 if (ret > 0) {
900 ret = btrfs_previous_item(root, path, key.objectid, key.type);
901 if (ret < 0)
7bfc837d 902 goto out;
1fcbac58 903 }
7bfc837d 904
0b86a832
CM		 905 while (1) {
906 l = path->nodes[0];
907 slot = path->slots[0];
908 if (slot >= btrfs_header_nritems(l)) {
909 ret = btrfs_next_leaf(root, path);
910 if (ret == 0)
911 continue;
912 if (ret < 0)
7bfc837d
MX		 913 goto out;
914
915 break;
0b86a832
CM		 916 }
917 btrfs_item_key_to_cpu(l, &key, slot);
918
919 if (key.objectid < device->devid)
920 goto next;
921
922 if (key.objectid > device->devid)
7bfc837d 923 break;
0b86a832 924
7bfc837d
MX		 925 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
926 goto next;
9779b72f 927
7bfc837d
MX		 928 if (key.offset > search_start) {
929 hole_size = key.offset - search_start;
9779b72f 930
7bfc837d
MX		 931 if (hole_size > max_hole_size) {
932 max_hole_start = search_start;
933 max_hole_size = hole_size;
934 }
9779b72f 935
7bfc837d
MX		 936 /*
937 * If this free space is greater than which we need,
938 * it must be the max free space that we have found
939 * until now, so max_hole_start must point to the start
940 * of this free space and the length of this free space
941 * is stored in max_hole_size. Thus, we return
942 * max_hole_start and max_hole_size and go back to the
943 * caller.
944 */
945 if (hole_size >= num_bytes) {
946 ret = 0;
947 goto out;
0b86a832
CM		 948 }
949 }
0b86a832 950
0b86a832 951 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
7bfc837d
MX		 952 extent_end = key.offset + btrfs_dev_extent_length(l,
953 dev_extent);
954 if (extent_end > search_start)
955 search_start = extent_end;
0b86a832
CM		 956next:
957 path->slots[0]++;
958 cond_resched();
959 }
0b86a832 960
38c01b96 961 /*
962 * At this point, search_start should be the end of
963 * allocated dev extents, and when shrinking the device,
964 * search_end may be smaller than search_start.
965 */
966 if (search_end > search_start)
967 hole_size = search_end - search_start;
968
7bfc837d
MX		 969 if (hole_size > max_hole_size) {
970 max_hole_start = search_start;
971 max_hole_size = hole_size;
0b86a832 972 }
0b86a832 973
7bfc837d
MX		 974 /* See above. */
975 if (hole_size < num_bytes)
976 ret = -ENOSPC;
977 else
978 ret = 0;
979
980out:
2b82032c 981 btrfs_free_path(path);
7bfc837d
MX		 982error:
983 *start = max_hole_start;
b2117a39 984 if (len)
7bfc837d 985 *len = max_hole_size;
0b86a832
CM		 986 return ret;
987}
988
b2950863 989static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 990 struct btrfs_device *device,
991 u64 start)
992{
993 int ret;
994 struct btrfs_path *path;
995 struct btrfs_root *root = device->dev_root;
996 struct btrfs_key key;
a061fc8d
CM		 997 struct btrfs_key found_key;
998 struct extent_buffer *leaf = NULL;
999 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 1000
1001 path = btrfs_alloc_path();
1002 if (!path)
1003 return -ENOMEM;
1004
1005 key.objectid = device->devid;
1006 key.offset = start;
1007 key.type = BTRFS_DEV_EXTENT_KEY;
924cd8fb 1008again:
8f18cf13 1009 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 1010 if (ret > 0) {
1011 ret = btrfs_previous_item(root, path, key.objectid,
1012 BTRFS_DEV_EXTENT_KEY);
b0b802d7
TI		 1013 if (ret)
1014 goto out;
a061fc8d
CM		 1015 leaf = path->nodes[0];
1016 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1017 extent = btrfs_item_ptr(leaf, path->slots[0],
1018 struct btrfs_dev_extent);
1019 BUG_ON(found_key.offset > start || found_key.offset +
1020 btrfs_dev_extent_length(leaf, extent) < start);
924cd8fb
MX		 1021 key = found_key;
1022 btrfs_release_path(path);
1023 goto again;
a061fc8d
CM		 1024 } else if (ret == 0) {
1025 leaf = path->nodes[0];
1026 extent = btrfs_item_ptr(leaf, path->slots[0],
1027 struct btrfs_dev_extent);
1028 }
8f18cf13
CM		 1029 BUG_ON(ret);
1030
2bf64758
JB		 1031 if (device->bytes_used > 0) {
1032 u64 len = btrfs_dev_extent_length(leaf, extent);
1033 device->bytes_used -= len;
1034 spin_lock(&root->fs_info->free_chunk_lock);
1035 root->fs_info->free_chunk_space += len;
1036 spin_unlock(&root->fs_info->free_chunk_lock);
1037 }
8f18cf13 1038 ret = btrfs_del_item(trans, root, path);
8f18cf13 1039
b0b802d7 1040out:
8f18cf13
CM		 1041 btrfs_free_path(path);
1042 return ret;
1043}
1044
2b82032c 1045int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 1046 struct btrfs_device *device,
e17cade2 1047 u64 chunk_tree, u64 chunk_objectid,
2b82032c 1048 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 1049{
1050 int ret;
1051 struct btrfs_path *path;
1052 struct btrfs_root *root = device->dev_root;
1053 struct btrfs_dev_extent *extent;
1054 struct extent_buffer *leaf;
1055 struct btrfs_key key;
1056
dfe25020 1057 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 1058 path = btrfs_alloc_path();
1059 if (!path)
1060 return -ENOMEM;
1061
0b86a832 1062 key.objectid = device->devid;
2b82032c 1063 key.offset = start;
0b86a832
CM		 1064 key.type = BTRFS_DEV_EXTENT_KEY;
1065 ret = btrfs_insert_empty_item(trans, root, path, &key,
1066 sizeof(*extent));
1067 BUG_ON(ret);
1068
1069 leaf = path->nodes[0];
1070 extent = btrfs_item_ptr(leaf, path->slots[0],
1071 struct btrfs_dev_extent);
e17cade2
CM		 1072 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1073 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1074 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1075
1076 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1077 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1078 BTRFS_UUID_SIZE);
1079
0b86a832
CM		 1080 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1081 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 1082 btrfs_free_path(path);
1083 return ret;
1084}
1085
a1b32a59
CM		 1086static noinline int find_next_chunk(struct btrfs_root *root,
1087 u64 objectid, u64 *offset)
0b86a832
CM		 1088{
1089 struct btrfs_path *path;
1090 int ret;
1091 struct btrfs_key key;
e17cade2 1092 struct btrfs_chunk *chunk;
0b86a832
CM		 1093 struct btrfs_key found_key;
1094
1095 path = btrfs_alloc_path();
92b8e897
MF		 1096 if (!path)
1097 return -ENOMEM;
0b86a832 1098
e17cade2 1099 key.objectid = objectid;
0b86a832
CM		 1100 key.offset = (u64)-1;
1101 key.type = BTRFS_CHUNK_ITEM_KEY;
1102
1103 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1104 if (ret < 0)
1105 goto error;
1106
1107 BUG_ON(ret == 0);
1108
1109 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1110 if (ret) {
e17cade2 1111 *offset = 0;
0b86a832
CM		 1112 } else {
1113 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1114 path->slots[0]);
e17cade2
CM		 1115 if (found_key.objectid != objectid)
1116 *offset = 0;
1117 else {
1118 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1119 struct btrfs_chunk);
1120 *offset = found_key.offset +
1121 btrfs_chunk_length(path->nodes[0], chunk);
1122 }
0b86a832
CM		 1123 }
1124 ret = 0;
1125error:
1126 btrfs_free_path(path);
1127 return ret;
1128}
1129
2b82032c 1130static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 1131{
1132 int ret;
1133 struct btrfs_key key;
1134 struct btrfs_key found_key;
2b82032c
YZ		 1135 struct btrfs_path *path;
1136
1137 root = root->fs_info->chunk_root;
1138
1139 path = btrfs_alloc_path();
1140 if (!path)
1141 return -ENOMEM;
0b86a832
CM		 1142
1143 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1144 key.type = BTRFS_DEV_ITEM_KEY;
1145 key.offset = (u64)-1;
1146
1147 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1148 if (ret < 0)
1149 goto error;
1150
1151 BUG_ON(ret == 0);
1152
1153 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1154 BTRFS_DEV_ITEM_KEY);
1155 if (ret) {
1156 *objectid = 1;
1157 } else {
1158 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1159 path->slots[0]);
1160 *objectid = found_key.offset + 1;
1161 }
1162 ret = 0;
1163error:
2b82032c 1164 btrfs_free_path(path);
0b86a832
CM		 1165 return ret;
1166}
1167
1168/*
1169 * the device information is stored in the chunk root
1170 * the btrfs_device struct should be fully filled in
1171 */
1172int btrfs_add_device(struct btrfs_trans_handle *trans,
1173 struct btrfs_root *root,
1174 struct btrfs_device *device)
1175{
1176 int ret;
1177 struct btrfs_path *path;
1178 struct btrfs_dev_item *dev_item;
1179 struct extent_buffer *leaf;
1180 struct btrfs_key key;
1181 unsigned long ptr;
0b86a832
CM		 1182
1183 root = root->fs_info->chunk_root;
1184
1185 path = btrfs_alloc_path();
1186 if (!path)
1187 return -ENOMEM;
1188
0b86a832
CM		 1189 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1190 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1191 key.offset = device->devid;
0b86a832
CM		 1192
1193 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1194 sizeof(*dev_item));
0b86a832
CM		 1195 if (ret)
1196 goto out;
1197
1198 leaf = path->nodes[0];
1199 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1200
1201 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1202 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1203 btrfs_set_device_type(leaf, dev_item, device->type);
1204 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1205 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1206 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1207 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1208 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1209 btrfs_set_device_group(leaf, dev_item, 0);
1210 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1211 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1212 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1213
0b86a832 1214 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1215 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1216 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1217 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1218 btrfs_mark_buffer_dirty(leaf);
0b86a832 1219
2b82032c 1220 ret = 0;
0b86a832
CM		 1221out:
1222 btrfs_free_path(path);
1223 return ret;
1224}
8f18cf13 1225
a061fc8d
CM		 1226static int btrfs_rm_dev_item(struct btrfs_root *root,
1227 struct btrfs_device *device)
1228{
1229 int ret;
1230 struct btrfs_path *path;
a061fc8d 1231 struct btrfs_key key;
a061fc8d
CM		 1232 struct btrfs_trans_handle *trans;
1233
1234 root = root->fs_info->chunk_root;
1235
1236 path = btrfs_alloc_path();
1237 if (!path)
1238 return -ENOMEM;
1239
a22285a6 1240 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 1241 if (IS_ERR(trans)) {
1242 btrfs_free_path(path);
1243 return PTR_ERR(trans);
1244 }
a061fc8d
CM		 1245 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1246 key.type = BTRFS_DEV_ITEM_KEY;
1247 key.offset = device->devid;
7d9eb12c 1248 lock_chunks(root);
a061fc8d
CM		 1249
1250 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1251 if (ret < 0)
1252 goto out;
1253
1254 if (ret > 0) {
1255 ret = -ENOENT;
1256 goto out;
1257 }
1258
1259 ret = btrfs_del_item(trans, root, path);
1260 if (ret)
1261 goto out;
a061fc8d
CM		 1262out:
1263 btrfs_free_path(path);
7d9eb12c 1264 unlock_chunks(root);
a061fc8d
CM		 1265 btrfs_commit_transaction(trans, root);
1266 return ret;
1267}
1268
1269int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1270{
1271 struct btrfs_device *device;
2b82032c 1272 struct btrfs_device *next_device;
a061fc8d 1273 struct block_device *bdev;
dfe25020 1274 struct buffer_head *bh = NULL;
a061fc8d 1275 struct btrfs_super_block *disk_super;
1f78160c 1276 struct btrfs_fs_devices *cur_devices;
a061fc8d
CM		 1277 u64 all_avail;
1278 u64 devid;
2b82032c
YZ		 1279 u64 num_devices;
1280 u8 *dev_uuid;
a061fc8d 1281 int ret = 0;
1f78160c 1282 bool clear_super = false;
a061fc8d 1283
a061fc8d
CM		 1284 mutex_lock(&uuid_mutex);
1285
1286 all_avail = root->fs_info->avail_data_alloc_bits |
1287 root->fs_info->avail_system_alloc_bits |
1288 root->fs_info->avail_metadata_alloc_bits;
1289
1290 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
035fe03a 1291 root->fs_info->fs_devices->num_devices <= 4) {
d397712b
CM		 1292 printk(KERN_ERR "btrfs: unable to go below four devices "
1293 "on raid10\n");
a061fc8d
CM		 1294 ret = -EINVAL;
1295 goto out;
1296 }
1297
1298 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
035fe03a 1299 root->fs_info->fs_devices->num_devices <= 2) {
d397712b
CM		 1300 printk(KERN_ERR "btrfs: unable to go below two "
1301 "devices on raid1\n");
a061fc8d
CM		 1302 ret = -EINVAL;
1303 goto out;
1304 }
1305
dfe25020 1306 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1307 struct list_head *devices;
1308 struct btrfs_device *tmp;
a061fc8d 1309
dfe25020
CM		 1310 device = NULL;
1311 devices = &root->fs_info->fs_devices->devices;
46224705
XG		 1312 /*
1313 * It is safe to read the devices since the volume_mutex
1314 * is held.
1315 */
c6e30871 1316 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1317 if (tmp->in_fs_metadata && !tmp->bdev) {
1318 device = tmp;
1319 break;
1320 }
1321 }
1322 bdev = NULL;
1323 bh = NULL;
1324 disk_super = NULL;
1325 if (!device) {
d397712b
CM		 1326 printk(KERN_ERR "btrfs: no missing devices found to "
1327 "remove\n");
dfe25020
CM		 1328 goto out;
1329 }
dfe25020 1330 } else {
d4d77629
TH		 1331 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1332 root->fs_info->bdev_holder);
dfe25020
CM		 1333 if (IS_ERR(bdev)) {
1334 ret = PTR_ERR(bdev);
1335 goto out;
1336 }
a061fc8d 1337
2b82032c 1338 set_blocksize(bdev, 4096);
a512bbf8 1339 bh = btrfs_read_dev_super(bdev);
dfe25020 1340 if (!bh) {
20b45077 1341 ret = -EINVAL;
dfe25020
CM		 1342 goto error_close;
1343 }
1344 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 1345 devid = btrfs_stack_device_id(&disk_super->dev_item);
2b82032c
YZ		 1346 dev_uuid = disk_super->dev_item.uuid;
1347 device = btrfs_find_device(root, devid, dev_uuid,
1348 disk_super->fsid);
dfe25020
CM		 1349 if (!device) {
1350 ret = -ENOENT;
1351 goto error_brelse;
1352 }
2b82032c 1353 }
dfe25020 1354
2b82032c 1355 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1356 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1357 "device\n");
2b82032c
YZ		 1358 ret = -EINVAL;
1359 goto error_brelse;
1360 }
1361
1362 if (device->writeable) {
0c1daee0 1363 lock_chunks(root);
2b82032c 1364 list_del_init(&device->dev_alloc_list);
0c1daee0 1365 unlock_chunks(root);
2b82032c 1366 root->fs_info->fs_devices->rw_devices--;
1f78160c 1367 clear_super = true;
dfe25020 1368 }
a061fc8d
CM		 1369
1370 ret = btrfs_shrink_device(device, 0);
1371 if (ret)
9b3517e9 1372 goto error_undo;
a061fc8d 1373
a061fc8d
CM		 1374 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1375 if (ret)
9b3517e9 1376 goto error_undo;
a061fc8d 1377
2bf64758
JB		 1378 spin_lock(&root->fs_info->free_chunk_lock);
1379 root->fs_info->free_chunk_space = device->total_bytes -
1380 device->bytes_used;
1381 spin_unlock(&root->fs_info->free_chunk_lock);
1382
2b82032c 1383 device->in_fs_metadata = 0;
a2de733c 1384 btrfs_scrub_cancel_dev(root, device);
e5e9a520
CM		 1385
1386 /*
1387 * the device list mutex makes sure that we don't change
1388 * the device list while someone else is writing out all
1389 * the device supers.
1390 */
1f78160c
XG		 1391
1392 cur_devices = device->fs_devices;
e5e9a520 1393 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1394 list_del_rcu(&device->dev_list);
e5e9a520 1395
e4404d6e 1396 device->fs_devices->num_devices--;
2b82032c 1397
cd02dca5
CM		 1398 if (device->missing)
1399 root->fs_info->fs_devices->missing_devices--;
1400
2b82032c
YZ		 1401 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1402 struct btrfs_device, dev_list);
1403 if (device->bdev == root->fs_info->sb->s_bdev)
1404 root->fs_info->sb->s_bdev = next_device->bdev;
1405 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1406 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1407
1f78160c 1408 if (device->bdev)
e4404d6e 1409 device->fs_devices->open_devices--;
1f78160c
XG		 1410
1411 call_rcu(&device->rcu, free_device);
1412 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1413
6c41761f
DS		 1414 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1415 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
2b82032c 1416
1f78160c 1417 if (cur_devices->open_devices == 0) {
e4404d6e
YZ		 1418 struct btrfs_fs_devices *fs_devices;
1419 fs_devices = root->fs_info->fs_devices;
1420 while (fs_devices) {
1f78160c 1421 if (fs_devices->seed == cur_devices)
e4404d6e
YZ		 1422 break;
1423 fs_devices = fs_devices->seed;
2b82032c 1424 }
1f78160c
XG		 1425 fs_devices->seed = cur_devices->seed;
1426 cur_devices->seed = NULL;
0c1daee0 1427 lock_chunks(root);
1f78160c 1428 __btrfs_close_devices(cur_devices);
0c1daee0 1429 unlock_chunks(root);
1f78160c 1430 free_fs_devices(cur_devices);
2b82032c
YZ		 1431 }
1432
1433 /*
1434 * at this point, the device is zero sized. We want to
1435 * remove it from the devices list and zero out the old super
1436 */
1f78160c 1437 if (clear_super) {
dfe25020
CM		 1438 /* make sure this device isn't detected as part of
1439 * the FS anymore
1440 */
1441 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1442 set_buffer_dirty(bh);
1443 sync_dirty_buffer(bh);
dfe25020 1444 }
a061fc8d 1445
a061fc8d 1446 ret = 0;
a061fc8d
CM		 1447
1448error_brelse:
1449 brelse(bh);
1450error_close:
dfe25020 1451 if (bdev)
e525fd89 1452 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
a061fc8d
CM		 1453out:
1454 mutex_unlock(&uuid_mutex);
a061fc8d 1455 return ret;
9b3517e9
ID		 1456error_undo:
1457 if (device->writeable) {
0c1daee0 1458 lock_chunks(root);
9b3517e9
ID		 1459 list_add(&device->dev_alloc_list,
1460 &root->fs_info->fs_devices->alloc_list);
0c1daee0 1461 unlock_chunks(root);
9b3517e9
ID		 1462 root->fs_info->fs_devices->rw_devices++;
1463 }
1464 goto error_brelse;
a061fc8d
CM		 1465}
1466
2b82032c
YZ		 1467/*
1468 * does all the dirty work required for changing file system's UUID.
1469 */
1470static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1471 struct btrfs_root *root)
1472{
1473 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1474 struct btrfs_fs_devices *old_devices;
e4404d6e 1475 struct btrfs_fs_devices *seed_devices;
6c41761f 1476 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2b82032c
YZ		 1477 struct btrfs_device *device;
1478 u64 super_flags;
1479
1480 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1481 if (!fs_devices->seeding)
2b82032c
YZ		 1482 return -EINVAL;
1483
e4404d6e
YZ		 1484 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1485 if (!seed_devices)
2b82032c
YZ		 1486 return -ENOMEM;
1487
e4404d6e
YZ		 1488 old_devices = clone_fs_devices(fs_devices);
1489 if (IS_ERR(old_devices)) {
1490 kfree(seed_devices);
1491 return PTR_ERR(old_devices);
2b82032c 1492 }
e4404d6e 1493
2b82032c
YZ		 1494 list_add(&old_devices->list, &fs_uuids);
1495
e4404d6e
YZ		 1496 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1497 seed_devices->opened = 1;
1498 INIT_LIST_HEAD(&seed_devices->devices);
1499 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1500 mutex_init(&seed_devices->device_list_mutex);
c9513edb
XG		 1501
1502 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c
XG		 1503 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1504 synchronize_rcu);
c9513edb
XG		 1505 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1506
e4404d6e
YZ		 1507 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1508 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1509 device->fs_devices = seed_devices;
1510 }
1511
2b82032c
YZ		 1512 fs_devices->seeding = 0;
1513 fs_devices->num_devices = 0;
1514 fs_devices->open_devices = 0;
e4404d6e 1515 fs_devices->seed = seed_devices;
2b82032c
YZ		 1516
1517 generate_random_uuid(fs_devices->fsid);
1518 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1519 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1520 super_flags = btrfs_super_flags(disk_super) &
1521 ~BTRFS_SUPER_FLAG_SEEDING;
1522 btrfs_set_super_flags(disk_super, super_flags);
1523
1524 return 0;
1525}
1526
1527/*
1528 * strore the expected generation for seed devices in device items.
1529 */
1530static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1531 struct btrfs_root *root)
1532{
1533 struct btrfs_path *path;
1534 struct extent_buffer *leaf;
1535 struct btrfs_dev_item *dev_item;
1536 struct btrfs_device *device;
1537 struct btrfs_key key;
1538 u8 fs_uuid[BTRFS_UUID_SIZE];
1539 u8 dev_uuid[BTRFS_UUID_SIZE];
1540 u64 devid;
1541 int ret;
1542
1543 path = btrfs_alloc_path();
1544 if (!path)
1545 return -ENOMEM;
1546
1547 root = root->fs_info->chunk_root;
1548 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1549 key.offset = 0;
1550 key.type = BTRFS_DEV_ITEM_KEY;
1551
1552 while (1) {
1553 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1554 if (ret < 0)
1555 goto error;
1556
1557 leaf = path->nodes[0];
1558next_slot:
1559 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1560 ret = btrfs_next_leaf(root, path);
1561 if (ret > 0)
1562 break;
1563 if (ret < 0)
1564 goto error;
1565 leaf = path->nodes[0];
1566 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 1567 btrfs_release_path(path);
2b82032c
YZ		 1568 continue;
1569 }
1570
1571 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1572 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1573 key.type != BTRFS_DEV_ITEM_KEY)
1574 break;
1575
1576 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1577 struct btrfs_dev_item);
1578 devid = btrfs_device_id(leaf, dev_item);
1579 read_extent_buffer(leaf, dev_uuid,
1580 (unsigned long)btrfs_device_uuid(dev_item),
1581 BTRFS_UUID_SIZE);
1582 read_extent_buffer(leaf, fs_uuid,
1583 (unsigned long)btrfs_device_fsid(dev_item),
1584 BTRFS_UUID_SIZE);
1585 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1586 BUG_ON(!device);
1587
1588 if (device->fs_devices->seeding) {
1589 btrfs_set_device_generation(leaf, dev_item,
1590 device->generation);
1591 btrfs_mark_buffer_dirty(leaf);
1592 }
1593
1594 path->slots[0]++;
1595 goto next_slot;
1596 }
1597 ret = 0;
1598error:
1599 btrfs_free_path(path);
1600 return ret;
1601}
1602
788f20eb
CM		 1603int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1604{
d5e2003c 1605 struct request_queue *q;
788f20eb
CM		 1606 struct btrfs_trans_handle *trans;
1607 struct btrfs_device *device;
1608 struct block_device *bdev;
788f20eb 1609 struct list_head *devices;
2b82032c 1610 struct super_block *sb = root->fs_info->sb;
788f20eb 1611 u64 total_bytes;
2b82032c 1612 int seeding_dev = 0;
788f20eb
CM		 1613 int ret = 0;
1614
2b82032c
YZ		 1615 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1616 return -EINVAL;
788f20eb 1617
a5d16333 1618 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
d4d77629 1619 root->fs_info->bdev_holder);
7f59203a
JB		 1620 if (IS_ERR(bdev))
1621 return PTR_ERR(bdev);
a2135011 1622
2b82032c
YZ		 1623 if (root->fs_info->fs_devices->seeding) {
1624 seeding_dev = 1;
1625 down_write(&sb->s_umount);
1626 mutex_lock(&uuid_mutex);
1627 }
1628
8c8bee1d 1629 filemap_write_and_wait(bdev->bd_inode->i_mapping);
a2135011 1630
788f20eb 1631 devices = &root->fs_info->fs_devices->devices;
e5e9a520
CM		 1632 /*
1633 * we have the volume lock, so we don't need the extra
1634 * device list mutex while reading the list here.
1635 */
c6e30871 1636 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1637 if (device->bdev == bdev) {
1638 ret = -EEXIST;
2b82032c 1639 goto error;
788f20eb
CM		 1640 }
1641 }
1642
1643 device = kzalloc(sizeof(*device), GFP_NOFS);
1644 if (!device) {
1645 /* we can safely leave the fs_devices entry around */
1646 ret = -ENOMEM;
2b82032c 1647 goto error;
788f20eb
CM		 1648 }
1649
788f20eb
CM		 1650 device->name = kstrdup(device_path, GFP_NOFS);
1651 if (!device->name) {
1652 kfree(device);
2b82032c
YZ		 1653 ret = -ENOMEM;
1654 goto error;
788f20eb 1655 }
2b82032c
YZ		 1656
1657 ret = find_next_devid(root, &device->devid);
1658 if (ret) {
67100f25 1659 kfree(device->name);
2b82032c
YZ		 1660 kfree(device);
1661 goto error;
1662 }
1663
a22285a6 1664 trans = btrfs_start_transaction(root, 0);
98d5dc13 1665 if (IS_ERR(trans)) {
67100f25 1666 kfree(device->name);
98d5dc13
TI		 1667 kfree(device);
1668 ret = PTR_ERR(trans);
1669 goto error;
1670 }
1671
2b82032c
YZ		 1672 lock_chunks(root);
1673
d5e2003c
JB		 1674 q = bdev_get_queue(bdev);
1675 if (blk_queue_discard(q))
1676 device->can_discard = 1;
2b82032c
YZ		 1677 device->writeable = 1;
1678 device->work.func = pending_bios_fn;
1679 generate_random_uuid(device->uuid);
1680 spin_lock_init(&device->io_lock);
1681 device->generation = trans->transid;
788f20eb
CM		 1682 device->io_width = root->sectorsize;
1683 device->io_align = root->sectorsize;
1684 device->sector_size = root->sectorsize;
1685 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1686 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1687 device->dev_root = root->fs_info->dev_root;
1688 device->bdev = bdev;
dfe25020 1689 device->in_fs_metadata = 1;
fb01aa85 1690 device->mode = FMODE_EXCL;
2b82032c 1691 set_blocksize(device->bdev, 4096);
788f20eb 1692
2b82032c
YZ		 1693 if (seeding_dev) {
1694 sb->s_flags &= ~MS_RDONLY;
1695 ret = btrfs_prepare_sprout(trans, root);
1696 BUG_ON(ret);
1697 }
788f20eb 1698
2b82032c 1699 device->fs_devices = root->fs_info->fs_devices;
e5e9a520
CM		 1700
1701 /*
1702 * we don't want write_supers to jump in here with our device
1703 * half setup
1704 */
1705 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1706 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2b82032c
YZ		 1707 list_add(&device->dev_alloc_list,
1708 &root->fs_info->fs_devices->alloc_list);
1709 root->fs_info->fs_devices->num_devices++;
1710 root->fs_info->fs_devices->open_devices++;
1711 root->fs_info->fs_devices->rw_devices++;
d5e2003c
JB		 1712 if (device->can_discard)
1713 root->fs_info->fs_devices->num_can_discard++;
2b82032c 1714 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1715
2bf64758
JB		 1716 spin_lock(&root->fs_info->free_chunk_lock);
1717 root->fs_info->free_chunk_space += device->total_bytes;
1718 spin_unlock(&root->fs_info->free_chunk_lock);
1719
c289811c
CM		 1720 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1721 root->fs_info->fs_devices->rotating = 1;
1722
6c41761f
DS		 1723 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1724 btrfs_set_super_total_bytes(root->fs_info->super_copy,
788f20eb
CM		 1725 total_bytes + device->total_bytes);
1726
6c41761f
DS		 1727 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1728 btrfs_set_super_num_devices(root->fs_info->super_copy,
788f20eb 1729 total_bytes + 1);
e5e9a520 1730 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 1731
2b82032c
YZ		 1732 if (seeding_dev) {
1733 ret = init_first_rw_device(trans, root, device);
1734 BUG_ON(ret);
1735 ret = btrfs_finish_sprout(trans, root);
1736 BUG_ON(ret);
1737 } else {
1738 ret = btrfs_add_device(trans, root, device);
1739 }
1740
913d952e
CM		 1741 /*
1742 * we've got more storage, clear any full flags on the space
1743 * infos
1744 */
1745 btrfs_clear_space_info_full(root->fs_info);
1746
7d9eb12c 1747 unlock_chunks(root);
2b82032c 1748 btrfs_commit_transaction(trans, root);
a2135011 1749
2b82032c
YZ		 1750 if (seeding_dev) {
1751 mutex_unlock(&uuid_mutex);
1752 up_write(&sb->s_umount);
788f20eb 1753
2b82032c
YZ		 1754 ret = btrfs_relocate_sys_chunks(root);
1755 BUG_ON(ret);
1756 }
c9e9f97b 1757
2b82032c
YZ		 1758 return ret;
1759error:
e525fd89 1760 blkdev_put(bdev, FMODE_EXCL);
2b82032c
YZ		 1761 if (seeding_dev) {
1762 mutex_unlock(&uuid_mutex);
1763 up_write(&sb->s_umount);
1764 }
c9e9f97b 1765 return ret;
788f20eb
CM		 1766}
1767
d397712b
CM		 1768static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1769 struct btrfs_device *device)
0b86a832
CM		 1770{
1771 int ret;
1772 struct btrfs_path *path;
1773 struct btrfs_root *root;
1774 struct btrfs_dev_item *dev_item;
1775 struct extent_buffer *leaf;
1776 struct btrfs_key key;
1777
1778 root = device->dev_root->fs_info->chunk_root;
1779
1780 path = btrfs_alloc_path();
1781 if (!path)
1782 return -ENOMEM;
1783
1784 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1785 key.type = BTRFS_DEV_ITEM_KEY;
1786 key.offset = device->devid;
1787
1788 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1789 if (ret < 0)
1790 goto out;
1791
1792 if (ret > 0) {
1793 ret = -ENOENT;
1794 goto out;
1795 }
1796
1797 leaf = path->nodes[0];
1798 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1799
1800 btrfs_set_device_id(leaf, dev_item, device->devid);
1801 btrfs_set_device_type(leaf, dev_item, device->type);
1802 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1803 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1804 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 1805 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 1806 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1807 btrfs_mark_buffer_dirty(leaf);
1808
1809out:
1810 btrfs_free_path(path);
1811 return ret;
1812}
1813
7d9eb12c 1814static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1815 struct btrfs_device *device, u64 new_size)
1816{
1817 struct btrfs_super_block *super_copy =
6c41761f 1818 device->dev_root->fs_info->super_copy;
8f18cf13
CM		 1819 u64 old_total = btrfs_super_total_bytes(super_copy);
1820 u64 diff = new_size - device->total_bytes;
1821
2b82032c
YZ		 1822 if (!device->writeable)
1823 return -EACCES;
1824 if (new_size <= device->total_bytes)
1825 return -EINVAL;
1826
8f18cf13 1827 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1828 device->fs_devices->total_rw_bytes += diff;
1829
1830 device->total_bytes = new_size;
9779b72f 1831 device->disk_total_bytes = new_size;
4184ea7f
CM		 1832 btrfs_clear_space_info_full(device->dev_root->fs_info);
1833
8f18cf13
CM		 1834 return btrfs_update_device(trans, device);
1835}
1836
7d9eb12c
CM		 1837int btrfs_grow_device(struct btrfs_trans_handle *trans,
1838 struct btrfs_device *device, u64 new_size)
1839{
1840 int ret;
1841 lock_chunks(device->dev_root);
1842 ret = __btrfs_grow_device(trans, device, new_size);
1843 unlock_chunks(device->dev_root);
1844 return ret;
1845}
1846
8f18cf13
CM		 1847static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1848 struct btrfs_root *root,
1849 u64 chunk_tree, u64 chunk_objectid,
1850 u64 chunk_offset)
1851{
1852 int ret;
1853 struct btrfs_path *path;
1854 struct btrfs_key key;
1855
1856 root = root->fs_info->chunk_root;
1857 path = btrfs_alloc_path();
1858 if (!path)
1859 return -ENOMEM;
1860
1861 key.objectid = chunk_objectid;
1862 key.offset = chunk_offset;
1863 key.type = BTRFS_CHUNK_ITEM_KEY;
1864
1865 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1866 BUG_ON(ret);
1867
1868 ret = btrfs_del_item(trans, root, path);
8f18cf13
CM		 1869
1870 btrfs_free_path(path);
65a246c5 1871 return ret;
8f18cf13
CM		 1872}
1873
b2950863 1874static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1875 chunk_offset)
1876{
6c41761f 1877 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
8f18cf13
CM		 1878 struct btrfs_disk_key *disk_key;
1879 struct btrfs_chunk *chunk;
1880 u8 *ptr;
1881 int ret = 0;
1882 u32 num_stripes;
1883 u32 array_size;
1884 u32 len = 0;
1885 u32 cur;
1886 struct btrfs_key key;
1887
1888 array_size = btrfs_super_sys_array_size(super_copy);
1889
1890 ptr = super_copy->sys_chunk_array;
1891 cur = 0;
1892
1893 while (cur < array_size) {
1894 disk_key = (struct btrfs_disk_key *)ptr;
1895 btrfs_disk_key_to_cpu(&key, disk_key);
1896
1897 len = sizeof(*disk_key);
1898
1899 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1900 chunk = (struct btrfs_chunk *)(ptr + len);
1901 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1902 len += btrfs_chunk_item_size(num_stripes);
1903 } else {
1904 ret = -EIO;
1905 break;
1906 }
1907 if (key.objectid == chunk_objectid &&
1908 key.offset == chunk_offset) {
1909 memmove(ptr, ptr + len, array_size - (cur + len));
1910 array_size -= len;
1911 btrfs_set_super_sys_array_size(super_copy, array_size);
1912 } else {
1913 ptr += len;
1914 cur += len;
1915 }
1916 }
1917 return ret;
1918}
1919
b2950863 1920static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1921 u64 chunk_tree, u64 chunk_objectid,
1922 u64 chunk_offset)
1923{
1924 struct extent_map_tree *em_tree;
1925 struct btrfs_root *extent_root;
1926 struct btrfs_trans_handle *trans;
1927 struct extent_map *em;
1928 struct map_lookup *map;
1929 int ret;
1930 int i;
1931
1932 root = root->fs_info->chunk_root;
1933 extent_root = root->fs_info->extent_root;
1934 em_tree = &root->fs_info->mapping_tree.map_tree;
1935
ba1bf481
JB		 1936 ret = btrfs_can_relocate(extent_root, chunk_offset);
1937 if (ret)
1938 return -ENOSPC;
1939
8f18cf13 1940 /* step one, relocate all the extents inside this chunk */
1a40e23b 1941 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
a22285a6
YZ		 1942 if (ret)
1943 return ret;
8f18cf13 1944
a22285a6 1945 trans = btrfs_start_transaction(root, 0);
98d5dc13 1946 BUG_ON(IS_ERR(trans));
8f18cf13 1947
7d9eb12c
CM		 1948 lock_chunks(root);
1949
8f18cf13
CM		 1950 /*
1951 * step two, delete the device extents and the
1952 * chunk tree entries
1953 */
890871be 1954 read_lock(&em_tree->lock);
8f18cf13 1955 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 1956 read_unlock(&em_tree->lock);
8f18cf13 1957
a061fc8d
CM		 1958 BUG_ON(em->start > chunk_offset ||
1959 em->start + em->len < chunk_offset);
8f18cf13
CM		 1960 map = (struct map_lookup *)em->bdev;
1961
1962 for (i = 0; i < map->num_stripes; i++) {
1963 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1964 map->stripes[i].physical);
1965 BUG_ON(ret);
a061fc8d 1966
dfe25020
CM		 1967 if (map->stripes[i].dev) {
1968 ret = btrfs_update_device(trans, map->stripes[i].dev);
1969 BUG_ON(ret);
1970 }
8f18cf13
CM		 1971 }
1972 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1973 chunk_offset);
1974
1975 BUG_ON(ret);
1976
1abe9b8a 1977 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1978
8f18cf13
CM		 1979 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1980 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1981 BUG_ON(ret);
8f18cf13
CM		 1982 }
1983
2b82032c
YZ		 1984 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1985 BUG_ON(ret);
1986
890871be 1987 write_lock(&em_tree->lock);
2b82032c 1988 remove_extent_mapping(em_tree, em);
890871be 1989 write_unlock(&em_tree->lock);
2b82032c
YZ		 1990
1991 kfree(map);
1992 em->bdev = NULL;
1993
1994 /* once for the tree */
1995 free_extent_map(em);
1996 /* once for us */
1997 free_extent_map(em);
1998
1999 unlock_chunks(root);
2000 btrfs_end_transaction(trans, root);
2001 return 0;
2002}
2003
2004static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2005{
2006 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2007 struct btrfs_path *path;
2008 struct extent_buffer *leaf;
2009 struct btrfs_chunk *chunk;
2010 struct btrfs_key key;
2011 struct btrfs_key found_key;
2012 u64 chunk_tree = chunk_root->root_key.objectid;
2013 u64 chunk_type;
ba1bf481
JB		 2014 bool retried = false;
2015 int failed = 0;
2b82032c
YZ		 2016 int ret;
2017
2018 path = btrfs_alloc_path();
2019 if (!path)
2020 return -ENOMEM;
2021
ba1bf481 2022again:
2b82032c
YZ		 2023 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2024 key.offset = (u64)-1;
2025 key.type = BTRFS_CHUNK_ITEM_KEY;
2026
2027 while (1) {
2028 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2029 if (ret < 0)
2030 goto error;
2031 BUG_ON(ret == 0);
2032
2033 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2034 key.type);
2035 if (ret < 0)
2036 goto error;
2037 if (ret > 0)
2038 break;
1a40e23b 2039
2b82032c
YZ		 2040 leaf = path->nodes[0];
2041 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 2042
2b82032c
YZ		 2043 chunk = btrfs_item_ptr(leaf, path->slots[0],
2044 struct btrfs_chunk);
2045 chunk_type = btrfs_chunk_type(leaf, chunk);
b3b4aa74 2046 btrfs_release_path(path);
8f18cf13 2047
2b82032c
YZ		 2048 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2049 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2050 found_key.objectid,
2051 found_key.offset);
ba1bf481
JB		 2052 if (ret == -ENOSPC)
2053 failed++;
2054 else if (ret)
2055 BUG();
2b82032c 2056 }
8f18cf13 2057
2b82032c
YZ		 2058 if (found_key.offset == 0)
2059 break;
2060 key.offset = found_key.offset - 1;
2061 }
2062 ret = 0;
ba1bf481
JB		 2063 if (failed && !retried) {
2064 failed = 0;
2065 retried = true;
2066 goto again;
2067 } else if (failed && retried) {
2068 WARN_ON(1);
2069 ret = -ENOSPC;
2070 }
2b82032c
YZ		 2071error:
2072 btrfs_free_path(path);
2073 return ret;
8f18cf13
CM		 2074}
2075
c9e9f97b
ID		 2076/*
2077 * Should be called with both balance and volume mutexes held to
2078 * serialize other volume operations (add_dev/rm_dev/resize) with
2079 * restriper. Same goes for unset_balance_control.
2080 */
2081static void set_balance_control(struct btrfs_balance_control *bctl)
2082{
2083 struct btrfs_fs_info *fs_info = bctl->fs_info;
2084
2085 BUG_ON(fs_info->balance_ctl);
2086
2087 spin_lock(&fs_info->balance_lock);
2088 fs_info->balance_ctl = bctl;
2089 spin_unlock(&fs_info->balance_lock);
2090}
2091
2092static void unset_balance_control(struct btrfs_fs_info *fs_info)
2093{
2094 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2095
2096 BUG_ON(!fs_info->balance_ctl);
2097
2098 spin_lock(&fs_info->balance_lock);
2099 fs_info->balance_ctl = NULL;
2100 spin_unlock(&fs_info->balance_lock);
2101
2102 kfree(bctl);
2103}
2104
ed25e9b2
ID		 2105/*
2106 * Balance filters. Return 1 if chunk should be filtered out
2107 * (should not be balanced).
2108 */
2109static int chunk_profiles_filter(u64 chunk_profile,
2110 struct btrfs_balance_args *bargs)
2111{
2112 chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK;
2113
2114 if (chunk_profile == 0)
2115 chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2116
2117 if (bargs->profiles & chunk_profile)
2118 return 0;
2119
2120 return 1;
2121}
2122
5ce5b3c0
ID		 2123static u64 div_factor_fine(u64 num, int factor)
2124{
2125 if (factor <= 0)
2126 return 0;
2127 if (factor >= 100)
2128 return num;
2129
2130 num *= factor;
2131 do_div(num, 100);
2132 return num;
2133}
2134
2135static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2136 struct btrfs_balance_args *bargs)
2137{
2138 struct btrfs_block_group_cache *cache;
2139 u64 chunk_used, user_thresh;
2140 int ret = 1;
2141
2142 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2143 chunk_used = btrfs_block_group_used(&cache->item);
2144
2145 user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2146 if (chunk_used < user_thresh)
2147 ret = 0;
2148
2149 btrfs_put_block_group(cache);
2150 return ret;
2151}
2152
409d404b
ID		 2153static int chunk_devid_filter(struct extent_buffer *leaf,
2154 struct btrfs_chunk *chunk,
2155 struct btrfs_balance_args *bargs)
2156{
2157 struct btrfs_stripe *stripe;
2158 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2159 int i;
2160
2161 for (i = 0; i < num_stripes; i++) {
2162 stripe = btrfs_stripe_nr(chunk, i);
2163 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2164 return 0;
2165 }
2166
2167 return 1;
2168}
2169
f43ffb60
ID		 2170static int should_balance_chunk(struct btrfs_root *root,
2171 struct extent_buffer *leaf,
2172 struct btrfs_chunk *chunk, u64 chunk_offset)
2173{
2174 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2175 struct btrfs_balance_args *bargs = NULL;
2176 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2177
2178 /* type filter */
2179 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2180 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2181 return 0;
2182 }
2183
2184 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2185 bargs = &bctl->data;
2186 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2187 bargs = &bctl->sys;
2188 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2189 bargs = &bctl->meta;
2190
ed25e9b2
ID		 2191 /* profiles filter */
2192 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2193 chunk_profiles_filter(chunk_type, bargs)) {
2194 return 0;
5ce5b3c0
ID		 2195 }
2196
2197 /* usage filter */
2198 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2199 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2200 return 0;
409d404b
ID		 2201 }
2202
2203 /* devid filter */
2204 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2205 chunk_devid_filter(leaf, chunk, bargs)) {
2206 return 0;
ed25e9b2
ID		 2207 }
2208
f43ffb60
ID		 2209 return 1;
2210}
2211
ec44a35c
CM		 2212static u64 div_factor(u64 num, int factor)
2213{
2214 if (factor == 10)
2215 return num;
2216 num *= factor;
2217 do_div(num, 10);
2218 return num;
2219}
2220
c9e9f97b 2221static int __btrfs_balance(struct btrfs_fs_info *fs_info)
ec44a35c 2222{
c9e9f97b
ID		 2223 struct btrfs_root *chunk_root = fs_info->chunk_root;
2224 struct btrfs_root *dev_root = fs_info->dev_root;
2225 struct list_head *devices;
ec44a35c
CM		 2226 struct btrfs_device *device;
2227 u64 old_size;
2228 u64 size_to_free;
f43ffb60 2229 struct btrfs_chunk *chunk;
ec44a35c
CM		 2230 struct btrfs_path *path;
2231 struct btrfs_key key;
ec44a35c 2232 struct btrfs_key found_key;
c9e9f97b 2233 struct btrfs_trans_handle *trans;
f43ffb60
ID		 2234 struct extent_buffer *leaf;
2235 int slot;
c9e9f97b
ID		 2236 int ret;
2237 int enospc_errors = 0;
ec44a35c 2238
ec44a35c 2239 /* step one make some room on all the devices */
c9e9f97b 2240 devices = &fs_info->fs_devices->devices;
c6e30871 2241 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 2242 old_size = device->total_bytes;
2243 size_to_free = div_factor(old_size, 1);
2244 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 2245 if (!device->writeable ||
2246 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 2247 continue;
2248
2249 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 2250 if (ret == -ENOSPC)
2251 break;
ec44a35c
CM		 2252 BUG_ON(ret);
2253
a22285a6 2254 trans = btrfs_start_transaction(dev_root, 0);
98d5dc13 2255 BUG_ON(IS_ERR(trans));
ec44a35c
CM		 2256
2257 ret = btrfs_grow_device(trans, device, old_size);
2258 BUG_ON(ret);
2259
2260 btrfs_end_transaction(trans, dev_root);
2261 }
2262
2263 /* step two, relocate all the chunks */
2264 path = btrfs_alloc_path();
17e9f796
MF		 2265 if (!path) {
2266 ret = -ENOMEM;
2267 goto error;
2268 }
ec44a35c
CM		 2269 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2270 key.offset = (u64)-1;
2271 key.type = BTRFS_CHUNK_ITEM_KEY;
2272
d397712b 2273 while (1) {
ec44a35c
CM		 2274 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2275 if (ret < 0)
2276 goto error;
2277
2278 /*
2279 * this shouldn't happen, it means the last relocate
2280 * failed
2281 */
2282 if (ret == 0)
c9e9f97b 2283 BUG(); /* FIXME break ? */
ec44a35c
CM		 2284
2285 ret = btrfs_previous_item(chunk_root, path, 0,
2286 BTRFS_CHUNK_ITEM_KEY);
c9e9f97b
ID		 2287 if (ret) {
2288 ret = 0;
ec44a35c 2289 break;
c9e9f97b 2290 }
7d9eb12c 2291
f43ffb60
ID		 2292 leaf = path->nodes[0];
2293 slot = path->slots[0];
2294 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2295
ec44a35c
CM		 2296 if (found_key.objectid != key.objectid)
2297 break;
7d9eb12c 2298
ec44a35c 2299 /* chunk zero is special */
ba1bf481 2300 if (found_key.offset == 0)
ec44a35c
CM		 2301 break;
2302
f43ffb60
ID		 2303 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2304
2305 ret = should_balance_chunk(chunk_root, leaf, chunk,
2306 found_key.offset);
b3b4aa74 2307 btrfs_release_path(path);
f43ffb60
ID		 2308 if (!ret)
2309 goto loop;
2310
ec44a35c
CM		 2311 ret = btrfs_relocate_chunk(chunk_root,
2312 chunk_root->root_key.objectid,
2313 found_key.objectid,
2314 found_key.offset);
508794eb
JB		 2315 if (ret && ret != -ENOSPC)
2316 goto error;
c9e9f97b
ID		 2317 if (ret == -ENOSPC)
2318 enospc_errors++;
f43ffb60 2319loop:
ba1bf481 2320 key.offset = found_key.offset - 1;
ec44a35c 2321 }
c9e9f97b 2322
ec44a35c
CM		 2323error:
2324 btrfs_free_path(path);
c9e9f97b
ID		 2325 if (enospc_errors) {
2326 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2327 enospc_errors);
2328 if (!ret)
2329 ret = -ENOSPC;
2330 }
2331
2332 return ret;
2333}
2334
2335static void __cancel_balance(struct btrfs_fs_info *fs_info)
2336{
2337 unset_balance_control(fs_info);
2338}
2339
2340void update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
2341 struct btrfs_ioctl_balance_args *bargs);
2342
2343/*
2344 * Should be called with both balance and volume mutexes held
2345 */
2346int btrfs_balance(struct btrfs_balance_control *bctl,
2347 struct btrfs_ioctl_balance_args *bargs)
2348{
2349 struct btrfs_fs_info *fs_info = bctl->fs_info;
f43ffb60 2350 u64 allowed;
c9e9f97b
ID		 2351 int ret;
2352
2353 if (btrfs_fs_closing(fs_info)) {
2354 ret = -EINVAL;
2355 goto out;
2356 }
2357
f43ffb60
ID		 2358 /*
2359 * In case of mixed groups both data and meta should be picked,
2360 * and identical options should be given for both of them.
2361 */
2362 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2363 if ((allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2364 (bctl->flags & (BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA))) {
2365 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2366 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2367 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2368 printk(KERN_ERR "btrfs: with mixed groups data and "
2369 "metadata balance options must be the same\n");
2370 ret = -EINVAL;
2371 goto out;
2372 }
2373 }
2374
c9e9f97b
ID		 2375 set_balance_control(bctl);
2376
2377 mutex_unlock(&fs_info->balance_mutex);
2378
2379 ret = __btrfs_balance(fs_info);
2380
2381 mutex_lock(&fs_info->balance_mutex);
2382
2383 if (bargs) {
2384 memset(bargs, 0, sizeof(*bargs));
2385 update_ioctl_balance_args(fs_info, bargs);
2386 }
2387
2388 __cancel_balance(fs_info);
2389
2390 return ret;
2391out:
2392 kfree(bctl);
ec44a35c
CM		 2393 return ret;
2394}
2395
8f18cf13
CM		 2396/*
2397 * shrinking a device means finding all of the device extents past
2398 * the new size, and then following the back refs to the chunks.
2399 * The chunk relocation code actually frees the device extent
2400 */
2401int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2402{
2403 struct btrfs_trans_handle *trans;
2404 struct btrfs_root *root = device->dev_root;
2405 struct btrfs_dev_extent *dev_extent = NULL;
2406 struct btrfs_path *path;
2407 u64 length;
2408 u64 chunk_tree;
2409 u64 chunk_objectid;
2410 u64 chunk_offset;
2411 int ret;
2412 int slot;
ba1bf481
JB		 2413 int failed = 0;
2414 bool retried = false;
8f18cf13
CM		 2415 struct extent_buffer *l;
2416 struct btrfs_key key;
6c41761f 2417 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
8f18cf13 2418 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 2419 u64 old_size = device->total_bytes;
8f18cf13
CM		 2420 u64 diff = device->total_bytes - new_size;
2421
2b82032c
YZ		 2422 if (new_size >= device->total_bytes)
2423 return -EINVAL;
8f18cf13
CM		 2424
2425 path = btrfs_alloc_path();
2426 if (!path)
2427 return -ENOMEM;
2428
8f18cf13
CM		 2429 path->reada = 2;
2430
7d9eb12c
CM		 2431 lock_chunks(root);
2432
8f18cf13 2433 device->total_bytes = new_size;
2bf64758 2434 if (device->writeable) {
2b82032c 2435 device->fs_devices->total_rw_bytes -= diff;
2bf64758
JB		 2436 spin_lock(&root->fs_info->free_chunk_lock);
2437 root->fs_info->free_chunk_space -= diff;
2438 spin_unlock(&root->fs_info->free_chunk_lock);
2439 }
7d9eb12c 2440 unlock_chunks(root);
8f18cf13 2441
ba1bf481 2442again:
8f18cf13
CM		 2443 key.objectid = device->devid;
2444 key.offset = (u64)-1;
2445 key.type = BTRFS_DEV_EXTENT_KEY;
2446
2447 while (1) {
2448 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2449 if (ret < 0)
2450 goto done;
2451
2452 ret = btrfs_previous_item(root, path, 0, key.type);
2453 if (ret < 0)
2454 goto done;
2455 if (ret) {
2456 ret = 0;
b3b4aa74 2457 btrfs_release_path(path);
bf1fb512 2458 break;
8f18cf13
CM		 2459 }
2460
2461 l = path->nodes[0];
2462 slot = path->slots[0];
2463 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2464
ba1bf481 2465 if (key.objectid != device->devid) {
b3b4aa74 2466 btrfs_release_path(path);
bf1fb512 2467 break;
ba1bf481 2468 }
8f18cf13
CM		 2469
2470 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2471 length = btrfs_dev_extent_length(l, dev_extent);
2472
ba1bf481 2473 if (key.offset + length <= new_size) {
b3b4aa74 2474 btrfs_release_path(path);
d6397bae 2475 break;
ba1bf481 2476 }
8f18cf13
CM		 2477
2478 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2479 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2480 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
b3b4aa74 2481 btrfs_release_path(path);
8f18cf13
CM		 2482
2483 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2484 chunk_offset);
ba1bf481 2485 if (ret && ret != -ENOSPC)
8f18cf13 2486 goto done;
ba1bf481
JB		 2487 if (ret == -ENOSPC)
2488 failed++;
2489 key.offset -= 1;
2490 }
2491
2492 if (failed && !retried) {
2493 failed = 0;
2494 retried = true;
2495 goto again;
2496 } else if (failed && retried) {
2497 ret = -ENOSPC;
2498 lock_chunks(root);
2499
2500 device->total_bytes = old_size;
2501 if (device->writeable)
2502 device->fs_devices->total_rw_bytes += diff;
2bf64758
JB		 2503 spin_lock(&root->fs_info->free_chunk_lock);
2504 root->fs_info->free_chunk_space += diff;
2505 spin_unlock(&root->fs_info->free_chunk_lock);
ba1bf481
JB		 2506 unlock_chunks(root);
2507 goto done;
8f18cf13
CM		 2508 }
2509
d6397bae 2510 /* Shrinking succeeded, else we would be at "done". */
a22285a6 2511 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 2512 if (IS_ERR(trans)) {
2513 ret = PTR_ERR(trans);
2514 goto done;
2515 }
2516
d6397bae
CB		 2517 lock_chunks(root);
2518
2519 device->disk_total_bytes = new_size;
2520 /* Now btrfs_update_device() will change the on-disk size. */
2521 ret = btrfs_update_device(trans, device);
2522 if (ret) {
2523 unlock_chunks(root);
2524 btrfs_end_transaction(trans, root);
2525 goto done;
2526 }
2527 WARN_ON(diff > old_total);
2528 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2529 unlock_chunks(root);
2530 btrfs_end_transaction(trans, root);
8f18cf13
CM		 2531done:
2532 btrfs_free_path(path);
2533 return ret;
2534}
2535
b2950863 2536static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2537 struct btrfs_root *root,
2538 struct btrfs_key *key,
2539 struct btrfs_chunk *chunk, int item_size)
2540{
6c41761f 2541 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
0b86a832
CM		 2542 struct btrfs_disk_key disk_key;
2543 u32 array_size;
2544 u8 *ptr;
2545
2546 array_size = btrfs_super_sys_array_size(super_copy);
2547 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2548 return -EFBIG;
2549
2550 ptr = super_copy->sys_chunk_array + array_size;
2551 btrfs_cpu_key_to_disk(&disk_key, key);
2552 memcpy(ptr, &disk_key, sizeof(disk_key));
2553 ptr += sizeof(disk_key);
2554 memcpy(ptr, chunk, item_size);
2555 item_size += sizeof(disk_key);
2556 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2557 return 0;
2558}
2559
73c5de00
AJ		 2560/*
2561 * sort the devices in descending order by max_avail, total_avail
2562 */
2563static int btrfs_cmp_device_info(const void *a, const void *b)
9b3f68b9 2564{
73c5de00
AJ		 2565 const struct btrfs_device_info *di_a = a;
2566 const struct btrfs_device_info *di_b = b;
9b3f68b9 2567
73c5de00 2568 if (di_a->max_avail > di_b->max_avail)
b2117a39 2569 return -1;
73c5de00 2570 if (di_a->max_avail < di_b->max_avail)
b2117a39 2571 return 1;
73c5de00
AJ		 2572 if (di_a->total_avail > di_b->total_avail)
2573 return -1;
2574 if (di_a->total_avail < di_b->total_avail)
2575 return 1;
2576 return 0;
b2117a39 2577}
0b86a832 2578
73c5de00
AJ		 2579static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2580 struct btrfs_root *extent_root,
2581 struct map_lookup **map_ret,
2582 u64 *num_bytes_out, u64 *stripe_size_out,
2583 u64 start, u64 type)
b2117a39 2584{
73c5de00
AJ		 2585 struct btrfs_fs_info *info = extent_root->fs_info;
2586 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2587 struct list_head *cur;
2588 struct map_lookup *map = NULL;
2589 struct extent_map_tree *em_tree;
2590 struct extent_map *em;
2591 struct btrfs_device_info *devices_info = NULL;
2592 u64 total_avail;
2593 int num_stripes; /* total number of stripes to allocate */
2594 int sub_stripes; /* sub_stripes info for map */
2595 int dev_stripes; /* stripes per dev */
2596 int devs_max; /* max devs to use */
2597 int devs_min; /* min devs needed */
2598 int devs_increment; /* ndevs has to be a multiple of this */
2599 int ncopies; /* how many copies to data has */
2600 int ret;
2601 u64 max_stripe_size;
2602 u64 max_chunk_size;
2603 u64 stripe_size;
2604 u64 num_bytes;
2605 int ndevs;
2606 int i;
2607 int j;
593060d7 2608
73c5de00
AJ		 2609 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2610 (type & BTRFS_BLOCK_GROUP_DUP)) {
2611 WARN_ON(1);
2612 type &= ~BTRFS_BLOCK_GROUP_DUP;
321aecc6 2613 }
9b3f68b9 2614
73c5de00
AJ		 2615 if (list_empty(&fs_devices->alloc_list))
2616 return -ENOSPC;
b2117a39 2617
73c5de00
AJ		 2618 sub_stripes = 1;
2619 dev_stripes = 1;
2620 devs_increment = 1;
2621 ncopies = 1;
2622 devs_max = 0; /* 0 == as many as possible */
2623 devs_min = 1;
2624
2625 /*
2626 * define the properties of each RAID type.
2627 * FIXME: move this to a global table and use it in all RAID
2628 * calculation code
2629 */
2630 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2631 dev_stripes = 2;
b2117a39 2632 ncopies = 2;
73c5de00
AJ		 2633 devs_max = 1;
2634 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2635 devs_min = 2;
2636 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2637 devs_increment = 2;
b2117a39 2638 ncopies = 2;
73c5de00
AJ		 2639 devs_max = 2;
2640 devs_min = 2;
2641 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2642 sub_stripes = 2;
2643 devs_increment = 2;
2644 ncopies = 2;
2645 devs_min = 4;
2646 } else {
2647 devs_max = 1;
2648 }
b2117a39 2649
9b3f68b9 2650 if (type & BTRFS_BLOCK_GROUP_DATA) {
73c5de00
AJ		 2651 max_stripe_size = 1024 * 1024 * 1024;
2652 max_chunk_size = 10 * max_stripe_size;
9b3f68b9 2653 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
73c5de00
AJ		 2654 max_stripe_size = 256 * 1024 * 1024;
2655 max_chunk_size = max_stripe_size;
a40a90a0 2656 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
73c5de00
AJ		 2657 max_stripe_size = 8 * 1024 * 1024;
2658 max_chunk_size = 2 * max_stripe_size;
2659 } else {
2660 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2661 type);
2662 BUG_ON(1);
9b3f68b9
CM		 2663 }
2664
2b82032c
YZ		 2665 /* we don't want a chunk larger than 10% of writeable space */
2666 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2667 max_chunk_size);
9b3f68b9 2668
73c5de00
AJ		 2669 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2670 GFP_NOFS);
2671 if (!devices_info)
2672 return -ENOMEM;
0cad8a11 2673
73c5de00 2674 cur = fs_devices->alloc_list.next;
9b3f68b9 2675
9f680ce0 2676 /*
73c5de00
AJ		 2677 * in the first pass through the devices list, we gather information
2678 * about the available holes on each device.
9f680ce0 2679 */
73c5de00
AJ		 2680 ndevs = 0;
2681 while (cur != &fs_devices->alloc_list) {
2682 struct btrfs_device *device;
2683 u64 max_avail;
2684 u64 dev_offset;
b2117a39 2685
73c5de00 2686 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
9f680ce0 2687
73c5de00 2688 cur = cur->next;
b2117a39 2689
73c5de00
AJ		 2690 if (!device->writeable) {
2691 printk(KERN_ERR
2692 "btrfs: read-only device in alloc_list\n");
2693 WARN_ON(1);
2694 continue;
2695 }
b2117a39 2696
73c5de00
AJ		 2697 if (!device->in_fs_metadata)
2698 continue;
b2117a39 2699
73c5de00
AJ		 2700 if (device->total_bytes > device->bytes_used)
2701 total_avail = device->total_bytes - device->bytes_used;
2702 else
2703 total_avail = 0;
38c01b96 2704
2705 /* If there is no space on this device, skip it. */
2706 if (total_avail == 0)
2707 continue;
b2117a39 2708
73c5de00
AJ		 2709 ret = find_free_dev_extent(trans, device,
2710 max_stripe_size * dev_stripes,
2711 &dev_offset, &max_avail);
2712 if (ret && ret != -ENOSPC)
2713 goto error;
b2117a39 2714
73c5de00
AJ		 2715 if (ret == 0)
2716 max_avail = max_stripe_size * dev_stripes;
b2117a39 2717
73c5de00
AJ		 2718 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2719 continue;
b2117a39 2720
73c5de00
AJ		 2721 devices_info[ndevs].dev_offset = dev_offset;
2722 devices_info[ndevs].max_avail = max_avail;
2723 devices_info[ndevs].total_avail = total_avail;
2724 devices_info[ndevs].dev = device;
2725 ++ndevs;
2726 }
b2117a39 2727
73c5de00
AJ		 2728 /*
2729 * now sort the devices by hole size / available space
2730 */
2731 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2732 btrfs_cmp_device_info, NULL);
b2117a39 2733
73c5de00
AJ		 2734 /* round down to number of usable stripes */
2735 ndevs -= ndevs % devs_increment;
b2117a39 2736
73c5de00
AJ		 2737 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2738 ret = -ENOSPC;
2739 goto error;
b2117a39 2740 }
9f680ce0 2741
73c5de00
AJ		 2742 if (devs_max && ndevs > devs_max)
2743 ndevs = devs_max;
2744 /*
2745 * the primary goal is to maximize the number of stripes, so use as many
2746 * devices as possible, even if the stripes are not maximum sized.
2747 */
2748 stripe_size = devices_info[ndevs-1].max_avail;
2749 num_stripes = ndevs * dev_stripes;
b2117a39 2750
73c5de00
AJ		 2751 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2752 stripe_size = max_chunk_size * ncopies;
2753 do_div(stripe_size, num_stripes);
b2117a39 2754 }
b2117a39 2755
73c5de00
AJ		 2756 do_div(stripe_size, dev_stripes);
2757 do_div(stripe_size, BTRFS_STRIPE_LEN);
2758 stripe_size *= BTRFS_STRIPE_LEN;
b2117a39
MX		 2759
2760 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2761 if (!map) {
2762 ret = -ENOMEM;
2763 goto error;
2764 }
2765 map->num_stripes = num_stripes;
9b3f68b9 2766
73c5de00
AJ		 2767 for (i = 0; i < ndevs; ++i) {
2768 for (j = 0; j < dev_stripes; ++j) {
2769 int s = i * dev_stripes + j;
2770 map->stripes[s].dev = devices_info[i].dev;
2771 map->stripes[s].physical = devices_info[i].dev_offset +
2772 j * stripe_size;
6324fbf3 2773 }
6324fbf3 2774 }
2b82032c 2775 map->sector_size = extent_root->sectorsize;
b2117a39
MX		 2776 map->stripe_len = BTRFS_STRIPE_LEN;
2777 map->io_align = BTRFS_STRIPE_LEN;
2778 map->io_width = BTRFS_STRIPE_LEN;
2b82032c 2779 map->type = type;
2b82032c 2780 map->sub_stripes = sub_stripes;
0b86a832 2781
2b82032c 2782 *map_ret = map;
73c5de00 2783 num_bytes = stripe_size * (num_stripes / ncopies);
0b86a832 2784
73c5de00
AJ		 2785 *stripe_size_out = stripe_size;
2786 *num_bytes_out = num_bytes;
0b86a832 2787
73c5de00 2788 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
1abe9b8a 2789
172ddd60 2790 em = alloc_extent_map();
2b82032c 2791 if (!em) {
b2117a39
MX		 2792 ret = -ENOMEM;
2793 goto error;
593060d7 2794 }
2b82032c
YZ		 2795 em->bdev = (struct block_device *)map;
2796 em->start = start;
73c5de00 2797 em->len = num_bytes;
2b82032c
YZ		 2798 em->block_start = 0;
2799 em->block_len = em->len;
593060d7 2800
2b82032c 2801 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 2802 write_lock(&em_tree->lock);
2b82032c 2803 ret = add_extent_mapping(em_tree, em);
890871be 2804 write_unlock(&em_tree->lock);
2b82032c
YZ		 2805 BUG_ON(ret);
2806 free_extent_map(em);
0b86a832 2807
2b82032c
YZ		 2808 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2809 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
73c5de00 2810 start, num_bytes);
2b82032c 2811 BUG_ON(ret);
611f0e00 2812
73c5de00
AJ		 2813 for (i = 0; i < map->num_stripes; ++i) {
2814 struct btrfs_device *device;
2815 u64 dev_offset;
2816
2817 device = map->stripes[i].dev;
2818 dev_offset = map->stripes[i].physical;
0b86a832
CM		 2819
2820 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2821 info->chunk_root->root_key.objectid,
2822 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
73c5de00 2823 start, dev_offset, stripe_size);
0b86a832 2824 BUG_ON(ret);
2b82032c
YZ		 2825 }
2826
b2117a39 2827 kfree(devices_info);
2b82032c 2828 return 0;
b2117a39
MX		 2829
2830error:
2831 kfree(map);
2832 kfree(devices_info);
2833 return ret;
2b82032c
YZ		 2834}
2835
2836static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2837 struct btrfs_root *extent_root,
2838 struct map_lookup *map, u64 chunk_offset,
2839 u64 chunk_size, u64 stripe_size)
2840{
2841 u64 dev_offset;
2842 struct btrfs_key key;
2843 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2844 struct btrfs_device *device;
2845 struct btrfs_chunk *chunk;
2846 struct btrfs_stripe *stripe;
2847 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2848 int index = 0;
2849 int ret;
2850
2851 chunk = kzalloc(item_size, GFP_NOFS);
2852 if (!chunk)
2853 return -ENOMEM;
2854
2855 index = 0;
2856 while (index < map->num_stripes) {
2857 device = map->stripes[index].dev;
2858 device->bytes_used += stripe_size;
0b86a832
CM		 2859 ret = btrfs_update_device(trans, device);
2860 BUG_ON(ret);
2b82032c
YZ		 2861 index++;
2862 }
2863
2bf64758
JB		 2864 spin_lock(&extent_root->fs_info->free_chunk_lock);
2865 extent_root->fs_info->free_chunk_space -= (stripe_size *
2866 map->num_stripes);
2867 spin_unlock(&extent_root->fs_info->free_chunk_lock);
2868
2b82032c
YZ		 2869 index = 0;
2870 stripe = &chunk->stripe;
2871 while (index < map->num_stripes) {
2872 device = map->stripes[index].dev;
2873 dev_offset = map->stripes[index].physical;
0b86a832 2874
e17cade2
CM		 2875 btrfs_set_stack_stripe_devid(stripe, device->devid);
2876 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2877 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2878 stripe++;
0b86a832
CM		 2879 index++;
2880 }
2881
2b82032c 2882 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2883 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2884 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2885 btrfs_set_stack_chunk_type(chunk, map->type);
2886 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2887 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2888 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2889 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2890 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2891
2b82032c
YZ		 2892 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2893 key.type = BTRFS_CHUNK_ITEM_KEY;
2894 key.offset = chunk_offset;
0b86a832 2895
2b82032c
YZ		 2896 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2897 BUG_ON(ret);
0b86a832 2898
2b82032c
YZ		 2899 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2900 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2901 item_size);
8f18cf13
CM		 2902 BUG_ON(ret);
2903 }
1abe9b8a 2904
0b86a832 2905 kfree(chunk);
2b82032c
YZ		 2906 return 0;
2907}
0b86a832 2908
2b82032c
YZ		 2909/*
2910 * Chunk allocation falls into two parts. The first part does works
2911 * that make the new allocated chunk useable, but not do any operation
2912 * that modifies the chunk tree. The second part does the works that
2913 * require modifying the chunk tree. This division is important for the
2914 * bootstrap process of adding storage to a seed btrfs.
2915 */
2916int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2917 struct btrfs_root *extent_root, u64 type)
2918{
2919 u64 chunk_offset;
2920 u64 chunk_size;
2921 u64 stripe_size;
2922 struct map_lookup *map;
2923 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2924 int ret;
2925
2926 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2927 &chunk_offset);
2928 if (ret)
2929 return ret;
2930
2931 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2932 &stripe_size, chunk_offset, type);
2933 if (ret)
2934 return ret;
2935
2936 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2937 chunk_size, stripe_size);
2938 BUG_ON(ret);
2939 return 0;
2940}
2941
d397712b 2942static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2943 struct btrfs_root *root,
2944 struct btrfs_device *device)
2945{
2946 u64 chunk_offset;
2947 u64 sys_chunk_offset;
2948 u64 chunk_size;
2949 u64 sys_chunk_size;
2950 u64 stripe_size;
2951 u64 sys_stripe_size;
2952 u64 alloc_profile;
2953 struct map_lookup *map;
2954 struct map_lookup *sys_map;
2955 struct btrfs_fs_info *fs_info = root->fs_info;
2956 struct btrfs_root *extent_root = fs_info->extent_root;
2957 int ret;
2958
2959 ret = find_next_chunk(fs_info->chunk_root,
2960 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
92b8e897
MF		 2961 if (ret)
2962 return ret;
2b82032c
YZ		 2963
2964 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
6fef8df1 2965 fs_info->avail_metadata_alloc_bits;
2b82032c
YZ		 2966 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2967
2968 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2969 &stripe_size, chunk_offset, alloc_profile);
2970 BUG_ON(ret);
2971
2972 sys_chunk_offset = chunk_offset + chunk_size;
2973
2974 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
6fef8df1 2975 fs_info->avail_system_alloc_bits;
2b82032c
YZ		 2976 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2977
2978 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2979 &sys_chunk_size, &sys_stripe_size,
2980 sys_chunk_offset, alloc_profile);
2981 BUG_ON(ret);
2982
2983 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2984 BUG_ON(ret);
2985
2986 /*
2987 * Modifying chunk tree needs allocating new blocks from both
2988 * system block group and metadata block group. So we only can
2989 * do operations require modifying the chunk tree after both
2990 * block groups were created.
2991 */
2992 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2993 chunk_size, stripe_size);
2994 BUG_ON(ret);
2995
2996 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2997 sys_chunk_offset, sys_chunk_size,
2998 sys_stripe_size);
b248a415 2999 BUG_ON(ret);
2b82032c
YZ		 3000 return 0;
3001}
3002
3003int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3004{
3005 struct extent_map *em;
3006 struct map_lookup *map;
3007 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3008 int readonly = 0;
3009 int i;
3010
890871be 3011 read_lock(&map_tree->map_tree.lock);
2b82032c 3012 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 3013 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 3014 if (!em)
3015 return 1;
3016
f48b9075
JB		 3017 if (btrfs_test_opt(root, DEGRADED)) {
3018 free_extent_map(em);
3019 return 0;
3020 }
3021
2b82032c
YZ		 3022 map = (struct map_lookup *)em->bdev;
3023 for (i = 0; i < map->num_stripes; i++) {
3024 if (!map->stripes[i].dev->writeable) {
3025 readonly = 1;
3026 break;
3027 }
3028 }
0b86a832 3029 free_extent_map(em);
2b82032c 3030 return readonly;
0b86a832
CM		 3031}
3032
3033void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3034{
a8067e02 3035 extent_map_tree_init(&tree->map_tree);
0b86a832
CM		 3036}
3037
3038void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3039{
3040 struct extent_map *em;
3041
d397712b 3042 while (1) {
890871be 3043 write_lock(&tree->map_tree.lock);
0b86a832
CM		 3044 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3045 if (em)
3046 remove_extent_mapping(&tree->map_tree, em);
890871be 3047 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 3048 if (!em)
3049 break;
3050 kfree(em->bdev);
3051 /* once for us */
3052 free_extent_map(em);
3053 /* once for the tree */
3054 free_extent_map(em);
3055 }
3056}
3057
f188591e
CM		 3058int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3059{
3060 struct extent_map *em;
3061 struct map_lookup *map;
3062 struct extent_map_tree *em_tree = &map_tree->map_tree;
3063 int ret;
3064
890871be 3065 read_lock(&em_tree->lock);
f188591e 3066 em = lookup_extent_mapping(em_tree, logical, len);
890871be 3067 read_unlock(&em_tree->lock);
f188591e
CM		 3068 BUG_ON(!em);
3069
3070 BUG_ON(em->start > logical || em->start + em->len < logical);
3071 map = (struct map_lookup *)em->bdev;
3072 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3073 ret = map->num_stripes;
321aecc6
CM		 3074 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3075 ret = map->sub_stripes;
f188591e
CM		 3076 else
3077 ret = 1;
3078 free_extent_map(em);
f188591e
CM		 3079 return ret;
3080}
3081
dfe25020
CM		 3082static int find_live_mirror(struct map_lookup *map, int first, int num,
3083 int optimal)
3084{
3085 int i;
3086 if (map->stripes[optimal].dev->bdev)
3087 return optimal;
3088 for (i = first; i < first + num; i++) {
3089 if (map->stripes[i].dev->bdev)
3090 return i;
3091 }
3092 /* we couldn't find one that doesn't fail. Just return something
3093 * and the io error handling code will clean up eventually
3094 */
3095 return optimal;
3096}
3097
f2d8d74d
CM		 3098static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3099 u64 logical, u64 *length,
a1d3c478 3100 struct btrfs_bio **bbio_ret,
7eaceacc 3101 int mirror_num)
0b86a832
CM		 3102{
3103 struct extent_map *em;
3104 struct map_lookup *map;
3105 struct extent_map_tree *em_tree = &map_tree->map_tree;
3106 u64 offset;
593060d7 3107 u64 stripe_offset;
fce3bb9a 3108 u64 stripe_end_offset;
593060d7 3109 u64 stripe_nr;
fce3bb9a
LD		 3110 u64 stripe_nr_orig;
3111 u64 stripe_nr_end;
cea9e445 3112 int stripes_allocated = 8;
321aecc6 3113 int stripes_required = 1;
593060d7 3114 int stripe_index;
cea9e445 3115 int i;
f2d8d74d 3116 int num_stripes;
a236aed1 3117 int max_errors = 0;
a1d3c478 3118 struct btrfs_bio *bbio = NULL;
0b86a832 3119
a1d3c478 3120 if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
cea9e445 3121 stripes_allocated = 1;
cea9e445 3122again:
a1d3c478
JS		 3123 if (bbio_ret) {
3124 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
cea9e445 3125 GFP_NOFS);
a1d3c478 3126 if (!bbio)
cea9e445 3127 return -ENOMEM;
a236aed1 3128
a1d3c478 3129 atomic_set(&bbio->error, 0);
cea9e445 3130 }
0b86a832 3131
890871be 3132 read_lock(&em_tree->lock);
0b86a832 3133 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 3134 read_unlock(&em_tree->lock);
f2d8d74d 3135
3b951516 3136 if (!em) {
d397712b
CM		 3137 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3138 (unsigned long long)logical,
3139 (unsigned long long)*length);
f2d8d74d 3140 BUG();
3b951516 3141 }
0b86a832
CM		 3142
3143 BUG_ON(em->start > logical || em->start + em->len < logical);
3144 map = (struct map_lookup *)em->bdev;
3145 offset = logical - em->start;
593060d7 3146
f188591e
CM		 3147 if (mirror_num > map->num_stripes)
3148 mirror_num = 0;
3149
a1d3c478 3150 /* if our btrfs_bio struct is too small, back off and try again */
7b6d91da 3151 if (rw & REQ_WRITE) {
321aecc6
CM		 3152 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3153 BTRFS_BLOCK_GROUP_DUP)) {
3154 stripes_required = map->num_stripes;
a236aed1 3155 max_errors = 1;
321aecc6
CM		 3156 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3157 stripes_required = map->sub_stripes;
a236aed1 3158 max_errors = 1;
321aecc6
CM		 3159 }
3160 }
fce3bb9a 3161 if (rw & REQ_DISCARD) {
52ba6929 3162 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK)
fce3bb9a 3163 stripes_required = map->num_stripes;
fce3bb9a 3164 }
a1d3c478 3165 if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
321aecc6 3166 stripes_allocated < stripes_required) {
cea9e445 3167 stripes_allocated = map->num_stripes;
cea9e445 3168 free_extent_map(em);
a1d3c478 3169 kfree(bbio);
cea9e445
CM		 3170 goto again;
3171 }
593060d7
CM		 3172 stripe_nr = offset;
3173 /*
3174 * stripe_nr counts the total number of stripes we have to stride
3175 * to get to this block
3176 */
3177 do_div(stripe_nr, map->stripe_len);
3178
3179 stripe_offset = stripe_nr * map->stripe_len;
3180 BUG_ON(offset < stripe_offset);
3181
3182 /* stripe_offset is the offset of this block in its stripe*/
3183 stripe_offset = offset - stripe_offset;
3184
fce3bb9a
LD		 3185 if (rw & REQ_DISCARD)
3186 *length = min_t(u64, em->len - offset, *length);
52ba6929 3187 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
cea9e445
CM		 3188 /* we limit the length of each bio to what fits in a stripe */
3189 *length = min_t(u64, em->len - offset,
fce3bb9a 3190 map->stripe_len - stripe_offset);
cea9e445
CM		 3191 } else {
3192 *length = em->len - offset;
3193 }
f2d8d74d 3194
a1d3c478 3195 if (!bbio_ret)
cea9e445
CM		 3196 goto out;
3197
f2d8d74d 3198 num_stripes = 1;
cea9e445 3199 stripe_index = 0;
fce3bb9a
LD		 3200 stripe_nr_orig = stripe_nr;
3201 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3202 (~(map->stripe_len - 1));
3203 do_div(stripe_nr_end, map->stripe_len);
3204 stripe_end_offset = stripe_nr_end * map->stripe_len -
3205 (offset + *length);
3206 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3207 if (rw & REQ_DISCARD)
3208 num_stripes = min_t(u64, map->num_stripes,
3209 stripe_nr_end - stripe_nr_orig);
3210 stripe_index = do_div(stripe_nr, map->num_stripes);
3211 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
212a17ab 3212 if (rw & (REQ_WRITE | REQ_DISCARD))
f2d8d74d 3213 num_stripes = map->num_stripes;
2fff734f 3214 else if (mirror_num)
f188591e 3215 stripe_index = mirror_num - 1;
dfe25020
CM		 3216 else {
3217 stripe_index = find_live_mirror(map, 0,
3218 map->num_stripes,
3219 current->pid % map->num_stripes);
a1d3c478 3220 mirror_num = stripe_index + 1;
dfe25020 3221 }
2fff734f 3222
611f0e00 3223 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
a1d3c478 3224 if (rw & (REQ_WRITE | REQ_DISCARD)) {
f2d8d74d 3225 num_stripes = map->num_stripes;
a1d3c478 3226 } else if (mirror_num) {
f188591e 3227 stripe_index = mirror_num - 1;
a1d3c478
JS		 3228 } else {
3229 mirror_num = 1;
3230 }
2fff734f 3231
321aecc6
CM		 3232 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3233 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 3234
3235 stripe_index = do_div(stripe_nr, factor);
3236 stripe_index *= map->sub_stripes;
3237
7eaceacc 3238 if (rw & REQ_WRITE)
f2d8d74d 3239 num_stripes = map->sub_stripes;
fce3bb9a
LD		 3240 else if (rw & REQ_DISCARD)
3241 num_stripes = min_t(u64, map->sub_stripes *
3242 (stripe_nr_end - stripe_nr_orig),
3243 map->num_stripes);
321aecc6
CM		 3244 else if (mirror_num)
3245 stripe_index += mirror_num - 1;
dfe25020
CM		 3246 else {
3247 stripe_index = find_live_mirror(map, stripe_index,
3248 map->sub_stripes, stripe_index +
3249 current->pid % map->sub_stripes);
a1d3c478 3250 mirror_num = stripe_index + 1;
dfe25020 3251 }
8790d502
CM		 3252 } else {
3253 /*
3254 * after this do_div call, stripe_nr is the number of stripes
3255 * on this device we have to walk to find the data, and
3256 * stripe_index is the number of our device in the stripe array
3257 */
3258 stripe_index = do_div(stripe_nr, map->num_stripes);
a1d3c478 3259 mirror_num = stripe_index + 1;
8790d502 3260 }
593060d7 3261 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 3262
fce3bb9a
LD		 3263 if (rw & REQ_DISCARD) {
3264 for (i = 0; i < num_stripes; i++) {
a1d3c478 3265 bbio->stripes[i].physical =
f2d8d74d
CM		 3266 map->stripes[stripe_index].physical +
3267 stripe_offset + stripe_nr * map->stripe_len;
a1d3c478 3268 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
fce3bb9a
LD		 3269
3270 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3271 u64 stripes;
d9d04879 3272 u32 last_stripe = 0;
fce3bb9a
LD		 3273 int j;
3274
d9d04879
CM		 3275 div_u64_rem(stripe_nr_end - 1,
3276 map->num_stripes,
3277 &last_stripe);
3278
fce3bb9a 3279 for (j = 0; j < map->num_stripes; j++) {
d9d04879
CM		 3280 u32 test;
3281
3282 div_u64_rem(stripe_nr_end - 1 - j,
3283 map->num_stripes, &test);
3284 if (test == stripe_index)
fce3bb9a
LD		 3285 break;
3286 }
3287 stripes = stripe_nr_end - 1 - j;
3288 do_div(stripes, map->num_stripes);
a1d3c478 3289 bbio->stripes[i].length = map->stripe_len *
fce3bb9a
LD		 3290 (stripes - stripe_nr + 1);
3291
3292 if (i == 0) {
a1d3c478 3293 bbio->stripes[i].length -=
fce3bb9a
LD		 3294 stripe_offset;
3295 stripe_offset = 0;
3296 }
3297 if (stripe_index == last_stripe)
a1d3c478 3298 bbio->stripes[i].length -=
fce3bb9a
LD		 3299 stripe_end_offset;
3300 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3301 u64 stripes;
3302 int j;
3303 int factor = map->num_stripes /
3304 map->sub_stripes;
d9d04879
CM		 3305 u32 last_stripe = 0;
3306
3307 div_u64_rem(stripe_nr_end - 1,
3308 factor, &last_stripe);
fce3bb9a
LD		 3309 last_stripe *= map->sub_stripes;
3310
3311 for (j = 0; j < factor; j++) {
d9d04879
CM		 3312 u32 test;
3313
3314 div_u64_rem(stripe_nr_end - 1 - j,
3315 factor, &test);
3316
3317 if (test ==
fce3bb9a
LD		 3318 stripe_index / map->sub_stripes)
3319 break;
3320 }
3321 stripes = stripe_nr_end - 1 - j;
3322 do_div(stripes, factor);
a1d3c478 3323 bbio->stripes[i].length = map->stripe_len *
fce3bb9a
LD		 3324 (stripes - stripe_nr + 1);
3325
3326 if (i < map->sub_stripes) {
a1d3c478 3327 bbio->stripes[i].length -=
fce3bb9a
LD		 3328 stripe_offset;
3329 if (i == map->sub_stripes - 1)
3330 stripe_offset = 0;
3331 }
3332 if (stripe_index >= last_stripe &&
3333 stripe_index <= (last_stripe +
3334 map->sub_stripes - 1)) {
a1d3c478 3335 bbio->stripes[i].length -=
fce3bb9a
LD		 3336 stripe_end_offset;
3337 }
3338 } else
a1d3c478 3339 bbio->stripes[i].length = *length;
fce3bb9a
LD		 3340
3341 stripe_index++;
3342 if (stripe_index == map->num_stripes) {
3343 /* This could only happen for RAID0/10 */
3344 stripe_index = 0;
3345 stripe_nr++;
3346 }
3347 }
3348 } else {
3349 for (i = 0; i < num_stripes; i++) {
a1d3c478 3350 bbio->stripes[i].physical =
212a17ab
LT		 3351 map->stripes[stripe_index].physical +
3352 stripe_offset +
3353 stripe_nr * map->stripe_len;
a1d3c478 3354 bbio->stripes[i].dev =
212a17ab 3355 map->stripes[stripe_index].dev;
fce3bb9a 3356 stripe_index++;
f2d8d74d 3357 }
593060d7 3358 }
a1d3c478
JS		 3359 if (bbio_ret) {
3360 *bbio_ret = bbio;
3361 bbio->num_stripes = num_stripes;
3362 bbio->max_errors = max_errors;
3363 bbio->mirror_num = mirror_num;
f2d8d74d 3364 }
cea9e445 3365out:
0b86a832 3366 free_extent_map(em);
0b86a832
CM		 3367 return 0;
3368}
3369
f2d8d74d
CM		 3370int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3371 u64 logical, u64 *length,
a1d3c478 3372 struct btrfs_bio **bbio_ret, int mirror_num)
f2d8d74d 3373{
a1d3c478 3374 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
7eaceacc 3375 mirror_num);
f2d8d74d
CM		 3376}
3377
a512bbf8
YZ		 3378int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3379 u64 chunk_start, u64 physical, u64 devid,
3380 u64 **logical, int *naddrs, int *stripe_len)
3381{
3382 struct extent_map_tree *em_tree = &map_tree->map_tree;
3383 struct extent_map *em;
3384 struct map_lookup *map;
3385 u64 *buf;
3386 u64 bytenr;
3387 u64 length;
3388 u64 stripe_nr;
3389 int i, j, nr = 0;
3390
890871be 3391 read_lock(&em_tree->lock);
a512bbf8 3392 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 3393 read_unlock(&em_tree->lock);
a512bbf8
YZ		 3394
3395 BUG_ON(!em || em->start != chunk_start);
3396 map = (struct map_lookup *)em->bdev;
3397
3398 length = em->len;
3399 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3400 do_div(length, map->num_stripes / map->sub_stripes);
3401 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3402 do_div(length, map->num_stripes);
3403
3404 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3405 BUG_ON(!buf);
3406
3407 for (i = 0; i < map->num_stripes; i++) {
3408 if (devid && map->stripes[i].dev->devid != devid)
3409 continue;
3410 if (map->stripes[i].physical > physical ||
3411 map->stripes[i].physical + length <= physical)
3412 continue;
3413
3414 stripe_nr = physical - map->stripes[i].physical;
3415 do_div(stripe_nr, map->stripe_len);
3416
3417 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3418 stripe_nr = stripe_nr * map->num_stripes + i;
3419 do_div(stripe_nr, map->sub_stripes);
3420 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3421 stripe_nr = stripe_nr * map->num_stripes + i;
3422 }
3423 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 3424 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 3425 for (j = 0; j < nr; j++) {
3426 if (buf[j] == bytenr)
3427 break;
3428 }
934d375b
CM		 3429 if (j == nr) {
3430 WARN_ON(nr >= map->num_stripes);
a512bbf8 3431 buf[nr++] = bytenr;
934d375b 3432 }
a512bbf8
YZ		 3433 }
3434
a512bbf8
YZ		 3435 *logical = buf;
3436 *naddrs = nr;
3437 *stripe_len = map->stripe_len;
3438
3439 free_extent_map(em);
3440 return 0;
f2d8d74d
CM		 3441}
3442
a1d3c478 3443static void btrfs_end_bio(struct bio *bio, int err)
8790d502 3444{
a1d3c478 3445 struct btrfs_bio *bbio = bio->bi_private;
7d2b4daa 3446 int is_orig_bio = 0;
8790d502 3447
8790d502 3448 if (err)
a1d3c478 3449 atomic_inc(&bbio->error);
8790d502 3450
a1d3c478 3451 if (bio == bbio->orig_bio)
7d2b4daa
CM		 3452 is_orig_bio = 1;
3453
a1d3c478 3454 if (atomic_dec_and_test(&bbio->stripes_pending)) {
7d2b4daa
CM		 3455 if (!is_orig_bio) {
3456 bio_put(bio);
a1d3c478 3457 bio = bbio->orig_bio;
7d2b4daa 3458 }
a1d3c478
JS		 3459 bio->bi_private = bbio->private;
3460 bio->bi_end_io = bbio->end_io;
2774b2ca
JS		 3461 bio->bi_bdev = (struct block_device *)
3462 (unsigned long)bbio->mirror_num;
a236aed1
CM		 3463 /* only send an error to the higher layers if it is
3464 * beyond the tolerance of the multi-bio
3465 */
a1d3c478 3466 if (atomic_read(&bbio->error) > bbio->max_errors) {
a236aed1 3467 err = -EIO;
5dbc8fca 3468 } else {
1259ab75
CM		 3469 /*
3470 * this bio is actually up to date, we didn't
3471 * go over the max number of errors
3472 */
3473 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 3474 err = 0;
1259ab75 3475 }
a1d3c478 3476 kfree(bbio);
8790d502
CM		 3477
3478 bio_endio(bio, err);
7d2b4daa 3479 } else if (!is_orig_bio) {
8790d502
CM		 3480 bio_put(bio);
3481 }
8790d502
CM		 3482}
3483
8b712842
CM		 3484struct async_sched {
3485 struct bio *bio;
3486 int rw;
3487 struct btrfs_fs_info *info;
3488 struct btrfs_work work;
3489};
3490
3491/*
3492 * see run_scheduled_bios for a description of why bios are collected for
3493 * async submit.
3494 *
3495 * This will add one bio to the pending list for a device and make sure
3496 * the work struct is scheduled.
3497 */
d397712b 3498static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 3499 struct btrfs_device *device,
3500 int rw, struct bio *bio)
8b712842
CM		 3501{
3502 int should_queue = 1;
ffbd517d 3503 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 3504
3505 /* don't bother with additional async steps for reads, right now */
7b6d91da 3506 if (!(rw & REQ_WRITE)) {
492bb6de 3507 bio_get(bio);
8b712842 3508 submit_bio(rw, bio);
492bb6de 3509 bio_put(bio);
8b712842
CM		 3510 return 0;
3511 }
3512
3513 /*
0986fe9e 3514 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 3515 * higher layers. Otherwise, the async bio makes it appear we have
3516 * made progress against dirty pages when we've really just put it
3517 * on a queue for later
3518 */
0986fe9e 3519 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 3520 WARN_ON(bio->bi_next);
8b712842
CM		 3521 bio->bi_next = NULL;
3522 bio->bi_rw |= rw;
3523
3524 spin_lock(&device->io_lock);
7b6d91da 3525 if (bio->bi_rw & REQ_SYNC)
ffbd517d
CM		 3526 pending_bios = &device->pending_sync_bios;
3527 else
3528 pending_bios = &device->pending_bios;
8b712842 3529
ffbd517d
CM		 3530 if (pending_bios->tail)
3531 pending_bios->tail->bi_next = bio;
8b712842 3532
ffbd517d
CM		 3533 pending_bios->tail = bio;
3534 if (!pending_bios->head)
3535 pending_bios->head = bio;
8b712842
CM		 3536 if (device->running_pending)
3537 should_queue = 0;
3538
3539 spin_unlock(&device->io_lock);
3540
3541 if (should_queue)
1cc127b5
CM		 3542 btrfs_queue_worker(&root->fs_info->submit_workers,
3543 &device->work);
8b712842
CM		 3544 return 0;
3545}
3546
f188591e 3547int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 3548 int mirror_num, int async_submit)
0b86a832
CM		 3549{
3550 struct btrfs_mapping_tree *map_tree;
3551 struct btrfs_device *dev;
8790d502 3552 struct bio *first_bio = bio;
a62b9401 3553 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 3554 u64 length = 0;
3555 u64 map_length;
0b86a832 3556 int ret;
8790d502
CM		 3557 int dev_nr = 0;
3558 int total_devs = 1;
a1d3c478 3559 struct btrfs_bio *bbio = NULL;
0b86a832 3560
f2d8d74d 3561 length = bio->bi_size;
0b86a832
CM		 3562 map_tree = &root->fs_info->mapping_tree;
3563 map_length = length;
cea9e445 3564
a1d3c478 3565 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
f188591e 3566 mirror_num);
cea9e445
CM		 3567 BUG_ON(ret);
3568
a1d3c478 3569 total_devs = bbio->num_stripes;
cea9e445 3570 if (map_length < length) {
d397712b
CM		 3571 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3572 "len %llu\n", (unsigned long long)logical,
3573 (unsigned long long)length,
3574 (unsigned long long)map_length);
cea9e445
CM		 3575 BUG();
3576 }
a1d3c478
JS		 3577
3578 bbio->orig_bio = first_bio;
3579 bbio->private = first_bio->bi_private;
3580 bbio->end_io = first_bio->bi_end_io;
3581 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
cea9e445 3582
d397712b 3583 while (dev_nr < total_devs) {
a1d3c478
JS		 3584 if (dev_nr < total_devs - 1) {
3585 bio = bio_clone(first_bio, GFP_NOFS);
3586 BUG_ON(!bio);
3587 } else {
3588 bio = first_bio;
8790d502 3589 }
a1d3c478
JS		 3590 bio->bi_private = bbio;
3591 bio->bi_end_io = btrfs_end_bio;
3592 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3593 dev = bbio->stripes[dev_nr].dev;
18e503d6 3594 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
a1d3c478
JS		 3595 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3596 "(%s id %llu), size=%u\n", rw,
3597 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3598 dev->name, dev->devid, bio->bi_size);
dfe25020 3599 bio->bi_bdev = dev->bdev;
8b712842
CM		 3600 if (async_submit)
3601 schedule_bio(root, dev, rw, bio);
3602 else
3603 submit_bio(rw, bio);
dfe25020
CM		 3604 } else {
3605 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3606 bio->bi_sector = logical >> 9;
dfe25020 3607 bio_endio(bio, -EIO);
dfe25020 3608 }
8790d502
CM		 3609 dev_nr++;
3610 }
0b86a832
CM		 3611 return 0;
3612}
3613
a443755f 3614struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 3615 u8 *uuid, u8 *fsid)
0b86a832 3616{
2b82032c
YZ		 3617 struct btrfs_device *device;
3618 struct btrfs_fs_devices *cur_devices;
3619
3620 cur_devices = root->fs_info->fs_devices;
3621 while (cur_devices) {
3622 if (!fsid ||
3623 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3624 device = __find_device(&cur_devices->devices,
3625 devid, uuid);
3626 if (device)
3627 return device;
3628 }
3629 cur_devices = cur_devices->seed;
3630 }
3631 return NULL;
0b86a832
CM		 3632}
3633
dfe25020
CM		 3634static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3635 u64 devid, u8 *dev_uuid)
3636{
3637 struct btrfs_device *device;
3638 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3639
3640 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 3641 if (!device)
3642 return NULL;
dfe25020
CM		 3643 list_add(&device->dev_list,
3644 &fs_devices->devices);
dfe25020
CM		 3645 device->dev_root = root->fs_info->dev_root;
3646 device->devid = devid;
8b712842 3647 device->work.func = pending_bios_fn;
e4404d6e 3648 device->fs_devices = fs_devices;
cd02dca5 3649 device->missing = 1;
dfe25020 3650 fs_devices->num_devices++;
cd02dca5 3651 fs_devices->missing_devices++;
dfe25020 3652 spin_lock_init(&device->io_lock);
d20f7043 3653 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 3654 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3655 return device;
3656}
3657
0b86a832
CM		 3658static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3659 struct extent_buffer *leaf,
3660 struct btrfs_chunk *chunk)
3661{
3662 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3663 struct map_lookup *map;
3664 struct extent_map *em;
3665 u64 logical;
3666 u64 length;
3667 u64 devid;
a443755f 3668 u8 uuid[BTRFS_UUID_SIZE];
593060d7 3669 int num_stripes;
0b86a832 3670 int ret;
593060d7 3671 int i;
0b86a832 3672
e17cade2
CM		 3673 logical = key->offset;
3674 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3675
890871be 3676 read_lock(&map_tree->map_tree.lock);
0b86a832 3677 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 3678 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3679
3680 /* already mapped? */
3681 if (em && em->start <= logical && em->start + em->len > logical) {
3682 free_extent_map(em);
0b86a832
CM		 3683 return 0;
3684 } else if (em) {
3685 free_extent_map(em);
3686 }
0b86a832 3687
172ddd60 3688 em = alloc_extent_map();
0b86a832
CM		 3689 if (!em)
3690 return -ENOMEM;
593060d7
CM		 3691 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3692 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3693 if (!map) {
3694 free_extent_map(em);
3695 return -ENOMEM;
3696 }
3697
3698 em->bdev = (struct block_device *)map;
3699 em->start = logical;
3700 em->len = length;
3701 em->block_start = 0;
c8b97818 3702 em->block_len = em->len;
0b86a832 3703
593060d7
CM		 3704 map->num_stripes = num_stripes;
3705 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3706 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3707 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3708 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3709 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3710 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3711 for (i = 0; i < num_stripes; i++) {
3712 map->stripes[i].physical =
3713 btrfs_stripe_offset_nr(leaf, chunk, i);
3714 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3715 read_extent_buffer(leaf, uuid, (unsigned long)
3716 btrfs_stripe_dev_uuid_nr(chunk, i),
3717 BTRFS_UUID_SIZE);
2b82032c
YZ		 3718 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3719 NULL);
dfe25020 3720 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3721 kfree(map);
3722 free_extent_map(em);
3723 return -EIO;
3724 }
dfe25020
CM		 3725 if (!map->stripes[i].dev) {
3726 map->stripes[i].dev =
3727 add_missing_dev(root, devid, uuid);
3728 if (!map->stripes[i].dev) {
3729 kfree(map);
3730 free_extent_map(em);
3731 return -EIO;
3732 }
3733 }
3734 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3735 }
3736
890871be 3737 write_lock(&map_tree->map_tree.lock);
0b86a832 3738 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 3739 write_unlock(&map_tree->map_tree.lock);
b248a415 3740 BUG_ON(ret);
0b86a832
CM		 3741 free_extent_map(em);
3742
3743 return 0;
3744}
3745
3746static int fill_device_from_item(struct extent_buffer *leaf,
3747 struct btrfs_dev_item *dev_item,
3748 struct btrfs_device *device)
3749{
3750 unsigned long ptr;
0b86a832
CM		 3751
3752 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 3753 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3754 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 3755 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3756 device->type = btrfs_device_type(leaf, dev_item);
3757 device->io_align = btrfs_device_io_align(leaf, dev_item);
3758 device->io_width = btrfs_device_io_width(leaf, dev_item);
3759 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3760
3761 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3762 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3763
0b86a832
CM		 3764 return 0;
3765}
3766
2b82032c
YZ		 3767static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3768{
3769 struct btrfs_fs_devices *fs_devices;
3770 int ret;
3771
3772 mutex_lock(&uuid_mutex);
3773
3774 fs_devices = root->fs_info->fs_devices->seed;
3775 while (fs_devices) {
3776 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3777 ret = 0;
3778 goto out;
3779 }
3780 fs_devices = fs_devices->seed;
3781 }
3782
3783 fs_devices = find_fsid(fsid);
3784 if (!fs_devices) {
3785 ret = -ENOENT;
3786 goto out;
3787 }
e4404d6e
YZ		 3788
3789 fs_devices = clone_fs_devices(fs_devices);
3790 if (IS_ERR(fs_devices)) {
3791 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3792 goto out;
3793 }
3794
97288f2c 3795 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3796 root->fs_info->bdev_holder);
2b82032c
YZ		 3797 if (ret)
3798 goto out;
3799
3800 if (!fs_devices->seeding) {
3801 __btrfs_close_devices(fs_devices);
e4404d6e 3802 free_fs_devices(fs_devices);
2b82032c
YZ		 3803 ret = -EINVAL;
3804 goto out;
3805 }
3806
3807 fs_devices->seed = root->fs_info->fs_devices->seed;
3808 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3809out:
3810 mutex_unlock(&uuid_mutex);
3811 return ret;
3812}
3813
0d81ba5d 3814static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3815 struct extent_buffer *leaf,
3816 struct btrfs_dev_item *dev_item)
3817{
3818 struct btrfs_device *device;
3819 u64 devid;
3820 int ret;
2b82032c 3821 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3822 u8 dev_uuid[BTRFS_UUID_SIZE];
3823
0b86a832 3824 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3825 read_extent_buffer(leaf, dev_uuid,
3826 (unsigned long)btrfs_device_uuid(dev_item),
3827 BTRFS_UUID_SIZE);
2b82032c
YZ		 3828 read_extent_buffer(leaf, fs_uuid,
3829 (unsigned long)btrfs_device_fsid(dev_item),
3830 BTRFS_UUID_SIZE);
3831
3832 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3833 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3834 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3835 return ret;
2b82032c
YZ		 3836 }
3837
3838 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3839 if (!device || !device->bdev) {
e4404d6e 3840 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3841 return -EIO;
3842
3843 if (!device) {
d397712b
CM		 3844 printk(KERN_WARNING "warning devid %llu missing\n",
3845 (unsigned long long)devid);
2b82032c
YZ		 3846 device = add_missing_dev(root, devid, dev_uuid);
3847 if (!device)
3848 return -ENOMEM;
cd02dca5
CM		 3849 } else if (!device->missing) {
3850 /*
3851 * this happens when a device that was properly setup
3852 * in the device info lists suddenly goes bad.
3853 * device->bdev is NULL, and so we have to set
3854 * device->missing to one here
3855 */
3856 root->fs_info->fs_devices->missing_devices++;
3857 device->missing = 1;
2b82032c
YZ		 3858 }
3859 }
3860
3861 if (device->fs_devices != root->fs_info->fs_devices) {
3862 BUG_ON(device->writeable);
3863 if (device->generation !=
3864 btrfs_device_generation(leaf, dev_item))
3865 return -EINVAL;
6324fbf3 3866 }
0b86a832
CM		 3867
3868 fill_device_from_item(leaf, dev_item, device);
3869 device->dev_root = root->fs_info->dev_root;
dfe25020 3870 device->in_fs_metadata = 1;
2bf64758 3871 if (device->writeable) {
2b82032c 3872 device->fs_devices->total_rw_bytes += device->total_bytes;
2bf64758
JB		 3873 spin_lock(&root->fs_info->free_chunk_lock);
3874 root->fs_info->free_chunk_space += device->total_bytes -
3875 device->bytes_used;
3876 spin_unlock(&root->fs_info->free_chunk_lock);
3877 }
0b86a832 3878 ret = 0;
0b86a832
CM		 3879 return ret;
3880}
3881
e4404d6e 3882int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832 3883{
6c41761f 3884 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
a061fc8d 3885 struct extent_buffer *sb;
0b86a832 3886 struct btrfs_disk_key *disk_key;
0b86a832 3887 struct btrfs_chunk *chunk;
84eed90f
CM		 3888 u8 *ptr;
3889 unsigned long sb_ptr;
3890 int ret = 0;
0b86a832
CM		 3891 u32 num_stripes;
3892 u32 array_size;
3893 u32 len = 0;
0b86a832 3894 u32 cur;
84eed90f 3895 struct btrfs_key key;
0b86a832 3896
e4404d6e 3897 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3898 BTRFS_SUPER_INFO_SIZE);
3899 if (!sb)
3900 return -ENOMEM;
3901 btrfs_set_buffer_uptodate(sb);
85d4e461 3902 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4008c04a 3903
a061fc8d 3904 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3905 array_size = btrfs_super_sys_array_size(super_copy);
3906
0b86a832
CM		 3907 ptr = super_copy->sys_chunk_array;
3908 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3909 cur = 0;
3910
3911 while (cur < array_size) {
3912 disk_key = (struct btrfs_disk_key *)ptr;
3913 btrfs_disk_key_to_cpu(&key, disk_key);
3914
a061fc8d 3915 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3916 sb_ptr += len;
3917 cur += len;
3918
0d81ba5d 3919 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3920 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3921 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3922 if (ret)
3923 break;
0b86a832
CM		 3924 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3925 len = btrfs_chunk_item_size(num_stripes);
3926 } else {
84eed90f
CM		 3927 ret = -EIO;
3928 break;
0b86a832
CM		 3929 }
3930 ptr += len;
3931 sb_ptr += len;
3932 cur += len;
3933 }
a061fc8d 3934 free_extent_buffer(sb);
84eed90f 3935 return ret;
0b86a832
CM		 3936}
3937
3938int btrfs_read_chunk_tree(struct btrfs_root *root)
3939{
3940 struct btrfs_path *path;
3941 struct extent_buffer *leaf;
3942 struct btrfs_key key;
3943 struct btrfs_key found_key;
3944 int ret;
3945 int slot;
3946
3947 root = root->fs_info->chunk_root;
3948
3949 path = btrfs_alloc_path();
3950 if (!path)
3951 return -ENOMEM;
3952
3953 /* first we search for all of the device items, and then we
3954 * read in all of the chunk items. This way we can create chunk
3955 * mappings that reference all of the devices that are afound
3956 */
3957 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3958 key.offset = 0;
3959 key.type = 0;
3960again:
3961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 3962 if (ret < 0)
3963 goto error;
d397712b 3964 while (1) {
0b86a832
CM		 3965 leaf = path->nodes[0];
3966 slot = path->slots[0];
3967 if (slot >= btrfs_header_nritems(leaf)) {
3968 ret = btrfs_next_leaf(root, path);
3969 if (ret == 0)
3970 continue;
3971 if (ret < 0)
3972 goto error;
3973 break;
3974 }
3975 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3976 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3977 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3978 break;
3979 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3980 struct btrfs_dev_item *dev_item;
3981 dev_item = btrfs_item_ptr(leaf, slot,
3982 struct btrfs_dev_item);
0d81ba5d 3983 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3984 if (ret)
3985 goto error;
0b86a832
CM		 3986 }
3987 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3988 struct btrfs_chunk *chunk;
3989 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3990 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3991 if (ret)
3992 goto error;
0b86a832
CM		 3993 }
3994 path->slots[0]++;
3995 }
3996 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3997 key.objectid = 0;
b3b4aa74 3998 btrfs_release_path(path);
0b86a832
CM		 3999 goto again;
4000 }
0b86a832
CM		 4001 ret = 0;
4002error:
2b82032c 4003 btrfs_free_path(path);
0b86a832
CM		 4004 return ret;
4005}
kernel source for telekom puls (alcatel/mediatek)