]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/btrfs/volumes.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Add a clear_cache mount option
[net-next-2.6.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>
593060d7 25#include <asm/div64.h>
4b4e25f2 26#include "compat.h"
0b86a832
CM		 27#include "ctree.h"
28#include "extent_map.h"
29#include "disk-io.h"
30#include "transaction.h"
31#include "print-tree.h"
32#include "volumes.h"
8b712842 33#include "async-thread.h"
0b86a832 34
593060d7
CM		 35struct map_lookup {
36 u64 type;
37 int io_align;
38 int io_width;
39 int stripe_len;
40 int sector_size;
41 int num_stripes;
321aecc6 42 int sub_stripes;
cea9e445 43 struct btrfs_bio_stripe stripes[];
593060d7
CM		 44};
45
2b82032c
YZ		 46static int init_first_rw_device(struct btrfs_trans_handle *trans,
47 struct btrfs_root *root,
48 struct btrfs_device *device);
49static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50
593060d7 51#define map_lookup_size(n) (sizeof(struct map_lookup) + \
cea9e445 52 (sizeof(struct btrfs_bio_stripe) * (n)))
593060d7 53
8a4b83cc
CM		 54static DEFINE_MUTEX(uuid_mutex);
55static LIST_HEAD(fs_uuids);
56
a061fc8d
CM		 57void btrfs_lock_volumes(void)
58{
59 mutex_lock(&uuid_mutex);
60}
61
62void btrfs_unlock_volumes(void)
63{
64 mutex_unlock(&uuid_mutex);
65}
66
7d9eb12c
CM		 67static void lock_chunks(struct btrfs_root *root)
68{
7d9eb12c
CM		 69 mutex_lock(&root->fs_info->chunk_mutex);
70}
71
72static void unlock_chunks(struct btrfs_root *root)
73{
7d9eb12c
CM		 74 mutex_unlock(&root->fs_info->chunk_mutex);
75}
76
e4404d6e
YZ		 77static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78{
79 struct btrfs_device *device;
80 WARN_ON(fs_devices->opened);
81 while (!list_empty(&fs_devices->devices)) {
82 device = list_entry(fs_devices->devices.next,
83 struct btrfs_device, dev_list);
84 list_del(&device->dev_list);
85 kfree(device->name);
86 kfree(device);
87 }
88 kfree(fs_devices);
89}
90
8a4b83cc
CM		 91int btrfs_cleanup_fs_uuids(void)
92{
93 struct btrfs_fs_devices *fs_devices;
8a4b83cc 94
2b82032c
YZ		 95 while (!list_empty(&fs_uuids)) {
96 fs_devices = list_entry(fs_uuids.next,
97 struct btrfs_fs_devices, list);
98 list_del(&fs_devices->list);
e4404d6e 99 free_fs_devices(fs_devices);
8a4b83cc
CM		 100 }
101 return 0;
102}
103
a1b32a59
CM		 104static noinline struct btrfs_device *__find_device(struct list_head *head,
105 u64 devid, u8 *uuid)
8a4b83cc
CM		 106{
107 struct btrfs_device *dev;
8a4b83cc 108
c6e30871 109 list_for_each_entry(dev, head, dev_list) {
a443755f 110 if (dev->devid == devid &&
8f18cf13 111 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 112 return dev;
a443755f 113 }
8a4b83cc
CM		 114 }
115 return NULL;
116}
117
a1b32a59 118static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 119{
8a4b83cc
CM		 120 struct btrfs_fs_devices *fs_devices;
121
c6e30871 122 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 123 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
124 return fs_devices;
125 }
126 return NULL;
127}
128
ffbd517d
CM		 129static void requeue_list(struct btrfs_pending_bios *pending_bios,
130 struct bio *head, struct bio *tail)
131{
132
133 struct bio *old_head;
134
135 old_head = pending_bios->head;
136 pending_bios->head = head;
137 if (pending_bios->tail)
138 tail->bi_next = old_head;
139 else
140 pending_bios->tail = tail;
141}
142
8b712842
CM		 143/*
144 * we try to collect pending bios for a device so we don't get a large
145 * number of procs sending bios down to the same device. This greatly
146 * improves the schedulers ability to collect and merge the bios.
147 *
148 * But, it also turns into a long list of bios to process and that is sure
149 * to eventually make the worker thread block. The solution here is to
150 * make some progress and then put this work struct back at the end of
151 * the list if the block device is congested. This way, multiple devices
152 * can make progress from a single worker thread.
153 */
d397712b 154static noinline int run_scheduled_bios(struct btrfs_device *device)
8b712842
CM		 155{
156 struct bio *pending;
157 struct backing_dev_info *bdi;
b64a2851 158 struct btrfs_fs_info *fs_info;
ffbd517d 159 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 160 struct bio *tail;
161 struct bio *cur;
162 int again = 0;
ffbd517d
CM		 163 unsigned long num_run;
164 unsigned long num_sync_run;
d644d8a1 165 unsigned long batch_run = 0;
b64a2851 166 unsigned long limit;
b765ead5 167 unsigned long last_waited = 0;
d84275c9 168 int force_reg = 0;
8b712842 169
bedf762b 170 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 171 fs_info = device->dev_root->fs_info;
172 limit = btrfs_async_submit_limit(fs_info);
173 limit = limit * 2 / 3;
174
ffbd517d
CM		 175 /* we want to make sure that every time we switch from the sync
176 * list to the normal list, we unplug
177 */
178 num_sync_run = 0;
179
8b712842
CM		 180loop:
181 spin_lock(&device->io_lock);
182
a6837051 183loop_lock:
d84275c9 184 num_run = 0;
ffbd517d 185
8b712842
CM		 186 /* take all the bios off the list at once and process them
187 * later on (without the lock held). But, remember the
188 * tail and other pointers so the bios can be properly reinserted
189 * into the list if we hit congestion
190 */
d84275c9 191 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 192 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 193 force_reg = 1;
194 } else {
ffbd517d 195 pending_bios = &device->pending_bios;
d84275c9
CM		 196 force_reg = 0;
197 }
ffbd517d
CM		 198
199 pending = pending_bios->head;
200 tail = pending_bios->tail;
8b712842 201 WARN_ON(pending && !tail);
8b712842
CM		 202
203 /*
204 * if pending was null this time around, no bios need processing
205 * at all and we can stop. Otherwise it'll loop back up again
206 * and do an additional check so no bios are missed.
207 *
208 * device->running_pending is used to synchronize with the
209 * schedule_bio code.
210 */
ffbd517d
CM		 211 if (device->pending_sync_bios.head == NULL &&
212 device->pending_bios.head == NULL) {
8b712842
CM		 213 again = 0;
214 device->running_pending = 0;
ffbd517d
CM		 215 } else {
216 again = 1;
217 device->running_pending = 1;
8b712842 218 }
ffbd517d
CM		 219
220 pending_bios->head = NULL;
221 pending_bios->tail = NULL;
222
8b712842
CM		 223 spin_unlock(&device->io_lock);
224
ffbd517d
CM		 225 /*
226 * if we're doing the regular priority list, make sure we unplug
227 * for any high prio bios we've sent down
228 */
229 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
230 num_sync_run = 0;
231 blk_run_backing_dev(bdi, NULL);
232 }
233
d397712b 234 while (pending) {
ffbd517d
CM		 235
236 rmb();
d84275c9
CM		 237 /* we want to work on both lists, but do more bios on the
238 * sync list than the regular list
239 */
240 if ((num_run > 32 &&
241 pending_bios != &device->pending_sync_bios &&
242 device->pending_sync_bios.head) ||
243 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
244 device->pending_bios.head)) {
ffbd517d
CM		 245 spin_lock(&device->io_lock);
246 requeue_list(pending_bios, pending, tail);
247 goto loop_lock;
248 }
249
8b712842
CM		 250 cur = pending;
251 pending = pending->bi_next;
252 cur->bi_next = NULL;
b64a2851
CM		 253 atomic_dec(&fs_info->nr_async_bios);
254
255 if (atomic_read(&fs_info->nr_async_bios) < limit &&
256 waitqueue_active(&fs_info->async_submit_wait))
257 wake_up(&fs_info->async_submit_wait);
492bb6de
CM		 258
259 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
d644d8a1 260
7b6d91da 261 if (cur->bi_rw & REQ_SYNC)
ffbd517d
CM		 262 num_sync_run++;
263
5ff7ba3a
CM		 264 submit_bio(cur->bi_rw, cur);
265 num_run++;
266 batch_run++;
ffbd517d
CM		 267 if (need_resched()) {
268 if (num_sync_run) {
269 blk_run_backing_dev(bdi, NULL);
270 num_sync_run = 0;
271 }
272 cond_resched();
273 }
8b712842
CM		 274
275 /*
276 * we made progress, there is more work to do and the bdi
277 * is now congested. Back off and let other work structs
278 * run instead
279 */
57fd5a5f 280 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 281 fs_info->fs_devices->open_devices > 1) {
b765ead5 282 struct io_context *ioc;
8b712842 283
b765ead5
CM		 284 ioc = current->io_context;
285
286 /*
287 * the main goal here is that we don't want to
288 * block if we're going to be able to submit
289 * more requests without blocking.
290 *
291 * This code does two great things, it pokes into
292 * the elevator code from a filesystem _and_
293 * it makes assumptions about how batching works.
294 */
295 if (ioc && ioc->nr_batch_requests > 0 &&
296 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
297 (last_waited == 0 ||
298 ioc->last_waited == last_waited)) {
299 /*
300 * we want to go through our batch of
301 * requests and stop. So, we copy out
302 * the ioc->last_waited time and test
303 * against it before looping
304 */
305 last_waited = ioc->last_waited;
ffbd517d
CM		 306 if (need_resched()) {
307 if (num_sync_run) {
308 blk_run_backing_dev(bdi, NULL);
309 num_sync_run = 0;
310 }
311 cond_resched();
312 }
b765ead5
CM		 313 continue;
314 }
8b712842 315 spin_lock(&device->io_lock);
ffbd517d 316 requeue_list(pending_bios, pending, tail);
a6837051 317 device->running_pending = 1;
8b712842
CM		 318
319 spin_unlock(&device->io_lock);
320 btrfs_requeue_work(&device->work);
321 goto done;
322 }
323 }
ffbd517d
CM		 324
325 if (num_sync_run) {
326 num_sync_run = 0;
327 blk_run_backing_dev(bdi, NULL);
328 }
bedf762b
CM		 329 /*
330 * IO has already been through a long path to get here. Checksumming,
331 * async helper threads, perhaps compression. We've done a pretty
332 * good job of collecting a batch of IO and should just unplug
333 * the device right away.
334 *
335 * This will help anyone who is waiting on the IO, they might have
336 * already unplugged, but managed to do so before the bio they
337 * cared about found its way down here.
338 */
339 blk_run_backing_dev(bdi, NULL);
51684082
CM		 340
341 cond_resched();
342 if (again)
343 goto loop;
344
345 spin_lock(&device->io_lock);
346 if (device->pending_bios.head || device->pending_sync_bios.head)
347 goto loop_lock;
348 spin_unlock(&device->io_lock);
349
8b712842
CM		 350done:
351 return 0;
352}
353
b2950863 354static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 355{
356 struct btrfs_device *device;
357
358 device = container_of(work, struct btrfs_device, work);
359 run_scheduled_bios(device);
360}
361
a1b32a59 362static noinline int device_list_add(const char *path,
8a4b83cc
CM		 363 struct btrfs_super_block *disk_super,
364 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
365{
366 struct btrfs_device *device;
367 struct btrfs_fs_devices *fs_devices;
368 u64 found_transid = btrfs_super_generation(disk_super);
3a0524dc 369 char *name;
8a4b83cc
CM		 370
371 fs_devices = find_fsid(disk_super->fsid);
372 if (!fs_devices) {
515dc322 373 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 374 if (!fs_devices)
375 return -ENOMEM;
376 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 377 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 378 list_add(&fs_devices->list, &fs_uuids);
379 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
380 fs_devices->latest_devid = devid;
381 fs_devices->latest_trans = found_transid;
e5e9a520 382 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 383 device = NULL;
384 } else {
a443755f
CM		 385 device = __find_device(&fs_devices->devices, devid,
386 disk_super->dev_item.uuid);
8a4b83cc
CM		 387 }
388 if (!device) {
2b82032c
YZ		 389 if (fs_devices->opened)
390 return -EBUSY;
391
8a4b83cc
CM		 392 device = kzalloc(sizeof(*device), GFP_NOFS);
393 if (!device) {
394 /* we can safely leave the fs_devices entry around */
395 return -ENOMEM;
396 }
397 device->devid = devid;
8b712842 398 device->work.func = pending_bios_fn;
a443755f
CM		 399 memcpy(device->uuid, disk_super->dev_item.uuid,
400 BTRFS_UUID_SIZE);
f2984462 401 device->barriers = 1;
b248a415 402 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 403 device->name = kstrdup(path, GFP_NOFS);
404 if (!device->name) {
405 kfree(device);
406 return -ENOMEM;
407 }
2b82032c 408 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520
CM		 409
410 mutex_lock(&fs_devices->device_list_mutex);
8a4b83cc 411 list_add(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 412 mutex_unlock(&fs_devices->device_list_mutex);
413
2b82032c 414 device->fs_devices = fs_devices;
8a4b83cc 415 fs_devices->num_devices++;
3a0524dc
TH		 416 } else if (strcmp(device->name, path)) {
417 name = kstrdup(path, GFP_NOFS);
418 if (!name)
419 return -ENOMEM;
420 kfree(device->name);
421 device->name = name;
8a4b83cc
CM		 422 }
423
424 if (found_transid > fs_devices->latest_trans) {
425 fs_devices->latest_devid = devid;
426 fs_devices->latest_trans = found_transid;
427 }
8a4b83cc
CM		 428 *fs_devices_ret = fs_devices;
429 return 0;
430}
431
e4404d6e
YZ		 432static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
433{
434 struct btrfs_fs_devices *fs_devices;
435 struct btrfs_device *device;
436 struct btrfs_device *orig_dev;
437
438 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
439 if (!fs_devices)
440 return ERR_PTR(-ENOMEM);
441
442 INIT_LIST_HEAD(&fs_devices->devices);
443 INIT_LIST_HEAD(&fs_devices->alloc_list);
444 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 445 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 446 fs_devices->latest_devid = orig->latest_devid;
447 fs_devices->latest_trans = orig->latest_trans;
448 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
449
e5e9a520 450 mutex_lock(&orig->device_list_mutex);
e4404d6e
YZ		 451 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
452 device = kzalloc(sizeof(*device), GFP_NOFS);
453 if (!device)
454 goto error;
455
456 device->name = kstrdup(orig_dev->name, GFP_NOFS);
fd2696f3
JL		 457 if (!device->name) {
458 kfree(device);
e4404d6e 459 goto error;
fd2696f3 460 }
e4404d6e
YZ		 461
462 device->devid = orig_dev->devid;
463 device->work.func = pending_bios_fn;
464 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
465 device->barriers = 1;
466 spin_lock_init(&device->io_lock);
467 INIT_LIST_HEAD(&device->dev_list);
468 INIT_LIST_HEAD(&device->dev_alloc_list);
469
470 list_add(&device->dev_list, &fs_devices->devices);
471 device->fs_devices = fs_devices;
472 fs_devices->num_devices++;
473 }
e5e9a520 474 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 475 return fs_devices;
476error:
e5e9a520 477 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 478 free_fs_devices(fs_devices);
479 return ERR_PTR(-ENOMEM);
480}
481
dfe25020
CM		 482int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
483{
c6e30871 484 struct btrfs_device *device, *next;
dfe25020
CM		 485
486 mutex_lock(&uuid_mutex);
487again:
e5e9a520 488 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 489 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 490 if (device->in_fs_metadata)
491 continue;
492
493 if (device->bdev) {
15916de8 494 close_bdev_exclusive(device->bdev, device->mode);
2b82032c
YZ		 495 device->bdev = NULL;
496 fs_devices->open_devices--;
497 }
498 if (device->writeable) {
499 list_del_init(&device->dev_alloc_list);
500 device->writeable = 0;
501 fs_devices->rw_devices--;
502 }
e4404d6e
YZ		 503 list_del_init(&device->dev_list);
504 fs_devices->num_devices--;
505 kfree(device->name);
506 kfree(device);
dfe25020 507 }
e5e9a520 508 mutex_unlock(&fs_devices->device_list_mutex);
2b82032c
YZ		 509
510 if (fs_devices->seed) {
511 fs_devices = fs_devices->seed;
2b82032c
YZ		 512 goto again;
513 }
514
dfe25020
CM		 515 mutex_unlock(&uuid_mutex);
516 return 0;
517}
a0af469b 518
2b82032c 519static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 520{
8a4b83cc 521 struct btrfs_device *device;
e4404d6e 522
2b82032c
YZ		 523 if (--fs_devices->opened > 0)
524 return 0;
8a4b83cc 525
c6e30871 526 list_for_each_entry(device, &fs_devices->devices, dev_list) {
8a4b83cc 527 if (device->bdev) {
15916de8 528 close_bdev_exclusive(device->bdev, device->mode);
a0af469b 529 fs_devices->open_devices--;
8a4b83cc 530 }
2b82032c
YZ		 531 if (device->writeable) {
532 list_del_init(&device->dev_alloc_list);
533 fs_devices->rw_devices--;
534 }
535
8a4b83cc 536 device->bdev = NULL;
2b82032c 537 device->writeable = 0;
dfe25020 538 device->in_fs_metadata = 0;
8a4b83cc 539 }
e4404d6e
YZ		 540 WARN_ON(fs_devices->open_devices);
541 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 542 fs_devices->opened = 0;
543 fs_devices->seeding = 0;
2b82032c 544
8a4b83cc
CM		 545 return 0;
546}
547
2b82032c
YZ		 548int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
549{
e4404d6e 550 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 551 int ret;
552
553 mutex_lock(&uuid_mutex);
554 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 555 if (!fs_devices->opened) {
556 seed_devices = fs_devices->seed;
557 fs_devices->seed = NULL;
558 }
2b82032c 559 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 560
561 while (seed_devices) {
562 fs_devices = seed_devices;
563 seed_devices = fs_devices->seed;
564 __btrfs_close_devices(fs_devices);
565 free_fs_devices(fs_devices);
566 }
2b82032c
YZ		 567 return ret;
568}
569
e4404d6e
YZ		 570static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
571 fmode_t flags, void *holder)
8a4b83cc
CM		 572{
573 struct block_device *bdev;
574 struct list_head *head = &fs_devices->devices;
8a4b83cc 575 struct btrfs_device *device;
a0af469b
CM		 576 struct block_device *latest_bdev = NULL;
577 struct buffer_head *bh;
578 struct btrfs_super_block *disk_super;
579 u64 latest_devid = 0;
580 u64 latest_transid = 0;
a0af469b 581 u64 devid;
2b82032c 582 int seeding = 1;
a0af469b 583 int ret = 0;
8a4b83cc 584
c6e30871 585 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 586 if (device->bdev)
587 continue;
dfe25020
CM		 588 if (!device->name)
589 continue;
590
15916de8 591 bdev = open_bdev_exclusive(device->name, flags, holder);
8a4b83cc 592 if (IS_ERR(bdev)) {
d397712b 593 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 594 goto error;
8a4b83cc 595 }
a061fc8d 596 set_blocksize(bdev, 4096);
a0af469b 597
a512bbf8 598 bh = btrfs_read_dev_super(bdev);
a0af469b
CM		 599 if (!bh)
600 goto error_close;
601
602 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 603 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 604 if (devid != device->devid)
605 goto error_brelse;
606
2b82032c
YZ		 607 if (memcmp(device->uuid, disk_super->dev_item.uuid,
608 BTRFS_UUID_SIZE))
609 goto error_brelse;
610
611 device->generation = btrfs_super_generation(disk_super);
612 if (!latest_transid || device->generation > latest_transid) {
a0af469b 613 latest_devid = devid;
2b82032c 614 latest_transid = device->generation;
a0af469b
CM		 615 latest_bdev = bdev;
616 }
617
2b82032c
YZ		 618 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
619 device->writeable = 0;
620 } else {
621 device->writeable = !bdev_read_only(bdev);
622 seeding = 0;
623 }
624
8a4b83cc 625 device->bdev = bdev;
dfe25020 626 device->in_fs_metadata = 0;
15916de8
CM		 627 device->mode = flags;
628
c289811c
CM		 629 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
630 fs_devices->rotating = 1;
631
a0af469b 632 fs_devices->open_devices++;
2b82032c
YZ		 633 if (device->writeable) {
634 fs_devices->rw_devices++;
635 list_add(&device->dev_alloc_list,
636 &fs_devices->alloc_list);
637 }
a0af469b 638 continue;
a061fc8d 639
a0af469b
CM		 640error_brelse:
641 brelse(bh);
642error_close:
97288f2c 643 close_bdev_exclusive(bdev, FMODE_READ);
a0af469b
CM		 644error:
645 continue;
8a4b83cc 646 }
a0af469b
CM		 647 if (fs_devices->open_devices == 0) {
648 ret = -EIO;
649 goto out;
650 }
2b82032c
YZ		 651 fs_devices->seeding = seeding;
652 fs_devices->opened = 1;
a0af469b
CM		 653 fs_devices->latest_bdev = latest_bdev;
654 fs_devices->latest_devid = latest_devid;
655 fs_devices->latest_trans = latest_transid;
2b82032c 656 fs_devices->total_rw_bytes = 0;
a0af469b 657out:
2b82032c
YZ		 658 return ret;
659}
660
661int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 662 fmode_t flags, void *holder)
2b82032c
YZ		 663{
664 int ret;
665
666 mutex_lock(&uuid_mutex);
667 if (fs_devices->opened) {
e4404d6e
YZ		 668 fs_devices->opened++;
669 ret = 0;
2b82032c 670 } else {
15916de8 671 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 672 }
8a4b83cc 673 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 674 return ret;
675}
676
97288f2c 677int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 678 struct btrfs_fs_devices **fs_devices_ret)
679{
680 struct btrfs_super_block *disk_super;
681 struct block_device *bdev;
682 struct buffer_head *bh;
683 int ret;
684 u64 devid;
f2984462 685 u64 transid;
8a4b83cc
CM		 686
687 mutex_lock(&uuid_mutex);
688
15916de8 689 bdev = open_bdev_exclusive(path, flags, holder);
8a4b83cc
CM		 690
691 if (IS_ERR(bdev)) {
8a4b83cc
CM		 692 ret = PTR_ERR(bdev);
693 goto error;
694 }
695
696 ret = set_blocksize(bdev, 4096);
697 if (ret)
698 goto error_close;
a512bbf8 699 bh = btrfs_read_dev_super(bdev);
8a4b83cc
CM		 700 if (!bh) {
701 ret = -EIO;
702 goto error_close;
703 }
704 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 705 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 706 transid = btrfs_super_generation(disk_super);
7ae9c09d 707 if (disk_super->label[0])
d397712b 708 printk(KERN_INFO "device label %s ", disk_super->label);
7ae9c09d
CM		 709 else {
710 /* FIXME, make a readl uuid parser */
d397712b 711 printk(KERN_INFO "device fsid %llx-%llx ",
7ae9c09d
CM		 712 *(unsigned long long *)disk_super->fsid,
713 *(unsigned long long *)(disk_super->fsid + 8));
714 }
119e10cf 715 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 716 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 717 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
718
8a4b83cc
CM		 719 brelse(bh);
720error_close:
15916de8 721 close_bdev_exclusive(bdev, flags);
8a4b83cc
CM		 722error:
723 mutex_unlock(&uuid_mutex);
724 return ret;
725}
0b86a832
CM		 726
727/*
728 * this uses a pretty simple search, the expectation is that it is
729 * called very infrequently and that a given device has a small number
730 * of extents
731 */
ba1bf481
JB		 732int find_free_dev_extent(struct btrfs_trans_handle *trans,
733 struct btrfs_device *device, u64 num_bytes,
734 u64 *start, u64 *max_avail)
0b86a832
CM		 735{
736 struct btrfs_key key;
737 struct btrfs_root *root = device->dev_root;
738 struct btrfs_dev_extent *dev_extent = NULL;
2b82032c 739 struct btrfs_path *path;
0b86a832
CM		 740 u64 hole_size = 0;
741 u64 last_byte = 0;
742 u64 search_start = 0;
743 u64 search_end = device->total_bytes;
744 int ret;
745 int slot = 0;
746 int start_found;
747 struct extent_buffer *l;
748
2b82032c
YZ		 749 path = btrfs_alloc_path();
750 if (!path)
751 return -ENOMEM;
0b86a832 752 path->reada = 2;
2b82032c 753 start_found = 0;
0b86a832
CM		 754
755 /* FIXME use last free of some kind */
756
8a4b83cc
CM		 757 /* we don't want to overwrite the superblock on the drive,
758 * so we make sure to start at an offset of at least 1MB
759 */
760 search_start = max((u64)1024 * 1024, search_start);
8f18cf13
CM		 761
762 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
763 search_start = max(root->fs_info->alloc_start, search_start);
764
0b86a832
CM		 765 key.objectid = device->devid;
766 key.offset = search_start;
767 key.type = BTRFS_DEV_EXTENT_KEY;
768 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
769 if (ret < 0)
770 goto error;
1fcbac58
YZ		 771 if (ret > 0) {
772 ret = btrfs_previous_item(root, path, key.objectid, key.type);
773 if (ret < 0)
774 goto error;
775 if (ret > 0)
776 start_found = 1;
777 }
0b86a832
CM		 778 l = path->nodes[0];
779 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
780 while (1) {
781 l = path->nodes[0];
782 slot = path->slots[0];
783 if (slot >= btrfs_header_nritems(l)) {
784 ret = btrfs_next_leaf(root, path);
785 if (ret == 0)
786 continue;
787 if (ret < 0)
788 goto error;
789no_more_items:
790 if (!start_found) {
791 if (search_start >= search_end) {
792 ret = -ENOSPC;
793 goto error;
794 }
795 *start = search_start;
796 start_found = 1;
797 goto check_pending;
798 }
799 *start = last_byte > search_start ?
800 last_byte : search_start;
801 if (search_end <= *start) {
802 ret = -ENOSPC;
803 goto error;
804 }
805 goto check_pending;
806 }
807 btrfs_item_key_to_cpu(l, &key, slot);
808
809 if (key.objectid < device->devid)
810 goto next;
811
812 if (key.objectid > device->devid)
813 goto no_more_items;
814
815 if (key.offset >= search_start && key.offset > last_byte &&
816 start_found) {
817 if (last_byte < search_start)
818 last_byte = search_start;
819 hole_size = key.offset - last_byte;
9779b72f
CM		 820
821 if (hole_size > *max_avail)
822 *max_avail = hole_size;
823
0b86a832
CM		 824 if (key.offset > last_byte &&
825 hole_size >= num_bytes) {
826 *start = last_byte;
827 goto check_pending;
828 }
829 }
d397712b 830 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
0b86a832 831 goto next;
0b86a832
CM		 832
833 start_found = 1;
834 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
835 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
836next:
837 path->slots[0]++;
838 cond_resched();
839 }
840check_pending:
841 /* we have to make sure we didn't find an extent that has already
842 * been allocated by the map tree or the original allocation
843 */
0b86a832
CM		 844 BUG_ON(*start < search_start);
845
6324fbf3 846 if (*start + num_bytes > search_end) {
0b86a832
CM		 847 ret = -ENOSPC;
848 goto error;
849 }
850 /* check for pending inserts here */
2b82032c 851 ret = 0;
0b86a832
CM		 852
853error:
2b82032c 854 btrfs_free_path(path);
0b86a832
CM		 855 return ret;
856}
857
b2950863 858static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 859 struct btrfs_device *device,
860 u64 start)
861{
862 int ret;
863 struct btrfs_path *path;
864 struct btrfs_root *root = device->dev_root;
865 struct btrfs_key key;
a061fc8d
CM		 866 struct btrfs_key found_key;
867 struct extent_buffer *leaf = NULL;
868 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 869
870 path = btrfs_alloc_path();
871 if (!path)
872 return -ENOMEM;
873
874 key.objectid = device->devid;
875 key.offset = start;
876 key.type = BTRFS_DEV_EXTENT_KEY;
877
878 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 879 if (ret > 0) {
880 ret = btrfs_previous_item(root, path, key.objectid,
881 BTRFS_DEV_EXTENT_KEY);
882 BUG_ON(ret);
883 leaf = path->nodes[0];
884 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
885 extent = btrfs_item_ptr(leaf, path->slots[0],
886 struct btrfs_dev_extent);
887 BUG_ON(found_key.offset > start || found_key.offset +
888 btrfs_dev_extent_length(leaf, extent) < start);
889 ret = 0;
890 } else if (ret == 0) {
891 leaf = path->nodes[0];
892 extent = btrfs_item_ptr(leaf, path->slots[0],
893 struct btrfs_dev_extent);
894 }
8f18cf13
CM		 895 BUG_ON(ret);
896
dfe25020
CM		 897 if (device->bytes_used > 0)
898 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
8f18cf13
CM		 899 ret = btrfs_del_item(trans, root, path);
900 BUG_ON(ret);
901
902 btrfs_free_path(path);
903 return ret;
904}
905
2b82032c 906int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 907 struct btrfs_device *device,
e17cade2 908 u64 chunk_tree, u64 chunk_objectid,
2b82032c 909 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 910{
911 int ret;
912 struct btrfs_path *path;
913 struct btrfs_root *root = device->dev_root;
914 struct btrfs_dev_extent *extent;
915 struct extent_buffer *leaf;
916 struct btrfs_key key;
917
dfe25020 918 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 919 path = btrfs_alloc_path();
920 if (!path)
921 return -ENOMEM;
922
0b86a832 923 key.objectid = device->devid;
2b82032c 924 key.offset = start;
0b86a832
CM		 925 key.type = BTRFS_DEV_EXTENT_KEY;
926 ret = btrfs_insert_empty_item(trans, root, path, &key,
927 sizeof(*extent));
928 BUG_ON(ret);
929
930 leaf = path->nodes[0];
931 extent = btrfs_item_ptr(leaf, path->slots[0],
932 struct btrfs_dev_extent);
e17cade2
CM		 933 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
934 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
935 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
936
937 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
938 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
939 BTRFS_UUID_SIZE);
940
0b86a832
CM		 941 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
942 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 943 btrfs_free_path(path);
944 return ret;
945}
946
a1b32a59
CM		 947static noinline int find_next_chunk(struct btrfs_root *root,
948 u64 objectid, u64 *offset)
0b86a832
CM		 949{
950 struct btrfs_path *path;
951 int ret;
952 struct btrfs_key key;
e17cade2 953 struct btrfs_chunk *chunk;
0b86a832
CM		 954 struct btrfs_key found_key;
955
956 path = btrfs_alloc_path();
957 BUG_ON(!path);
958
e17cade2 959 key.objectid = objectid;
0b86a832
CM		 960 key.offset = (u64)-1;
961 key.type = BTRFS_CHUNK_ITEM_KEY;
962
963 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
964 if (ret < 0)
965 goto error;
966
967 BUG_ON(ret == 0);
968
969 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
970 if (ret) {
e17cade2 971 *offset = 0;
0b86a832
CM		 972 } else {
973 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
974 path->slots[0]);
e17cade2
CM		 975 if (found_key.objectid != objectid)
976 *offset = 0;
977 else {
978 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
979 struct btrfs_chunk);
980 *offset = found_key.offset +
981 btrfs_chunk_length(path->nodes[0], chunk);
982 }
0b86a832
CM		 983 }
984 ret = 0;
985error:
986 btrfs_free_path(path);
987 return ret;
988}
989
2b82032c 990static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 991{
992 int ret;
993 struct btrfs_key key;
994 struct btrfs_key found_key;
2b82032c
YZ		 995 struct btrfs_path *path;
996
997 root = root->fs_info->chunk_root;
998
999 path = btrfs_alloc_path();
1000 if (!path)
1001 return -ENOMEM;
0b86a832
CM		 1002
1003 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1004 key.type = BTRFS_DEV_ITEM_KEY;
1005 key.offset = (u64)-1;
1006
1007 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1008 if (ret < 0)
1009 goto error;
1010
1011 BUG_ON(ret == 0);
1012
1013 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1014 BTRFS_DEV_ITEM_KEY);
1015 if (ret) {
1016 *objectid = 1;
1017 } else {
1018 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1019 path->slots[0]);
1020 *objectid = found_key.offset + 1;
1021 }
1022 ret = 0;
1023error:
2b82032c 1024 btrfs_free_path(path);
0b86a832
CM		 1025 return ret;
1026}
1027
1028/*
1029 * the device information is stored in the chunk root
1030 * the btrfs_device struct should be fully filled in
1031 */
1032int btrfs_add_device(struct btrfs_trans_handle *trans,
1033 struct btrfs_root *root,
1034 struct btrfs_device *device)
1035{
1036 int ret;
1037 struct btrfs_path *path;
1038 struct btrfs_dev_item *dev_item;
1039 struct extent_buffer *leaf;
1040 struct btrfs_key key;
1041 unsigned long ptr;
0b86a832
CM		 1042
1043 root = root->fs_info->chunk_root;
1044
1045 path = btrfs_alloc_path();
1046 if (!path)
1047 return -ENOMEM;
1048
0b86a832
CM		 1049 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1050 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1051 key.offset = device->devid;
0b86a832
CM		 1052
1053 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1054 sizeof(*dev_item));
0b86a832
CM		 1055 if (ret)
1056 goto out;
1057
1058 leaf = path->nodes[0];
1059 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1060
1061 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1062 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1063 btrfs_set_device_type(leaf, dev_item, device->type);
1064 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1065 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1066 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1067 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1068 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1069 btrfs_set_device_group(leaf, dev_item, 0);
1070 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1071 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1072 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1073
0b86a832 1074 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1075 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1076 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1077 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1078 btrfs_mark_buffer_dirty(leaf);
0b86a832 1079
2b82032c 1080 ret = 0;
0b86a832
CM		 1081out:
1082 btrfs_free_path(path);
1083 return ret;
1084}
8f18cf13 1085
a061fc8d
CM		 1086static int btrfs_rm_dev_item(struct btrfs_root *root,
1087 struct btrfs_device *device)
1088{
1089 int ret;
1090 struct btrfs_path *path;
a061fc8d 1091 struct btrfs_key key;
a061fc8d
CM		 1092 struct btrfs_trans_handle *trans;
1093
1094 root = root->fs_info->chunk_root;
1095
1096 path = btrfs_alloc_path();
1097 if (!path)
1098 return -ENOMEM;
1099
a22285a6 1100 trans = btrfs_start_transaction(root, 0);
a061fc8d
CM		 1101 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1102 key.type = BTRFS_DEV_ITEM_KEY;
1103 key.offset = device->devid;
7d9eb12c 1104 lock_chunks(root);
a061fc8d
CM		 1105
1106 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 if (ret < 0)
1108 goto out;
1109
1110 if (ret > 0) {
1111 ret = -ENOENT;
1112 goto out;
1113 }
1114
1115 ret = btrfs_del_item(trans, root, path);
1116 if (ret)
1117 goto out;
a061fc8d
CM		 1118out:
1119 btrfs_free_path(path);
7d9eb12c 1120 unlock_chunks(root);
a061fc8d
CM		 1121 btrfs_commit_transaction(trans, root);
1122 return ret;
1123}
1124
1125int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1126{
1127 struct btrfs_device *device;
2b82032c 1128 struct btrfs_device *next_device;
a061fc8d 1129 struct block_device *bdev;
dfe25020 1130 struct buffer_head *bh = NULL;
a061fc8d
CM		 1131 struct btrfs_super_block *disk_super;
1132 u64 all_avail;
1133 u64 devid;
2b82032c
YZ		 1134 u64 num_devices;
1135 u8 *dev_uuid;
a061fc8d
CM		 1136 int ret = 0;
1137
a061fc8d 1138 mutex_lock(&uuid_mutex);
7d9eb12c 1139 mutex_lock(&root->fs_info->volume_mutex);
a061fc8d
CM		 1140
1141 all_avail = root->fs_info->avail_data_alloc_bits |
1142 root->fs_info->avail_system_alloc_bits |
1143 root->fs_info->avail_metadata_alloc_bits;
1144
1145 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
035fe03a 1146 root->fs_info->fs_devices->num_devices <= 4) {
d397712b
CM		 1147 printk(KERN_ERR "btrfs: unable to go below four devices "
1148 "on raid10\n");
a061fc8d
CM		 1149 ret = -EINVAL;
1150 goto out;
1151 }
1152
1153 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
035fe03a 1154 root->fs_info->fs_devices->num_devices <= 2) {
d397712b
CM		 1155 printk(KERN_ERR "btrfs: unable to go below two "
1156 "devices on raid1\n");
a061fc8d
CM		 1157 ret = -EINVAL;
1158 goto out;
1159 }
1160
dfe25020 1161 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1162 struct list_head *devices;
1163 struct btrfs_device *tmp;
a061fc8d 1164
dfe25020
CM		 1165 device = NULL;
1166 devices = &root->fs_info->fs_devices->devices;
e5e9a520 1167 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
c6e30871 1168 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1169 if (tmp->in_fs_metadata && !tmp->bdev) {
1170 device = tmp;
1171 break;
1172 }
1173 }
e5e9a520 1174 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
dfe25020
CM		 1175 bdev = NULL;
1176 bh = NULL;
1177 disk_super = NULL;
1178 if (!device) {
d397712b
CM		 1179 printk(KERN_ERR "btrfs: no missing devices found to "
1180 "remove\n");
dfe25020
CM		 1181 goto out;
1182 }
dfe25020 1183 } else {
97288f2c 1184 bdev = open_bdev_exclusive(device_path, FMODE_READ,
dfe25020
CM		 1185 root->fs_info->bdev_holder);
1186 if (IS_ERR(bdev)) {
1187 ret = PTR_ERR(bdev);
1188 goto out;
1189 }
a061fc8d 1190
2b82032c 1191 set_blocksize(bdev, 4096);
a512bbf8 1192 bh = btrfs_read_dev_super(bdev);
dfe25020
CM		 1193 if (!bh) {
1194 ret = -EIO;
1195 goto error_close;
1196 }
1197 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 1198 devid = btrfs_stack_device_id(&disk_super->dev_item);
2b82032c
YZ		 1199 dev_uuid = disk_super->dev_item.uuid;
1200 device = btrfs_find_device(root, devid, dev_uuid,
1201 disk_super->fsid);
dfe25020
CM		 1202 if (!device) {
1203 ret = -ENOENT;
1204 goto error_brelse;
1205 }
2b82032c 1206 }
dfe25020 1207
2b82032c 1208 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1209 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1210 "device\n");
2b82032c
YZ		 1211 ret = -EINVAL;
1212 goto error_brelse;
1213 }
1214
1215 if (device->writeable) {
1216 list_del_init(&device->dev_alloc_list);
1217 root->fs_info->fs_devices->rw_devices--;
dfe25020 1218 }
a061fc8d
CM		 1219
1220 ret = btrfs_shrink_device(device, 0);
1221 if (ret)
1222 goto error_brelse;
1223
a061fc8d
CM		 1224 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1225 if (ret)
1226 goto error_brelse;
1227
2b82032c 1228 device->in_fs_metadata = 0;
e5e9a520
CM		 1229
1230 /*
1231 * the device list mutex makes sure that we don't change
1232 * the device list while someone else is writing out all
1233 * the device supers.
1234 */
1235 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1236 list_del_init(&device->dev_list);
e5e9a520
CM		 1237 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1238
e4404d6e 1239 device->fs_devices->num_devices--;
2b82032c
YZ		 1240
1241 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1242 struct btrfs_device, dev_list);
1243 if (device->bdev == root->fs_info->sb->s_bdev)
1244 root->fs_info->sb->s_bdev = next_device->bdev;
1245 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1246 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1247
e4404d6e
YZ		 1248 if (device->bdev) {
1249 close_bdev_exclusive(device->bdev, device->mode);
1250 device->bdev = NULL;
1251 device->fs_devices->open_devices--;
1252 }
1253
2b82032c
YZ		 1254 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1255 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1256
e4404d6e
YZ		 1257 if (device->fs_devices->open_devices == 0) {
1258 struct btrfs_fs_devices *fs_devices;
1259 fs_devices = root->fs_info->fs_devices;
1260 while (fs_devices) {
1261 if (fs_devices->seed == device->fs_devices)
1262 break;
1263 fs_devices = fs_devices->seed;
2b82032c 1264 }
e4404d6e
YZ		 1265 fs_devices->seed = device->fs_devices->seed;
1266 device->fs_devices->seed = NULL;
1267 __btrfs_close_devices(device->fs_devices);
1268 free_fs_devices(device->fs_devices);
2b82032c
YZ		 1269 }
1270
1271 /*
1272 * at this point, the device is zero sized. We want to
1273 * remove it from the devices list and zero out the old super
1274 */
1275 if (device->writeable) {
dfe25020
CM		 1276 /* make sure this device isn't detected as part of
1277 * the FS anymore
1278 */
1279 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1280 set_buffer_dirty(bh);
1281 sync_dirty_buffer(bh);
dfe25020 1282 }
a061fc8d
CM		 1283
1284 kfree(device->name);
1285 kfree(device);
1286 ret = 0;
a061fc8d
CM		 1287
1288error_brelse:
1289 brelse(bh);
1290error_close:
dfe25020 1291 if (bdev)
97288f2c 1292 close_bdev_exclusive(bdev, FMODE_READ);
a061fc8d 1293out:
7d9eb12c 1294 mutex_unlock(&root->fs_info->volume_mutex);
a061fc8d 1295 mutex_unlock(&uuid_mutex);
a061fc8d
CM		 1296 return ret;
1297}
1298
2b82032c
YZ		 1299/*
1300 * does all the dirty work required for changing file system's UUID.
1301 */
1302static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1303 struct btrfs_root *root)
1304{
1305 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1306 struct btrfs_fs_devices *old_devices;
e4404d6e 1307 struct btrfs_fs_devices *seed_devices;
2b82032c
YZ		 1308 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1309 struct btrfs_device *device;
1310 u64 super_flags;
1311
1312 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1313 if (!fs_devices->seeding)
2b82032c
YZ		 1314 return -EINVAL;
1315
e4404d6e
YZ		 1316 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1317 if (!seed_devices)
2b82032c
YZ		 1318 return -ENOMEM;
1319
e4404d6e
YZ		 1320 old_devices = clone_fs_devices(fs_devices);
1321 if (IS_ERR(old_devices)) {
1322 kfree(seed_devices);
1323 return PTR_ERR(old_devices);
2b82032c 1324 }
e4404d6e 1325
2b82032c
YZ		 1326 list_add(&old_devices->list, &fs_uuids);
1327
e4404d6e
YZ		 1328 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1329 seed_devices->opened = 1;
1330 INIT_LIST_HEAD(&seed_devices->devices);
1331 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1332 mutex_init(&seed_devices->device_list_mutex);
e4404d6e
YZ		 1333 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1334 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1335 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1336 device->fs_devices = seed_devices;
1337 }
1338
2b82032c
YZ		 1339 fs_devices->seeding = 0;
1340 fs_devices->num_devices = 0;
1341 fs_devices->open_devices = 0;
e4404d6e 1342 fs_devices->seed = seed_devices;
2b82032c
YZ		 1343
1344 generate_random_uuid(fs_devices->fsid);
1345 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1346 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1347 super_flags = btrfs_super_flags(disk_super) &
1348 ~BTRFS_SUPER_FLAG_SEEDING;
1349 btrfs_set_super_flags(disk_super, super_flags);
1350
1351 return 0;
1352}
1353
1354/*
1355 * strore the expected generation for seed devices in device items.
1356 */
1357static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1358 struct btrfs_root *root)
1359{
1360 struct btrfs_path *path;
1361 struct extent_buffer *leaf;
1362 struct btrfs_dev_item *dev_item;
1363 struct btrfs_device *device;
1364 struct btrfs_key key;
1365 u8 fs_uuid[BTRFS_UUID_SIZE];
1366 u8 dev_uuid[BTRFS_UUID_SIZE];
1367 u64 devid;
1368 int ret;
1369
1370 path = btrfs_alloc_path();
1371 if (!path)
1372 return -ENOMEM;
1373
1374 root = root->fs_info->chunk_root;
1375 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1376 key.offset = 0;
1377 key.type = BTRFS_DEV_ITEM_KEY;
1378
1379 while (1) {
1380 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1381 if (ret < 0)
1382 goto error;
1383
1384 leaf = path->nodes[0];
1385next_slot:
1386 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1387 ret = btrfs_next_leaf(root, path);
1388 if (ret > 0)
1389 break;
1390 if (ret < 0)
1391 goto error;
1392 leaf = path->nodes[0];
1393 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1394 btrfs_release_path(root, path);
1395 continue;
1396 }
1397
1398 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1399 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1400 key.type != BTRFS_DEV_ITEM_KEY)
1401 break;
1402
1403 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1404 struct btrfs_dev_item);
1405 devid = btrfs_device_id(leaf, dev_item);
1406 read_extent_buffer(leaf, dev_uuid,
1407 (unsigned long)btrfs_device_uuid(dev_item),
1408 BTRFS_UUID_SIZE);
1409 read_extent_buffer(leaf, fs_uuid,
1410 (unsigned long)btrfs_device_fsid(dev_item),
1411 BTRFS_UUID_SIZE);
1412 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1413 BUG_ON(!device);
1414
1415 if (device->fs_devices->seeding) {
1416 btrfs_set_device_generation(leaf, dev_item,
1417 device->generation);
1418 btrfs_mark_buffer_dirty(leaf);
1419 }
1420
1421 path->slots[0]++;
1422 goto next_slot;
1423 }
1424 ret = 0;
1425error:
1426 btrfs_free_path(path);
1427 return ret;
1428}
1429
788f20eb
CM		 1430int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1431{
1432 struct btrfs_trans_handle *trans;
1433 struct btrfs_device *device;
1434 struct block_device *bdev;
788f20eb 1435 struct list_head *devices;
2b82032c 1436 struct super_block *sb = root->fs_info->sb;
788f20eb 1437 u64 total_bytes;
2b82032c 1438 int seeding_dev = 0;
788f20eb
CM		 1439 int ret = 0;
1440
2b82032c
YZ		 1441 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1442 return -EINVAL;
788f20eb 1443
15916de8 1444 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
7f59203a
JB		 1445 if (IS_ERR(bdev))
1446 return PTR_ERR(bdev);
a2135011 1447
2b82032c
YZ		 1448 if (root->fs_info->fs_devices->seeding) {
1449 seeding_dev = 1;
1450 down_write(&sb->s_umount);
1451 mutex_lock(&uuid_mutex);
1452 }
1453
8c8bee1d 1454 filemap_write_and_wait(bdev->bd_inode->i_mapping);
7d9eb12c 1455 mutex_lock(&root->fs_info->volume_mutex);
a2135011 1456
788f20eb 1457 devices = &root->fs_info->fs_devices->devices;
e5e9a520
CM		 1458 /*
1459 * we have the volume lock, so we don't need the extra
1460 * device list mutex while reading the list here.
1461 */
c6e30871 1462 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1463 if (device->bdev == bdev) {
1464 ret = -EEXIST;
2b82032c 1465 goto error;
788f20eb
CM		 1466 }
1467 }
1468
1469 device = kzalloc(sizeof(*device), GFP_NOFS);
1470 if (!device) {
1471 /* we can safely leave the fs_devices entry around */
1472 ret = -ENOMEM;
2b82032c 1473 goto error;
788f20eb
CM		 1474 }
1475
788f20eb
CM		 1476 device->name = kstrdup(device_path, GFP_NOFS);
1477 if (!device->name) {
1478 kfree(device);
2b82032c
YZ		 1479 ret = -ENOMEM;
1480 goto error;
788f20eb 1481 }
2b82032c
YZ		 1482
1483 ret = find_next_devid(root, &device->devid);
1484 if (ret) {
1485 kfree(device);
1486 goto error;
1487 }
1488
a22285a6 1489 trans = btrfs_start_transaction(root, 0);
2b82032c
YZ		 1490 lock_chunks(root);
1491
1492 device->barriers = 1;
1493 device->writeable = 1;
1494 device->work.func = pending_bios_fn;
1495 generate_random_uuid(device->uuid);
1496 spin_lock_init(&device->io_lock);
1497 device->generation = trans->transid;
788f20eb
CM		 1498 device->io_width = root->sectorsize;
1499 device->io_align = root->sectorsize;
1500 device->sector_size = root->sectorsize;
1501 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1502 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1503 device->dev_root = root->fs_info->dev_root;
1504 device->bdev = bdev;
dfe25020 1505 device->in_fs_metadata = 1;
15916de8 1506 device->mode = 0;
2b82032c 1507 set_blocksize(device->bdev, 4096);
788f20eb 1508
2b82032c
YZ		 1509 if (seeding_dev) {
1510 sb->s_flags &= ~MS_RDONLY;
1511 ret = btrfs_prepare_sprout(trans, root);
1512 BUG_ON(ret);
1513 }
788f20eb 1514
2b82032c 1515 device->fs_devices = root->fs_info->fs_devices;
e5e9a520
CM		 1516
1517 /*
1518 * we don't want write_supers to jump in here with our device
1519 * half setup
1520 */
1521 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2b82032c
YZ		 1522 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1523 list_add(&device->dev_alloc_list,
1524 &root->fs_info->fs_devices->alloc_list);
1525 root->fs_info->fs_devices->num_devices++;
1526 root->fs_info->fs_devices->open_devices++;
1527 root->fs_info->fs_devices->rw_devices++;
1528 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1529
c289811c
CM		 1530 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1531 root->fs_info->fs_devices->rotating = 1;
1532
788f20eb
CM		 1533 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1534 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1535 total_bytes + device->total_bytes);
1536
1537 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1538 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1539 total_bytes + 1);
e5e9a520 1540 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 1541
2b82032c
YZ		 1542 if (seeding_dev) {
1543 ret = init_first_rw_device(trans, root, device);
1544 BUG_ON(ret);
1545 ret = btrfs_finish_sprout(trans, root);
1546 BUG_ON(ret);
1547 } else {
1548 ret = btrfs_add_device(trans, root, device);
1549 }
1550
913d952e
CM		 1551 /*
1552 * we've got more storage, clear any full flags on the space
1553 * infos
1554 */
1555 btrfs_clear_space_info_full(root->fs_info);
1556
7d9eb12c 1557 unlock_chunks(root);
2b82032c 1558 btrfs_commit_transaction(trans, root);
a2135011 1559
2b82032c
YZ		 1560 if (seeding_dev) {
1561 mutex_unlock(&uuid_mutex);
1562 up_write(&sb->s_umount);
788f20eb 1563
2b82032c
YZ		 1564 ret = btrfs_relocate_sys_chunks(root);
1565 BUG_ON(ret);
1566 }
1567out:
1568 mutex_unlock(&root->fs_info->volume_mutex);
1569 return ret;
1570error:
15916de8 1571 close_bdev_exclusive(bdev, 0);
2b82032c
YZ		 1572 if (seeding_dev) {
1573 mutex_unlock(&uuid_mutex);
1574 up_write(&sb->s_umount);
1575 }
788f20eb
CM		 1576 goto out;
1577}
1578
d397712b
CM		 1579static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1580 struct btrfs_device *device)
0b86a832
CM		 1581{
1582 int ret;
1583 struct btrfs_path *path;
1584 struct btrfs_root *root;
1585 struct btrfs_dev_item *dev_item;
1586 struct extent_buffer *leaf;
1587 struct btrfs_key key;
1588
1589 root = device->dev_root->fs_info->chunk_root;
1590
1591 path = btrfs_alloc_path();
1592 if (!path)
1593 return -ENOMEM;
1594
1595 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1596 key.type = BTRFS_DEV_ITEM_KEY;
1597 key.offset = device->devid;
1598
1599 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1600 if (ret < 0)
1601 goto out;
1602
1603 if (ret > 0) {
1604 ret = -ENOENT;
1605 goto out;
1606 }
1607
1608 leaf = path->nodes[0];
1609 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1610
1611 btrfs_set_device_id(leaf, dev_item, device->devid);
1612 btrfs_set_device_type(leaf, dev_item, device->type);
1613 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1614 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1615 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 1616 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 1617 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1618 btrfs_mark_buffer_dirty(leaf);
1619
1620out:
1621 btrfs_free_path(path);
1622 return ret;
1623}
1624
7d9eb12c 1625static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1626 struct btrfs_device *device, u64 new_size)
1627{
1628 struct btrfs_super_block *super_copy =
1629 &device->dev_root->fs_info->super_copy;
1630 u64 old_total = btrfs_super_total_bytes(super_copy);
1631 u64 diff = new_size - device->total_bytes;
1632
2b82032c
YZ		 1633 if (!device->writeable)
1634 return -EACCES;
1635 if (new_size <= device->total_bytes)
1636 return -EINVAL;
1637
8f18cf13 1638 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1639 device->fs_devices->total_rw_bytes += diff;
1640
1641 device->total_bytes = new_size;
9779b72f 1642 device->disk_total_bytes = new_size;
4184ea7f
CM		 1643 btrfs_clear_space_info_full(device->dev_root->fs_info);
1644
8f18cf13
CM		 1645 return btrfs_update_device(trans, device);
1646}
1647
7d9eb12c
CM		 1648int btrfs_grow_device(struct btrfs_trans_handle *trans,
1649 struct btrfs_device *device, u64 new_size)
1650{
1651 int ret;
1652 lock_chunks(device->dev_root);
1653 ret = __btrfs_grow_device(trans, device, new_size);
1654 unlock_chunks(device->dev_root);
1655 return ret;
1656}
1657
8f18cf13
CM		 1658static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root,
1660 u64 chunk_tree, u64 chunk_objectid,
1661 u64 chunk_offset)
1662{
1663 int ret;
1664 struct btrfs_path *path;
1665 struct btrfs_key key;
1666
1667 root = root->fs_info->chunk_root;
1668 path = btrfs_alloc_path();
1669 if (!path)
1670 return -ENOMEM;
1671
1672 key.objectid = chunk_objectid;
1673 key.offset = chunk_offset;
1674 key.type = BTRFS_CHUNK_ITEM_KEY;
1675
1676 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1677 BUG_ON(ret);
1678
1679 ret = btrfs_del_item(trans, root, path);
1680 BUG_ON(ret);
1681
1682 btrfs_free_path(path);
1683 return 0;
1684}
1685
b2950863 1686static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1687 chunk_offset)
1688{
1689 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1690 struct btrfs_disk_key *disk_key;
1691 struct btrfs_chunk *chunk;
1692 u8 *ptr;
1693 int ret = 0;
1694 u32 num_stripes;
1695 u32 array_size;
1696 u32 len = 0;
1697 u32 cur;
1698 struct btrfs_key key;
1699
1700 array_size = btrfs_super_sys_array_size(super_copy);
1701
1702 ptr = super_copy->sys_chunk_array;
1703 cur = 0;
1704
1705 while (cur < array_size) {
1706 disk_key = (struct btrfs_disk_key *)ptr;
1707 btrfs_disk_key_to_cpu(&key, disk_key);
1708
1709 len = sizeof(*disk_key);
1710
1711 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1712 chunk = (struct btrfs_chunk *)(ptr + len);
1713 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1714 len += btrfs_chunk_item_size(num_stripes);
1715 } else {
1716 ret = -EIO;
1717 break;
1718 }
1719 if (key.objectid == chunk_objectid &&
1720 key.offset == chunk_offset) {
1721 memmove(ptr, ptr + len, array_size - (cur + len));
1722 array_size -= len;
1723 btrfs_set_super_sys_array_size(super_copy, array_size);
1724 } else {
1725 ptr += len;
1726 cur += len;
1727 }
1728 }
1729 return ret;
1730}
1731
b2950863 1732static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1733 u64 chunk_tree, u64 chunk_objectid,
1734 u64 chunk_offset)
1735{
1736 struct extent_map_tree *em_tree;
1737 struct btrfs_root *extent_root;
1738 struct btrfs_trans_handle *trans;
1739 struct extent_map *em;
1740 struct map_lookup *map;
1741 int ret;
1742 int i;
1743
1744 root = root->fs_info->chunk_root;
1745 extent_root = root->fs_info->extent_root;
1746 em_tree = &root->fs_info->mapping_tree.map_tree;
1747
ba1bf481
JB		 1748 ret = btrfs_can_relocate(extent_root, chunk_offset);
1749 if (ret)
1750 return -ENOSPC;
1751
8f18cf13 1752 /* step one, relocate all the extents inside this chunk */
1a40e23b 1753 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
a22285a6
YZ		 1754 if (ret)
1755 return ret;
8f18cf13 1756
a22285a6 1757 trans = btrfs_start_transaction(root, 0);
8f18cf13
CM		 1758 BUG_ON(!trans);
1759
7d9eb12c
CM		 1760 lock_chunks(root);
1761
8f18cf13
CM		 1762 /*
1763 * step two, delete the device extents and the
1764 * chunk tree entries
1765 */
890871be 1766 read_lock(&em_tree->lock);
8f18cf13 1767 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 1768 read_unlock(&em_tree->lock);
8f18cf13 1769
a061fc8d
CM		 1770 BUG_ON(em->start > chunk_offset ||
1771 em->start + em->len < chunk_offset);
8f18cf13
CM		 1772 map = (struct map_lookup *)em->bdev;
1773
1774 for (i = 0; i < map->num_stripes; i++) {
1775 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1776 map->stripes[i].physical);
1777 BUG_ON(ret);
a061fc8d 1778
dfe25020
CM		 1779 if (map->stripes[i].dev) {
1780 ret = btrfs_update_device(trans, map->stripes[i].dev);
1781 BUG_ON(ret);
1782 }
8f18cf13
CM		 1783 }
1784 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1785 chunk_offset);
1786
1787 BUG_ON(ret);
1788
1789 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1790 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1791 BUG_ON(ret);
8f18cf13
CM		 1792 }
1793
2b82032c
YZ		 1794 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1795 BUG_ON(ret);
1796
890871be 1797 write_lock(&em_tree->lock);
2b82032c 1798 remove_extent_mapping(em_tree, em);
890871be 1799 write_unlock(&em_tree->lock);
2b82032c
YZ		 1800
1801 kfree(map);
1802 em->bdev = NULL;
1803
1804 /* once for the tree */
1805 free_extent_map(em);
1806 /* once for us */
1807 free_extent_map(em);
1808
1809 unlock_chunks(root);
1810 btrfs_end_transaction(trans, root);
1811 return 0;
1812}
1813
1814static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1815{
1816 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1817 struct btrfs_path *path;
1818 struct extent_buffer *leaf;
1819 struct btrfs_chunk *chunk;
1820 struct btrfs_key key;
1821 struct btrfs_key found_key;
1822 u64 chunk_tree = chunk_root->root_key.objectid;
1823 u64 chunk_type;
ba1bf481
JB		 1824 bool retried = false;
1825 int failed = 0;
2b82032c
YZ		 1826 int ret;
1827
1828 path = btrfs_alloc_path();
1829 if (!path)
1830 return -ENOMEM;
1831
ba1bf481 1832again:
2b82032c
YZ		 1833 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1834 key.offset = (u64)-1;
1835 key.type = BTRFS_CHUNK_ITEM_KEY;
1836
1837 while (1) {
1838 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1839 if (ret < 0)
1840 goto error;
1841 BUG_ON(ret == 0);
1842
1843 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1844 key.type);
1845 if (ret < 0)
1846 goto error;
1847 if (ret > 0)
1848 break;
1a40e23b 1849
2b82032c
YZ		 1850 leaf = path->nodes[0];
1851 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 1852
2b82032c
YZ		 1853 chunk = btrfs_item_ptr(leaf, path->slots[0],
1854 struct btrfs_chunk);
1855 chunk_type = btrfs_chunk_type(leaf, chunk);
1856 btrfs_release_path(chunk_root, path);
8f18cf13 1857
2b82032c
YZ		 1858 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1859 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1860 found_key.objectid,
1861 found_key.offset);
ba1bf481
JB		 1862 if (ret == -ENOSPC)
1863 failed++;
1864 else if (ret)
1865 BUG();
2b82032c 1866 }
8f18cf13 1867
2b82032c
YZ		 1868 if (found_key.offset == 0)
1869 break;
1870 key.offset = found_key.offset - 1;
1871 }
1872 ret = 0;
ba1bf481
JB		 1873 if (failed && !retried) {
1874 failed = 0;
1875 retried = true;
1876 goto again;
1877 } else if (failed && retried) {
1878 WARN_ON(1);
1879 ret = -ENOSPC;
1880 }
2b82032c
YZ		 1881error:
1882 btrfs_free_path(path);
1883 return ret;
8f18cf13
CM		 1884}
1885
ec44a35c
CM		 1886static u64 div_factor(u64 num, int factor)
1887{
1888 if (factor == 10)
1889 return num;
1890 num *= factor;
1891 do_div(num, 10);
1892 return num;
1893}
1894
ec44a35c
CM		 1895int btrfs_balance(struct btrfs_root *dev_root)
1896{
1897 int ret;
ec44a35c
CM		 1898 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1899 struct btrfs_device *device;
1900 u64 old_size;
1901 u64 size_to_free;
1902 struct btrfs_path *path;
1903 struct btrfs_key key;
1904 struct btrfs_chunk *chunk;
1905 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1906 struct btrfs_trans_handle *trans;
1907 struct btrfs_key found_key;
1908
2b82032c
YZ		 1909 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1910 return -EROFS;
ec44a35c 1911
7d9eb12c 1912 mutex_lock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1913 dev_root = dev_root->fs_info->dev_root;
1914
ec44a35c 1915 /* step one make some room on all the devices */
c6e30871 1916 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 1917 old_size = device->total_bytes;
1918 size_to_free = div_factor(old_size, 1);
1919 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 1920 if (!device->writeable ||
1921 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 1922 continue;
1923
1924 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 1925 if (ret == -ENOSPC)
1926 break;
ec44a35c
CM		 1927 BUG_ON(ret);
1928
a22285a6 1929 trans = btrfs_start_transaction(dev_root, 0);
ec44a35c
CM		 1930 BUG_ON(!trans);
1931
1932 ret = btrfs_grow_device(trans, device, old_size);
1933 BUG_ON(ret);
1934
1935 btrfs_end_transaction(trans, dev_root);
1936 }
1937
1938 /* step two, relocate all the chunks */
1939 path = btrfs_alloc_path();
1940 BUG_ON(!path);
1941
1942 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1943 key.offset = (u64)-1;
1944 key.type = BTRFS_CHUNK_ITEM_KEY;
1945
d397712b 1946 while (1) {
ec44a35c
CM		 1947 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1948 if (ret < 0)
1949 goto error;
1950
1951 /*
1952 * this shouldn't happen, it means the last relocate
1953 * failed
1954 */
1955 if (ret == 0)
1956 break;
1957
1958 ret = btrfs_previous_item(chunk_root, path, 0,
1959 BTRFS_CHUNK_ITEM_KEY);
7d9eb12c 1960 if (ret)
ec44a35c 1961 break;
7d9eb12c 1962
ec44a35c
CM		 1963 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1964 path->slots[0]);
1965 if (found_key.objectid != key.objectid)
1966 break;
7d9eb12c 1967
ec44a35c
CM		 1968 chunk = btrfs_item_ptr(path->nodes[0],
1969 path->slots[0],
1970 struct btrfs_chunk);
ec44a35c 1971 /* chunk zero is special */
ba1bf481 1972 if (found_key.offset == 0)
ec44a35c
CM		 1973 break;
1974
7d9eb12c 1975 btrfs_release_path(chunk_root, path);
ec44a35c
CM		 1976 ret = btrfs_relocate_chunk(chunk_root,
1977 chunk_root->root_key.objectid,
1978 found_key.objectid,
1979 found_key.offset);
ba1bf481
JB		 1980 BUG_ON(ret && ret != -ENOSPC);
1981 key.offset = found_key.offset - 1;
ec44a35c
CM		 1982 }
1983 ret = 0;
1984error:
1985 btrfs_free_path(path);
7d9eb12c 1986 mutex_unlock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1987 return ret;
1988}
1989
8f18cf13
CM		 1990/*
1991 * shrinking a device means finding all of the device extents past
1992 * the new size, and then following the back refs to the chunks.
1993 * The chunk relocation code actually frees the device extent
1994 */
1995int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1996{
1997 struct btrfs_trans_handle *trans;
1998 struct btrfs_root *root = device->dev_root;
1999 struct btrfs_dev_extent *dev_extent = NULL;
2000 struct btrfs_path *path;
2001 u64 length;
2002 u64 chunk_tree;
2003 u64 chunk_objectid;
2004 u64 chunk_offset;
2005 int ret;
2006 int slot;
ba1bf481
JB		 2007 int failed = 0;
2008 bool retried = false;
8f18cf13
CM		 2009 struct extent_buffer *l;
2010 struct btrfs_key key;
2011 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2012 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 2013 u64 old_size = device->total_bytes;
8f18cf13
CM		 2014 u64 diff = device->total_bytes - new_size;
2015
2b82032c
YZ		 2016 if (new_size >= device->total_bytes)
2017 return -EINVAL;
8f18cf13
CM		 2018
2019 path = btrfs_alloc_path();
2020 if (!path)
2021 return -ENOMEM;
2022
8f18cf13
CM		 2023 path->reada = 2;
2024
7d9eb12c
CM		 2025 lock_chunks(root);
2026
8f18cf13 2027 device->total_bytes = new_size;
2b82032c
YZ		 2028 if (device->writeable)
2029 device->fs_devices->total_rw_bytes -= diff;
7d9eb12c 2030 unlock_chunks(root);
8f18cf13 2031
ba1bf481 2032again:
8f18cf13
CM		 2033 key.objectid = device->devid;
2034 key.offset = (u64)-1;
2035 key.type = BTRFS_DEV_EXTENT_KEY;
2036
2037 while (1) {
2038 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2039 if (ret < 0)
2040 goto done;
2041
2042 ret = btrfs_previous_item(root, path, 0, key.type);
2043 if (ret < 0)
2044 goto done;
2045 if (ret) {
2046 ret = 0;
ba1bf481 2047 btrfs_release_path(root, path);
bf1fb512 2048 break;
8f18cf13
CM		 2049 }
2050
2051 l = path->nodes[0];
2052 slot = path->slots[0];
2053 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2054
ba1bf481
JB		 2055 if (key.objectid != device->devid) {
2056 btrfs_release_path(root, path);
bf1fb512 2057 break;
ba1bf481 2058 }
8f18cf13
CM		 2059
2060 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2061 length = btrfs_dev_extent_length(l, dev_extent);
2062
ba1bf481
JB		 2063 if (key.offset + length <= new_size) {
2064 btrfs_release_path(root, path);
d6397bae 2065 break;
ba1bf481 2066 }
8f18cf13
CM		 2067
2068 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2069 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2070 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2071 btrfs_release_path(root, path);
2072
2073 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2074 chunk_offset);
ba1bf481 2075 if (ret && ret != -ENOSPC)
8f18cf13 2076 goto done;
ba1bf481
JB		 2077 if (ret == -ENOSPC)
2078 failed++;
2079 key.offset -= 1;
2080 }
2081
2082 if (failed && !retried) {
2083 failed = 0;
2084 retried = true;
2085 goto again;
2086 } else if (failed && retried) {
2087 ret = -ENOSPC;
2088 lock_chunks(root);
2089
2090 device->total_bytes = old_size;
2091 if (device->writeable)
2092 device->fs_devices->total_rw_bytes += diff;
2093 unlock_chunks(root);
2094 goto done;
8f18cf13
CM		 2095 }
2096
d6397bae 2097 /* Shrinking succeeded, else we would be at "done". */
a22285a6 2098 trans = btrfs_start_transaction(root, 0);
d6397bae
CB		 2099 lock_chunks(root);
2100
2101 device->disk_total_bytes = new_size;
2102 /* Now btrfs_update_device() will change the on-disk size. */
2103 ret = btrfs_update_device(trans, device);
2104 if (ret) {
2105 unlock_chunks(root);
2106 btrfs_end_transaction(trans, root);
2107 goto done;
2108 }
2109 WARN_ON(diff > old_total);
2110 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2111 unlock_chunks(root);
2112 btrfs_end_transaction(trans, root);
8f18cf13
CM		 2113done:
2114 btrfs_free_path(path);
2115 return ret;
2116}
2117
b2950863 2118static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2119 struct btrfs_root *root,
2120 struct btrfs_key *key,
2121 struct btrfs_chunk *chunk, int item_size)
2122{
2123 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2124 struct btrfs_disk_key disk_key;
2125 u32 array_size;
2126 u8 *ptr;
2127
2128 array_size = btrfs_super_sys_array_size(super_copy);
2129 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2130 return -EFBIG;
2131
2132 ptr = super_copy->sys_chunk_array + array_size;
2133 btrfs_cpu_key_to_disk(&disk_key, key);
2134 memcpy(ptr, &disk_key, sizeof(disk_key));
2135 ptr += sizeof(disk_key);
2136 memcpy(ptr, chunk, item_size);
2137 item_size += sizeof(disk_key);
2138 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2139 return 0;
2140}
2141
d397712b 2142static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
a1b32a59 2143 int num_stripes, int sub_stripes)
9b3f68b9
CM		 2144{
2145 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2146 return calc_size;
2147 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2148 return calc_size * (num_stripes / sub_stripes);
2149 else
2150 return calc_size * num_stripes;
2151}
2152
2b82032c
YZ		 2153static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2154 struct btrfs_root *extent_root,
2155 struct map_lookup **map_ret,
2156 u64 *num_bytes, u64 *stripe_size,
2157 u64 start, u64 type)
0b86a832 2158{
593060d7 2159 struct btrfs_fs_info *info = extent_root->fs_info;
0b86a832 2160 struct btrfs_device *device = NULL;
2b82032c 2161 struct btrfs_fs_devices *fs_devices = info->fs_devices;
6324fbf3 2162 struct list_head *cur;
2b82032c 2163 struct map_lookup *map = NULL;
0b86a832 2164 struct extent_map_tree *em_tree;
0b86a832 2165 struct extent_map *em;
2b82032c 2166 struct list_head private_devs;
a40a90a0 2167 int min_stripe_size = 1 * 1024 * 1024;
0b86a832 2168 u64 calc_size = 1024 * 1024 * 1024;
9b3f68b9
CM		 2169 u64 max_chunk_size = calc_size;
2170 u64 min_free;
6324fbf3
CM		 2171 u64 avail;
2172 u64 max_avail = 0;
2b82032c 2173 u64 dev_offset;
6324fbf3 2174 int num_stripes = 1;
a40a90a0 2175 int min_stripes = 1;
321aecc6 2176 int sub_stripes = 0;
6324fbf3 2177 int looped = 0;
0b86a832 2178 int ret;
6324fbf3 2179 int index;
593060d7 2180 int stripe_len = 64 * 1024;
0b86a832 2181
ec44a35c
CM		 2182 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2183 (type & BTRFS_BLOCK_GROUP_DUP)) {
2184 WARN_ON(1);
2185 type &= ~BTRFS_BLOCK_GROUP_DUP;
2186 }
2b82032c 2187 if (list_empty(&fs_devices->alloc_list))
6324fbf3 2188 return -ENOSPC;
593060d7 2189
a40a90a0 2190 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2b82032c 2191 num_stripes = fs_devices->rw_devices;
a40a90a0
CM		 2192 min_stripes = 2;
2193 }
2194 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
611f0e00 2195 num_stripes = 2;
a40a90a0
CM		 2196 min_stripes = 2;
2197 }
8790d502 2198 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
f3eae7e8 2199 if (fs_devices->rw_devices < 2)
9b3f68b9 2200 return -ENOSPC;
f3eae7e8 2201 num_stripes = 2;
a40a90a0 2202 min_stripes = 2;
8790d502 2203 }
321aecc6 2204 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2b82032c 2205 num_stripes = fs_devices->rw_devices;
321aecc6
CM		 2206 if (num_stripes < 4)
2207 return -ENOSPC;
2208 num_stripes &= ~(u32)1;
2209 sub_stripes = 2;
a40a90a0 2210 min_stripes = 4;
321aecc6 2211 }
9b3f68b9
CM		 2212
2213 if (type & BTRFS_BLOCK_GROUP_DATA) {
2214 max_chunk_size = 10 * calc_size;
a40a90a0 2215 min_stripe_size = 64 * 1024 * 1024;
9b3f68b9 2216 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
83d3c969 2217 max_chunk_size = 256 * 1024 * 1024;
a40a90a0
CM		 2218 min_stripe_size = 32 * 1024 * 1024;
2219 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2220 calc_size = 8 * 1024 * 1024;
2221 max_chunk_size = calc_size * 2;
2222 min_stripe_size = 1 * 1024 * 1024;
9b3f68b9
CM		 2223 }
2224
2b82032c
YZ		 2225 /* we don't want a chunk larger than 10% of writeable space */
2226 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2227 max_chunk_size);
9b3f68b9 2228
a40a90a0 2229again:
9779b72f 2230 max_avail = 0;
2b82032c
YZ		 2231 if (!map || map->num_stripes != num_stripes) {
2232 kfree(map);
2233 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2234 if (!map)
2235 return -ENOMEM;
2236 map->num_stripes = num_stripes;
2237 }
2238
9b3f68b9
CM		 2239 if (calc_size * num_stripes > max_chunk_size) {
2240 calc_size = max_chunk_size;
2241 do_div(calc_size, num_stripes);
2242 do_div(calc_size, stripe_len);
2243 calc_size *= stripe_len;
2244 }
0cad8a11 2245
9b3f68b9 2246 /* we don't want tiny stripes */
0cad8a11
JB		 2247 if (!looped)
2248 calc_size = max_t(u64, min_stripe_size, calc_size);
9b3f68b9 2249
9f680ce0
CM		 2250 /*
2251 * we're about to do_div by the stripe_len so lets make sure
2252 * we end up with something bigger than a stripe
2253 */
2254 calc_size = max_t(u64, calc_size, stripe_len * 4);
2255
9b3f68b9
CM		 2256 do_div(calc_size, stripe_len);
2257 calc_size *= stripe_len;
2258
2b82032c 2259 cur = fs_devices->alloc_list.next;
6324fbf3 2260 index = 0;
611f0e00
CM		 2261
2262 if (type & BTRFS_BLOCK_GROUP_DUP)
2263 min_free = calc_size * 2;
9b3f68b9
CM		 2264 else
2265 min_free = calc_size;
611f0e00 2266
0f9dd46c
JB		 2267 /*
2268 * we add 1MB because we never use the first 1MB of the device, unless
2269 * we've looped, then we are likely allocating the maximum amount of
2270 * space left already
2271 */
2272 if (!looped)
2273 min_free += 1024 * 1024;
ad5bd91e 2274
2b82032c 2275 INIT_LIST_HEAD(&private_devs);
d397712b 2276 while (index < num_stripes) {
b3075717 2277 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2b82032c 2278 BUG_ON(!device->writeable);
dfe25020
CM		 2279 if (device->total_bytes > device->bytes_used)
2280 avail = device->total_bytes - device->bytes_used;
2281 else
2282 avail = 0;
6324fbf3 2283 cur = cur->next;
8f18cf13 2284
dfe25020 2285 if (device->in_fs_metadata && avail >= min_free) {
2b82032c 2286 ret = find_free_dev_extent(trans, device,
9779b72f
CM		 2287 min_free, &dev_offset,
2288 &max_avail);
8f18cf13
CM		 2289 if (ret == 0) {
2290 list_move_tail(&device->dev_alloc_list,
2291 &private_devs);
2b82032c
YZ		 2292 map->stripes[index].dev = device;
2293 map->stripes[index].physical = dev_offset;
611f0e00 2294 index++;
2b82032c
YZ		 2295 if (type & BTRFS_BLOCK_GROUP_DUP) {
2296 map->stripes[index].dev = device;
2297 map->stripes[index].physical =
2298 dev_offset + calc_size;
8f18cf13 2299 index++;
2b82032c 2300 }
8f18cf13 2301 }
dfe25020 2302 } else if (device->in_fs_metadata && avail > max_avail)
a40a90a0 2303 max_avail = avail;
2b82032c 2304 if (cur == &fs_devices->alloc_list)
6324fbf3
CM		 2305 break;
2306 }
2b82032c 2307 list_splice(&private_devs, &fs_devices->alloc_list);
6324fbf3 2308 if (index < num_stripes) {
a40a90a0
CM		 2309 if (index >= min_stripes) {
2310 num_stripes = index;
2311 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2312 num_stripes /= sub_stripes;
2313 num_stripes *= sub_stripes;
2314 }
2315 looped = 1;
2316 goto again;
2317 }
6324fbf3
CM		 2318 if (!looped && max_avail > 0) {
2319 looped = 1;
2320 calc_size = max_avail;
2321 goto again;
2322 }
2b82032c 2323 kfree(map);
6324fbf3
CM		 2324 return -ENOSPC;
2325 }
2b82032c
YZ		 2326 map->sector_size = extent_root->sectorsize;
2327 map->stripe_len = stripe_len;
2328 map->io_align = stripe_len;
2329 map->io_width = stripe_len;
2330 map->type = type;
2331 map->num_stripes = num_stripes;
2332 map->sub_stripes = sub_stripes;
0b86a832 2333
2b82032c
YZ		 2334 *map_ret = map;
2335 *stripe_size = calc_size;
2336 *num_bytes = chunk_bytes_by_type(type, calc_size,
2337 num_stripes, sub_stripes);
0b86a832 2338
2b82032c
YZ		 2339 em = alloc_extent_map(GFP_NOFS);
2340 if (!em) {
2341 kfree(map);
593060d7
CM		 2342 return -ENOMEM;
2343 }
2b82032c
YZ		 2344 em->bdev = (struct block_device *)map;
2345 em->start = start;
2346 em->len = *num_bytes;
2347 em->block_start = 0;
2348 em->block_len = em->len;
593060d7 2349
2b82032c 2350 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 2351 write_lock(&em_tree->lock);
2b82032c 2352 ret = add_extent_mapping(em_tree, em);
890871be 2353 write_unlock(&em_tree->lock);
2b82032c
YZ		 2354 BUG_ON(ret);
2355 free_extent_map(em);
0b86a832 2356
2b82032c
YZ		 2357 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2358 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2359 start, *num_bytes);
2360 BUG_ON(ret);
611f0e00 2361
2b82032c
YZ		 2362 index = 0;
2363 while (index < map->num_stripes) {
2364 device = map->stripes[index].dev;
2365 dev_offset = map->stripes[index].physical;
0b86a832
CM		 2366
2367 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2368 info->chunk_root->root_key.objectid,
2369 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2370 start, dev_offset, calc_size);
0b86a832 2371 BUG_ON(ret);
2b82032c
YZ		 2372 index++;
2373 }
2374
2375 return 0;
2376}
2377
2378static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2379 struct btrfs_root *extent_root,
2380 struct map_lookup *map, u64 chunk_offset,
2381 u64 chunk_size, u64 stripe_size)
2382{
2383 u64 dev_offset;
2384 struct btrfs_key key;
2385 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2386 struct btrfs_device *device;
2387 struct btrfs_chunk *chunk;
2388 struct btrfs_stripe *stripe;
2389 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2390 int index = 0;
2391 int ret;
2392
2393 chunk = kzalloc(item_size, GFP_NOFS);
2394 if (!chunk)
2395 return -ENOMEM;
2396
2397 index = 0;
2398 while (index < map->num_stripes) {
2399 device = map->stripes[index].dev;
2400 device->bytes_used += stripe_size;
0b86a832
CM		 2401 ret = btrfs_update_device(trans, device);
2402 BUG_ON(ret);
2b82032c
YZ		 2403 index++;
2404 }
2405
2406 index = 0;
2407 stripe = &chunk->stripe;
2408 while (index < map->num_stripes) {
2409 device = map->stripes[index].dev;
2410 dev_offset = map->stripes[index].physical;
0b86a832 2411
e17cade2
CM		 2412 btrfs_set_stack_stripe_devid(stripe, device->devid);
2413 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2414 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2415 stripe++;
0b86a832
CM		 2416 index++;
2417 }
2418
2b82032c 2419 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2420 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2421 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2422 btrfs_set_stack_chunk_type(chunk, map->type);
2423 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2424 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2425 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2426 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2427 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2428
2b82032c
YZ		 2429 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2430 key.type = BTRFS_CHUNK_ITEM_KEY;
2431 key.offset = chunk_offset;
0b86a832 2432
2b82032c
YZ		 2433 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2434 BUG_ON(ret);
0b86a832 2435
2b82032c
YZ		 2436 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2437 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2438 item_size);
8f18cf13
CM		 2439 BUG_ON(ret);
2440 }
0b86a832 2441 kfree(chunk);
2b82032c
YZ		 2442 return 0;
2443}
0b86a832 2444
2b82032c
YZ		 2445/*
2446 * Chunk allocation falls into two parts. The first part does works
2447 * that make the new allocated chunk useable, but not do any operation
2448 * that modifies the chunk tree. The second part does the works that
2449 * require modifying the chunk tree. This division is important for the
2450 * bootstrap process of adding storage to a seed btrfs.
2451 */
2452int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2453 struct btrfs_root *extent_root, u64 type)
2454{
2455 u64 chunk_offset;
2456 u64 chunk_size;
2457 u64 stripe_size;
2458 struct map_lookup *map;
2459 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2460 int ret;
2461
2462 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2463 &chunk_offset);
2464 if (ret)
2465 return ret;
2466
2467 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2468 &stripe_size, chunk_offset, type);
2469 if (ret)
2470 return ret;
2471
2472 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2473 chunk_size, stripe_size);
2474 BUG_ON(ret);
2475 return 0;
2476}
2477
d397712b 2478static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2479 struct btrfs_root *root,
2480 struct btrfs_device *device)
2481{
2482 u64 chunk_offset;
2483 u64 sys_chunk_offset;
2484 u64 chunk_size;
2485 u64 sys_chunk_size;
2486 u64 stripe_size;
2487 u64 sys_stripe_size;
2488 u64 alloc_profile;
2489 struct map_lookup *map;
2490 struct map_lookup *sys_map;
2491 struct btrfs_fs_info *fs_info = root->fs_info;
2492 struct btrfs_root *extent_root = fs_info->extent_root;
2493 int ret;
2494
2495 ret = find_next_chunk(fs_info->chunk_root,
2496 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2497 BUG_ON(ret);
2498
2499 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2500 (fs_info->metadata_alloc_profile &
2501 fs_info->avail_metadata_alloc_bits);
2502 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2503
2504 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2505 &stripe_size, chunk_offset, alloc_profile);
2506 BUG_ON(ret);
2507
2508 sys_chunk_offset = chunk_offset + chunk_size;
2509
2510 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2511 (fs_info->system_alloc_profile &
2512 fs_info->avail_system_alloc_bits);
2513 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2514
2515 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2516 &sys_chunk_size, &sys_stripe_size,
2517 sys_chunk_offset, alloc_profile);
2518 BUG_ON(ret);
2519
2520 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2521 BUG_ON(ret);
2522
2523 /*
2524 * Modifying chunk tree needs allocating new blocks from both
2525 * system block group and metadata block group. So we only can
2526 * do operations require modifying the chunk tree after both
2527 * block groups were created.
2528 */
2529 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2530 chunk_size, stripe_size);
2531 BUG_ON(ret);
2532
2533 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2534 sys_chunk_offset, sys_chunk_size,
2535 sys_stripe_size);
b248a415 2536 BUG_ON(ret);
2b82032c
YZ		 2537 return 0;
2538}
2539
2540int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2541{
2542 struct extent_map *em;
2543 struct map_lookup *map;
2544 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2545 int readonly = 0;
2546 int i;
2547
890871be 2548 read_lock(&map_tree->map_tree.lock);
2b82032c 2549 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 2550 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 2551 if (!em)
2552 return 1;
2553
f48b9075
JB		 2554 if (btrfs_test_opt(root, DEGRADED)) {
2555 free_extent_map(em);
2556 return 0;
2557 }
2558
2b82032c
YZ		 2559 map = (struct map_lookup *)em->bdev;
2560 for (i = 0; i < map->num_stripes; i++) {
2561 if (!map->stripes[i].dev->writeable) {
2562 readonly = 1;
2563 break;
2564 }
2565 }
0b86a832 2566 free_extent_map(em);
2b82032c 2567 return readonly;
0b86a832
CM		 2568}
2569
2570void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2571{
2572 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2573}
2574
2575void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2576{
2577 struct extent_map *em;
2578
d397712b 2579 while (1) {
890871be 2580 write_lock(&tree->map_tree.lock);
0b86a832
CM		 2581 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2582 if (em)
2583 remove_extent_mapping(&tree->map_tree, em);
890871be 2584 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 2585 if (!em)
2586 break;
2587 kfree(em->bdev);
2588 /* once for us */
2589 free_extent_map(em);
2590 /* once for the tree */
2591 free_extent_map(em);
2592 }
2593}
2594
f188591e
CM		 2595int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2596{
2597 struct extent_map *em;
2598 struct map_lookup *map;
2599 struct extent_map_tree *em_tree = &map_tree->map_tree;
2600 int ret;
2601
890871be 2602 read_lock(&em_tree->lock);
f188591e 2603 em = lookup_extent_mapping(em_tree, logical, len);
890871be 2604 read_unlock(&em_tree->lock);
f188591e
CM		 2605 BUG_ON(!em);
2606
2607 BUG_ON(em->start > logical || em->start + em->len < logical);
2608 map = (struct map_lookup *)em->bdev;
2609 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2610 ret = map->num_stripes;
321aecc6
CM		 2611 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2612 ret = map->sub_stripes;
f188591e
CM		 2613 else
2614 ret = 1;
2615 free_extent_map(em);
f188591e
CM		 2616 return ret;
2617}
2618
dfe25020
CM		 2619static int find_live_mirror(struct map_lookup *map, int first, int num,
2620 int optimal)
2621{
2622 int i;
2623 if (map->stripes[optimal].dev->bdev)
2624 return optimal;
2625 for (i = first; i < first + num; i++) {
2626 if (map->stripes[i].dev->bdev)
2627 return i;
2628 }
2629 /* we couldn't find one that doesn't fail. Just return something
2630 * and the io error handling code will clean up eventually
2631 */
2632 return optimal;
2633}
2634
f2d8d74d
CM		 2635static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2636 u64 logical, u64 *length,
2637 struct btrfs_multi_bio **multi_ret,
2638 int mirror_num, struct page *unplug_page)
0b86a832
CM		 2639{
2640 struct extent_map *em;
2641 struct map_lookup *map;
2642 struct extent_map_tree *em_tree = &map_tree->map_tree;
2643 u64 offset;
593060d7
CM		 2644 u64 stripe_offset;
2645 u64 stripe_nr;
cea9e445 2646 int stripes_allocated = 8;
321aecc6 2647 int stripes_required = 1;
593060d7 2648 int stripe_index;
cea9e445 2649 int i;
f2d8d74d 2650 int num_stripes;
a236aed1 2651 int max_errors = 0;
cea9e445 2652 struct btrfs_multi_bio *multi = NULL;
0b86a832 2653
7b6d91da 2654 if (multi_ret && !(rw & REQ_WRITE))
cea9e445 2655 stripes_allocated = 1;
cea9e445
CM		 2656again:
2657 if (multi_ret) {
2658 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2659 GFP_NOFS);
2660 if (!multi)
2661 return -ENOMEM;
a236aed1
CM		 2662
2663 atomic_set(&multi->error, 0);
cea9e445 2664 }
0b86a832 2665
890871be 2666 read_lock(&em_tree->lock);
0b86a832 2667 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 2668 read_unlock(&em_tree->lock);
f2d8d74d 2669
2423fdfb
JS		 2670 if (!em && unplug_page) {
2671 kfree(multi);
f2d8d74d 2672 return 0;
2423fdfb 2673 }
f2d8d74d 2674
3b951516 2675 if (!em) {
d397712b
CM		 2676 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2677 (unsigned long long)logical,
2678 (unsigned long long)*length);
f2d8d74d 2679 BUG();
3b951516 2680 }
0b86a832
CM		 2681
2682 BUG_ON(em->start > logical || em->start + em->len < logical);
2683 map = (struct map_lookup *)em->bdev;
2684 offset = logical - em->start;
593060d7 2685
f188591e
CM		 2686 if (mirror_num > map->num_stripes)
2687 mirror_num = 0;
2688
cea9e445 2689 /* if our multi bio struct is too small, back off and try again */
7b6d91da 2690 if (rw & REQ_WRITE) {
321aecc6
CM		 2691 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2692 BTRFS_BLOCK_GROUP_DUP)) {
2693 stripes_required = map->num_stripes;
a236aed1 2694 max_errors = 1;
321aecc6
CM		 2695 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2696 stripes_required = map->sub_stripes;
a236aed1 2697 max_errors = 1;
321aecc6
CM		 2698 }
2699 }
7b6d91da 2700 if (multi_ret && (rw & REQ_WRITE) &&
321aecc6 2701 stripes_allocated < stripes_required) {
cea9e445 2702 stripes_allocated = map->num_stripes;
cea9e445
CM		 2703 free_extent_map(em);
2704 kfree(multi);
2705 goto again;
2706 }
593060d7
CM		 2707 stripe_nr = offset;
2708 /*
2709 * stripe_nr counts the total number of stripes we have to stride
2710 * to get to this block
2711 */
2712 do_div(stripe_nr, map->stripe_len);
2713
2714 stripe_offset = stripe_nr * map->stripe_len;
2715 BUG_ON(offset < stripe_offset);
2716
2717 /* stripe_offset is the offset of this block in its stripe*/
2718 stripe_offset = offset - stripe_offset;
2719
cea9e445 2720 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 2721 BTRFS_BLOCK_GROUP_RAID10 |
cea9e445
CM		 2722 BTRFS_BLOCK_GROUP_DUP)) {
2723 /* we limit the length of each bio to what fits in a stripe */
2724 *length = min_t(u64, em->len - offset,
2725 map->stripe_len - stripe_offset);
2726 } else {
2727 *length = em->len - offset;
2728 }
f2d8d74d
CM		 2729
2730 if (!multi_ret && !unplug_page)
cea9e445
CM		 2731 goto out;
2732
f2d8d74d 2733 num_stripes = 1;
cea9e445 2734 stripe_index = 0;
8790d502 2735 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
7b6d91da 2736 if (unplug_page || (rw & REQ_WRITE))
f2d8d74d 2737 num_stripes = map->num_stripes;
2fff734f 2738 else if (mirror_num)
f188591e 2739 stripe_index = mirror_num - 1;
dfe25020
CM		 2740 else {
2741 stripe_index = find_live_mirror(map, 0,
2742 map->num_stripes,
2743 current->pid % map->num_stripes);
2744 }
2fff734f 2745
611f0e00 2746 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
7b6d91da 2747 if (rw & REQ_WRITE)
f2d8d74d 2748 num_stripes = map->num_stripes;
f188591e
CM		 2749 else if (mirror_num)
2750 stripe_index = mirror_num - 1;
2fff734f 2751
321aecc6
CM		 2752 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2753 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 2754
2755 stripe_index = do_div(stripe_nr, factor);
2756 stripe_index *= map->sub_stripes;
2757
7b6d91da 2758 if (unplug_page || (rw & REQ_WRITE))
f2d8d74d 2759 num_stripes = map->sub_stripes;
321aecc6
CM		 2760 else if (mirror_num)
2761 stripe_index += mirror_num - 1;
dfe25020
CM		 2762 else {
2763 stripe_index = find_live_mirror(map, stripe_index,
2764 map->sub_stripes, stripe_index +
2765 current->pid % map->sub_stripes);
2766 }
8790d502
CM		 2767 } else {
2768 /*
2769 * after this do_div call, stripe_nr is the number of stripes
2770 * on this device we have to walk to find the data, and
2771 * stripe_index is the number of our device in the stripe array
2772 */
2773 stripe_index = do_div(stripe_nr, map->num_stripes);
2774 }
593060d7 2775 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 2776
f2d8d74d
CM		 2777 for (i = 0; i < num_stripes; i++) {
2778 if (unplug_page) {
2779 struct btrfs_device *device;
2780 struct backing_dev_info *bdi;
2781
2782 device = map->stripes[stripe_index].dev;
dfe25020
CM		 2783 if (device->bdev) {
2784 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 2785 if (bdi->unplug_io_fn)
dfe25020 2786 bdi->unplug_io_fn(bdi, unplug_page);
f2d8d74d
CM		 2787 }
2788 } else {
2789 multi->stripes[i].physical =
2790 map->stripes[stripe_index].physical +
2791 stripe_offset + stripe_nr * map->stripe_len;
2792 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2793 }
cea9e445 2794 stripe_index++;
593060d7 2795 }
f2d8d74d
CM		 2796 if (multi_ret) {
2797 *multi_ret = multi;
2798 multi->num_stripes = num_stripes;
a236aed1 2799 multi->max_errors = max_errors;
f2d8d74d 2800 }
cea9e445 2801out:
0b86a832 2802 free_extent_map(em);
0b86a832
CM		 2803 return 0;
2804}
2805
f2d8d74d
CM		 2806int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2807 u64 logical, u64 *length,
2808 struct btrfs_multi_bio **multi_ret, int mirror_num)
2809{
2810 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2811 mirror_num, NULL);
2812}
2813
a512bbf8
YZ		 2814int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2815 u64 chunk_start, u64 physical, u64 devid,
2816 u64 **logical, int *naddrs, int *stripe_len)
2817{
2818 struct extent_map_tree *em_tree = &map_tree->map_tree;
2819 struct extent_map *em;
2820 struct map_lookup *map;
2821 u64 *buf;
2822 u64 bytenr;
2823 u64 length;
2824 u64 stripe_nr;
2825 int i, j, nr = 0;
2826
890871be 2827 read_lock(&em_tree->lock);
a512bbf8 2828 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 2829 read_unlock(&em_tree->lock);
a512bbf8
YZ		 2830
2831 BUG_ON(!em || em->start != chunk_start);
2832 map = (struct map_lookup *)em->bdev;
2833
2834 length = em->len;
2835 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2836 do_div(length, map->num_stripes / map->sub_stripes);
2837 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2838 do_div(length, map->num_stripes);
2839
2840 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2841 BUG_ON(!buf);
2842
2843 for (i = 0; i < map->num_stripes; i++) {
2844 if (devid && map->stripes[i].dev->devid != devid)
2845 continue;
2846 if (map->stripes[i].physical > physical ||
2847 map->stripes[i].physical + length <= physical)
2848 continue;
2849
2850 stripe_nr = physical - map->stripes[i].physical;
2851 do_div(stripe_nr, map->stripe_len);
2852
2853 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2854 stripe_nr = stripe_nr * map->num_stripes + i;
2855 do_div(stripe_nr, map->sub_stripes);
2856 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2857 stripe_nr = stripe_nr * map->num_stripes + i;
2858 }
2859 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 2860 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 2861 for (j = 0; j < nr; j++) {
2862 if (buf[j] == bytenr)
2863 break;
2864 }
934d375b
CM		 2865 if (j == nr) {
2866 WARN_ON(nr >= map->num_stripes);
a512bbf8 2867 buf[nr++] = bytenr;
934d375b 2868 }
a512bbf8
YZ		 2869 }
2870
a512bbf8
YZ		 2871 *logical = buf;
2872 *naddrs = nr;
2873 *stripe_len = map->stripe_len;
2874
2875 free_extent_map(em);
2876 return 0;
2877}
2878
f2d8d74d
CM		 2879int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2880 u64 logical, struct page *page)
2881{
2882 u64 length = PAGE_CACHE_SIZE;
2883 return __btrfs_map_block(map_tree, READ, logical, &length,
2884 NULL, 0, page);
2885}
2886
8790d502 2887static void end_bio_multi_stripe(struct bio *bio, int err)
8790d502 2888{
cea9e445 2889 struct btrfs_multi_bio *multi = bio->bi_private;
7d2b4daa 2890 int is_orig_bio = 0;
8790d502 2891
8790d502 2892 if (err)
a236aed1 2893 atomic_inc(&multi->error);
8790d502 2894
7d2b4daa
CM		 2895 if (bio == multi->orig_bio)
2896 is_orig_bio = 1;
2897
cea9e445 2898 if (atomic_dec_and_test(&multi->stripes_pending)) {
7d2b4daa
CM		 2899 if (!is_orig_bio) {
2900 bio_put(bio);
2901 bio = multi->orig_bio;
2902 }
8790d502
CM		 2903 bio->bi_private = multi->private;
2904 bio->bi_end_io = multi->end_io;
a236aed1
CM		 2905 /* only send an error to the higher layers if it is
2906 * beyond the tolerance of the multi-bio
2907 */
1259ab75 2908 if (atomic_read(&multi->error) > multi->max_errors) {
a236aed1 2909 err = -EIO;
1259ab75
CM		 2910 } else if (err) {
2911 /*
2912 * this bio is actually up to date, we didn't
2913 * go over the max number of errors
2914 */
2915 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 2916 err = 0;
1259ab75 2917 }
8790d502
CM		 2918 kfree(multi);
2919
2920 bio_endio(bio, err);
7d2b4daa 2921 } else if (!is_orig_bio) {
8790d502
CM		 2922 bio_put(bio);
2923 }
8790d502
CM		 2924}
2925
8b712842
CM		 2926struct async_sched {
2927 struct bio *bio;
2928 int rw;
2929 struct btrfs_fs_info *info;
2930 struct btrfs_work work;
2931};
2932
2933/*
2934 * see run_scheduled_bios for a description of why bios are collected for
2935 * async submit.
2936 *
2937 * This will add one bio to the pending list for a device and make sure
2938 * the work struct is scheduled.
2939 */
d397712b 2940static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 2941 struct btrfs_device *device,
2942 int rw, struct bio *bio)
8b712842
CM		 2943{
2944 int should_queue = 1;
ffbd517d 2945 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 2946
2947 /* don't bother with additional async steps for reads, right now */
7b6d91da 2948 if (!(rw & REQ_WRITE)) {
492bb6de 2949 bio_get(bio);
8b712842 2950 submit_bio(rw, bio);
492bb6de 2951 bio_put(bio);
8b712842
CM		 2952 return 0;
2953 }
2954
2955 /*
0986fe9e 2956 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 2957 * higher layers. Otherwise, the async bio makes it appear we have
2958 * made progress against dirty pages when we've really just put it
2959 * on a queue for later
2960 */
0986fe9e 2961 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 2962 WARN_ON(bio->bi_next);
8b712842
CM		 2963 bio->bi_next = NULL;
2964 bio->bi_rw |= rw;
2965
2966 spin_lock(&device->io_lock);
7b6d91da 2967 if (bio->bi_rw & REQ_SYNC)
ffbd517d
CM		 2968 pending_bios = &device->pending_sync_bios;
2969 else
2970 pending_bios = &device->pending_bios;
8b712842 2971
ffbd517d
CM		 2972 if (pending_bios->tail)
2973 pending_bios->tail->bi_next = bio;
8b712842 2974
ffbd517d
CM		 2975 pending_bios->tail = bio;
2976 if (!pending_bios->head)
2977 pending_bios->head = bio;
8b712842
CM		 2978 if (device->running_pending)
2979 should_queue = 0;
2980
2981 spin_unlock(&device->io_lock);
2982
2983 if (should_queue)
1cc127b5
CM		 2984 btrfs_queue_worker(&root->fs_info->submit_workers,
2985 &device->work);
8b712842
CM		 2986 return 0;
2987}
2988
f188591e 2989int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 2990 int mirror_num, int async_submit)
0b86a832
CM		 2991{
2992 struct btrfs_mapping_tree *map_tree;
2993 struct btrfs_device *dev;
8790d502 2994 struct bio *first_bio = bio;
a62b9401 2995 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 2996 u64 length = 0;
2997 u64 map_length;
cea9e445 2998 struct btrfs_multi_bio *multi = NULL;
0b86a832 2999 int ret;
8790d502
CM		 3000 int dev_nr = 0;
3001 int total_devs = 1;
0b86a832 3002
f2d8d74d 3003 length = bio->bi_size;
0b86a832
CM		 3004 map_tree = &root->fs_info->mapping_tree;
3005 map_length = length;
cea9e445 3006
f188591e
CM		 3007 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3008 mirror_num);
cea9e445
CM		 3009 BUG_ON(ret);
3010
3011 total_devs = multi->num_stripes;
3012 if (map_length < length) {
d397712b
CM		 3013 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3014 "len %llu\n", (unsigned long long)logical,
3015 (unsigned long long)length,
3016 (unsigned long long)map_length);
cea9e445
CM		 3017 BUG();
3018 }
3019 multi->end_io = first_bio->bi_end_io;
3020 multi->private = first_bio->bi_private;
7d2b4daa 3021 multi->orig_bio = first_bio;
cea9e445
CM		 3022 atomic_set(&multi->stripes_pending, multi->num_stripes);
3023
d397712b 3024 while (dev_nr < total_devs) {
8790d502 3025 if (total_devs > 1) {
8790d502
CM		 3026 if (dev_nr < total_devs - 1) {
3027 bio = bio_clone(first_bio, GFP_NOFS);
3028 BUG_ON(!bio);
3029 } else {
3030 bio = first_bio;
3031 }
3032 bio->bi_private = multi;
3033 bio->bi_end_io = end_bio_multi_stripe;
3034 }
cea9e445
CM		 3035 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3036 dev = multi->stripes[dev_nr].dev;
2b82032c 3037 BUG_ON(rw == WRITE && !dev->writeable);
dfe25020
CM		 3038 if (dev && dev->bdev) {
3039 bio->bi_bdev = dev->bdev;
8b712842
CM		 3040 if (async_submit)
3041 schedule_bio(root, dev, rw, bio);
3042 else
3043 submit_bio(rw, bio);
dfe25020
CM		 3044 } else {
3045 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3046 bio->bi_sector = logical >> 9;
dfe25020 3047 bio_endio(bio, -EIO);
dfe25020 3048 }
8790d502
CM		 3049 dev_nr++;
3050 }
cea9e445
CM		 3051 if (total_devs == 1)
3052 kfree(multi);
0b86a832
CM		 3053 return 0;
3054}
3055
a443755f 3056struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 3057 u8 *uuid, u8 *fsid)
0b86a832 3058{
2b82032c
YZ		 3059 struct btrfs_device *device;
3060 struct btrfs_fs_devices *cur_devices;
3061
3062 cur_devices = root->fs_info->fs_devices;
3063 while (cur_devices) {
3064 if (!fsid ||
3065 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3066 device = __find_device(&cur_devices->devices,
3067 devid, uuid);
3068 if (device)
3069 return device;
3070 }
3071 cur_devices = cur_devices->seed;
3072 }
3073 return NULL;
0b86a832
CM		 3074}
3075
dfe25020
CM		 3076static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3077 u64 devid, u8 *dev_uuid)
3078{
3079 struct btrfs_device *device;
3080 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3081
3082 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 3083 if (!device)
3084 return NULL;
dfe25020
CM		 3085 list_add(&device->dev_list,
3086 &fs_devices->devices);
dfe25020
CM		 3087 device->barriers = 1;
3088 device->dev_root = root->fs_info->dev_root;
3089 device->devid = devid;
8b712842 3090 device->work.func = pending_bios_fn;
e4404d6e 3091 device->fs_devices = fs_devices;
dfe25020
CM		 3092 fs_devices->num_devices++;
3093 spin_lock_init(&device->io_lock);
d20f7043 3094 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 3095 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3096 return device;
3097}
3098
0b86a832
CM		 3099static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3100 struct extent_buffer *leaf,
3101 struct btrfs_chunk *chunk)
3102{
3103 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3104 struct map_lookup *map;
3105 struct extent_map *em;
3106 u64 logical;
3107 u64 length;
3108 u64 devid;
a443755f 3109 u8 uuid[BTRFS_UUID_SIZE];
593060d7 3110 int num_stripes;
0b86a832 3111 int ret;
593060d7 3112 int i;
0b86a832 3113
e17cade2
CM		 3114 logical = key->offset;
3115 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3116
890871be 3117 read_lock(&map_tree->map_tree.lock);
0b86a832 3118 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 3119 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3120
3121 /* already mapped? */
3122 if (em && em->start <= logical && em->start + em->len > logical) {
3123 free_extent_map(em);
0b86a832
CM		 3124 return 0;
3125 } else if (em) {
3126 free_extent_map(em);
3127 }
0b86a832 3128
0b86a832
CM		 3129 em = alloc_extent_map(GFP_NOFS);
3130 if (!em)
3131 return -ENOMEM;
593060d7
CM		 3132 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3133 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3134 if (!map) {
3135 free_extent_map(em);
3136 return -ENOMEM;
3137 }
3138
3139 em->bdev = (struct block_device *)map;
3140 em->start = logical;
3141 em->len = length;
3142 em->block_start = 0;
c8b97818 3143 em->block_len = em->len;
0b86a832 3144
593060d7
CM		 3145 map->num_stripes = num_stripes;
3146 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3147 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3148 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3149 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3150 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3151 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3152 for (i = 0; i < num_stripes; i++) {
3153 map->stripes[i].physical =
3154 btrfs_stripe_offset_nr(leaf, chunk, i);
3155 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3156 read_extent_buffer(leaf, uuid, (unsigned long)
3157 btrfs_stripe_dev_uuid_nr(chunk, i),
3158 BTRFS_UUID_SIZE);
2b82032c
YZ		 3159 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3160 NULL);
dfe25020 3161 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3162 kfree(map);
3163 free_extent_map(em);
3164 return -EIO;
3165 }
dfe25020
CM		 3166 if (!map->stripes[i].dev) {
3167 map->stripes[i].dev =
3168 add_missing_dev(root, devid, uuid);
3169 if (!map->stripes[i].dev) {
3170 kfree(map);
3171 free_extent_map(em);
3172 return -EIO;
3173 }
3174 }
3175 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3176 }
3177
890871be 3178 write_lock(&map_tree->map_tree.lock);
0b86a832 3179 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 3180 write_unlock(&map_tree->map_tree.lock);
b248a415 3181 BUG_ON(ret);
0b86a832
CM		 3182 free_extent_map(em);
3183
3184 return 0;
3185}
3186
3187static int fill_device_from_item(struct extent_buffer *leaf,
3188 struct btrfs_dev_item *dev_item,
3189 struct btrfs_device *device)
3190{
3191 unsigned long ptr;
0b86a832
CM		 3192
3193 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 3194 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3195 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 3196 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3197 device->type = btrfs_device_type(leaf, dev_item);
3198 device->io_align = btrfs_device_io_align(leaf, dev_item);
3199 device->io_width = btrfs_device_io_width(leaf, dev_item);
3200 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3201
3202 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3203 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3204
0b86a832
CM		 3205 return 0;
3206}
3207
2b82032c
YZ		 3208static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3209{
3210 struct btrfs_fs_devices *fs_devices;
3211 int ret;
3212
3213 mutex_lock(&uuid_mutex);
3214
3215 fs_devices = root->fs_info->fs_devices->seed;
3216 while (fs_devices) {
3217 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3218 ret = 0;
3219 goto out;
3220 }
3221 fs_devices = fs_devices->seed;
3222 }
3223
3224 fs_devices = find_fsid(fsid);
3225 if (!fs_devices) {
3226 ret = -ENOENT;
3227 goto out;
3228 }
e4404d6e
YZ		 3229
3230 fs_devices = clone_fs_devices(fs_devices);
3231 if (IS_ERR(fs_devices)) {
3232 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3233 goto out;
3234 }
3235
97288f2c 3236 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3237 root->fs_info->bdev_holder);
2b82032c
YZ		 3238 if (ret)
3239 goto out;
3240
3241 if (!fs_devices->seeding) {
3242 __btrfs_close_devices(fs_devices);
e4404d6e 3243 free_fs_devices(fs_devices);
2b82032c
YZ		 3244 ret = -EINVAL;
3245 goto out;
3246 }
3247
3248 fs_devices->seed = root->fs_info->fs_devices->seed;
3249 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3250out:
3251 mutex_unlock(&uuid_mutex);
3252 return ret;
3253}
3254
0d81ba5d 3255static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3256 struct extent_buffer *leaf,
3257 struct btrfs_dev_item *dev_item)
3258{
3259 struct btrfs_device *device;
3260 u64 devid;
3261 int ret;
2b82032c 3262 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3263 u8 dev_uuid[BTRFS_UUID_SIZE];
3264
0b86a832 3265 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3266 read_extent_buffer(leaf, dev_uuid,
3267 (unsigned long)btrfs_device_uuid(dev_item),
3268 BTRFS_UUID_SIZE);
2b82032c
YZ		 3269 read_extent_buffer(leaf, fs_uuid,
3270 (unsigned long)btrfs_device_fsid(dev_item),
3271 BTRFS_UUID_SIZE);
3272
3273 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3274 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3275 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3276 return ret;
2b82032c
YZ		 3277 }
3278
3279 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3280 if (!device || !device->bdev) {
e4404d6e 3281 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3282 return -EIO;
3283
3284 if (!device) {
d397712b
CM		 3285 printk(KERN_WARNING "warning devid %llu missing\n",
3286 (unsigned long long)devid);
2b82032c
YZ		 3287 device = add_missing_dev(root, devid, dev_uuid);
3288 if (!device)
3289 return -ENOMEM;
3290 }
3291 }
3292
3293 if (device->fs_devices != root->fs_info->fs_devices) {
3294 BUG_ON(device->writeable);
3295 if (device->generation !=
3296 btrfs_device_generation(leaf, dev_item))
3297 return -EINVAL;
6324fbf3 3298 }
0b86a832
CM		 3299
3300 fill_device_from_item(leaf, dev_item, device);
3301 device->dev_root = root->fs_info->dev_root;
dfe25020 3302 device->in_fs_metadata = 1;
2b82032c
YZ		 3303 if (device->writeable)
3304 device->fs_devices->total_rw_bytes += device->total_bytes;
0b86a832 3305 ret = 0;
0b86a832
CM		 3306 return ret;
3307}
3308
0d81ba5d
CM		 3309int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3310{
3311 struct btrfs_dev_item *dev_item;
3312
3313 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3314 dev_item);
3315 return read_one_dev(root, buf, dev_item);
3316}
3317
e4404d6e 3318int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832
CM		 3319{
3320 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
a061fc8d 3321 struct extent_buffer *sb;
0b86a832 3322 struct btrfs_disk_key *disk_key;
0b86a832 3323 struct btrfs_chunk *chunk;
84eed90f
CM		 3324 u8 *ptr;
3325 unsigned long sb_ptr;
3326 int ret = 0;
0b86a832
CM		 3327 u32 num_stripes;
3328 u32 array_size;
3329 u32 len = 0;
0b86a832 3330 u32 cur;
84eed90f 3331 struct btrfs_key key;
0b86a832 3332
e4404d6e 3333 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3334 BTRFS_SUPER_INFO_SIZE);
3335 if (!sb)
3336 return -ENOMEM;
3337 btrfs_set_buffer_uptodate(sb);
4008c04a
CM		 3338 btrfs_set_buffer_lockdep_class(sb, 0);
3339
a061fc8d 3340 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3341 array_size = btrfs_super_sys_array_size(super_copy);
3342
0b86a832
CM		 3343 ptr = super_copy->sys_chunk_array;
3344 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3345 cur = 0;
3346
3347 while (cur < array_size) {
3348 disk_key = (struct btrfs_disk_key *)ptr;
3349 btrfs_disk_key_to_cpu(&key, disk_key);
3350
a061fc8d 3351 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3352 sb_ptr += len;
3353 cur += len;
3354
0d81ba5d 3355 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3356 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3357 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3358 if (ret)
3359 break;
0b86a832
CM		 3360 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3361 len = btrfs_chunk_item_size(num_stripes);
3362 } else {
84eed90f
CM		 3363 ret = -EIO;
3364 break;
0b86a832
CM		 3365 }
3366 ptr += len;
3367 sb_ptr += len;
3368 cur += len;
3369 }
a061fc8d 3370 free_extent_buffer(sb);
84eed90f 3371 return ret;
0b86a832
CM		 3372}
3373
3374int btrfs_read_chunk_tree(struct btrfs_root *root)
3375{
3376 struct btrfs_path *path;
3377 struct extent_buffer *leaf;
3378 struct btrfs_key key;
3379 struct btrfs_key found_key;
3380 int ret;
3381 int slot;
3382
3383 root = root->fs_info->chunk_root;
3384
3385 path = btrfs_alloc_path();
3386 if (!path)
3387 return -ENOMEM;
3388
3389 /* first we search for all of the device items, and then we
3390 * read in all of the chunk items. This way we can create chunk
3391 * mappings that reference all of the devices that are afound
3392 */
3393 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3394 key.offset = 0;
3395 key.type = 0;
3396again:
3397 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 3398 if (ret < 0)
3399 goto error;
d397712b 3400 while (1) {
0b86a832
CM		 3401 leaf = path->nodes[0];
3402 slot = path->slots[0];
3403 if (slot >= btrfs_header_nritems(leaf)) {
3404 ret = btrfs_next_leaf(root, path);
3405 if (ret == 0)
3406 continue;
3407 if (ret < 0)
3408 goto error;
3409 break;
3410 }
3411 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3412 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3413 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3414 break;
3415 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3416 struct btrfs_dev_item *dev_item;
3417 dev_item = btrfs_item_ptr(leaf, slot,
3418 struct btrfs_dev_item);
0d81ba5d 3419 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3420 if (ret)
3421 goto error;
0b86a832
CM		 3422 }
3423 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3424 struct btrfs_chunk *chunk;
3425 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3426 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3427 if (ret)
3428 goto error;
0b86a832
CM		 3429 }
3430 path->slots[0]++;
3431 }
3432 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3433 key.objectid = 0;
3434 btrfs_release_path(root, path);
3435 goto again;
3436 }
0b86a832
CM		 3437 ret = 0;
3438error:
2b82032c 3439 btrfs_free_path(path);
0b86a832
CM		 3440 return ret;
3441}
Clone of davem's net-next-2.6 tree