projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: make metadata chunks smaller
[linux-2.6-block.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>
8a4b83cc 20#include <linux/buffer_head.h>
f2d8d74d 21#include <linux/blkdev.h>
788f20eb 22#include <linux/random.h>
b765ead5 23#include <linux/iocontext.h>
593060d7 24#include <asm/div64.h>
4b4e25f2 25#include "compat.h"
0b86a832
CM		 26#include "ctree.h"
27#include "extent_map.h"
28#include "disk-io.h"
29#include "transaction.h"
30#include "print-tree.h"
31#include "volumes.h"
8b712842 32#include "async-thread.h"
0b86a832 33
593060d7
CM		 34struct map_lookup {
35 u64 type;
36 int io_align;
37 int io_width;
38 int stripe_len;
39 int sector_size;
40 int num_stripes;
321aecc6 41 int sub_stripes;
cea9e445 42 struct btrfs_bio_stripe stripes[];
593060d7
CM		 43};
44
2b82032c
YZ		 45static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49
593060d7 50#define map_lookup_size(n) (sizeof(struct map_lookup) + \
cea9e445 51 (sizeof(struct btrfs_bio_stripe) * (n)))
593060d7 52
8a4b83cc
CM		 53static DEFINE_MUTEX(uuid_mutex);
54static LIST_HEAD(fs_uuids);
55
a061fc8d
CM		 56void btrfs_lock_volumes(void)
57{
58 mutex_lock(&uuid_mutex);
59}
60
61void btrfs_unlock_volumes(void)
62{
63 mutex_unlock(&uuid_mutex);
64}
65
7d9eb12c
CM		 66static void lock_chunks(struct btrfs_root *root)
67{
7d9eb12c
CM		 68 mutex_lock(&root->fs_info->chunk_mutex);
69}
70
71static void unlock_chunks(struct btrfs_root *root)
72{
7d9eb12c
CM		 73 mutex_unlock(&root->fs_info->chunk_mutex);
74}
75
e4404d6e
YZ		 76static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
77{
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
84 kfree(device->name);
85 kfree(device);
86 }
87 kfree(fs_devices);
88}
89
8a4b83cc
CM		 90int btrfs_cleanup_fs_uuids(void)
91{
92 struct btrfs_fs_devices *fs_devices;
8a4b83cc 93
2b82032c
YZ		 94 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
e4404d6e 98 free_fs_devices(fs_devices);
8a4b83cc
CM		 99 }
100 return 0;
101}
102
a1b32a59
CM		 103static noinline struct btrfs_device *__find_device(struct list_head *head,
104 u64 devid, u8 *uuid)
8a4b83cc
CM		 105{
106 struct btrfs_device *dev;
8a4b83cc 107
c6e30871 108 list_for_each_entry(dev, head, dev_list) {
a443755f 109 if (dev->devid == devid &&
8f18cf13 110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 111 return dev;
a443755f 112 }
8a4b83cc
CM		 113 }
114 return NULL;
115}
116
a1b32a59 117static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 118{
8a4b83cc
CM		 119 struct btrfs_fs_devices *fs_devices;
120
c6e30871 121 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
123 return fs_devices;
124 }
125 return NULL;
126}
127
ffbd517d
CM		 128static void requeue_list(struct btrfs_pending_bios *pending_bios,
129 struct bio *head, struct bio *tail)
130{
131
132 struct bio *old_head;
133
134 old_head = pending_bios->head;
135 pending_bios->head = head;
136 if (pending_bios->tail)
137 tail->bi_next = old_head;
138 else
139 pending_bios->tail = tail;
140}
141
8b712842
CM		 142/*
143 * we try to collect pending bios for a device so we don't get a large
144 * number of procs sending bios down to the same device. This greatly
145 * improves the schedulers ability to collect and merge the bios.
146 *
147 * But, it also turns into a long list of bios to process and that is sure
148 * to eventually make the worker thread block. The solution here is to
149 * make some progress and then put this work struct back at the end of
150 * the list if the block device is congested. This way, multiple devices
151 * can make progress from a single worker thread.
152 */
d397712b 153static noinline int run_scheduled_bios(struct btrfs_device *device)
8b712842
CM		 154{
155 struct bio *pending;
156 struct backing_dev_info *bdi;
b64a2851 157 struct btrfs_fs_info *fs_info;
ffbd517d 158 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 159 struct bio *tail;
160 struct bio *cur;
161 int again = 0;
ffbd517d
CM		 162 unsigned long num_run;
163 unsigned long num_sync_run;
d644d8a1 164 unsigned long batch_run = 0;
b64a2851 165 unsigned long limit;
b765ead5 166 unsigned long last_waited = 0;
d84275c9 167 int force_reg = 0;
8b712842 168
bedf762b 169 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 170 fs_info = device->dev_root->fs_info;
171 limit = btrfs_async_submit_limit(fs_info);
172 limit = limit * 2 / 3;
173
ffbd517d
CM		 174 /* we want to make sure that every time we switch from the sync
175 * list to the normal list, we unplug
176 */
177 num_sync_run = 0;
178
8b712842
CM		 179loop:
180 spin_lock(&device->io_lock);
181
a6837051 182loop_lock:
d84275c9 183 num_run = 0;
ffbd517d 184
8b712842
CM		 185 /* take all the bios off the list at once and process them
186 * later on (without the lock held). But, remember the
187 * tail and other pointers so the bios can be properly reinserted
188 * into the list if we hit congestion
189 */
d84275c9 190 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 191 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 192 force_reg = 1;
193 } else {
ffbd517d 194 pending_bios = &device->pending_bios;
d84275c9
CM		 195 force_reg = 0;
196 }
ffbd517d
CM		 197
198 pending = pending_bios->head;
199 tail = pending_bios->tail;
8b712842 200 WARN_ON(pending && !tail);
8b712842
CM		 201
202 /*
203 * if pending was null this time around, no bios need processing
204 * at all and we can stop. Otherwise it'll loop back up again
205 * and do an additional check so no bios are missed.
206 *
207 * device->running_pending is used to synchronize with the
208 * schedule_bio code.
209 */
ffbd517d
CM		 210 if (device->pending_sync_bios.head == NULL &&
211 device->pending_bios.head == NULL) {
8b712842
CM		 212 again = 0;
213 device->running_pending = 0;
ffbd517d
CM		 214 } else {
215 again = 1;
216 device->running_pending = 1;
8b712842 217 }
ffbd517d
CM		 218
219 pending_bios->head = NULL;
220 pending_bios->tail = NULL;
221
8b712842
CM		 222 spin_unlock(&device->io_lock);
223
ffbd517d
CM		 224 /*
225 * if we're doing the regular priority list, make sure we unplug
226 * for any high prio bios we've sent down
227 */
228 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
229 num_sync_run = 0;
230 blk_run_backing_dev(bdi, NULL);
231 }
232
d397712b 233 while (pending) {
ffbd517d
CM		 234
235 rmb();
d84275c9
CM		 236 /* we want to work on both lists, but do more bios on the
237 * sync list than the regular list
238 */
239 if ((num_run > 32 &&
240 pending_bios != &device->pending_sync_bios &&
241 device->pending_sync_bios.head) ||
242 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
243 device->pending_bios.head)) {
ffbd517d
CM		 244 spin_lock(&device->io_lock);
245 requeue_list(pending_bios, pending, tail);
246 goto loop_lock;
247 }
248
8b712842
CM		 249 cur = pending;
250 pending = pending->bi_next;
251 cur->bi_next = NULL;
b64a2851
CM		 252 atomic_dec(&fs_info->nr_async_bios);
253
254 if (atomic_read(&fs_info->nr_async_bios) < limit &&
255 waitqueue_active(&fs_info->async_submit_wait))
256 wake_up(&fs_info->async_submit_wait);
492bb6de
CM		 257
258 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
8b712842
CM		 259 submit_bio(cur->bi_rw, cur);
260 num_run++;
d644d8a1
CM		 261 batch_run++;
262
1f98a13f 263 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
ffbd517d
CM		 264 num_sync_run++;
265
266 if (need_resched()) {
267 if (num_sync_run) {
268 blk_run_backing_dev(bdi, NULL);
269 num_sync_run = 0;
270 }
271 cond_resched();
272 }
8b712842
CM		 273
274 /*
275 * we made progress, there is more work to do and the bdi
276 * is now congested. Back off and let other work structs
277 * run instead
278 */
57fd5a5f 279 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 280 fs_info->fs_devices->open_devices > 1) {
b765ead5 281 struct io_context *ioc;
8b712842 282
b765ead5
CM		 283 ioc = current->io_context;
284
285 /*
286 * the main goal here is that we don't want to
287 * block if we're going to be able to submit
288 * more requests without blocking.
289 *
290 * This code does two great things, it pokes into
291 * the elevator code from a filesystem _and_
292 * it makes assumptions about how batching works.
293 */
294 if (ioc && ioc->nr_batch_requests > 0 &&
295 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
296 (last_waited == 0 ||
297 ioc->last_waited == last_waited)) {
298 /*
299 * we want to go through our batch of
300 * requests and stop. So, we copy out
301 * the ioc->last_waited time and test
302 * against it before looping
303 */
304 last_waited = ioc->last_waited;
ffbd517d
CM		 305 if (need_resched()) {
306 if (num_sync_run) {
307 blk_run_backing_dev(bdi, NULL);
308 num_sync_run = 0;
309 }
310 cond_resched();
311 }
b765ead5
CM		 312 continue;
313 }
8b712842 314 spin_lock(&device->io_lock);
ffbd517d 315 requeue_list(pending_bios, pending, tail);
a6837051 316 device->running_pending = 1;
8b712842
CM		 317
318 spin_unlock(&device->io_lock);
319 btrfs_requeue_work(&device->work);
320 goto done;
321 }
322 }
ffbd517d
CM		 323
324 if (num_sync_run) {
325 num_sync_run = 0;
326 blk_run_backing_dev(bdi, NULL);
327 }
328
329 cond_resched();
8b712842
CM		 330 if (again)
331 goto loop;
a6837051
CM		 332
333 spin_lock(&device->io_lock);
ffbd517d 334 if (device->pending_bios.head || device->pending_sync_bios.head)
a6837051
CM		 335 goto loop_lock;
336 spin_unlock(&device->io_lock);
bedf762b
CM		 337
338 /*
339 * IO has already been through a long path to get here. Checksumming,
340 * async helper threads, perhaps compression. We've done a pretty
341 * good job of collecting a batch of IO and should just unplug
342 * the device right away.
343 *
344 * This will help anyone who is waiting on the IO, they might have
345 * already unplugged, but managed to do so before the bio they
346 * cared about found its way down here.
347 */
348 blk_run_backing_dev(bdi, NULL);
8b712842
CM		 349done:
350 return 0;
351}
352
b2950863 353static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 354{
355 struct btrfs_device *device;
356
357 device = container_of(work, struct btrfs_device, work);
358 run_scheduled_bios(device);
359}
360
a1b32a59 361static noinline int device_list_add(const char *path,
8a4b83cc
CM		 362 struct btrfs_super_block *disk_super,
363 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
364{
365 struct btrfs_device *device;
366 struct btrfs_fs_devices *fs_devices;
367 u64 found_transid = btrfs_super_generation(disk_super);
368
369 fs_devices = find_fsid(disk_super->fsid);
370 if (!fs_devices) {
515dc322 371 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 372 if (!fs_devices)
373 return -ENOMEM;
374 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 375 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 376 list_add(&fs_devices->list, &fs_uuids);
377 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
378 fs_devices->latest_devid = devid;
379 fs_devices->latest_trans = found_transid;
e5e9a520 380 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 381 device = NULL;
382 } else {
a443755f
CM		 383 device = __find_device(&fs_devices->devices, devid,
384 disk_super->dev_item.uuid);
8a4b83cc
CM		 385 }
386 if (!device) {
2b82032c
YZ		 387 if (fs_devices->opened)
388 return -EBUSY;
389
8a4b83cc
CM		 390 device = kzalloc(sizeof(*device), GFP_NOFS);
391 if (!device) {
392 /* we can safely leave the fs_devices entry around */
393 return -ENOMEM;
394 }
395 device->devid = devid;
8b712842 396 device->work.func = pending_bios_fn;
a443755f
CM		 397 memcpy(device->uuid, disk_super->dev_item.uuid,
398 BTRFS_UUID_SIZE);
f2984462 399 device->barriers = 1;
b248a415 400 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 401 device->name = kstrdup(path, GFP_NOFS);
402 if (!device->name) {
403 kfree(device);
404 return -ENOMEM;
405 }
2b82032c 406 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520
CM		 407
408 mutex_lock(&fs_devices->device_list_mutex);
8a4b83cc 409 list_add(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 410 mutex_unlock(&fs_devices->device_list_mutex);
411
2b82032c 412 device->fs_devices = fs_devices;
8a4b83cc
CM		 413 fs_devices->num_devices++;
414 }
415
416 if (found_transid > fs_devices->latest_trans) {
417 fs_devices->latest_devid = devid;
418 fs_devices->latest_trans = found_transid;
419 }
8a4b83cc
CM		 420 *fs_devices_ret = fs_devices;
421 return 0;
422}
423
e4404d6e
YZ		 424static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
425{
426 struct btrfs_fs_devices *fs_devices;
427 struct btrfs_device *device;
428 struct btrfs_device *orig_dev;
429
430 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
431 if (!fs_devices)
432 return ERR_PTR(-ENOMEM);
433
434 INIT_LIST_HEAD(&fs_devices->devices);
435 INIT_LIST_HEAD(&fs_devices->alloc_list);
436 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 437 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 438 fs_devices->latest_devid = orig->latest_devid;
439 fs_devices->latest_trans = orig->latest_trans;
440 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
441
e5e9a520 442 mutex_lock(&orig->device_list_mutex);
e4404d6e
YZ		 443 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
444 device = kzalloc(sizeof(*device), GFP_NOFS);
445 if (!device)
446 goto error;
447
448 device->name = kstrdup(orig_dev->name, GFP_NOFS);
fd2696f3
JL		 449 if (!device->name) {
450 kfree(device);
e4404d6e 451 goto error;
fd2696f3 452 }
e4404d6e
YZ		 453
454 device->devid = orig_dev->devid;
455 device->work.func = pending_bios_fn;
456 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
457 device->barriers = 1;
458 spin_lock_init(&device->io_lock);
459 INIT_LIST_HEAD(&device->dev_list);
460 INIT_LIST_HEAD(&device->dev_alloc_list);
461
462 list_add(&device->dev_list, &fs_devices->devices);
463 device->fs_devices = fs_devices;
464 fs_devices->num_devices++;
465 }
e5e9a520 466 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 467 return fs_devices;
468error:
e5e9a520 469 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 470 free_fs_devices(fs_devices);
471 return ERR_PTR(-ENOMEM);
472}
473
dfe25020
CM		 474int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
475{
c6e30871 476 struct btrfs_device *device, *next;
dfe25020
CM		 477
478 mutex_lock(&uuid_mutex);
479again:
e5e9a520 480 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 481 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 482 if (device->in_fs_metadata)
483 continue;
484
485 if (device->bdev) {
15916de8 486 close_bdev_exclusive(device->bdev, device->mode);
2b82032c
YZ		 487 device->bdev = NULL;
488 fs_devices->open_devices--;
489 }
490 if (device->writeable) {
491 list_del_init(&device->dev_alloc_list);
492 device->writeable = 0;
493 fs_devices->rw_devices--;
494 }
e4404d6e
YZ		 495 list_del_init(&device->dev_list);
496 fs_devices->num_devices--;
497 kfree(device->name);
498 kfree(device);
dfe25020 499 }
e5e9a520 500 mutex_unlock(&fs_devices->device_list_mutex);
2b82032c
YZ		 501
502 if (fs_devices->seed) {
503 fs_devices = fs_devices->seed;
2b82032c
YZ		 504 goto again;
505 }
506
dfe25020
CM		 507 mutex_unlock(&uuid_mutex);
508 return 0;
509}
a0af469b 510
2b82032c 511static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 512{
8a4b83cc 513 struct btrfs_device *device;
e4404d6e 514
2b82032c
YZ		 515 if (--fs_devices->opened > 0)
516 return 0;
8a4b83cc 517
c6e30871 518 list_for_each_entry(device, &fs_devices->devices, dev_list) {
8a4b83cc 519 if (device->bdev) {
15916de8 520 close_bdev_exclusive(device->bdev, device->mode);
a0af469b 521 fs_devices->open_devices--;
8a4b83cc 522 }
2b82032c
YZ		 523 if (device->writeable) {
524 list_del_init(&device->dev_alloc_list);
525 fs_devices->rw_devices--;
526 }
527
8a4b83cc 528 device->bdev = NULL;
2b82032c 529 device->writeable = 0;
dfe25020 530 device->in_fs_metadata = 0;
8a4b83cc 531 }
e4404d6e
YZ		 532 WARN_ON(fs_devices->open_devices);
533 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 534 fs_devices->opened = 0;
535 fs_devices->seeding = 0;
2b82032c 536
8a4b83cc
CM		 537 return 0;
538}
539
2b82032c
YZ		 540int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
541{
e4404d6e 542 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 543 int ret;
544
545 mutex_lock(&uuid_mutex);
546 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 547 if (!fs_devices->opened) {
548 seed_devices = fs_devices->seed;
549 fs_devices->seed = NULL;
550 }
2b82032c 551 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 552
553 while (seed_devices) {
554 fs_devices = seed_devices;
555 seed_devices = fs_devices->seed;
556 __btrfs_close_devices(fs_devices);
557 free_fs_devices(fs_devices);
558 }
2b82032c
YZ		 559 return ret;
560}
561
e4404d6e
YZ		 562static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563 fmode_t flags, void *holder)
8a4b83cc
CM		 564{
565 struct block_device *bdev;
566 struct list_head *head = &fs_devices->devices;
8a4b83cc 567 struct btrfs_device *device;
a0af469b
CM		 568 struct block_device *latest_bdev = NULL;
569 struct buffer_head *bh;
570 struct btrfs_super_block *disk_super;
571 u64 latest_devid = 0;
572 u64 latest_transid = 0;
a0af469b 573 u64 devid;
2b82032c 574 int seeding = 1;
a0af469b 575 int ret = 0;
8a4b83cc 576
c6e30871 577 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 578 if (device->bdev)
579 continue;
dfe25020
CM		 580 if (!device->name)
581 continue;
582
15916de8 583 bdev = open_bdev_exclusive(device->name, flags, holder);
8a4b83cc 584 if (IS_ERR(bdev)) {
d397712b 585 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 586 goto error;
8a4b83cc 587 }
a061fc8d 588 set_blocksize(bdev, 4096);
a0af469b 589
a512bbf8 590 bh = btrfs_read_dev_super(bdev);
a0af469b
CM		 591 if (!bh)
592 goto error_close;
593
594 disk_super = (struct btrfs_super_block *)bh->b_data;
a0af469b
CM		 595 devid = le64_to_cpu(disk_super->dev_item.devid);
596 if (devid != device->devid)
597 goto error_brelse;
598
2b82032c
YZ		 599 if (memcmp(device->uuid, disk_super->dev_item.uuid,
600 BTRFS_UUID_SIZE))
601 goto error_brelse;
602
603 device->generation = btrfs_super_generation(disk_super);
604 if (!latest_transid || device->generation > latest_transid) {
a0af469b 605 latest_devid = devid;
2b82032c 606 latest_transid = device->generation;
a0af469b
CM		 607 latest_bdev = bdev;
608 }
609
2b82032c
YZ		 610 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
611 device->writeable = 0;
612 } else {
613 device->writeable = !bdev_read_only(bdev);
614 seeding = 0;
615 }
616
8a4b83cc 617 device->bdev = bdev;
dfe25020 618 device->in_fs_metadata = 0;
15916de8
CM		 619 device->mode = flags;
620
c289811c
CM		 621 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
622 fs_devices->rotating = 1;
623
a0af469b 624 fs_devices->open_devices++;
2b82032c
YZ		 625 if (device->writeable) {
626 fs_devices->rw_devices++;
627 list_add(&device->dev_alloc_list,
628 &fs_devices->alloc_list);
629 }
a0af469b 630 continue;
a061fc8d 631
a0af469b
CM		 632error_brelse:
633 brelse(bh);
634error_close:
97288f2c 635 close_bdev_exclusive(bdev, FMODE_READ);
a0af469b
CM		 636error:
637 continue;
8a4b83cc 638 }
a0af469b
CM		 639 if (fs_devices->open_devices == 0) {
640 ret = -EIO;
641 goto out;
642 }
2b82032c
YZ		 643 fs_devices->seeding = seeding;
644 fs_devices->opened = 1;
a0af469b
CM		 645 fs_devices->latest_bdev = latest_bdev;
646 fs_devices->latest_devid = latest_devid;
647 fs_devices->latest_trans = latest_transid;
2b82032c 648 fs_devices->total_rw_bytes = 0;
a0af469b 649out:
2b82032c
YZ		 650 return ret;
651}
652
653int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 654 fmode_t flags, void *holder)
2b82032c
YZ		 655{
656 int ret;
657
658 mutex_lock(&uuid_mutex);
659 if (fs_devices->opened) {
e4404d6e
YZ		 660 fs_devices->opened++;
661 ret = 0;
2b82032c 662 } else {
15916de8 663 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 664 }
8a4b83cc 665 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 666 return ret;
667}
668
97288f2c 669int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 670 struct btrfs_fs_devices **fs_devices_ret)
671{
672 struct btrfs_super_block *disk_super;
673 struct block_device *bdev;
674 struct buffer_head *bh;
675 int ret;
676 u64 devid;
f2984462 677 u64 transid;
8a4b83cc
CM		 678
679 mutex_lock(&uuid_mutex);
680
15916de8 681 bdev = open_bdev_exclusive(path, flags, holder);
8a4b83cc
CM		 682
683 if (IS_ERR(bdev)) {
8a4b83cc
CM		 684 ret = PTR_ERR(bdev);
685 goto error;
686 }
687
688 ret = set_blocksize(bdev, 4096);
689 if (ret)
690 goto error_close;
a512bbf8 691 bh = btrfs_read_dev_super(bdev);
8a4b83cc
CM		 692 if (!bh) {
693 ret = -EIO;
694 goto error_close;
695 }
696 disk_super = (struct btrfs_super_block *)bh->b_data;
8a4b83cc 697 devid = le64_to_cpu(disk_super->dev_item.devid);
f2984462 698 transid = btrfs_super_generation(disk_super);
7ae9c09d 699 if (disk_super->label[0])
d397712b 700 printk(KERN_INFO "device label %s ", disk_super->label);
7ae9c09d
CM		 701 else {
702 /* FIXME, make a readl uuid parser */
d397712b 703 printk(KERN_INFO "device fsid %llx-%llx ",
7ae9c09d
CM		 704 *(unsigned long long *)disk_super->fsid,
705 *(unsigned long long *)(disk_super->fsid + 8));
706 }
119e10cf 707 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 708 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 709 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
710
8a4b83cc
CM		 711 brelse(bh);
712error_close:
15916de8 713 close_bdev_exclusive(bdev, flags);
8a4b83cc
CM		 714error:
715 mutex_unlock(&uuid_mutex);
716 return ret;
717}
0b86a832
CM		 718
719/*
720 * this uses a pretty simple search, the expectation is that it is
721 * called very infrequently and that a given device has a small number
722 * of extents
723 */
ba1bf481
JB		 724int find_free_dev_extent(struct btrfs_trans_handle *trans,
725 struct btrfs_device *device, u64 num_bytes,
726 u64 *start, u64 *max_avail)
0b86a832
CM		 727{
728 struct btrfs_key key;
729 struct btrfs_root *root = device->dev_root;
730 struct btrfs_dev_extent *dev_extent = NULL;
2b82032c 731 struct btrfs_path *path;
0b86a832
CM		 732 u64 hole_size = 0;
733 u64 last_byte = 0;
734 u64 search_start = 0;
735 u64 search_end = device->total_bytes;
736 int ret;
737 int slot = 0;
738 int start_found;
739 struct extent_buffer *l;
740
2b82032c
YZ		 741 path = btrfs_alloc_path();
742 if (!path)
743 return -ENOMEM;
0b86a832 744 path->reada = 2;
2b82032c 745 start_found = 0;
0b86a832
CM		 746
747 /* FIXME use last free of some kind */
748
8a4b83cc
CM		 749 /* we don't want to overwrite the superblock on the drive,
750 * so we make sure to start at an offset of at least 1MB
751 */
752 search_start = max((u64)1024 * 1024, search_start);
8f18cf13
CM		 753
754 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
755 search_start = max(root->fs_info->alloc_start, search_start);
756
0b86a832
CM		 757 key.objectid = device->devid;
758 key.offset = search_start;
759 key.type = BTRFS_DEV_EXTENT_KEY;
760 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
761 if (ret < 0)
762 goto error;
1fcbac58
YZ		 763 if (ret > 0) {
764 ret = btrfs_previous_item(root, path, key.objectid, key.type);
765 if (ret < 0)
766 goto error;
767 if (ret > 0)
768 start_found = 1;
769 }
0b86a832
CM		 770 l = path->nodes[0];
771 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
772 while (1) {
773 l = path->nodes[0];
774 slot = path->slots[0];
775 if (slot >= btrfs_header_nritems(l)) {
776 ret = btrfs_next_leaf(root, path);
777 if (ret == 0)
778 continue;
779 if (ret < 0)
780 goto error;
781no_more_items:
782 if (!start_found) {
783 if (search_start >= search_end) {
784 ret = -ENOSPC;
785 goto error;
786 }
787 *start = search_start;
788 start_found = 1;
789 goto check_pending;
790 }
791 *start = last_byte > search_start ?
792 last_byte : search_start;
793 if (search_end <= *start) {
794 ret = -ENOSPC;
795 goto error;
796 }
797 goto check_pending;
798 }
799 btrfs_item_key_to_cpu(l, &key, slot);
800
801 if (key.objectid < device->devid)
802 goto next;
803
804 if (key.objectid > device->devid)
805 goto no_more_items;
806
807 if (key.offset >= search_start && key.offset > last_byte &&
808 start_found) {
809 if (last_byte < search_start)
810 last_byte = search_start;
811 hole_size = key.offset - last_byte;
9779b72f
CM		 812
813 if (hole_size > *max_avail)
814 *max_avail = hole_size;
815
0b86a832
CM		 816 if (key.offset > last_byte &&
817 hole_size >= num_bytes) {
818 *start = last_byte;
819 goto check_pending;
820 }
821 }
d397712b 822 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
0b86a832 823 goto next;
0b86a832
CM		 824
825 start_found = 1;
826 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
827 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
828next:
829 path->slots[0]++;
830 cond_resched();
831 }
832check_pending:
833 /* we have to make sure we didn't find an extent that has already
834 * been allocated by the map tree or the original allocation
835 */
0b86a832
CM		 836 BUG_ON(*start < search_start);
837
6324fbf3 838 if (*start + num_bytes > search_end) {
0b86a832
CM		 839 ret = -ENOSPC;
840 goto error;
841 }
842 /* check for pending inserts here */
2b82032c 843 ret = 0;
0b86a832
CM		 844
845error:
2b82032c 846 btrfs_free_path(path);
0b86a832
CM		 847 return ret;
848}
849
b2950863 850static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 851 struct btrfs_device *device,
852 u64 start)
853{
854 int ret;
855 struct btrfs_path *path;
856 struct btrfs_root *root = device->dev_root;
857 struct btrfs_key key;
a061fc8d
CM		 858 struct btrfs_key found_key;
859 struct extent_buffer *leaf = NULL;
860 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 861
862 path = btrfs_alloc_path();
863 if (!path)
864 return -ENOMEM;
865
866 key.objectid = device->devid;
867 key.offset = start;
868 key.type = BTRFS_DEV_EXTENT_KEY;
869
870 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 871 if (ret > 0) {
872 ret = btrfs_previous_item(root, path, key.objectid,
873 BTRFS_DEV_EXTENT_KEY);
874 BUG_ON(ret);
875 leaf = path->nodes[0];
876 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
877 extent = btrfs_item_ptr(leaf, path->slots[0],
878 struct btrfs_dev_extent);
879 BUG_ON(found_key.offset > start || found_key.offset +
880 btrfs_dev_extent_length(leaf, extent) < start);
881 ret = 0;
882 } else if (ret == 0) {
883 leaf = path->nodes[0];
884 extent = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_dev_extent);
886 }
8f18cf13
CM		 887 BUG_ON(ret);
888
dfe25020
CM		 889 if (device->bytes_used > 0)
890 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
8f18cf13
CM		 891 ret = btrfs_del_item(trans, root, path);
892 BUG_ON(ret);
893
894 btrfs_free_path(path);
895 return ret;
896}
897
2b82032c 898int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 899 struct btrfs_device *device,
e17cade2 900 u64 chunk_tree, u64 chunk_objectid,
2b82032c 901 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 902{
903 int ret;
904 struct btrfs_path *path;
905 struct btrfs_root *root = device->dev_root;
906 struct btrfs_dev_extent *extent;
907 struct extent_buffer *leaf;
908 struct btrfs_key key;
909
dfe25020 910 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 911 path = btrfs_alloc_path();
912 if (!path)
913 return -ENOMEM;
914
0b86a832 915 key.objectid = device->devid;
2b82032c 916 key.offset = start;
0b86a832
CM		 917 key.type = BTRFS_DEV_EXTENT_KEY;
918 ret = btrfs_insert_empty_item(trans, root, path, &key,
919 sizeof(*extent));
920 BUG_ON(ret);
921
922 leaf = path->nodes[0];
923 extent = btrfs_item_ptr(leaf, path->slots[0],
924 struct btrfs_dev_extent);
e17cade2
CM		 925 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
926 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
927 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
928
929 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
930 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
931 BTRFS_UUID_SIZE);
932
0b86a832
CM		 933 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
934 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 935 btrfs_free_path(path);
936 return ret;
937}
938
a1b32a59
CM		 939static noinline int find_next_chunk(struct btrfs_root *root,
940 u64 objectid, u64 *offset)
0b86a832
CM		 941{
942 struct btrfs_path *path;
943 int ret;
944 struct btrfs_key key;
e17cade2 945 struct btrfs_chunk *chunk;
0b86a832
CM		 946 struct btrfs_key found_key;
947
948 path = btrfs_alloc_path();
949 BUG_ON(!path);
950
e17cade2 951 key.objectid = objectid;
0b86a832
CM		 952 key.offset = (u64)-1;
953 key.type = BTRFS_CHUNK_ITEM_KEY;
954
955 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
956 if (ret < 0)
957 goto error;
958
959 BUG_ON(ret == 0);
960
961 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
962 if (ret) {
e17cade2 963 *offset = 0;
0b86a832
CM		 964 } else {
965 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
966 path->slots[0]);
e17cade2
CM		 967 if (found_key.objectid != objectid)
968 *offset = 0;
969 else {
970 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
971 struct btrfs_chunk);
972 *offset = found_key.offset +
973 btrfs_chunk_length(path->nodes[0], chunk);
974 }
0b86a832
CM		 975 }
976 ret = 0;
977error:
978 btrfs_free_path(path);
979 return ret;
980}
981
2b82032c 982static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 983{
984 int ret;
985 struct btrfs_key key;
986 struct btrfs_key found_key;
2b82032c
YZ		 987 struct btrfs_path *path;
988
989 root = root->fs_info->chunk_root;
990
991 path = btrfs_alloc_path();
992 if (!path)
993 return -ENOMEM;
0b86a832
CM		 994
995 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
996 key.type = BTRFS_DEV_ITEM_KEY;
997 key.offset = (u64)-1;
998
999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1000 if (ret < 0)
1001 goto error;
1002
1003 BUG_ON(ret == 0);
1004
1005 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1006 BTRFS_DEV_ITEM_KEY);
1007 if (ret) {
1008 *objectid = 1;
1009 } else {
1010 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1011 path->slots[0]);
1012 *objectid = found_key.offset + 1;
1013 }
1014 ret = 0;
1015error:
2b82032c 1016 btrfs_free_path(path);
0b86a832
CM		 1017 return ret;
1018}
1019
1020/*
1021 * the device information is stored in the chunk root
1022 * the btrfs_device struct should be fully filled in
1023 */
1024int btrfs_add_device(struct btrfs_trans_handle *trans,
1025 struct btrfs_root *root,
1026 struct btrfs_device *device)
1027{
1028 int ret;
1029 struct btrfs_path *path;
1030 struct btrfs_dev_item *dev_item;
1031 struct extent_buffer *leaf;
1032 struct btrfs_key key;
1033 unsigned long ptr;
0b86a832
CM		 1034
1035 root = root->fs_info->chunk_root;
1036
1037 path = btrfs_alloc_path();
1038 if (!path)
1039 return -ENOMEM;
1040
0b86a832
CM		 1041 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1042 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1043 key.offset = device->devid;
0b86a832
CM		 1044
1045 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1046 sizeof(*dev_item));
0b86a832
CM		 1047 if (ret)
1048 goto out;
1049
1050 leaf = path->nodes[0];
1051 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1052
1053 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1054 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1055 btrfs_set_device_type(leaf, dev_item, device->type);
1056 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1057 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1058 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1059 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1060 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1061 btrfs_set_device_group(leaf, dev_item, 0);
1062 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1063 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1064 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1065
0b86a832 1066 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1067 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1068 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1069 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1070 btrfs_mark_buffer_dirty(leaf);
0b86a832 1071
2b82032c 1072 ret = 0;
0b86a832
CM		 1073out:
1074 btrfs_free_path(path);
1075 return ret;
1076}
8f18cf13 1077
a061fc8d
CM		 1078static int btrfs_rm_dev_item(struct btrfs_root *root,
1079 struct btrfs_device *device)
1080{
1081 int ret;
1082 struct btrfs_path *path;
a061fc8d 1083 struct btrfs_key key;
a061fc8d
CM		 1084 struct btrfs_trans_handle *trans;
1085
1086 root = root->fs_info->chunk_root;
1087
1088 path = btrfs_alloc_path();
1089 if (!path)
1090 return -ENOMEM;
1091
1092 trans = btrfs_start_transaction(root, 1);
1093 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1094 key.type = BTRFS_DEV_ITEM_KEY;
1095 key.offset = device->devid;
7d9eb12c 1096 lock_chunks(root);
a061fc8d
CM		 1097
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 if (ret < 0)
1100 goto out;
1101
1102 if (ret > 0) {
1103 ret = -ENOENT;
1104 goto out;
1105 }
1106
1107 ret = btrfs_del_item(trans, root, path);
1108 if (ret)
1109 goto out;
a061fc8d
CM		 1110out:
1111 btrfs_free_path(path);
7d9eb12c 1112 unlock_chunks(root);
a061fc8d
CM		 1113 btrfs_commit_transaction(trans, root);
1114 return ret;
1115}
1116
1117int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1118{
1119 struct btrfs_device *device;
2b82032c 1120 struct btrfs_device *next_device;
a061fc8d 1121 struct block_device *bdev;
dfe25020 1122 struct buffer_head *bh = NULL;
a061fc8d
CM		 1123 struct btrfs_super_block *disk_super;
1124 u64 all_avail;
1125 u64 devid;
2b82032c
YZ		 1126 u64 num_devices;
1127 u8 *dev_uuid;
a061fc8d
CM		 1128 int ret = 0;
1129
a061fc8d 1130 mutex_lock(&uuid_mutex);
7d9eb12c 1131 mutex_lock(&root->fs_info->volume_mutex);
a061fc8d
CM		 1132
1133 all_avail = root->fs_info->avail_data_alloc_bits |
1134 root->fs_info->avail_system_alloc_bits |
1135 root->fs_info->avail_metadata_alloc_bits;
1136
1137 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
2b82032c 1138 root->fs_info->fs_devices->rw_devices <= 4) {
d397712b
CM		 1139 printk(KERN_ERR "btrfs: unable to go below four devices "
1140 "on raid10\n");
a061fc8d
CM		 1141 ret = -EINVAL;
1142 goto out;
1143 }
1144
1145 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
2b82032c 1146 root->fs_info->fs_devices->rw_devices <= 2) {
d397712b
CM		 1147 printk(KERN_ERR "btrfs: unable to go below two "
1148 "devices on raid1\n");
a061fc8d
CM		 1149 ret = -EINVAL;
1150 goto out;
1151 }
1152
dfe25020 1153 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1154 struct list_head *devices;
1155 struct btrfs_device *tmp;
a061fc8d 1156
dfe25020
CM		 1157 device = NULL;
1158 devices = &root->fs_info->fs_devices->devices;
e5e9a520 1159 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
c6e30871 1160 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1161 if (tmp->in_fs_metadata && !tmp->bdev) {
1162 device = tmp;
1163 break;
1164 }
1165 }
e5e9a520 1166 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
dfe25020
CM		 1167 bdev = NULL;
1168 bh = NULL;
1169 disk_super = NULL;
1170 if (!device) {
d397712b
CM		 1171 printk(KERN_ERR "btrfs: no missing devices found to "
1172 "remove\n");
dfe25020
CM		 1173 goto out;
1174 }
dfe25020 1175 } else {
97288f2c 1176 bdev = open_bdev_exclusive(device_path, FMODE_READ,
dfe25020
CM		 1177 root->fs_info->bdev_holder);
1178 if (IS_ERR(bdev)) {
1179 ret = PTR_ERR(bdev);
1180 goto out;
1181 }
a061fc8d 1182
2b82032c 1183 set_blocksize(bdev, 4096);
a512bbf8 1184 bh = btrfs_read_dev_super(bdev);
dfe25020
CM		 1185 if (!bh) {
1186 ret = -EIO;
1187 goto error_close;
1188 }
1189 disk_super = (struct btrfs_super_block *)bh->b_data;
dfe25020 1190 devid = le64_to_cpu(disk_super->dev_item.devid);
2b82032c
YZ		 1191 dev_uuid = disk_super->dev_item.uuid;
1192 device = btrfs_find_device(root, devid, dev_uuid,
1193 disk_super->fsid);
dfe25020
CM		 1194 if (!device) {
1195 ret = -ENOENT;
1196 goto error_brelse;
1197 }
2b82032c 1198 }
dfe25020 1199
2b82032c 1200 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1201 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1202 "device\n");
2b82032c
YZ		 1203 ret = -EINVAL;
1204 goto error_brelse;
1205 }
1206
1207 if (device->writeable) {
1208 list_del_init(&device->dev_alloc_list);
1209 root->fs_info->fs_devices->rw_devices--;
dfe25020 1210 }
a061fc8d
CM		 1211
1212 ret = btrfs_shrink_device(device, 0);
1213 if (ret)
1214 goto error_brelse;
1215
a061fc8d
CM		 1216 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1217 if (ret)
1218 goto error_brelse;
1219
2b82032c 1220 device->in_fs_metadata = 0;
e5e9a520
CM		 1221
1222 /*
1223 * the device list mutex makes sure that we don't change
1224 * the device list while someone else is writing out all
1225 * the device supers.
1226 */
1227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1228 list_del_init(&device->dev_list);
e5e9a520
CM		 1229 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1230
e4404d6e 1231 device->fs_devices->num_devices--;
2b82032c
YZ		 1232
1233 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1234 struct btrfs_device, dev_list);
1235 if (device->bdev == root->fs_info->sb->s_bdev)
1236 root->fs_info->sb->s_bdev = next_device->bdev;
1237 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1238 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1239
e4404d6e
YZ		 1240 if (device->bdev) {
1241 close_bdev_exclusive(device->bdev, device->mode);
1242 device->bdev = NULL;
1243 device->fs_devices->open_devices--;
1244 }
1245
2b82032c
YZ		 1246 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1247 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1248
e4404d6e
YZ		 1249 if (device->fs_devices->open_devices == 0) {
1250 struct btrfs_fs_devices *fs_devices;
1251 fs_devices = root->fs_info->fs_devices;
1252 while (fs_devices) {
1253 if (fs_devices->seed == device->fs_devices)
1254 break;
1255 fs_devices = fs_devices->seed;
2b82032c 1256 }
e4404d6e
YZ		 1257 fs_devices->seed = device->fs_devices->seed;
1258 device->fs_devices->seed = NULL;
1259 __btrfs_close_devices(device->fs_devices);
1260 free_fs_devices(device->fs_devices);
2b82032c
YZ		 1261 }
1262
1263 /*
1264 * at this point, the device is zero sized. We want to
1265 * remove it from the devices list and zero out the old super
1266 */
1267 if (device->writeable) {
dfe25020
CM		 1268 /* make sure this device isn't detected as part of
1269 * the FS anymore
1270 */
1271 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1272 set_buffer_dirty(bh);
1273 sync_dirty_buffer(bh);
dfe25020 1274 }
a061fc8d
CM		 1275
1276 kfree(device->name);
1277 kfree(device);
1278 ret = 0;
a061fc8d
CM		 1279
1280error_brelse:
1281 brelse(bh);
1282error_close:
dfe25020 1283 if (bdev)
97288f2c 1284 close_bdev_exclusive(bdev, FMODE_READ);
a061fc8d 1285out:
7d9eb12c 1286 mutex_unlock(&root->fs_info->volume_mutex);
a061fc8d 1287 mutex_unlock(&uuid_mutex);
a061fc8d
CM		 1288 return ret;
1289}
1290
2b82032c
YZ		 1291/*
1292 * does all the dirty work required for changing file system's UUID.
1293 */
1294static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1295 struct btrfs_root *root)
1296{
1297 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1298 struct btrfs_fs_devices *old_devices;
e4404d6e 1299 struct btrfs_fs_devices *seed_devices;
2b82032c
YZ		 1300 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1301 struct btrfs_device *device;
1302 u64 super_flags;
1303
1304 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1305 if (!fs_devices->seeding)
2b82032c
YZ		 1306 return -EINVAL;
1307
e4404d6e
YZ		 1308 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1309 if (!seed_devices)
2b82032c
YZ		 1310 return -ENOMEM;
1311
e4404d6e
YZ		 1312 old_devices = clone_fs_devices(fs_devices);
1313 if (IS_ERR(old_devices)) {
1314 kfree(seed_devices);
1315 return PTR_ERR(old_devices);
2b82032c 1316 }
e4404d6e 1317
2b82032c
YZ		 1318 list_add(&old_devices->list, &fs_uuids);
1319
e4404d6e
YZ		 1320 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1321 seed_devices->opened = 1;
1322 INIT_LIST_HEAD(&seed_devices->devices);
1323 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1324 mutex_init(&seed_devices->device_list_mutex);
e4404d6e
YZ		 1325 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1326 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1327 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1328 device->fs_devices = seed_devices;
1329 }
1330
2b82032c
YZ		 1331 fs_devices->seeding = 0;
1332 fs_devices->num_devices = 0;
1333 fs_devices->open_devices = 0;
e4404d6e 1334 fs_devices->seed = seed_devices;
2b82032c
YZ		 1335
1336 generate_random_uuid(fs_devices->fsid);
1337 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1338 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1339 super_flags = btrfs_super_flags(disk_super) &
1340 ~BTRFS_SUPER_FLAG_SEEDING;
1341 btrfs_set_super_flags(disk_super, super_flags);
1342
1343 return 0;
1344}
1345
1346/*
1347 * strore the expected generation for seed devices in device items.
1348 */
1349static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1350 struct btrfs_root *root)
1351{
1352 struct btrfs_path *path;
1353 struct extent_buffer *leaf;
1354 struct btrfs_dev_item *dev_item;
1355 struct btrfs_device *device;
1356 struct btrfs_key key;
1357 u8 fs_uuid[BTRFS_UUID_SIZE];
1358 u8 dev_uuid[BTRFS_UUID_SIZE];
1359 u64 devid;
1360 int ret;
1361
1362 path = btrfs_alloc_path();
1363 if (!path)
1364 return -ENOMEM;
1365
1366 root = root->fs_info->chunk_root;
1367 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1368 key.offset = 0;
1369 key.type = BTRFS_DEV_ITEM_KEY;
1370
1371 while (1) {
1372 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1373 if (ret < 0)
1374 goto error;
1375
1376 leaf = path->nodes[0];
1377next_slot:
1378 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1379 ret = btrfs_next_leaf(root, path);
1380 if (ret > 0)
1381 break;
1382 if (ret < 0)
1383 goto error;
1384 leaf = path->nodes[0];
1385 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1386 btrfs_release_path(root, path);
1387 continue;
1388 }
1389
1390 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1391 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1392 key.type != BTRFS_DEV_ITEM_KEY)
1393 break;
1394
1395 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1396 struct btrfs_dev_item);
1397 devid = btrfs_device_id(leaf, dev_item);
1398 read_extent_buffer(leaf, dev_uuid,
1399 (unsigned long)btrfs_device_uuid(dev_item),
1400 BTRFS_UUID_SIZE);
1401 read_extent_buffer(leaf, fs_uuid,
1402 (unsigned long)btrfs_device_fsid(dev_item),
1403 BTRFS_UUID_SIZE);
1404 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1405 BUG_ON(!device);
1406
1407 if (device->fs_devices->seeding) {
1408 btrfs_set_device_generation(leaf, dev_item,
1409 device->generation);
1410 btrfs_mark_buffer_dirty(leaf);
1411 }
1412
1413 path->slots[0]++;
1414 goto next_slot;
1415 }
1416 ret = 0;
1417error:
1418 btrfs_free_path(path);
1419 return ret;
1420}
1421
788f20eb
CM		 1422int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1423{
1424 struct btrfs_trans_handle *trans;
1425 struct btrfs_device *device;
1426 struct block_device *bdev;
788f20eb 1427 struct list_head *devices;
2b82032c 1428 struct super_block *sb = root->fs_info->sb;
788f20eb 1429 u64 total_bytes;
2b82032c 1430 int seeding_dev = 0;
788f20eb
CM		 1431 int ret = 0;
1432
2b82032c
YZ		 1433 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1434 return -EINVAL;
788f20eb 1435
15916de8 1436 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
d397712b 1437 if (!bdev)
788f20eb 1438 return -EIO;
a2135011 1439
2b82032c
YZ		 1440 if (root->fs_info->fs_devices->seeding) {
1441 seeding_dev = 1;
1442 down_write(&sb->s_umount);
1443 mutex_lock(&uuid_mutex);
1444 }
1445
8c8bee1d 1446 filemap_write_and_wait(bdev->bd_inode->i_mapping);
7d9eb12c 1447 mutex_lock(&root->fs_info->volume_mutex);
a2135011 1448
788f20eb 1449 devices = &root->fs_info->fs_devices->devices;
e5e9a520
CM		 1450 /*
1451 * we have the volume lock, so we don't need the extra
1452 * device list mutex while reading the list here.
1453 */
c6e30871 1454 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1455 if (device->bdev == bdev) {
1456 ret = -EEXIST;
2b82032c 1457 goto error;
788f20eb
CM		 1458 }
1459 }
1460
1461 device = kzalloc(sizeof(*device), GFP_NOFS);
1462 if (!device) {
1463 /* we can safely leave the fs_devices entry around */
1464 ret = -ENOMEM;
2b82032c 1465 goto error;
788f20eb
CM		 1466 }
1467
788f20eb
CM		 1468 device->name = kstrdup(device_path, GFP_NOFS);
1469 if (!device->name) {
1470 kfree(device);
2b82032c
YZ		 1471 ret = -ENOMEM;
1472 goto error;
788f20eb 1473 }
2b82032c
YZ		 1474
1475 ret = find_next_devid(root, &device->devid);
1476 if (ret) {
1477 kfree(device);
1478 goto error;
1479 }
1480
1481 trans = btrfs_start_transaction(root, 1);
1482 lock_chunks(root);
1483
1484 device->barriers = 1;
1485 device->writeable = 1;
1486 device->work.func = pending_bios_fn;
1487 generate_random_uuid(device->uuid);
1488 spin_lock_init(&device->io_lock);
1489 device->generation = trans->transid;
788f20eb
CM		 1490 device->io_width = root->sectorsize;
1491 device->io_align = root->sectorsize;
1492 device->sector_size = root->sectorsize;
1493 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1494 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1495 device->dev_root = root->fs_info->dev_root;
1496 device->bdev = bdev;
dfe25020 1497 device->in_fs_metadata = 1;
15916de8 1498 device->mode = 0;
2b82032c 1499 set_blocksize(device->bdev, 4096);
788f20eb 1500
2b82032c
YZ		 1501 if (seeding_dev) {
1502 sb->s_flags &= ~MS_RDONLY;
1503 ret = btrfs_prepare_sprout(trans, root);
1504 BUG_ON(ret);
1505 }
788f20eb 1506
2b82032c 1507 device->fs_devices = root->fs_info->fs_devices;
e5e9a520
CM		 1508
1509 /*
1510 * we don't want write_supers to jump in here with our device
1511 * half setup
1512 */
1513 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2b82032c
YZ		 1514 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1515 list_add(&device->dev_alloc_list,
1516 &root->fs_info->fs_devices->alloc_list);
1517 root->fs_info->fs_devices->num_devices++;
1518 root->fs_info->fs_devices->open_devices++;
1519 root->fs_info->fs_devices->rw_devices++;
1520 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1521
c289811c
CM		 1522 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1523 root->fs_info->fs_devices->rotating = 1;
1524
788f20eb
CM		 1525 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1526 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1527 total_bytes + device->total_bytes);
1528
1529 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1530 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1531 total_bytes + 1);
e5e9a520 1532 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 1533
2b82032c
YZ		 1534 if (seeding_dev) {
1535 ret = init_first_rw_device(trans, root, device);
1536 BUG_ON(ret);
1537 ret = btrfs_finish_sprout(trans, root);
1538 BUG_ON(ret);
1539 } else {
1540 ret = btrfs_add_device(trans, root, device);
1541 }
1542
913d952e
CM		 1543 /*
1544 * we've got more storage, clear any full flags on the space
1545 * infos
1546 */
1547 btrfs_clear_space_info_full(root->fs_info);
1548
7d9eb12c 1549 unlock_chunks(root);
2b82032c 1550 btrfs_commit_transaction(trans, root);
a2135011 1551
2b82032c
YZ		 1552 if (seeding_dev) {
1553 mutex_unlock(&uuid_mutex);
1554 up_write(&sb->s_umount);
788f20eb 1555
2b82032c
YZ		 1556 ret = btrfs_relocate_sys_chunks(root);
1557 BUG_ON(ret);
1558 }
1559out:
1560 mutex_unlock(&root->fs_info->volume_mutex);
1561 return ret;
1562error:
15916de8 1563 close_bdev_exclusive(bdev, 0);
2b82032c
YZ		 1564 if (seeding_dev) {
1565 mutex_unlock(&uuid_mutex);
1566 up_write(&sb->s_umount);
1567 }
788f20eb
CM		 1568 goto out;
1569}
1570
d397712b
CM		 1571static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1572 struct btrfs_device *device)
0b86a832
CM		 1573{
1574 int ret;
1575 struct btrfs_path *path;
1576 struct btrfs_root *root;
1577 struct btrfs_dev_item *dev_item;
1578 struct extent_buffer *leaf;
1579 struct btrfs_key key;
1580
1581 root = device->dev_root->fs_info->chunk_root;
1582
1583 path = btrfs_alloc_path();
1584 if (!path)
1585 return -ENOMEM;
1586
1587 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1588 key.type = BTRFS_DEV_ITEM_KEY;
1589 key.offset = device->devid;
1590
1591 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1592 if (ret < 0)
1593 goto out;
1594
1595 if (ret > 0) {
1596 ret = -ENOENT;
1597 goto out;
1598 }
1599
1600 leaf = path->nodes[0];
1601 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1602
1603 btrfs_set_device_id(leaf, dev_item, device->devid);
1604 btrfs_set_device_type(leaf, dev_item, device->type);
1605 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1606 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1607 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 1608 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 1609 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1610 btrfs_mark_buffer_dirty(leaf);
1611
1612out:
1613 btrfs_free_path(path);
1614 return ret;
1615}
1616
7d9eb12c 1617static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1618 struct btrfs_device *device, u64 new_size)
1619{
1620 struct btrfs_super_block *super_copy =
1621 &device->dev_root->fs_info->super_copy;
1622 u64 old_total = btrfs_super_total_bytes(super_copy);
1623 u64 diff = new_size - device->total_bytes;
1624
2b82032c
YZ		 1625 if (!device->writeable)
1626 return -EACCES;
1627 if (new_size <= device->total_bytes)
1628 return -EINVAL;
1629
8f18cf13 1630 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1631 device->fs_devices->total_rw_bytes += diff;
1632
1633 device->total_bytes = new_size;
9779b72f 1634 device->disk_total_bytes = new_size;
4184ea7f
CM		 1635 btrfs_clear_space_info_full(device->dev_root->fs_info);
1636
8f18cf13
CM		 1637 return btrfs_update_device(trans, device);
1638}
1639
7d9eb12c
CM		 1640int btrfs_grow_device(struct btrfs_trans_handle *trans,
1641 struct btrfs_device *device, u64 new_size)
1642{
1643 int ret;
1644 lock_chunks(device->dev_root);
1645 ret = __btrfs_grow_device(trans, device, new_size);
1646 unlock_chunks(device->dev_root);
1647 return ret;
1648}
1649
8f18cf13
CM		 1650static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1651 struct btrfs_root *root,
1652 u64 chunk_tree, u64 chunk_objectid,
1653 u64 chunk_offset)
1654{
1655 int ret;
1656 struct btrfs_path *path;
1657 struct btrfs_key key;
1658
1659 root = root->fs_info->chunk_root;
1660 path = btrfs_alloc_path();
1661 if (!path)
1662 return -ENOMEM;
1663
1664 key.objectid = chunk_objectid;
1665 key.offset = chunk_offset;
1666 key.type = BTRFS_CHUNK_ITEM_KEY;
1667
1668 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1669 BUG_ON(ret);
1670
1671 ret = btrfs_del_item(trans, root, path);
1672 BUG_ON(ret);
1673
1674 btrfs_free_path(path);
1675 return 0;
1676}
1677
b2950863 1678static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1679 chunk_offset)
1680{
1681 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1682 struct btrfs_disk_key *disk_key;
1683 struct btrfs_chunk *chunk;
1684 u8 *ptr;
1685 int ret = 0;
1686 u32 num_stripes;
1687 u32 array_size;
1688 u32 len = 0;
1689 u32 cur;
1690 struct btrfs_key key;
1691
1692 array_size = btrfs_super_sys_array_size(super_copy);
1693
1694 ptr = super_copy->sys_chunk_array;
1695 cur = 0;
1696
1697 while (cur < array_size) {
1698 disk_key = (struct btrfs_disk_key *)ptr;
1699 btrfs_disk_key_to_cpu(&key, disk_key);
1700
1701 len = sizeof(*disk_key);
1702
1703 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1704 chunk = (struct btrfs_chunk *)(ptr + len);
1705 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1706 len += btrfs_chunk_item_size(num_stripes);
1707 } else {
1708 ret = -EIO;
1709 break;
1710 }
1711 if (key.objectid == chunk_objectid &&
1712 key.offset == chunk_offset) {
1713 memmove(ptr, ptr + len, array_size - (cur + len));
1714 array_size -= len;
1715 btrfs_set_super_sys_array_size(super_copy, array_size);
1716 } else {
1717 ptr += len;
1718 cur += len;
1719 }
1720 }
1721 return ret;
1722}
1723
b2950863 1724static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1725 u64 chunk_tree, u64 chunk_objectid,
1726 u64 chunk_offset)
1727{
1728 struct extent_map_tree *em_tree;
1729 struct btrfs_root *extent_root;
1730 struct btrfs_trans_handle *trans;
1731 struct extent_map *em;
1732 struct map_lookup *map;
1733 int ret;
1734 int i;
1735
1736 root = root->fs_info->chunk_root;
1737 extent_root = root->fs_info->extent_root;
1738 em_tree = &root->fs_info->mapping_tree.map_tree;
1739
ba1bf481
JB		 1740 ret = btrfs_can_relocate(extent_root, chunk_offset);
1741 if (ret)
1742 return -ENOSPC;
1743
8f18cf13 1744 /* step one, relocate all the extents inside this chunk */
1a40e23b 1745 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
8f18cf13
CM		 1746 BUG_ON(ret);
1747
1748 trans = btrfs_start_transaction(root, 1);
1749 BUG_ON(!trans);
1750
7d9eb12c
CM		 1751 lock_chunks(root);
1752
8f18cf13
CM		 1753 /*
1754 * step two, delete the device extents and the
1755 * chunk tree entries
1756 */
890871be 1757 read_lock(&em_tree->lock);
8f18cf13 1758 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 1759 read_unlock(&em_tree->lock);
8f18cf13 1760
a061fc8d
CM		 1761 BUG_ON(em->start > chunk_offset ||
1762 em->start + em->len < chunk_offset);
8f18cf13
CM		 1763 map = (struct map_lookup *)em->bdev;
1764
1765 for (i = 0; i < map->num_stripes; i++) {
1766 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1767 map->stripes[i].physical);
1768 BUG_ON(ret);
a061fc8d 1769
dfe25020
CM		 1770 if (map->stripes[i].dev) {
1771 ret = btrfs_update_device(trans, map->stripes[i].dev);
1772 BUG_ON(ret);
1773 }
8f18cf13
CM		 1774 }
1775 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1776 chunk_offset);
1777
1778 BUG_ON(ret);
1779
1780 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1781 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1782 BUG_ON(ret);
8f18cf13
CM		 1783 }
1784
2b82032c
YZ		 1785 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1786 BUG_ON(ret);
1787
890871be 1788 write_lock(&em_tree->lock);
2b82032c 1789 remove_extent_mapping(em_tree, em);
890871be 1790 write_unlock(&em_tree->lock);
2b82032c
YZ		 1791
1792 kfree(map);
1793 em->bdev = NULL;
1794
1795 /* once for the tree */
1796 free_extent_map(em);
1797 /* once for us */
1798 free_extent_map(em);
1799
1800 unlock_chunks(root);
1801 btrfs_end_transaction(trans, root);
1802 return 0;
1803}
1804
1805static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1806{
1807 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1808 struct btrfs_path *path;
1809 struct extent_buffer *leaf;
1810 struct btrfs_chunk *chunk;
1811 struct btrfs_key key;
1812 struct btrfs_key found_key;
1813 u64 chunk_tree = chunk_root->root_key.objectid;
1814 u64 chunk_type;
ba1bf481
JB		 1815 bool retried = false;
1816 int failed = 0;
2b82032c
YZ		 1817 int ret;
1818
1819 path = btrfs_alloc_path();
1820 if (!path)
1821 return -ENOMEM;
1822
ba1bf481 1823again:
2b82032c
YZ		 1824 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1825 key.offset = (u64)-1;
1826 key.type = BTRFS_CHUNK_ITEM_KEY;
1827
1828 while (1) {
1829 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1830 if (ret < 0)
1831 goto error;
1832 BUG_ON(ret == 0);
1833
1834 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1835 key.type);
1836 if (ret < 0)
1837 goto error;
1838 if (ret > 0)
1839 break;
1a40e23b 1840
2b82032c
YZ		 1841 leaf = path->nodes[0];
1842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 1843
2b82032c
YZ		 1844 chunk = btrfs_item_ptr(leaf, path->slots[0],
1845 struct btrfs_chunk);
1846 chunk_type = btrfs_chunk_type(leaf, chunk);
1847 btrfs_release_path(chunk_root, path);
8f18cf13 1848
2b82032c
YZ		 1849 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1850 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1851 found_key.objectid,
1852 found_key.offset);
ba1bf481
JB		 1853 if (ret == -ENOSPC)
1854 failed++;
1855 else if (ret)
1856 BUG();
2b82032c 1857 }
8f18cf13 1858
2b82032c
YZ		 1859 if (found_key.offset == 0)
1860 break;
1861 key.offset = found_key.offset - 1;
1862 }
1863 ret = 0;
ba1bf481
JB		 1864 if (failed && !retried) {
1865 failed = 0;
1866 retried = true;
1867 goto again;
1868 } else if (failed && retried) {
1869 WARN_ON(1);
1870 ret = -ENOSPC;
1871 }
2b82032c
YZ		 1872error:
1873 btrfs_free_path(path);
1874 return ret;
8f18cf13
CM		 1875}
1876
ec44a35c
CM		 1877static u64 div_factor(u64 num, int factor)
1878{
1879 if (factor == 10)
1880 return num;
1881 num *= factor;
1882 do_div(num, 10);
1883 return num;
1884}
1885
ec44a35c
CM		 1886int btrfs_balance(struct btrfs_root *dev_root)
1887{
1888 int ret;
ec44a35c
CM		 1889 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1890 struct btrfs_device *device;
1891 u64 old_size;
1892 u64 size_to_free;
1893 struct btrfs_path *path;
1894 struct btrfs_key key;
1895 struct btrfs_chunk *chunk;
1896 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1897 struct btrfs_trans_handle *trans;
1898 struct btrfs_key found_key;
1899
2b82032c
YZ		 1900 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1901 return -EROFS;
ec44a35c 1902
7d9eb12c 1903 mutex_lock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1904 dev_root = dev_root->fs_info->dev_root;
1905
ec44a35c 1906 /* step one make some room on all the devices */
c6e30871 1907 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 1908 old_size = device->total_bytes;
1909 size_to_free = div_factor(old_size, 1);
1910 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 1911 if (!device->writeable ||
1912 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 1913 continue;
1914
1915 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 1916 if (ret == -ENOSPC)
1917 break;
ec44a35c
CM		 1918 BUG_ON(ret);
1919
1920 trans = btrfs_start_transaction(dev_root, 1);
1921 BUG_ON(!trans);
1922
1923 ret = btrfs_grow_device(trans, device, old_size);
1924 BUG_ON(ret);
1925
1926 btrfs_end_transaction(trans, dev_root);
1927 }
1928
1929 /* step two, relocate all the chunks */
1930 path = btrfs_alloc_path();
1931 BUG_ON(!path);
1932
1933 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1934 key.offset = (u64)-1;
1935 key.type = BTRFS_CHUNK_ITEM_KEY;
1936
d397712b 1937 while (1) {
ec44a35c
CM		 1938 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1939 if (ret < 0)
1940 goto error;
1941
1942 /*
1943 * this shouldn't happen, it means the last relocate
1944 * failed
1945 */
1946 if (ret == 0)
1947 break;
1948
1949 ret = btrfs_previous_item(chunk_root, path, 0,
1950 BTRFS_CHUNK_ITEM_KEY);
7d9eb12c 1951 if (ret)
ec44a35c 1952 break;
7d9eb12c 1953
ec44a35c
CM		 1954 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1955 path->slots[0]);
1956 if (found_key.objectid != key.objectid)
1957 break;
7d9eb12c 1958
ec44a35c
CM		 1959 chunk = btrfs_item_ptr(path->nodes[0],
1960 path->slots[0],
1961 struct btrfs_chunk);
ec44a35c 1962 /* chunk zero is special */
ba1bf481 1963 if (found_key.offset == 0)
ec44a35c
CM		 1964 break;
1965
7d9eb12c 1966 btrfs_release_path(chunk_root, path);
ec44a35c
CM		 1967 ret = btrfs_relocate_chunk(chunk_root,
1968 chunk_root->root_key.objectid,
1969 found_key.objectid,
1970 found_key.offset);
ba1bf481
JB		 1971 BUG_ON(ret && ret != -ENOSPC);
1972 key.offset = found_key.offset - 1;
ec44a35c
CM		 1973 }
1974 ret = 0;
1975error:
1976 btrfs_free_path(path);
7d9eb12c 1977 mutex_unlock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1978 return ret;
1979}
1980
8f18cf13
CM		 1981/*
1982 * shrinking a device means finding all of the device extents past
1983 * the new size, and then following the back refs to the chunks.
1984 * The chunk relocation code actually frees the device extent
1985 */
1986int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1987{
1988 struct btrfs_trans_handle *trans;
1989 struct btrfs_root *root = device->dev_root;
1990 struct btrfs_dev_extent *dev_extent = NULL;
1991 struct btrfs_path *path;
1992 u64 length;
1993 u64 chunk_tree;
1994 u64 chunk_objectid;
1995 u64 chunk_offset;
1996 int ret;
1997 int slot;
ba1bf481
JB		 1998 int failed = 0;
1999 bool retried = false;
8f18cf13
CM		 2000 struct extent_buffer *l;
2001 struct btrfs_key key;
2002 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2003 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 2004 u64 old_size = device->total_bytes;
8f18cf13
CM		 2005 u64 diff = device->total_bytes - new_size;
2006
2b82032c
YZ		 2007 if (new_size >= device->total_bytes)
2008 return -EINVAL;
8f18cf13
CM		 2009
2010 path = btrfs_alloc_path();
2011 if (!path)
2012 return -ENOMEM;
2013
8f18cf13
CM		 2014 path->reada = 2;
2015
7d9eb12c
CM		 2016 lock_chunks(root);
2017
8f18cf13 2018 device->total_bytes = new_size;
2b82032c
YZ		 2019 if (device->writeable)
2020 device->fs_devices->total_rw_bytes -= diff;
7d9eb12c 2021 unlock_chunks(root);
8f18cf13 2022
ba1bf481 2023again:
8f18cf13
CM		 2024 key.objectid = device->devid;
2025 key.offset = (u64)-1;
2026 key.type = BTRFS_DEV_EXTENT_KEY;
2027
2028 while (1) {
2029 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2030 if (ret < 0)
2031 goto done;
2032
2033 ret = btrfs_previous_item(root, path, 0, key.type);
2034 if (ret < 0)
2035 goto done;
2036 if (ret) {
2037 ret = 0;
ba1bf481 2038 btrfs_release_path(root, path);
bf1fb512 2039 break;
8f18cf13
CM		 2040 }
2041
2042 l = path->nodes[0];
2043 slot = path->slots[0];
2044 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2045
ba1bf481
JB		 2046 if (key.objectid != device->devid) {
2047 btrfs_release_path(root, path);
bf1fb512 2048 break;
ba1bf481 2049 }
8f18cf13
CM		 2050
2051 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2052 length = btrfs_dev_extent_length(l, dev_extent);
2053
ba1bf481
JB		 2054 if (key.offset + length <= new_size) {
2055 btrfs_release_path(root, path);
d6397bae 2056 break;
ba1bf481 2057 }
8f18cf13
CM		 2058
2059 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2060 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2061 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2062 btrfs_release_path(root, path);
2063
2064 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2065 chunk_offset);
ba1bf481 2066 if (ret && ret != -ENOSPC)
8f18cf13 2067 goto done;
ba1bf481
JB		 2068 if (ret == -ENOSPC)
2069 failed++;
2070 key.offset -= 1;
2071 }
2072
2073 if (failed && !retried) {
2074 failed = 0;
2075 retried = true;
2076 goto again;
2077 } else if (failed && retried) {
2078 ret = -ENOSPC;
2079 lock_chunks(root);
2080
2081 device->total_bytes = old_size;
2082 if (device->writeable)
2083 device->fs_devices->total_rw_bytes += diff;
2084 unlock_chunks(root);
2085 goto done;
8f18cf13
CM		 2086 }
2087
d6397bae
CB		 2088 /* Shrinking succeeded, else we would be at "done". */
2089 trans = btrfs_start_transaction(root, 1);
2090 if (!trans) {
2091 ret = -ENOMEM;
2092 goto done;
2093 }
2094 lock_chunks(root);
2095
2096 device->disk_total_bytes = new_size;
2097 /* Now btrfs_update_device() will change the on-disk size. */
2098 ret = btrfs_update_device(trans, device);
2099 if (ret) {
2100 unlock_chunks(root);
2101 btrfs_end_transaction(trans, root);
2102 goto done;
2103 }
2104 WARN_ON(diff > old_total);
2105 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2106 unlock_chunks(root);
2107 btrfs_end_transaction(trans, root);
8f18cf13
CM		 2108done:
2109 btrfs_free_path(path);
2110 return ret;
2111}
2112
b2950863 2113static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2114 struct btrfs_root *root,
2115 struct btrfs_key *key,
2116 struct btrfs_chunk *chunk, int item_size)
2117{
2118 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2119 struct btrfs_disk_key disk_key;
2120 u32 array_size;
2121 u8 *ptr;
2122
2123 array_size = btrfs_super_sys_array_size(super_copy);
2124 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2125 return -EFBIG;
2126
2127 ptr = super_copy->sys_chunk_array + array_size;
2128 btrfs_cpu_key_to_disk(&disk_key, key);
2129 memcpy(ptr, &disk_key, sizeof(disk_key));
2130 ptr += sizeof(disk_key);
2131 memcpy(ptr, chunk, item_size);
2132 item_size += sizeof(disk_key);
2133 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2134 return 0;
2135}
2136
d397712b 2137static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
a1b32a59 2138 int num_stripes, int sub_stripes)
9b3f68b9
CM		 2139{
2140 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2141 return calc_size;
2142 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2143 return calc_size * (num_stripes / sub_stripes);
2144 else
2145 return calc_size * num_stripes;
2146}
2147
2b82032c
YZ		 2148static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *extent_root,
2150 struct map_lookup **map_ret,
2151 u64 *num_bytes, u64 *stripe_size,
2152 u64 start, u64 type)
0b86a832 2153{
593060d7 2154 struct btrfs_fs_info *info = extent_root->fs_info;
0b86a832 2155 struct btrfs_device *device = NULL;
2b82032c 2156 struct btrfs_fs_devices *fs_devices = info->fs_devices;
6324fbf3 2157 struct list_head *cur;
2b82032c 2158 struct map_lookup *map = NULL;
0b86a832 2159 struct extent_map_tree *em_tree;
0b86a832 2160 struct extent_map *em;
2b82032c 2161 struct list_head private_devs;
a40a90a0 2162 int min_stripe_size = 1 * 1024 * 1024;
0b86a832 2163 u64 calc_size = 1024 * 1024 * 1024;
9b3f68b9
CM		 2164 u64 max_chunk_size = calc_size;
2165 u64 min_free;
6324fbf3
CM		 2166 u64 avail;
2167 u64 max_avail = 0;
2b82032c 2168 u64 dev_offset;
6324fbf3 2169 int num_stripes = 1;
a40a90a0 2170 int min_stripes = 1;
321aecc6 2171 int sub_stripes = 0;
6324fbf3 2172 int looped = 0;
0b86a832 2173 int ret;
6324fbf3 2174 int index;
593060d7 2175 int stripe_len = 64 * 1024;
0b86a832 2176
ec44a35c
CM		 2177 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2178 (type & BTRFS_BLOCK_GROUP_DUP)) {
2179 WARN_ON(1);
2180 type &= ~BTRFS_BLOCK_GROUP_DUP;
2181 }
2b82032c 2182 if (list_empty(&fs_devices->alloc_list))
6324fbf3 2183 return -ENOSPC;
593060d7 2184
a40a90a0 2185 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2b82032c 2186 num_stripes = fs_devices->rw_devices;
a40a90a0
CM		 2187 min_stripes = 2;
2188 }
2189 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
611f0e00 2190 num_stripes = 2;
a40a90a0
CM		 2191 min_stripes = 2;
2192 }
8790d502 2193 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2b82032c 2194 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
9b3f68b9
CM		 2195 if (num_stripes < 2)
2196 return -ENOSPC;
a40a90a0 2197 min_stripes = 2;
8790d502 2198 }
321aecc6 2199 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2b82032c 2200 num_stripes = fs_devices->rw_devices;
321aecc6
CM		 2201 if (num_stripes < 4)
2202 return -ENOSPC;
2203 num_stripes &= ~(u32)1;
2204 sub_stripes = 2;
a40a90a0 2205 min_stripes = 4;
321aecc6 2206 }
9b3f68b9
CM		 2207
2208 if (type & BTRFS_BLOCK_GROUP_DATA) {
2209 max_chunk_size = 10 * calc_size;
a40a90a0 2210 min_stripe_size = 64 * 1024 * 1024;
9b3f68b9 2211 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
83d3c969 2212 max_chunk_size = 256 * 1024 * 1024;
a40a90a0
CM		 2213 min_stripe_size = 32 * 1024 * 1024;
2214 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2215 calc_size = 8 * 1024 * 1024;
2216 max_chunk_size = calc_size * 2;
2217 min_stripe_size = 1 * 1024 * 1024;
9b3f68b9
CM		 2218 }
2219
2b82032c
YZ		 2220 /* we don't want a chunk larger than 10% of writeable space */
2221 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2222 max_chunk_size);
9b3f68b9 2223
a40a90a0 2224again:
9779b72f 2225 max_avail = 0;
2b82032c
YZ		 2226 if (!map || map->num_stripes != num_stripes) {
2227 kfree(map);
2228 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2229 if (!map)
2230 return -ENOMEM;
2231 map->num_stripes = num_stripes;
2232 }
2233
9b3f68b9
CM		 2234 if (calc_size * num_stripes > max_chunk_size) {
2235 calc_size = max_chunk_size;
2236 do_div(calc_size, num_stripes);
2237 do_div(calc_size, stripe_len);
2238 calc_size *= stripe_len;
2239 }
2240 /* we don't want tiny stripes */
a40a90a0 2241 calc_size = max_t(u64, min_stripe_size, calc_size);
9b3f68b9 2242
9b3f68b9
CM		 2243 do_div(calc_size, stripe_len);
2244 calc_size *= stripe_len;
2245
2b82032c 2246 cur = fs_devices->alloc_list.next;
6324fbf3 2247 index = 0;
611f0e00
CM		 2248
2249 if (type & BTRFS_BLOCK_GROUP_DUP)
2250 min_free = calc_size * 2;
9b3f68b9
CM		 2251 else
2252 min_free = calc_size;
611f0e00 2253
0f9dd46c
JB		 2254 /*
2255 * we add 1MB because we never use the first 1MB of the device, unless
2256 * we've looped, then we are likely allocating the maximum amount of
2257 * space left already
2258 */
2259 if (!looped)
2260 min_free += 1024 * 1024;
ad5bd91e 2261
2b82032c 2262 INIT_LIST_HEAD(&private_devs);
d397712b 2263 while (index < num_stripes) {
b3075717 2264 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2b82032c 2265 BUG_ON(!device->writeable);
dfe25020
CM		 2266 if (device->total_bytes > device->bytes_used)
2267 avail = device->total_bytes - device->bytes_used;
2268 else
2269 avail = 0;
6324fbf3 2270 cur = cur->next;
8f18cf13 2271
dfe25020 2272 if (device->in_fs_metadata && avail >= min_free) {
2b82032c 2273 ret = find_free_dev_extent(trans, device,
9779b72f
CM		 2274 min_free, &dev_offset,
2275 &max_avail);
8f18cf13
CM		 2276 if (ret == 0) {
2277 list_move_tail(&device->dev_alloc_list,
2278 &private_devs);
2b82032c
YZ		 2279 map->stripes[index].dev = device;
2280 map->stripes[index].physical = dev_offset;
611f0e00 2281 index++;
2b82032c
YZ		 2282 if (type & BTRFS_BLOCK_GROUP_DUP) {
2283 map->stripes[index].dev = device;
2284 map->stripes[index].physical =
2285 dev_offset + calc_size;
8f18cf13 2286 index++;
2b82032c 2287 }
8f18cf13 2288 }
dfe25020 2289 } else if (device->in_fs_metadata && avail > max_avail)
a40a90a0 2290 max_avail = avail;
2b82032c 2291 if (cur == &fs_devices->alloc_list)
6324fbf3
CM		 2292 break;
2293 }
2b82032c 2294 list_splice(&private_devs, &fs_devices->alloc_list);
6324fbf3 2295 if (index < num_stripes) {
a40a90a0
CM		 2296 if (index >= min_stripes) {
2297 num_stripes = index;
2298 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2299 num_stripes /= sub_stripes;
2300 num_stripes *= sub_stripes;
2301 }
2302 looped = 1;
2303 goto again;
2304 }
6324fbf3
CM		 2305 if (!looped && max_avail > 0) {
2306 looped = 1;
2307 calc_size = max_avail;
2308 goto again;
2309 }
2b82032c 2310 kfree(map);
6324fbf3
CM		 2311 return -ENOSPC;
2312 }
2b82032c
YZ		 2313 map->sector_size = extent_root->sectorsize;
2314 map->stripe_len = stripe_len;
2315 map->io_align = stripe_len;
2316 map->io_width = stripe_len;
2317 map->type = type;
2318 map->num_stripes = num_stripes;
2319 map->sub_stripes = sub_stripes;
0b86a832 2320
2b82032c
YZ		 2321 *map_ret = map;
2322 *stripe_size = calc_size;
2323 *num_bytes = chunk_bytes_by_type(type, calc_size,
2324 num_stripes, sub_stripes);
0b86a832 2325
2b82032c
YZ		 2326 em = alloc_extent_map(GFP_NOFS);
2327 if (!em) {
2328 kfree(map);
593060d7
CM		 2329 return -ENOMEM;
2330 }
2b82032c
YZ		 2331 em->bdev = (struct block_device *)map;
2332 em->start = start;
2333 em->len = *num_bytes;
2334 em->block_start = 0;
2335 em->block_len = em->len;
593060d7 2336
2b82032c 2337 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 2338 write_lock(&em_tree->lock);
2b82032c 2339 ret = add_extent_mapping(em_tree, em);
890871be 2340 write_unlock(&em_tree->lock);
2b82032c
YZ		 2341 BUG_ON(ret);
2342 free_extent_map(em);
0b86a832 2343
2b82032c
YZ		 2344 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2345 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2346 start, *num_bytes);
2347 BUG_ON(ret);
611f0e00 2348
2b82032c
YZ		 2349 index = 0;
2350 while (index < map->num_stripes) {
2351 device = map->stripes[index].dev;
2352 dev_offset = map->stripes[index].physical;
0b86a832
CM		 2353
2354 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2355 info->chunk_root->root_key.objectid,
2356 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2357 start, dev_offset, calc_size);
0b86a832 2358 BUG_ON(ret);
2b82032c
YZ		 2359 index++;
2360 }
2361
2362 return 0;
2363}
2364
2365static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2366 struct btrfs_root *extent_root,
2367 struct map_lookup *map, u64 chunk_offset,
2368 u64 chunk_size, u64 stripe_size)
2369{
2370 u64 dev_offset;
2371 struct btrfs_key key;
2372 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2373 struct btrfs_device *device;
2374 struct btrfs_chunk *chunk;
2375 struct btrfs_stripe *stripe;
2376 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2377 int index = 0;
2378 int ret;
2379
2380 chunk = kzalloc(item_size, GFP_NOFS);
2381 if (!chunk)
2382 return -ENOMEM;
2383
2384 index = 0;
2385 while (index < map->num_stripes) {
2386 device = map->stripes[index].dev;
2387 device->bytes_used += stripe_size;
0b86a832
CM		 2388 ret = btrfs_update_device(trans, device);
2389 BUG_ON(ret);
2b82032c
YZ		 2390 index++;
2391 }
2392
2393 index = 0;
2394 stripe = &chunk->stripe;
2395 while (index < map->num_stripes) {
2396 device = map->stripes[index].dev;
2397 dev_offset = map->stripes[index].physical;
0b86a832 2398
e17cade2
CM		 2399 btrfs_set_stack_stripe_devid(stripe, device->devid);
2400 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2401 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2402 stripe++;
0b86a832
CM		 2403 index++;
2404 }
2405
2b82032c 2406 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2407 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2408 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2409 btrfs_set_stack_chunk_type(chunk, map->type);
2410 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2411 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2412 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2413 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2414 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2415
2b82032c
YZ		 2416 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2417 key.type = BTRFS_CHUNK_ITEM_KEY;
2418 key.offset = chunk_offset;
0b86a832 2419
2b82032c
YZ		 2420 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2421 BUG_ON(ret);
0b86a832 2422
2b82032c
YZ		 2423 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2424 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2425 item_size);
8f18cf13
CM		 2426 BUG_ON(ret);
2427 }
0b86a832 2428 kfree(chunk);
2b82032c
YZ		 2429 return 0;
2430}
0b86a832 2431
2b82032c
YZ		 2432/*
2433 * Chunk allocation falls into two parts. The first part does works
2434 * that make the new allocated chunk useable, but not do any operation
2435 * that modifies the chunk tree. The second part does the works that
2436 * require modifying the chunk tree. This division is important for the
2437 * bootstrap process of adding storage to a seed btrfs.
2438 */
2439int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2440 struct btrfs_root *extent_root, u64 type)
2441{
2442 u64 chunk_offset;
2443 u64 chunk_size;
2444 u64 stripe_size;
2445 struct map_lookup *map;
2446 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2447 int ret;
2448
2449 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2450 &chunk_offset);
2451 if (ret)
2452 return ret;
2453
2454 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2455 &stripe_size, chunk_offset, type);
2456 if (ret)
2457 return ret;
2458
2459 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2460 chunk_size, stripe_size);
2461 BUG_ON(ret);
2462 return 0;
2463}
2464
d397712b 2465static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2466 struct btrfs_root *root,
2467 struct btrfs_device *device)
2468{
2469 u64 chunk_offset;
2470 u64 sys_chunk_offset;
2471 u64 chunk_size;
2472 u64 sys_chunk_size;
2473 u64 stripe_size;
2474 u64 sys_stripe_size;
2475 u64 alloc_profile;
2476 struct map_lookup *map;
2477 struct map_lookup *sys_map;
2478 struct btrfs_fs_info *fs_info = root->fs_info;
2479 struct btrfs_root *extent_root = fs_info->extent_root;
2480 int ret;
2481
2482 ret = find_next_chunk(fs_info->chunk_root,
2483 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2484 BUG_ON(ret);
2485
2486 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2487 (fs_info->metadata_alloc_profile &
2488 fs_info->avail_metadata_alloc_bits);
2489 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2490
2491 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2492 &stripe_size, chunk_offset, alloc_profile);
2493 BUG_ON(ret);
2494
2495 sys_chunk_offset = chunk_offset + chunk_size;
2496
2497 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2498 (fs_info->system_alloc_profile &
2499 fs_info->avail_system_alloc_bits);
2500 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2501
2502 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2503 &sys_chunk_size, &sys_stripe_size,
2504 sys_chunk_offset, alloc_profile);
2505 BUG_ON(ret);
2506
2507 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2508 BUG_ON(ret);
2509
2510 /*
2511 * Modifying chunk tree needs allocating new blocks from both
2512 * system block group and metadata block group. So we only can
2513 * do operations require modifying the chunk tree after both
2514 * block groups were created.
2515 */
2516 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2517 chunk_size, stripe_size);
2518 BUG_ON(ret);
2519
2520 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2521 sys_chunk_offset, sys_chunk_size,
2522 sys_stripe_size);
b248a415 2523 BUG_ON(ret);
2b82032c
YZ		 2524 return 0;
2525}
2526
2527int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2528{
2529 struct extent_map *em;
2530 struct map_lookup *map;
2531 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2532 int readonly = 0;
2533 int i;
2534
890871be 2535 read_lock(&map_tree->map_tree.lock);
2b82032c 2536 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 2537 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 2538 if (!em)
2539 return 1;
2540
2541 map = (struct map_lookup *)em->bdev;
2542 for (i = 0; i < map->num_stripes; i++) {
2543 if (!map->stripes[i].dev->writeable) {
2544 readonly = 1;
2545 break;
2546 }
2547 }
0b86a832 2548 free_extent_map(em);
2b82032c 2549 return readonly;
0b86a832
CM		 2550}
2551
2552void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2553{
2554 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2555}
2556
2557void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2558{
2559 struct extent_map *em;
2560
d397712b 2561 while (1) {
890871be 2562 write_lock(&tree->map_tree.lock);
0b86a832
CM		 2563 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2564 if (em)
2565 remove_extent_mapping(&tree->map_tree, em);
890871be 2566 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 2567 if (!em)
2568 break;
2569 kfree(em->bdev);
2570 /* once for us */
2571 free_extent_map(em);
2572 /* once for the tree */
2573 free_extent_map(em);
2574 }
2575}
2576
f188591e
CM		 2577int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2578{
2579 struct extent_map *em;
2580 struct map_lookup *map;
2581 struct extent_map_tree *em_tree = &map_tree->map_tree;
2582 int ret;
2583
890871be 2584 read_lock(&em_tree->lock);
f188591e 2585 em = lookup_extent_mapping(em_tree, logical, len);
890871be 2586 read_unlock(&em_tree->lock);
f188591e
CM		 2587 BUG_ON(!em);
2588
2589 BUG_ON(em->start > logical || em->start + em->len < logical);
2590 map = (struct map_lookup *)em->bdev;
2591 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2592 ret = map->num_stripes;
321aecc6
CM		 2593 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2594 ret = map->sub_stripes;
f188591e
CM		 2595 else
2596 ret = 1;
2597 free_extent_map(em);
f188591e
CM		 2598 return ret;
2599}
2600
dfe25020
CM		 2601static int find_live_mirror(struct map_lookup *map, int first, int num,
2602 int optimal)
2603{
2604 int i;
2605 if (map->stripes[optimal].dev->bdev)
2606 return optimal;
2607 for (i = first; i < first + num; i++) {
2608 if (map->stripes[i].dev->bdev)
2609 return i;
2610 }
2611 /* we couldn't find one that doesn't fail. Just return something
2612 * and the io error handling code will clean up eventually
2613 */
2614 return optimal;
2615}
2616
f2d8d74d
CM		 2617static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2618 u64 logical, u64 *length,
2619 struct btrfs_multi_bio **multi_ret,
2620 int mirror_num, struct page *unplug_page)
0b86a832
CM		 2621{
2622 struct extent_map *em;
2623 struct map_lookup *map;
2624 struct extent_map_tree *em_tree = &map_tree->map_tree;
2625 u64 offset;
593060d7
CM		 2626 u64 stripe_offset;
2627 u64 stripe_nr;
cea9e445 2628 int stripes_allocated = 8;
321aecc6 2629 int stripes_required = 1;
593060d7 2630 int stripe_index;
cea9e445 2631 int i;
f2d8d74d 2632 int num_stripes;
a236aed1 2633 int max_errors = 0;
cea9e445 2634 struct btrfs_multi_bio *multi = NULL;
0b86a832 2635
d397712b 2636 if (multi_ret && !(rw & (1 << BIO_RW)))
cea9e445 2637 stripes_allocated = 1;
cea9e445
CM		 2638again:
2639 if (multi_ret) {
2640 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2641 GFP_NOFS);
2642 if (!multi)
2643 return -ENOMEM;
a236aed1
CM		 2644
2645 atomic_set(&multi->error, 0);
cea9e445 2646 }
0b86a832 2647
890871be 2648 read_lock(&em_tree->lock);
0b86a832 2649 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 2650 read_unlock(&em_tree->lock);
f2d8d74d
CM		 2651
2652 if (!em && unplug_page)
2653 return 0;
2654
3b951516 2655 if (!em) {
d397712b
CM		 2656 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2657 (unsigned long long)logical,
2658 (unsigned long long)*length);
f2d8d74d 2659 BUG();
3b951516 2660 }
0b86a832
CM		 2661
2662 BUG_ON(em->start > logical || em->start + em->len < logical);
2663 map = (struct map_lookup *)em->bdev;
2664 offset = logical - em->start;
593060d7 2665
f188591e
CM		 2666 if (mirror_num > map->num_stripes)
2667 mirror_num = 0;
2668
cea9e445 2669 /* if our multi bio struct is too small, back off and try again */
321aecc6
CM		 2670 if (rw & (1 << BIO_RW)) {
2671 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2672 BTRFS_BLOCK_GROUP_DUP)) {
2673 stripes_required = map->num_stripes;
a236aed1 2674 max_errors = 1;
321aecc6
CM		 2675 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2676 stripes_required = map->sub_stripes;
a236aed1 2677 max_errors = 1;
321aecc6
CM		 2678 }
2679 }
ffbd517d 2680 if (multi_ret && (rw & (1 << BIO_RW)) &&
321aecc6 2681 stripes_allocated < stripes_required) {
cea9e445 2682 stripes_allocated = map->num_stripes;
cea9e445
CM		 2683 free_extent_map(em);
2684 kfree(multi);
2685 goto again;
2686 }
593060d7
CM		 2687 stripe_nr = offset;
2688 /*
2689 * stripe_nr counts the total number of stripes we have to stride
2690 * to get to this block
2691 */
2692 do_div(stripe_nr, map->stripe_len);
2693
2694 stripe_offset = stripe_nr * map->stripe_len;
2695 BUG_ON(offset < stripe_offset);
2696
2697 /* stripe_offset is the offset of this block in its stripe*/
2698 stripe_offset = offset - stripe_offset;
2699
cea9e445 2700 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 2701 BTRFS_BLOCK_GROUP_RAID10 |
cea9e445
CM		 2702 BTRFS_BLOCK_GROUP_DUP)) {
2703 /* we limit the length of each bio to what fits in a stripe */
2704 *length = min_t(u64, em->len - offset,
2705 map->stripe_len - stripe_offset);
2706 } else {
2707 *length = em->len - offset;
2708 }
f2d8d74d
CM		 2709
2710 if (!multi_ret && !unplug_page)
cea9e445
CM		 2711 goto out;
2712
f2d8d74d 2713 num_stripes = 1;
cea9e445 2714 stripe_index = 0;
8790d502 2715 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
f2d8d74d
CM		 2716 if (unplug_page || (rw & (1 << BIO_RW)))
2717 num_stripes = map->num_stripes;
2fff734f 2718 else if (mirror_num)
f188591e 2719 stripe_index = mirror_num - 1;
dfe25020
CM		 2720 else {
2721 stripe_index = find_live_mirror(map, 0,
2722 map->num_stripes,
2723 current->pid % map->num_stripes);
2724 }
2fff734f 2725
611f0e00 2726 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
cea9e445 2727 if (rw & (1 << BIO_RW))
f2d8d74d 2728 num_stripes = map->num_stripes;
f188591e
CM		 2729 else if (mirror_num)
2730 stripe_index = mirror_num - 1;
2fff734f 2731
321aecc6
CM		 2732 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2733 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 2734
2735 stripe_index = do_div(stripe_nr, factor);
2736 stripe_index *= map->sub_stripes;
2737
f2d8d74d
CM		 2738 if (unplug_page || (rw & (1 << BIO_RW)))
2739 num_stripes = map->sub_stripes;
321aecc6
CM		 2740 else if (mirror_num)
2741 stripe_index += mirror_num - 1;
dfe25020
CM		 2742 else {
2743 stripe_index = find_live_mirror(map, stripe_index,
2744 map->sub_stripes, stripe_index +
2745 current->pid % map->sub_stripes);
2746 }
8790d502
CM		 2747 } else {
2748 /*
2749 * after this do_div call, stripe_nr is the number of stripes
2750 * on this device we have to walk to find the data, and
2751 * stripe_index is the number of our device in the stripe array
2752 */
2753 stripe_index = do_div(stripe_nr, map->num_stripes);
2754 }
593060d7 2755 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 2756
f2d8d74d
CM		 2757 for (i = 0; i < num_stripes; i++) {
2758 if (unplug_page) {
2759 struct btrfs_device *device;
2760 struct backing_dev_info *bdi;
2761
2762 device = map->stripes[stripe_index].dev;
dfe25020
CM		 2763 if (device->bdev) {
2764 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 2765 if (bdi->unplug_io_fn)
dfe25020 2766 bdi->unplug_io_fn(bdi, unplug_page);
f2d8d74d
CM		 2767 }
2768 } else {
2769 multi->stripes[i].physical =
2770 map->stripes[stripe_index].physical +
2771 stripe_offset + stripe_nr * map->stripe_len;
2772 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2773 }
cea9e445 2774 stripe_index++;
593060d7 2775 }
f2d8d74d
CM		 2776 if (multi_ret) {
2777 *multi_ret = multi;
2778 multi->num_stripes = num_stripes;
a236aed1 2779 multi->max_errors = max_errors;
f2d8d74d 2780 }
cea9e445 2781out:
0b86a832 2782 free_extent_map(em);
0b86a832
CM		 2783 return 0;
2784}
2785
f2d8d74d
CM		 2786int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2787 u64 logical, u64 *length,
2788 struct btrfs_multi_bio **multi_ret, int mirror_num)
2789{
2790 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2791 mirror_num, NULL);
2792}
2793
a512bbf8
YZ		 2794int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2795 u64 chunk_start, u64 physical, u64 devid,
2796 u64 **logical, int *naddrs, int *stripe_len)
2797{
2798 struct extent_map_tree *em_tree = &map_tree->map_tree;
2799 struct extent_map *em;
2800 struct map_lookup *map;
2801 u64 *buf;
2802 u64 bytenr;
2803 u64 length;
2804 u64 stripe_nr;
2805 int i, j, nr = 0;
2806
890871be 2807 read_lock(&em_tree->lock);
a512bbf8 2808 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 2809 read_unlock(&em_tree->lock);
a512bbf8
YZ		 2810
2811 BUG_ON(!em || em->start != chunk_start);
2812 map = (struct map_lookup *)em->bdev;
2813
2814 length = em->len;
2815 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2816 do_div(length, map->num_stripes / map->sub_stripes);
2817 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2818 do_div(length, map->num_stripes);
2819
2820 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2821 BUG_ON(!buf);
2822
2823 for (i = 0; i < map->num_stripes; i++) {
2824 if (devid && map->stripes[i].dev->devid != devid)
2825 continue;
2826 if (map->stripes[i].physical > physical ||
2827 map->stripes[i].physical + length <= physical)
2828 continue;
2829
2830 stripe_nr = physical - map->stripes[i].physical;
2831 do_div(stripe_nr, map->stripe_len);
2832
2833 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2834 stripe_nr = stripe_nr * map->num_stripes + i;
2835 do_div(stripe_nr, map->sub_stripes);
2836 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2837 stripe_nr = stripe_nr * map->num_stripes + i;
2838 }
2839 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 2840 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 2841 for (j = 0; j < nr; j++) {
2842 if (buf[j] == bytenr)
2843 break;
2844 }
934d375b
CM		 2845 if (j == nr) {
2846 WARN_ON(nr >= map->num_stripes);
a512bbf8 2847 buf[nr++] = bytenr;
934d375b 2848 }
a512bbf8
YZ		 2849 }
2850
a512bbf8
YZ		 2851 *logical = buf;
2852 *naddrs = nr;
2853 *stripe_len = map->stripe_len;
2854
2855 free_extent_map(em);
2856 return 0;
2857}
2858
f2d8d74d
CM		 2859int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2860 u64 logical, struct page *page)
2861{
2862 u64 length = PAGE_CACHE_SIZE;
2863 return __btrfs_map_block(map_tree, READ, logical, &length,
2864 NULL, 0, page);
2865}
2866
8790d502 2867static void end_bio_multi_stripe(struct bio *bio, int err)
8790d502 2868{
cea9e445 2869 struct btrfs_multi_bio *multi = bio->bi_private;
7d2b4daa 2870 int is_orig_bio = 0;
8790d502 2871
8790d502 2872 if (err)
a236aed1 2873 atomic_inc(&multi->error);
8790d502 2874
7d2b4daa
CM		 2875 if (bio == multi->orig_bio)
2876 is_orig_bio = 1;
2877
cea9e445 2878 if (atomic_dec_and_test(&multi->stripes_pending)) {
7d2b4daa
CM		 2879 if (!is_orig_bio) {
2880 bio_put(bio);
2881 bio = multi->orig_bio;
2882 }
8790d502
CM		 2883 bio->bi_private = multi->private;
2884 bio->bi_end_io = multi->end_io;
a236aed1
CM		 2885 /* only send an error to the higher layers if it is
2886 * beyond the tolerance of the multi-bio
2887 */
1259ab75 2888 if (atomic_read(&multi->error) > multi->max_errors) {
a236aed1 2889 err = -EIO;
1259ab75
CM		 2890 } else if (err) {
2891 /*
2892 * this bio is actually up to date, we didn't
2893 * go over the max number of errors
2894 */
2895 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 2896 err = 0;
1259ab75 2897 }
8790d502
CM		 2898 kfree(multi);
2899
2900 bio_endio(bio, err);
7d2b4daa 2901 } else if (!is_orig_bio) {
8790d502
CM		 2902 bio_put(bio);
2903 }
8790d502
CM		 2904}
2905
8b712842
CM		 2906struct async_sched {
2907 struct bio *bio;
2908 int rw;
2909 struct btrfs_fs_info *info;
2910 struct btrfs_work work;
2911};
2912
2913/*
2914 * see run_scheduled_bios for a description of why bios are collected for
2915 * async submit.
2916 *
2917 * This will add one bio to the pending list for a device and make sure
2918 * the work struct is scheduled.
2919 */
d397712b 2920static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 2921 struct btrfs_device *device,
2922 int rw, struct bio *bio)
8b712842
CM		 2923{
2924 int should_queue = 1;
ffbd517d 2925 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 2926
2927 /* don't bother with additional async steps for reads, right now */
2928 if (!(rw & (1 << BIO_RW))) {
492bb6de 2929 bio_get(bio);
8b712842 2930 submit_bio(rw, bio);
492bb6de 2931 bio_put(bio);
8b712842
CM		 2932 return 0;
2933 }
2934
2935 /*
0986fe9e 2936 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 2937 * higher layers. Otherwise, the async bio makes it appear we have
2938 * made progress against dirty pages when we've really just put it
2939 * on a queue for later
2940 */
0986fe9e 2941 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 2942 WARN_ON(bio->bi_next);
8b712842
CM		 2943 bio->bi_next = NULL;
2944 bio->bi_rw |= rw;
2945
2946 spin_lock(&device->io_lock);
1f98a13f 2947 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
ffbd517d
CM		 2948 pending_bios = &device->pending_sync_bios;
2949 else
2950 pending_bios = &device->pending_bios;
8b712842 2951
ffbd517d
CM		 2952 if (pending_bios->tail)
2953 pending_bios->tail->bi_next = bio;
8b712842 2954
ffbd517d
CM		 2955 pending_bios->tail = bio;
2956 if (!pending_bios->head)
2957 pending_bios->head = bio;
8b712842
CM		 2958 if (device->running_pending)
2959 should_queue = 0;
2960
2961 spin_unlock(&device->io_lock);
2962
2963 if (should_queue)
1cc127b5
CM		 2964 btrfs_queue_worker(&root->fs_info->submit_workers,
2965 &device->work);
8b712842
CM		 2966 return 0;
2967}
2968
f188591e 2969int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 2970 int mirror_num, int async_submit)
0b86a832
CM		 2971{
2972 struct btrfs_mapping_tree *map_tree;
2973 struct btrfs_device *dev;
8790d502 2974 struct bio *first_bio = bio;
a62b9401 2975 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 2976 u64 length = 0;
2977 u64 map_length;
cea9e445 2978 struct btrfs_multi_bio *multi = NULL;
0b86a832 2979 int ret;
8790d502
CM		 2980 int dev_nr = 0;
2981 int total_devs = 1;
0b86a832 2982
f2d8d74d 2983 length = bio->bi_size;
0b86a832
CM		 2984 map_tree = &root->fs_info->mapping_tree;
2985 map_length = length;
cea9e445 2986
f188591e
CM		 2987 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2988 mirror_num);
cea9e445
CM		 2989 BUG_ON(ret);
2990
2991 total_devs = multi->num_stripes;
2992 if (map_length < length) {
d397712b
CM		 2993 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2994 "len %llu\n", (unsigned long long)logical,
2995 (unsigned long long)length,
2996 (unsigned long long)map_length);
cea9e445
CM		 2997 BUG();
2998 }
2999 multi->end_io = first_bio->bi_end_io;
3000 multi->private = first_bio->bi_private;
7d2b4daa 3001 multi->orig_bio = first_bio;
cea9e445
CM		 3002 atomic_set(&multi->stripes_pending, multi->num_stripes);
3003
d397712b 3004 while (dev_nr < total_devs) {
8790d502 3005 if (total_devs > 1) {
8790d502
CM		 3006 if (dev_nr < total_devs - 1) {
3007 bio = bio_clone(first_bio, GFP_NOFS);
3008 BUG_ON(!bio);
3009 } else {
3010 bio = first_bio;
3011 }
3012 bio->bi_private = multi;
3013 bio->bi_end_io = end_bio_multi_stripe;
3014 }
cea9e445
CM		 3015 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3016 dev = multi->stripes[dev_nr].dev;
2b82032c 3017 BUG_ON(rw == WRITE && !dev->writeable);
dfe25020
CM		 3018 if (dev && dev->bdev) {
3019 bio->bi_bdev = dev->bdev;
8b712842
CM		 3020 if (async_submit)
3021 schedule_bio(root, dev, rw, bio);
3022 else
3023 submit_bio(rw, bio);
dfe25020
CM		 3024 } else {
3025 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3026 bio->bi_sector = logical >> 9;
dfe25020 3027 bio_endio(bio, -EIO);
dfe25020 3028 }
8790d502
CM		 3029 dev_nr++;
3030 }
cea9e445
CM		 3031 if (total_devs == 1)
3032 kfree(multi);
0b86a832
CM		 3033 return 0;
3034}
3035
a443755f 3036struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 3037 u8 *uuid, u8 *fsid)
0b86a832 3038{
2b82032c
YZ		 3039 struct btrfs_device *device;
3040 struct btrfs_fs_devices *cur_devices;
3041
3042 cur_devices = root->fs_info->fs_devices;
3043 while (cur_devices) {
3044 if (!fsid ||
3045 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3046 device = __find_device(&cur_devices->devices,
3047 devid, uuid);
3048 if (device)
3049 return device;
3050 }
3051 cur_devices = cur_devices->seed;
3052 }
3053 return NULL;
0b86a832
CM		 3054}
3055
dfe25020
CM		 3056static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3057 u64 devid, u8 *dev_uuid)
3058{
3059 struct btrfs_device *device;
3060 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3061
3062 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 3063 if (!device)
3064 return NULL;
dfe25020
CM		 3065 list_add(&device->dev_list,
3066 &fs_devices->devices);
dfe25020
CM		 3067 device->barriers = 1;
3068 device->dev_root = root->fs_info->dev_root;
3069 device->devid = devid;
8b712842 3070 device->work.func = pending_bios_fn;
e4404d6e 3071 device->fs_devices = fs_devices;
dfe25020
CM		 3072 fs_devices->num_devices++;
3073 spin_lock_init(&device->io_lock);
d20f7043 3074 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 3075 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3076 return device;
3077}
3078
0b86a832
CM		 3079static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3080 struct extent_buffer *leaf,
3081 struct btrfs_chunk *chunk)
3082{
3083 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3084 struct map_lookup *map;
3085 struct extent_map *em;
3086 u64 logical;
3087 u64 length;
3088 u64 devid;
a443755f 3089 u8 uuid[BTRFS_UUID_SIZE];
593060d7 3090 int num_stripes;
0b86a832 3091 int ret;
593060d7 3092 int i;
0b86a832 3093
e17cade2
CM		 3094 logical = key->offset;
3095 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3096
890871be 3097 read_lock(&map_tree->map_tree.lock);
0b86a832 3098 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 3099 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3100
3101 /* already mapped? */
3102 if (em && em->start <= logical && em->start + em->len > logical) {
3103 free_extent_map(em);
0b86a832
CM		 3104 return 0;
3105 } else if (em) {
3106 free_extent_map(em);
3107 }
0b86a832 3108
0b86a832
CM		 3109 em = alloc_extent_map(GFP_NOFS);
3110 if (!em)
3111 return -ENOMEM;
593060d7
CM		 3112 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3113 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3114 if (!map) {
3115 free_extent_map(em);
3116 return -ENOMEM;
3117 }
3118
3119 em->bdev = (struct block_device *)map;
3120 em->start = logical;
3121 em->len = length;
3122 em->block_start = 0;
c8b97818 3123 em->block_len = em->len;
0b86a832 3124
593060d7
CM		 3125 map->num_stripes = num_stripes;
3126 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3127 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3128 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3129 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3130 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3131 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3132 for (i = 0; i < num_stripes; i++) {
3133 map->stripes[i].physical =
3134 btrfs_stripe_offset_nr(leaf, chunk, i);
3135 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3136 read_extent_buffer(leaf, uuid, (unsigned long)
3137 btrfs_stripe_dev_uuid_nr(chunk, i),
3138 BTRFS_UUID_SIZE);
2b82032c
YZ		 3139 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3140 NULL);
dfe25020 3141 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3142 kfree(map);
3143 free_extent_map(em);
3144 return -EIO;
3145 }
dfe25020
CM		 3146 if (!map->stripes[i].dev) {
3147 map->stripes[i].dev =
3148 add_missing_dev(root, devid, uuid);
3149 if (!map->stripes[i].dev) {
3150 kfree(map);
3151 free_extent_map(em);
3152 return -EIO;
3153 }
3154 }
3155 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3156 }
3157
890871be 3158 write_lock(&map_tree->map_tree.lock);
0b86a832 3159 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 3160 write_unlock(&map_tree->map_tree.lock);
b248a415 3161 BUG_ON(ret);
0b86a832
CM		 3162 free_extent_map(em);
3163
3164 return 0;
3165}
3166
3167static int fill_device_from_item(struct extent_buffer *leaf,
3168 struct btrfs_dev_item *dev_item,
3169 struct btrfs_device *device)
3170{
3171 unsigned long ptr;
0b86a832
CM		 3172
3173 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 3174 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3175 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 3176 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3177 device->type = btrfs_device_type(leaf, dev_item);
3178 device->io_align = btrfs_device_io_align(leaf, dev_item);
3179 device->io_width = btrfs_device_io_width(leaf, dev_item);
3180 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3181
3182 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3183 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3184
0b86a832
CM		 3185 return 0;
3186}
3187
2b82032c
YZ		 3188static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3189{
3190 struct btrfs_fs_devices *fs_devices;
3191 int ret;
3192
3193 mutex_lock(&uuid_mutex);
3194
3195 fs_devices = root->fs_info->fs_devices->seed;
3196 while (fs_devices) {
3197 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3198 ret = 0;
3199 goto out;
3200 }
3201 fs_devices = fs_devices->seed;
3202 }
3203
3204 fs_devices = find_fsid(fsid);
3205 if (!fs_devices) {
3206 ret = -ENOENT;
3207 goto out;
3208 }
e4404d6e
YZ		 3209
3210 fs_devices = clone_fs_devices(fs_devices);
3211 if (IS_ERR(fs_devices)) {
3212 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3213 goto out;
3214 }
3215
97288f2c 3216 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3217 root->fs_info->bdev_holder);
2b82032c
YZ		 3218 if (ret)
3219 goto out;
3220
3221 if (!fs_devices->seeding) {
3222 __btrfs_close_devices(fs_devices);
e4404d6e 3223 free_fs_devices(fs_devices);
2b82032c
YZ		 3224 ret = -EINVAL;
3225 goto out;
3226 }
3227
3228 fs_devices->seed = root->fs_info->fs_devices->seed;
3229 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3230out:
3231 mutex_unlock(&uuid_mutex);
3232 return ret;
3233}
3234
0d81ba5d 3235static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3236 struct extent_buffer *leaf,
3237 struct btrfs_dev_item *dev_item)
3238{
3239 struct btrfs_device *device;
3240 u64 devid;
3241 int ret;
2b82032c 3242 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3243 u8 dev_uuid[BTRFS_UUID_SIZE];
3244
0b86a832 3245 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3246 read_extent_buffer(leaf, dev_uuid,
3247 (unsigned long)btrfs_device_uuid(dev_item),
3248 BTRFS_UUID_SIZE);
2b82032c
YZ		 3249 read_extent_buffer(leaf, fs_uuid,
3250 (unsigned long)btrfs_device_fsid(dev_item),
3251 BTRFS_UUID_SIZE);
3252
3253 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3254 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3255 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3256 return ret;
2b82032c
YZ		 3257 }
3258
3259 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3260 if (!device || !device->bdev) {
e4404d6e 3261 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3262 return -EIO;
3263
3264 if (!device) {
d397712b
CM		 3265 printk(KERN_WARNING "warning devid %llu missing\n",
3266 (unsigned long long)devid);
2b82032c
YZ		 3267 device = add_missing_dev(root, devid, dev_uuid);
3268 if (!device)
3269 return -ENOMEM;
3270 }
3271 }
3272
3273 if (device->fs_devices != root->fs_info->fs_devices) {
3274 BUG_ON(device->writeable);
3275 if (device->generation !=
3276 btrfs_device_generation(leaf, dev_item))
3277 return -EINVAL;
6324fbf3 3278 }
0b86a832
CM		 3279
3280 fill_device_from_item(leaf, dev_item, device);
3281 device->dev_root = root->fs_info->dev_root;
dfe25020 3282 device->in_fs_metadata = 1;
2b82032c
YZ		 3283 if (device->writeable)
3284 device->fs_devices->total_rw_bytes += device->total_bytes;
0b86a832 3285 ret = 0;
0b86a832
CM		 3286 return ret;
3287}
3288
0d81ba5d
CM		 3289int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3290{
3291 struct btrfs_dev_item *dev_item;
3292
3293 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3294 dev_item);
3295 return read_one_dev(root, buf, dev_item);
3296}
3297
e4404d6e 3298int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832
CM		 3299{
3300 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
a061fc8d 3301 struct extent_buffer *sb;
0b86a832 3302 struct btrfs_disk_key *disk_key;
0b86a832 3303 struct btrfs_chunk *chunk;
84eed90f
CM		 3304 u8 *ptr;
3305 unsigned long sb_ptr;
3306 int ret = 0;
0b86a832
CM		 3307 u32 num_stripes;
3308 u32 array_size;
3309 u32 len = 0;
0b86a832 3310 u32 cur;
84eed90f 3311 struct btrfs_key key;
0b86a832 3312
e4404d6e 3313 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3314 BTRFS_SUPER_INFO_SIZE);
3315 if (!sb)
3316 return -ENOMEM;
3317 btrfs_set_buffer_uptodate(sb);
4008c04a
CM		 3318 btrfs_set_buffer_lockdep_class(sb, 0);
3319
a061fc8d 3320 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3321 array_size = btrfs_super_sys_array_size(super_copy);
3322
0b86a832
CM		 3323 ptr = super_copy->sys_chunk_array;
3324 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3325 cur = 0;
3326
3327 while (cur < array_size) {
3328 disk_key = (struct btrfs_disk_key *)ptr;
3329 btrfs_disk_key_to_cpu(&key, disk_key);
3330
a061fc8d 3331 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3332 sb_ptr += len;
3333 cur += len;
3334
0d81ba5d 3335 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3336 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3337 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3338 if (ret)
3339 break;
0b86a832
CM		 3340 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3341 len = btrfs_chunk_item_size(num_stripes);
3342 } else {
84eed90f
CM		 3343 ret = -EIO;
3344 break;
0b86a832
CM		 3345 }
3346 ptr += len;
3347 sb_ptr += len;
3348 cur += len;
3349 }
a061fc8d 3350 free_extent_buffer(sb);
84eed90f 3351 return ret;
0b86a832
CM		 3352}
3353
3354int btrfs_read_chunk_tree(struct btrfs_root *root)
3355{
3356 struct btrfs_path *path;
3357 struct extent_buffer *leaf;
3358 struct btrfs_key key;
3359 struct btrfs_key found_key;
3360 int ret;
3361 int slot;
3362
3363 root = root->fs_info->chunk_root;
3364
3365 path = btrfs_alloc_path();
3366 if (!path)
3367 return -ENOMEM;
3368
3369 /* first we search for all of the device items, and then we
3370 * read in all of the chunk items. This way we can create chunk
3371 * mappings that reference all of the devices that are afound
3372 */
3373 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3374 key.offset = 0;
3375 key.type = 0;
3376again:
3377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
d397712b 3378 while (1) {
0b86a832
CM		 3379 leaf = path->nodes[0];
3380 slot = path->slots[0];
3381 if (slot >= btrfs_header_nritems(leaf)) {
3382 ret = btrfs_next_leaf(root, path);
3383 if (ret == 0)
3384 continue;
3385 if (ret < 0)
3386 goto error;
3387 break;
3388 }
3389 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3390 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3391 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3392 break;
3393 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3394 struct btrfs_dev_item *dev_item;
3395 dev_item = btrfs_item_ptr(leaf, slot,
3396 struct btrfs_dev_item);
0d81ba5d 3397 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3398 if (ret)
3399 goto error;
0b86a832
CM		 3400 }
3401 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3402 struct btrfs_chunk *chunk;
3403 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3404 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3405 if (ret)
3406 goto error;
0b86a832
CM		 3407 }
3408 path->slots[0]++;
3409 }
3410 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3411 key.objectid = 0;
3412 btrfs_release_path(root, path);
3413 goto again;
3414 }
0b86a832
CM		 3415 ret = 0;
3416error:
2b82032c 3417 btrfs_free_path(path);
0b86a832
CM		 3418 return ret;
3419}
Linux 6.x block layer and io_uring tree(s)