projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: fix possible deadlock by clearing __GFP_FS flag
[linux-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>
5a0e3ad6 20#include <linux/slab.h>
8a4b83cc 21#include <linux/buffer_head.h>
f2d8d74d 22#include <linux/blkdev.h>
788f20eb 23#include <linux/random.h>
b765ead5 24#include <linux/iocontext.h>
6f88a440 25#include <linux/capability.h>
593060d7 26#include <asm/div64.h>
4b4e25f2 27#include "compat.h"
0b86a832
CM		 28#include "ctree.h"
29#include "extent_map.h"
30#include "disk-io.h"
31#include "transaction.h"
32#include "print-tree.h"
33#include "volumes.h"
8b712842 34#include "async-thread.h"
0b86a832 35
2b82032c
YZ		 36static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
40
593060d7 41#define map_lookup_size(n) (sizeof(struct map_lookup) + \
cea9e445 42 (sizeof(struct btrfs_bio_stripe) * (n)))
593060d7 43
8a4b83cc
CM		 44static DEFINE_MUTEX(uuid_mutex);
45static LIST_HEAD(fs_uuids);
46
a061fc8d
CM		 47void btrfs_lock_volumes(void)
48{
49 mutex_lock(&uuid_mutex);
50}
51
52void btrfs_unlock_volumes(void)
53{
54 mutex_unlock(&uuid_mutex);
55}
56
7d9eb12c
CM		 57static void lock_chunks(struct btrfs_root *root)
58{
7d9eb12c
CM		 59 mutex_lock(&root->fs_info->chunk_mutex);
60}
61
62static void unlock_chunks(struct btrfs_root *root)
63{
7d9eb12c
CM		 64 mutex_unlock(&root->fs_info->chunk_mutex);
65}
66
e4404d6e
YZ		 67static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
68{
69 struct btrfs_device *device;
70 WARN_ON(fs_devices->opened);
71 while (!list_empty(&fs_devices->devices)) {
72 device = list_entry(fs_devices->devices.next,
73 struct btrfs_device, dev_list);
74 list_del(&device->dev_list);
75 kfree(device->name);
76 kfree(device);
77 }
78 kfree(fs_devices);
79}
80
8a4b83cc
CM		 81int btrfs_cleanup_fs_uuids(void)
82{
83 struct btrfs_fs_devices *fs_devices;
8a4b83cc 84
2b82032c
YZ		 85 while (!list_empty(&fs_uuids)) {
86 fs_devices = list_entry(fs_uuids.next,
87 struct btrfs_fs_devices, list);
88 list_del(&fs_devices->list);
e4404d6e 89 free_fs_devices(fs_devices);
8a4b83cc
CM		 90 }
91 return 0;
92}
93
a1b32a59
CM		 94static noinline struct btrfs_device *__find_device(struct list_head *head,
95 u64 devid, u8 *uuid)
8a4b83cc
CM		 96{
97 struct btrfs_device *dev;
8a4b83cc 98
c6e30871 99 list_for_each_entry(dev, head, dev_list) {
a443755f 100 if (dev->devid == devid &&
8f18cf13 101 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 102 return dev;
a443755f 103 }
8a4b83cc
CM		 104 }
105 return NULL;
106}
107
a1b32a59 108static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 109{
8a4b83cc
CM		 110 struct btrfs_fs_devices *fs_devices;
111
c6e30871 112 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 113 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
114 return fs_devices;
115 }
116 return NULL;
117}
118
ffbd517d
CM		 119static void requeue_list(struct btrfs_pending_bios *pending_bios,
120 struct bio *head, struct bio *tail)
121{
122
123 struct bio *old_head;
124
125 old_head = pending_bios->head;
126 pending_bios->head = head;
127 if (pending_bios->tail)
128 tail->bi_next = old_head;
129 else
130 pending_bios->tail = tail;
131}
132
8b712842
CM		 133/*
134 * we try to collect pending bios for a device so we don't get a large
135 * number of procs sending bios down to the same device. This greatly
136 * improves the schedulers ability to collect and merge the bios.
137 *
138 * But, it also turns into a long list of bios to process and that is sure
139 * to eventually make the worker thread block. The solution here is to
140 * make some progress and then put this work struct back at the end of
141 * the list if the block device is congested. This way, multiple devices
142 * can make progress from a single worker thread.
143 */
d397712b 144static noinline int run_scheduled_bios(struct btrfs_device *device)
8b712842
CM		 145{
146 struct bio *pending;
147 struct backing_dev_info *bdi;
b64a2851 148 struct btrfs_fs_info *fs_info;
ffbd517d 149 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 150 struct bio *tail;
151 struct bio *cur;
152 int again = 0;
ffbd517d
CM		 153 unsigned long num_run;
154 unsigned long num_sync_run;
d644d8a1 155 unsigned long batch_run = 0;
b64a2851 156 unsigned long limit;
b765ead5 157 unsigned long last_waited = 0;
d84275c9 158 int force_reg = 0;
8b712842 159
bedf762b 160 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 161 fs_info = device->dev_root->fs_info;
162 limit = btrfs_async_submit_limit(fs_info);
163 limit = limit * 2 / 3;
164
ffbd517d
CM		 165 /* we want to make sure that every time we switch from the sync
166 * list to the normal list, we unplug
167 */
168 num_sync_run = 0;
169
8b712842
CM		 170loop:
171 spin_lock(&device->io_lock);
172
a6837051 173loop_lock:
d84275c9 174 num_run = 0;
ffbd517d 175
8b712842
CM		 176 /* take all the bios off the list at once and process them
177 * later on (without the lock held). But, remember the
178 * tail and other pointers so the bios can be properly reinserted
179 * into the list if we hit congestion
180 */
d84275c9 181 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 182 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 183 force_reg = 1;
184 } else {
ffbd517d 185 pending_bios = &device->pending_bios;
d84275c9
CM		 186 force_reg = 0;
187 }
ffbd517d
CM		 188
189 pending = pending_bios->head;
190 tail = pending_bios->tail;
8b712842 191 WARN_ON(pending && !tail);
8b712842
CM		 192
193 /*
194 * if pending was null this time around, no bios need processing
195 * at all and we can stop. Otherwise it'll loop back up again
196 * and do an additional check so no bios are missed.
197 *
198 * device->running_pending is used to synchronize with the
199 * schedule_bio code.
200 */
ffbd517d
CM		 201 if (device->pending_sync_bios.head == NULL &&
202 device->pending_bios.head == NULL) {
8b712842
CM		 203 again = 0;
204 device->running_pending = 0;
ffbd517d
CM		 205 } else {
206 again = 1;
207 device->running_pending = 1;
8b712842 208 }
ffbd517d
CM		 209
210 pending_bios->head = NULL;
211 pending_bios->tail = NULL;
212
8b712842
CM		 213 spin_unlock(&device->io_lock);
214
ffbd517d
CM		 215 /*
216 * if we're doing the regular priority list, make sure we unplug
217 * for any high prio bios we've sent down
218 */
219 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
220 num_sync_run = 0;
221 blk_run_backing_dev(bdi, NULL);
222 }
223
d397712b 224 while (pending) {
ffbd517d
CM		 225
226 rmb();
d84275c9
CM		 227 /* we want to work on both lists, but do more bios on the
228 * sync list than the regular list
229 */
230 if ((num_run > 32 &&
231 pending_bios != &device->pending_sync_bios &&
232 device->pending_sync_bios.head) ||
233 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
234 device->pending_bios.head)) {
ffbd517d
CM		 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);
d644d8a1 250
7b6d91da 251 if (cur->bi_rw & REQ_SYNC)
ffbd517d
CM		 252 num_sync_run++;
253
5ff7ba3a
CM		 254 submit_bio(cur->bi_rw, cur);
255 num_run++;
256 batch_run++;
ffbd517d
CM		 257 if (need_resched()) {
258 if (num_sync_run) {
259 blk_run_backing_dev(bdi, NULL);
260 num_sync_run = 0;
261 }
262 cond_resched();
263 }
8b712842
CM		 264
265 /*
266 * we made progress, there is more work to do and the bdi
267 * is now congested. Back off and let other work structs
268 * run instead
269 */
57fd5a5f 270 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 271 fs_info->fs_devices->open_devices > 1) {
b765ead5 272 struct io_context *ioc;
8b712842 273
b765ead5
CM		 274 ioc = current->io_context;
275
276 /*
277 * the main goal here is that we don't want to
278 * block if we're going to be able to submit
279 * more requests without blocking.
280 *
281 * This code does two great things, it pokes into
282 * the elevator code from a filesystem _and_
283 * it makes assumptions about how batching works.
284 */
285 if (ioc && ioc->nr_batch_requests > 0 &&
286 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
287 (last_waited == 0 ||
288 ioc->last_waited == last_waited)) {
289 /*
290 * we want to go through our batch of
291 * requests and stop. So, we copy out
292 * the ioc->last_waited time and test
293 * against it before looping
294 */
295 last_waited = ioc->last_waited;
ffbd517d
CM		 296 if (need_resched()) {
297 if (num_sync_run) {
298 blk_run_backing_dev(bdi, NULL);
299 num_sync_run = 0;
300 }
301 cond_resched();
302 }
b765ead5
CM		 303 continue;
304 }
8b712842 305 spin_lock(&device->io_lock);
ffbd517d 306 requeue_list(pending_bios, pending, tail);
a6837051 307 device->running_pending = 1;
8b712842
CM		 308
309 spin_unlock(&device->io_lock);
310 btrfs_requeue_work(&device->work);
311 goto done;
312 }
313 }
ffbd517d
CM		 314
315 if (num_sync_run) {
316 num_sync_run = 0;
317 blk_run_backing_dev(bdi, NULL);
318 }
bedf762b
CM		 319 /*
320 * IO has already been through a long path to get here. Checksumming,
321 * async helper threads, perhaps compression. We've done a pretty
322 * good job of collecting a batch of IO and should just unplug
323 * the device right away.
324 *
325 * This will help anyone who is waiting on the IO, they might have
326 * already unplugged, but managed to do so before the bio they
327 * cared about found its way down here.
328 */
329 blk_run_backing_dev(bdi, NULL);
51684082
CM		 330
331 cond_resched();
332 if (again)
333 goto loop;
334
335 spin_lock(&device->io_lock);
336 if (device->pending_bios.head || device->pending_sync_bios.head)
337 goto loop_lock;
338 spin_unlock(&device->io_lock);
339
8b712842
CM		 340done:
341 return 0;
342}
343
b2950863 344static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 345{
346 struct btrfs_device *device;
347
348 device = container_of(work, struct btrfs_device, work);
349 run_scheduled_bios(device);
350}
351
a1b32a59 352static noinline int device_list_add(const char *path,
8a4b83cc
CM		 353 struct btrfs_super_block *disk_super,
354 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
355{
356 struct btrfs_device *device;
357 struct btrfs_fs_devices *fs_devices;
358 u64 found_transid = btrfs_super_generation(disk_super);
3a0524dc 359 char *name;
8a4b83cc
CM		 360
361 fs_devices = find_fsid(disk_super->fsid);
362 if (!fs_devices) {
515dc322 363 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 364 if (!fs_devices)
365 return -ENOMEM;
366 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 367 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 368 list_add(&fs_devices->list, &fs_uuids);
369 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
370 fs_devices->latest_devid = devid;
371 fs_devices->latest_trans = found_transid;
e5e9a520 372 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 373 device = NULL;
374 } else {
a443755f
CM		 375 device = __find_device(&fs_devices->devices, devid,
376 disk_super->dev_item.uuid);
8a4b83cc
CM		 377 }
378 if (!device) {
2b82032c
YZ		 379 if (fs_devices->opened)
380 return -EBUSY;
381
8a4b83cc
CM		 382 device = kzalloc(sizeof(*device), GFP_NOFS);
383 if (!device) {
384 /* we can safely leave the fs_devices entry around */
385 return -ENOMEM;
386 }
387 device->devid = devid;
8b712842 388 device->work.func = pending_bios_fn;
a443755f
CM		 389 memcpy(device->uuid, disk_super->dev_item.uuid,
390 BTRFS_UUID_SIZE);
b248a415 391 spin_lock_init(&device->io_lock);
8a4b83cc
CM		 392 device->name = kstrdup(path, GFP_NOFS);
393 if (!device->name) {
394 kfree(device);
395 return -ENOMEM;
396 }
2b82032c 397 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520
CM		 398
399 mutex_lock(&fs_devices->device_list_mutex);
8a4b83cc 400 list_add(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 401 mutex_unlock(&fs_devices->device_list_mutex);
402
2b82032c 403 device->fs_devices = fs_devices;
8a4b83cc 404 fs_devices->num_devices++;
cd02dca5 405 } else if (!device->name || strcmp(device->name, path)) {
3a0524dc
TH		 406 name = kstrdup(path, GFP_NOFS);
407 if (!name)
408 return -ENOMEM;
409 kfree(device->name);
410 device->name = name;
cd02dca5
CM		 411 if (device->missing) {
412 fs_devices->missing_devices--;
413 device->missing = 0;
414 }
8a4b83cc
CM		 415 }
416
417 if (found_transid > fs_devices->latest_trans) {
418 fs_devices->latest_devid = devid;
419 fs_devices->latest_trans = found_transid;
420 }
8a4b83cc
CM		 421 *fs_devices_ret = fs_devices;
422 return 0;
423}
424
e4404d6e
YZ		 425static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
426{
427 struct btrfs_fs_devices *fs_devices;
428 struct btrfs_device *device;
429 struct btrfs_device *orig_dev;
430
431 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
432 if (!fs_devices)
433 return ERR_PTR(-ENOMEM);
434
435 INIT_LIST_HEAD(&fs_devices->devices);
436 INIT_LIST_HEAD(&fs_devices->alloc_list);
437 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 438 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 439 fs_devices->latest_devid = orig->latest_devid;
440 fs_devices->latest_trans = orig->latest_trans;
441 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
442
e5e9a520 443 mutex_lock(&orig->device_list_mutex);
e4404d6e
YZ		 444 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
445 device = kzalloc(sizeof(*device), GFP_NOFS);
446 if (!device)
447 goto error;
448
449 device->name = kstrdup(orig_dev->name, GFP_NOFS);
fd2696f3
JL		 450 if (!device->name) {
451 kfree(device);
e4404d6e 452 goto error;
fd2696f3 453 }
e4404d6e
YZ		 454
455 device->devid = orig_dev->devid;
456 device->work.func = pending_bios_fn;
457 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
e4404d6e
YZ		 458 spin_lock_init(&device->io_lock);
459 INIT_LIST_HEAD(&device->dev_list);
460 INIT_LIST_HEAD(&device->dev_alloc_list);
461
462 list_add(&device->dev_list, &fs_devices->devices);
463 device->fs_devices = fs_devices;
464 fs_devices->num_devices++;
465 }
e5e9a520 466 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 467 return fs_devices;
468error:
e5e9a520 469 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 470 free_fs_devices(fs_devices);
471 return ERR_PTR(-ENOMEM);
472}
473
dfe25020
CM		 474int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
475{
c6e30871 476 struct btrfs_device *device, *next;
dfe25020
CM		 477
478 mutex_lock(&uuid_mutex);
479again:
e5e9a520 480 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 481 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
2b82032c
YZ		 482 if (device->in_fs_metadata)
483 continue;
484
485 if (device->bdev) {
d4d77629 486 blkdev_put(device->bdev, device->mode);
2b82032c
YZ		 487 device->bdev = NULL;
488 fs_devices->open_devices--;
489 }
490 if (device->writeable) {
491 list_del_init(&device->dev_alloc_list);
492 device->writeable = 0;
493 fs_devices->rw_devices--;
494 }
e4404d6e
YZ		 495 list_del_init(&device->dev_list);
496 fs_devices->num_devices--;
497 kfree(device->name);
498 kfree(device);
dfe25020 499 }
e5e9a520 500 mutex_unlock(&fs_devices->device_list_mutex);
2b82032c
YZ		 501
502 if (fs_devices->seed) {
503 fs_devices = fs_devices->seed;
2b82032c
YZ		 504 goto again;
505 }
506
dfe25020
CM		 507 mutex_unlock(&uuid_mutex);
508 return 0;
509}
a0af469b 510
2b82032c 511static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 512{
8a4b83cc 513 struct btrfs_device *device;
e4404d6e 514
2b82032c
YZ		 515 if (--fs_devices->opened > 0)
516 return 0;
8a4b83cc 517
c6e30871 518 list_for_each_entry(device, &fs_devices->devices, dev_list) {
8a4b83cc 519 if (device->bdev) {
d4d77629 520 blkdev_put(device->bdev, device->mode);
a0af469b 521 fs_devices->open_devices--;
8a4b83cc 522 }
2b82032c
YZ		 523 if (device->writeable) {
524 list_del_init(&device->dev_alloc_list);
525 fs_devices->rw_devices--;
526 }
527
8a4b83cc 528 device->bdev = NULL;
2b82032c 529 device->writeable = 0;
dfe25020 530 device->in_fs_metadata = 0;
8a4b83cc 531 }
e4404d6e
YZ		 532 WARN_ON(fs_devices->open_devices);
533 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 534 fs_devices->opened = 0;
535 fs_devices->seeding = 0;
2b82032c 536
8a4b83cc
CM		 537 return 0;
538}
539
2b82032c
YZ		 540int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
541{
e4404d6e 542 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 543 int ret;
544
545 mutex_lock(&uuid_mutex);
546 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 547 if (!fs_devices->opened) {
548 seed_devices = fs_devices->seed;
549 fs_devices->seed = NULL;
550 }
2b82032c 551 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 552
553 while (seed_devices) {
554 fs_devices = seed_devices;
555 seed_devices = fs_devices->seed;
556 __btrfs_close_devices(fs_devices);
557 free_fs_devices(fs_devices);
558 }
2b82032c
YZ		 559 return ret;
560}
561
e4404d6e
YZ		 562static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563 fmode_t flags, void *holder)
8a4b83cc
CM		 564{
565 struct block_device *bdev;
566 struct list_head *head = &fs_devices->devices;
8a4b83cc 567 struct btrfs_device *device;
a0af469b
CM		 568 struct block_device *latest_bdev = NULL;
569 struct buffer_head *bh;
570 struct btrfs_super_block *disk_super;
571 u64 latest_devid = 0;
572 u64 latest_transid = 0;
a0af469b 573 u64 devid;
2b82032c 574 int seeding = 1;
a0af469b 575 int ret = 0;
8a4b83cc 576
d4d77629
TH		 577 flags |= FMODE_EXCL;
578
c6e30871 579 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 580 if (device->bdev)
581 continue;
dfe25020
CM		 582 if (!device->name)
583 continue;
584
d4d77629 585 bdev = blkdev_get_by_path(device->name, flags, holder);
8a4b83cc 586 if (IS_ERR(bdev)) {
d397712b 587 printk(KERN_INFO "open %s failed\n", device->name);
a0af469b 588 goto error;
8a4b83cc 589 }
a061fc8d 590 set_blocksize(bdev, 4096);
a0af469b 591
a512bbf8 592 bh = btrfs_read_dev_super(bdev);
20b45077
DY		 593 if (!bh) {
594 ret = -EINVAL;
a0af469b 595 goto error_close;
20b45077 596 }
a0af469b
CM		 597
598 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 599 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 600 if (devid != device->devid)
601 goto error_brelse;
602
2b82032c
YZ		 603 if (memcmp(device->uuid, disk_super->dev_item.uuid,
604 BTRFS_UUID_SIZE))
605 goto error_brelse;
606
607 device->generation = btrfs_super_generation(disk_super);
608 if (!latest_transid || device->generation > latest_transid) {
a0af469b 609 latest_devid = devid;
2b82032c 610 latest_transid = device->generation;
a0af469b
CM		 611 latest_bdev = bdev;
612 }
613
2b82032c
YZ		 614 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
615 device->writeable = 0;
616 } else {
617 device->writeable = !bdev_read_only(bdev);
618 seeding = 0;
619 }
620
8a4b83cc 621 device->bdev = bdev;
dfe25020 622 device->in_fs_metadata = 0;
15916de8
CM		 623 device->mode = flags;
624
c289811c
CM		 625 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
626 fs_devices->rotating = 1;
627
a0af469b 628 fs_devices->open_devices++;
2b82032c
YZ		 629 if (device->writeable) {
630 fs_devices->rw_devices++;
631 list_add(&device->dev_alloc_list,
632 &fs_devices->alloc_list);
633 }
a0af469b 634 continue;
a061fc8d 635
a0af469b
CM		 636error_brelse:
637 brelse(bh);
638error_close:
d4d77629 639 blkdev_put(bdev, flags);
a0af469b
CM		 640error:
641 continue;
8a4b83cc 642 }
a0af469b
CM		 643 if (fs_devices->open_devices == 0) {
644 ret = -EIO;
645 goto out;
646 }
2b82032c
YZ		 647 fs_devices->seeding = seeding;
648 fs_devices->opened = 1;
a0af469b
CM		 649 fs_devices->latest_bdev = latest_bdev;
650 fs_devices->latest_devid = latest_devid;
651 fs_devices->latest_trans = latest_transid;
2b82032c 652 fs_devices->total_rw_bytes = 0;
a0af469b 653out:
2b82032c
YZ		 654 return ret;
655}
656
657int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 658 fmode_t flags, void *holder)
2b82032c
YZ		 659{
660 int ret;
661
662 mutex_lock(&uuid_mutex);
663 if (fs_devices->opened) {
e4404d6e
YZ		 664 fs_devices->opened++;
665 ret = 0;
2b82032c 666 } else {
15916de8 667 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 668 }
8a4b83cc 669 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 670 return ret;
671}
672
97288f2c 673int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 674 struct btrfs_fs_devices **fs_devices_ret)
675{
676 struct btrfs_super_block *disk_super;
677 struct block_device *bdev;
678 struct buffer_head *bh;
679 int ret;
680 u64 devid;
f2984462 681 u64 transid;
8a4b83cc
CM		 682
683 mutex_lock(&uuid_mutex);
684
d4d77629
TH		 685 flags |= FMODE_EXCL;
686 bdev = blkdev_get_by_path(path, flags, holder);
8a4b83cc
CM		 687
688 if (IS_ERR(bdev)) {
8a4b83cc
CM		 689 ret = PTR_ERR(bdev);
690 goto error;
691 }
692
693 ret = set_blocksize(bdev, 4096);
694 if (ret)
695 goto error_close;
a512bbf8 696 bh = btrfs_read_dev_super(bdev);
8a4b83cc 697 if (!bh) {
20b45077 698 ret = -EINVAL;
8a4b83cc
CM		 699 goto error_close;
700 }
701 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 702 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 703 transid = btrfs_super_generation(disk_super);
7ae9c09d 704 if (disk_super->label[0])
d397712b 705 printk(KERN_INFO "device label %s ", disk_super->label);
7ae9c09d
CM		 706 else {
707 /* FIXME, make a readl uuid parser */
d397712b 708 printk(KERN_INFO "device fsid %llx-%llx ",
7ae9c09d
CM		 709 *(unsigned long long *)disk_super->fsid,
710 *(unsigned long long *)(disk_super->fsid + 8));
711 }
119e10cf 712 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 713 (unsigned long long)devid, (unsigned long long)transid, path);
8a4b83cc
CM		 714 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
715
8a4b83cc
CM		 716 brelse(bh);
717error_close:
d4d77629 718 blkdev_put(bdev, flags);
8a4b83cc
CM		 719error:
720 mutex_unlock(&uuid_mutex);
721 return ret;
722}
0b86a832 723
6d07bcec
MX		 724/* helper to account the used device space in the range */
725int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
726 u64 end, u64 *length)
727{
728 struct btrfs_key key;
729 struct btrfs_root *root = device->dev_root;
730 struct btrfs_dev_extent *dev_extent;
731 struct btrfs_path *path;
732 u64 extent_end;
733 int ret;
734 int slot;
735 struct extent_buffer *l;
736
737 *length = 0;
738
739 if (start >= device->total_bytes)
740 return 0;
741
742 path = btrfs_alloc_path();
743 if (!path)
744 return -ENOMEM;
745 path->reada = 2;
746
747 key.objectid = device->devid;
748 key.offset = start;
749 key.type = BTRFS_DEV_EXTENT_KEY;
750
751 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
752 if (ret < 0)
753 goto out;
754 if (ret > 0) {
755 ret = btrfs_previous_item(root, path, key.objectid, key.type);
756 if (ret < 0)
757 goto out;
758 }
759
760 while (1) {
761 l = path->nodes[0];
762 slot = path->slots[0];
763 if (slot >= btrfs_header_nritems(l)) {
764 ret = btrfs_next_leaf(root, path);
765 if (ret == 0)
766 continue;
767 if (ret < 0)
768 goto out;
769
770 break;
771 }
772 btrfs_item_key_to_cpu(l, &key, slot);
773
774 if (key.objectid < device->devid)
775 goto next;
776
777 if (key.objectid > device->devid)
778 break;
779
780 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
781 goto next;
782
783 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
784 extent_end = key.offset + btrfs_dev_extent_length(l,
785 dev_extent);
786 if (key.offset <= start && extent_end > end) {
787 *length = end - start + 1;
788 break;
789 } else if (key.offset <= start && extent_end > start)
790 *length += extent_end - start;
791 else if (key.offset > start && extent_end <= end)
792 *length += extent_end - key.offset;
793 else if (key.offset > start && key.offset <= end) {
794 *length += end - key.offset + 1;
795 break;
796 } else if (key.offset > end)
797 break;
798
799next:
800 path->slots[0]++;
801 }
802 ret = 0;
803out:
804 btrfs_free_path(path);
805 return ret;
806}
807
0b86a832 808/*
7bfc837d
MX		 809 * find_free_dev_extent - find free space in the specified device
810 * @trans: transaction handler
811 * @device: the device which we search the free space in
812 * @num_bytes: the size of the free space that we need
813 * @start: store the start of the free space.
814 * @len: the size of the free space. that we find, or the size of the max
815 * free space if we don't find suitable free space
816 *
0b86a832
CM		 817 * this uses a pretty simple search, the expectation is that it is
818 * called very infrequently and that a given device has a small number
819 * of extents
7bfc837d
MX		 820 *
821 * @start is used to store the start of the free space if we find. But if we
822 * don't find suitable free space, it will be used to store the start position
823 * of the max free space.
824 *
825 * @len is used to store the size of the free space that we find.
826 * But if we don't find suitable free space, it is used to store the size of
827 * the max free space.
0b86a832 828 */
ba1bf481
JB		 829int find_free_dev_extent(struct btrfs_trans_handle *trans,
830 struct btrfs_device *device, u64 num_bytes,
7bfc837d 831 u64 *start, u64 *len)
0b86a832
CM		 832{
833 struct btrfs_key key;
834 struct btrfs_root *root = device->dev_root;
7bfc837d 835 struct btrfs_dev_extent *dev_extent;
2b82032c 836 struct btrfs_path *path;
7bfc837d
MX		 837 u64 hole_size;
838 u64 max_hole_start;
839 u64 max_hole_size;
840 u64 extent_end;
841 u64 search_start;
0b86a832
CM		 842 u64 search_end = device->total_bytes;
843 int ret;
7bfc837d 844 int slot;
0b86a832
CM		 845 struct extent_buffer *l;
846
0b86a832
CM		 847 /* FIXME use last free of some kind */
848
8a4b83cc
CM		 849 /* we don't want to overwrite the superblock on the drive,
850 * so we make sure to start at an offset of at least 1MB
851 */
7bfc837d 852 search_start = 1024 * 1024;
8f18cf13 853
7bfc837d 854 if (root->fs_info->alloc_start + num_bytes <= search_end)
8f18cf13
CM		 855 search_start = max(root->fs_info->alloc_start, search_start);
856
7bfc837d
MX		 857 max_hole_start = search_start;
858 max_hole_size = 0;
859
860 if (search_start >= search_end) {
861 ret = -ENOSPC;
862 goto error;
863 }
864
865 path = btrfs_alloc_path();
866 if (!path) {
867 ret = -ENOMEM;
868 goto error;
869 }
870 path->reada = 2;
871
0b86a832
CM		 872 key.objectid = device->devid;
873 key.offset = search_start;
874 key.type = BTRFS_DEV_EXTENT_KEY;
7bfc837d 875
0b86a832
CM		 876 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
877 if (ret < 0)
7bfc837d 878 goto out;
1fcbac58
YZ		 879 if (ret > 0) {
880 ret = btrfs_previous_item(root, path, key.objectid, key.type);
881 if (ret < 0)
7bfc837d 882 goto out;
1fcbac58 883 }
7bfc837d 884
0b86a832
CM		 885 while (1) {
886 l = path->nodes[0];
887 slot = path->slots[0];
888 if (slot >= btrfs_header_nritems(l)) {
889 ret = btrfs_next_leaf(root, path);
890 if (ret == 0)
891 continue;
892 if (ret < 0)
7bfc837d
MX		 893 goto out;
894
895 break;
0b86a832
CM		 896 }
897 btrfs_item_key_to_cpu(l, &key, slot);
898
899 if (key.objectid < device->devid)
900 goto next;
901
902 if (key.objectid > device->devid)
7bfc837d 903 break;
0b86a832 904
7bfc837d
MX		 905 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
906 goto next;
9779b72f 907
7bfc837d
MX		 908 if (key.offset > search_start) {
909 hole_size = key.offset - search_start;
9779b72f 910
7bfc837d
MX		 911 if (hole_size > max_hole_size) {
912 max_hole_start = search_start;
913 max_hole_size = hole_size;
914 }
9779b72f 915
7bfc837d
MX		 916 /*
917 * If this free space is greater than which we need,
918 * it must be the max free space that we have found
919 * until now, so max_hole_start must point to the start
920 * of this free space and the length of this free space
921 * is stored in max_hole_size. Thus, we return
922 * max_hole_start and max_hole_size and go back to the
923 * caller.
924 */
925 if (hole_size >= num_bytes) {
926 ret = 0;
927 goto out;
0b86a832
CM		 928 }
929 }
0b86a832 930
0b86a832 931 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
7bfc837d
MX		 932 extent_end = key.offset + btrfs_dev_extent_length(l,
933 dev_extent);
934 if (extent_end > search_start)
935 search_start = extent_end;
0b86a832
CM		 936next:
937 path->slots[0]++;
938 cond_resched();
939 }
0b86a832 940
7bfc837d
MX		 941 hole_size = search_end- search_start;
942 if (hole_size > max_hole_size) {
943 max_hole_start = search_start;
944 max_hole_size = hole_size;
0b86a832 945 }
0b86a832 946
7bfc837d
MX		 947 /* See above. */
948 if (hole_size < num_bytes)
949 ret = -ENOSPC;
950 else
951 ret = 0;
952
953out:
2b82032c 954 btrfs_free_path(path);
7bfc837d
MX		 955error:
956 *start = max_hole_start;
b2117a39 957 if (len)
7bfc837d 958 *len = max_hole_size;
0b86a832
CM		 959 return ret;
960}
961
b2950863 962static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 963 struct btrfs_device *device,
964 u64 start)
965{
966 int ret;
967 struct btrfs_path *path;
968 struct btrfs_root *root = device->dev_root;
969 struct btrfs_key key;
a061fc8d
CM		 970 struct btrfs_key found_key;
971 struct extent_buffer *leaf = NULL;
972 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 973
974 path = btrfs_alloc_path();
975 if (!path)
976 return -ENOMEM;
977
978 key.objectid = device->devid;
979 key.offset = start;
980 key.type = BTRFS_DEV_EXTENT_KEY;
981
982 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 983 if (ret > 0) {
984 ret = btrfs_previous_item(root, path, key.objectid,
985 BTRFS_DEV_EXTENT_KEY);
986 BUG_ON(ret);
987 leaf = path->nodes[0];
988 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
989 extent = btrfs_item_ptr(leaf, path->slots[0],
990 struct btrfs_dev_extent);
991 BUG_ON(found_key.offset > start || found_key.offset +
992 btrfs_dev_extent_length(leaf, extent) < start);
993 ret = 0;
994 } else if (ret == 0) {
995 leaf = path->nodes[0];
996 extent = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_dev_extent);
998 }
8f18cf13
CM		 999 BUG_ON(ret);
1000
dfe25020
CM		 1001 if (device->bytes_used > 0)
1002 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
8f18cf13
CM		 1003 ret = btrfs_del_item(trans, root, path);
1004 BUG_ON(ret);
1005
1006 btrfs_free_path(path);
1007 return ret;
1008}
1009
2b82032c 1010int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 1011 struct btrfs_device *device,
e17cade2 1012 u64 chunk_tree, u64 chunk_objectid,
2b82032c 1013 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 1014{
1015 int ret;
1016 struct btrfs_path *path;
1017 struct btrfs_root *root = device->dev_root;
1018 struct btrfs_dev_extent *extent;
1019 struct extent_buffer *leaf;
1020 struct btrfs_key key;
1021
dfe25020 1022 WARN_ON(!device->in_fs_metadata);
0b86a832
CM		 1023 path = btrfs_alloc_path();
1024 if (!path)
1025 return -ENOMEM;
1026
0b86a832 1027 key.objectid = device->devid;
2b82032c 1028 key.offset = start;
0b86a832
CM		 1029 key.type = BTRFS_DEV_EXTENT_KEY;
1030 ret = btrfs_insert_empty_item(trans, root, path, &key,
1031 sizeof(*extent));
1032 BUG_ON(ret);
1033
1034 leaf = path->nodes[0];
1035 extent = btrfs_item_ptr(leaf, path->slots[0],
1036 struct btrfs_dev_extent);
e17cade2
CM		 1037 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1038 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1039 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1040
1041 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1042 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1043 BTRFS_UUID_SIZE);
1044
0b86a832
CM		 1045 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1046 btrfs_mark_buffer_dirty(leaf);
0b86a832
CM		 1047 btrfs_free_path(path);
1048 return ret;
1049}
1050
a1b32a59
CM		 1051static noinline int find_next_chunk(struct btrfs_root *root,
1052 u64 objectid, u64 *offset)
0b86a832
CM		 1053{
1054 struct btrfs_path *path;
1055 int ret;
1056 struct btrfs_key key;
e17cade2 1057 struct btrfs_chunk *chunk;
0b86a832
CM		 1058 struct btrfs_key found_key;
1059
1060 path = btrfs_alloc_path();
1061 BUG_ON(!path);
1062
e17cade2 1063 key.objectid = objectid;
0b86a832
CM		 1064 key.offset = (u64)-1;
1065 key.type = BTRFS_CHUNK_ITEM_KEY;
1066
1067 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1068 if (ret < 0)
1069 goto error;
1070
1071 BUG_ON(ret == 0);
1072
1073 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1074 if (ret) {
e17cade2 1075 *offset = 0;
0b86a832
CM		 1076 } else {
1077 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1078 path->slots[0]);
e17cade2
CM		 1079 if (found_key.objectid != objectid)
1080 *offset = 0;
1081 else {
1082 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1083 struct btrfs_chunk);
1084 *offset = found_key.offset +
1085 btrfs_chunk_length(path->nodes[0], chunk);
1086 }
0b86a832
CM		 1087 }
1088 ret = 0;
1089error:
1090 btrfs_free_path(path);
1091 return ret;
1092}
1093
2b82032c 1094static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 1095{
1096 int ret;
1097 struct btrfs_key key;
1098 struct btrfs_key found_key;
2b82032c
YZ		 1099 struct btrfs_path *path;
1100
1101 root = root->fs_info->chunk_root;
1102
1103 path = btrfs_alloc_path();
1104 if (!path)
1105 return -ENOMEM;
0b86a832
CM		 1106
1107 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1108 key.type = BTRFS_DEV_ITEM_KEY;
1109 key.offset = (u64)-1;
1110
1111 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1112 if (ret < 0)
1113 goto error;
1114
1115 BUG_ON(ret == 0);
1116
1117 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1118 BTRFS_DEV_ITEM_KEY);
1119 if (ret) {
1120 *objectid = 1;
1121 } else {
1122 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1123 path->slots[0]);
1124 *objectid = found_key.offset + 1;
1125 }
1126 ret = 0;
1127error:
2b82032c 1128 btrfs_free_path(path);
0b86a832
CM		 1129 return ret;
1130}
1131
1132/*
1133 * the device information is stored in the chunk root
1134 * the btrfs_device struct should be fully filled in
1135 */
1136int btrfs_add_device(struct btrfs_trans_handle *trans,
1137 struct btrfs_root *root,
1138 struct btrfs_device *device)
1139{
1140 int ret;
1141 struct btrfs_path *path;
1142 struct btrfs_dev_item *dev_item;
1143 struct extent_buffer *leaf;
1144 struct btrfs_key key;
1145 unsigned long ptr;
0b86a832
CM		 1146
1147 root = root->fs_info->chunk_root;
1148
1149 path = btrfs_alloc_path();
1150 if (!path)
1151 return -ENOMEM;
1152
0b86a832
CM		 1153 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1154 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1155 key.offset = device->devid;
0b86a832
CM		 1156
1157 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1158 sizeof(*dev_item));
0b86a832
CM		 1159 if (ret)
1160 goto out;
1161
1162 leaf = path->nodes[0];
1163 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1164
1165 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1166 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1167 btrfs_set_device_type(leaf, dev_item, device->type);
1168 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1169 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1170 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1171 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1172 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1173 btrfs_set_device_group(leaf, dev_item, 0);
1174 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1175 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1176 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1177
0b86a832 1178 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1179 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1180 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1181 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1182 btrfs_mark_buffer_dirty(leaf);
0b86a832 1183
2b82032c 1184 ret = 0;
0b86a832
CM		 1185out:
1186 btrfs_free_path(path);
1187 return ret;
1188}
8f18cf13 1189
a061fc8d
CM		 1190static int btrfs_rm_dev_item(struct btrfs_root *root,
1191 struct btrfs_device *device)
1192{
1193 int ret;
1194 struct btrfs_path *path;
a061fc8d 1195 struct btrfs_key key;
a061fc8d
CM		 1196 struct btrfs_trans_handle *trans;
1197
1198 root = root->fs_info->chunk_root;
1199
1200 path = btrfs_alloc_path();
1201 if (!path)
1202 return -ENOMEM;
1203
a22285a6 1204 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 1205 if (IS_ERR(trans)) {
1206 btrfs_free_path(path);
1207 return PTR_ERR(trans);
1208 }
a061fc8d
CM		 1209 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1210 key.type = BTRFS_DEV_ITEM_KEY;
1211 key.offset = device->devid;
7d9eb12c 1212 lock_chunks(root);
a061fc8d
CM		 1213
1214 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1215 if (ret < 0)
1216 goto out;
1217
1218 if (ret > 0) {
1219 ret = -ENOENT;
1220 goto out;
1221 }
1222
1223 ret = btrfs_del_item(trans, root, path);
1224 if (ret)
1225 goto out;
a061fc8d
CM		 1226out:
1227 btrfs_free_path(path);
7d9eb12c 1228 unlock_chunks(root);
a061fc8d
CM		 1229 btrfs_commit_transaction(trans, root);
1230 return ret;
1231}
1232
1233int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1234{
1235 struct btrfs_device *device;
2b82032c 1236 struct btrfs_device *next_device;
a061fc8d 1237 struct block_device *bdev;
dfe25020 1238 struct buffer_head *bh = NULL;
a061fc8d
CM		 1239 struct btrfs_super_block *disk_super;
1240 u64 all_avail;
1241 u64 devid;
2b82032c
YZ		 1242 u64 num_devices;
1243 u8 *dev_uuid;
a061fc8d
CM		 1244 int ret = 0;
1245
a061fc8d 1246 mutex_lock(&uuid_mutex);
7d9eb12c 1247 mutex_lock(&root->fs_info->volume_mutex);
a061fc8d
CM		 1248
1249 all_avail = root->fs_info->avail_data_alloc_bits |
1250 root->fs_info->avail_system_alloc_bits |
1251 root->fs_info->avail_metadata_alloc_bits;
1252
1253 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
035fe03a 1254 root->fs_info->fs_devices->num_devices <= 4) {
d397712b
CM		 1255 printk(KERN_ERR "btrfs: unable to go below four devices "
1256 "on raid10\n");
a061fc8d
CM		 1257 ret = -EINVAL;
1258 goto out;
1259 }
1260
1261 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
035fe03a 1262 root->fs_info->fs_devices->num_devices <= 2) {
d397712b
CM		 1263 printk(KERN_ERR "btrfs: unable to go below two "
1264 "devices on raid1\n");
a061fc8d
CM		 1265 ret = -EINVAL;
1266 goto out;
1267 }
1268
dfe25020 1269 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1270 struct list_head *devices;
1271 struct btrfs_device *tmp;
a061fc8d 1272
dfe25020
CM		 1273 device = NULL;
1274 devices = &root->fs_info->fs_devices->devices;
e5e9a520 1275 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
c6e30871 1276 list_for_each_entry(tmp, devices, dev_list) {
dfe25020
CM		 1277 if (tmp->in_fs_metadata && !tmp->bdev) {
1278 device = tmp;
1279 break;
1280 }
1281 }
e5e9a520 1282 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
dfe25020
CM		 1283 bdev = NULL;
1284 bh = NULL;
1285 disk_super = NULL;
1286 if (!device) {
d397712b
CM		 1287 printk(KERN_ERR "btrfs: no missing devices found to "
1288 "remove\n");
dfe25020
CM		 1289 goto out;
1290 }
dfe25020 1291 } else {
d4d77629
TH		 1292 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1293 root->fs_info->bdev_holder);
dfe25020
CM		 1294 if (IS_ERR(bdev)) {
1295 ret = PTR_ERR(bdev);
1296 goto out;
1297 }
a061fc8d 1298
2b82032c 1299 set_blocksize(bdev, 4096);
a512bbf8 1300 bh = btrfs_read_dev_super(bdev);
dfe25020 1301 if (!bh) {
20b45077 1302 ret = -EINVAL;
dfe25020
CM		 1303 goto error_close;
1304 }
1305 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 1306 devid = btrfs_stack_device_id(&disk_super->dev_item);
2b82032c
YZ		 1307 dev_uuid = disk_super->dev_item.uuid;
1308 device = btrfs_find_device(root, devid, dev_uuid,
1309 disk_super->fsid);
dfe25020
CM		 1310 if (!device) {
1311 ret = -ENOENT;
1312 goto error_brelse;
1313 }
2b82032c 1314 }
dfe25020 1315
2b82032c 1316 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1317 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1318 "device\n");
2b82032c
YZ		 1319 ret = -EINVAL;
1320 goto error_brelse;
1321 }
1322
1323 if (device->writeable) {
1324 list_del_init(&device->dev_alloc_list);
1325 root->fs_info->fs_devices->rw_devices--;
dfe25020 1326 }
a061fc8d
CM		 1327
1328 ret = btrfs_shrink_device(device, 0);
1329 if (ret)
9b3517e9 1330 goto error_undo;
a061fc8d 1331
a061fc8d
CM		 1332 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1333 if (ret)
9b3517e9 1334 goto error_undo;
a061fc8d 1335
2b82032c 1336 device->in_fs_metadata = 0;
e5e9a520
CM		 1337
1338 /*
1339 * the device list mutex makes sure that we don't change
1340 * the device list while someone else is writing out all
1341 * the device supers.
1342 */
1343 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1344 list_del_init(&device->dev_list);
e5e9a520
CM		 1345 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1346
e4404d6e 1347 device->fs_devices->num_devices--;
2b82032c 1348
cd02dca5
CM		 1349 if (device->missing)
1350 root->fs_info->fs_devices->missing_devices--;
1351
2b82032c
YZ		 1352 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1353 struct btrfs_device, dev_list);
1354 if (device->bdev == root->fs_info->sb->s_bdev)
1355 root->fs_info->sb->s_bdev = next_device->bdev;
1356 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1357 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1358
e4404d6e 1359 if (device->bdev) {
d4d77629 1360 blkdev_put(device->bdev, device->mode);
e4404d6e
YZ		 1361 device->bdev = NULL;
1362 device->fs_devices->open_devices--;
1363 }
1364
2b82032c
YZ		 1365 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1366 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1367
e4404d6e
YZ		 1368 if (device->fs_devices->open_devices == 0) {
1369 struct btrfs_fs_devices *fs_devices;
1370 fs_devices = root->fs_info->fs_devices;
1371 while (fs_devices) {
1372 if (fs_devices->seed == device->fs_devices)
1373 break;
1374 fs_devices = fs_devices->seed;
2b82032c 1375 }
e4404d6e
YZ		 1376 fs_devices->seed = device->fs_devices->seed;
1377 device->fs_devices->seed = NULL;
1378 __btrfs_close_devices(device->fs_devices);
1379 free_fs_devices(device->fs_devices);
2b82032c
YZ		 1380 }
1381
1382 /*
1383 * at this point, the device is zero sized. We want to
1384 * remove it from the devices list and zero out the old super
1385 */
1386 if (device->writeable) {
dfe25020
CM		 1387 /* make sure this device isn't detected as part of
1388 * the FS anymore
1389 */
1390 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1391 set_buffer_dirty(bh);
1392 sync_dirty_buffer(bh);
dfe25020 1393 }
a061fc8d
CM		 1394
1395 kfree(device->name);
1396 kfree(device);
1397 ret = 0;
a061fc8d
CM		 1398
1399error_brelse:
1400 brelse(bh);
1401error_close:
dfe25020 1402 if (bdev)
e525fd89 1403 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
a061fc8d 1404out:
7d9eb12c 1405 mutex_unlock(&root->fs_info->volume_mutex);
a061fc8d 1406 mutex_unlock(&uuid_mutex);
a061fc8d 1407 return ret;
9b3517e9
ID		 1408error_undo:
1409 if (device->writeable) {
1410 list_add(&device->dev_alloc_list,
1411 &root->fs_info->fs_devices->alloc_list);
1412 root->fs_info->fs_devices->rw_devices++;
1413 }
1414 goto error_brelse;
a061fc8d
CM		 1415}
1416
2b82032c
YZ		 1417/*
1418 * does all the dirty work required for changing file system's UUID.
1419 */
1420static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1421 struct btrfs_root *root)
1422{
1423 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1424 struct btrfs_fs_devices *old_devices;
e4404d6e 1425 struct btrfs_fs_devices *seed_devices;
2b82032c
YZ		 1426 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1427 struct btrfs_device *device;
1428 u64 super_flags;
1429
1430 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1431 if (!fs_devices->seeding)
2b82032c
YZ		 1432 return -EINVAL;
1433
e4404d6e
YZ		 1434 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1435 if (!seed_devices)
2b82032c
YZ		 1436 return -ENOMEM;
1437
e4404d6e
YZ		 1438 old_devices = clone_fs_devices(fs_devices);
1439 if (IS_ERR(old_devices)) {
1440 kfree(seed_devices);
1441 return PTR_ERR(old_devices);
2b82032c 1442 }
e4404d6e 1443
2b82032c
YZ		 1444 list_add(&old_devices->list, &fs_uuids);
1445
e4404d6e
YZ		 1446 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1447 seed_devices->opened = 1;
1448 INIT_LIST_HEAD(&seed_devices->devices);
1449 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1450 mutex_init(&seed_devices->device_list_mutex);
e4404d6e
YZ		 1451 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1452 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1453 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1454 device->fs_devices = seed_devices;
1455 }
1456
2b82032c
YZ		 1457 fs_devices->seeding = 0;
1458 fs_devices->num_devices = 0;
1459 fs_devices->open_devices = 0;
e4404d6e 1460 fs_devices->seed = seed_devices;
2b82032c
YZ		 1461
1462 generate_random_uuid(fs_devices->fsid);
1463 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1464 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1465 super_flags = btrfs_super_flags(disk_super) &
1466 ~BTRFS_SUPER_FLAG_SEEDING;
1467 btrfs_set_super_flags(disk_super, super_flags);
1468
1469 return 0;
1470}
1471
1472/*
1473 * strore the expected generation for seed devices in device items.
1474 */
1475static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1476 struct btrfs_root *root)
1477{
1478 struct btrfs_path *path;
1479 struct extent_buffer *leaf;
1480 struct btrfs_dev_item *dev_item;
1481 struct btrfs_device *device;
1482 struct btrfs_key key;
1483 u8 fs_uuid[BTRFS_UUID_SIZE];
1484 u8 dev_uuid[BTRFS_UUID_SIZE];
1485 u64 devid;
1486 int ret;
1487
1488 path = btrfs_alloc_path();
1489 if (!path)
1490 return -ENOMEM;
1491
1492 root = root->fs_info->chunk_root;
1493 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1494 key.offset = 0;
1495 key.type = BTRFS_DEV_ITEM_KEY;
1496
1497 while (1) {
1498 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1499 if (ret < 0)
1500 goto error;
1501
1502 leaf = path->nodes[0];
1503next_slot:
1504 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1505 ret = btrfs_next_leaf(root, path);
1506 if (ret > 0)
1507 break;
1508 if (ret < 0)
1509 goto error;
1510 leaf = path->nodes[0];
1511 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512 btrfs_release_path(root, path);
1513 continue;
1514 }
1515
1516 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1517 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1518 key.type != BTRFS_DEV_ITEM_KEY)
1519 break;
1520
1521 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1522 struct btrfs_dev_item);
1523 devid = btrfs_device_id(leaf, dev_item);
1524 read_extent_buffer(leaf, dev_uuid,
1525 (unsigned long)btrfs_device_uuid(dev_item),
1526 BTRFS_UUID_SIZE);
1527 read_extent_buffer(leaf, fs_uuid,
1528 (unsigned long)btrfs_device_fsid(dev_item),
1529 BTRFS_UUID_SIZE);
1530 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1531 BUG_ON(!device);
1532
1533 if (device->fs_devices->seeding) {
1534 btrfs_set_device_generation(leaf, dev_item,
1535 device->generation);
1536 btrfs_mark_buffer_dirty(leaf);
1537 }
1538
1539 path->slots[0]++;
1540 goto next_slot;
1541 }
1542 ret = 0;
1543error:
1544 btrfs_free_path(path);
1545 return ret;
1546}
1547
788f20eb
CM		 1548int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1549{
1550 struct btrfs_trans_handle *trans;
1551 struct btrfs_device *device;
1552 struct block_device *bdev;
788f20eb 1553 struct list_head *devices;
2b82032c 1554 struct super_block *sb = root->fs_info->sb;
788f20eb 1555 u64 total_bytes;
2b82032c 1556 int seeding_dev = 0;
788f20eb
CM		 1557 int ret = 0;
1558
2b82032c
YZ		 1559 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1560 return -EINVAL;
788f20eb 1561
d4d77629
TH		 1562 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1563 root->fs_info->bdev_holder);
7f59203a
JB		 1564 if (IS_ERR(bdev))
1565 return PTR_ERR(bdev);
a2135011 1566
2b82032c
YZ		 1567 if (root->fs_info->fs_devices->seeding) {
1568 seeding_dev = 1;
1569 down_write(&sb->s_umount);
1570 mutex_lock(&uuid_mutex);
1571 }
1572
8c8bee1d 1573 filemap_write_and_wait(bdev->bd_inode->i_mapping);
7d9eb12c 1574 mutex_lock(&root->fs_info->volume_mutex);
a2135011 1575
788f20eb 1576 devices = &root->fs_info->fs_devices->devices;
e5e9a520
CM		 1577 /*
1578 * we have the volume lock, so we don't need the extra
1579 * device list mutex while reading the list here.
1580 */
c6e30871 1581 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1582 if (device->bdev == bdev) {
1583 ret = -EEXIST;
2b82032c 1584 goto error;
788f20eb
CM		 1585 }
1586 }
1587
1588 device = kzalloc(sizeof(*device), GFP_NOFS);
1589 if (!device) {
1590 /* we can safely leave the fs_devices entry around */
1591 ret = -ENOMEM;
2b82032c 1592 goto error;
788f20eb
CM		 1593 }
1594
788f20eb
CM		 1595 device->name = kstrdup(device_path, GFP_NOFS);
1596 if (!device->name) {
1597 kfree(device);
2b82032c
YZ		 1598 ret = -ENOMEM;
1599 goto error;
788f20eb 1600 }
2b82032c
YZ		 1601
1602 ret = find_next_devid(root, &device->devid);
1603 if (ret) {
67100f25 1604 kfree(device->name);
2b82032c
YZ		 1605 kfree(device);
1606 goto error;
1607 }
1608
a22285a6 1609 trans = btrfs_start_transaction(root, 0);
98d5dc13 1610 if (IS_ERR(trans)) {
67100f25 1611 kfree(device->name);
98d5dc13
TI		 1612 kfree(device);
1613 ret = PTR_ERR(trans);
1614 goto error;
1615 }
1616
2b82032c
YZ		 1617 lock_chunks(root);
1618
2b82032c
YZ		 1619 device->writeable = 1;
1620 device->work.func = pending_bios_fn;
1621 generate_random_uuid(device->uuid);
1622 spin_lock_init(&device->io_lock);
1623 device->generation = trans->transid;
788f20eb
CM		 1624 device->io_width = root->sectorsize;
1625 device->io_align = root->sectorsize;
1626 device->sector_size = root->sectorsize;
1627 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1628 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1629 device->dev_root = root->fs_info->dev_root;
1630 device->bdev = bdev;
dfe25020 1631 device->in_fs_metadata = 1;
fb01aa85 1632 device->mode = FMODE_EXCL;
2b82032c 1633 set_blocksize(device->bdev, 4096);
788f20eb 1634
2b82032c
YZ		 1635 if (seeding_dev) {
1636 sb->s_flags &= ~MS_RDONLY;
1637 ret = btrfs_prepare_sprout(trans, root);
1638 BUG_ON(ret);
1639 }
788f20eb 1640
2b82032c 1641 device->fs_devices = root->fs_info->fs_devices;
e5e9a520
CM		 1642
1643 /*
1644 * we don't want write_supers to jump in here with our device
1645 * half setup
1646 */
1647 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2b82032c
YZ		 1648 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1649 list_add(&device->dev_alloc_list,
1650 &root->fs_info->fs_devices->alloc_list);
1651 root->fs_info->fs_devices->num_devices++;
1652 root->fs_info->fs_devices->open_devices++;
1653 root->fs_info->fs_devices->rw_devices++;
1654 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1655
c289811c
CM		 1656 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1657 root->fs_info->fs_devices->rotating = 1;
1658
788f20eb
CM		 1659 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1660 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1661 total_bytes + device->total_bytes);
1662
1663 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1664 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1665 total_bytes + 1);
e5e9a520 1666 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 1667
2b82032c
YZ		 1668 if (seeding_dev) {
1669 ret = init_first_rw_device(trans, root, device);
1670 BUG_ON(ret);
1671 ret = btrfs_finish_sprout(trans, root);
1672 BUG_ON(ret);
1673 } else {
1674 ret = btrfs_add_device(trans, root, device);
1675 }
1676
913d952e
CM		 1677 /*
1678 * we've got more storage, clear any full flags on the space
1679 * infos
1680 */
1681 btrfs_clear_space_info_full(root->fs_info);
1682
7d9eb12c 1683 unlock_chunks(root);
2b82032c 1684 btrfs_commit_transaction(trans, root);
a2135011 1685
2b82032c
YZ		 1686 if (seeding_dev) {
1687 mutex_unlock(&uuid_mutex);
1688 up_write(&sb->s_umount);
788f20eb 1689
2b82032c
YZ		 1690 ret = btrfs_relocate_sys_chunks(root);
1691 BUG_ON(ret);
1692 }
1693out:
1694 mutex_unlock(&root->fs_info->volume_mutex);
1695 return ret;
1696error:
e525fd89 1697 blkdev_put(bdev, FMODE_EXCL);
2b82032c
YZ		 1698 if (seeding_dev) {
1699 mutex_unlock(&uuid_mutex);
1700 up_write(&sb->s_umount);
1701 }
788f20eb
CM		 1702 goto out;
1703}
1704
d397712b
CM		 1705static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1706 struct btrfs_device *device)
0b86a832
CM		 1707{
1708 int ret;
1709 struct btrfs_path *path;
1710 struct btrfs_root *root;
1711 struct btrfs_dev_item *dev_item;
1712 struct extent_buffer *leaf;
1713 struct btrfs_key key;
1714
1715 root = device->dev_root->fs_info->chunk_root;
1716
1717 path = btrfs_alloc_path();
1718 if (!path)
1719 return -ENOMEM;
1720
1721 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1722 key.type = BTRFS_DEV_ITEM_KEY;
1723 key.offset = device->devid;
1724
1725 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1726 if (ret < 0)
1727 goto out;
1728
1729 if (ret > 0) {
1730 ret = -ENOENT;
1731 goto out;
1732 }
1733
1734 leaf = path->nodes[0];
1735 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1736
1737 btrfs_set_device_id(leaf, dev_item, device->devid);
1738 btrfs_set_device_type(leaf, dev_item, device->type);
1739 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1740 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1741 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 1742 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 1743 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1744 btrfs_mark_buffer_dirty(leaf);
1745
1746out:
1747 btrfs_free_path(path);
1748 return ret;
1749}
1750
7d9eb12c 1751static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1752 struct btrfs_device *device, u64 new_size)
1753{
1754 struct btrfs_super_block *super_copy =
1755 &device->dev_root->fs_info->super_copy;
1756 u64 old_total = btrfs_super_total_bytes(super_copy);
1757 u64 diff = new_size - device->total_bytes;
1758
2b82032c
YZ		 1759 if (!device->writeable)
1760 return -EACCES;
1761 if (new_size <= device->total_bytes)
1762 return -EINVAL;
1763
8f18cf13 1764 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 1765 device->fs_devices->total_rw_bytes += diff;
1766
1767 device->total_bytes = new_size;
9779b72f 1768 device->disk_total_bytes = new_size;
4184ea7f
CM		 1769 btrfs_clear_space_info_full(device->dev_root->fs_info);
1770
8f18cf13
CM		 1771 return btrfs_update_device(trans, device);
1772}
1773
7d9eb12c
CM		 1774int btrfs_grow_device(struct btrfs_trans_handle *trans,
1775 struct btrfs_device *device, u64 new_size)
1776{
1777 int ret;
1778 lock_chunks(device->dev_root);
1779 ret = __btrfs_grow_device(trans, device, new_size);
1780 unlock_chunks(device->dev_root);
1781 return ret;
1782}
1783
8f18cf13
CM		 1784static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 u64 chunk_tree, u64 chunk_objectid,
1787 u64 chunk_offset)
1788{
1789 int ret;
1790 struct btrfs_path *path;
1791 struct btrfs_key key;
1792
1793 root = root->fs_info->chunk_root;
1794 path = btrfs_alloc_path();
1795 if (!path)
1796 return -ENOMEM;
1797
1798 key.objectid = chunk_objectid;
1799 key.offset = chunk_offset;
1800 key.type = BTRFS_CHUNK_ITEM_KEY;
1801
1802 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1803 BUG_ON(ret);
1804
1805 ret = btrfs_del_item(trans, root, path);
1806 BUG_ON(ret);
1807
1808 btrfs_free_path(path);
1809 return 0;
1810}
1811
b2950863 1812static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 1813 chunk_offset)
1814{
1815 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1816 struct btrfs_disk_key *disk_key;
1817 struct btrfs_chunk *chunk;
1818 u8 *ptr;
1819 int ret = 0;
1820 u32 num_stripes;
1821 u32 array_size;
1822 u32 len = 0;
1823 u32 cur;
1824 struct btrfs_key key;
1825
1826 array_size = btrfs_super_sys_array_size(super_copy);
1827
1828 ptr = super_copy->sys_chunk_array;
1829 cur = 0;
1830
1831 while (cur < array_size) {
1832 disk_key = (struct btrfs_disk_key *)ptr;
1833 btrfs_disk_key_to_cpu(&key, disk_key);
1834
1835 len = sizeof(*disk_key);
1836
1837 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1838 chunk = (struct btrfs_chunk *)(ptr + len);
1839 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1840 len += btrfs_chunk_item_size(num_stripes);
1841 } else {
1842 ret = -EIO;
1843 break;
1844 }
1845 if (key.objectid == chunk_objectid &&
1846 key.offset == chunk_offset) {
1847 memmove(ptr, ptr + len, array_size - (cur + len));
1848 array_size -= len;
1849 btrfs_set_super_sys_array_size(super_copy, array_size);
1850 } else {
1851 ptr += len;
1852 cur += len;
1853 }
1854 }
1855 return ret;
1856}
1857
b2950863 1858static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 1859 u64 chunk_tree, u64 chunk_objectid,
1860 u64 chunk_offset)
1861{
1862 struct extent_map_tree *em_tree;
1863 struct btrfs_root *extent_root;
1864 struct btrfs_trans_handle *trans;
1865 struct extent_map *em;
1866 struct map_lookup *map;
1867 int ret;
1868 int i;
1869
1870 root = root->fs_info->chunk_root;
1871 extent_root = root->fs_info->extent_root;
1872 em_tree = &root->fs_info->mapping_tree.map_tree;
1873
ba1bf481
JB		 1874 ret = btrfs_can_relocate(extent_root, chunk_offset);
1875 if (ret)
1876 return -ENOSPC;
1877
8f18cf13 1878 /* step one, relocate all the extents inside this chunk */
1a40e23b 1879 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
a22285a6
YZ		 1880 if (ret)
1881 return ret;
8f18cf13 1882
a22285a6 1883 trans = btrfs_start_transaction(root, 0);
98d5dc13 1884 BUG_ON(IS_ERR(trans));
8f18cf13 1885
7d9eb12c
CM		 1886 lock_chunks(root);
1887
8f18cf13
CM		 1888 /*
1889 * step two, delete the device extents and the
1890 * chunk tree entries
1891 */
890871be 1892 read_lock(&em_tree->lock);
8f18cf13 1893 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 1894 read_unlock(&em_tree->lock);
8f18cf13 1895
a061fc8d
CM		 1896 BUG_ON(em->start > chunk_offset ||
1897 em->start + em->len < chunk_offset);
8f18cf13
CM		 1898 map = (struct map_lookup *)em->bdev;
1899
1900 for (i = 0; i < map->num_stripes; i++) {
1901 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1902 map->stripes[i].physical);
1903 BUG_ON(ret);
a061fc8d 1904
dfe25020
CM		 1905 if (map->stripes[i].dev) {
1906 ret = btrfs_update_device(trans, map->stripes[i].dev);
1907 BUG_ON(ret);
1908 }
8f18cf13
CM		 1909 }
1910 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1911 chunk_offset);
1912
1913 BUG_ON(ret);
1914
1abe9b8a 1915 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1916
8f18cf13
CM		 1917 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1918 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1919 BUG_ON(ret);
8f18cf13
CM		 1920 }
1921
2b82032c
YZ		 1922 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1923 BUG_ON(ret);
1924
890871be 1925 write_lock(&em_tree->lock);
2b82032c 1926 remove_extent_mapping(em_tree, em);
890871be 1927 write_unlock(&em_tree->lock);
2b82032c
YZ		 1928
1929 kfree(map);
1930 em->bdev = NULL;
1931
1932 /* once for the tree */
1933 free_extent_map(em);
1934 /* once for us */
1935 free_extent_map(em);
1936
1937 unlock_chunks(root);
1938 btrfs_end_transaction(trans, root);
1939 return 0;
1940}
1941
1942static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1943{
1944 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1945 struct btrfs_path *path;
1946 struct extent_buffer *leaf;
1947 struct btrfs_chunk *chunk;
1948 struct btrfs_key key;
1949 struct btrfs_key found_key;
1950 u64 chunk_tree = chunk_root->root_key.objectid;
1951 u64 chunk_type;
ba1bf481
JB		 1952 bool retried = false;
1953 int failed = 0;
2b82032c
YZ		 1954 int ret;
1955
1956 path = btrfs_alloc_path();
1957 if (!path)
1958 return -ENOMEM;
1959
ba1bf481 1960again:
2b82032c
YZ		 1961 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1962 key.offset = (u64)-1;
1963 key.type = BTRFS_CHUNK_ITEM_KEY;
1964
1965 while (1) {
1966 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1967 if (ret < 0)
1968 goto error;
1969 BUG_ON(ret == 0);
1970
1971 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1972 key.type);
1973 if (ret < 0)
1974 goto error;
1975 if (ret > 0)
1976 break;
1a40e23b 1977
2b82032c
YZ		 1978 leaf = path->nodes[0];
1979 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 1980
2b82032c
YZ		 1981 chunk = btrfs_item_ptr(leaf, path->slots[0],
1982 struct btrfs_chunk);
1983 chunk_type = btrfs_chunk_type(leaf, chunk);
1984 btrfs_release_path(chunk_root, path);
8f18cf13 1985
2b82032c
YZ		 1986 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1987 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1988 found_key.objectid,
1989 found_key.offset);
ba1bf481
JB		 1990 if (ret == -ENOSPC)
1991 failed++;
1992 else if (ret)
1993 BUG();
2b82032c 1994 }
8f18cf13 1995
2b82032c
YZ		 1996 if (found_key.offset == 0)
1997 break;
1998 key.offset = found_key.offset - 1;
1999 }
2000 ret = 0;
ba1bf481
JB		 2001 if (failed && !retried) {
2002 failed = 0;
2003 retried = true;
2004 goto again;
2005 } else if (failed && retried) {
2006 WARN_ON(1);
2007 ret = -ENOSPC;
2008 }
2b82032c
YZ		 2009error:
2010 btrfs_free_path(path);
2011 return ret;
8f18cf13
CM		 2012}
2013
ec44a35c
CM		 2014static u64 div_factor(u64 num, int factor)
2015{
2016 if (factor == 10)
2017 return num;
2018 num *= factor;
2019 do_div(num, 10);
2020 return num;
2021}
2022
ec44a35c
CM		 2023int btrfs_balance(struct btrfs_root *dev_root)
2024{
2025 int ret;
ec44a35c
CM		 2026 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2027 struct btrfs_device *device;
2028 u64 old_size;
2029 u64 size_to_free;
2030 struct btrfs_path *path;
2031 struct btrfs_key key;
ec44a35c
CM		 2032 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2033 struct btrfs_trans_handle *trans;
2034 struct btrfs_key found_key;
2035
2b82032c
YZ		 2036 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2037 return -EROFS;
ec44a35c 2038
6f88a440
BH		 2039 if (!capable(CAP_SYS_ADMIN))
2040 return -EPERM;
2041
7d9eb12c 2042 mutex_lock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 2043 dev_root = dev_root->fs_info->dev_root;
2044
ec44a35c 2045 /* step one make some room on all the devices */
c6e30871 2046 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 2047 old_size = device->total_bytes;
2048 size_to_free = div_factor(old_size, 1);
2049 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c
YZ		 2050 if (!device->writeable ||
2051 device->total_bytes - device->bytes_used > size_to_free)
ec44a35c
CM		 2052 continue;
2053
2054 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 2055 if (ret == -ENOSPC)
2056 break;
ec44a35c
CM		 2057 BUG_ON(ret);
2058
a22285a6 2059 trans = btrfs_start_transaction(dev_root, 0);
98d5dc13 2060 BUG_ON(IS_ERR(trans));
ec44a35c
CM		 2061
2062 ret = btrfs_grow_device(trans, device, old_size);
2063 BUG_ON(ret);
2064
2065 btrfs_end_transaction(trans, dev_root);
2066 }
2067
2068 /* step two, relocate all the chunks */
2069 path = btrfs_alloc_path();
2070 BUG_ON(!path);
2071
2072 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2073 key.offset = (u64)-1;
2074 key.type = BTRFS_CHUNK_ITEM_KEY;
2075
d397712b 2076 while (1) {
ec44a35c
CM		 2077 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2078 if (ret < 0)
2079 goto error;
2080
2081 /*
2082 * this shouldn't happen, it means the last relocate
2083 * failed
2084 */
2085 if (ret == 0)
2086 break;
2087
2088 ret = btrfs_previous_item(chunk_root, path, 0,
2089 BTRFS_CHUNK_ITEM_KEY);
7d9eb12c 2090 if (ret)
ec44a35c 2091 break;
7d9eb12c 2092
ec44a35c
CM		 2093 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2094 path->slots[0]);
2095 if (found_key.objectid != key.objectid)
2096 break;
7d9eb12c 2097
ec44a35c 2098 /* chunk zero is special */
ba1bf481 2099 if (found_key.offset == 0)
ec44a35c
CM		 2100 break;
2101
7d9eb12c 2102 btrfs_release_path(chunk_root, path);
ec44a35c
CM		 2103 ret = btrfs_relocate_chunk(chunk_root,
2104 chunk_root->root_key.objectid,
2105 found_key.objectid,
2106 found_key.offset);
ba1bf481
JB		 2107 BUG_ON(ret && ret != -ENOSPC);
2108 key.offset = found_key.offset - 1;
ec44a35c
CM		 2109 }
2110 ret = 0;
2111error:
2112 btrfs_free_path(path);
7d9eb12c 2113 mutex_unlock(&dev_root->fs_info->volume_mutex);
ec44a35c
CM		 2114 return ret;
2115}
2116
8f18cf13
CM		 2117/*
2118 * shrinking a device means finding all of the device extents past
2119 * the new size, and then following the back refs to the chunks.
2120 * The chunk relocation code actually frees the device extent
2121 */
2122int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2123{
2124 struct btrfs_trans_handle *trans;
2125 struct btrfs_root *root = device->dev_root;
2126 struct btrfs_dev_extent *dev_extent = NULL;
2127 struct btrfs_path *path;
2128 u64 length;
2129 u64 chunk_tree;
2130 u64 chunk_objectid;
2131 u64 chunk_offset;
2132 int ret;
2133 int slot;
ba1bf481
JB		 2134 int failed = 0;
2135 bool retried = false;
8f18cf13
CM		 2136 struct extent_buffer *l;
2137 struct btrfs_key key;
2138 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2139 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 2140 u64 old_size = device->total_bytes;
8f18cf13
CM		 2141 u64 diff = device->total_bytes - new_size;
2142
2b82032c
YZ		 2143 if (new_size >= device->total_bytes)
2144 return -EINVAL;
8f18cf13
CM		 2145
2146 path = btrfs_alloc_path();
2147 if (!path)
2148 return -ENOMEM;
2149
8f18cf13
CM		 2150 path->reada = 2;
2151
7d9eb12c
CM		 2152 lock_chunks(root);
2153
8f18cf13 2154 device->total_bytes = new_size;
2b82032c
YZ		 2155 if (device->writeable)
2156 device->fs_devices->total_rw_bytes -= diff;
7d9eb12c 2157 unlock_chunks(root);
8f18cf13 2158
ba1bf481 2159again:
8f18cf13
CM		 2160 key.objectid = device->devid;
2161 key.offset = (u64)-1;
2162 key.type = BTRFS_DEV_EXTENT_KEY;
2163
2164 while (1) {
2165 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2166 if (ret < 0)
2167 goto done;
2168
2169 ret = btrfs_previous_item(root, path, 0, key.type);
2170 if (ret < 0)
2171 goto done;
2172 if (ret) {
2173 ret = 0;
ba1bf481 2174 btrfs_release_path(root, path);
bf1fb512 2175 break;
8f18cf13
CM		 2176 }
2177
2178 l = path->nodes[0];
2179 slot = path->slots[0];
2180 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2181
ba1bf481
JB		 2182 if (key.objectid != device->devid) {
2183 btrfs_release_path(root, path);
bf1fb512 2184 break;
ba1bf481 2185 }
8f18cf13
CM		 2186
2187 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2188 length = btrfs_dev_extent_length(l, dev_extent);
2189
ba1bf481
JB		 2190 if (key.offset + length <= new_size) {
2191 btrfs_release_path(root, path);
d6397bae 2192 break;
ba1bf481 2193 }
8f18cf13
CM		 2194
2195 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2196 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2197 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2198 btrfs_release_path(root, path);
2199
2200 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2201 chunk_offset);
ba1bf481 2202 if (ret && ret != -ENOSPC)
8f18cf13 2203 goto done;
ba1bf481
JB		 2204 if (ret == -ENOSPC)
2205 failed++;
2206 key.offset -= 1;
2207 }
2208
2209 if (failed && !retried) {
2210 failed = 0;
2211 retried = true;
2212 goto again;
2213 } else if (failed && retried) {
2214 ret = -ENOSPC;
2215 lock_chunks(root);
2216
2217 device->total_bytes = old_size;
2218 if (device->writeable)
2219 device->fs_devices->total_rw_bytes += diff;
2220 unlock_chunks(root);
2221 goto done;
8f18cf13
CM		 2222 }
2223
d6397bae 2224 /* Shrinking succeeded, else we would be at "done". */
a22285a6 2225 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 2226 if (IS_ERR(trans)) {
2227 ret = PTR_ERR(trans);
2228 goto done;
2229 }
2230
d6397bae
CB		 2231 lock_chunks(root);
2232
2233 device->disk_total_bytes = new_size;
2234 /* Now btrfs_update_device() will change the on-disk size. */
2235 ret = btrfs_update_device(trans, device);
2236 if (ret) {
2237 unlock_chunks(root);
2238 btrfs_end_transaction(trans, root);
2239 goto done;
2240 }
2241 WARN_ON(diff > old_total);
2242 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2243 unlock_chunks(root);
2244 btrfs_end_transaction(trans, root);
8f18cf13
CM		 2245done:
2246 btrfs_free_path(path);
2247 return ret;
2248}
2249
b2950863 2250static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
0b86a832
CM		 2251 struct btrfs_root *root,
2252 struct btrfs_key *key,
2253 struct btrfs_chunk *chunk, int item_size)
2254{
2255 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2256 struct btrfs_disk_key disk_key;
2257 u32 array_size;
2258 u8 *ptr;
2259
2260 array_size = btrfs_super_sys_array_size(super_copy);
2261 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2262 return -EFBIG;
2263
2264 ptr = super_copy->sys_chunk_array + array_size;
2265 btrfs_cpu_key_to_disk(&disk_key, key);
2266 memcpy(ptr, &disk_key, sizeof(disk_key));
2267 ptr += sizeof(disk_key);
2268 memcpy(ptr, chunk, item_size);
2269 item_size += sizeof(disk_key);
2270 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2271 return 0;
2272}
2273
d397712b 2274static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
a1b32a59 2275 int num_stripes, int sub_stripes)
9b3f68b9
CM		 2276{
2277 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2278 return calc_size;
2279 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2280 return calc_size * (num_stripes / sub_stripes);
2281 else
2282 return calc_size * num_stripes;
2283}
2284
b2117a39
MX		 2285/* Used to sort the devices by max_avail(descending sort) */
2286int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
0b86a832 2287{
b2117a39
MX		 2288 if (((struct btrfs_device_info *)dev_info1)->max_avail >
2289 ((struct btrfs_device_info *)dev_info2)->max_avail)
2290 return -1;
2291 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2292 ((struct btrfs_device_info *)dev_info2)->max_avail)
2293 return 1;
2294 else
2295 return 0;
2296}
0b86a832 2297
b2117a39
MX		 2298static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2299 int *num_stripes, int *min_stripes,
2300 int *sub_stripes)
2301{
2302 *num_stripes = 1;
2303 *min_stripes = 1;
2304 *sub_stripes = 0;
593060d7 2305
a40a90a0 2306 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
b2117a39
MX		 2307 *num_stripes = fs_devices->rw_devices;
2308 *min_stripes = 2;
a40a90a0
CM		 2309 }
2310 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
b2117a39
MX		 2311 *num_stripes = 2;
2312 *min_stripes = 2;
a40a90a0 2313 }
8790d502 2314 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
f3eae7e8 2315 if (fs_devices->rw_devices < 2)
9b3f68b9 2316 return -ENOSPC;
b2117a39
MX		 2317 *num_stripes = 2;
2318 *min_stripes = 2;
8790d502 2319 }
321aecc6 2320 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
b2117a39
MX		 2321 *num_stripes = fs_devices->rw_devices;
2322 if (*num_stripes < 4)
321aecc6 2323 return -ENOSPC;
b2117a39
MX		 2324 *num_stripes &= ~(u32)1;
2325 *sub_stripes = 2;
2326 *min_stripes = 4;
321aecc6 2327 }
9b3f68b9 2328
b2117a39
MX		 2329 return 0;
2330}
2331
2332static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2333 u64 proposed_size, u64 type,
2334 int num_stripes, int small_stripe)
2335{
2336 int min_stripe_size = 1 * 1024 * 1024;
2337 u64 calc_size = proposed_size;
2338 u64 max_chunk_size = calc_size;
2339 int ncopies = 1;
2340
2341 if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2342 BTRFS_BLOCK_GROUP_DUP |
2343 BTRFS_BLOCK_GROUP_RAID10))
2344 ncopies = 2;
2345
9b3f68b9
CM		 2346 if (type & BTRFS_BLOCK_GROUP_DATA) {
2347 max_chunk_size = 10 * calc_size;
a40a90a0 2348 min_stripe_size = 64 * 1024 * 1024;
9b3f68b9 2349 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
83d3c969 2350 max_chunk_size = 256 * 1024 * 1024;
a40a90a0
CM		 2351 min_stripe_size = 32 * 1024 * 1024;
2352 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2353 calc_size = 8 * 1024 * 1024;
2354 max_chunk_size = calc_size * 2;
2355 min_stripe_size = 1 * 1024 * 1024;
9b3f68b9
CM		 2356 }
2357
2b82032c
YZ		 2358 /* we don't want a chunk larger than 10% of writeable space */
2359 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2360 max_chunk_size);
9b3f68b9 2361
1974a3b4
MX		 2362 if (calc_size * num_stripes > max_chunk_size * ncopies) {
2363 calc_size = max_chunk_size * ncopies;
9b3f68b9 2364 do_div(calc_size, num_stripes);
b2117a39
MX		 2365 do_div(calc_size, BTRFS_STRIPE_LEN);
2366 calc_size *= BTRFS_STRIPE_LEN;
9b3f68b9 2367 }
0cad8a11 2368
9b3f68b9 2369 /* we don't want tiny stripes */
b2117a39 2370 if (!small_stripe)
0cad8a11 2371 calc_size = max_t(u64, min_stripe_size, calc_size);
9b3f68b9 2372
9f680ce0 2373 /*
b2117a39 2374 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
9f680ce0
CM		 2375 * we end up with something bigger than a stripe
2376 */
b2117a39
MX		 2377 calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2378
2379 do_div(calc_size, BTRFS_STRIPE_LEN);
2380 calc_size *= BTRFS_STRIPE_LEN;
2381
2382 return calc_size;
2383}
2384
2385static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2386 int num_stripes)
2387{
2388 struct map_lookup *new;
2389 size_t len = map_lookup_size(num_stripes);
2390
2391 BUG_ON(map->num_stripes < num_stripes);
2392
2393 if (map->num_stripes == num_stripes)
2394 return map;
2395
2396 new = kmalloc(len, GFP_NOFS);
2397 if (!new) {
2398 /* just change map->num_stripes */
2399 map->num_stripes = num_stripes;
2400 return map;
2401 }
2402
2403 memcpy(new, map, len);
2404 new->num_stripes = num_stripes;
2405 kfree(map);
2406 return new;
2407}
2408
2409/*
2410 * helper to allocate device space from btrfs_device_info, in which we stored
2411 * max free space information of every device. It is used when we can not
2412 * allocate chunks by default size.
2413 *
2414 * By this helper, we can allocate a new chunk as larger as possible.
2415 */
2416static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2417 struct btrfs_fs_devices *fs_devices,
2418 struct btrfs_device_info *devices,
2419 int nr_device, u64 type,
2420 struct map_lookup **map_lookup,
2421 int min_stripes, u64 *stripe_size)
2422{
2423 int i, index, sort_again = 0;
2424 int min_devices = min_stripes;
2425 u64 max_avail, min_free;
2426 struct map_lookup *map = *map_lookup;
2427 int ret;
9f680ce0 2428
b2117a39
MX		 2429 if (nr_device < min_stripes)
2430 return -ENOSPC;
2431
2432 btrfs_descending_sort_devices(devices, nr_device);
2433
2434 max_avail = devices[0].max_avail;
2435 if (!max_avail)
2436 return -ENOSPC;
2437
2438 for (i = 0; i < nr_device; i++) {
2439 /*
2440 * if dev_offset = 0, it means the free space of this device
2441 * is less than what we need, and we didn't search max avail
2442 * extent on this device, so do it now.
2443 */
2444 if (!devices[i].dev_offset) {
2445 ret = find_free_dev_extent(trans, devices[i].dev,
2446 max_avail,
2447 &devices[i].dev_offset,
2448 &devices[i].max_avail);
2449 if (ret != 0 && ret != -ENOSPC)
2450 return ret;
2451 sort_again = 1;
2452 }
2453 }
2454
2455 /* we update the max avail free extent of each devices, sort again */
2456 if (sort_again)
2457 btrfs_descending_sort_devices(devices, nr_device);
2458
2459 if (type & BTRFS_BLOCK_GROUP_DUP)
2460 min_devices = 1;
2461
2462 if (!devices[min_devices - 1].max_avail)
2463 return -ENOSPC;
2464
2465 max_avail = devices[min_devices - 1].max_avail;
2466 if (type & BTRFS_BLOCK_GROUP_DUP)
2467 do_div(max_avail, 2);
2468
2469 max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2470 min_stripes, 1);
2471 if (type & BTRFS_BLOCK_GROUP_DUP)
2472 min_free = max_avail * 2;
2473 else
2474 min_free = max_avail;
2475
2476 if (min_free > devices[min_devices - 1].max_avail)
2477 return -ENOSPC;
2478
2479 map = __shrink_map_lookup_stripes(map, min_stripes);
2480 *stripe_size = max_avail;
2481
2482 index = 0;
2483 for (i = 0; i < min_stripes; i++) {
2484 map->stripes[i].dev = devices[index].dev;
2485 map->stripes[i].physical = devices[index].dev_offset;
2486 if (type & BTRFS_BLOCK_GROUP_DUP) {
2487 i++;
2488 map->stripes[i].dev = devices[index].dev;
2489 map->stripes[i].physical = devices[index].dev_offset +
2490 max_avail;
2491 }
2492 index++;
2493 }
2494 *map_lookup = map;
9f680ce0 2495
b2117a39
MX		 2496 return 0;
2497}
2498
2499static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2500 struct btrfs_root *extent_root,
2501 struct map_lookup **map_ret,
2502 u64 *num_bytes, u64 *stripe_size,
2503 u64 start, u64 type)
2504{
2505 struct btrfs_fs_info *info = extent_root->fs_info;
2506 struct btrfs_device *device = NULL;
2507 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2508 struct list_head *cur;
2509 struct map_lookup *map;
2510 struct extent_map_tree *em_tree;
2511 struct extent_map *em;
2512 struct btrfs_device_info *devices_info;
2513 struct list_head private_devs;
2514 u64 calc_size = 1024 * 1024 * 1024;
2515 u64 min_free;
2516 u64 avail;
2517 u64 dev_offset;
2518 int num_stripes;
2519 int min_stripes;
2520 int sub_stripes;
2521 int min_devices; /* the min number of devices we need */
2522 int i;
2523 int ret;
2524 int index;
2525
2526 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2527 (type & BTRFS_BLOCK_GROUP_DUP)) {
2528 WARN_ON(1);
2529 type &= ~BTRFS_BLOCK_GROUP_DUP;
2530 }
2531 if (list_empty(&fs_devices->alloc_list))
2532 return -ENOSPC;
2533
2534 ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2535 &min_stripes, &sub_stripes);
2536 if (ret)
2537 return ret;
2538
2539 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2540 GFP_NOFS);
2541 if (!devices_info)
2542 return -ENOMEM;
2543
2544 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2545 if (!map) {
2546 ret = -ENOMEM;
2547 goto error;
2548 }
2549 map->num_stripes = num_stripes;
9b3f68b9 2550
2b82032c 2551 cur = fs_devices->alloc_list.next;
6324fbf3 2552 index = 0;
b2117a39 2553 i = 0;
611f0e00 2554
b2117a39
MX		 2555 calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2556 num_stripes, 0);
2557
2558 if (type & BTRFS_BLOCK_GROUP_DUP) {
611f0e00 2559 min_free = calc_size * 2;
b2117a39
MX		 2560 min_devices = 1;
2561 } else {
9b3f68b9 2562 min_free = calc_size;
b2117a39
MX		 2563 min_devices = min_stripes;
2564 }
ad5bd91e 2565
2b82032c 2566 INIT_LIST_HEAD(&private_devs);
d397712b 2567 while (index < num_stripes) {
b3075717 2568 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2b82032c 2569 BUG_ON(!device->writeable);
dfe25020
CM		 2570 if (device->total_bytes > device->bytes_used)
2571 avail = device->total_bytes - device->bytes_used;
2572 else
2573 avail = 0;
6324fbf3 2574 cur = cur->next;
8f18cf13 2575
dfe25020 2576 if (device->in_fs_metadata && avail >= min_free) {
b2117a39
MX		 2577 ret = find_free_dev_extent(trans, device, min_free,
2578 &devices_info[i].dev_offset,
2579 &devices_info[i].max_avail);
8f18cf13
CM		 2580 if (ret == 0) {
2581 list_move_tail(&device->dev_alloc_list,
2582 &private_devs);
2b82032c 2583 map->stripes[index].dev = device;
b2117a39
MX		 2584 map->stripes[index].physical =
2585 devices_info[i].dev_offset;
611f0e00 2586 index++;
2b82032c
YZ		 2587 if (type & BTRFS_BLOCK_GROUP_DUP) {
2588 map->stripes[index].dev = device;
2589 map->stripes[index].physical =
b2117a39
MX		 2590 devices_info[i].dev_offset +
2591 calc_size;
8f18cf13 2592 index++;
2b82032c 2593 }
b2117a39
MX		 2594 } else if (ret != -ENOSPC)
2595 goto error;
2596
2597 devices_info[i].dev = device;
2598 i++;
2599 } else if (device->in_fs_metadata &&
2600 avail >= BTRFS_STRIPE_LEN) {
2601 devices_info[i].dev = device;
2602 devices_info[i].max_avail = avail;
2603 i++;
2604 }
2605
2b82032c 2606 if (cur == &fs_devices->alloc_list)
6324fbf3
CM		 2607 break;
2608 }
b2117a39 2609
2b82032c 2610 list_splice(&private_devs, &fs_devices->alloc_list);
6324fbf3 2611 if (index < num_stripes) {
a40a90a0
CM		 2612 if (index >= min_stripes) {
2613 num_stripes = index;
2614 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2615 num_stripes /= sub_stripes;
2616 num_stripes *= sub_stripes;
2617 }
b2117a39
MX		 2618
2619 map = __shrink_map_lookup_stripes(map, num_stripes);
2620 } else if (i >= min_devices) {
2621 ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2622 devices_info, i, type,
2623 &map, min_stripes,
2624 &calc_size);
2625 if (ret)
2626 goto error;
2627 } else {
2628 ret = -ENOSPC;
2629 goto error;
6324fbf3 2630 }
6324fbf3 2631 }
2b82032c 2632 map->sector_size = extent_root->sectorsize;
b2117a39
MX		 2633 map->stripe_len = BTRFS_STRIPE_LEN;
2634 map->io_align = BTRFS_STRIPE_LEN;
2635 map->io_width = BTRFS_STRIPE_LEN;
2b82032c 2636 map->type = type;
2b82032c 2637 map->sub_stripes = sub_stripes;
0b86a832 2638
2b82032c
YZ		 2639 *map_ret = map;
2640 *stripe_size = calc_size;
2641 *num_bytes = chunk_bytes_by_type(type, calc_size,
b2117a39 2642 map->num_stripes, sub_stripes);
0b86a832 2643
1abe9b8a 2644 trace_btrfs_chunk_alloc(info->chunk_root, map, start, *num_bytes);
2645
2b82032c
YZ		 2646 em = alloc_extent_map(GFP_NOFS);
2647 if (!em) {
b2117a39
MX		 2648 ret = -ENOMEM;
2649 goto error;
593060d7 2650 }
2b82032c
YZ		 2651 em->bdev = (struct block_device *)map;
2652 em->start = start;
2653 em->len = *num_bytes;
2654 em->block_start = 0;
2655 em->block_len = em->len;
593060d7 2656
2b82032c 2657 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 2658 write_lock(&em_tree->lock);
2b82032c 2659 ret = add_extent_mapping(em_tree, em);
890871be 2660 write_unlock(&em_tree->lock);
2b82032c
YZ		 2661 BUG_ON(ret);
2662 free_extent_map(em);
0b86a832 2663
2b82032c
YZ		 2664 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2665 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2666 start, *num_bytes);
2667 BUG_ON(ret);
611f0e00 2668
2b82032c
YZ		 2669 index = 0;
2670 while (index < map->num_stripes) {
2671 device = map->stripes[index].dev;
2672 dev_offset = map->stripes[index].physical;
0b86a832
CM		 2673
2674 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 2675 info->chunk_root->root_key.objectid,
2676 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2677 start, dev_offset, calc_size);
0b86a832 2678 BUG_ON(ret);
2b82032c
YZ		 2679 index++;
2680 }
2681
b2117a39 2682 kfree(devices_info);
2b82032c 2683 return 0;
b2117a39
MX		 2684
2685error:
2686 kfree(map);
2687 kfree(devices_info);
2688 return ret;
2b82032c
YZ		 2689}
2690
2691static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2692 struct btrfs_root *extent_root,
2693 struct map_lookup *map, u64 chunk_offset,
2694 u64 chunk_size, u64 stripe_size)
2695{
2696 u64 dev_offset;
2697 struct btrfs_key key;
2698 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2699 struct btrfs_device *device;
2700 struct btrfs_chunk *chunk;
2701 struct btrfs_stripe *stripe;
2702 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2703 int index = 0;
2704 int ret;
2705
2706 chunk = kzalloc(item_size, GFP_NOFS);
2707 if (!chunk)
2708 return -ENOMEM;
2709
2710 index = 0;
2711 while (index < map->num_stripes) {
2712 device = map->stripes[index].dev;
2713 device->bytes_used += stripe_size;
0b86a832
CM		 2714 ret = btrfs_update_device(trans, device);
2715 BUG_ON(ret);
2b82032c
YZ		 2716 index++;
2717 }
2718
2719 index = 0;
2720 stripe = &chunk->stripe;
2721 while (index < map->num_stripes) {
2722 device = map->stripes[index].dev;
2723 dev_offset = map->stripes[index].physical;
0b86a832 2724
e17cade2
CM		 2725 btrfs_set_stack_stripe_devid(stripe, device->devid);
2726 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2727 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 2728 stripe++;
0b86a832
CM		 2729 index++;
2730 }
2731
2b82032c 2732 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 2733 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 2734 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2735 btrfs_set_stack_chunk_type(chunk, map->type);
2736 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2737 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2738 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 2739 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 2740 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 2741
2b82032c
YZ		 2742 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2743 key.type = BTRFS_CHUNK_ITEM_KEY;
2744 key.offset = chunk_offset;
0b86a832 2745
2b82032c
YZ		 2746 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2747 BUG_ON(ret);
0b86a832 2748
2b82032c
YZ		 2749 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2750 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2751 item_size);
8f18cf13
CM		 2752 BUG_ON(ret);
2753 }
1abe9b8a 2754
0b86a832 2755 kfree(chunk);
2b82032c
YZ		 2756 return 0;
2757}
0b86a832 2758
2b82032c
YZ		 2759/*
2760 * Chunk allocation falls into two parts. The first part does works
2761 * that make the new allocated chunk useable, but not do any operation
2762 * that modifies the chunk tree. The second part does the works that
2763 * require modifying the chunk tree. This division is important for the
2764 * bootstrap process of adding storage to a seed btrfs.
2765 */
2766int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2767 struct btrfs_root *extent_root, u64 type)
2768{
2769 u64 chunk_offset;
2770 u64 chunk_size;
2771 u64 stripe_size;
2772 struct map_lookup *map;
2773 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2774 int ret;
2775
2776 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2777 &chunk_offset);
2778 if (ret)
2779 return ret;
2780
2781 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2782 &stripe_size, chunk_offset, type);
2783 if (ret)
2784 return ret;
2785
2786 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2787 chunk_size, stripe_size);
2788 BUG_ON(ret);
2789 return 0;
2790}
2791
d397712b 2792static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 2793 struct btrfs_root *root,
2794 struct btrfs_device *device)
2795{
2796 u64 chunk_offset;
2797 u64 sys_chunk_offset;
2798 u64 chunk_size;
2799 u64 sys_chunk_size;
2800 u64 stripe_size;
2801 u64 sys_stripe_size;
2802 u64 alloc_profile;
2803 struct map_lookup *map;
2804 struct map_lookup *sys_map;
2805 struct btrfs_fs_info *fs_info = root->fs_info;
2806 struct btrfs_root *extent_root = fs_info->extent_root;
2807 int ret;
2808
2809 ret = find_next_chunk(fs_info->chunk_root,
2810 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2811 BUG_ON(ret);
2812
2813 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2814 (fs_info->metadata_alloc_profile &
2815 fs_info->avail_metadata_alloc_bits);
2816 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2817
2818 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2819 &stripe_size, chunk_offset, alloc_profile);
2820 BUG_ON(ret);
2821
2822 sys_chunk_offset = chunk_offset + chunk_size;
2823
2824 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2825 (fs_info->system_alloc_profile &
2826 fs_info->avail_system_alloc_bits);
2827 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2828
2829 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2830 &sys_chunk_size, &sys_stripe_size,
2831 sys_chunk_offset, alloc_profile);
2832 BUG_ON(ret);
2833
2834 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2835 BUG_ON(ret);
2836
2837 /*
2838 * Modifying chunk tree needs allocating new blocks from both
2839 * system block group and metadata block group. So we only can
2840 * do operations require modifying the chunk tree after both
2841 * block groups were created.
2842 */
2843 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2844 chunk_size, stripe_size);
2845 BUG_ON(ret);
2846
2847 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2848 sys_chunk_offset, sys_chunk_size,
2849 sys_stripe_size);
b248a415 2850 BUG_ON(ret);
2b82032c
YZ		 2851 return 0;
2852}
2853
2854int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2855{
2856 struct extent_map *em;
2857 struct map_lookup *map;
2858 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2859 int readonly = 0;
2860 int i;
2861
890871be 2862 read_lock(&map_tree->map_tree.lock);
2b82032c 2863 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 2864 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 2865 if (!em)
2866 return 1;
2867
f48b9075
JB		 2868 if (btrfs_test_opt(root, DEGRADED)) {
2869 free_extent_map(em);
2870 return 0;
2871 }
2872
2b82032c
YZ		 2873 map = (struct map_lookup *)em->bdev;
2874 for (i = 0; i < map->num_stripes; i++) {
2875 if (!map->stripes[i].dev->writeable) {
2876 readonly = 1;
2877 break;
2878 }
2879 }
0b86a832 2880 free_extent_map(em);
2b82032c 2881 return readonly;
0b86a832
CM		 2882}
2883
2884void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2885{
2886 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2887}
2888
2889void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2890{
2891 struct extent_map *em;
2892
d397712b 2893 while (1) {
890871be 2894 write_lock(&tree->map_tree.lock);
0b86a832
CM		 2895 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2896 if (em)
2897 remove_extent_mapping(&tree->map_tree, em);
890871be 2898 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 2899 if (!em)
2900 break;
2901 kfree(em->bdev);
2902 /* once for us */
2903 free_extent_map(em);
2904 /* once for the tree */
2905 free_extent_map(em);
2906 }
2907}
2908
f188591e
CM		 2909int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2910{
2911 struct extent_map *em;
2912 struct map_lookup *map;
2913 struct extent_map_tree *em_tree = &map_tree->map_tree;
2914 int ret;
2915
890871be 2916 read_lock(&em_tree->lock);
f188591e 2917 em = lookup_extent_mapping(em_tree, logical, len);
890871be 2918 read_unlock(&em_tree->lock);
f188591e
CM		 2919 BUG_ON(!em);
2920
2921 BUG_ON(em->start > logical || em->start + em->len < logical);
2922 map = (struct map_lookup *)em->bdev;
2923 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2924 ret = map->num_stripes;
321aecc6
CM		 2925 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2926 ret = map->sub_stripes;
f188591e
CM		 2927 else
2928 ret = 1;
2929 free_extent_map(em);
f188591e
CM		 2930 return ret;
2931}
2932
dfe25020
CM		 2933static int find_live_mirror(struct map_lookup *map, int first, int num,
2934 int optimal)
2935{
2936 int i;
2937 if (map->stripes[optimal].dev->bdev)
2938 return optimal;
2939 for (i = first; i < first + num; i++) {
2940 if (map->stripes[i].dev->bdev)
2941 return i;
2942 }
2943 /* we couldn't find one that doesn't fail. Just return something
2944 * and the io error handling code will clean up eventually
2945 */
2946 return optimal;
2947}
2948
f2d8d74d
CM		 2949static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2950 u64 logical, u64 *length,
2951 struct btrfs_multi_bio **multi_ret,
2952 int mirror_num, struct page *unplug_page)
0b86a832
CM		 2953{
2954 struct extent_map *em;
2955 struct map_lookup *map;
2956 struct extent_map_tree *em_tree = &map_tree->map_tree;
2957 u64 offset;
593060d7 2958 u64 stripe_offset;
fce3bb9a 2959 u64 stripe_end_offset;
593060d7 2960 u64 stripe_nr;
fce3bb9a
LD		 2961 u64 stripe_nr_orig;
2962 u64 stripe_nr_end;
cea9e445 2963 int stripes_allocated = 8;
321aecc6 2964 int stripes_required = 1;
593060d7 2965 int stripe_index;
cea9e445 2966 int i;
f2d8d74d 2967 int num_stripes;
a236aed1 2968 int max_errors = 0;
cea9e445 2969 struct btrfs_multi_bio *multi = NULL;
0b86a832 2970
fce3bb9a 2971 if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
cea9e445 2972 stripes_allocated = 1;
cea9e445
CM		 2973again:
2974 if (multi_ret) {
2975 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2976 GFP_NOFS);
2977 if (!multi)
2978 return -ENOMEM;
a236aed1
CM		 2979
2980 atomic_set(&multi->error, 0);
cea9e445 2981 }
0b86a832 2982
890871be 2983 read_lock(&em_tree->lock);
0b86a832 2984 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 2985 read_unlock(&em_tree->lock);
f2d8d74d 2986
2423fdfb
JS		 2987 if (!em && unplug_page) {
2988 kfree(multi);
f2d8d74d 2989 return 0;
2423fdfb 2990 }
f2d8d74d 2991
3b951516 2992 if (!em) {
d397712b
CM		 2993 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2994 (unsigned long long)logical,
2995 (unsigned long long)*length);
f2d8d74d 2996 BUG();
3b951516 2997 }
0b86a832
CM		 2998
2999 BUG_ON(em->start > logical || em->start + em->len < logical);
3000 map = (struct map_lookup *)em->bdev;
3001 offset = logical - em->start;
593060d7 3002
f188591e
CM		 3003 if (mirror_num > map->num_stripes)
3004 mirror_num = 0;
3005
cea9e445 3006 /* if our multi bio struct is too small, back off and try again */
7b6d91da 3007 if (rw & REQ_WRITE) {
321aecc6
CM		 3008 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3009 BTRFS_BLOCK_GROUP_DUP)) {
3010 stripes_required = map->num_stripes;
a236aed1 3011 max_errors = 1;
321aecc6
CM		 3012 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3013 stripes_required = map->sub_stripes;
a236aed1 3014 max_errors = 1;
321aecc6
CM		 3015 }
3016 }
fce3bb9a
LD		 3017 if (rw & REQ_DISCARD) {
3018 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3019 BTRFS_BLOCK_GROUP_RAID1 |
3020 BTRFS_BLOCK_GROUP_DUP |
3021 BTRFS_BLOCK_GROUP_RAID10)) {
3022 stripes_required = map->num_stripes;
3023 }
3024 }
3025 if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
321aecc6 3026 stripes_allocated < stripes_required) {
cea9e445 3027 stripes_allocated = map->num_stripes;
cea9e445
CM		 3028 free_extent_map(em);
3029 kfree(multi);
3030 goto again;
3031 }
593060d7
CM		 3032 stripe_nr = offset;
3033 /*
3034 * stripe_nr counts the total number of stripes we have to stride
3035 * to get to this block
3036 */
3037 do_div(stripe_nr, map->stripe_len);
3038
3039 stripe_offset = stripe_nr * map->stripe_len;
3040 BUG_ON(offset < stripe_offset);
3041
3042 /* stripe_offset is the offset of this block in its stripe*/
3043 stripe_offset = offset - stripe_offset;
3044
fce3bb9a
LD		 3045 if (rw & REQ_DISCARD)
3046 *length = min_t(u64, em->len - offset, *length);
3047 else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3048 BTRFS_BLOCK_GROUP_RAID1 |
3049 BTRFS_BLOCK_GROUP_RAID10 |
3050 BTRFS_BLOCK_GROUP_DUP)) {
cea9e445
CM		 3051 /* we limit the length of each bio to what fits in a stripe */
3052 *length = min_t(u64, em->len - offset,
fce3bb9a 3053 map->stripe_len - stripe_offset);
cea9e445
CM		 3054 } else {
3055 *length = em->len - offset;
3056 }
f2d8d74d
CM		 3057
3058 if (!multi_ret && !unplug_page)
cea9e445
CM		 3059 goto out;
3060
f2d8d74d 3061 num_stripes = 1;
cea9e445 3062 stripe_index = 0;
fce3bb9a
LD		 3063 stripe_nr_orig = stripe_nr;
3064 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3065 (~(map->stripe_len - 1));
3066 do_div(stripe_nr_end, map->stripe_len);
3067 stripe_end_offset = stripe_nr_end * map->stripe_len -
3068 (offset + *length);
3069 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3070 if (rw & REQ_DISCARD)
3071 num_stripes = min_t(u64, map->num_stripes,
3072 stripe_nr_end - stripe_nr_orig);
3073 stripe_index = do_div(stripe_nr, map->num_stripes);
3074 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3075 if (unplug_page || (rw & (REQ_WRITE | REQ_DISCARD)))
f2d8d74d 3076 num_stripes = map->num_stripes;
2fff734f 3077 else if (mirror_num)
f188591e 3078 stripe_index = mirror_num - 1;
dfe25020
CM		 3079 else {
3080 stripe_index = find_live_mirror(map, 0,
3081 map->num_stripes,
3082 current->pid % map->num_stripes);
3083 }
2fff734f 3084
611f0e00 3085 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
fce3bb9a 3086 if (rw & (REQ_WRITE | REQ_DISCARD))
f2d8d74d 3087 num_stripes = map->num_stripes;
f188591e
CM		 3088 else if (mirror_num)
3089 stripe_index = mirror_num - 1;
2fff734f 3090
321aecc6
CM		 3091 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3092 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 3093
3094 stripe_index = do_div(stripe_nr, factor);
3095 stripe_index *= map->sub_stripes;
3096
7b6d91da 3097 if (unplug_page || (rw & REQ_WRITE))
f2d8d74d 3098 num_stripes = map->sub_stripes;
fce3bb9a
LD		 3099 else if (rw & REQ_DISCARD)
3100 num_stripes = min_t(u64, map->sub_stripes *
3101 (stripe_nr_end - stripe_nr_orig),
3102 map->num_stripes);
321aecc6
CM		 3103 else if (mirror_num)
3104 stripe_index += mirror_num - 1;
dfe25020
CM		 3105 else {
3106 stripe_index = find_live_mirror(map, stripe_index,
3107 map->sub_stripes, stripe_index +
3108 current->pid % map->sub_stripes);
3109 }
8790d502
CM		 3110 } else {
3111 /*
3112 * after this do_div call, stripe_nr is the number of stripes
3113 * on this device we have to walk to find the data, and
3114 * stripe_index is the number of our device in the stripe array
3115 */
3116 stripe_index = do_div(stripe_nr, map->num_stripes);
3117 }
593060d7 3118 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 3119
fce3bb9a
LD		 3120 if (rw & REQ_DISCARD) {
3121 for (i = 0; i < num_stripes; i++) {
f2d8d74d
CM		 3122 multi->stripes[i].physical =
3123 map->stripes[stripe_index].physical +
3124 stripe_offset + stripe_nr * map->stripe_len;
3125 multi->stripes[i].dev = map->stripes[stripe_index].dev;
fce3bb9a
LD		 3126
3127 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3128 u64 stripes;
3129 int last_stripe = (stripe_nr_end - 1) %
3130 map->num_stripes;
3131 int j;
3132
3133 for (j = 0; j < map->num_stripes; j++) {
3134 if ((stripe_nr_end - 1 - j) %
3135 map->num_stripes == stripe_index)
3136 break;
3137 }
3138 stripes = stripe_nr_end - 1 - j;
3139 do_div(stripes, map->num_stripes);
3140 multi->stripes[i].length = map->stripe_len *
3141 (stripes - stripe_nr + 1);
3142
3143 if (i == 0) {
3144 multi->stripes[i].length -=
3145 stripe_offset;
3146 stripe_offset = 0;
3147 }
3148 if (stripe_index == last_stripe)
3149 multi->stripes[i].length -=
3150 stripe_end_offset;
3151 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3152 u64 stripes;
3153 int j;
3154 int factor = map->num_stripes /
3155 map->sub_stripes;
3156 int last_stripe = (stripe_nr_end - 1) % factor;
3157 last_stripe *= map->sub_stripes;
3158
3159 for (j = 0; j < factor; j++) {
3160 if ((stripe_nr_end - 1 - j) % factor ==
3161 stripe_index / map->sub_stripes)
3162 break;
3163 }
3164 stripes = stripe_nr_end - 1 - j;
3165 do_div(stripes, factor);
3166 multi->stripes[i].length = map->stripe_len *
3167 (stripes - stripe_nr + 1);
3168
3169 if (i < map->sub_stripes) {
3170 multi->stripes[i].length -=
3171 stripe_offset;
3172 if (i == map->sub_stripes - 1)
3173 stripe_offset = 0;
3174 }
3175 if (stripe_index >= last_stripe &&
3176 stripe_index <= (last_stripe +
3177 map->sub_stripes - 1)) {
3178 multi->stripes[i].length -=
3179 stripe_end_offset;
3180 }
3181 } else
3182 multi->stripes[i].length = *length;
3183
3184 stripe_index++;
3185 if (stripe_index == map->num_stripes) {
3186 /* This could only happen for RAID0/10 */
3187 stripe_index = 0;
3188 stripe_nr++;
3189 }
3190 }
3191 } else {
3192 for (i = 0; i < num_stripes; i++) {
3193 if (unplug_page) {
3194 struct btrfs_device *device;
3195 struct backing_dev_info *bdi;
3196
3197 device = map->stripes[stripe_index].dev;
3198 if (device->bdev) {
3199 bdi = blk_get_backing_dev_info(device->
3200 bdev);
3201 if (bdi->unplug_io_fn)
3202 bdi->unplug_io_fn(bdi,
3203 unplug_page);
3204 }
3205 } else {
3206 multi->stripes[i].physical =
3207 map->stripes[stripe_index].physical +
3208 stripe_offset +
3209 stripe_nr * map->stripe_len;
3210 multi->stripes[i].dev =
3211 map->stripes[stripe_index].dev;
3212 }
3213 stripe_index++;
f2d8d74d 3214 }
593060d7 3215 }
f2d8d74d
CM		 3216 if (multi_ret) {
3217 *multi_ret = multi;
3218 multi->num_stripes = num_stripes;
a236aed1 3219 multi->max_errors = max_errors;
f2d8d74d 3220 }
cea9e445 3221out:
0b86a832 3222 free_extent_map(em);
0b86a832
CM		 3223 return 0;
3224}
3225
f2d8d74d
CM		 3226int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3227 u64 logical, u64 *length,
3228 struct btrfs_multi_bio **multi_ret, int mirror_num)
3229{
3230 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3231 mirror_num, NULL);
3232}
3233
a512bbf8
YZ		 3234int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3235 u64 chunk_start, u64 physical, u64 devid,
3236 u64 **logical, int *naddrs, int *stripe_len)
3237{
3238 struct extent_map_tree *em_tree = &map_tree->map_tree;
3239 struct extent_map *em;
3240 struct map_lookup *map;
3241 u64 *buf;
3242 u64 bytenr;
3243 u64 length;
3244 u64 stripe_nr;
3245 int i, j, nr = 0;
3246
890871be 3247 read_lock(&em_tree->lock);
a512bbf8 3248 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 3249 read_unlock(&em_tree->lock);
a512bbf8
YZ		 3250
3251 BUG_ON(!em || em->start != chunk_start);
3252 map = (struct map_lookup *)em->bdev;
3253
3254 length = em->len;
3255 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3256 do_div(length, map->num_stripes / map->sub_stripes);
3257 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3258 do_div(length, map->num_stripes);
3259
3260 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3261 BUG_ON(!buf);
3262
3263 for (i = 0; i < map->num_stripes; i++) {
3264 if (devid && map->stripes[i].dev->devid != devid)
3265 continue;
3266 if (map->stripes[i].physical > physical ||
3267 map->stripes[i].physical + length <= physical)
3268 continue;
3269
3270 stripe_nr = physical - map->stripes[i].physical;
3271 do_div(stripe_nr, map->stripe_len);
3272
3273 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3274 stripe_nr = stripe_nr * map->num_stripes + i;
3275 do_div(stripe_nr, map->sub_stripes);
3276 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3277 stripe_nr = stripe_nr * map->num_stripes + i;
3278 }
3279 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 3280 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 3281 for (j = 0; j < nr; j++) {
3282 if (buf[j] == bytenr)
3283 break;
3284 }
934d375b
CM		 3285 if (j == nr) {
3286 WARN_ON(nr >= map->num_stripes);
a512bbf8 3287 buf[nr++] = bytenr;
934d375b 3288 }
a512bbf8
YZ		 3289 }
3290
a512bbf8
YZ		 3291 *logical = buf;
3292 *naddrs = nr;
3293 *stripe_len = map->stripe_len;
3294
3295 free_extent_map(em);
3296 return 0;
3297}
3298
f2d8d74d
CM		 3299int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3300 u64 logical, struct page *page)
3301{
3302 u64 length = PAGE_CACHE_SIZE;
3303 return __btrfs_map_block(map_tree, READ, logical, &length,
3304 NULL, 0, page);
3305}
3306
8790d502 3307static void end_bio_multi_stripe(struct bio *bio, int err)
8790d502 3308{
cea9e445 3309 struct btrfs_multi_bio *multi = bio->bi_private;
7d2b4daa 3310 int is_orig_bio = 0;
8790d502 3311
8790d502 3312 if (err)
a236aed1 3313 atomic_inc(&multi->error);
8790d502 3314
7d2b4daa
CM		 3315 if (bio == multi->orig_bio)
3316 is_orig_bio = 1;
3317
cea9e445 3318 if (atomic_dec_and_test(&multi->stripes_pending)) {
7d2b4daa
CM		 3319 if (!is_orig_bio) {
3320 bio_put(bio);
3321 bio = multi->orig_bio;
3322 }
8790d502
CM		 3323 bio->bi_private = multi->private;
3324 bio->bi_end_io = multi->end_io;
a236aed1
CM		 3325 /* only send an error to the higher layers if it is
3326 * beyond the tolerance of the multi-bio
3327 */
1259ab75 3328 if (atomic_read(&multi->error) > multi->max_errors) {
a236aed1 3329 err = -EIO;
1259ab75
CM		 3330 } else if (err) {
3331 /*
3332 * this bio is actually up to date, we didn't
3333 * go over the max number of errors
3334 */
3335 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 3336 err = 0;
1259ab75 3337 }
8790d502
CM		 3338 kfree(multi);
3339
3340 bio_endio(bio, err);
7d2b4daa 3341 } else if (!is_orig_bio) {
8790d502
CM		 3342 bio_put(bio);
3343 }
8790d502
CM		 3344}
3345
8b712842
CM		 3346struct async_sched {
3347 struct bio *bio;
3348 int rw;
3349 struct btrfs_fs_info *info;
3350 struct btrfs_work work;
3351};
3352
3353/*
3354 * see run_scheduled_bios for a description of why bios are collected for
3355 * async submit.
3356 *
3357 * This will add one bio to the pending list for a device and make sure
3358 * the work struct is scheduled.
3359 */
d397712b 3360static noinline int schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 3361 struct btrfs_device *device,
3362 int rw, struct bio *bio)
8b712842
CM		 3363{
3364 int should_queue = 1;
ffbd517d 3365 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 3366
3367 /* don't bother with additional async steps for reads, right now */
7b6d91da 3368 if (!(rw & REQ_WRITE)) {
492bb6de 3369 bio_get(bio);
8b712842 3370 submit_bio(rw, bio);
492bb6de 3371 bio_put(bio);
8b712842
CM		 3372 return 0;
3373 }
3374
3375 /*
0986fe9e 3376 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 3377 * higher layers. Otherwise, the async bio makes it appear we have
3378 * made progress against dirty pages when we've really just put it
3379 * on a queue for later
3380 */
0986fe9e 3381 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 3382 WARN_ON(bio->bi_next);
8b712842
CM		 3383 bio->bi_next = NULL;
3384 bio->bi_rw |= rw;
3385
3386 spin_lock(&device->io_lock);
7b6d91da 3387 if (bio->bi_rw & REQ_SYNC)
ffbd517d
CM		 3388 pending_bios = &device->pending_sync_bios;
3389 else
3390 pending_bios = &device->pending_bios;
8b712842 3391
ffbd517d
CM		 3392 if (pending_bios->tail)
3393 pending_bios->tail->bi_next = bio;
8b712842 3394
ffbd517d
CM		 3395 pending_bios->tail = bio;
3396 if (!pending_bios->head)
3397 pending_bios->head = bio;
8b712842
CM		 3398 if (device->running_pending)
3399 should_queue = 0;
3400
3401 spin_unlock(&device->io_lock);
3402
3403 if (should_queue)
1cc127b5
CM		 3404 btrfs_queue_worker(&root->fs_info->submit_workers,
3405 &device->work);
8b712842
CM		 3406 return 0;
3407}
3408
f188591e 3409int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 3410 int mirror_num, int async_submit)
0b86a832
CM		 3411{
3412 struct btrfs_mapping_tree *map_tree;
3413 struct btrfs_device *dev;
8790d502 3414 struct bio *first_bio = bio;
a62b9401 3415 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 3416 u64 length = 0;
3417 u64 map_length;
cea9e445 3418 struct btrfs_multi_bio *multi = NULL;
0b86a832 3419 int ret;
8790d502
CM		 3420 int dev_nr = 0;
3421 int total_devs = 1;
0b86a832 3422
f2d8d74d 3423 length = bio->bi_size;
0b86a832
CM		 3424 map_tree = &root->fs_info->mapping_tree;
3425 map_length = length;
cea9e445 3426
f188591e
CM		 3427 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3428 mirror_num);
cea9e445
CM		 3429 BUG_ON(ret);
3430
3431 total_devs = multi->num_stripes;
3432 if (map_length < length) {
d397712b
CM		 3433 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3434 "len %llu\n", (unsigned long long)logical,
3435 (unsigned long long)length,
3436 (unsigned long long)map_length);
cea9e445
CM		 3437 BUG();
3438 }
3439 multi->end_io = first_bio->bi_end_io;
3440 multi->private = first_bio->bi_private;
7d2b4daa 3441 multi->orig_bio = first_bio;
cea9e445
CM		 3442 atomic_set(&multi->stripes_pending, multi->num_stripes);
3443
d397712b 3444 while (dev_nr < total_devs) {
8790d502 3445 if (total_devs > 1) {
8790d502
CM		 3446 if (dev_nr < total_devs - 1) {
3447 bio = bio_clone(first_bio, GFP_NOFS);
3448 BUG_ON(!bio);
3449 } else {
3450 bio = first_bio;
3451 }
3452 bio->bi_private = multi;
3453 bio->bi_end_io = end_bio_multi_stripe;
3454 }
cea9e445
CM		 3455 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3456 dev = multi->stripes[dev_nr].dev;
18e503d6 3457 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
dfe25020 3458 bio->bi_bdev = dev->bdev;
8b712842
CM		 3459 if (async_submit)
3460 schedule_bio(root, dev, rw, bio);
3461 else
3462 submit_bio(rw, bio);
dfe25020
CM		 3463 } else {
3464 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3465 bio->bi_sector = logical >> 9;
dfe25020 3466 bio_endio(bio, -EIO);
dfe25020 3467 }
8790d502
CM		 3468 dev_nr++;
3469 }
cea9e445
CM		 3470 if (total_devs == 1)
3471 kfree(multi);
0b86a832
CM		 3472 return 0;
3473}
3474
a443755f 3475struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2b82032c 3476 u8 *uuid, u8 *fsid)
0b86a832 3477{
2b82032c
YZ		 3478 struct btrfs_device *device;
3479 struct btrfs_fs_devices *cur_devices;
3480
3481 cur_devices = root->fs_info->fs_devices;
3482 while (cur_devices) {
3483 if (!fsid ||
3484 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3485 device = __find_device(&cur_devices->devices,
3486 devid, uuid);
3487 if (device)
3488 return device;
3489 }
3490 cur_devices = cur_devices->seed;
3491 }
3492 return NULL;
0b86a832
CM		 3493}
3494
dfe25020
CM		 3495static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3496 u64 devid, u8 *dev_uuid)
3497{
3498 struct btrfs_device *device;
3499 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3500
3501 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 3502 if (!device)
3503 return NULL;
dfe25020
CM		 3504 list_add(&device->dev_list,
3505 &fs_devices->devices);
dfe25020
CM		 3506 device->dev_root = root->fs_info->dev_root;
3507 device->devid = devid;
8b712842 3508 device->work.func = pending_bios_fn;
e4404d6e 3509 device->fs_devices = fs_devices;
cd02dca5 3510 device->missing = 1;
dfe25020 3511 fs_devices->num_devices++;
cd02dca5 3512 fs_devices->missing_devices++;
dfe25020 3513 spin_lock_init(&device->io_lock);
d20f7043 3514 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 3515 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3516 return device;
3517}
3518
0b86a832
CM		 3519static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3520 struct extent_buffer *leaf,
3521 struct btrfs_chunk *chunk)
3522{
3523 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3524 struct map_lookup *map;
3525 struct extent_map *em;
3526 u64 logical;
3527 u64 length;
3528 u64 devid;
a443755f 3529 u8 uuid[BTRFS_UUID_SIZE];
593060d7 3530 int num_stripes;
0b86a832 3531 int ret;
593060d7 3532 int i;
0b86a832 3533
e17cade2
CM		 3534 logical = key->offset;
3535 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 3536
890871be 3537 read_lock(&map_tree->map_tree.lock);
0b86a832 3538 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 3539 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 3540
3541 /* already mapped? */
3542 if (em && em->start <= logical && em->start + em->len > logical) {
3543 free_extent_map(em);
0b86a832
CM		 3544 return 0;
3545 } else if (em) {
3546 free_extent_map(em);
3547 }
0b86a832 3548
0b86a832
CM		 3549 em = alloc_extent_map(GFP_NOFS);
3550 if (!em)
3551 return -ENOMEM;
593060d7
CM		 3552 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3553 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 3554 if (!map) {
3555 free_extent_map(em);
3556 return -ENOMEM;
3557 }
3558
3559 em->bdev = (struct block_device *)map;
3560 em->start = logical;
3561 em->len = length;
3562 em->block_start = 0;
c8b97818 3563 em->block_len = em->len;
0b86a832 3564
593060d7
CM		 3565 map->num_stripes = num_stripes;
3566 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3567 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3568 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3569 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3570 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 3571 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 3572 for (i = 0; i < num_stripes; i++) {
3573 map->stripes[i].physical =
3574 btrfs_stripe_offset_nr(leaf, chunk, i);
3575 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 3576 read_extent_buffer(leaf, uuid, (unsigned long)
3577 btrfs_stripe_dev_uuid_nr(chunk, i),
3578 BTRFS_UUID_SIZE);
2b82032c
YZ		 3579 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3580 NULL);
dfe25020 3581 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 3582 kfree(map);
3583 free_extent_map(em);
3584 return -EIO;
3585 }
dfe25020
CM		 3586 if (!map->stripes[i].dev) {
3587 map->stripes[i].dev =
3588 add_missing_dev(root, devid, uuid);
3589 if (!map->stripes[i].dev) {
3590 kfree(map);
3591 free_extent_map(em);
3592 return -EIO;
3593 }
3594 }
3595 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 3596 }
3597
890871be 3598 write_lock(&map_tree->map_tree.lock);
0b86a832 3599 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 3600 write_unlock(&map_tree->map_tree.lock);
b248a415 3601 BUG_ON(ret);
0b86a832
CM		 3602 free_extent_map(em);
3603
3604 return 0;
3605}
3606
3607static int fill_device_from_item(struct extent_buffer *leaf,
3608 struct btrfs_dev_item *dev_item,
3609 struct btrfs_device *device)
3610{
3611 unsigned long ptr;
0b86a832
CM		 3612
3613 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 3614 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3615 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 3616 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3617 device->type = btrfs_device_type(leaf, dev_item);
3618 device->io_align = btrfs_device_io_align(leaf, dev_item);
3619 device->io_width = btrfs_device_io_width(leaf, dev_item);
3620 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
0b86a832
CM		 3621
3622 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 3623 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832 3624
0b86a832
CM		 3625 return 0;
3626}
3627
2b82032c
YZ		 3628static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3629{
3630 struct btrfs_fs_devices *fs_devices;
3631 int ret;
3632
3633 mutex_lock(&uuid_mutex);
3634
3635 fs_devices = root->fs_info->fs_devices->seed;
3636 while (fs_devices) {
3637 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3638 ret = 0;
3639 goto out;
3640 }
3641 fs_devices = fs_devices->seed;
3642 }
3643
3644 fs_devices = find_fsid(fsid);
3645 if (!fs_devices) {
3646 ret = -ENOENT;
3647 goto out;
3648 }
e4404d6e
YZ		 3649
3650 fs_devices = clone_fs_devices(fs_devices);
3651 if (IS_ERR(fs_devices)) {
3652 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 3653 goto out;
3654 }
3655
97288f2c 3656 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 3657 root->fs_info->bdev_holder);
2b82032c
YZ		 3658 if (ret)
3659 goto out;
3660
3661 if (!fs_devices->seeding) {
3662 __btrfs_close_devices(fs_devices);
e4404d6e 3663 free_fs_devices(fs_devices);
2b82032c
YZ		 3664 ret = -EINVAL;
3665 goto out;
3666 }
3667
3668 fs_devices->seed = root->fs_info->fs_devices->seed;
3669 root->fs_info->fs_devices->seed = fs_devices;
2b82032c
YZ		 3670out:
3671 mutex_unlock(&uuid_mutex);
3672 return ret;
3673}
3674
0d81ba5d 3675static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 3676 struct extent_buffer *leaf,
3677 struct btrfs_dev_item *dev_item)
3678{
3679 struct btrfs_device *device;
3680 u64 devid;
3681 int ret;
2b82032c 3682 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 3683 u8 dev_uuid[BTRFS_UUID_SIZE];
3684
0b86a832 3685 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 3686 read_extent_buffer(leaf, dev_uuid,
3687 (unsigned long)btrfs_device_uuid(dev_item),
3688 BTRFS_UUID_SIZE);
2b82032c
YZ		 3689 read_extent_buffer(leaf, fs_uuid,
3690 (unsigned long)btrfs_device_fsid(dev_item),
3691 BTRFS_UUID_SIZE);
3692
3693 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3694 ret = open_seed_devices(root, fs_uuid);
e4404d6e 3695 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 3696 return ret;
2b82032c
YZ		 3697 }
3698
3699 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3700 if (!device || !device->bdev) {
e4404d6e 3701 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 3702 return -EIO;
3703
3704 if (!device) {
d397712b
CM		 3705 printk(KERN_WARNING "warning devid %llu missing\n",
3706 (unsigned long long)devid);
2b82032c
YZ		 3707 device = add_missing_dev(root, devid, dev_uuid);
3708 if (!device)
3709 return -ENOMEM;
cd02dca5
CM		 3710 } else if (!device->missing) {
3711 /*
3712 * this happens when a device that was properly setup
3713 * in the device info lists suddenly goes bad.
3714 * device->bdev is NULL, and so we have to set
3715 * device->missing to one here
3716 */
3717 root->fs_info->fs_devices->missing_devices++;
3718 device->missing = 1;
2b82032c
YZ		 3719 }
3720 }
3721
3722 if (device->fs_devices != root->fs_info->fs_devices) {
3723 BUG_ON(device->writeable);
3724 if (device->generation !=
3725 btrfs_device_generation(leaf, dev_item))
3726 return -EINVAL;
6324fbf3 3727 }
0b86a832
CM		 3728
3729 fill_device_from_item(leaf, dev_item, device);
3730 device->dev_root = root->fs_info->dev_root;
dfe25020 3731 device->in_fs_metadata = 1;
2b82032c
YZ		 3732 if (device->writeable)
3733 device->fs_devices->total_rw_bytes += device->total_bytes;
0b86a832 3734 ret = 0;
0b86a832
CM		 3735 return ret;
3736}
3737
0d81ba5d
CM		 3738int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3739{
3740 struct btrfs_dev_item *dev_item;
3741
3742 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3743 dev_item);
3744 return read_one_dev(root, buf, dev_item);
3745}
3746
e4404d6e 3747int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832
CM		 3748{
3749 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
a061fc8d 3750 struct extent_buffer *sb;
0b86a832 3751 struct btrfs_disk_key *disk_key;
0b86a832 3752 struct btrfs_chunk *chunk;
84eed90f
CM		 3753 u8 *ptr;
3754 unsigned long sb_ptr;
3755 int ret = 0;
0b86a832
CM		 3756 u32 num_stripes;
3757 u32 array_size;
3758 u32 len = 0;
0b86a832 3759 u32 cur;
84eed90f 3760 struct btrfs_key key;
0b86a832 3761
e4404d6e 3762 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 3763 BTRFS_SUPER_INFO_SIZE);
3764 if (!sb)
3765 return -ENOMEM;
3766 btrfs_set_buffer_uptodate(sb);
4008c04a
CM		 3767 btrfs_set_buffer_lockdep_class(sb, 0);
3768
a061fc8d 3769 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 3770 array_size = btrfs_super_sys_array_size(super_copy);
3771
0b86a832
CM		 3772 ptr = super_copy->sys_chunk_array;
3773 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3774 cur = 0;
3775
3776 while (cur < array_size) {
3777 disk_key = (struct btrfs_disk_key *)ptr;
3778 btrfs_disk_key_to_cpu(&key, disk_key);
3779
a061fc8d 3780 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 3781 sb_ptr += len;
3782 cur += len;
3783
0d81ba5d 3784 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 3785 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 3786 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 3787 if (ret)
3788 break;
0b86a832
CM		 3789 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3790 len = btrfs_chunk_item_size(num_stripes);
3791 } else {
84eed90f
CM		 3792 ret = -EIO;
3793 break;
0b86a832
CM		 3794 }
3795 ptr += len;
3796 sb_ptr += len;
3797 cur += len;
3798 }
a061fc8d 3799 free_extent_buffer(sb);
84eed90f 3800 return ret;
0b86a832
CM		 3801}
3802
3803int btrfs_read_chunk_tree(struct btrfs_root *root)
3804{
3805 struct btrfs_path *path;
3806 struct extent_buffer *leaf;
3807 struct btrfs_key key;
3808 struct btrfs_key found_key;
3809 int ret;
3810 int slot;
3811
3812 root = root->fs_info->chunk_root;
3813
3814 path = btrfs_alloc_path();
3815 if (!path)
3816 return -ENOMEM;
3817
3818 /* first we search for all of the device items, and then we
3819 * read in all of the chunk items. This way we can create chunk
3820 * mappings that reference all of the devices that are afound
3821 */
3822 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3823 key.offset = 0;
3824 key.type = 0;
3825again:
3826 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 3827 if (ret < 0)
3828 goto error;
d397712b 3829 while (1) {
0b86a832
CM		 3830 leaf = path->nodes[0];
3831 slot = path->slots[0];
3832 if (slot >= btrfs_header_nritems(leaf)) {
3833 ret = btrfs_next_leaf(root, path);
3834 if (ret == 0)
3835 continue;
3836 if (ret < 0)
3837 goto error;
3838 break;
3839 }
3840 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3841 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3842 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3843 break;
3844 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3845 struct btrfs_dev_item *dev_item;
3846 dev_item = btrfs_item_ptr(leaf, slot,
3847 struct btrfs_dev_item);
0d81ba5d 3848 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 3849 if (ret)
3850 goto error;
0b86a832
CM		 3851 }
3852 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3853 struct btrfs_chunk *chunk;
3854 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3855 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 3856 if (ret)
3857 goto error;
0b86a832
CM		 3858 }
3859 path->slots[0]++;
3860 }
3861 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3862 key.objectid = 0;
3863 btrfs_release_path(root, path);
3864 goto again;
3865 }
0b86a832
CM		 3866 ret = 0;
3867error:
2b82032c 3868 btrfs_free_path(path);
0b86a832
CM		 3869 return ret;
3870}
Linux 6.x block layer and io_uring tree(s)