projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix deadlocks and stalls on dead root removal
[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;
b64a2851 164 unsigned long limit;
b765ead5 165 unsigned long last_waited = 0;
8b712842 166
bedf762b 167 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 168 fs_info = device->dev_root->fs_info;
169 limit = btrfs_async_submit_limit(fs_info);
170 limit = limit * 2 / 3;
171
ffbd517d
CM		 172 /* we want to make sure that every time we switch from the sync
173 * list to the normal list, we unplug
174 */
175 num_sync_run = 0;
176
8b712842
CM		 177loop:
178 spin_lock(&device->io_lock);
ffbd517d 179 num_run = 0;
8b712842 180
a6837051 181loop_lock:
ffbd517d 182
8b712842
CM		 183 /* take all the bios off the list at once and process them
184 * later on (without the lock held). But, remember the
185 * tail and other pointers so the bios can be properly reinserted
186 * into the list if we hit congestion
187 */
ffbd517d
CM		 188 if (device->pending_sync_bios.head)
189 pending_bios = &device->pending_sync_bios;
190 else
191 pending_bios = &device->pending_bios;
192
193 pending = pending_bios->head;
194 tail = pending_bios->tail;
8b712842 195 WARN_ON(pending && !tail);
8b712842
CM		 196
197 /*
198 * if pending was null this time around, no bios need processing
199 * at all and we can stop. Otherwise it'll loop back up again
200 * and do an additional check so no bios are missed.
201 *
202 * device->running_pending is used to synchronize with the
203 * schedule_bio code.
204 */
ffbd517d
CM		 205 if (device->pending_sync_bios.head == NULL &&
206 device->pending_bios.head == NULL) {
8b712842
CM		 207 again = 0;
208 device->running_pending = 0;
ffbd517d
CM		 209 } else {
210 again = 1;
211 device->running_pending = 1;
8b712842 212 }
ffbd517d
CM		 213
214 pending_bios->head = NULL;
215 pending_bios->tail = NULL;
216
8b712842
CM		 217 spin_unlock(&device->io_lock);
218
ffbd517d
CM		 219 /*
220 * if we're doing the regular priority list, make sure we unplug
221 * for any high prio bios we've sent down
222 */
223 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
224 num_sync_run = 0;
225 blk_run_backing_dev(bdi, NULL);
226 }
227
d397712b 228 while (pending) {
ffbd517d
CM		 229
230 rmb();
231 if (pending_bios != &device->pending_sync_bios &&
232 device->pending_sync_bios.head &&
233 num_run > 16) {
234 cond_resched();
235 spin_lock(&device->io_lock);
236 requeue_list(pending_bios, pending, tail);
237 goto loop_lock;
238 }
239
8b712842
CM		 240 cur = pending;
241 pending = pending->bi_next;
242 cur->bi_next = NULL;
b64a2851
CM		 243 atomic_dec(&fs_info->nr_async_bios);
244
245 if (atomic_read(&fs_info->nr_async_bios) < limit &&
246 waitqueue_active(&fs_info->async_submit_wait))
247 wake_up(&fs_info->async_submit_wait);
492bb6de
CM		 248
249 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
8b712842
CM		 250 submit_bio(cur->bi_rw, cur);
251 num_run++;
ffbd517d
CM		 252 if (bio_sync(cur))
253 num_sync_run++;
254
255 if (need_resched()) {
256 if (num_sync_run) {
257 blk_run_backing_dev(bdi, NULL);
258 num_sync_run = 0;
259 }
260 cond_resched();
261 }
8b712842
CM		 262
263 /*
264 * we made progress, there is more work to do and the bdi
265 * is now congested. Back off and let other work structs
266 * run instead
267 */
a6837051 268 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
5f2cc086 269 fs_info->fs_devices->open_devices > 1) {
b765ead5 270 struct io_context *ioc;
8b712842 271
b765ead5
CM		 272 ioc = current->io_context;
273
274 /*
275 * the main goal here is that we don't want to
276 * block if we're going to be able to submit
277 * more requests without blocking.
278 *
279 * This code does two great things, it pokes into
280 * the elevator code from a filesystem _and_
281 * it makes assumptions about how batching works.
282 */
283 if (ioc && ioc->nr_batch_requests > 0 &&
284 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
285 (last_waited == 0 ||
286 ioc->last_waited == last_waited)) {
287 /*
288 * we want to go through our batch of
289 * requests and stop. So, we copy out
290 * the ioc->last_waited time and test
291 * against it before looping
292 */
293 last_waited = ioc->last_waited;
ffbd517d
CM		 294 if (need_resched()) {
295 if (num_sync_run) {
296 blk_run_backing_dev(bdi, NULL);
297 num_sync_run = 0;
298 }
299 cond_resched();
300 }
b765ead5
CM		 301 continue;
302 }
8b712842 303 spin_lock(&device->io_lock);
ffbd517d 304 requeue_list(pending_bios, pending, tail);
a6837051 305 device->running_pending = 1;
8b712842
CM		 306
307 spin_unlock(&device->io_lock);
308 btrfs_requeue_work(&device->work);
309 goto done;
310 }
311 }
ffbd517d
CM		 312
313 if (num_sync_run) {
314 num_sync_run = 0;
315 blk_run_backing_dev(bdi, NULL);
316 }
317
318 cond_resched();
8b712842
CM		 319 if (again)
320 goto loop;
a6837051
CM		 321
322 spin_lock(&device->io_lock);
ffbd517d 323 if (device->pending_bios.head || device->pending_sync_bios.head)
a6837051
CM		 324 goto loop_lock;
325 spin_unlock(&device->io_lock);
bedf762b
CM		 326
327 /*
328 * IO has already been through a long path to get here. Checksumming,
329 * async helper threads, perhaps compression. We've done a pretty
330 * good job of collecting a batch of IO and should just unplug
331 * the device right away.
332 *
333 * This will help anyone who is waiting on the IO, they might have
334 * already unplugged, but managed to do so before the bio they
335 * cared about found its way down here.
336 */
337 blk_run_backing_dev(bdi, NULL);
8b712842
CM		 338done:
339 return 0;
340}
341
b2950863 342static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 343{
344 struct btrfs_device *device;
345
346 device = container_of(work, struct btrfs_device, work);
347 run_scheduled_bios(device);
348}
349
a1b32a59 350static noinline int device_list_add(const char *path,
8a4b83cc
CM		 351 struct btrfs_super_block *disk_super,
352 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
353{
354 struct btrfs_device *device;
355 struct btrfs_fs_devices *fs_devices;
356 u64 found_transid = btrfs_super_generation(disk_super);
357
358 fs_devices = find_fsid(disk_super->fsid);
359 if (!fs_devices) {
515dc322 360 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 361 if (!fs_devices)
362 return -ENOMEM;
363 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 364 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 365 list_add(&fs_devices->list, &fs_uuids);
366 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
367 fs_devices->latest_devid = devid;
368 fs_devices->latest_trans = found_transid;
8a4b83cc
CM		 369 device = NULL;
370 } else {
a443755f
CM		 371 device = __find_device(&fs_devices->devices, devid,
372 disk_super->dev_item.uuid);
8a4b83cc
CM		 373 }
374 if (!device) {
2b82032c
YZ		 375 if (fs_devices->opened)
376 return -EBUSY;
377
8a4b83cc
CM		 378 device = kzalloc(sizeof(*device), GFP_NOFS);
379 if (!device) {
380 /* we can safely leave the fs_devices entry around */
381 return -ENOMEM;
382 }
383 device->devid = devid;
8b712842 384 device->work.func = pending_bios_fn;
a443755f
CM		 385 memcpy(device->uuid, disk_super->dev_item.uuid,
386 BTRFS_UUID_SIZE);
f2984462 387 device->barriers = 1;
b248a415 388 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 389 device->name = kstrdup(path, GFP_NOFS);
390 if (!device->name) {
391 kfree(device);
392 return -ENOMEM;
393 }
2b82032c 394 INIT_LIST_HEAD(&device->dev_alloc_list);
8a4b83cc 395 list_add(&device->dev_list, &fs_devices->devices);
2b82032c 396 device->fs_devices = fs_devices;
8a4b83cc
CM		 397 fs_devices->num_devices++;
398 }
399
400 if (found_transid > fs_devices->latest_trans) {
401 fs_devices->latest_devid = devid;
402 fs_devices->latest_trans = found_transid;
403 }
8a4b83cc
CM		 404 *fs_devices_ret = fs_devices;
405 return 0;
406}
407
e4404d6e
YZ		 408static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
409{
410 struct btrfs_fs_devices *fs_devices;
411 struct btrfs_device *device;
412 struct btrfs_device *orig_dev;
413
414 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
415 if (!fs_devices)
416 return ERR_PTR(-ENOMEM);
417
418 INIT_LIST_HEAD(&fs_devices->devices);
419 INIT_LIST_HEAD(&fs_devices->alloc_list);
420 INIT_LIST_HEAD(&fs_devices->list);
421 fs_devices->latest_devid = orig->latest_devid;
422 fs_devices->latest_trans = orig->latest_trans;
423 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
424
425 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
426 device = kzalloc(sizeof(*device), GFP_NOFS);
427 if (!device)
428 goto error;
429
430 device->name = kstrdup(orig_dev->name, GFP_NOFS);
431 if (!device->name)
432 goto error;
433
434 device->devid = orig_dev->devid;
435 device->work.func = pending_bios_fn;
436 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
437 device->barriers = 1;
438 spin_lock_init(&device->io_lock);
439 INIT_LIST_HEAD(&device->dev_list);
440 INIT_LIST_HEAD(&device->dev_alloc_list);
441
442 list_add(&device->dev_list, &fs_devices->devices);
443 device->fs_devices = fs_devices;
444 fs_devices->num_devices++;
445 }
446 return fs_devices;
447error:
448 free_fs_devices(fs_devices);
449 return ERR_PTR(-ENOMEM);
450}
451
dfe25020
CM		 452int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
453{
c6e30871 454 struct btrfs_device *device, *next;
dfe25020
CM		 455
456 mutex_lock(&uuid_mutex);
457again:
c6e30871 458 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 459 if (device->in_fs_metadata)
460 continue;
461
462 if (device->bdev) {
15916de8 463 close_bdev_exclusive(device->bdev, device->mode);
2b82032c
YZ		 464 device->bdev = NULL;
465 fs_devices->open_devices--;
466 }
467 if (device->writeable) {
468 list_del_init(&device->dev_alloc_list);
469 device->writeable = 0;
470 fs_devices->rw_devices--;
471 }
e4404d6e
YZ		 472 list_del_init(&device->dev_list);
473 fs_devices->num_devices--;
474 kfree(device->name);
475 kfree(device);
dfe25020 476 }
2b82032c
YZ		 477
478 if (fs_devices->seed) {
479 fs_devices = fs_devices->seed;
2b82032c
YZ		 480 goto again;
481 }
482
dfe25020
CM		 483 mutex_unlock(&uuid_mutex);
484 return 0;
485}
a0af469b 486
2b82032c 487static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 488{
8a4b83cc 489 struct btrfs_device *device;
e4404d6e 490
2b82032c
YZ		 491 if (--fs_devices->opened > 0)
492 return 0;
8a4b83cc 493
c6e30871 494 list_for_each_entry(device, &fs_devices->devices, dev_list) {
8a4b83cc 495 if (device->bdev) {
15916de8 496 close_bdev_exclusive(device->bdev, device->mode);
a0af469b 497 fs_devices->open_devices--;
8a4b83cc 498 }
2b82032c
YZ		 499 if (device->writeable) {
500 list_del_init(&device->dev_alloc_list);
501 fs_devices->rw_devices--;
502 }
503
8a4b83cc 504 device->bdev = NULL;
2b82032c 505 device->writeable = 0;
dfe25020 506 device->in_fs_metadata = 0;
8a4b83cc 507 }
e4404d6e
YZ		 508 WARN_ON(fs_devices->open_devices);
509 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 510 fs_devices->opened = 0;
511 fs_devices->seeding = 0;
2b82032c 512
8a4b83cc
CM		 513 return 0;
514}
515
2b82032c
YZ		 516int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
517{
e4404d6e 518 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 519 int ret;
520
521 mutex_lock(&uuid_mutex);
522 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 523 if (!fs_devices->opened) {
524 seed_devices = fs_devices->seed;
525 fs_devices->seed = NULL;
526 }
2b82032c 527 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 528
529 while (seed_devices) {
530 fs_devices = seed_devices;
531 seed_devices = fs_devices->seed;
532 __btrfs_close_devices(fs_devices);
533 free_fs_devices(fs_devices);
534 }
2b82032c
YZ		 535 return ret;
536}
537
e4404d6e
YZ		 538static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
539 fmode_t flags, void *holder)
8a4b83cc
CM		 540{
541 struct block_device *bdev;
542 struct list_head *head = &fs_devices->devices;
8a4b83cc 543 struct btrfs_device *device;
a0af469b
CM		 544 struct block_device *latest_bdev = NULL;
545 struct buffer_head *bh;
546 struct btrfs_super_block *disk_super;
547 u64 latest_devid = 0;
548 u64 latest_transid = 0;
a0af469b 549 u64 devid;
2b82032c 550 int seeding = 1;
a0af469b 551 int ret = 0;
8a4b83cc 552
c6e30871 553 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 554 if (device->bdev)
555 continue;
dfe25020
CM		 556 if (!device->name)
557 continue;
558
15916de8 559 bdev = open_bdev_exclusive(device->name, flags, holder);
8a4b83cc 560 if (IS_ERR(bdev)) {
d397712b 561 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 562 goto error;
8a4b83cc 563 }
a061fc8d 564 set_blocksize(bdev, 4096);
a0af469b 565
a512bbf8 566 bh = btrfs_read_dev_super(bdev);
a0af469b
CM		 567 if (!bh)
568 goto error_close;
569
570 disk_super = (struct btrfs_super_block *)bh->b_data;
a0af469b
CM		 571 devid = le64_to_cpu(disk_super->dev_item.devid);
572 if (devid != device->devid)
573 goto error_brelse;
574
2b82032c
YZ		 575 if (memcmp(device->uuid, disk_super->dev_item.uuid,
576 BTRFS_UUID_SIZE))
577 goto error_brelse;
578
579 device->generation = btrfs_super_generation(disk_super);
580 if (!latest_transid || device->generation > latest_transid) {
a0af469b 581 latest_devid = devid;
2b82032c 582 latest_transid = device->generation;
a0af469b
CM		 583 latest_bdev = bdev;
584 }
585
2b82032c
YZ		 586 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
587 device->writeable = 0;
588 } else {
589 device->writeable = !bdev_read_only(bdev);
590 seeding = 0;
591 }
592
8a4b83cc 593 device->bdev = bdev;
dfe25020 594 device->in_fs_metadata = 0;
15916de8
CM		 595 device->mode = flags;
596
a0af469b 597 fs_devices->open_devices++;
2b82032c
YZ		 598 if (device->writeable) {
599 fs_devices->rw_devices++;
600 list_add(&device->dev_alloc_list,
601 &fs_devices->alloc_list);
602 }
a0af469b 603 continue;
a061fc8d 604
a0af469b
CM		 605error_brelse:
606 brelse(bh);
607error_close:
97288f2c 608 close_bdev_exclusive(bdev, FMODE_READ);
a0af469b
CM		 609error:
610 continue;
8a4b83cc 611 }
a0af469b
CM		 612 if (fs_devices->open_devices == 0) {
613 ret = -EIO;
614 goto out;
615 }
2b82032c
YZ		 616 fs_devices->seeding = seeding;
617 fs_devices->opened = 1;
a0af469b
CM		 618 fs_devices->latest_bdev = latest_bdev;
619 fs_devices->latest_devid = latest_devid;
620 fs_devices->latest_trans = latest_transid;
2b82032c 621 fs_devices->total_rw_bytes = 0;
a0af469b 622out:
2b82032c
YZ		 623 return ret;
624}
625
626int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 627 fmode_t flags, void *holder)
2b82032c
YZ		 628{
629 int ret;
630
631 mutex_lock(&uuid_mutex);
632 if (fs_devices->opened) {
e4404d6e
YZ		 633 fs_devices->opened++;
634 ret = 0;
2b82032c 635 } else {
15916de8 636 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 637 }
8a4b83cc 638 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 639 return ret;
640}
641
97288f2c 642int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 643 struct btrfs_fs_devices **fs_devices_ret)
644{
645 struct btrfs_super_block *disk_super;
646 struct block_device *bdev;
647 struct buffer_head *bh;
648 int ret;
649 u64 devid;
f2984462 650 u64 transid;
8a4b83cc
CM		 651
652 mutex_lock(&uuid_mutex);
653
15916de8 654 bdev = open_bdev_exclusive(path, flags, holder);
8a4b83cc
CM		 655
656 if (IS_ERR(bdev)) {
8a4b83cc
CM		 657 ret = PTR_ERR(bdev);
658 goto error;
659 }
660
661 ret = set_blocksize(bdev, 4096);
662 if (ret)
663 goto error_close;
a512bbf8 664 bh = btrfs_read_dev_super(bdev);
8a4b83cc
CM		 665 if (!bh) {
666 ret = -EIO;
667 goto error_close;
668 }
669 disk_super = (struct btrfs_super_block *)bh->b_data;
8a4b83cc 670 devid = le64_to_cpu(disk_super->dev_item.devid);
f2984462 671 transid = btrfs_super_generation(disk_super);
7ae9c09d 672 if (disk_super->label[0])
d397712b 673 printk(KERN_INFO "device label %s ", disk_super->label);
7ae9c09d
CM		 674 else {
675 /* FIXME, make a readl uuid parser */
d397712b 676 printk(KERN_INFO "device fsid %llx-%llx ",
7ae9c09d
CM		 677 *(unsigned long long *)disk_super->fsid,
678 *(unsigned long long *)(disk_super->fsid + 8));
679 }
119e10cf 680 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 681 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 682 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
683
8a4b83cc
CM		 684 brelse(bh);
685error_close:
15916de8 686 close_bdev_exclusive(bdev, flags);
8a4b83cc
CM		 687error:
688 mutex_unlock(&uuid_mutex);
689 return ret;
690}
0b86a832
CM		 691
692/*
693 * this uses a pretty simple search, the expectation is that it is
694 * called very infrequently and that a given device has a small number
695 * of extents
696 */
a1b32a59
CM		 697static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
698 struct btrfs_device *device,
a1b32a59 699 u64 num_bytes, u64 *start)
0b86a832
CM		 700{
701 struct btrfs_key key;
702 struct btrfs_root *root = device->dev_root;
703 struct btrfs_dev_extent *dev_extent = NULL;
2b82032c 704 struct btrfs_path *path;
0b86a832
CM		 705 u64 hole_size = 0;
706 u64 last_byte = 0;
707 u64 search_start = 0;
708 u64 search_end = device->total_bytes;
709 int ret;
710 int slot = 0;
711 int start_found;
712 struct extent_buffer *l;
713
2b82032c
YZ		 714 path = btrfs_alloc_path();
715 if (!path)
716 return -ENOMEM;
0b86a832 717 path->reada = 2;
2b82032c 718 start_found = 0;
0b86a832
CM		 719
720 /* FIXME use last free of some kind */
721
8a4b83cc
CM		 722 /* we don't want to overwrite the superblock on the drive,
723 * so we make sure to start at an offset of at least 1MB
724 */
725 search_start = max((u64)1024 * 1024, search_start);
8f18cf13
CM		 726
727 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
728 search_start = max(root->fs_info->alloc_start, search_start);
729
0b86a832
CM		 730 key.objectid = device->devid;
731 key.offset = search_start;
732 key.type = BTRFS_DEV_EXTENT_KEY;
733 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
734 if (ret < 0)
735 goto error;
736 ret = btrfs_previous_item(root, path, 0, key.type);
737 if (ret < 0)
738 goto error;
739 l = path->nodes[0];
740 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
741 while (1) {
742 l = path->nodes[0];
743 slot = path->slots[0];
744 if (slot >= btrfs_header_nritems(l)) {
745 ret = btrfs_next_leaf(root, path);
746 if (ret == 0)
747 continue;
748 if (ret < 0)
749 goto error;
750no_more_items:
751 if (!start_found) {
752 if (search_start >= search_end) {
753 ret = -ENOSPC;
754 goto error;
755 }
756 *start = search_start;
757 start_found = 1;
758 goto check_pending;
759 }
760 *start = last_byte > search_start ?
761 last_byte : search_start;
762 if (search_end <= *start) {
763 ret = -ENOSPC;
764 goto error;
765 }
766 goto check_pending;
767 }
768 btrfs_item_key_to_cpu(l, &key, slot);
769
770 if (key.objectid < device->devid)
771 goto next;
772
773 if (key.objectid > device->devid)
774 goto no_more_items;
775
776 if (key.offset >= search_start && key.offset > last_byte &&
777 start_found) {
778 if (last_byte < search_start)
779 last_byte = search_start;
780 hole_size = key.offset - last_byte;
781 if (key.offset > last_byte &&
782 hole_size >= num_bytes) {
783 *start = last_byte;
784 goto check_pending;
785 }
786 }
d397712b 787 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
0b86a832 788 goto next;
0b86a832
CM		 789
790 start_found = 1;
791 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
792 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
793next:
794 path->slots[0]++;
795 cond_resched();
796 }
797check_pending:
798 /* we have to make sure we didn't find an extent that has already
799 * been allocated by the map tree or the original allocation
800 */
0b86a832
CM		 801 BUG_ON(*start < search_start);
802
6324fbf3 803 if (*start + num_bytes > search_end) {
0b86a832
CM		 804 ret = -ENOSPC;
805 goto error;
806 }
807 /* check for pending inserts here */
2b82032c 808 ret = 0;
0b86a832
CM		 809
810error:
2b82032c 811 btrfs_free_path(path);
0b86a832
CM		 812 return ret;
813}
814
b2950863 815static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 816 struct btrfs_device *device,
817 u64 start)
818{
819 int ret;
820 struct btrfs_path *path;
821 struct btrfs_root *root = device->dev_root;
822 struct btrfs_key key;
a061fc8d
CM		 823 struct btrfs_key found_key;
824 struct extent_buffer *leaf = NULL;
825 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 826
827 path = btrfs_alloc_path();
828 if (!path)
829 return -ENOMEM;
830
831 key.objectid = device->devid;
832 key.offset = start;
833 key.type = BTRFS_DEV_EXTENT_KEY;
834
835 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 836 if (ret > 0) {
837 ret = btrfs_previous_item(root, path, key.objectid,
838 BTRFS_DEV_EXTENT_KEY);
839 BUG_ON(ret);
840 leaf = path->nodes[0];
841 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
842 extent = btrfs_item_ptr(leaf, path->slots[0],
843 struct btrfs_dev_extent);
844 BUG_ON(found_key.offset > start || found_key.offset +
845 btrfs_dev_extent_length(leaf, extent) < start);
846 ret = 0;
847 } else if (ret == 0) {
848 leaf = path->nodes[0];
849 extent = btrfs_item_ptr(leaf, path->slots[0],
850 struct btrfs_dev_extent);
851 }
8f18cf13
CM		 852 BUG_ON(ret);
853
dfe25020
CM		 854 if (device->bytes_used > 0)
855 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
8f18cf13
CM		 856 ret = btrfs_del_item(trans, root, path);
857 BUG_ON(ret);
858
859 btrfs_free_path(path);
860 return ret;
861}
862
2b82032c 863int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 864 struct btrfs_device *device,
e17cade2 865 u64 chunk_tree, u64 chunk_objectid,
2b82032c 866 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 867{
868 int ret;
869 struct btrfs_path *path;
870 struct btrfs_root *root = device->dev_root;
871 struct btrfs_dev_extent *extent;
872 struct extent_buffer *leaf;
873 struct btrfs_key key;
874
dfe25020 875 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 876 path = btrfs_alloc_path();
877 if (!path)
878 return -ENOMEM;
879
0b86a832 880 key.objectid = device->devid;
2b82032c 881 key.offset = start;
0b86a832
CM		 882 key.type = BTRFS_DEV_EXTENT_KEY;
883 ret = btrfs_insert_empty_item(trans, root, path, &key,
884 sizeof(*extent));
885 BUG_ON(ret);
886
887 leaf = path->nodes[0];
888 extent = btrfs_item_ptr(leaf, path->slots[0],
889 struct btrfs_dev_extent);
e17cade2
CM		 890 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
891 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
892 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
893
894 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
895 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
896 BTRFS_UUID_SIZE);
897
0b86a832
CM		 898 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
899 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 900 btrfs_free_path(path);
901 return ret;
902}
903
a1b32a59
CM		 904static noinline int find_next_chunk(struct btrfs_root *root,
905 u64 objectid, u64 *offset)
0b86a832
CM		 906{
907 struct btrfs_path *path;
908 int ret;
909 struct btrfs_key key;
e17cade2 910 struct btrfs_chunk *chunk;
0b86a832
CM		 911 struct btrfs_key found_key;
912
913 path = btrfs_alloc_path();
914 BUG_ON(!path);
915
e17cade2 916 key.objectid = objectid;
0b86a832
CM		 917 key.offset = (u64)-1;
918 key.type = BTRFS_CHUNK_ITEM_KEY;
919
920 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
921 if (ret < 0)
922 goto error;
923
924 BUG_ON(ret == 0);
925
926 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
927 if (ret) {
e17cade2 928 *offset = 0;
0b86a832
CM		 929 } else {
930 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
931 path->slots[0]);
e17cade2
CM		 932 if (found_key.objectid != objectid)
933 *offset = 0;
934 else {
935 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
936 struct btrfs_chunk);
937 *offset = found_key.offset +
938 btrfs_chunk_length(path->nodes[0], chunk);
939 }
0b86a832
CM		 940 }
941 ret = 0;
942error:
943 btrfs_free_path(path);
944 return ret;
945}
946
2b82032c 947static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 948{
949 int ret;
950 struct btrfs_key key;
951 struct btrfs_key found_key;
2b82032c
YZ		 952 struct btrfs_path *path;
953
954 root = root->fs_info->chunk_root;
955
956 path = btrfs_alloc_path();
957 if (!path)
958 return -ENOMEM;
0b86a832
CM		 959
960 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
961 key.type = BTRFS_DEV_ITEM_KEY;
962 key.offset = (u64)-1;
963
964 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
965 if (ret < 0)
966 goto error;
967
968 BUG_ON(ret == 0);
969
970 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
971 BTRFS_DEV_ITEM_KEY);
972 if (ret) {
973 *objectid = 1;
974 } else {
975 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
976 path->slots[0]);
977 *objectid = found_key.offset + 1;
978 }
979 ret = 0;
980error:
2b82032c 981 btrfs_free_path(path);
0b86a832
CM		 982 return ret;
983}
984
985/*
986 * the device information is stored in the chunk root
987 * the btrfs_device struct should be fully filled in
988 */
989int btrfs_add_device(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_device *device)
992{
993 int ret;
994 struct btrfs_path *path;
995 struct btrfs_dev_item *dev_item;
996 struct extent_buffer *leaf;
997 struct btrfs_key key;
998 unsigned long ptr;
0b86a832
CM		 999
1000 root = root->fs_info->chunk_root;
1001
1002 path = btrfs_alloc_path();
1003 if (!path)
1004 return -ENOMEM;
1005
0b86a832
CM		 1006 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1007 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1008 key.offset = device->devid;
0b86a832
CM		 1009
1010 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1011 sizeof(*dev_item));
0b86a832
CM		 1012 if (ret)
1013 goto out;
1014
1015 leaf = path->nodes[0];
1016 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1017
1018 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1019 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1020 btrfs_set_device_type(leaf, dev_item, device->type);
1021 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1022 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1023 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1024 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1025 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1026 btrfs_set_device_group(leaf, dev_item, 0);
1027 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1028 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1029 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1030
0b86a832 1031 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1032 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1033 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1034 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1035 btrfs_mark_buffer_dirty(leaf);
0b86a832 1036
2b82032c 1037 ret = 0;
0b86a832
CM		 1038out:
1039 btrfs_free_path(path);
1040 return ret;
1041}
8f18cf13 1042
a061fc8d
CM		 1043static int btrfs_rm_dev_item(struct btrfs_root *root,
1044 struct btrfs_device *device)
1045{
1046 int ret;
1047 struct btrfs_path *path;
a061fc8d 1048 struct btrfs_key key;
a061fc8d
CM		 1049 struct btrfs_trans_handle *trans;
1050
1051 root = root->fs_info->chunk_root;
1052
1053 path = btrfs_alloc_path();
1054 if (!path)
1055 return -ENOMEM;
1056
1057 trans = btrfs_start_transaction(root, 1);
1058 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1059 key.type = BTRFS_DEV_ITEM_KEY;
1060 key.offset = device->devid;
7d9eb12c 1061 lock_chunks(root);
a061fc8d
CM		 1062
1063 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064 if (ret < 0)
1065 goto out;
1066
1067 if (ret > 0) {
1068 ret = -ENOENT;
1069 goto out;
1070 }
1071
1072 ret = btrfs_del_item(trans, root, path);
1073 if (ret)
1074 goto out;
a061fc8d
CM		 1075out:
1076 btrfs_free_path(path);
7d9eb12c 1077 unlock_chunks(root);
a061fc8d
CM		 1078 btrfs_commit_transaction(trans, root);
1079 return ret;
1080}
1081
1082int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1083{
1084 struct btrfs_device *device;
2b82032c 1085 struct btrfs_device *next_device;
a061fc8d 1086 struct block_device *bdev;
dfe25020 1087 struct buffer_head *bh = NULL;
a061fc8d
CM		 1088 struct btrfs_super_block *disk_super;
1089 u64 all_avail;
1090 u64 devid;
2b82032c
YZ		 1091 u64 num_devices;
1092 u8 *dev_uuid;
a061fc8d
CM		 1093 int ret = 0;
1094
a061fc8d 1095 mutex_lock(&uuid_mutex);
7d9eb12c 1096 mutex_lock(&root->fs_info->volume_mutex);
a061fc8d
CM		 1097
1098 all_avail = root->fs_info->avail_data_alloc_bits |
1099 root->fs_info->avail_system_alloc_bits |
1100 root->fs_info->avail_metadata_alloc_bits;
1101
1102 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
2b82032c 1103 root->fs_info->fs_devices->rw_devices <= 4) {
d397712b
CM		 1104 printk(KERN_ERR "btrfs: unable to go below four devices "
1105 "on raid10\n");
a061fc8d
CM		 1106 ret = -EINVAL;
1107 goto out;
1108 }
1109
1110 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
2b82032c 1111 root->fs_info->fs_devices->rw_devices <= 2) {
d397712b
CM		 1112 printk(KERN_ERR "btrfs: unable to go below two "
1113 "devices on raid1\n");
a061fc8d
CM		 1114 ret = -EINVAL;
1115 goto out;
1116 }
1117
dfe25020 1118 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1119 struct list_head *devices;
1120 struct btrfs_device *tmp;
a061fc8d 1121
dfe25020
CM		 1122 device = NULL;
1123 devices = &root->fs_info->fs_devices->devices;
c6e30871 1124 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1125 if (tmp->in_fs_metadata && !tmp->bdev) {
1126 device = tmp;
1127 break;
1128 }
1129 }
1130 bdev = NULL;
1131 bh = NULL;
1132 disk_super = NULL;
1133 if (!device) {
d397712b
CM		 1134 printk(KERN_ERR "btrfs: no missing devices found to "
1135 "remove\n");
dfe25020
CM		 1136 goto out;
1137 }
dfe25020 1138 } else {
97288f2c 1139 bdev = open_bdev_exclusive(device_path, FMODE_READ,
dfe25020
CM		 1140 root->fs_info->bdev_holder);
1141 if (IS_ERR(bdev)) {
1142 ret = PTR_ERR(bdev);
1143 goto out;
1144 }
a061fc8d 1145
2b82032c 1146 set_blocksize(bdev, 4096);
a512bbf8 1147 bh = btrfs_read_dev_super(bdev);
dfe25020
CM		 1148 if (!bh) {
1149 ret = -EIO;
1150 goto error_close;
1151 }
1152 disk_super = (struct btrfs_super_block *)bh->b_data;
dfe25020 1153 devid = le64_to_cpu(disk_super->dev_item.devid);
2b82032c
YZ		 1154 dev_uuid = disk_super->dev_item.uuid;
1155 device = btrfs_find_device(root, devid, dev_uuid,
1156 disk_super->fsid);
dfe25020
CM		 1157 if (!device) {
1158 ret = -ENOENT;
1159 goto error_brelse;
1160 }
2b82032c 1161 }
dfe25020 1162
2b82032c 1163 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1164 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1165 "device\n");
2b82032c
YZ		 1166 ret = -EINVAL;
1167 goto error_brelse;
1168 }
1169
1170 if (device->writeable) {
1171 list_del_init(&device->dev_alloc_list);
1172 root->fs_info->fs_devices->rw_devices--;
dfe25020 1173 }
a061fc8d
CM		 1174
1175 ret = btrfs_shrink_device(device, 0);
1176 if (ret)
1177 goto error_brelse;
1178
a061fc8d
CM		 1179 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1180 if (ret)
1181 goto error_brelse;
1182
2b82032c 1183 device->in_fs_metadata = 0;
e4404d6e
YZ		 1184 list_del_init(&device->dev_list);
1185 device->fs_devices->num_devices--;
2b82032c
YZ		 1186
1187 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1188 struct btrfs_device, dev_list);
1189 if (device->bdev == root->fs_info->sb->s_bdev)
1190 root->fs_info->sb->s_bdev = next_device->bdev;
1191 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1192 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1193
e4404d6e
YZ		 1194 if (device->bdev) {
1195 close_bdev_exclusive(device->bdev, device->mode);
1196 device->bdev = NULL;
1197 device->fs_devices->open_devices--;
1198 }
1199
2b82032c
YZ		 1200 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1201 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1202
e4404d6e
YZ		 1203 if (device->fs_devices->open_devices == 0) {
1204 struct btrfs_fs_devices *fs_devices;
1205 fs_devices = root->fs_info->fs_devices;
1206 while (fs_devices) {
1207 if (fs_devices->seed == device->fs_devices)
1208 break;
1209 fs_devices = fs_devices->seed;
2b82032c 1210 }
e4404d6e
YZ		 1211 fs_devices->seed = device->fs_devices->seed;
1212 device->fs_devices->seed = NULL;
1213 __btrfs_close_devices(device->fs_devices);
1214 free_fs_devices(device->fs_devices);
2b82032c
YZ		 1215 }
1216
1217 /*
1218 * at this point, the device is zero sized. We want to
1219 * remove it from the devices list and zero out the old super
1220 */
1221 if (device->writeable) {
dfe25020
CM		 1222 /* make sure this device isn't detected as part of
1223 * the FS anymore
1224 */
1225 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1226 set_buffer_dirty(bh);
1227 sync_dirty_buffer(bh);
dfe25020 1228 }
a061fc8d
CM		 1229
1230 kfree(device->name);
1231 kfree(device);
1232 ret = 0;
a061fc8d
CM		 1233
1234error_brelse:
1235 brelse(bh);
1236error_close:
dfe25020 1237 if (bdev)
97288f2c 1238 close_bdev_exclusive(bdev, FMODE_READ);
a061fc8d 1239out:
7d9eb12c 1240 mutex_unlock(&root->fs_info->volume_mutex);
a061fc8d 1241 mutex_unlock(&uuid_mutex);
a061fc8d
CM		 1242 return ret;
1243}
1244
2b82032c
YZ		 1245/*
1246 * does all the dirty work required for changing file system's UUID.
1247 */
1248static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1249 struct btrfs_root *root)
1250{
1251 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1252 struct btrfs_fs_devices *old_devices;
e4404d6e 1253 struct btrfs_fs_devices *seed_devices;
2b82032c
YZ		 1254 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1255 struct btrfs_device *device;
1256 u64 super_flags;
1257
1258 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1259 if (!fs_devices->seeding)
2b82032c
YZ		 1260 return -EINVAL;
1261
e4404d6e
YZ		 1262 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1263 if (!seed_devices)
2b82032c
YZ		 1264 return -ENOMEM;
1265
e4404d6e
YZ		 1266 old_devices = clone_fs_devices(fs_devices);
1267 if (IS_ERR(old_devices)) {
1268 kfree(seed_devices);
1269 return PTR_ERR(old_devices);
2b82032c 1270 }
e4404d6e 1271
2b82032c
YZ		 1272 list_add(&old_devices->list, &fs_uuids);
1273
e4404d6e
YZ		 1274 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1275 seed_devices->opened = 1;
1276 INIT_LIST_HEAD(&seed_devices->devices);
1277 INIT_LIST_HEAD(&seed_devices->alloc_list);
1278 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1279 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1280 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1281 device->fs_devices = seed_devices;
1282 }
1283
2b82032c
YZ		 1284 fs_devices->seeding = 0;
1285 fs_devices->num_devices = 0;
1286 fs_devices->open_devices = 0;
e4404d6e 1287 fs_devices->seed = seed_devices;
2b82032c
YZ		 1288
1289 generate_random_uuid(fs_devices->fsid);
1290 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1291 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1292 super_flags = btrfs_super_flags(disk_super) &
1293 ~BTRFS_SUPER_FLAG_SEEDING;
1294 btrfs_set_super_flags(disk_super, super_flags);
1295
1296 return 0;
1297}
1298
1299/*
1300 * strore the expected generation for seed devices in device items.
1301 */
1302static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1303 struct btrfs_root *root)
1304{
1305 struct btrfs_path *path;
1306 struct extent_buffer *leaf;
1307 struct btrfs_dev_item *dev_item;
1308 struct btrfs_device *device;
1309 struct btrfs_key key;
1310 u8 fs_uuid[BTRFS_UUID_SIZE];
1311 u8 dev_uuid[BTRFS_UUID_SIZE];
1312 u64 devid;
1313 int ret;
1314
1315 path = btrfs_alloc_path();
1316 if (!path)
1317 return -ENOMEM;
1318
1319 root = root->fs_info->chunk_root;
1320 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1321 key.offset = 0;
1322 key.type = BTRFS_DEV_ITEM_KEY;
1323
1324 while (1) {
1325 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1326 if (ret < 0)
1327 goto error;
1328
1329 leaf = path->nodes[0];
1330next_slot:
1331 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 ret = btrfs_next_leaf(root, path);
1333 if (ret > 0)
1334 break;
1335 if (ret < 0)
1336 goto error;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 btrfs_release_path(root, path);
1340 continue;
1341 }
1342
1343 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1344 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1345 key.type != BTRFS_DEV_ITEM_KEY)
1346 break;
1347
1348 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_dev_item);
1350 devid = btrfs_device_id(leaf, dev_item);
1351 read_extent_buffer(leaf, dev_uuid,
1352 (unsigned long)btrfs_device_uuid(dev_item),
1353 BTRFS_UUID_SIZE);
1354 read_extent_buffer(leaf, fs_uuid,
1355 (unsigned long)btrfs_device_fsid(dev_item),
1356 BTRFS_UUID_SIZE);
1357 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1358 BUG_ON(!device);
1359
1360 if (device->fs_devices->seeding) {
1361 btrfs_set_device_generation(leaf, dev_item,
1362 device->generation);
1363 btrfs_mark_buffer_dirty(leaf);
1364 }
1365
1366 path->slots[0]++;
1367 goto next_slot;
1368 }
1369 ret = 0;
1370error:
1371 btrfs_free_path(path);
1372 return ret;
1373}
1374
788f20eb
CM		 1375int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1376{
1377 struct btrfs_trans_handle *trans;
1378 struct btrfs_device *device;
1379 struct block_device *bdev;
788f20eb 1380 struct list_head *devices;
2b82032c 1381 struct super_block *sb = root->fs_info->sb;
788f20eb 1382 u64 total_bytes;
2b82032c 1383 int seeding_dev = 0;
788f20eb
CM		 1384 int ret = 0;
1385
2b82032c
YZ		 1386 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1387 return -EINVAL;
788f20eb 1388
15916de8 1389 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
d397712b 1390 if (!bdev)
788f20eb 1391 return -EIO;
a2135011 1392
2b82032c
YZ		 1393 if (root->fs_info->fs_devices->seeding) {
1394 seeding_dev = 1;
1395 down_write(&sb->s_umount);
1396 mutex_lock(&uuid_mutex);
1397 }
1398
8c8bee1d 1399 filemap_write_and_wait(bdev->bd_inode->i_mapping);
7d9eb12c 1400 mutex_lock(&root->fs_info->volume_mutex);
a2135011 1401
788f20eb 1402 devices = &root->fs_info->fs_devices->devices;
c6e30871 1403 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1404 if (device->bdev == bdev) {
1405 ret = -EEXIST;
2b82032c 1406 goto error;
788f20eb
CM		 1407 }
1408 }
1409
1410 device = kzalloc(sizeof(*device), GFP_NOFS);
1411 if (!device) {
1412 /* we can safely leave the fs_devices entry around */
1413 ret = -ENOMEM;
2b82032c 1414 goto error;
788f20eb
CM		 1415 }
1416
788f20eb
CM		 1417 device->name = kstrdup(device_path, GFP_NOFS);
1418 if (!device->name) {
1419 kfree(device);
2b82032c
YZ		 1420 ret = -ENOMEM;
1421 goto error;
788f20eb 1422 }
2b82032c
YZ		 1423
1424 ret = find_next_devid(root, &device->devid);
1425 if (ret) {
1426 kfree(device);
1427 goto error;
1428 }
1429
1430 trans = btrfs_start_transaction(root, 1);
1431 lock_chunks(root);
1432
1433 device->barriers = 1;
1434 device->writeable = 1;
1435 device->work.func = pending_bios_fn;
1436 generate_random_uuid(device->uuid);
1437 spin_lock_init(&device->io_lock);
1438 device->generation = trans->transid;
788f20eb
CM		 1439 device->io_width = root->sectorsize;
1440 device->io_align = root->sectorsize;
1441 device->sector_size = root->sectorsize;
1442 device->total_bytes = i_size_read(bdev->bd_inode);
1443 device->dev_root = root->fs_info->dev_root;
1444 device->bdev = bdev;
dfe25020 1445 device->in_fs_metadata = 1;
15916de8 1446 device->mode = 0;
2b82032c 1447 set_blocksize(device->bdev, 4096);
788f20eb 1448
2b82032c
YZ		 1449 if (seeding_dev) {
1450 sb->s_flags &= ~MS_RDONLY;
1451 ret = btrfs_prepare_sprout(trans, root);
1452 BUG_ON(ret);
1453 }
788f20eb 1454
2b82032c
YZ		 1455 device->fs_devices = root->fs_info->fs_devices;
1456 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1457 list_add(&device->dev_alloc_list,
1458 &root->fs_info->fs_devices->alloc_list);
1459 root->fs_info->fs_devices->num_devices++;
1460 root->fs_info->fs_devices->open_devices++;
1461 root->fs_info->fs_devices->rw_devices++;
1462 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1463
788f20eb
CM		 1464 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1465 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1466 total_bytes + device->total_bytes);
1467
1468 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1469 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1470 total_bytes + 1);
1471
2b82032c
YZ		 1472 if (seeding_dev) {
1473 ret = init_first_rw_device(trans, root, device);
1474 BUG_ON(ret);
1475 ret = btrfs_finish_sprout(trans, root);
1476 BUG_ON(ret);
1477 } else {
1478 ret = btrfs_add_device(trans, root, device);
1479 }
1480
913d952e
CM		 1481 /*
1482 * we've got more storage, clear any full flags on the space
1483 * infos
1484 */
1485 btrfs_clear_space_info_full(root->fs_info);
1486
7d9eb12c 1487 unlock_chunks(root);
2b82032c 1488 btrfs_commit_transaction(trans, root);
a2135011 1489
2b82032c
YZ		 1490 if (seeding_dev) {
1491 mutex_unlock(&uuid_mutex);
1492 up_write(&sb->s_umount);
788f20eb 1493
2b82032c
YZ		 1494 ret = btrfs_relocate_sys_chunks(root);
1495 BUG_ON(ret);
1496 }
1497out:
1498 mutex_unlock(&root->fs_info->volume_mutex);
1499 return ret;
1500error:
15916de8 1501 close_bdev_exclusive(bdev, 0);
2b82032c
YZ		 1502 if (seeding_dev) {
1503 mutex_unlock(&uuid_mutex);
1504 up_write(&sb->s_umount);
1505 }
788f20eb
CM		 1506 goto out;
1507}
1508
d397712b
CM		 1509static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1510 struct btrfs_device *device)
0b86a832
CM		 1511{
1512 int ret;
1513 struct btrfs_path *path;
1514 struct btrfs_root *root;
1515 struct btrfs_dev_item *dev_item;
1516 struct extent_buffer *leaf;
1517 struct btrfs_key key;
1518
1519 root = device->dev_root->fs_info->chunk_root;
1520
1521 path = btrfs_alloc_path();
1522 if (!path)
1523 return -ENOMEM;
1524
1525 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1526 key.type = BTRFS_DEV_ITEM_KEY;
1527 key.offset = device->devid;
1528
1529 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1530 if (ret < 0)
1531 goto out;
1532
1533 if (ret > 0) {
1534 ret = -ENOENT;
1535 goto out;
1536 }
1537
1538 leaf = path->nodes[0];
1539 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1540
1541 btrfs_set_device_id(leaf, dev_item, device->devid);
1542 btrfs_set_device_type(leaf, dev_item, device->type);
1543 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1544 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1545 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1546 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1547 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1548 btrfs_mark_buffer_dirty(leaf);
1549
1550out:
1551 btrfs_free_path(path);
1552 return ret;
1553}
1554
7d9eb12c 1555static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1556 struct btrfs_device *device, u64 new_size)
1557{
1558 struct btrfs_super_block *super_copy =
1559 &device->dev_root->fs_info->super_copy;
1560 u64 old_total = btrfs_super_total_bytes(super_copy);
1561 u64 diff = new_size - device->total_bytes;
1562
2b82032c
YZ		 1563 if (!device->writeable)
1564 return -EACCES;
1565 if (new_size <= device->total_bytes)
1566 return -EINVAL;
1567
8f18cf13 1568 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1569 device->fs_devices->total_rw_bytes += diff;
1570
1571 device->total_bytes = new_size;
4184ea7f
CM		 1572 btrfs_clear_space_info_full(device->dev_root->fs_info);
1573
8f18cf13
CM		 1574 return btrfs_update_device(trans, device);
1575}
1576
7d9eb12c
CM		 1577int btrfs_grow_device(struct btrfs_trans_handle *trans,
1578 struct btrfs_device *device, u64 new_size)
1579{
1580 int ret;
1581 lock_chunks(device->dev_root);
1582 ret = __btrfs_grow_device(trans, device, new_size);
1583 unlock_chunks(device->dev_root);
1584 return ret;
1585}
1586
8f18cf13
CM		 1587static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1588 struct btrfs_root *root,
1589 u64 chunk_tree, u64 chunk_objectid,
1590 u64 chunk_offset)
1591{
1592 int ret;
1593 struct btrfs_path *path;
1594 struct btrfs_key key;
1595
1596 root = root->fs_info->chunk_root;
1597 path = btrfs_alloc_path();
1598 if (!path)
1599 return -ENOMEM;
1600
1601 key.objectid = chunk_objectid;
1602 key.offset = chunk_offset;
1603 key.type = BTRFS_CHUNK_ITEM_KEY;
1604
1605 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1606 BUG_ON(ret);
1607
1608 ret = btrfs_del_item(trans, root, path);
1609 BUG_ON(ret);
1610
1611 btrfs_free_path(path);
1612 return 0;
1613}
1614
b2950863 1615static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1616 chunk_offset)
1617{
1618 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1619 struct btrfs_disk_key *disk_key;
1620 struct btrfs_chunk *chunk;
1621 u8 *ptr;
1622 int ret = 0;
1623 u32 num_stripes;
1624 u32 array_size;
1625 u32 len = 0;
1626 u32 cur;
1627 struct btrfs_key key;
1628
1629 array_size = btrfs_super_sys_array_size(super_copy);
1630
1631 ptr = super_copy->sys_chunk_array;
1632 cur = 0;
1633
1634 while (cur < array_size) {
1635 disk_key = (struct btrfs_disk_key *)ptr;
1636 btrfs_disk_key_to_cpu(&key, disk_key);
1637
1638 len = sizeof(*disk_key);
1639
1640 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1641 chunk = (struct btrfs_chunk *)(ptr + len);
1642 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1643 len += btrfs_chunk_item_size(num_stripes);
1644 } else {
1645 ret = -EIO;
1646 break;
1647 }
1648 if (key.objectid == chunk_objectid &&
1649 key.offset == chunk_offset) {
1650 memmove(ptr, ptr + len, array_size - (cur + len));
1651 array_size -= len;
1652 btrfs_set_super_sys_array_size(super_copy, array_size);
1653 } else {
1654 ptr += len;
1655 cur += len;
1656 }
1657 }
1658 return ret;
1659}
1660
b2950863 1661static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1662 u64 chunk_tree, u64 chunk_objectid,
1663 u64 chunk_offset)
1664{
1665 struct extent_map_tree *em_tree;
1666 struct btrfs_root *extent_root;
1667 struct btrfs_trans_handle *trans;
1668 struct extent_map *em;
1669 struct map_lookup *map;
1670 int ret;
1671 int i;
1672
d397712b 1673 printk(KERN_INFO "btrfs relocating chunk %llu\n",
323da79c 1674 (unsigned long long)chunk_offset);
8f18cf13
CM		 1675 root = root->fs_info->chunk_root;
1676 extent_root = root->fs_info->extent_root;
1677 em_tree = &root->fs_info->mapping_tree.map_tree;
1678
1679 /* step one, relocate all the extents inside this chunk */
1a40e23b 1680 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
8f18cf13
CM		 1681 BUG_ON(ret);
1682
1683 trans = btrfs_start_transaction(root, 1);
1684 BUG_ON(!trans);
1685
7d9eb12c
CM		 1686 lock_chunks(root);
1687
8f18cf13
CM		 1688 /*
1689 * step two, delete the device extents and the
1690 * chunk tree entries
1691 */
1692 spin_lock(&em_tree->lock);
1693 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1694 spin_unlock(&em_tree->lock);
1695
a061fc8d
CM		 1696 BUG_ON(em->start > chunk_offset ||
1697 em->start + em->len < chunk_offset);
8f18cf13
CM		 1698 map = (struct map_lookup *)em->bdev;
1699
1700 for (i = 0; i < map->num_stripes; i++) {
1701 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1702 map->stripes[i].physical);
1703 BUG_ON(ret);
a061fc8d 1704
dfe25020
CM		 1705 if (map->stripes[i].dev) {
1706 ret = btrfs_update_device(trans, map->stripes[i].dev);
1707 BUG_ON(ret);
1708 }
8f18cf13
CM		 1709 }
1710 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1711 chunk_offset);
1712
1713 BUG_ON(ret);
1714
1715 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1716 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1717 BUG_ON(ret);
8f18cf13
CM		 1718 }
1719
2b82032c
YZ		 1720 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1721 BUG_ON(ret);
1722
1723 spin_lock(&em_tree->lock);
1724 remove_extent_mapping(em_tree, em);
1725 spin_unlock(&em_tree->lock);
1726
1727 kfree(map);
1728 em->bdev = NULL;
1729
1730 /* once for the tree */
1731 free_extent_map(em);
1732 /* once for us */
1733 free_extent_map(em);
1734
1735 unlock_chunks(root);
1736 btrfs_end_transaction(trans, root);
1737 return 0;
1738}
1739
1740static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1741{
1742 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1743 struct btrfs_path *path;
1744 struct extent_buffer *leaf;
1745 struct btrfs_chunk *chunk;
1746 struct btrfs_key key;
1747 struct btrfs_key found_key;
1748 u64 chunk_tree = chunk_root->root_key.objectid;
1749 u64 chunk_type;
1750 int ret;
1751
1752 path = btrfs_alloc_path();
1753 if (!path)
1754 return -ENOMEM;
1755
1756 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1757 key.offset = (u64)-1;
1758 key.type = BTRFS_CHUNK_ITEM_KEY;
1759
1760 while (1) {
1761 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1762 if (ret < 0)
1763 goto error;
1764 BUG_ON(ret == 0);
1765
1766 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1767 key.type);
1768 if (ret < 0)
1769 goto error;
1770 if (ret > 0)
1771 break;
1a40e23b 1772
2b82032c
YZ		 1773 leaf = path->nodes[0];
1774 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 1775
2b82032c
YZ		 1776 chunk = btrfs_item_ptr(leaf, path->slots[0],
1777 struct btrfs_chunk);
1778 chunk_type = btrfs_chunk_type(leaf, chunk);
1779 btrfs_release_path(chunk_root, path);
8f18cf13 1780
2b82032c
YZ		 1781 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1782 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1783 found_key.objectid,
1784 found_key.offset);
1785 BUG_ON(ret);
1786 }
8f18cf13 1787
2b82032c
YZ		 1788 if (found_key.offset == 0)
1789 break;
1790 key.offset = found_key.offset - 1;
1791 }
1792 ret = 0;
1793error:
1794 btrfs_free_path(path);
1795 return ret;
8f18cf13
CM		 1796}
1797
ec44a35c
CM		 1798static u64 div_factor(u64 num, int factor)
1799{
1800 if (factor == 10)
1801 return num;
1802 num *= factor;
1803 do_div(num, 10);
1804 return num;
1805}
1806
ec44a35c
CM		 1807int btrfs_balance(struct btrfs_root *dev_root)
1808{
1809 int ret;
ec44a35c
CM		 1810 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1811 struct btrfs_device *device;
1812 u64 old_size;
1813 u64 size_to_free;
1814 struct btrfs_path *path;
1815 struct btrfs_key key;
1816 struct btrfs_chunk *chunk;
1817 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1818 struct btrfs_trans_handle *trans;
1819 struct btrfs_key found_key;
1820
2b82032c
YZ		 1821 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1822 return -EROFS;
ec44a35c 1823
7d9eb12c 1824 mutex_lock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1825 dev_root = dev_root->fs_info->dev_root;
1826
ec44a35c 1827 /* step one make some room on all the devices */
c6e30871 1828 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 1829 old_size = device->total_bytes;
1830 size_to_free = div_factor(old_size, 1);
1831 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 1832 if (!device->writeable ||
1833 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 1834 continue;
1835
1836 ret = btrfs_shrink_device(device, old_size - size_to_free);
1837 BUG_ON(ret);
1838
1839 trans = btrfs_start_transaction(dev_root, 1);
1840 BUG_ON(!trans);
1841
1842 ret = btrfs_grow_device(trans, device, old_size);
1843 BUG_ON(ret);
1844
1845 btrfs_end_transaction(trans, dev_root);
1846 }
1847
1848 /* step two, relocate all the chunks */
1849 path = btrfs_alloc_path();
1850 BUG_ON(!path);
1851
1852 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1853 key.offset = (u64)-1;
1854 key.type = BTRFS_CHUNK_ITEM_KEY;
1855
d397712b 1856 while (1) {
ec44a35c
CM		 1857 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1858 if (ret < 0)
1859 goto error;
1860
1861 /*
1862 * this shouldn't happen, it means the last relocate
1863 * failed
1864 */
1865 if (ret == 0)
1866 break;
1867
1868 ret = btrfs_previous_item(chunk_root, path, 0,
1869 BTRFS_CHUNK_ITEM_KEY);
7d9eb12c 1870 if (ret)
ec44a35c 1871 break;
7d9eb12c 1872
ec44a35c
CM		 1873 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1874 path->slots[0]);
1875 if (found_key.objectid != key.objectid)
1876 break;
7d9eb12c 1877
ec44a35c
CM		 1878 chunk = btrfs_item_ptr(path->nodes[0],
1879 path->slots[0],
1880 struct btrfs_chunk);
1881 key.offset = found_key.offset;
1882 /* chunk zero is special */
1883 if (key.offset == 0)
1884 break;
1885
7d9eb12c 1886 btrfs_release_path(chunk_root, path);
ec44a35c
CM		 1887 ret = btrfs_relocate_chunk(chunk_root,
1888 chunk_root->root_key.objectid,
1889 found_key.objectid,
1890 found_key.offset);
1891 BUG_ON(ret);
ec44a35c
CM		 1892 }
1893 ret = 0;
1894error:
1895 btrfs_free_path(path);
7d9eb12c 1896 mutex_unlock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 1897 return ret;
1898}
1899
8f18cf13
CM		 1900/*
1901 * shrinking a device means finding all of the device extents past
1902 * the new size, and then following the back refs to the chunks.
1903 * The chunk relocation code actually frees the device extent
1904 */
1905int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1906{
1907 struct btrfs_trans_handle *trans;
1908 struct btrfs_root *root = device->dev_root;
1909 struct btrfs_dev_extent *dev_extent = NULL;
1910 struct btrfs_path *path;
1911 u64 length;
1912 u64 chunk_tree;
1913 u64 chunk_objectid;
1914 u64 chunk_offset;
1915 int ret;
1916 int slot;
1917 struct extent_buffer *l;
1918 struct btrfs_key key;
1919 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1920 u64 old_total = btrfs_super_total_bytes(super_copy);
1921 u64 diff = device->total_bytes - new_size;
1922
2b82032c
YZ		 1923 if (new_size >= device->total_bytes)
1924 return -EINVAL;
8f18cf13
CM		 1925
1926 path = btrfs_alloc_path();
1927 if (!path)
1928 return -ENOMEM;
1929
1930 trans = btrfs_start_transaction(root, 1);
1931 if (!trans) {
1932 ret = -ENOMEM;
1933 goto done;
1934 }
1935
1936 path->reada = 2;
1937
7d9eb12c
CM		 1938 lock_chunks(root);
1939
8f18cf13 1940 device->total_bytes = new_size;
2b82032c
YZ		 1941 if (device->writeable)
1942 device->fs_devices->total_rw_bytes -= diff;
8f18cf13
CM		 1943 ret = btrfs_update_device(trans, device);
1944 if (ret) {
7d9eb12c 1945 unlock_chunks(root);
8f18cf13
CM		 1946 btrfs_end_transaction(trans, root);
1947 goto done;
1948 }
1949 WARN_ON(diff > old_total);
1950 btrfs_set_super_total_bytes(super_copy, old_total - diff);
7d9eb12c 1951 unlock_chunks(root);
8f18cf13
CM		 1952 btrfs_end_transaction(trans, root);
1953
1954 key.objectid = device->devid;
1955 key.offset = (u64)-1;
1956 key.type = BTRFS_DEV_EXTENT_KEY;
1957
1958 while (1) {
1959 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1960 if (ret < 0)
1961 goto done;
1962
1963 ret = btrfs_previous_item(root, path, 0, key.type);
1964 if (ret < 0)
1965 goto done;
1966 if (ret) {
1967 ret = 0;
1968 goto done;
1969 }
1970
1971 l = path->nodes[0];
1972 slot = path->slots[0];
1973 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1974
1975 if (key.objectid != device->devid)
1976 goto done;
1977
1978 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1979 length = btrfs_dev_extent_length(l, dev_extent);
1980
1981 if (key.offset + length <= new_size)
1982 goto done;
1983
1984 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1985 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1986 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1987 btrfs_release_path(root, path);
1988
1989 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1990 chunk_offset);
1991 if (ret)
1992 goto done;
1993 }
1994
1995done:
1996 btrfs_free_path(path);
1997 return ret;
1998}
1999
b2950863 2000static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2001 struct btrfs_root *root,
2002 struct btrfs_key *key,
2003 struct btrfs_chunk *chunk, int item_size)
2004{
2005 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2006 struct btrfs_disk_key disk_key;
2007 u32 array_size;
2008 u8 *ptr;
2009
2010 array_size = btrfs_super_sys_array_size(super_copy);
2011 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2012 return -EFBIG;
2013
2014 ptr = super_copy->sys_chunk_array + array_size;
2015 btrfs_cpu_key_to_disk(&disk_key, key);
2016 memcpy(ptr, &disk_key, sizeof(disk_key));
2017 ptr += sizeof(disk_key);
2018 memcpy(ptr, chunk, item_size);
2019 item_size += sizeof(disk_key);
2020 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2021 return 0;
2022}
2023
d397712b 2024static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
a1b32a59 2025 int num_stripes, int sub_stripes)
9b3f68b9
CM		 2026{
2027 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2028 return calc_size;
2029 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2030 return calc_size * (num_stripes / sub_stripes);
2031 else
2032 return calc_size * num_stripes;
2033}
2034
2b82032c
YZ		 2035static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *extent_root,
2037 struct map_lookup **map_ret,
2038 u64 *num_bytes, u64 *stripe_size,
2039 u64 start, u64 type)
0b86a832 2040{
593060d7 2041 struct btrfs_fs_info *info = extent_root->fs_info;
0b86a832 2042 struct btrfs_device *device = NULL;
2b82032c 2043 struct btrfs_fs_devices *fs_devices = info->fs_devices;
6324fbf3 2044 struct list_head *cur;
2b82032c 2045 struct map_lookup *map = NULL;
0b86a832 2046 struct extent_map_tree *em_tree;
0b86a832 2047 struct extent_map *em;
2b82032c 2048 struct list_head private_devs;
a40a90a0 2049 int min_stripe_size = 1 * 1024 * 1024;
0b86a832 2050 u64 calc_size = 1024 * 1024 * 1024;
9b3f68b9
CM		 2051 u64 max_chunk_size = calc_size;
2052 u64 min_free;
6324fbf3
CM		 2053 u64 avail;
2054 u64 max_avail = 0;
2b82032c 2055 u64 dev_offset;
6324fbf3 2056 int num_stripes = 1;
a40a90a0 2057 int min_stripes = 1;
321aecc6 2058 int sub_stripes = 0;
6324fbf3 2059 int looped = 0;
0b86a832 2060 int ret;
6324fbf3 2061 int index;
593060d7 2062 int stripe_len = 64 * 1024;
0b86a832 2063
ec44a35c
CM		 2064 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2065 (type & BTRFS_BLOCK_GROUP_DUP)) {
2066 WARN_ON(1);
2067 type &= ~BTRFS_BLOCK_GROUP_DUP;
2068 }
2b82032c 2069 if (list_empty(&fs_devices->alloc_list))
6324fbf3 2070 return -ENOSPC;
593060d7 2071
a40a90a0 2072 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2b82032c 2073 num_stripes = fs_devices->rw_devices;
a40a90a0
CM		 2074 min_stripes = 2;
2075 }
2076 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
611f0e00 2077 num_stripes = 2;
a40a90a0
CM		 2078 min_stripes = 2;
2079 }
8790d502 2080 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2b82032c 2081 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
9b3f68b9
CM		 2082 if (num_stripes < 2)
2083 return -ENOSPC;
a40a90a0 2084 min_stripes = 2;
8790d502 2085 }
321aecc6 2086 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2b82032c 2087 num_stripes = fs_devices->rw_devices;
321aecc6
CM		 2088 if (num_stripes < 4)
2089 return -ENOSPC;
2090 num_stripes &= ~(u32)1;
2091 sub_stripes = 2;
a40a90a0 2092 min_stripes = 4;
321aecc6 2093 }
9b3f68b9
CM		 2094
2095 if (type & BTRFS_BLOCK_GROUP_DATA) {
2096 max_chunk_size = 10 * calc_size;
a40a90a0 2097 min_stripe_size = 64 * 1024 * 1024;
9b3f68b9
CM		 2098 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2099 max_chunk_size = 4 * calc_size;
a40a90a0
CM		 2100 min_stripe_size = 32 * 1024 * 1024;
2101 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2102 calc_size = 8 * 1024 * 1024;
2103 max_chunk_size = calc_size * 2;
2104 min_stripe_size = 1 * 1024 * 1024;
9b3f68b9
CM		 2105 }
2106
2b82032c
YZ		 2107 /* we don't want a chunk larger than 10% of writeable space */
2108 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2109 max_chunk_size);
9b3f68b9 2110
a40a90a0 2111again:
2b82032c
YZ		 2112 if (!map || map->num_stripes != num_stripes) {
2113 kfree(map);
2114 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2115 if (!map)
2116 return -ENOMEM;
2117 map->num_stripes = num_stripes;
2118 }
2119
9b3f68b9
CM		 2120 if (calc_size * num_stripes > max_chunk_size) {
2121 calc_size = max_chunk_size;
2122 do_div(calc_size, num_stripes);
2123 do_div(calc_size, stripe_len);
2124 calc_size *= stripe_len;
2125 }
2126 /* we don't want tiny stripes */
a40a90a0 2127 calc_size = max_t(u64, min_stripe_size, calc_size);
9b3f68b9 2128
9b3f68b9
CM		 2129 do_div(calc_size, stripe_len);
2130 calc_size *= stripe_len;
2131
2b82032c 2132 cur = fs_devices->alloc_list.next;
6324fbf3 2133 index = 0;
611f0e00
CM		 2134
2135 if (type & BTRFS_BLOCK_GROUP_DUP)
2136 min_free = calc_size * 2;
9b3f68b9
CM		 2137 else
2138 min_free = calc_size;
611f0e00 2139
0f9dd46c
JB		 2140 /*
2141 * we add 1MB because we never use the first 1MB of the device, unless
2142 * we've looped, then we are likely allocating the maximum amount of
2143 * space left already
2144 */
2145 if (!looped)
2146 min_free += 1024 * 1024;
ad5bd91e 2147
2b82032c 2148 INIT_LIST_HEAD(&private_devs);
d397712b 2149 while (index < num_stripes) {
b3075717 2150 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2b82032c 2151 BUG_ON(!device->writeable);
dfe25020
CM		 2152 if (device->total_bytes > device->bytes_used)
2153 avail = device->total_bytes - device->bytes_used;
2154 else
2155 avail = 0;
6324fbf3 2156 cur = cur->next;
8f18cf13 2157
dfe25020 2158 if (device->in_fs_metadata && avail >= min_free) {
2b82032c
YZ		 2159 ret = find_free_dev_extent(trans, device,
2160 min_free, &dev_offset);
8f18cf13
CM		 2161 if (ret == 0) {
2162 list_move_tail(&device->dev_alloc_list,
2163 &private_devs);
2b82032c
YZ		 2164 map->stripes[index].dev = device;
2165 map->stripes[index].physical = dev_offset;
611f0e00 2166 index++;
2b82032c
YZ		 2167 if (type & BTRFS_BLOCK_GROUP_DUP) {
2168 map->stripes[index].dev = device;
2169 map->stripes[index].physical =
2170 dev_offset + calc_size;
8f18cf13 2171 index++;
2b82032c 2172 }
8f18cf13 2173 }
dfe25020 2174 } else if (device->in_fs_metadata && avail > max_avail)
a40a90a0 2175 max_avail = avail;
2b82032c 2176 if (cur == &fs_devices->alloc_list)
6324fbf3
CM		 2177 break;
2178 }
2b82032c 2179 list_splice(&private_devs, &fs_devices->alloc_list);
6324fbf3 2180 if (index < num_stripes) {
a40a90a0
CM		 2181 if (index >= min_stripes) {
2182 num_stripes = index;
2183 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2184 num_stripes /= sub_stripes;
2185 num_stripes *= sub_stripes;
2186 }
2187 looped = 1;
2188 goto again;
2189 }
6324fbf3
CM		 2190 if (!looped && max_avail > 0) {
2191 looped = 1;
2192 calc_size = max_avail;
2193 goto again;
2194 }
2b82032c 2195 kfree(map);
6324fbf3
CM		 2196 return -ENOSPC;
2197 }
2b82032c
YZ		 2198 map->sector_size = extent_root->sectorsize;
2199 map->stripe_len = stripe_len;
2200 map->io_align = stripe_len;
2201 map->io_width = stripe_len;
2202 map->type = type;
2203 map->num_stripes = num_stripes;
2204 map->sub_stripes = sub_stripes;
0b86a832 2205
2b82032c
YZ		 2206 *map_ret = map;
2207 *stripe_size = calc_size;
2208 *num_bytes = chunk_bytes_by_type(type, calc_size,
2209 num_stripes, sub_stripes);
0b86a832 2210
2b82032c
YZ		 2211 em = alloc_extent_map(GFP_NOFS);
2212 if (!em) {
2213 kfree(map);
593060d7
CM		 2214 return -ENOMEM;
2215 }
2b82032c
YZ		 2216 em->bdev = (struct block_device *)map;
2217 em->start = start;
2218 em->len = *num_bytes;
2219 em->block_start = 0;
2220 em->block_len = em->len;
593060d7 2221
2b82032c
YZ		 2222 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2223 spin_lock(&em_tree->lock);
2224 ret = add_extent_mapping(em_tree, em);
2225 spin_unlock(&em_tree->lock);
2226 BUG_ON(ret);
2227 free_extent_map(em);
0b86a832 2228
2b82032c
YZ		 2229 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2230 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2231 start, *num_bytes);
2232 BUG_ON(ret);
611f0e00 2233
2b82032c
YZ		 2234 index = 0;
2235 while (index < map->num_stripes) {
2236 device = map->stripes[index].dev;
2237 dev_offset = map->stripes[index].physical;
0b86a832
CM		 2238
2239 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2240 info->chunk_root->root_key.objectid,
2241 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2242 start, dev_offset, calc_size);
0b86a832 2243 BUG_ON(ret);
2b82032c
YZ		 2244 index++;
2245 }
2246
2247 return 0;
2248}
2249
2250static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2251 struct btrfs_root *extent_root,
2252 struct map_lookup *map, u64 chunk_offset,
2253 u64 chunk_size, u64 stripe_size)
2254{
2255 u64 dev_offset;
2256 struct btrfs_key key;
2257 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2258 struct btrfs_device *device;
2259 struct btrfs_chunk *chunk;
2260 struct btrfs_stripe *stripe;
2261 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2262 int index = 0;
2263 int ret;
2264
2265 chunk = kzalloc(item_size, GFP_NOFS);
2266 if (!chunk)
2267 return -ENOMEM;
2268
2269 index = 0;
2270 while (index < map->num_stripes) {
2271 device = map->stripes[index].dev;
2272 device->bytes_used += stripe_size;
0b86a832
CM		 2273 ret = btrfs_update_device(trans, device);
2274 BUG_ON(ret);
2b82032c
YZ		 2275 index++;
2276 }
2277
2278 index = 0;
2279 stripe = &chunk->stripe;
2280 while (index < map->num_stripes) {
2281 device = map->stripes[index].dev;
2282 dev_offset = map->stripes[index].physical;
0b86a832 2283
e17cade2
CM		 2284 btrfs_set_stack_stripe_devid(stripe, device->devid);
2285 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2286 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2287 stripe++;
0b86a832
CM		 2288 index++;
2289 }
2290
2b82032c 2291 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2292 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2293 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2294 btrfs_set_stack_chunk_type(chunk, map->type);
2295 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2296 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2297 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2298 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2299 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2300
2b82032c
YZ		 2301 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2302 key.type = BTRFS_CHUNK_ITEM_KEY;
2303 key.offset = chunk_offset;
0b86a832 2304
2b82032c
YZ		 2305 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2306 BUG_ON(ret);
0b86a832 2307
2b82032c
YZ		 2308 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2309 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2310 item_size);
8f18cf13
CM		 2311 BUG_ON(ret);
2312 }
0b86a832 2313 kfree(chunk);
2b82032c
YZ		 2314 return 0;
2315}
0b86a832 2316
2b82032c
YZ		 2317/*
2318 * Chunk allocation falls into two parts. The first part does works
2319 * that make the new allocated chunk useable, but not do any operation
2320 * that modifies the chunk tree. The second part does the works that
2321 * require modifying the chunk tree. This division is important for the
2322 * bootstrap process of adding storage to a seed btrfs.
2323 */
2324int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2325 struct btrfs_root *extent_root, u64 type)
2326{
2327 u64 chunk_offset;
2328 u64 chunk_size;
2329 u64 stripe_size;
2330 struct map_lookup *map;
2331 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2332 int ret;
2333
2334 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2335 &chunk_offset);
2336 if (ret)
2337 return ret;
2338
2339 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2340 &stripe_size, chunk_offset, type);
2341 if (ret)
2342 return ret;
2343
2344 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2345 chunk_size, stripe_size);
2346 BUG_ON(ret);
2347 return 0;
2348}
2349
d397712b 2350static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2351 struct btrfs_root *root,
2352 struct btrfs_device *device)
2353{
2354 u64 chunk_offset;
2355 u64 sys_chunk_offset;
2356 u64 chunk_size;
2357 u64 sys_chunk_size;
2358 u64 stripe_size;
2359 u64 sys_stripe_size;
2360 u64 alloc_profile;
2361 struct map_lookup *map;
2362 struct map_lookup *sys_map;
2363 struct btrfs_fs_info *fs_info = root->fs_info;
2364 struct btrfs_root *extent_root = fs_info->extent_root;
2365 int ret;
2366
2367 ret = find_next_chunk(fs_info->chunk_root,
2368 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2369 BUG_ON(ret);
2370
2371 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2372 (fs_info->metadata_alloc_profile &
2373 fs_info->avail_metadata_alloc_bits);
2374 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2375
2376 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2377 &stripe_size, chunk_offset, alloc_profile);
2378 BUG_ON(ret);
2379
2380 sys_chunk_offset = chunk_offset + chunk_size;
2381
2382 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2383 (fs_info->system_alloc_profile &
2384 fs_info->avail_system_alloc_bits);
2385 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2386
2387 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2388 &sys_chunk_size, &sys_stripe_size,
2389 sys_chunk_offset, alloc_profile);
2390 BUG_ON(ret);
2391
2392 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2393 BUG_ON(ret);
2394
2395 /*
2396 * Modifying chunk tree needs allocating new blocks from both
2397 * system block group and metadata block group. So we only can
2398 * do operations require modifying the chunk tree after both
2399 * block groups were created.
2400 */
2401 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2402 chunk_size, stripe_size);
2403 BUG_ON(ret);
2404
2405 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2406 sys_chunk_offset, sys_chunk_size,
2407 sys_stripe_size);
b248a415 2408 BUG_ON(ret);
2b82032c
YZ		 2409 return 0;
2410}
2411
2412int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2413{
2414 struct extent_map *em;
2415 struct map_lookup *map;
2416 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2417 int readonly = 0;
2418 int i;
2419
2420 spin_lock(&map_tree->map_tree.lock);
2421 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2422 spin_unlock(&map_tree->map_tree.lock);
2423 if (!em)
2424 return 1;
2425
2426 map = (struct map_lookup *)em->bdev;
2427 for (i = 0; i < map->num_stripes; i++) {
2428 if (!map->stripes[i].dev->writeable) {
2429 readonly = 1;
2430 break;
2431 }
2432 }
0b86a832 2433 free_extent_map(em);
2b82032c 2434 return readonly;
0b86a832
CM		 2435}
2436
2437void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2438{
2439 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2440}
2441
2442void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2443{
2444 struct extent_map *em;
2445
d397712b 2446 while (1) {
0b86a832
CM		 2447 spin_lock(&tree->map_tree.lock);
2448 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2449 if (em)
2450 remove_extent_mapping(&tree->map_tree, em);
2451 spin_unlock(&tree->map_tree.lock);
2452 if (!em)
2453 break;
2454 kfree(em->bdev);
2455 /* once for us */
2456 free_extent_map(em);
2457 /* once for the tree */
2458 free_extent_map(em);
2459 }
2460}
2461
f188591e
CM		 2462int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2463{
2464 struct extent_map *em;
2465 struct map_lookup *map;
2466 struct extent_map_tree *em_tree = &map_tree->map_tree;
2467 int ret;
2468
2469 spin_lock(&em_tree->lock);
2470 em = lookup_extent_mapping(em_tree, logical, len);
b248a415 2471 spin_unlock(&em_tree->lock);
f188591e
CM		 2472 BUG_ON(!em);
2473
2474 BUG_ON(em->start > logical || em->start + em->len < logical);
2475 map = (struct map_lookup *)em->bdev;
2476 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2477 ret = map->num_stripes;
321aecc6
CM		 2478 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2479 ret = map->sub_stripes;
f188591e
CM		 2480 else
2481 ret = 1;
2482 free_extent_map(em);
f188591e
CM		 2483 return ret;
2484}
2485
dfe25020
CM		 2486static int find_live_mirror(struct map_lookup *map, int first, int num,
2487 int optimal)
2488{
2489 int i;
2490 if (map->stripes[optimal].dev->bdev)
2491 return optimal;
2492 for (i = first; i < first + num; i++) {
2493 if (map->stripes[i].dev->bdev)
2494 return i;
2495 }
2496 /* we couldn't find one that doesn't fail. Just return something
2497 * and the io error handling code will clean up eventually
2498 */
2499 return optimal;
2500}
2501
f2d8d74d
CM		 2502static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2503 u64 logical, u64 *length,
2504 struct btrfs_multi_bio **multi_ret,
2505 int mirror_num, struct page *unplug_page)
0b86a832
CM		 2506{
2507 struct extent_map *em;
2508 struct map_lookup *map;
2509 struct extent_map_tree *em_tree = &map_tree->map_tree;
2510 u64 offset;
593060d7
CM		 2511 u64 stripe_offset;
2512 u64 stripe_nr;
cea9e445 2513 int stripes_allocated = 8;
321aecc6 2514 int stripes_required = 1;
593060d7 2515 int stripe_index;
cea9e445 2516 int i;
f2d8d74d 2517 int num_stripes;
a236aed1 2518 int max_errors = 0;
cea9e445 2519 struct btrfs_multi_bio *multi = NULL;
0b86a832 2520
d397712b 2521 if (multi_ret && !(rw & (1 << BIO_RW)))
cea9e445 2522 stripes_allocated = 1;
cea9e445
CM		 2523again:
2524 if (multi_ret) {
2525 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2526 GFP_NOFS);
2527 if (!multi)
2528 return -ENOMEM;
a236aed1
CM		 2529
2530 atomic_set(&multi->error, 0);
cea9e445 2531 }
0b86a832
CM		 2532
2533 spin_lock(&em_tree->lock);
2534 em = lookup_extent_mapping(em_tree, logical, *length);
b248a415 2535 spin_unlock(&em_tree->lock);
f2d8d74d
CM		 2536
2537 if (!em && unplug_page)
2538 return 0;
2539
3b951516 2540 if (!em) {
d397712b
CM		 2541 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2542 (unsigned long long)logical,
2543 (unsigned long long)*length);
f2d8d74d 2544 BUG();
3b951516 2545 }
0b86a832
CM		 2546
2547 BUG_ON(em->start > logical || em->start + em->len < logical);
2548 map = (struct map_lookup *)em->bdev;
2549 offset = logical - em->start;
593060d7 2550
f188591e
CM		 2551 if (mirror_num > map->num_stripes)
2552 mirror_num = 0;
2553
cea9e445 2554 /* if our multi bio struct is too small, back off and try again */
321aecc6
CM		 2555 if (rw & (1 << BIO_RW)) {
2556 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2557 BTRFS_BLOCK_GROUP_DUP)) {
2558 stripes_required = map->num_stripes;
a236aed1 2559 max_errors = 1;
321aecc6
CM		 2560 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2561 stripes_required = map->sub_stripes;
a236aed1 2562 max_errors = 1;
321aecc6
CM		 2563 }
2564 }
ffbd517d 2565 if (multi_ret && (rw & (1 << BIO_RW)) &&
321aecc6 2566 stripes_allocated < stripes_required) {
cea9e445 2567 stripes_allocated = map->num_stripes;
cea9e445
CM		 2568 free_extent_map(em);
2569 kfree(multi);
2570 goto again;
2571 }
593060d7
CM		 2572 stripe_nr = offset;
2573 /*
2574 * stripe_nr counts the total number of stripes we have to stride
2575 * to get to this block
2576 */
2577 do_div(stripe_nr, map->stripe_len);
2578
2579 stripe_offset = stripe_nr * map->stripe_len;
2580 BUG_ON(offset < stripe_offset);
2581
2582 /* stripe_offset is the offset of this block in its stripe*/
2583 stripe_offset = offset - stripe_offset;
2584
cea9e445 2585 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 2586 BTRFS_BLOCK_GROUP_RAID10 |
cea9e445
CM		 2587 BTRFS_BLOCK_GROUP_DUP)) {
2588 /* we limit the length of each bio to what fits in a stripe */
2589 *length = min_t(u64, em->len - offset,
2590 map->stripe_len - stripe_offset);
2591 } else {
2592 *length = em->len - offset;
2593 }
f2d8d74d
CM		 2594
2595 if (!multi_ret && !unplug_page)
cea9e445
CM		 2596 goto out;
2597
f2d8d74d 2598 num_stripes = 1;
cea9e445 2599 stripe_index = 0;
8790d502 2600 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
f2d8d74d
CM		 2601 if (unplug_page || (rw & (1 << BIO_RW)))
2602 num_stripes = map->num_stripes;
2fff734f 2603 else if (mirror_num)
f188591e 2604 stripe_index = mirror_num - 1;
dfe25020
CM		 2605 else {
2606 stripe_index = find_live_mirror(map, 0,
2607 map->num_stripes,
2608 current->pid % map->num_stripes);
2609 }
2fff734f 2610
611f0e00 2611 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
cea9e445 2612 if (rw & (1 << BIO_RW))
f2d8d74d 2613 num_stripes = map->num_stripes;
f188591e
CM		 2614 else if (mirror_num)
2615 stripe_index = mirror_num - 1;
2fff734f 2616
321aecc6
CM		 2617 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2618 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 2619
2620 stripe_index = do_div(stripe_nr, factor);
2621 stripe_index *= map->sub_stripes;
2622
f2d8d74d
CM		 2623 if (unplug_page || (rw & (1 << BIO_RW)))
2624 num_stripes = map->sub_stripes;
321aecc6
CM		 2625 else if (mirror_num)
2626 stripe_index += mirror_num - 1;
dfe25020
CM		 2627 else {
2628 stripe_index = find_live_mirror(map, stripe_index,
2629 map->sub_stripes, stripe_index +
2630 current->pid % map->sub_stripes);
2631 }
8790d502
CM		 2632 } else {
2633 /*
2634 * after this do_div call, stripe_nr is the number of stripes
2635 * on this device we have to walk to find the data, and
2636 * stripe_index is the number of our device in the stripe array
2637 */
2638 stripe_index = do_div(stripe_nr, map->num_stripes);
2639 }
593060d7 2640 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 2641
f2d8d74d
CM		 2642 for (i = 0; i < num_stripes; i++) {
2643 if (unplug_page) {
2644 struct btrfs_device *device;
2645 struct backing_dev_info *bdi;
2646
2647 device = map->stripes[stripe_index].dev;
dfe25020
CM		 2648 if (device->bdev) {
2649 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 2650 if (bdi->unplug_io_fn)
dfe25020 2651 bdi->unplug_io_fn(bdi, unplug_page);
f2d8d74d
CM		 2652 }
2653 } else {
2654 multi->stripes[i].physical =
2655 map->stripes[stripe_index].physical +
2656 stripe_offset + stripe_nr * map->stripe_len;
2657 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2658 }
cea9e445 2659 stripe_index++;
593060d7 2660 }
f2d8d74d
CM		 2661 if (multi_ret) {
2662 *multi_ret = multi;
2663 multi->num_stripes = num_stripes;
a236aed1 2664 multi->max_errors = max_errors;
f2d8d74d 2665 }
cea9e445 2666out:
0b86a832 2667 free_extent_map(em);
0b86a832
CM		 2668 return 0;
2669}
2670
f2d8d74d
CM		 2671int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2672 u64 logical, u64 *length,
2673 struct btrfs_multi_bio **multi_ret, int mirror_num)
2674{
2675 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2676 mirror_num, NULL);
2677}
2678
a512bbf8
YZ		 2679int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2680 u64 chunk_start, u64 physical, u64 devid,
2681 u64 **logical, int *naddrs, int *stripe_len)
2682{
2683 struct extent_map_tree *em_tree = &map_tree->map_tree;
2684 struct extent_map *em;
2685 struct map_lookup *map;
2686 u64 *buf;
2687 u64 bytenr;
2688 u64 length;
2689 u64 stripe_nr;
2690 int i, j, nr = 0;
2691
2692 spin_lock(&em_tree->lock);
2693 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2694 spin_unlock(&em_tree->lock);
2695
2696 BUG_ON(!em || em->start != chunk_start);
2697 map = (struct map_lookup *)em->bdev;
2698
2699 length = em->len;
2700 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2701 do_div(length, map->num_stripes / map->sub_stripes);
2702 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2703 do_div(length, map->num_stripes);
2704
2705 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2706 BUG_ON(!buf);
2707
2708 for (i = 0; i < map->num_stripes; i++) {
2709 if (devid && map->stripes[i].dev->devid != devid)
2710 continue;
2711 if (map->stripes[i].physical > physical ||
2712 map->stripes[i].physical + length <= physical)
2713 continue;
2714
2715 stripe_nr = physical - map->stripes[i].physical;
2716 do_div(stripe_nr, map->stripe_len);
2717
2718 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2719 stripe_nr = stripe_nr * map->num_stripes + i;
2720 do_div(stripe_nr, map->sub_stripes);
2721 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2722 stripe_nr = stripe_nr * map->num_stripes + i;
2723 }
2724 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 2725 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 2726 for (j = 0; j < nr; j++) {
2727 if (buf[j] == bytenr)
2728 break;
2729 }
934d375b
CM		 2730 if (j == nr) {
2731 WARN_ON(nr >= map->num_stripes);
a512bbf8 2732 buf[nr++] = bytenr;
934d375b 2733 }
a512bbf8
YZ		 2734 }
2735
2736 for (i = 0; i > nr; i++) {
2737 struct btrfs_multi_bio *multi;
2738 struct btrfs_bio_stripe *stripe;
2739 int ret;
2740
2741 length = 1;
2742 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2743 &length, &multi, 0);
2744 BUG_ON(ret);
2745
2746 stripe = multi->stripes;
2747 for (j = 0; j < multi->num_stripes; j++) {
2748 if (stripe->physical >= physical &&
2749 physical < stripe->physical + length)
2750 break;
2751 }
2752 BUG_ON(j >= multi->num_stripes);
2753 kfree(multi);
2754 }
2755
2756 *logical = buf;
2757 *naddrs = nr;
2758 *stripe_len = map->stripe_len;
2759
2760 free_extent_map(em);
2761 return 0;
2762}
2763
f2d8d74d
CM		 2764int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2765 u64 logical, struct page *page)
2766{
2767 u64 length = PAGE_CACHE_SIZE;
2768 return __btrfs_map_block(map_tree, READ, logical, &length,
2769 NULL, 0, page);
2770}
2771
8790d502 2772static void end_bio_multi_stripe(struct bio *bio, int err)
8790d502 2773{
cea9e445 2774 struct btrfs_multi_bio *multi = bio->bi_private;
7d2b4daa 2775 int is_orig_bio = 0;
8790d502 2776
8790d502 2777 if (err)
a236aed1 2778 atomic_inc(&multi->error);
8790d502 2779
7d2b4daa
CM		 2780 if (bio == multi->orig_bio)
2781 is_orig_bio = 1;
2782
cea9e445 2783 if (atomic_dec_and_test(&multi->stripes_pending)) {
7d2b4daa
CM		 2784 if (!is_orig_bio) {
2785 bio_put(bio);
2786 bio = multi->orig_bio;
2787 }
8790d502
CM		 2788 bio->bi_private = multi->private;
2789 bio->bi_end_io = multi->end_io;
a236aed1
CM		 2790 /* only send an error to the higher layers if it is
2791 * beyond the tolerance of the multi-bio
2792 */
1259ab75 2793 if (atomic_read(&multi->error) > multi->max_errors) {
a236aed1 2794 err = -EIO;
1259ab75
CM		 2795 } else if (err) {
2796 /*
2797 * this bio is actually up to date, we didn't
2798 * go over the max number of errors
2799 */
2800 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 2801 err = 0;
1259ab75 2802 }
8790d502
CM		 2803 kfree(multi);
2804
2805 bio_endio(bio, err);
7d2b4daa 2806 } else if (!is_orig_bio) {
8790d502
CM		 2807 bio_put(bio);
2808 }
8790d502
CM		 2809}
2810
8b712842
CM		 2811struct async_sched {
2812 struct bio *bio;
2813 int rw;
2814 struct btrfs_fs_info *info;
2815 struct btrfs_work work;
2816};
2817
2818/*
2819 * see run_scheduled_bios for a description of why bios are collected for
2820 * async submit.
2821 *
2822 * This will add one bio to the pending list for a device and make sure
2823 * the work struct is scheduled.
2824 */
d397712b 2825static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 2826 struct btrfs_device *device,
2827 int rw, struct bio *bio)
8b712842
CM		 2828{
2829 int should_queue = 1;
ffbd517d 2830 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 2831
2832 /* don't bother with additional async steps for reads, right now */
2833 if (!(rw & (1 << BIO_RW))) {
492bb6de 2834 bio_get(bio);
8b712842 2835 submit_bio(rw, bio);
492bb6de 2836 bio_put(bio);
8b712842
CM		 2837 return 0;
2838 }
2839
2840 /*
0986fe9e 2841 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 2842 * higher layers. Otherwise, the async bio makes it appear we have
2843 * made progress against dirty pages when we've really just put it
2844 * on a queue for later
2845 */
0986fe9e 2846 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 2847 WARN_ON(bio->bi_next);
8b712842
CM		 2848 bio->bi_next = NULL;
2849 bio->bi_rw |= rw;
2850
2851 spin_lock(&device->io_lock);
ffbd517d
CM		 2852 if (bio_sync(bio))
2853 pending_bios = &device->pending_sync_bios;
2854 else
2855 pending_bios = &device->pending_bios;
8b712842 2856
ffbd517d
CM		 2857 if (pending_bios->tail)
2858 pending_bios->tail->bi_next = bio;
8b712842 2859
ffbd517d
CM		 2860 pending_bios->tail = bio;
2861 if (!pending_bios->head)
2862 pending_bios->head = bio;
8b712842
CM		 2863 if (device->running_pending)
2864 should_queue = 0;
2865
2866 spin_unlock(&device->io_lock);
2867
2868 if (should_queue)
1cc127b5
CM		 2869 btrfs_queue_worker(&root->fs_info->submit_workers,
2870 &device->work);
8b712842
CM		 2871 return 0;
2872}
2873
f188591e 2874int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 2875 int mirror_num, int async_submit)
0b86a832
CM		 2876{
2877 struct btrfs_mapping_tree *map_tree;
2878 struct btrfs_device *dev;
8790d502 2879 struct bio *first_bio = bio;
a62b9401 2880 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 2881 u64 length = 0;
2882 u64 map_length;
cea9e445 2883 struct btrfs_multi_bio *multi = NULL;
0b86a832 2884 int ret;
8790d502
CM		 2885 int dev_nr = 0;
2886 int total_devs = 1;
0b86a832 2887
f2d8d74d 2888 length = bio->bi_size;
0b86a832
CM		 2889 map_tree = &root->fs_info->mapping_tree;
2890 map_length = length;
cea9e445 2891
f188591e
CM		 2892 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2893 mirror_num);
cea9e445
CM		 2894 BUG_ON(ret);
2895
2896 total_devs = multi->num_stripes;
2897 if (map_length < length) {
d397712b
CM		 2898 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2899 "len %llu\n", (unsigned long long)logical,
2900 (unsigned long long)length,
2901 (unsigned long long)map_length);
cea9e445
CM		 2902 BUG();
2903 }
2904 multi->end_io = first_bio->bi_end_io;
2905 multi->private = first_bio->bi_private;
7d2b4daa 2906 multi->orig_bio = first_bio;
cea9e445
CM		 2907 atomic_set(&multi->stripes_pending, multi->num_stripes);
2908
d397712b 2909 while (dev_nr < total_devs) {
8790d502 2910 if (total_devs > 1) {
8790d502
CM		 2911 if (dev_nr < total_devs - 1) {
2912 bio = bio_clone(first_bio, GFP_NOFS);
2913 BUG_ON(!bio);
2914 } else {
2915 bio = first_bio;
2916 }
2917 bio->bi_private = multi;
2918 bio->bi_end_io = end_bio_multi_stripe;
2919 }
cea9e445
CM		 2920 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2921 dev = multi->stripes[dev_nr].dev;
2b82032c 2922 BUG_ON(rw == WRITE && !dev->writeable);
dfe25020
CM		 2923 if (dev && dev->bdev) {
2924 bio->bi_bdev = dev->bdev;
8b712842
CM		 2925 if (async_submit)
2926 schedule_bio(root, dev, rw, bio);
2927 else
2928 submit_bio(rw, bio);
dfe25020
CM		 2929 } else {
2930 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2931 bio->bi_sector = logical >> 9;
dfe25020 2932 bio_endio(bio, -EIO);
dfe25020 2933 }
8790d502
CM		 2934 dev_nr++;
2935 }
cea9e445
CM		 2936 if (total_devs == 1)
2937 kfree(multi);
0b86a832
CM		 2938 return 0;
2939}
2940
a443755f 2941struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 2942 u8 *uuid, u8 *fsid)
0b86a832 2943{
2b82032c
YZ		 2944 struct btrfs_device *device;
2945 struct btrfs_fs_devices *cur_devices;
2946
2947 cur_devices = root->fs_info->fs_devices;
2948 while (cur_devices) {
2949 if (!fsid ||
2950 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2951 device = __find_device(&cur_devices->devices,
2952 devid, uuid);
2953 if (device)
2954 return device;
2955 }
2956 cur_devices = cur_devices->seed;
2957 }
2958 return NULL;
0b86a832
CM		 2959}
2960
dfe25020
CM		 2961static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2962 u64 devid, u8 *dev_uuid)
2963{
2964 struct btrfs_device *device;
2965 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2966
2967 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 2968 if (!device)
2969 return NULL;
dfe25020
CM		 2970 list_add(&device->dev_list,
2971 &fs_devices->devices);
dfe25020
CM		 2972 device->barriers = 1;
2973 device->dev_root = root->fs_info->dev_root;
2974 device->devid = devid;
8b712842 2975 device->work.func = pending_bios_fn;
e4404d6e 2976 device->fs_devices = fs_devices;
dfe25020
CM		 2977 fs_devices->num_devices++;
2978 spin_lock_init(&device->io_lock);
d20f7043 2979 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 2980 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2981 return device;
2982}
2983
0b86a832
CM		 2984static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2985 struct extent_buffer *leaf,
2986 struct btrfs_chunk *chunk)
2987{
2988 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2989 struct map_lookup *map;
2990 struct extent_map *em;
2991 u64 logical;
2992 u64 length;
2993 u64 devid;
a443755f 2994 u8 uuid[BTRFS_UUID_SIZE];
593060d7 2995 int num_stripes;
0b86a832 2996 int ret;
593060d7 2997 int i;
0b86a832 2998
e17cade2
CM		 2999 logical = key->offset;
3000 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3001
0b86a832
CM		 3002 spin_lock(&map_tree->map_tree.lock);
3003 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
b248a415 3004 spin_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3005
3006 /* already mapped? */
3007 if (em && em->start <= logical && em->start + em->len > logical) {
3008 free_extent_map(em);
0b86a832
CM		 3009 return 0;
3010 } else if (em) {
3011 free_extent_map(em);
3012 }
0b86a832 3013
0b86a832
CM		 3014 em = alloc_extent_map(GFP_NOFS);
3015 if (!em)
3016 return -ENOMEM;
593060d7
CM		 3017 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3018 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3019 if (!map) {
3020 free_extent_map(em);
3021 return -ENOMEM;
3022 }
3023
3024 em->bdev = (struct block_device *)map;
3025 em->start = logical;
3026 em->len = length;
3027 em->block_start = 0;
c8b97818 3028 em->block_len = em->len;
0b86a832 3029
593060d7
CM		 3030 map->num_stripes = num_stripes;
3031 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3032 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3033 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3034 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3035 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3036 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3037 for (i = 0; i < num_stripes; i++) {
3038 map->stripes[i].physical =
3039 btrfs_stripe_offset_nr(leaf, chunk, i);
3040 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3041 read_extent_buffer(leaf, uuid, (unsigned long)
3042 btrfs_stripe_dev_uuid_nr(chunk, i),
3043 BTRFS_UUID_SIZE);
2b82032c
YZ		 3044 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3045 NULL);
dfe25020 3046 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3047 kfree(map);
3048 free_extent_map(em);
3049 return -EIO;
3050 }
dfe25020
CM		 3051 if (!map->stripes[i].dev) {
3052 map->stripes[i].dev =
3053 add_missing_dev(root, devid, uuid);
3054 if (!map->stripes[i].dev) {
3055 kfree(map);
3056 free_extent_map(em);
3057 return -EIO;
3058 }
3059 }
3060 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3061 }
3062
3063 spin_lock(&map_tree->map_tree.lock);
3064 ret = add_extent_mapping(&map_tree->map_tree, em);
0b86a832 3065 spin_unlock(&map_tree->map_tree.lock);
b248a415 3066 BUG_ON(ret);
0b86a832
CM		 3067 free_extent_map(em);
3068
3069 return 0;
3070}
3071
3072static int fill_device_from_item(struct extent_buffer *leaf,
3073 struct btrfs_dev_item *dev_item,
3074 struct btrfs_device *device)
3075{
3076 unsigned long ptr;
0b86a832
CM		 3077
3078 device->devid = btrfs_device_id(leaf, dev_item);
3079 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3080 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3081 device->type = btrfs_device_type(leaf, dev_item);
3082 device->io_align = btrfs_device_io_align(leaf, dev_item);
3083 device->io_width = btrfs_device_io_width(leaf, dev_item);
3084 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3085
3086 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3087 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3088
0b86a832
CM		 3089 return 0;
3090}
3091
2b82032c
YZ		 3092static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3093{
3094 struct btrfs_fs_devices *fs_devices;
3095 int ret;
3096
3097 mutex_lock(&uuid_mutex);
3098
3099 fs_devices = root->fs_info->fs_devices->seed;
3100 while (fs_devices) {
3101 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3102 ret = 0;
3103 goto out;
3104 }
3105 fs_devices = fs_devices->seed;
3106 }
3107
3108 fs_devices = find_fsid(fsid);
3109 if (!fs_devices) {
3110 ret = -ENOENT;
3111 goto out;
3112 }
e4404d6e
YZ		 3113
3114 fs_devices = clone_fs_devices(fs_devices);
3115 if (IS_ERR(fs_devices)) {
3116 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3117 goto out;
3118 }
3119
97288f2c 3120 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3121 root->fs_info->bdev_holder);
2b82032c
YZ		 3122 if (ret)
3123 goto out;
3124
3125 if (!fs_devices->seeding) {
3126 __btrfs_close_devices(fs_devices);
e4404d6e 3127 free_fs_devices(fs_devices);
2b82032c
YZ		 3128 ret = -EINVAL;
3129 goto out;
3130 }
3131
3132 fs_devices->seed = root->fs_info->fs_devices->seed;
3133 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3134out:
3135 mutex_unlock(&uuid_mutex);
3136 return ret;
3137}
3138
0d81ba5d 3139static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3140 struct extent_buffer *leaf,
3141 struct btrfs_dev_item *dev_item)
3142{
3143 struct btrfs_device *device;
3144 u64 devid;
3145 int ret;
2b82032c 3146 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3147 u8 dev_uuid[BTRFS_UUID_SIZE];
3148
0b86a832 3149 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3150 read_extent_buffer(leaf, dev_uuid,
3151 (unsigned long)btrfs_device_uuid(dev_item),
3152 BTRFS_UUID_SIZE);
2b82032c
YZ		 3153 read_extent_buffer(leaf, fs_uuid,
3154 (unsigned long)btrfs_device_fsid(dev_item),
3155 BTRFS_UUID_SIZE);
3156
3157 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3158 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3159 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3160 return ret;
2b82032c
YZ		 3161 }
3162
3163 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3164 if (!device || !device->bdev) {
e4404d6e 3165 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3166 return -EIO;
3167
3168 if (!device) {
d397712b
CM		 3169 printk(KERN_WARNING "warning devid %llu missing\n",
3170 (unsigned long long)devid);
2b82032c
YZ		 3171 device = add_missing_dev(root, devid, dev_uuid);
3172 if (!device)
3173 return -ENOMEM;
3174 }
3175 }
3176
3177 if (device->fs_devices != root->fs_info->fs_devices) {
3178 BUG_ON(device->writeable);
3179 if (device->generation !=
3180 btrfs_device_generation(leaf, dev_item))
3181 return -EINVAL;
6324fbf3 3182 }
0b86a832
CM		 3183
3184 fill_device_from_item(leaf, dev_item, device);
3185 device->dev_root = root->fs_info->dev_root;
dfe25020 3186 device->in_fs_metadata = 1;
2b82032c
YZ		 3187 if (device->writeable)
3188 device->fs_devices->total_rw_bytes += device->total_bytes;
0b86a832 3189 ret = 0;
0b86a832
CM		 3190 return ret;
3191}
3192
0d81ba5d
CM		 3193int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3194{
3195 struct btrfs_dev_item *dev_item;
3196
3197 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3198 dev_item);
3199 return read_one_dev(root, buf, dev_item);
3200}
3201
e4404d6e 3202int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832
CM		 3203{
3204 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
a061fc8d 3205 struct extent_buffer *sb;
0b86a832 3206 struct btrfs_disk_key *disk_key;
0b86a832 3207 struct btrfs_chunk *chunk;
84eed90f
CM		 3208 u8 *ptr;
3209 unsigned long sb_ptr;
3210 int ret = 0;
0b86a832
CM		 3211 u32 num_stripes;
3212 u32 array_size;
3213 u32 len = 0;
0b86a832 3214 u32 cur;
84eed90f 3215 struct btrfs_key key;
0b86a832 3216
e4404d6e 3217 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3218 BTRFS_SUPER_INFO_SIZE);
3219 if (!sb)
3220 return -ENOMEM;
3221 btrfs_set_buffer_uptodate(sb);
4008c04a
CM		 3222 btrfs_set_buffer_lockdep_class(sb, 0);
3223
a061fc8d 3224 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3225 array_size = btrfs_super_sys_array_size(super_copy);
3226
0b86a832
CM		 3227 ptr = super_copy->sys_chunk_array;
3228 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3229 cur = 0;
3230
3231 while (cur < array_size) {
3232 disk_key = (struct btrfs_disk_key *)ptr;
3233 btrfs_disk_key_to_cpu(&key, disk_key);
3234
a061fc8d 3235 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3236 sb_ptr += len;
3237 cur += len;
3238
0d81ba5d 3239 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3240 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3241 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3242 if (ret)
3243 break;
0b86a832
CM		 3244 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3245 len = btrfs_chunk_item_size(num_stripes);
3246 } else {
84eed90f
CM		 3247 ret = -EIO;
3248 break;
0b86a832
CM		 3249 }
3250 ptr += len;
3251 sb_ptr += len;
3252 cur += len;
3253 }
a061fc8d 3254 free_extent_buffer(sb);
84eed90f 3255 return ret;
0b86a832
CM		 3256}
3257
3258int btrfs_read_chunk_tree(struct btrfs_root *root)
3259{
3260 struct btrfs_path *path;
3261 struct extent_buffer *leaf;
3262 struct btrfs_key key;
3263 struct btrfs_key found_key;
3264 int ret;
3265 int slot;
3266
3267 root = root->fs_info->chunk_root;
3268
3269 path = btrfs_alloc_path();
3270 if (!path)
3271 return -ENOMEM;
3272
3273 /* first we search for all of the device items, and then we
3274 * read in all of the chunk items. This way we can create chunk
3275 * mappings that reference all of the devices that are afound
3276 */
3277 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3278 key.offset = 0;
3279 key.type = 0;
3280again:
3281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
d397712b 3282 while (1) {
0b86a832
CM		 3283 leaf = path->nodes[0];
3284 slot = path->slots[0];
3285 if (slot >= btrfs_header_nritems(leaf)) {
3286 ret = btrfs_next_leaf(root, path);
3287 if (ret == 0)
3288 continue;
3289 if (ret < 0)
3290 goto error;
3291 break;
3292 }
3293 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3294 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3295 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3296 break;
3297 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3298 struct btrfs_dev_item *dev_item;
3299 dev_item = btrfs_item_ptr(leaf, slot,
3300 struct btrfs_dev_item);
0d81ba5d 3301 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3302 if (ret)
3303 goto error;
0b86a832
CM		 3304 }
3305 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3306 struct btrfs_chunk *chunk;
3307 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3308 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3309 if (ret)
3310 goto error;
0b86a832
CM		 3311 }
3312 path->slots[0]++;
3313 }
3314 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3315 key.objectid = 0;
3316 btrfs_release_path(root, path);
3317 goto again;
3318 }
0b86a832
CM		 3319 ret = 0;
3320error:
2b82032c 3321 btrfs_free_path(path);
0b86a832
CM		 3322 return ret;
3323}
Linux 6.x block layer and io_uring tree(s)