projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: remove unnecessary DEFINE_WAIT() declarations
[linux-2.6-block.git] / fs / btrfs / volumes.c
CommitLineData
0b86a832
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
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>
442a4f63 26#include <linux/ratelimit.h>
59641015 27#include <linux/kthread.h>
4b4e25f2 28#include "compat.h"
0b86a832
CM		 29#include "ctree.h"
30#include "extent_map.h"
31#include "disk-io.h"
32#include "transaction.h"
33#include "print-tree.h"
34#include "volumes.h"
8b712842 35#include "async-thread.h"
21adbd5c 36#include "check-integrity.h"
606686ee 37#include "rcu-string.h"
3fed40cc 38#include "math.h"
8dabb742 39#include "dev-replace.h"
0b86a832 40
2b82032c
YZ		 41static int init_first_rw_device(struct btrfs_trans_handle *trans,
42 struct btrfs_root *root,
43 struct btrfs_device *device);
44static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
733f4fbb
SB		 45static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
46static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
2b82032c 47
8a4b83cc
CM		 48static DEFINE_MUTEX(uuid_mutex);
49static LIST_HEAD(fs_uuids);
50
7d9eb12c
CM		 51static void lock_chunks(struct btrfs_root *root)
52{
7d9eb12c
CM		 53 mutex_lock(&root->fs_info->chunk_mutex);
54}
55
56static void unlock_chunks(struct btrfs_root *root)
57{
7d9eb12c
CM		 58 mutex_unlock(&root->fs_info->chunk_mutex);
59}
60
e4404d6e
YZ		 61static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62{
63 struct btrfs_device *device;
64 WARN_ON(fs_devices->opened);
65 while (!list_empty(&fs_devices->devices)) {
66 device = list_entry(fs_devices->devices.next,
67 struct btrfs_device, dev_list);
68 list_del(&device->dev_list);
606686ee 69 rcu_string_free(device->name);
e4404d6e
YZ		 70 kfree(device);
71 }
72 kfree(fs_devices);
73}
74
b8b8ff59
LC		 75static void btrfs_kobject_uevent(struct block_device *bdev,
76 enum kobject_action action)
77{
78 int ret;
79
80 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
81 if (ret)
82 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
83 action,
84 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
85 &disk_to_dev(bdev->bd_disk)->kobj);
86}
87
143bede5 88void btrfs_cleanup_fs_uuids(void)
8a4b83cc
CM		 89{
90 struct btrfs_fs_devices *fs_devices;
8a4b83cc 91
2b82032c
YZ		 92 while (!list_empty(&fs_uuids)) {
93 fs_devices = list_entry(fs_uuids.next,
94 struct btrfs_fs_devices, list);
95 list_del(&fs_devices->list);
e4404d6e 96 free_fs_devices(fs_devices);
8a4b83cc 97 }
8a4b83cc
CM		 98}
99
a1b32a59
CM		 100static noinline struct btrfs_device *__find_device(struct list_head *head,
101 u64 devid, u8 *uuid)
8a4b83cc
CM		 102{
103 struct btrfs_device *dev;
8a4b83cc 104
c6e30871 105 list_for_each_entry(dev, head, dev_list) {
a443755f 106 if (dev->devid == devid &&
8f18cf13 107 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 108 return dev;
a443755f 109 }
8a4b83cc
CM		 110 }
111 return NULL;
112}
113
a1b32a59 114static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 115{
8a4b83cc
CM		 116 struct btrfs_fs_devices *fs_devices;
117
c6e30871 118 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 119 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
120 return fs_devices;
121 }
122 return NULL;
123}
124
beaf8ab3
SB		 125static int
126btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
127 int flush, struct block_device **bdev,
128 struct buffer_head **bh)
129{
130 int ret;
131
132 *bdev = blkdev_get_by_path(device_path, flags, holder);
133
134 if (IS_ERR(*bdev)) {
135 ret = PTR_ERR(*bdev);
136 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
137 goto error;
138 }
139
140 if (flush)
141 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
142 ret = set_blocksize(*bdev, 4096);
143 if (ret) {
144 blkdev_put(*bdev, flags);
145 goto error;
146 }
147 invalidate_bdev(*bdev);
148 *bh = btrfs_read_dev_super(*bdev);
149 if (!*bh) {
150 ret = -EINVAL;
151 blkdev_put(*bdev, flags);
152 goto error;
153 }
154
155 return 0;
156
157error:
158 *bdev = NULL;
159 *bh = NULL;
160 return ret;
161}
162
ffbd517d
CM		 163static void requeue_list(struct btrfs_pending_bios *pending_bios,
164 struct bio *head, struct bio *tail)
165{
166
167 struct bio *old_head;
168
169 old_head = pending_bios->head;
170 pending_bios->head = head;
171 if (pending_bios->tail)
172 tail->bi_next = old_head;
173 else
174 pending_bios->tail = tail;
175}
176
8b712842
CM		 177/*
178 * we try to collect pending bios for a device so we don't get a large
179 * number of procs sending bios down to the same device. This greatly
180 * improves the schedulers ability to collect and merge the bios.
181 *
182 * But, it also turns into a long list of bios to process and that is sure
183 * to eventually make the worker thread block. The solution here is to
184 * make some progress and then put this work struct back at the end of
185 * the list if the block device is congested. This way, multiple devices
186 * can make progress from a single worker thread.
187 */
143bede5 188static noinline void run_scheduled_bios(struct btrfs_device *device)
8b712842
CM		 189{
190 struct bio *pending;
191 struct backing_dev_info *bdi;
b64a2851 192 struct btrfs_fs_info *fs_info;
ffbd517d 193 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 194 struct bio *tail;
195 struct bio *cur;
196 int again = 0;
ffbd517d 197 unsigned long num_run;
d644d8a1 198 unsigned long batch_run = 0;
b64a2851 199 unsigned long limit;
b765ead5 200 unsigned long last_waited = 0;
d84275c9 201 int force_reg = 0;
0e588859 202 int sync_pending = 0;
211588ad
CM		 203 struct blk_plug plug;
204
205 /*
206 * this function runs all the bios we've collected for
207 * a particular device. We don't want to wander off to
208 * another device without first sending all of these down.
209 * So, setup a plug here and finish it off before we return
210 */
211 blk_start_plug(&plug);
8b712842 212
bedf762b 213 bdi = blk_get_backing_dev_info(device->bdev);
b64a2851
CM		 214 fs_info = device->dev_root->fs_info;
215 limit = btrfs_async_submit_limit(fs_info);
216 limit = limit * 2 / 3;
217
8b712842
CM		 218loop:
219 spin_lock(&device->io_lock);
220
a6837051 221loop_lock:
d84275c9 222 num_run = 0;
ffbd517d 223
8b712842
CM		 224 /* take all the bios off the list at once and process them
225 * later on (without the lock held). But, remember the
226 * tail and other pointers so the bios can be properly reinserted
227 * into the list if we hit congestion
228 */
d84275c9 229 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 230 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 231 force_reg = 1;
232 } else {
ffbd517d 233 pending_bios = &device->pending_bios;
d84275c9
CM		 234 force_reg = 0;
235 }
ffbd517d
CM		 236
237 pending = pending_bios->head;
238 tail = pending_bios->tail;
8b712842 239 WARN_ON(pending && !tail);
8b712842
CM		 240
241 /*
242 * if pending was null this time around, no bios need processing
243 * at all and we can stop. Otherwise it'll loop back up again
244 * and do an additional check so no bios are missed.
245 *
246 * device->running_pending is used to synchronize with the
247 * schedule_bio code.
248 */
ffbd517d
CM		 249 if (device->pending_sync_bios.head == NULL &&
250 device->pending_bios.head == NULL) {
8b712842
CM		 251 again = 0;
252 device->running_pending = 0;
ffbd517d
CM		 253 } else {
254 again = 1;
255 device->running_pending = 1;
8b712842 256 }
ffbd517d
CM		 257
258 pending_bios->head = NULL;
259 pending_bios->tail = NULL;
260
8b712842
CM		 261 spin_unlock(&device->io_lock);
262
d397712b 263 while (pending) {
ffbd517d
CM		 264
265 rmb();
d84275c9
CM		 266 /* we want to work on both lists, but do more bios on the
267 * sync list than the regular list
268 */
269 if ((num_run > 32 &&
270 pending_bios != &device->pending_sync_bios &&
271 device->pending_sync_bios.head) ||
272 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
273 device->pending_bios.head)) {
ffbd517d
CM		 274 spin_lock(&device->io_lock);
275 requeue_list(pending_bios, pending, tail);
276 goto loop_lock;
277 }
278
8b712842
CM		 279 cur = pending;
280 pending = pending->bi_next;
281 cur->bi_next = NULL;
b64a2851 282
66657b31 283 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
b64a2851
CM		 284 waitqueue_active(&fs_info->async_submit_wait))
285 wake_up(&fs_info->async_submit_wait);
492bb6de
CM		 286
287 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
d644d8a1 288
2ab1ba68
CM		 289 /*
290 * if we're doing the sync list, record that our
291 * plug has some sync requests on it
292 *
293 * If we're doing the regular list and there are
294 * sync requests sitting around, unplug before
295 * we add more
296 */
297 if (pending_bios == &device->pending_sync_bios) {
298 sync_pending = 1;
299 } else if (sync_pending) {
300 blk_finish_plug(&plug);
301 blk_start_plug(&plug);
302 sync_pending = 0;
303 }
304
21adbd5c 305 btrfsic_submit_bio(cur->bi_rw, cur);
5ff7ba3a
CM		 306 num_run++;
307 batch_run++;
7eaceacc 308 if (need_resched())
ffbd517d 309 cond_resched();
8b712842
CM		 310
311 /*
312 * we made progress, there is more work to do and the bdi
313 * is now congested. Back off and let other work structs
314 * run instead
315 */
57fd5a5f 316 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 317 fs_info->fs_devices->open_devices > 1) {
b765ead5 318 struct io_context *ioc;
8b712842 319
b765ead5
CM		 320 ioc = current->io_context;
321
322 /*
323 * the main goal here is that we don't want to
324 * block if we're going to be able to submit
325 * more requests without blocking.
326 *
327 * This code does two great things, it pokes into
328 * the elevator code from a filesystem _and_
329 * it makes assumptions about how batching works.
330 */
331 if (ioc && ioc->nr_batch_requests > 0 &&
332 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
333 (last_waited == 0 ||
334 ioc->last_waited == last_waited)) {
335 /*
336 * we want to go through our batch of
337 * requests and stop. So, we copy out
338 * the ioc->last_waited time and test
339 * against it before looping
340 */
341 last_waited = ioc->last_waited;
7eaceacc 342 if (need_resched())
ffbd517d 343 cond_resched();
b765ead5
CM		 344 continue;
345 }
8b712842 346 spin_lock(&device->io_lock);
ffbd517d 347 requeue_list(pending_bios, pending, tail);
a6837051 348 device->running_pending = 1;
8b712842
CM		 349
350 spin_unlock(&device->io_lock);
351 btrfs_requeue_work(&device->work);
352 goto done;
353 }
d85c8a6f
CM		 354 /* unplug every 64 requests just for good measure */
355 if (batch_run % 64 == 0) {
356 blk_finish_plug(&plug);
357 blk_start_plug(&plug);
358 sync_pending = 0;
359 }
8b712842 360 }
ffbd517d 361
51684082
CM		 362 cond_resched();
363 if (again)
364 goto loop;
365
366 spin_lock(&device->io_lock);
367 if (device->pending_bios.head || device->pending_sync_bios.head)
368 goto loop_lock;
369 spin_unlock(&device->io_lock);
370
8b712842 371done:
211588ad 372 blk_finish_plug(&plug);
8b712842
CM		 373}
374
b2950863 375static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 376{
377 struct btrfs_device *device;
378
379 device = container_of(work, struct btrfs_device, work);
380 run_scheduled_bios(device);
381}
382
a1b32a59 383static noinline int device_list_add(const char *path,
8a4b83cc
CM		 384 struct btrfs_super_block *disk_super,
385 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
386{
387 struct btrfs_device *device;
388 struct btrfs_fs_devices *fs_devices;
606686ee 389 struct rcu_string *name;
8a4b83cc
CM		 390 u64 found_transid = btrfs_super_generation(disk_super);
391
392 fs_devices = find_fsid(disk_super->fsid);
393 if (!fs_devices) {
515dc322 394 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
8a4b83cc
CM		 395 if (!fs_devices)
396 return -ENOMEM;
397 INIT_LIST_HEAD(&fs_devices->devices);
b3075717 398 INIT_LIST_HEAD(&fs_devices->alloc_list);
8a4b83cc
CM		 399 list_add(&fs_devices->list, &fs_uuids);
400 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
401 fs_devices->latest_devid = devid;
402 fs_devices->latest_trans = found_transid;
e5e9a520 403 mutex_init(&fs_devices->device_list_mutex);
8a4b83cc
CM		 404 device = NULL;
405 } else {
a443755f
CM		 406 device = __find_device(&fs_devices->devices, devid,
407 disk_super->dev_item.uuid);
8a4b83cc
CM		 408 }
409 if (!device) {
2b82032c
YZ		 410 if (fs_devices->opened)
411 return -EBUSY;
412
8a4b83cc
CM		 413 device = kzalloc(sizeof(*device), GFP_NOFS);
414 if (!device) {
415 /* we can safely leave the fs_devices entry around */
416 return -ENOMEM;
417 }
418 device->devid = devid;
733f4fbb 419 device->dev_stats_valid = 0;
8b712842 420 device->work.func = pending_bios_fn;
a443755f
CM		 421 memcpy(device->uuid, disk_super->dev_item.uuid,
422 BTRFS_UUID_SIZE);
b248a415 423 spin_lock_init(&device->io_lock);
606686ee
JB		 424
425 name = rcu_string_strdup(path, GFP_NOFS);
426 if (!name) {
8a4b83cc
CM		 427 kfree(device);
428 return -ENOMEM;
429 }
606686ee 430 rcu_assign_pointer(device->name, name);
2b82032c 431 INIT_LIST_HEAD(&device->dev_alloc_list);
e5e9a520 432
90519d66
AJ		 433 /* init readahead state */
434 spin_lock_init(&device->reada_lock);
435 device->reada_curr_zone = NULL;
436 atomic_set(&device->reada_in_flight, 0);
437 device->reada_next = 0;
438 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
439 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
440
e5e9a520 441 mutex_lock(&fs_devices->device_list_mutex);
1f78160c 442 list_add_rcu(&device->dev_list, &fs_devices->devices);
e5e9a520
CM		 443 mutex_unlock(&fs_devices->device_list_mutex);
444
2b82032c 445 device->fs_devices = fs_devices;
8a4b83cc 446 fs_devices->num_devices++;
606686ee
JB		 447 } else if (!device->name || strcmp(device->name->str, path)) {
448 name = rcu_string_strdup(path, GFP_NOFS);
3a0524dc
TH		 449 if (!name)
450 return -ENOMEM;
606686ee
JB		 451 rcu_string_free(device->name);
452 rcu_assign_pointer(device->name, name);
cd02dca5
CM		 453 if (device->missing) {
454 fs_devices->missing_devices--;
455 device->missing = 0;
456 }
8a4b83cc
CM		 457 }
458
459 if (found_transid > fs_devices->latest_trans) {
460 fs_devices->latest_devid = devid;
461 fs_devices->latest_trans = found_transid;
462 }
8a4b83cc
CM		 463 *fs_devices_ret = fs_devices;
464 return 0;
465}
466
e4404d6e
YZ		 467static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
468{
469 struct btrfs_fs_devices *fs_devices;
470 struct btrfs_device *device;
471 struct btrfs_device *orig_dev;
472
473 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
474 if (!fs_devices)
475 return ERR_PTR(-ENOMEM);
476
477 INIT_LIST_HEAD(&fs_devices->devices);
478 INIT_LIST_HEAD(&fs_devices->alloc_list);
479 INIT_LIST_HEAD(&fs_devices->list);
e5e9a520 480 mutex_init(&fs_devices->device_list_mutex);
e4404d6e
YZ		 481 fs_devices->latest_devid = orig->latest_devid;
482 fs_devices->latest_trans = orig->latest_trans;
02db0844 483 fs_devices->total_devices = orig->total_devices;
e4404d6e
YZ		 484 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
485
46224705 486 /* We have held the volume lock, it is safe to get the devices. */
e4404d6e 487 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
606686ee
JB		 488 struct rcu_string *name;
489
e4404d6e
YZ		 490 device = kzalloc(sizeof(*device), GFP_NOFS);
491 if (!device)
492 goto error;
493
606686ee
JB		 494 /*
495 * This is ok to do without rcu read locked because we hold the
496 * uuid mutex so nothing we touch in here is going to disappear.
497 */
498 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
499 if (!name) {
fd2696f3 500 kfree(device);
e4404d6e 501 goto error;
fd2696f3 502 }
606686ee 503 rcu_assign_pointer(device->name, name);
e4404d6e
YZ		 504
505 device->devid = orig_dev->devid;
506 device->work.func = pending_bios_fn;
507 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
e4404d6e
YZ		 508 spin_lock_init(&device->io_lock);
509 INIT_LIST_HEAD(&device->dev_list);
510 INIT_LIST_HEAD(&device->dev_alloc_list);
511
512 list_add(&device->dev_list, &fs_devices->devices);
513 device->fs_devices = fs_devices;
514 fs_devices->num_devices++;
515 }
516 return fs_devices;
517error:
518 free_fs_devices(fs_devices);
519 return ERR_PTR(-ENOMEM);
520}
521
8dabb742
SB		 522void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
523 struct btrfs_fs_devices *fs_devices, int step)
dfe25020 524{
c6e30871 525 struct btrfs_device *device, *next;
dfe25020 526
a6b0d5c8
CM		 527 struct block_device *latest_bdev = NULL;
528 u64 latest_devid = 0;
529 u64 latest_transid = 0;
530
dfe25020
CM		 531 mutex_lock(&uuid_mutex);
532again:
46224705 533 /* This is the initialized path, it is safe to release the devices. */
c6e30871 534 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
a6b0d5c8 535 if (device->in_fs_metadata) {
63a212ab
SB		 536 if (!device->is_tgtdev_for_dev_replace &&
537 (!latest_transid ||
538 device->generation > latest_transid)) {
a6b0d5c8
CM		 539 latest_devid = device->devid;
540 latest_transid = device->generation;
541 latest_bdev = device->bdev;
542 }
2b82032c 543 continue;
a6b0d5c8 544 }
2b82032c 545
8dabb742
SB		 546 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
547 /*
548 * In the first step, keep the device which has
549 * the correct fsid and the devid that is used
550 * for the dev_replace procedure.
551 * In the second step, the dev_replace state is
552 * read from the device tree and it is known
553 * whether the procedure is really active or
554 * not, which means whether this device is
555 * used or whether it should be removed.
556 */
557 if (step == 0 || device->is_tgtdev_for_dev_replace) {
558 continue;
559 }
560 }
2b82032c 561 if (device->bdev) {
d4d77629 562 blkdev_put(device->bdev, device->mode);
2b82032c
YZ		 563 device->bdev = NULL;
564 fs_devices->open_devices--;
565 }
566 if (device->writeable) {
567 list_del_init(&device->dev_alloc_list);
568 device->writeable = 0;
8dabb742
SB		 569 if (!device->is_tgtdev_for_dev_replace)
570 fs_devices->rw_devices--;
2b82032c 571 }
e4404d6e
YZ		 572 list_del_init(&device->dev_list);
573 fs_devices->num_devices--;
606686ee 574 rcu_string_free(device->name);
e4404d6e 575 kfree(device);
dfe25020 576 }
2b82032c
YZ		 577
578 if (fs_devices->seed) {
579 fs_devices = fs_devices->seed;
2b82032c
YZ		 580 goto again;
581 }
582
a6b0d5c8
CM		 583 fs_devices->latest_bdev = latest_bdev;
584 fs_devices->latest_devid = latest_devid;
585 fs_devices->latest_trans = latest_transid;
586
dfe25020 587 mutex_unlock(&uuid_mutex);
dfe25020 588}
a0af469b 589
1f78160c
XG		 590static void __free_device(struct work_struct *work)
591{
592 struct btrfs_device *device;
593
594 device = container_of(work, struct btrfs_device, rcu_work);
595
596 if (device->bdev)
597 blkdev_put(device->bdev, device->mode);
598
606686ee 599 rcu_string_free(device->name);
1f78160c
XG		 600 kfree(device);
601}
602
603static void free_device(struct rcu_head *head)
604{
605 struct btrfs_device *device;
606
607 device = container_of(head, struct btrfs_device, rcu);
608
609 INIT_WORK(&device->rcu_work, __free_device);
610 schedule_work(&device->rcu_work);
611}
612
2b82032c 613static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 614{
8a4b83cc 615 struct btrfs_device *device;
e4404d6e 616
2b82032c
YZ		 617 if (--fs_devices->opened > 0)
618 return 0;
8a4b83cc 619
c9513edb 620 mutex_lock(&fs_devices->device_list_mutex);
c6e30871 621 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1f78160c 622 struct btrfs_device *new_device;
606686ee 623 struct rcu_string *name;
1f78160c
XG		 624
625 if (device->bdev)
a0af469b 626 fs_devices->open_devices--;
1f78160c 627
8dabb742 628 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
2b82032c
YZ		 629 list_del_init(&device->dev_alloc_list);
630 fs_devices->rw_devices--;
631 }
632
d5e2003c
JB		 633 if (device->can_discard)
634 fs_devices->num_can_discard--;
635
1f78160c 636 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
79787eaa 637 BUG_ON(!new_device); /* -ENOMEM */
1f78160c 638 memcpy(new_device, device, sizeof(*new_device));
606686ee
JB		 639
640 /* Safe because we are under uuid_mutex */
99f5944b
JB		 641 if (device->name) {
642 name = rcu_string_strdup(device->name->str, GFP_NOFS);
643 BUG_ON(device->name && !name); /* -ENOMEM */
644 rcu_assign_pointer(new_device->name, name);
645 }
1f78160c
XG		 646 new_device->bdev = NULL;
647 new_device->writeable = 0;
648 new_device->in_fs_metadata = 0;
d5e2003c 649 new_device->can_discard = 0;
1f78160c
XG		 650 list_replace_rcu(&device->dev_list, &new_device->dev_list);
651
652 call_rcu(&device->rcu, free_device);
8a4b83cc 653 }
c9513edb
XG		 654 mutex_unlock(&fs_devices->device_list_mutex);
655
e4404d6e
YZ		 656 WARN_ON(fs_devices->open_devices);
657 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 658 fs_devices->opened = 0;
659 fs_devices->seeding = 0;
2b82032c 660
8a4b83cc
CM		 661 return 0;
662}
663
2b82032c
YZ		 664int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
665{
e4404d6e 666 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 667 int ret;
668
669 mutex_lock(&uuid_mutex);
670 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 671 if (!fs_devices->opened) {
672 seed_devices = fs_devices->seed;
673 fs_devices->seed = NULL;
674 }
2b82032c 675 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 676
677 while (seed_devices) {
678 fs_devices = seed_devices;
679 seed_devices = fs_devices->seed;
680 __btrfs_close_devices(fs_devices);
681 free_fs_devices(fs_devices);
682 }
2b82032c
YZ		 683 return ret;
684}
685
e4404d6e
YZ		 686static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
687 fmode_t flags, void *holder)
8a4b83cc 688{
d5e2003c 689 struct request_queue *q;
8a4b83cc
CM		 690 struct block_device *bdev;
691 struct list_head *head = &fs_devices->devices;
8a4b83cc 692 struct btrfs_device *device;
a0af469b
CM		 693 struct block_device *latest_bdev = NULL;
694 struct buffer_head *bh;
695 struct btrfs_super_block *disk_super;
696 u64 latest_devid = 0;
697 u64 latest_transid = 0;
a0af469b 698 u64 devid;
2b82032c 699 int seeding = 1;
a0af469b 700 int ret = 0;
8a4b83cc 701
d4d77629
TH		 702 flags |= FMODE_EXCL;
703
c6e30871 704 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 705 if (device->bdev)
706 continue;
dfe25020
CM		 707 if (!device->name)
708 continue;
709
beaf8ab3
SB		 710 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
711 &bdev, &bh);
712 if (ret)
713 continue;
a0af469b
CM		 714
715 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 716 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 717 if (devid != device->devid)
718 goto error_brelse;
719
2b82032c
YZ		 720 if (memcmp(device->uuid, disk_super->dev_item.uuid,
721 BTRFS_UUID_SIZE))
722 goto error_brelse;
723
724 device->generation = btrfs_super_generation(disk_super);
725 if (!latest_transid || device->generation > latest_transid) {
a0af469b 726 latest_devid = devid;
2b82032c 727 latest_transid = device->generation;
a0af469b
CM		 728 latest_bdev = bdev;
729 }
730
2b82032c
YZ		 731 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
732 device->writeable = 0;
733 } else {
734 device->writeable = !bdev_read_only(bdev);
735 seeding = 0;
736 }
737
d5e2003c
JB		 738 q = bdev_get_queue(bdev);
739 if (blk_queue_discard(q)) {
740 device->can_discard = 1;
741 fs_devices->num_can_discard++;
742 }
743
8a4b83cc 744 device->bdev = bdev;
dfe25020 745 device->in_fs_metadata = 0;
15916de8
CM		 746 device->mode = flags;
747
c289811c
CM		 748 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
749 fs_devices->rotating = 1;
750
a0af469b 751 fs_devices->open_devices++;
8dabb742 752 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
2b82032c
YZ		 753 fs_devices->rw_devices++;
754 list_add(&device->dev_alloc_list,
755 &fs_devices->alloc_list);
756 }
4f6c9328 757 brelse(bh);
a0af469b 758 continue;
a061fc8d 759
a0af469b
CM		 760error_brelse:
761 brelse(bh);
d4d77629 762 blkdev_put(bdev, flags);
a0af469b 763 continue;
8a4b83cc 764 }
a0af469b 765 if (fs_devices->open_devices == 0) {
20bcd649 766 ret = -EINVAL;
a0af469b
CM		 767 goto out;
768 }
2b82032c
YZ		 769 fs_devices->seeding = seeding;
770 fs_devices->opened = 1;
a0af469b
CM		 771 fs_devices->latest_bdev = latest_bdev;
772 fs_devices->latest_devid = latest_devid;
773 fs_devices->latest_trans = latest_transid;
2b82032c 774 fs_devices->total_rw_bytes = 0;
a0af469b 775out:
2b82032c
YZ		 776 return ret;
777}
778
779int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 780 fmode_t flags, void *holder)
2b82032c
YZ		 781{
782 int ret;
783
784 mutex_lock(&uuid_mutex);
785 if (fs_devices->opened) {
e4404d6e
YZ		 786 fs_devices->opened++;
787 ret = 0;
2b82032c 788 } else {
15916de8 789 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 790 }
8a4b83cc 791 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 792 return ret;
793}
794
6f60cbd3
DS		 795/*
796 * Look for a btrfs signature on a device. This may be called out of the mount path
797 * and we are not allowed to call set_blocksize during the scan. The superblock
798 * is read via pagecache
799 */
97288f2c 800int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 801 struct btrfs_fs_devices **fs_devices_ret)
802{
803 struct btrfs_super_block *disk_super;
804 struct block_device *bdev;
6f60cbd3
DS		 805 struct page *page;
806 void *p;
807 int ret = -EINVAL;
8a4b83cc 808 u64 devid;
f2984462 809 u64 transid;
02db0844 810 u64 total_devices;
6f60cbd3
DS		 811 u64 bytenr;
812 pgoff_t index;
8a4b83cc 813
6f60cbd3
DS		 814 /*
815 * we would like to check all the supers, but that would make
816 * a btrfs mount succeed after a mkfs from a different FS.
817 * So, we need to add a special mount option to scan for
818 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
819 */
820 bytenr = btrfs_sb_offset(0);
d4d77629 821 flags |= FMODE_EXCL;
10f6327b 822 mutex_lock(&uuid_mutex);
6f60cbd3
DS		 823
824 bdev = blkdev_get_by_path(path, flags, holder);
825
826 if (IS_ERR(bdev)) {
827 ret = PTR_ERR(bdev);
828 printk(KERN_INFO "btrfs: open %s failed\n", path);
beaf8ab3 829 goto error;
6f60cbd3
DS		 830 }
831
832 /* make sure our super fits in the device */
833 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
834 goto error_bdev_put;
835
836 /* make sure our super fits in the page */
837 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
838 goto error_bdev_put;
839
840 /* make sure our super doesn't straddle pages on disk */
841 index = bytenr >> PAGE_CACHE_SHIFT;
842 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
843 goto error_bdev_put;
844
845 /* pull in the page with our super */
846 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
847 index, GFP_NOFS);
848
849 if (IS_ERR_OR_NULL(page))
850 goto error_bdev_put;
851
852 p = kmap(page);
853
854 /* align our pointer to the offset of the super block */
855 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
856
857 if (btrfs_super_bytenr(disk_super) != bytenr ||
858 strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
859 sizeof(disk_super->magic)))
860 goto error_unmap;
861
a343832f 862 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 863 transid = btrfs_super_generation(disk_super);
02db0844 864 total_devices = btrfs_super_num_devices(disk_super);
6f60cbd3 865
d03f918a
SB		 866 if (disk_super->label[0]) {
867 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
868 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
d397712b 869 printk(KERN_INFO "device label %s ", disk_super->label);
d03f918a 870 } else {
22b63a29 871 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
d03f918a 872 }
6f60cbd3 873
119e10cf 874 printk(KERN_CONT "devid %llu transid %llu %s\n",
d397712b 875 (unsigned long long)devid, (unsigned long long)transid, path);
6f60cbd3 876
8a4b83cc 877 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
02db0844
JB		 878 if (!ret && fs_devices_ret)
879 (*fs_devices_ret)->total_devices = total_devices;
6f60cbd3
DS		 880
881error_unmap:
882 kunmap(page);
883 page_cache_release(page);
884
885error_bdev_put:
d4d77629 886 blkdev_put(bdev, flags);
8a4b83cc 887error:
beaf8ab3 888 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 889 return ret;
890}
0b86a832 891
6d07bcec
MX		 892/* helper to account the used device space in the range */
893int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
894 u64 end, u64 *length)
895{
896 struct btrfs_key key;
897 struct btrfs_root *root = device->dev_root;
898 struct btrfs_dev_extent *dev_extent;
899 struct btrfs_path *path;
900 u64 extent_end;
901 int ret;
902 int slot;
903 struct extent_buffer *l;
904
905 *length = 0;
906
63a212ab 907 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
6d07bcec
MX		 908 return 0;
909
910 path = btrfs_alloc_path();
911 if (!path)
912 return -ENOMEM;
913 path->reada = 2;
914
915 key.objectid = device->devid;
916 key.offset = start;
917 key.type = BTRFS_DEV_EXTENT_KEY;
918
919 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
920 if (ret < 0)
921 goto out;
922 if (ret > 0) {
923 ret = btrfs_previous_item(root, path, key.objectid, key.type);
924 if (ret < 0)
925 goto out;
926 }
927
928 while (1) {
929 l = path->nodes[0];
930 slot = path->slots[0];
931 if (slot >= btrfs_header_nritems(l)) {
932 ret = btrfs_next_leaf(root, path);
933 if (ret == 0)
934 continue;
935 if (ret < 0)
936 goto out;
937
938 break;
939 }
940 btrfs_item_key_to_cpu(l, &key, slot);
941
942 if (key.objectid < device->devid)
943 goto next;
944
945 if (key.objectid > device->devid)
946 break;
947
948 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
949 goto next;
950
951 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
952 extent_end = key.offset + btrfs_dev_extent_length(l,
953 dev_extent);
954 if (key.offset <= start && extent_end > end) {
955 *length = end - start + 1;
956 break;
957 } else if (key.offset <= start && extent_end > start)
958 *length += extent_end - start;
959 else if (key.offset > start && extent_end <= end)
960 *length += extent_end - key.offset;
961 else if (key.offset > start && key.offset <= end) {
962 *length += end - key.offset + 1;
963 break;
964 } else if (key.offset > end)
965 break;
966
967next:
968 path->slots[0]++;
969 }
970 ret = 0;
971out:
972 btrfs_free_path(path);
973 return ret;
974}
975
0b86a832 976/*
7bfc837d 977 * find_free_dev_extent - find free space in the specified device
7bfc837d
MX		 978 * @device: the device which we search the free space in
979 * @num_bytes: the size of the free space that we need
980 * @start: store the start of the free space.
981 * @len: the size of the free space. that we find, or the size of the max
982 * free space if we don't find suitable free space
983 *
0b86a832
CM		 984 * this uses a pretty simple search, the expectation is that it is
985 * called very infrequently and that a given device has a small number
986 * of extents
7bfc837d
MX		 987 *
988 * @start is used to store the start of the free space if we find. But if we
989 * don't find suitable free space, it will be used to store the start position
990 * of the max free space.
991 *
992 * @len is used to store the size of the free space that we find.
993 * But if we don't find suitable free space, it is used to store the size of
994 * the max free space.
0b86a832 995 */
125ccb0a 996int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
7bfc837d 997 u64 *start, u64 *len)
0b86a832
CM		 998{
999 struct btrfs_key key;
1000 struct btrfs_root *root = device->dev_root;
7bfc837d 1001 struct btrfs_dev_extent *dev_extent;
2b82032c 1002 struct btrfs_path *path;
7bfc837d
MX		 1003 u64 hole_size;
1004 u64 max_hole_start;
1005 u64 max_hole_size;
1006 u64 extent_end;
1007 u64 search_start;
0b86a832
CM		 1008 u64 search_end = device->total_bytes;
1009 int ret;
7bfc837d 1010 int slot;
0b86a832
CM		 1011 struct extent_buffer *l;
1012
0b86a832
CM		 1013 /* FIXME use last free of some kind */
1014
8a4b83cc
CM		 1015 /* we don't want to overwrite the superblock on the drive,
1016 * so we make sure to start at an offset of at least 1MB
1017 */
a9c9bf68 1018 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
8f18cf13 1019
7bfc837d
MX		 1020 max_hole_start = search_start;
1021 max_hole_size = 0;
38c01b96 1022 hole_size = 0;
7bfc837d 1023
63a212ab 1024 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
7bfc837d
MX		 1025 ret = -ENOSPC;
1026 goto error;
1027 }
1028
1029 path = btrfs_alloc_path();
1030 if (!path) {
1031 ret = -ENOMEM;
1032 goto error;
1033 }
1034 path->reada = 2;
1035
0b86a832
CM		 1036 key.objectid = device->devid;
1037 key.offset = search_start;
1038 key.type = BTRFS_DEV_EXTENT_KEY;
7bfc837d 1039
125ccb0a 1040 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
0b86a832 1041 if (ret < 0)
7bfc837d 1042 goto out;
1fcbac58
YZ		 1043 if (ret > 0) {
1044 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1045 if (ret < 0)
7bfc837d 1046 goto out;
1fcbac58 1047 }
7bfc837d 1048
0b86a832
CM		 1049 while (1) {
1050 l = path->nodes[0];
1051 slot = path->slots[0];
1052 if (slot >= btrfs_header_nritems(l)) {
1053 ret = btrfs_next_leaf(root, path);
1054 if (ret == 0)
1055 continue;
1056 if (ret < 0)
7bfc837d
MX		 1057 goto out;
1058
1059 break;
0b86a832
CM		 1060 }
1061 btrfs_item_key_to_cpu(l, &key, slot);
1062
1063 if (key.objectid < device->devid)
1064 goto next;
1065
1066 if (key.objectid > device->devid)
7bfc837d 1067 break;
0b86a832 1068
7bfc837d
MX		 1069 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1070 goto next;
9779b72f 1071
7bfc837d
MX		 1072 if (key.offset > search_start) {
1073 hole_size = key.offset - search_start;
9779b72f 1074
7bfc837d
MX		 1075 if (hole_size > max_hole_size) {
1076 max_hole_start = search_start;
1077 max_hole_size = hole_size;
1078 }
9779b72f 1079
7bfc837d
MX		 1080 /*
1081 * If this free space is greater than which we need,
1082 * it must be the max free space that we have found
1083 * until now, so max_hole_start must point to the start
1084 * of this free space and the length of this free space
1085 * is stored in max_hole_size. Thus, we return
1086 * max_hole_start and max_hole_size and go back to the
1087 * caller.
1088 */
1089 if (hole_size >= num_bytes) {
1090 ret = 0;
1091 goto out;
0b86a832
CM		 1092 }
1093 }
0b86a832 1094
0b86a832 1095 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
7bfc837d
MX		 1096 extent_end = key.offset + btrfs_dev_extent_length(l,
1097 dev_extent);
1098 if (extent_end > search_start)
1099 search_start = extent_end;
0b86a832
CM		 1100next:
1101 path->slots[0]++;
1102 cond_resched();
1103 }
0b86a832 1104
38c01b96 1105 /*
1106 * At this point, search_start should be the end of
1107 * allocated dev extents, and when shrinking the device,
1108 * search_end may be smaller than search_start.
1109 */
1110 if (search_end > search_start)
1111 hole_size = search_end - search_start;
1112
7bfc837d
MX		 1113 if (hole_size > max_hole_size) {
1114 max_hole_start = search_start;
1115 max_hole_size = hole_size;
0b86a832 1116 }
0b86a832 1117
7bfc837d
MX		 1118 /* See above. */
1119 if (hole_size < num_bytes)
1120 ret = -ENOSPC;
1121 else
1122 ret = 0;
1123
1124out:
2b82032c 1125 btrfs_free_path(path);
7bfc837d
MX		 1126error:
1127 *start = max_hole_start;
b2117a39 1128 if (len)
7bfc837d 1129 *len = max_hole_size;
0b86a832
CM		 1130 return ret;
1131}
1132
b2950863 1133static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13
CM		 1134 struct btrfs_device *device,
1135 u64 start)
1136{
1137 int ret;
1138 struct btrfs_path *path;
1139 struct btrfs_root *root = device->dev_root;
1140 struct btrfs_key key;
a061fc8d
CM		 1141 struct btrfs_key found_key;
1142 struct extent_buffer *leaf = NULL;
1143 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 1144
1145 path = btrfs_alloc_path();
1146 if (!path)
1147 return -ENOMEM;
1148
1149 key.objectid = device->devid;
1150 key.offset = start;
1151 key.type = BTRFS_DEV_EXTENT_KEY;
924cd8fb 1152again:
8f18cf13 1153 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 1154 if (ret > 0) {
1155 ret = btrfs_previous_item(root, path, key.objectid,
1156 BTRFS_DEV_EXTENT_KEY);
b0b802d7
TI		 1157 if (ret)
1158 goto out;
a061fc8d
CM		 1159 leaf = path->nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1161 extent = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_dev_extent);
1163 BUG_ON(found_key.offset > start || found_key.offset +
1164 btrfs_dev_extent_length(leaf, extent) < start);
924cd8fb
MX		 1165 key = found_key;
1166 btrfs_release_path(path);
1167 goto again;
a061fc8d
CM		 1168 } else if (ret == 0) {
1169 leaf = path->nodes[0];
1170 extent = btrfs_item_ptr(leaf, path->slots[0],
1171 struct btrfs_dev_extent);
79787eaa
JM		 1172 } else {
1173 btrfs_error(root->fs_info, ret, "Slot search failed");
1174 goto out;
a061fc8d 1175 }
8f18cf13 1176
2bf64758
JB		 1177 if (device->bytes_used > 0) {
1178 u64 len = btrfs_dev_extent_length(leaf, extent);
1179 device->bytes_used -= len;
1180 spin_lock(&root->fs_info->free_chunk_lock);
1181 root->fs_info->free_chunk_space += len;
1182 spin_unlock(&root->fs_info->free_chunk_lock);
1183 }
8f18cf13 1184 ret = btrfs_del_item(trans, root, path);
79787eaa
JM		 1185 if (ret) {
1186 btrfs_error(root->fs_info, ret,
1187 "Failed to remove dev extent item");
1188 }
b0b802d7 1189out:
8f18cf13
CM		 1190 btrfs_free_path(path);
1191 return ret;
1192}
1193
2b82032c 1194int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
0b86a832 1195 struct btrfs_device *device,
e17cade2 1196 u64 chunk_tree, u64 chunk_objectid,
2b82032c 1197 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 1198{
1199 int ret;
1200 struct btrfs_path *path;
1201 struct btrfs_root *root = device->dev_root;
1202 struct btrfs_dev_extent *extent;
1203 struct extent_buffer *leaf;
1204 struct btrfs_key key;
1205
dfe25020 1206 WARN_ON(!device->in_fs_metadata);
63a212ab 1207 WARN_ON(device->is_tgtdev_for_dev_replace);
0b86a832
CM		 1208 path = btrfs_alloc_path();
1209 if (!path)
1210 return -ENOMEM;
1211
0b86a832 1212 key.objectid = device->devid;
2b82032c 1213 key.offset = start;
0b86a832
CM		 1214 key.type = BTRFS_DEV_EXTENT_KEY;
1215 ret = btrfs_insert_empty_item(trans, root, path, &key,
1216 sizeof(*extent));
2cdcecbc
MF		 1217 if (ret)
1218 goto out;
0b86a832
CM		 1219
1220 leaf = path->nodes[0];
1221 extent = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_dev_extent);
e17cade2
CM		 1223 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1224 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1225 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1226
1227 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1228 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1229 BTRFS_UUID_SIZE);
1230
0b86a832
CM		 1231 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1232 btrfs_mark_buffer_dirty(leaf);
2cdcecbc 1233out:
0b86a832
CM		 1234 btrfs_free_path(path);
1235 return ret;
1236}
1237
a1b32a59
CM		 1238static noinline int find_next_chunk(struct btrfs_root *root,
1239 u64 objectid, u64 *offset)
0b86a832
CM		 1240{
1241 struct btrfs_path *path;
1242 int ret;
1243 struct btrfs_key key;
e17cade2 1244 struct btrfs_chunk *chunk;
0b86a832
CM		 1245 struct btrfs_key found_key;
1246
1247 path = btrfs_alloc_path();
92b8e897
MF		 1248 if (!path)
1249 return -ENOMEM;
0b86a832 1250
e17cade2 1251 key.objectid = objectid;
0b86a832
CM		 1252 key.offset = (u64)-1;
1253 key.type = BTRFS_CHUNK_ITEM_KEY;
1254
1255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1256 if (ret < 0)
1257 goto error;
1258
79787eaa 1259 BUG_ON(ret == 0); /* Corruption */
0b86a832
CM		 1260
1261 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1262 if (ret) {
e17cade2 1263 *offset = 0;
0b86a832
CM		 1264 } else {
1265 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1266 path->slots[0]);
e17cade2
CM		 1267 if (found_key.objectid != objectid)
1268 *offset = 0;
1269 else {
1270 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1271 struct btrfs_chunk);
1272 *offset = found_key.offset +
1273 btrfs_chunk_length(path->nodes[0], chunk);
1274 }
0b86a832
CM		 1275 }
1276 ret = 0;
1277error:
1278 btrfs_free_path(path);
1279 return ret;
1280}
1281
2b82032c 1282static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
0b86a832
CM		 1283{
1284 int ret;
1285 struct btrfs_key key;
1286 struct btrfs_key found_key;
2b82032c
YZ		 1287 struct btrfs_path *path;
1288
1289 root = root->fs_info->chunk_root;
1290
1291 path = btrfs_alloc_path();
1292 if (!path)
1293 return -ENOMEM;
0b86a832
CM		 1294
1295 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1296 key.type = BTRFS_DEV_ITEM_KEY;
1297 key.offset = (u64)-1;
1298
1299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1300 if (ret < 0)
1301 goto error;
1302
79787eaa 1303 BUG_ON(ret == 0); /* Corruption */
0b86a832
CM		 1304
1305 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1306 BTRFS_DEV_ITEM_KEY);
1307 if (ret) {
1308 *objectid = 1;
1309 } else {
1310 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1311 path->slots[0]);
1312 *objectid = found_key.offset + 1;
1313 }
1314 ret = 0;
1315error:
2b82032c 1316 btrfs_free_path(path);
0b86a832
CM		 1317 return ret;
1318}
1319
1320/*
1321 * the device information is stored in the chunk root
1322 * the btrfs_device struct should be fully filled in
1323 */
1324int btrfs_add_device(struct btrfs_trans_handle *trans,
1325 struct btrfs_root *root,
1326 struct btrfs_device *device)
1327{
1328 int ret;
1329 struct btrfs_path *path;
1330 struct btrfs_dev_item *dev_item;
1331 struct extent_buffer *leaf;
1332 struct btrfs_key key;
1333 unsigned long ptr;
0b86a832
CM		 1334
1335 root = root->fs_info->chunk_root;
1336
1337 path = btrfs_alloc_path();
1338 if (!path)
1339 return -ENOMEM;
1340
0b86a832
CM		 1341 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1342 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1343 key.offset = device->devid;
0b86a832
CM		 1344
1345 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1346 sizeof(*dev_item));
0b86a832
CM		 1347 if (ret)
1348 goto out;
1349
1350 leaf = path->nodes[0];
1351 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1352
1353 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1354 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1355 btrfs_set_device_type(leaf, dev_item, device->type);
1356 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1357 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1358 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
0b86a832
CM		 1359 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1360 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
e17cade2
CM		 1361 btrfs_set_device_group(leaf, dev_item, 0);
1362 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1363 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1364 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1365
0b86a832 1366 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 1367 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2b82032c
YZ		 1368 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1369 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
0b86a832 1370 btrfs_mark_buffer_dirty(leaf);
0b86a832 1371
2b82032c 1372 ret = 0;
0b86a832
CM		 1373out:
1374 btrfs_free_path(path);
1375 return ret;
1376}
8f18cf13 1377
a061fc8d
CM		 1378static int btrfs_rm_dev_item(struct btrfs_root *root,
1379 struct btrfs_device *device)
1380{
1381 int ret;
1382 struct btrfs_path *path;
a061fc8d 1383 struct btrfs_key key;
a061fc8d
CM		 1384 struct btrfs_trans_handle *trans;
1385
1386 root = root->fs_info->chunk_root;
1387
1388 path = btrfs_alloc_path();
1389 if (!path)
1390 return -ENOMEM;
1391
a22285a6 1392 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 1393 if (IS_ERR(trans)) {
1394 btrfs_free_path(path);
1395 return PTR_ERR(trans);
1396 }
a061fc8d
CM		 1397 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1398 key.type = BTRFS_DEV_ITEM_KEY;
1399 key.offset = device->devid;
7d9eb12c 1400 lock_chunks(root);
a061fc8d
CM		 1401
1402 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 if (ret < 0)
1404 goto out;
1405
1406 if (ret > 0) {
1407 ret = -ENOENT;
1408 goto out;
1409 }
1410
1411 ret = btrfs_del_item(trans, root, path);
1412 if (ret)
1413 goto out;
a061fc8d
CM		 1414out:
1415 btrfs_free_path(path);
7d9eb12c 1416 unlock_chunks(root);
a061fc8d
CM		 1417 btrfs_commit_transaction(trans, root);
1418 return ret;
1419}
1420
1421int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1422{
1423 struct btrfs_device *device;
2b82032c 1424 struct btrfs_device *next_device;
a061fc8d 1425 struct block_device *bdev;
dfe25020 1426 struct buffer_head *bh = NULL;
a061fc8d 1427 struct btrfs_super_block *disk_super;
1f78160c 1428 struct btrfs_fs_devices *cur_devices;
a061fc8d
CM		 1429 u64 all_avail;
1430 u64 devid;
2b82032c
YZ		 1431 u64 num_devices;
1432 u8 *dev_uuid;
de98ced9 1433 unsigned seq;
a061fc8d 1434 int ret = 0;
1f78160c 1435 bool clear_super = false;
a061fc8d 1436
a061fc8d
CM		 1437 mutex_lock(&uuid_mutex);
1438
de98ced9
MX		 1439 do {
1440 seq = read_seqbegin(&root->fs_info->profiles_lock);
1441
1442 all_avail = root->fs_info->avail_data_alloc_bits |
1443 root->fs_info->avail_system_alloc_bits |
1444 root->fs_info->avail_metadata_alloc_bits;
1445 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
a061fc8d 1446
8dabb742
SB		 1447 num_devices = root->fs_info->fs_devices->num_devices;
1448 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1449 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1450 WARN_ON(num_devices < 1);
1451 num_devices--;
1452 }
1453 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1454
1455 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
d397712b
CM		 1456 printk(KERN_ERR "btrfs: unable to go below four devices "
1457 "on raid10\n");
a061fc8d
CM		 1458 ret = -EINVAL;
1459 goto out;
1460 }
1461
8dabb742 1462 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
d397712b
CM		 1463 printk(KERN_ERR "btrfs: unable to go below two "
1464 "devices on raid1\n");
a061fc8d
CM		 1465 ret = -EINVAL;
1466 goto out;
1467 }
1468
dfe25020 1469 if (strcmp(device_path, "missing") == 0) {
dfe25020
CM		 1470 struct list_head *devices;
1471 struct btrfs_device *tmp;
a061fc8d 1472
dfe25020
CM		 1473 device = NULL;
1474 devices = &root->fs_info->fs_devices->devices;
46224705
XG		 1475 /*
1476 * It is safe to read the devices since the volume_mutex
1477 * is held.
1478 */
c6e30871 1479 list_for_each_entry(tmp, devices, dev_list) {
63a212ab
SB		 1480 if (tmp->in_fs_metadata &&
1481 !tmp->is_tgtdev_for_dev_replace &&
1482 !tmp->bdev) {
dfe25020
CM		 1483 device = tmp;
1484 break;
1485 }
1486 }
1487 bdev = NULL;
1488 bh = NULL;
1489 disk_super = NULL;
1490 if (!device) {
d397712b
CM		 1491 printk(KERN_ERR "btrfs: no missing devices found to "
1492 "remove\n");
dfe25020
CM		 1493 goto out;
1494 }
dfe25020 1495 } else {
beaf8ab3 1496 ret = btrfs_get_bdev_and_sb(device_path,
cc975eb4 1497 FMODE_WRITE | FMODE_EXCL,
beaf8ab3
SB		 1498 root->fs_info->bdev_holder, 0,
1499 &bdev, &bh);
1500 if (ret)
dfe25020 1501 goto out;
dfe25020 1502 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 1503 devid = btrfs_stack_device_id(&disk_super->dev_item);
2b82032c 1504 dev_uuid = disk_super->dev_item.uuid;
aa1b8cd4 1505 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2b82032c 1506 disk_super->fsid);
dfe25020
CM		 1507 if (!device) {
1508 ret = -ENOENT;
1509 goto error_brelse;
1510 }
2b82032c 1511 }
dfe25020 1512
63a212ab
SB		 1513 if (device->is_tgtdev_for_dev_replace) {
1514 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1515 ret = -EINVAL;
1516 goto error_brelse;
1517 }
1518
2b82032c 1519 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
d397712b
CM		 1520 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1521 "device\n");
2b82032c
YZ		 1522 ret = -EINVAL;
1523 goto error_brelse;
1524 }
1525
1526 if (device->writeable) {
0c1daee0 1527 lock_chunks(root);
2b82032c 1528 list_del_init(&device->dev_alloc_list);
0c1daee0 1529 unlock_chunks(root);
2b82032c 1530 root->fs_info->fs_devices->rw_devices--;
1f78160c 1531 clear_super = true;
dfe25020 1532 }
a061fc8d
CM		 1533
1534 ret = btrfs_shrink_device(device, 0);
1535 if (ret)
9b3517e9 1536 goto error_undo;
a061fc8d 1537
63a212ab
SB		 1538 /*
1539 * TODO: the superblock still includes this device in its num_devices
1540 * counter although write_all_supers() is not locked out. This
1541 * could give a filesystem state which requires a degraded mount.
1542 */
a061fc8d
CM		 1543 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1544 if (ret)
9b3517e9 1545 goto error_undo;
a061fc8d 1546
2bf64758
JB		 1547 spin_lock(&root->fs_info->free_chunk_lock);
1548 root->fs_info->free_chunk_space = device->total_bytes -
1549 device->bytes_used;
1550 spin_unlock(&root->fs_info->free_chunk_lock);
1551
2b82032c 1552 device->in_fs_metadata = 0;
aa1b8cd4 1553 btrfs_scrub_cancel_dev(root->fs_info, device);
e5e9a520
CM		 1554
1555 /*
1556 * the device list mutex makes sure that we don't change
1557 * the device list while someone else is writing out all
1558 * the device supers.
1559 */
1f78160c
XG		 1560
1561 cur_devices = device->fs_devices;
e5e9a520 1562 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1563 list_del_rcu(&device->dev_list);
e5e9a520 1564
e4404d6e 1565 device->fs_devices->num_devices--;
02db0844 1566 device->fs_devices->total_devices--;
2b82032c 1567
cd02dca5
CM		 1568 if (device->missing)
1569 root->fs_info->fs_devices->missing_devices--;
1570
2b82032c
YZ		 1571 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1572 struct btrfs_device, dev_list);
1573 if (device->bdev == root->fs_info->sb->s_bdev)
1574 root->fs_info->sb->s_bdev = next_device->bdev;
1575 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1576 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1577
1f78160c 1578 if (device->bdev)
e4404d6e 1579 device->fs_devices->open_devices--;
1f78160c
XG		 1580
1581 call_rcu(&device->rcu, free_device);
1582 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
e4404d6e 1583
6c41761f
DS		 1584 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1585 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
2b82032c 1586
1f78160c 1587 if (cur_devices->open_devices == 0) {
e4404d6e
YZ		 1588 struct btrfs_fs_devices *fs_devices;
1589 fs_devices = root->fs_info->fs_devices;
1590 while (fs_devices) {
1f78160c 1591 if (fs_devices->seed == cur_devices)
e4404d6e
YZ		 1592 break;
1593 fs_devices = fs_devices->seed;
2b82032c 1594 }
1f78160c
XG		 1595 fs_devices->seed = cur_devices->seed;
1596 cur_devices->seed = NULL;
0c1daee0 1597 lock_chunks(root);
1f78160c 1598 __btrfs_close_devices(cur_devices);
0c1daee0 1599 unlock_chunks(root);
1f78160c 1600 free_fs_devices(cur_devices);
2b82032c
YZ		 1601 }
1602
5af3e8cc
SB		 1603 root->fs_info->num_tolerated_disk_barrier_failures =
1604 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1605
2b82032c
YZ		 1606 /*
1607 * at this point, the device is zero sized. We want to
1608 * remove it from the devices list and zero out the old super
1609 */
aa1b8cd4 1610 if (clear_super && disk_super) {
dfe25020
CM		 1611 /* make sure this device isn't detected as part of
1612 * the FS anymore
1613 */
1614 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1615 set_buffer_dirty(bh);
1616 sync_dirty_buffer(bh);
dfe25020 1617 }
a061fc8d 1618
a061fc8d 1619 ret = 0;
a061fc8d 1620
b8b8ff59 1621 /* Notify udev that device has changed */
3c911608
ES		 1622 if (bdev)
1623 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
b8b8ff59 1624
a061fc8d
CM		 1625error_brelse:
1626 brelse(bh);
dfe25020 1627 if (bdev)
e525fd89 1628 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
a061fc8d
CM		 1629out:
1630 mutex_unlock(&uuid_mutex);
a061fc8d 1631 return ret;
9b3517e9
ID		 1632error_undo:
1633 if (device->writeable) {
0c1daee0 1634 lock_chunks(root);
9b3517e9
ID		 1635 list_add(&device->dev_alloc_list,
1636 &root->fs_info->fs_devices->alloc_list);
0c1daee0 1637 unlock_chunks(root);
9b3517e9
ID		 1638 root->fs_info->fs_devices->rw_devices++;
1639 }
1640 goto error_brelse;
a061fc8d
CM		 1641}
1642
e93c89c1
SB		 1643void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1644 struct btrfs_device *srcdev)
1645{
1646 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1647 list_del_rcu(&srcdev->dev_list);
1648 list_del_rcu(&srcdev->dev_alloc_list);
1649 fs_info->fs_devices->num_devices--;
1650 if (srcdev->missing) {
1651 fs_info->fs_devices->missing_devices--;
1652 fs_info->fs_devices->rw_devices++;
1653 }
1654 if (srcdev->can_discard)
1655 fs_info->fs_devices->num_can_discard--;
1656 if (srcdev->bdev)
1657 fs_info->fs_devices->open_devices--;
1658
1659 call_rcu(&srcdev->rcu, free_device);
1660}
1661
1662void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1663 struct btrfs_device *tgtdev)
1664{
1665 struct btrfs_device *next_device;
1666
1667 WARN_ON(!tgtdev);
1668 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1669 if (tgtdev->bdev) {
1670 btrfs_scratch_superblock(tgtdev);
1671 fs_info->fs_devices->open_devices--;
1672 }
1673 fs_info->fs_devices->num_devices--;
1674 if (tgtdev->can_discard)
1675 fs_info->fs_devices->num_can_discard++;
1676
1677 next_device = list_entry(fs_info->fs_devices->devices.next,
1678 struct btrfs_device, dev_list);
1679 if (tgtdev->bdev == fs_info->sb->s_bdev)
1680 fs_info->sb->s_bdev = next_device->bdev;
1681 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1682 fs_info->fs_devices->latest_bdev = next_device->bdev;
1683 list_del_rcu(&tgtdev->dev_list);
1684
1685 call_rcu(&tgtdev->rcu, free_device);
1686
1687 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1688}
1689
7ba15b7d
SB		 1690int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1691 struct btrfs_device **device)
1692{
1693 int ret = 0;
1694 struct btrfs_super_block *disk_super;
1695 u64 devid;
1696 u8 *dev_uuid;
1697 struct block_device *bdev;
1698 struct buffer_head *bh;
1699
1700 *device = NULL;
1701 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1702 root->fs_info->bdev_holder, 0, &bdev, &bh);
1703 if (ret)
1704 return ret;
1705 disk_super = (struct btrfs_super_block *)bh->b_data;
1706 devid = btrfs_stack_device_id(&disk_super->dev_item);
1707 dev_uuid = disk_super->dev_item.uuid;
aa1b8cd4 1708 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
7ba15b7d
SB		 1709 disk_super->fsid);
1710 brelse(bh);
1711 if (!*device)
1712 ret = -ENOENT;
1713 blkdev_put(bdev, FMODE_READ);
1714 return ret;
1715}
1716
1717int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1718 char *device_path,
1719 struct btrfs_device **device)
1720{
1721 *device = NULL;
1722 if (strcmp(device_path, "missing") == 0) {
1723 struct list_head *devices;
1724 struct btrfs_device *tmp;
1725
1726 devices = &root->fs_info->fs_devices->devices;
1727 /*
1728 * It is safe to read the devices since the volume_mutex
1729 * is held by the caller.
1730 */
1731 list_for_each_entry(tmp, devices, dev_list) {
1732 if (tmp->in_fs_metadata && !tmp->bdev) {
1733 *device = tmp;
1734 break;
1735 }
1736 }
1737
1738 if (!*device) {
1739 pr_err("btrfs: no missing device found\n");
1740 return -ENOENT;
1741 }
1742
1743 return 0;
1744 } else {
1745 return btrfs_find_device_by_path(root, device_path, device);
1746 }
1747}
1748
2b82032c
YZ		 1749/*
1750 * does all the dirty work required for changing file system's UUID.
1751 */
125ccb0a 1752static int btrfs_prepare_sprout(struct btrfs_root *root)
2b82032c
YZ		 1753{
1754 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1755 struct btrfs_fs_devices *old_devices;
e4404d6e 1756 struct btrfs_fs_devices *seed_devices;
6c41761f 1757 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2b82032c
YZ		 1758 struct btrfs_device *device;
1759 u64 super_flags;
1760
1761 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 1762 if (!fs_devices->seeding)
2b82032c
YZ		 1763 return -EINVAL;
1764
e4404d6e
YZ		 1765 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1766 if (!seed_devices)
2b82032c
YZ		 1767 return -ENOMEM;
1768
e4404d6e
YZ		 1769 old_devices = clone_fs_devices(fs_devices);
1770 if (IS_ERR(old_devices)) {
1771 kfree(seed_devices);
1772 return PTR_ERR(old_devices);
2b82032c 1773 }
e4404d6e 1774
2b82032c
YZ		 1775 list_add(&old_devices->list, &fs_uuids);
1776
e4404d6e
YZ		 1777 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1778 seed_devices->opened = 1;
1779 INIT_LIST_HEAD(&seed_devices->devices);
1780 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 1781 mutex_init(&seed_devices->device_list_mutex);
c9513edb
XG		 1782
1783 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c
XG		 1784 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1785 synchronize_rcu);
c9513edb
XG		 1786 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1787
e4404d6e
YZ		 1788 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1789 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1790 device->fs_devices = seed_devices;
1791 }
1792
2b82032c
YZ		 1793 fs_devices->seeding = 0;
1794 fs_devices->num_devices = 0;
1795 fs_devices->open_devices = 0;
02db0844 1796 fs_devices->total_devices = 0;
e4404d6e 1797 fs_devices->seed = seed_devices;
2b82032c
YZ		 1798
1799 generate_random_uuid(fs_devices->fsid);
1800 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1801 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1802 super_flags = btrfs_super_flags(disk_super) &
1803 ~BTRFS_SUPER_FLAG_SEEDING;
1804 btrfs_set_super_flags(disk_super, super_flags);
1805
1806 return 0;
1807}
1808
1809/*
1810 * strore the expected generation for seed devices in device items.
1811 */
1812static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1813 struct btrfs_root *root)
1814{
1815 struct btrfs_path *path;
1816 struct extent_buffer *leaf;
1817 struct btrfs_dev_item *dev_item;
1818 struct btrfs_device *device;
1819 struct btrfs_key key;
1820 u8 fs_uuid[BTRFS_UUID_SIZE];
1821 u8 dev_uuid[BTRFS_UUID_SIZE];
1822 u64 devid;
1823 int ret;
1824
1825 path = btrfs_alloc_path();
1826 if (!path)
1827 return -ENOMEM;
1828
1829 root = root->fs_info->chunk_root;
1830 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1831 key.offset = 0;
1832 key.type = BTRFS_DEV_ITEM_KEY;
1833
1834 while (1) {
1835 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1836 if (ret < 0)
1837 goto error;
1838
1839 leaf = path->nodes[0];
1840next_slot:
1841 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1842 ret = btrfs_next_leaf(root, path);
1843 if (ret > 0)
1844 break;
1845 if (ret < 0)
1846 goto error;
1847 leaf = path->nodes[0];
1848 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 1849 btrfs_release_path(path);
2b82032c
YZ		 1850 continue;
1851 }
1852
1853 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1854 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1855 key.type != BTRFS_DEV_ITEM_KEY)
1856 break;
1857
1858 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1859 struct btrfs_dev_item);
1860 devid = btrfs_device_id(leaf, dev_item);
1861 read_extent_buffer(leaf, dev_uuid,
1862 (unsigned long)btrfs_device_uuid(dev_item),
1863 BTRFS_UUID_SIZE);
1864 read_extent_buffer(leaf, fs_uuid,
1865 (unsigned long)btrfs_device_fsid(dev_item),
1866 BTRFS_UUID_SIZE);
aa1b8cd4
SB		 1867 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1868 fs_uuid);
79787eaa 1869 BUG_ON(!device); /* Logic error */
2b82032c
YZ		 1870
1871 if (device->fs_devices->seeding) {
1872 btrfs_set_device_generation(leaf, dev_item,
1873 device->generation);
1874 btrfs_mark_buffer_dirty(leaf);
1875 }
1876
1877 path->slots[0]++;
1878 goto next_slot;
1879 }
1880 ret = 0;
1881error:
1882 btrfs_free_path(path);
1883 return ret;
1884}
1885
788f20eb
CM		 1886int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1887{
d5e2003c 1888 struct request_queue *q;
788f20eb
CM		 1889 struct btrfs_trans_handle *trans;
1890 struct btrfs_device *device;
1891 struct block_device *bdev;
788f20eb 1892 struct list_head *devices;
2b82032c 1893 struct super_block *sb = root->fs_info->sb;
606686ee 1894 struct rcu_string *name;
788f20eb 1895 u64 total_bytes;
2b82032c 1896 int seeding_dev = 0;
788f20eb
CM		 1897 int ret = 0;
1898
2b82032c 1899 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
f8c5d0b4 1900 return -EROFS;
788f20eb 1901
a5d16333 1902 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
d4d77629 1903 root->fs_info->bdev_holder);
7f59203a
JB		 1904 if (IS_ERR(bdev))
1905 return PTR_ERR(bdev);
a2135011 1906
2b82032c
YZ		 1907 if (root->fs_info->fs_devices->seeding) {
1908 seeding_dev = 1;
1909 down_write(&sb->s_umount);
1910 mutex_lock(&uuid_mutex);
1911 }
1912
8c8bee1d 1913 filemap_write_and_wait(bdev->bd_inode->i_mapping);
a2135011 1914
788f20eb 1915 devices = &root->fs_info->fs_devices->devices;
d25628bd
LB		 1916
1917 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
c6e30871 1918 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 1919 if (device->bdev == bdev) {
1920 ret = -EEXIST;
d25628bd
LB		 1921 mutex_unlock(
1922 &root->fs_info->fs_devices->device_list_mutex);
2b82032c 1923 goto error;
788f20eb
CM		 1924 }
1925 }
d25628bd 1926 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb
CM		 1927
1928 device = kzalloc(sizeof(*device), GFP_NOFS);
1929 if (!device) {
1930 /* we can safely leave the fs_devices entry around */
1931 ret = -ENOMEM;
2b82032c 1932 goto error;
788f20eb
CM		 1933 }
1934
606686ee
JB		 1935 name = rcu_string_strdup(device_path, GFP_NOFS);
1936 if (!name) {
788f20eb 1937 kfree(device);
2b82032c
YZ		 1938 ret = -ENOMEM;
1939 goto error;
788f20eb 1940 }
606686ee 1941 rcu_assign_pointer(device->name, name);
2b82032c
YZ		 1942
1943 ret = find_next_devid(root, &device->devid);
1944 if (ret) {
606686ee 1945 rcu_string_free(device->name);
2b82032c
YZ		 1946 kfree(device);
1947 goto error;
1948 }
1949
a22285a6 1950 trans = btrfs_start_transaction(root, 0);
98d5dc13 1951 if (IS_ERR(trans)) {
606686ee 1952 rcu_string_free(device->name);
98d5dc13
TI		 1953 kfree(device);
1954 ret = PTR_ERR(trans);
1955 goto error;
1956 }
1957
2b82032c
YZ		 1958 lock_chunks(root);
1959
d5e2003c
JB		 1960 q = bdev_get_queue(bdev);
1961 if (blk_queue_discard(q))
1962 device->can_discard = 1;
2b82032c
YZ		 1963 device->writeable = 1;
1964 device->work.func = pending_bios_fn;
1965 generate_random_uuid(device->uuid);
1966 spin_lock_init(&device->io_lock);
1967 device->generation = trans->transid;
788f20eb
CM		 1968 device->io_width = root->sectorsize;
1969 device->io_align = root->sectorsize;
1970 device->sector_size = root->sectorsize;
1971 device->total_bytes = i_size_read(bdev->bd_inode);
2cc3c559 1972 device->disk_total_bytes = device->total_bytes;
788f20eb
CM		 1973 device->dev_root = root->fs_info->dev_root;
1974 device->bdev = bdev;
dfe25020 1975 device->in_fs_metadata = 1;
63a212ab 1976 device->is_tgtdev_for_dev_replace = 0;
fb01aa85 1977 device->mode = FMODE_EXCL;
2b82032c 1978 set_blocksize(device->bdev, 4096);
788f20eb 1979
2b82032c
YZ		 1980 if (seeding_dev) {
1981 sb->s_flags &= ~MS_RDONLY;
125ccb0a 1982 ret = btrfs_prepare_sprout(root);
79787eaa 1983 BUG_ON(ret); /* -ENOMEM */
2b82032c 1984 }
788f20eb 1985
2b82032c 1986 device->fs_devices = root->fs_info->fs_devices;
e5e9a520 1987
e5e9a520 1988 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1f78160c 1989 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2b82032c
YZ		 1990 list_add(&device->dev_alloc_list,
1991 &root->fs_info->fs_devices->alloc_list);
1992 root->fs_info->fs_devices->num_devices++;
1993 root->fs_info->fs_devices->open_devices++;
1994 root->fs_info->fs_devices->rw_devices++;
02db0844 1995 root->fs_info->fs_devices->total_devices++;
d5e2003c
JB		 1996 if (device->can_discard)
1997 root->fs_info->fs_devices->num_can_discard++;
2b82032c 1998 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 1999
2bf64758
JB		 2000 spin_lock(&root->fs_info->free_chunk_lock);
2001 root->fs_info->free_chunk_space += device->total_bytes;
2002 spin_unlock(&root->fs_info->free_chunk_lock);
2003
c289811c
CM		 2004 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2005 root->fs_info->fs_devices->rotating = 1;
2006
6c41761f
DS		 2007 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2008 btrfs_set_super_total_bytes(root->fs_info->super_copy,
788f20eb
CM		 2009 total_bytes + device->total_bytes);
2010
6c41761f
DS		 2011 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2012 btrfs_set_super_num_devices(root->fs_info->super_copy,
788f20eb 2013 total_bytes + 1);
e5e9a520 2014 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
788f20eb 2015
2b82032c
YZ		 2016 if (seeding_dev) {
2017 ret = init_first_rw_device(trans, root, device);
005d6427
DS		 2018 if (ret) {
2019 btrfs_abort_transaction(trans, root, ret);
79787eaa 2020 goto error_trans;
005d6427 2021 }
2b82032c 2022 ret = btrfs_finish_sprout(trans, root);
005d6427
DS		 2023 if (ret) {
2024 btrfs_abort_transaction(trans, root, ret);
79787eaa 2025 goto error_trans;
005d6427 2026 }
2b82032c
YZ		 2027 } else {
2028 ret = btrfs_add_device(trans, root, device);
005d6427
DS		 2029 if (ret) {
2030 btrfs_abort_transaction(trans, root, ret);
79787eaa 2031 goto error_trans;
005d6427 2032 }
2b82032c
YZ		 2033 }
2034
913d952e
CM		 2035 /*
2036 * we've got more storage, clear any full flags on the space
2037 * infos
2038 */
2039 btrfs_clear_space_info_full(root->fs_info);
2040
7d9eb12c 2041 unlock_chunks(root);
5af3e8cc
SB		 2042 root->fs_info->num_tolerated_disk_barrier_failures =
2043 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
79787eaa 2044 ret = btrfs_commit_transaction(trans, root);
a2135011 2045
2b82032c
YZ		 2046 if (seeding_dev) {
2047 mutex_unlock(&uuid_mutex);
2048 up_write(&sb->s_umount);
788f20eb 2049
79787eaa
JM		 2050 if (ret) /* transaction commit */
2051 return ret;
2052
2b82032c 2053 ret = btrfs_relocate_sys_chunks(root);
79787eaa
JM		 2054 if (ret < 0)
2055 btrfs_error(root->fs_info, ret,
2056 "Failed to relocate sys chunks after "
2057 "device initialization. This can be fixed "
2058 "using the \"btrfs balance\" command.");
671415b7
MX		 2059 trans = btrfs_attach_transaction(root);
2060 if (IS_ERR(trans)) {
2061 if (PTR_ERR(trans) == -ENOENT)
2062 return 0;
2063 return PTR_ERR(trans);
2064 }
2065 ret = btrfs_commit_transaction(trans, root);
2b82032c 2066 }
c9e9f97b 2067
2b82032c 2068 return ret;
79787eaa
JM		 2069
2070error_trans:
2071 unlock_chunks(root);
79787eaa 2072 btrfs_end_transaction(trans, root);
606686ee 2073 rcu_string_free(device->name);
79787eaa 2074 kfree(device);
2b82032c 2075error:
e525fd89 2076 blkdev_put(bdev, FMODE_EXCL);
2b82032c
YZ		 2077 if (seeding_dev) {
2078 mutex_unlock(&uuid_mutex);
2079 up_write(&sb->s_umount);
2080 }
c9e9f97b 2081 return ret;
788f20eb
CM		 2082}
2083
e93c89c1
SB		 2084int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2085 struct btrfs_device **device_out)
2086{
2087 struct request_queue *q;
2088 struct btrfs_device *device;
2089 struct block_device *bdev;
2090 struct btrfs_fs_info *fs_info = root->fs_info;
2091 struct list_head *devices;
2092 struct rcu_string *name;
2093 int ret = 0;
2094
2095 *device_out = NULL;
2096 if (fs_info->fs_devices->seeding)
2097 return -EINVAL;
2098
2099 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2100 fs_info->bdev_holder);
2101 if (IS_ERR(bdev))
2102 return PTR_ERR(bdev);
2103
2104 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2105
2106 devices = &fs_info->fs_devices->devices;
2107 list_for_each_entry(device, devices, dev_list) {
2108 if (device->bdev == bdev) {
2109 ret = -EEXIST;
2110 goto error;
2111 }
2112 }
2113
2114 device = kzalloc(sizeof(*device), GFP_NOFS);
2115 if (!device) {
2116 ret = -ENOMEM;
2117 goto error;
2118 }
2119
2120 name = rcu_string_strdup(device_path, GFP_NOFS);
2121 if (!name) {
2122 kfree(device);
2123 ret = -ENOMEM;
2124 goto error;
2125 }
2126 rcu_assign_pointer(device->name, name);
2127
2128 q = bdev_get_queue(bdev);
2129 if (blk_queue_discard(q))
2130 device->can_discard = 1;
2131 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2132 device->writeable = 1;
2133 device->work.func = pending_bios_fn;
2134 generate_random_uuid(device->uuid);
2135 device->devid = BTRFS_DEV_REPLACE_DEVID;
2136 spin_lock_init(&device->io_lock);
2137 device->generation = 0;
2138 device->io_width = root->sectorsize;
2139 device->io_align = root->sectorsize;
2140 device->sector_size = root->sectorsize;
2141 device->total_bytes = i_size_read(bdev->bd_inode);
2142 device->disk_total_bytes = device->total_bytes;
2143 device->dev_root = fs_info->dev_root;
2144 device->bdev = bdev;
2145 device->in_fs_metadata = 1;
2146 device->is_tgtdev_for_dev_replace = 1;
2147 device->mode = FMODE_EXCL;
2148 set_blocksize(device->bdev, 4096);
2149 device->fs_devices = fs_info->fs_devices;
2150 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2151 fs_info->fs_devices->num_devices++;
2152 fs_info->fs_devices->open_devices++;
2153 if (device->can_discard)
2154 fs_info->fs_devices->num_can_discard++;
2155 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2156
2157 *device_out = device;
2158 return ret;
2159
2160error:
2161 blkdev_put(bdev, FMODE_EXCL);
2162 return ret;
2163}
2164
2165void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2166 struct btrfs_device *tgtdev)
2167{
2168 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2169 tgtdev->io_width = fs_info->dev_root->sectorsize;
2170 tgtdev->io_align = fs_info->dev_root->sectorsize;
2171 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2172 tgtdev->dev_root = fs_info->dev_root;
2173 tgtdev->in_fs_metadata = 1;
2174}
2175
d397712b
CM		 2176static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2177 struct btrfs_device *device)
0b86a832
CM		 2178{
2179 int ret;
2180 struct btrfs_path *path;
2181 struct btrfs_root *root;
2182 struct btrfs_dev_item *dev_item;
2183 struct extent_buffer *leaf;
2184 struct btrfs_key key;
2185
2186 root = device->dev_root->fs_info->chunk_root;
2187
2188 path = btrfs_alloc_path();
2189 if (!path)
2190 return -ENOMEM;
2191
2192 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2193 key.type = BTRFS_DEV_ITEM_KEY;
2194 key.offset = device->devid;
2195
2196 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2197 if (ret < 0)
2198 goto out;
2199
2200 if (ret > 0) {
2201 ret = -ENOENT;
2202 goto out;
2203 }
2204
2205 leaf = path->nodes[0];
2206 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2207
2208 btrfs_set_device_id(leaf, dev_item, device->devid);
2209 btrfs_set_device_type(leaf, dev_item, device->type);
2210 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2211 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2212 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
d6397bae 2213 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
0b86a832
CM		 2214 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2215 btrfs_mark_buffer_dirty(leaf);
2216
2217out:
2218 btrfs_free_path(path);
2219 return ret;
2220}
2221
7d9eb12c 2222static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 2223 struct btrfs_device *device, u64 new_size)
2224{
2225 struct btrfs_super_block *super_copy =
6c41761f 2226 device->dev_root->fs_info->super_copy;
8f18cf13
CM		 2227 u64 old_total = btrfs_super_total_bytes(super_copy);
2228 u64 diff = new_size - device->total_bytes;
2229
2b82032c
YZ		 2230 if (!device->writeable)
2231 return -EACCES;
63a212ab
SB		 2232 if (new_size <= device->total_bytes ||
2233 device->is_tgtdev_for_dev_replace)
2b82032c
YZ		 2234 return -EINVAL;
2235
8f18cf13 2236 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2b82032c
YZ		 2237 device->fs_devices->total_rw_bytes += diff;
2238
2239 device->total_bytes = new_size;
9779b72f 2240 device->disk_total_bytes = new_size;
4184ea7f
CM		 2241 btrfs_clear_space_info_full(device->dev_root->fs_info);
2242
8f18cf13
CM		 2243 return btrfs_update_device(trans, device);
2244}
2245
7d9eb12c
CM		 2246int btrfs_grow_device(struct btrfs_trans_handle *trans,
2247 struct btrfs_device *device, u64 new_size)
2248{
2249 int ret;
2250 lock_chunks(device->dev_root);
2251 ret = __btrfs_grow_device(trans, device, new_size);
2252 unlock_chunks(device->dev_root);
2253 return ret;
2254}
2255
8f18cf13
CM		 2256static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2257 struct btrfs_root *root,
2258 u64 chunk_tree, u64 chunk_objectid,
2259 u64 chunk_offset)
2260{
2261 int ret;
2262 struct btrfs_path *path;
2263 struct btrfs_key key;
2264
2265 root = root->fs_info->chunk_root;
2266 path = btrfs_alloc_path();
2267 if (!path)
2268 return -ENOMEM;
2269
2270 key.objectid = chunk_objectid;
2271 key.offset = chunk_offset;
2272 key.type = BTRFS_CHUNK_ITEM_KEY;
2273
2274 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
79787eaa
JM		 2275 if (ret < 0)
2276 goto out;
2277 else if (ret > 0) { /* Logic error or corruption */
2278 btrfs_error(root->fs_info, -ENOENT,
2279 "Failed lookup while freeing chunk.");
2280 ret = -ENOENT;
2281 goto out;
2282 }
8f18cf13
CM		 2283
2284 ret = btrfs_del_item(trans, root, path);
79787eaa
JM		 2285 if (ret < 0)
2286 btrfs_error(root->fs_info, ret,
2287 "Failed to delete chunk item.");
2288out:
8f18cf13 2289 btrfs_free_path(path);
65a246c5 2290 return ret;
8f18cf13
CM		 2291}
2292
b2950863 2293static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
8f18cf13
CM		 2294 chunk_offset)
2295{
6c41761f 2296 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
8f18cf13
CM		 2297 struct btrfs_disk_key *disk_key;
2298 struct btrfs_chunk *chunk;
2299 u8 *ptr;
2300 int ret = 0;
2301 u32 num_stripes;
2302 u32 array_size;
2303 u32 len = 0;
2304 u32 cur;
2305 struct btrfs_key key;
2306
2307 array_size = btrfs_super_sys_array_size(super_copy);
2308
2309 ptr = super_copy->sys_chunk_array;
2310 cur = 0;
2311
2312 while (cur < array_size) {
2313 disk_key = (struct btrfs_disk_key *)ptr;
2314 btrfs_disk_key_to_cpu(&key, disk_key);
2315
2316 len = sizeof(*disk_key);
2317
2318 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2319 chunk = (struct btrfs_chunk *)(ptr + len);
2320 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2321 len += btrfs_chunk_item_size(num_stripes);
2322 } else {
2323 ret = -EIO;
2324 break;
2325 }
2326 if (key.objectid == chunk_objectid &&
2327 key.offset == chunk_offset) {
2328 memmove(ptr, ptr + len, array_size - (cur + len));
2329 array_size -= len;
2330 btrfs_set_super_sys_array_size(super_copy, array_size);
2331 } else {
2332 ptr += len;
2333 cur += len;
2334 }
2335 }
2336 return ret;
2337}
2338
b2950863 2339static int btrfs_relocate_chunk(struct btrfs_root *root,
8f18cf13
CM		 2340 u64 chunk_tree, u64 chunk_objectid,
2341 u64 chunk_offset)
2342{
2343 struct extent_map_tree *em_tree;
2344 struct btrfs_root *extent_root;
2345 struct btrfs_trans_handle *trans;
2346 struct extent_map *em;
2347 struct map_lookup *map;
2348 int ret;
2349 int i;
2350
2351 root = root->fs_info->chunk_root;
2352 extent_root = root->fs_info->extent_root;
2353 em_tree = &root->fs_info->mapping_tree.map_tree;
2354
ba1bf481
JB		 2355 ret = btrfs_can_relocate(extent_root, chunk_offset);
2356 if (ret)
2357 return -ENOSPC;
2358
8f18cf13 2359 /* step one, relocate all the extents inside this chunk */
1a40e23b 2360 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
a22285a6
YZ		 2361 if (ret)
2362 return ret;
8f18cf13 2363
a22285a6 2364 trans = btrfs_start_transaction(root, 0);
98d5dc13 2365 BUG_ON(IS_ERR(trans));
8f18cf13 2366
7d9eb12c
CM		 2367 lock_chunks(root);
2368
8f18cf13
CM		 2369 /*
2370 * step two, delete the device extents and the
2371 * chunk tree entries
2372 */
890871be 2373 read_lock(&em_tree->lock);
8f18cf13 2374 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
890871be 2375 read_unlock(&em_tree->lock);
8f18cf13 2376
285190d9 2377 BUG_ON(!em || em->start > chunk_offset ||
a061fc8d 2378 em->start + em->len < chunk_offset);
8f18cf13
CM		 2379 map = (struct map_lookup *)em->bdev;
2380
2381 for (i = 0; i < map->num_stripes; i++) {
2382 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2383 map->stripes[i].physical);
2384 BUG_ON(ret);
a061fc8d 2385
dfe25020
CM		 2386 if (map->stripes[i].dev) {
2387 ret = btrfs_update_device(trans, map->stripes[i].dev);
2388 BUG_ON(ret);
2389 }
8f18cf13
CM		 2390 }
2391 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2392 chunk_offset);
2393
2394 BUG_ON(ret);
2395
1abe9b8a 2396 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2397
8f18cf13
CM		 2398 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2399 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2400 BUG_ON(ret);
8f18cf13
CM		 2401 }
2402
2b82032c
YZ		 2403 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2404 BUG_ON(ret);
2405
890871be 2406 write_lock(&em_tree->lock);
2b82032c 2407 remove_extent_mapping(em_tree, em);
890871be 2408 write_unlock(&em_tree->lock);
2b82032c
YZ		 2409
2410 kfree(map);
2411 em->bdev = NULL;
2412
2413 /* once for the tree */
2414 free_extent_map(em);
2415 /* once for us */
2416 free_extent_map(em);
2417
2418 unlock_chunks(root);
2419 btrfs_end_transaction(trans, root);
2420 return 0;
2421}
2422
2423static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2424{
2425 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2426 struct btrfs_path *path;
2427 struct extent_buffer *leaf;
2428 struct btrfs_chunk *chunk;
2429 struct btrfs_key key;
2430 struct btrfs_key found_key;
2431 u64 chunk_tree = chunk_root->root_key.objectid;
2432 u64 chunk_type;
ba1bf481
JB		 2433 bool retried = false;
2434 int failed = 0;
2b82032c
YZ		 2435 int ret;
2436
2437 path = btrfs_alloc_path();
2438 if (!path)
2439 return -ENOMEM;
2440
ba1bf481 2441again:
2b82032c
YZ		 2442 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2443 key.offset = (u64)-1;
2444 key.type = BTRFS_CHUNK_ITEM_KEY;
2445
2446 while (1) {
2447 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2448 if (ret < 0)
2449 goto error;
79787eaa 2450 BUG_ON(ret == 0); /* Corruption */
2b82032c
YZ		 2451
2452 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2453 key.type);
2454 if (ret < 0)
2455 goto error;
2456 if (ret > 0)
2457 break;
1a40e23b 2458
2b82032c
YZ		 2459 leaf = path->nodes[0];
2460 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 2461
2b82032c
YZ		 2462 chunk = btrfs_item_ptr(leaf, path->slots[0],
2463 struct btrfs_chunk);
2464 chunk_type = btrfs_chunk_type(leaf, chunk);
b3b4aa74 2465 btrfs_release_path(path);
8f18cf13 2466
2b82032c
YZ		 2467 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2468 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2469 found_key.objectid,
2470 found_key.offset);
ba1bf481
JB		 2471 if (ret == -ENOSPC)
2472 failed++;
2473 else if (ret)
2474 BUG();
2b82032c 2475 }
8f18cf13 2476
2b82032c
YZ		 2477 if (found_key.offset == 0)
2478 break;
2479 key.offset = found_key.offset - 1;
2480 }
2481 ret = 0;
ba1bf481
JB		 2482 if (failed && !retried) {
2483 failed = 0;
2484 retried = true;
2485 goto again;
2486 } else if (failed && retried) {
2487 WARN_ON(1);
2488 ret = -ENOSPC;
2489 }
2b82032c
YZ		 2490error:
2491 btrfs_free_path(path);
2492 return ret;
8f18cf13
CM		 2493}
2494
0940ebf6
ID		 2495static int insert_balance_item(struct btrfs_root *root,
2496 struct btrfs_balance_control *bctl)
2497{
2498 struct btrfs_trans_handle *trans;
2499 struct btrfs_balance_item *item;
2500 struct btrfs_disk_balance_args disk_bargs;
2501 struct btrfs_path *path;
2502 struct extent_buffer *leaf;
2503 struct btrfs_key key;
2504 int ret, err;
2505
2506 path = btrfs_alloc_path();
2507 if (!path)
2508 return -ENOMEM;
2509
2510 trans = btrfs_start_transaction(root, 0);
2511 if (IS_ERR(trans)) {
2512 btrfs_free_path(path);
2513 return PTR_ERR(trans);
2514 }
2515
2516 key.objectid = BTRFS_BALANCE_OBJECTID;
2517 key.type = BTRFS_BALANCE_ITEM_KEY;
2518 key.offset = 0;
2519
2520 ret = btrfs_insert_empty_item(trans, root, path, &key,
2521 sizeof(*item));
2522 if (ret)
2523 goto out;
2524
2525 leaf = path->nodes[0];
2526 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2527
2528 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2529
2530 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2531 btrfs_set_balance_data(leaf, item, &disk_bargs);
2532 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2533 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2534 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2535 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2536
2537 btrfs_set_balance_flags(leaf, item, bctl->flags);
2538
2539 btrfs_mark_buffer_dirty(leaf);
2540out:
2541 btrfs_free_path(path);
2542 err = btrfs_commit_transaction(trans, root);
2543 if (err && !ret)
2544 ret = err;
2545 return ret;
2546}
2547
2548static int del_balance_item(struct btrfs_root *root)
2549{
2550 struct btrfs_trans_handle *trans;
2551 struct btrfs_path *path;
2552 struct btrfs_key key;
2553 int ret, err;
2554
2555 path = btrfs_alloc_path();
2556 if (!path)
2557 return -ENOMEM;
2558
2559 trans = btrfs_start_transaction(root, 0);
2560 if (IS_ERR(trans)) {
2561 btrfs_free_path(path);
2562 return PTR_ERR(trans);
2563 }
2564
2565 key.objectid = BTRFS_BALANCE_OBJECTID;
2566 key.type = BTRFS_BALANCE_ITEM_KEY;
2567 key.offset = 0;
2568
2569 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2570 if (ret < 0)
2571 goto out;
2572 if (ret > 0) {
2573 ret = -ENOENT;
2574 goto out;
2575 }
2576
2577 ret = btrfs_del_item(trans, root, path);
2578out:
2579 btrfs_free_path(path);
2580 err = btrfs_commit_transaction(trans, root);
2581 if (err && !ret)
2582 ret = err;
2583 return ret;
2584}
2585
59641015
ID		 2586/*
2587 * This is a heuristic used to reduce the number of chunks balanced on
2588 * resume after balance was interrupted.
2589 */
2590static void update_balance_args(struct btrfs_balance_control *bctl)
2591{
2592 /*
2593 * Turn on soft mode for chunk types that were being converted.
2594 */
2595 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2596 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2597 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2598 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2599 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2600 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2601
2602 /*
2603 * Turn on usage filter if is not already used. The idea is
2604 * that chunks that we have already balanced should be
2605 * reasonably full. Don't do it for chunks that are being
2606 * converted - that will keep us from relocating unconverted
2607 * (albeit full) chunks.
2608 */
2609 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2610 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2611 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2612 bctl->data.usage = 90;
2613 }
2614 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2615 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2616 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2617 bctl->sys.usage = 90;
2618 }
2619 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2620 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2621 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2622 bctl->meta.usage = 90;
2623 }
2624}
2625
c9e9f97b
ID		 2626/*
2627 * Should be called with both balance and volume mutexes held to
2628 * serialize other volume operations (add_dev/rm_dev/resize) with
2629 * restriper. Same goes for unset_balance_control.
2630 */
2631static void set_balance_control(struct btrfs_balance_control *bctl)
2632{
2633 struct btrfs_fs_info *fs_info = bctl->fs_info;
2634
2635 BUG_ON(fs_info->balance_ctl);
2636
2637 spin_lock(&fs_info->balance_lock);
2638 fs_info->balance_ctl = bctl;
2639 spin_unlock(&fs_info->balance_lock);
2640}
2641
2642static void unset_balance_control(struct btrfs_fs_info *fs_info)
2643{
2644 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2645
2646 BUG_ON(!fs_info->balance_ctl);
2647
2648 spin_lock(&fs_info->balance_lock);
2649 fs_info->balance_ctl = NULL;
2650 spin_unlock(&fs_info->balance_lock);
2651
2652 kfree(bctl);
2653}
2654
ed25e9b2
ID		 2655/*
2656 * Balance filters. Return 1 if chunk should be filtered out
2657 * (should not be balanced).
2658 */
899c81ea 2659static int chunk_profiles_filter(u64 chunk_type,
ed25e9b2
ID		 2660 struct btrfs_balance_args *bargs)
2661{
899c81ea
ID		 2662 chunk_type = chunk_to_extended(chunk_type) &
2663 BTRFS_EXTENDED_PROFILE_MASK;
ed25e9b2 2664
899c81ea 2665 if (bargs->profiles & chunk_type)
ed25e9b2
ID		 2666 return 0;
2667
2668 return 1;
2669}
2670
5ce5b3c0
ID		 2671static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2672 struct btrfs_balance_args *bargs)
2673{
2674 struct btrfs_block_group_cache *cache;
2675 u64 chunk_used, user_thresh;
2676 int ret = 1;
2677
2678 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2679 chunk_used = btrfs_block_group_used(&cache->item);
2680
a105bb88
ID		 2681 if (bargs->usage == 0)
2682 user_thresh = 0;
2683 else if (bargs->usage > 100)
2684 user_thresh = cache->key.offset;
2685 else
2686 user_thresh = div_factor_fine(cache->key.offset,
2687 bargs->usage);
2688
5ce5b3c0
ID		 2689 if (chunk_used < user_thresh)
2690 ret = 0;
2691
2692 btrfs_put_block_group(cache);
2693 return ret;
2694}
2695
409d404b
ID		 2696static int chunk_devid_filter(struct extent_buffer *leaf,
2697 struct btrfs_chunk *chunk,
2698 struct btrfs_balance_args *bargs)
2699{
2700 struct btrfs_stripe *stripe;
2701 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2702 int i;
2703
2704 for (i = 0; i < num_stripes; i++) {
2705 stripe = btrfs_stripe_nr(chunk, i);
2706 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2707 return 0;
2708 }
2709
2710 return 1;
2711}
2712
94e60d5a
ID		 2713/* [pstart, pend) */
2714static int chunk_drange_filter(struct extent_buffer *leaf,
2715 struct btrfs_chunk *chunk,
2716 u64 chunk_offset,
2717 struct btrfs_balance_args *bargs)
2718{
2719 struct btrfs_stripe *stripe;
2720 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2721 u64 stripe_offset;
2722 u64 stripe_length;
2723 int factor;
2724 int i;
2725
2726 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2727 return 0;
2728
2729 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2730 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2731 factor = 2;
2732 else
2733 factor = 1;
2734 factor = num_stripes / factor;
2735
2736 for (i = 0; i < num_stripes; i++) {
2737 stripe = btrfs_stripe_nr(chunk, i);
2738 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2739 continue;
2740
2741 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2742 stripe_length = btrfs_chunk_length(leaf, chunk);
2743 do_div(stripe_length, factor);
2744
2745 if (stripe_offset < bargs->pend &&
2746 stripe_offset + stripe_length > bargs->pstart)
2747 return 0;
2748 }
2749
2750 return 1;
2751}
2752
ea67176a
ID		 2753/* [vstart, vend) */
2754static int chunk_vrange_filter(struct extent_buffer *leaf,
2755 struct btrfs_chunk *chunk,
2756 u64 chunk_offset,
2757 struct btrfs_balance_args *bargs)
2758{
2759 if (chunk_offset < bargs->vend &&
2760 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2761 /* at least part of the chunk is inside this vrange */
2762 return 0;
2763
2764 return 1;
2765}
2766
899c81ea 2767static int chunk_soft_convert_filter(u64 chunk_type,
cfa4c961
ID		 2768 struct btrfs_balance_args *bargs)
2769{
2770 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2771 return 0;
2772
899c81ea
ID		 2773 chunk_type = chunk_to_extended(chunk_type) &
2774 BTRFS_EXTENDED_PROFILE_MASK;
cfa4c961 2775
899c81ea 2776 if (bargs->target == chunk_type)
cfa4c961
ID		 2777 return 1;
2778
2779 return 0;
2780}
2781
f43ffb60
ID		 2782static int should_balance_chunk(struct btrfs_root *root,
2783 struct extent_buffer *leaf,
2784 struct btrfs_chunk *chunk, u64 chunk_offset)
2785{
2786 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2787 struct btrfs_balance_args *bargs = NULL;
2788 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2789
2790 /* type filter */
2791 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2792 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2793 return 0;
2794 }
2795
2796 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2797 bargs = &bctl->data;
2798 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2799 bargs = &bctl->sys;
2800 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2801 bargs = &bctl->meta;
2802
ed25e9b2
ID		 2803 /* profiles filter */
2804 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2805 chunk_profiles_filter(chunk_type, bargs)) {
2806 return 0;
5ce5b3c0
ID		 2807 }
2808
2809 /* usage filter */
2810 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2811 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2812 return 0;
409d404b
ID		 2813 }
2814
2815 /* devid filter */
2816 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2817 chunk_devid_filter(leaf, chunk, bargs)) {
2818 return 0;
94e60d5a
ID		 2819 }
2820
2821 /* drange filter, makes sense only with devid filter */
2822 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2823 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2824 return 0;
ea67176a
ID		 2825 }
2826
2827 /* vrange filter */
2828 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2829 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2830 return 0;
ed25e9b2
ID		 2831 }
2832
cfa4c961
ID		 2833 /* soft profile changing mode */
2834 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2835 chunk_soft_convert_filter(chunk_type, bargs)) {
2836 return 0;
2837 }
2838
f43ffb60
ID		 2839 return 1;
2840}
2841
c9e9f97b 2842static int __btrfs_balance(struct btrfs_fs_info *fs_info)
ec44a35c 2843{
19a39dce 2844 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
c9e9f97b
ID		 2845 struct btrfs_root *chunk_root = fs_info->chunk_root;
2846 struct btrfs_root *dev_root = fs_info->dev_root;
2847 struct list_head *devices;
ec44a35c
CM		 2848 struct btrfs_device *device;
2849 u64 old_size;
2850 u64 size_to_free;
f43ffb60 2851 struct btrfs_chunk *chunk;
ec44a35c
CM		 2852 struct btrfs_path *path;
2853 struct btrfs_key key;
ec44a35c 2854 struct btrfs_key found_key;
c9e9f97b 2855 struct btrfs_trans_handle *trans;
f43ffb60
ID		 2856 struct extent_buffer *leaf;
2857 int slot;
c9e9f97b
ID		 2858 int ret;
2859 int enospc_errors = 0;
19a39dce 2860 bool counting = true;
ec44a35c 2861
ec44a35c 2862 /* step one make some room on all the devices */
c9e9f97b 2863 devices = &fs_info->fs_devices->devices;
c6e30871 2864 list_for_each_entry(device, devices, dev_list) {
ec44a35c
CM		 2865 old_size = device->total_bytes;
2866 size_to_free = div_factor(old_size, 1);
2867 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2b82032c 2868 if (!device->writeable ||
63a212ab
SB		 2869 device->total_bytes - device->bytes_used > size_to_free ||
2870 device->is_tgtdev_for_dev_replace)
ec44a35c
CM		 2871 continue;
2872
2873 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 2874 if (ret == -ENOSPC)
2875 break;
ec44a35c
CM		 2876 BUG_ON(ret);
2877
a22285a6 2878 trans = btrfs_start_transaction(dev_root, 0);
98d5dc13 2879 BUG_ON(IS_ERR(trans));
ec44a35c
CM		 2880
2881 ret = btrfs_grow_device(trans, device, old_size);
2882 BUG_ON(ret);
2883
2884 btrfs_end_transaction(trans, dev_root);
2885 }
2886
2887 /* step two, relocate all the chunks */
2888 path = btrfs_alloc_path();
17e9f796
MF		 2889 if (!path) {
2890 ret = -ENOMEM;
2891 goto error;
2892 }
19a39dce
ID		 2893
2894 /* zero out stat counters */
2895 spin_lock(&fs_info->balance_lock);
2896 memset(&bctl->stat, 0, sizeof(bctl->stat));
2897 spin_unlock(&fs_info->balance_lock);
2898again:
ec44a35c
CM		 2899 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2900 key.offset = (u64)-1;
2901 key.type = BTRFS_CHUNK_ITEM_KEY;
2902
d397712b 2903 while (1) {
19a39dce 2904 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
a7e99c69 2905 atomic_read(&fs_info->balance_cancel_req)) {
837d5b6e
ID		 2906 ret = -ECANCELED;
2907 goto error;
2908 }
2909
ec44a35c
CM		 2910 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2911 if (ret < 0)
2912 goto error;
2913
2914 /*
2915 * this shouldn't happen, it means the last relocate
2916 * failed
2917 */
2918 if (ret == 0)
c9e9f97b 2919 BUG(); /* FIXME break ? */
ec44a35c
CM		 2920
2921 ret = btrfs_previous_item(chunk_root, path, 0,
2922 BTRFS_CHUNK_ITEM_KEY);
c9e9f97b
ID		 2923 if (ret) {
2924 ret = 0;
ec44a35c 2925 break;
c9e9f97b 2926 }
7d9eb12c 2927
f43ffb60
ID		 2928 leaf = path->nodes[0];
2929 slot = path->slots[0];
2930 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7d9eb12c 2931
ec44a35c
CM		 2932 if (found_key.objectid != key.objectid)
2933 break;
7d9eb12c 2934
ec44a35c 2935 /* chunk zero is special */
ba1bf481 2936 if (found_key.offset == 0)
ec44a35c
CM		 2937 break;
2938
f43ffb60
ID		 2939 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2940
19a39dce
ID		 2941 if (!counting) {
2942 spin_lock(&fs_info->balance_lock);
2943 bctl->stat.considered++;
2944 spin_unlock(&fs_info->balance_lock);
2945 }
2946
f43ffb60
ID		 2947 ret = should_balance_chunk(chunk_root, leaf, chunk,
2948 found_key.offset);
b3b4aa74 2949 btrfs_release_path(path);
f43ffb60
ID		 2950 if (!ret)
2951 goto loop;
2952
19a39dce
ID		 2953 if (counting) {
2954 spin_lock(&fs_info->balance_lock);
2955 bctl->stat.expected++;
2956 spin_unlock(&fs_info->balance_lock);
2957 goto loop;
2958 }
2959
ec44a35c
CM		 2960 ret = btrfs_relocate_chunk(chunk_root,
2961 chunk_root->root_key.objectid,
2962 found_key.objectid,
2963 found_key.offset);
508794eb
JB		 2964 if (ret && ret != -ENOSPC)
2965 goto error;
19a39dce 2966 if (ret == -ENOSPC) {
c9e9f97b 2967 enospc_errors++;
19a39dce
ID		 2968 } else {
2969 spin_lock(&fs_info->balance_lock);
2970 bctl->stat.completed++;
2971 spin_unlock(&fs_info->balance_lock);
2972 }
f43ffb60 2973loop:
ba1bf481 2974 key.offset = found_key.offset - 1;
ec44a35c 2975 }
c9e9f97b 2976
19a39dce
ID		 2977 if (counting) {
2978 btrfs_release_path(path);
2979 counting = false;
2980 goto again;
2981 }
ec44a35c
CM		 2982error:
2983 btrfs_free_path(path);
c9e9f97b
ID		 2984 if (enospc_errors) {
2985 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2986 enospc_errors);
2987 if (!ret)
2988 ret = -ENOSPC;
2989 }
2990
ec44a35c
CM		 2991 return ret;
2992}
2993
0c460c0d
ID		 2994/**
2995 * alloc_profile_is_valid - see if a given profile is valid and reduced
2996 * @flags: profile to validate
2997 * @extended: if true @flags is treated as an extended profile
2998 */
2999static int alloc_profile_is_valid(u64 flags, int extended)
3000{
3001 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3002 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3003
3004 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3005
3006 /* 1) check that all other bits are zeroed */
3007 if (flags & ~mask)
3008 return 0;
3009
3010 /* 2) see if profile is reduced */
3011 if (flags == 0)
3012 return !extended; /* "0" is valid for usual profiles */
3013
3014 /* true if exactly one bit set */
3015 return (flags & (flags - 1)) == 0;
3016}
3017
837d5b6e
ID		 3018static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3019{
a7e99c69
ID		 3020 /* cancel requested || normal exit path */
3021 return atomic_read(&fs_info->balance_cancel_req) ||
3022 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3023 atomic_read(&fs_info->balance_cancel_req) == 0);
837d5b6e
ID		 3024}
3025
c9e9f97b
ID		 3026static void __cancel_balance(struct btrfs_fs_info *fs_info)
3027{
0940ebf6
ID		 3028 int ret;
3029
c9e9f97b 3030 unset_balance_control(fs_info);
0940ebf6
ID		 3031 ret = del_balance_item(fs_info->tree_root);
3032 BUG_ON(ret);
ed0fb78f
ID		 3033
3034 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
c9e9f97b
ID		 3035}
3036
19a39dce 3037void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
c9e9f97b
ID		 3038 struct btrfs_ioctl_balance_args *bargs);
3039
3040/*
3041 * Should be called with both balance and volume mutexes held
3042 */
3043int btrfs_balance(struct btrfs_balance_control *bctl,
3044 struct btrfs_ioctl_balance_args *bargs)
3045{
3046 struct btrfs_fs_info *fs_info = bctl->fs_info;
f43ffb60 3047 u64 allowed;
e4837f8f 3048 int mixed = 0;
c9e9f97b 3049 int ret;
8dabb742 3050 u64 num_devices;
de98ced9 3051 unsigned seq;
c9e9f97b 3052
837d5b6e 3053 if (btrfs_fs_closing(fs_info) ||
a7e99c69
ID		 3054 atomic_read(&fs_info->balance_pause_req) ||
3055 atomic_read(&fs_info->balance_cancel_req)) {
c9e9f97b
ID		 3056 ret = -EINVAL;
3057 goto out;
3058 }
3059
e4837f8f
ID		 3060 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3061 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3062 mixed = 1;
3063
f43ffb60
ID		 3064 /*
3065 * In case of mixed groups both data and meta should be picked,
3066 * and identical options should be given for both of them.
3067 */
e4837f8f
ID		 3068 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3069 if (mixed && (bctl->flags & allowed)) {
f43ffb60
ID		 3070 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3071 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3072 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3073 printk(KERN_ERR "btrfs: with mixed groups data and "
3074 "metadata balance options must be the same\n");
3075 ret = -EINVAL;
3076 goto out;
3077 }
3078 }
3079
8dabb742
SB		 3080 num_devices = fs_info->fs_devices->num_devices;
3081 btrfs_dev_replace_lock(&fs_info->dev_replace);
3082 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3083 BUG_ON(num_devices < 1);
3084 num_devices--;
3085 }
3086 btrfs_dev_replace_unlock(&fs_info->dev_replace);
e4d8ec0f 3087 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
8dabb742 3088 if (num_devices == 1)
e4d8ec0f 3089 allowed |= BTRFS_BLOCK_GROUP_DUP;
8dabb742 3090 else if (num_devices < 4)
e4d8ec0f
ID		 3091 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3092 else
3093 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3094 BTRFS_BLOCK_GROUP_RAID10);
3095
6728b198
ID		 3096 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3097 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3098 (bctl->data.target & ~allowed))) {
e4d8ec0f
ID		 3099 printk(KERN_ERR "btrfs: unable to start balance with target "
3100 "data profile %llu\n",
3101 (unsigned long long)bctl->data.target);
3102 ret = -EINVAL;
3103 goto out;
3104 }
6728b198
ID		 3105 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3106 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3107 (bctl->meta.target & ~allowed))) {
e4d8ec0f
ID		 3108 printk(KERN_ERR "btrfs: unable to start balance with target "
3109 "metadata profile %llu\n",
3110 (unsigned long long)bctl->meta.target);
3111 ret = -EINVAL;
3112 goto out;
3113 }
6728b198
ID		 3114 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3115 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3116 (bctl->sys.target & ~allowed))) {
e4d8ec0f
ID		 3117 printk(KERN_ERR "btrfs: unable to start balance with target "
3118 "system profile %llu\n",
3119 (unsigned long long)bctl->sys.target);
3120 ret = -EINVAL;
3121 goto out;
3122 }
3123
e4837f8f
ID		 3124 /* allow dup'ed data chunks only in mixed mode */
3125 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
6728b198 3126 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
e4d8ec0f
ID		 3127 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3128 ret = -EINVAL;
3129 goto out;
3130 }
3131
3132 /* allow to reduce meta or sys integrity only if force set */
3133 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3134 BTRFS_BLOCK_GROUP_RAID10;
de98ced9
MX		 3135 do {
3136 seq = read_seqbegin(&fs_info->profiles_lock);
3137
3138 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139 (fs_info->avail_system_alloc_bits & allowed) &&
3140 !(bctl->sys.target & allowed)) ||
3141 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3142 (fs_info->avail_metadata_alloc_bits & allowed) &&
3143 !(bctl->meta.target & allowed))) {
3144 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3145 printk(KERN_INFO "btrfs: force reducing metadata "
3146 "integrity\n");
3147 } else {
3148 printk(KERN_ERR "btrfs: balance will reduce metadata "
3149 "integrity, use force if you want this\n");
3150 ret = -EINVAL;
3151 goto out;
3152 }
e4d8ec0f 3153 }
de98ced9 3154 } while (read_seqretry(&fs_info->profiles_lock, seq));
e4d8ec0f 3155
5af3e8cc
SB		 3156 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3157 int num_tolerated_disk_barrier_failures;
3158 u64 target = bctl->sys.target;
3159
3160 num_tolerated_disk_barrier_failures =
3161 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3162 if (num_tolerated_disk_barrier_failures > 0 &&
3163 (target &
3164 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3165 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3166 num_tolerated_disk_barrier_failures = 0;
3167 else if (num_tolerated_disk_barrier_failures > 1 &&
3168 (target &
3169 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3170 num_tolerated_disk_barrier_failures = 1;
3171
3172 fs_info->num_tolerated_disk_barrier_failures =
3173 num_tolerated_disk_barrier_failures;
3174 }
3175
0940ebf6 3176 ret = insert_balance_item(fs_info->tree_root, bctl);
59641015 3177 if (ret && ret != -EEXIST)
0940ebf6
ID		 3178 goto out;
3179
59641015
ID		 3180 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3181 BUG_ON(ret == -EEXIST);
3182 set_balance_control(bctl);
3183 } else {
3184 BUG_ON(ret != -EEXIST);
3185 spin_lock(&fs_info->balance_lock);
3186 update_balance_args(bctl);
3187 spin_unlock(&fs_info->balance_lock);
3188 }
c9e9f97b 3189
837d5b6e 3190 atomic_inc(&fs_info->balance_running);
c9e9f97b
ID		 3191 mutex_unlock(&fs_info->balance_mutex);
3192
3193 ret = __btrfs_balance(fs_info);
3194
3195 mutex_lock(&fs_info->balance_mutex);
837d5b6e 3196 atomic_dec(&fs_info->balance_running);
c9e9f97b
ID		 3197
3198 if (bargs) {
3199 memset(bargs, 0, sizeof(*bargs));
19a39dce 3200 update_ioctl_balance_args(fs_info, 0, bargs);
c9e9f97b
ID		 3201 }
3202
837d5b6e
ID		 3203 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3204 balance_need_close(fs_info)) {
3205 __cancel_balance(fs_info);
3206 }
3207
5af3e8cc
SB		 3208 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3209 fs_info->num_tolerated_disk_barrier_failures =
3210 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3211 }
3212
837d5b6e 3213 wake_up(&fs_info->balance_wait_q);
c9e9f97b
ID		 3214
3215 return ret;
3216out:
59641015
ID		 3217 if (bctl->flags & BTRFS_BALANCE_RESUME)
3218 __cancel_balance(fs_info);
ed0fb78f 3219 else {
59641015 3220 kfree(bctl);
ed0fb78f
ID		 3221 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3222 }
59641015
ID		 3223 return ret;
3224}
3225
3226static int balance_kthread(void *data)
3227{
2b6ba629 3228 struct btrfs_fs_info *fs_info = data;
9555c6c1 3229 int ret = 0;
59641015
ID		 3230
3231 mutex_lock(&fs_info->volume_mutex);
3232 mutex_lock(&fs_info->balance_mutex);
3233
2b6ba629 3234 if (fs_info->balance_ctl) {
9555c6c1 3235 printk(KERN_INFO "btrfs: continuing balance\n");
2b6ba629 3236 ret = btrfs_balance(fs_info->balance_ctl, NULL);
9555c6c1 3237 }
59641015
ID		 3238
3239 mutex_unlock(&fs_info->balance_mutex);
3240 mutex_unlock(&fs_info->volume_mutex);
2b6ba629 3241
59641015
ID		 3242 return ret;
3243}
3244
2b6ba629
ID		 3245int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3246{
3247 struct task_struct *tsk;
3248
3249 spin_lock(&fs_info->balance_lock);
3250 if (!fs_info->balance_ctl) {
3251 spin_unlock(&fs_info->balance_lock);
3252 return 0;
3253 }
3254 spin_unlock(&fs_info->balance_lock);
3255
3256 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3257 printk(KERN_INFO "btrfs: force skipping balance\n");
3258 return 0;
3259 }
3260
3261 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3262 if (IS_ERR(tsk))
3263 return PTR_ERR(tsk);
3264
3265 return 0;
3266}
3267
68310a5e 3268int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
59641015 3269{
59641015
ID		 3270 struct btrfs_balance_control *bctl;
3271 struct btrfs_balance_item *item;
3272 struct btrfs_disk_balance_args disk_bargs;
3273 struct btrfs_path *path;
3274 struct extent_buffer *leaf;
3275 struct btrfs_key key;
3276 int ret;
3277
3278 path = btrfs_alloc_path();
3279 if (!path)
3280 return -ENOMEM;
3281
59641015
ID		 3282 key.objectid = BTRFS_BALANCE_OBJECTID;
3283 key.type = BTRFS_BALANCE_ITEM_KEY;
3284 key.offset = 0;
3285
68310a5e 3286 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
59641015 3287 if (ret < 0)
68310a5e 3288 goto out;
59641015
ID		 3289 if (ret > 0) { /* ret = -ENOENT; */
3290 ret = 0;
68310a5e
ID		 3291 goto out;
3292 }
3293
3294 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3295 if (!bctl) {
3296 ret = -ENOMEM;
3297 goto out;
59641015
ID		 3298 }
3299
3300 leaf = path->nodes[0];
3301 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3302
68310a5e
ID		 3303 bctl->fs_info = fs_info;
3304 bctl->flags = btrfs_balance_flags(leaf, item);
3305 bctl->flags |= BTRFS_BALANCE_RESUME;
59641015
ID		 3306
3307 btrfs_balance_data(leaf, item, &disk_bargs);
3308 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3309 btrfs_balance_meta(leaf, item, &disk_bargs);
3310 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3311 btrfs_balance_sys(leaf, item, &disk_bargs);
3312 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3313
ed0fb78f
ID		 3314 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3315
68310a5e
ID		 3316 mutex_lock(&fs_info->volume_mutex);
3317 mutex_lock(&fs_info->balance_mutex);
59641015 3318
68310a5e
ID		 3319 set_balance_control(bctl);
3320
3321 mutex_unlock(&fs_info->balance_mutex);
3322 mutex_unlock(&fs_info->volume_mutex);
59641015
ID		 3323out:
3324 btrfs_free_path(path);
ec44a35c
CM		 3325 return ret;
3326}
3327
837d5b6e
ID		 3328int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3329{
3330 int ret = 0;
3331
3332 mutex_lock(&fs_info->balance_mutex);
3333 if (!fs_info->balance_ctl) {
3334 mutex_unlock(&fs_info->balance_mutex);
3335 return -ENOTCONN;
3336 }
3337
3338 if (atomic_read(&fs_info->balance_running)) {
3339 atomic_inc(&fs_info->balance_pause_req);
3340 mutex_unlock(&fs_info->balance_mutex);
3341
3342 wait_event(fs_info->balance_wait_q,
3343 atomic_read(&fs_info->balance_running) == 0);
3344
3345 mutex_lock(&fs_info->balance_mutex);
3346 /* we are good with balance_ctl ripped off from under us */
3347 BUG_ON(atomic_read(&fs_info->balance_running));
3348 atomic_dec(&fs_info->balance_pause_req);
3349 } else {
3350 ret = -ENOTCONN;
3351 }
3352
3353 mutex_unlock(&fs_info->balance_mutex);
3354 return ret;
3355}
3356
a7e99c69
ID		 3357int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3358{
3359 mutex_lock(&fs_info->balance_mutex);
3360 if (!fs_info->balance_ctl) {
3361 mutex_unlock(&fs_info->balance_mutex);
3362 return -ENOTCONN;
3363 }
3364
3365 atomic_inc(&fs_info->balance_cancel_req);
3366 /*
3367 * if we are running just wait and return, balance item is
3368 * deleted in btrfs_balance in this case
3369 */
3370 if (atomic_read(&fs_info->balance_running)) {
3371 mutex_unlock(&fs_info->balance_mutex);
3372 wait_event(fs_info->balance_wait_q,
3373 atomic_read(&fs_info->balance_running) == 0);
3374 mutex_lock(&fs_info->balance_mutex);
3375 } else {
3376 /* __cancel_balance needs volume_mutex */
3377 mutex_unlock(&fs_info->balance_mutex);
3378 mutex_lock(&fs_info->volume_mutex);
3379 mutex_lock(&fs_info->balance_mutex);
3380
3381 if (fs_info->balance_ctl)
3382 __cancel_balance(fs_info);
3383
3384 mutex_unlock(&fs_info->volume_mutex);
3385 }
3386
3387 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3388 atomic_dec(&fs_info->balance_cancel_req);
3389 mutex_unlock(&fs_info->balance_mutex);
3390 return 0;
3391}
3392
8f18cf13
CM		 3393/*
3394 * shrinking a device means finding all of the device extents past
3395 * the new size, and then following the back refs to the chunks.
3396 * The chunk relocation code actually frees the device extent
3397 */
3398int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3399{
3400 struct btrfs_trans_handle *trans;
3401 struct btrfs_root *root = device->dev_root;
3402 struct btrfs_dev_extent *dev_extent = NULL;
3403 struct btrfs_path *path;
3404 u64 length;
3405 u64 chunk_tree;
3406 u64 chunk_objectid;
3407 u64 chunk_offset;
3408 int ret;
3409 int slot;
ba1bf481
JB		 3410 int failed = 0;
3411 bool retried = false;
8f18cf13
CM		 3412 struct extent_buffer *l;
3413 struct btrfs_key key;
6c41761f 3414 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
8f18cf13 3415 u64 old_total = btrfs_super_total_bytes(super_copy);
ba1bf481 3416 u64 old_size = device->total_bytes;
8f18cf13
CM		 3417 u64 diff = device->total_bytes - new_size;
3418
63a212ab
SB		 3419 if (device->is_tgtdev_for_dev_replace)
3420 return -EINVAL;
3421
8f18cf13
CM		 3422 path = btrfs_alloc_path();
3423 if (!path)
3424 return -ENOMEM;
3425
8f18cf13
CM		 3426 path->reada = 2;
3427
7d9eb12c
CM		 3428 lock_chunks(root);
3429
8f18cf13 3430 device->total_bytes = new_size;
2bf64758 3431 if (device->writeable) {
2b82032c 3432 device->fs_devices->total_rw_bytes -= diff;
2bf64758
JB		 3433 spin_lock(&root->fs_info->free_chunk_lock);
3434 root->fs_info->free_chunk_space -= diff;
3435 spin_unlock(&root->fs_info->free_chunk_lock);
3436 }
7d9eb12c 3437 unlock_chunks(root);
8f18cf13 3438
ba1bf481 3439again:
8f18cf13
CM		 3440 key.objectid = device->devid;
3441 key.offset = (u64)-1;
3442 key.type = BTRFS_DEV_EXTENT_KEY;
3443
213e64da 3444 do {
8f18cf13
CM		 3445 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3446 if (ret < 0)
3447 goto done;
3448
3449 ret = btrfs_previous_item(root, path, 0, key.type);
3450 if (ret < 0)
3451 goto done;
3452 if (ret) {
3453 ret = 0;
b3b4aa74 3454 btrfs_release_path(path);
bf1fb512 3455 break;
8f18cf13
CM		 3456 }
3457
3458 l = path->nodes[0];
3459 slot = path->slots[0];
3460 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3461
ba1bf481 3462 if (key.objectid != device->devid) {
b3b4aa74 3463 btrfs_release_path(path);
bf1fb512 3464 break;
ba1bf481 3465 }
8f18cf13
CM		 3466
3467 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3468 length = btrfs_dev_extent_length(l, dev_extent);
3469
ba1bf481 3470 if (key.offset + length <= new_size) {
b3b4aa74 3471 btrfs_release_path(path);
d6397bae 3472 break;
ba1bf481 3473 }
8f18cf13
CM		 3474
3475 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3476 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3477 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
b3b4aa74 3478 btrfs_release_path(path);
8f18cf13
CM		 3479
3480 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3481 chunk_offset);
ba1bf481 3482 if (ret && ret != -ENOSPC)
8f18cf13 3483 goto done;
ba1bf481
JB		 3484 if (ret == -ENOSPC)
3485 failed++;
213e64da 3486 } while (key.offset-- > 0);
ba1bf481
JB		 3487
3488 if (failed && !retried) {
3489 failed = 0;
3490 retried = true;
3491 goto again;
3492 } else if (failed && retried) {
3493 ret = -ENOSPC;
3494 lock_chunks(root);
3495
3496 device->total_bytes = old_size;
3497 if (device->writeable)
3498 device->fs_devices->total_rw_bytes += diff;
2bf64758
JB		 3499 spin_lock(&root->fs_info->free_chunk_lock);
3500 root->fs_info->free_chunk_space += diff;
3501 spin_unlock(&root->fs_info->free_chunk_lock);
ba1bf481
JB		 3502 unlock_chunks(root);
3503 goto done;
8f18cf13
CM		 3504 }
3505
d6397bae 3506 /* Shrinking succeeded, else we would be at "done". */
a22285a6 3507 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 3508 if (IS_ERR(trans)) {
3509 ret = PTR_ERR(trans);
3510 goto done;
3511 }
3512
d6397bae
CB		 3513 lock_chunks(root);
3514
3515 device->disk_total_bytes = new_size;
3516 /* Now btrfs_update_device() will change the on-disk size. */
3517 ret = btrfs_update_device(trans, device);
3518 if (ret) {
3519 unlock_chunks(root);
3520 btrfs_end_transaction(trans, root);
3521 goto done;
3522 }
3523 WARN_ON(diff > old_total);
3524 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3525 unlock_chunks(root);
3526 btrfs_end_transaction(trans, root);
8f18cf13
CM		 3527done:
3528 btrfs_free_path(path);
3529 return ret;
3530}
3531
125ccb0a 3532static int btrfs_add_system_chunk(struct btrfs_root *root,
0b86a832
CM		 3533 struct btrfs_key *key,
3534 struct btrfs_chunk *chunk, int item_size)
3535{
6c41761f 3536 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
0b86a832
CM		 3537 struct btrfs_disk_key disk_key;
3538 u32 array_size;
3539 u8 *ptr;
3540
3541 array_size = btrfs_super_sys_array_size(super_copy);
3542 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3543 return -EFBIG;
3544
3545 ptr = super_copy->sys_chunk_array + array_size;
3546 btrfs_cpu_key_to_disk(&disk_key, key);
3547 memcpy(ptr, &disk_key, sizeof(disk_key));
3548 ptr += sizeof(disk_key);
3549 memcpy(ptr, chunk, item_size);
3550 item_size += sizeof(disk_key);
3551 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3552 return 0;
3553}
3554
73c5de00
AJ		 3555/*
3556 * sort the devices in descending order by max_avail, total_avail
3557 */
3558static int btrfs_cmp_device_info(const void *a, const void *b)
9b3f68b9 3559{
73c5de00
AJ		 3560 const struct btrfs_device_info *di_a = a;
3561 const struct btrfs_device_info *di_b = b;
9b3f68b9 3562
73c5de00 3563 if (di_a->max_avail > di_b->max_avail)
b2117a39 3564 return -1;
73c5de00 3565 if (di_a->max_avail < di_b->max_avail)
b2117a39 3566 return 1;
73c5de00
AJ		 3567 if (di_a->total_avail > di_b->total_avail)
3568 return -1;
3569 if (di_a->total_avail < di_b->total_avail)
3570 return 1;
3571 return 0;
b2117a39 3572}
0b86a832 3573
31e50229 3574struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
e6ec716f
MX		 3575 [BTRFS_RAID_RAID10] = {
3576 .sub_stripes = 2,
3577 .dev_stripes = 1,
3578 .devs_max = 0, /* 0 == as many as possible */
3579 .devs_min = 4,
3580 .devs_increment = 2,
3581 .ncopies = 2,
3582 },
3583 [BTRFS_RAID_RAID1] = {
3584 .sub_stripes = 1,
3585 .dev_stripes = 1,
3586 .devs_max = 2,
3587 .devs_min = 2,
3588 .devs_increment = 2,
3589 .ncopies = 2,
3590 },
3591 [BTRFS_RAID_DUP] = {
3592 .sub_stripes = 1,
3593 .dev_stripes = 2,
3594 .devs_max = 1,
3595 .devs_min = 1,
3596 .devs_increment = 1,
3597 .ncopies = 2,
3598 },
3599 [BTRFS_RAID_RAID0] = {
3600 .sub_stripes = 1,
3601 .dev_stripes = 1,
3602 .devs_max = 0,
3603 .devs_min = 2,
3604 .devs_increment = 1,
3605 .ncopies = 1,
3606 },
3607 [BTRFS_RAID_SINGLE] = {
3608 .sub_stripes = 1,
3609 .dev_stripes = 1,
3610 .devs_max = 1,
3611 .devs_min = 1,
3612 .devs_increment = 1,
3613 .ncopies = 1,
3614 },
31e50229 3615};
e6ec716f 3616
73c5de00
AJ		 3617static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3618 struct btrfs_root *extent_root,
3619 struct map_lookup **map_ret,
3620 u64 *num_bytes_out, u64 *stripe_size_out,
3621 u64 start, u64 type)
b2117a39 3622{
73c5de00
AJ		 3623 struct btrfs_fs_info *info = extent_root->fs_info;
3624 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3625 struct list_head *cur;
3626 struct map_lookup *map = NULL;
3627 struct extent_map_tree *em_tree;
3628 struct extent_map *em;
3629 struct btrfs_device_info *devices_info = NULL;
3630 u64 total_avail;
3631 int num_stripes; /* total number of stripes to allocate */
3632 int sub_stripes; /* sub_stripes info for map */
3633 int dev_stripes; /* stripes per dev */
3634 int devs_max; /* max devs to use */
3635 int devs_min; /* min devs needed */
3636 int devs_increment; /* ndevs has to be a multiple of this */
3637 int ncopies; /* how many copies to data has */
3638 int ret;
3639 u64 max_stripe_size;
3640 u64 max_chunk_size;
3641 u64 stripe_size;
3642 u64 num_bytes;
3643 int ndevs;
3644 int i;
3645 int j;
31e50229 3646 int index;
593060d7 3647
0c460c0d 3648 BUG_ON(!alloc_profile_is_valid(type, 0));
9b3f68b9 3649
73c5de00
AJ		 3650 if (list_empty(&fs_devices->alloc_list))
3651 return -ENOSPC;
b2117a39 3652
31e50229 3653 index = __get_raid_index(type);
73c5de00 3654
31e50229
LB		 3655 sub_stripes = btrfs_raid_array[index].sub_stripes;
3656 dev_stripes = btrfs_raid_array[index].dev_stripes;
3657 devs_max = btrfs_raid_array[index].devs_max;
3658 devs_min = btrfs_raid_array[index].devs_min;
3659 devs_increment = btrfs_raid_array[index].devs_increment;
3660 ncopies = btrfs_raid_array[index].ncopies;
b2117a39 3661
9b3f68b9 3662 if (type & BTRFS_BLOCK_GROUP_DATA) {
73c5de00
AJ		 3663 max_stripe_size = 1024 * 1024 * 1024;
3664 max_chunk_size = 10 * max_stripe_size;
9b3f68b9 3665 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1100373f
CM		 3666 /* for larger filesystems, use larger metadata chunks */
3667 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3668 max_stripe_size = 1024 * 1024 * 1024;
3669 else
3670 max_stripe_size = 256 * 1024 * 1024;
73c5de00 3671 max_chunk_size = max_stripe_size;
a40a90a0 3672 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
96bdc7dc 3673 max_stripe_size = 32 * 1024 * 1024;
73c5de00
AJ		 3674 max_chunk_size = 2 * max_stripe_size;
3675 } else {
3676 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3677 type);
3678 BUG_ON(1);
9b3f68b9
CM		 3679 }
3680
2b82032c
YZ		 3681 /* we don't want a chunk larger than 10% of writeable space */
3682 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3683 max_chunk_size);
9b3f68b9 3684
73c5de00
AJ		 3685 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3686 GFP_NOFS);
3687 if (!devices_info)
3688 return -ENOMEM;
0cad8a11 3689
73c5de00 3690 cur = fs_devices->alloc_list.next;
9b3f68b9 3691
9f680ce0 3692 /*
73c5de00
AJ		 3693 * in the first pass through the devices list, we gather information
3694 * about the available holes on each device.
9f680ce0 3695 */
73c5de00
AJ		 3696 ndevs = 0;
3697 while (cur != &fs_devices->alloc_list) {
3698 struct btrfs_device *device;
3699 u64 max_avail;
3700 u64 dev_offset;
b2117a39 3701
73c5de00 3702 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
9f680ce0 3703
73c5de00 3704 cur = cur->next;
b2117a39 3705
73c5de00 3706 if (!device->writeable) {
31b1a2bd 3707 WARN(1, KERN_ERR
73c5de00 3708 "btrfs: read-only device in alloc_list\n");
73c5de00
AJ		 3709 continue;
3710 }
b2117a39 3711
63a212ab
SB		 3712 if (!device->in_fs_metadata ||
3713 device->is_tgtdev_for_dev_replace)
73c5de00 3714 continue;
b2117a39 3715
73c5de00
AJ		 3716 if (device->total_bytes > device->bytes_used)
3717 total_avail = device->total_bytes - device->bytes_used;
3718 else
3719 total_avail = 0;
38c01b96 3720
3721 /* If there is no space on this device, skip it. */
3722 if (total_avail == 0)
3723 continue;
b2117a39 3724
125ccb0a 3725 ret = find_free_dev_extent(device,
73c5de00
AJ		 3726 max_stripe_size * dev_stripes,
3727 &dev_offset, &max_avail);
3728 if (ret && ret != -ENOSPC)
3729 goto error;
b2117a39 3730
73c5de00
AJ		 3731 if (ret == 0)
3732 max_avail = max_stripe_size * dev_stripes;
b2117a39 3733
73c5de00
AJ		 3734 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3735 continue;
b2117a39 3736
73c5de00
AJ		 3737 devices_info[ndevs].dev_offset = dev_offset;
3738 devices_info[ndevs].max_avail = max_avail;
3739 devices_info[ndevs].total_avail = total_avail;
3740 devices_info[ndevs].dev = device;
3741 ++ndevs;
8dabb742 3742 WARN_ON(ndevs > fs_devices->rw_devices);
73c5de00 3743 }
b2117a39 3744
73c5de00
AJ		 3745 /*
3746 * now sort the devices by hole size / available space
3747 */
3748 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3749 btrfs_cmp_device_info, NULL);
b2117a39 3750
73c5de00
AJ		 3751 /* round down to number of usable stripes */
3752 ndevs -= ndevs % devs_increment;
b2117a39 3753
73c5de00
AJ		 3754 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3755 ret = -ENOSPC;
3756 goto error;
b2117a39 3757 }
9f680ce0 3758
73c5de00
AJ		 3759 if (devs_max && ndevs > devs_max)
3760 ndevs = devs_max;
3761 /*
3762 * the primary goal is to maximize the number of stripes, so use as many
3763 * devices as possible, even if the stripes are not maximum sized.
3764 */
3765 stripe_size = devices_info[ndevs-1].max_avail;
3766 num_stripes = ndevs * dev_stripes;
b2117a39 3767
37db63a4 3768 if (stripe_size * ndevs > max_chunk_size * ncopies) {
73c5de00 3769 stripe_size = max_chunk_size * ncopies;
37db63a4 3770 do_div(stripe_size, ndevs);
b2117a39 3771 }
b2117a39 3772
73c5de00 3773 do_div(stripe_size, dev_stripes);
37db63a4
ID		 3774
3775 /* align to BTRFS_STRIPE_LEN */
73c5de00
AJ		 3776 do_div(stripe_size, BTRFS_STRIPE_LEN);
3777 stripe_size *= BTRFS_STRIPE_LEN;
b2117a39
MX		 3778
3779 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3780 if (!map) {
3781 ret = -ENOMEM;
3782 goto error;
3783 }
3784 map->num_stripes = num_stripes;
9b3f68b9 3785
73c5de00
AJ		 3786 for (i = 0; i < ndevs; ++i) {
3787 for (j = 0; j < dev_stripes; ++j) {
3788 int s = i * dev_stripes + j;
3789 map->stripes[s].dev = devices_info[i].dev;
3790 map->stripes[s].physical = devices_info[i].dev_offset +
3791 j * stripe_size;
6324fbf3 3792 }
6324fbf3 3793 }
2b82032c 3794 map->sector_size = extent_root->sectorsize;
b2117a39
MX		 3795 map->stripe_len = BTRFS_STRIPE_LEN;
3796 map->io_align = BTRFS_STRIPE_LEN;
3797 map->io_width = BTRFS_STRIPE_LEN;
2b82032c 3798 map->type = type;
2b82032c 3799 map->sub_stripes = sub_stripes;
0b86a832 3800
2b82032c 3801 *map_ret = map;
73c5de00 3802 num_bytes = stripe_size * (num_stripes / ncopies);
0b86a832 3803
73c5de00
AJ		 3804 *stripe_size_out = stripe_size;
3805 *num_bytes_out = num_bytes;
0b86a832 3806
73c5de00 3807 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
1abe9b8a 3808
172ddd60 3809 em = alloc_extent_map();
2b82032c 3810 if (!em) {
b2117a39
MX		 3811 ret = -ENOMEM;
3812 goto error;
593060d7 3813 }
2b82032c
YZ		 3814 em->bdev = (struct block_device *)map;
3815 em->start = start;
73c5de00 3816 em->len = num_bytes;
2b82032c
YZ		 3817 em->block_start = 0;
3818 em->block_len = em->len;
593060d7 3819
2b82032c 3820 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
890871be 3821 write_lock(&em_tree->lock);
2b82032c 3822 ret = add_extent_mapping(em_tree, em);
890871be 3823 write_unlock(&em_tree->lock);
0f5d42b2
JB		 3824 if (ret) {
3825 free_extent_map(em);
1dd4602f 3826 goto error;
0f5d42b2 3827 }
0b86a832 3828
73c5de00
AJ		 3829 for (i = 0; i < map->num_stripes; ++i) {
3830 struct btrfs_device *device;
3831 u64 dev_offset;
3832
3833 device = map->stripes[i].dev;
3834 dev_offset = map->stripes[i].physical;
0b86a832
CM		 3835
3836 ret = btrfs_alloc_dev_extent(trans, device,
2b82032c
YZ		 3837 info->chunk_root->root_key.objectid,
3838 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
73c5de00 3839 start, dev_offset, stripe_size);
04487488
JB		 3840 if (ret)
3841 goto error_dev_extent;
3842 }
3843
3844 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3845 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3846 start, num_bytes);
3847 if (ret) {
3848 i = map->num_stripes - 1;
3849 goto error_dev_extent;
2b82032c
YZ		 3850 }
3851
0f5d42b2 3852 free_extent_map(em);
b2117a39 3853 kfree(devices_info);
2b82032c 3854 return 0;
b2117a39 3855
04487488
JB		 3856error_dev_extent:
3857 for (; i >= 0; i--) {
3858 struct btrfs_device *device;
3859 int err;
3860
3861 device = map->stripes[i].dev;
3862 err = btrfs_free_dev_extent(trans, device, start);
3863 if (err) {
3864 btrfs_abort_transaction(trans, extent_root, err);
3865 break;
3866 }
3867 }
0f5d42b2
JB		 3868 write_lock(&em_tree->lock);
3869 remove_extent_mapping(em_tree, em);
3870 write_unlock(&em_tree->lock);
3871
3872 /* One for our allocation */
3873 free_extent_map(em);
3874 /* One for the tree reference */
3875 free_extent_map(em);
b2117a39
MX		 3876error:
3877 kfree(map);
3878 kfree(devices_info);
3879 return ret;
2b82032c
YZ		 3880}
3881
3882static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3883 struct btrfs_root *extent_root,
3884 struct map_lookup *map, u64 chunk_offset,
3885 u64 chunk_size, u64 stripe_size)
3886{
3887 u64 dev_offset;
3888 struct btrfs_key key;
3889 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3890 struct btrfs_device *device;
3891 struct btrfs_chunk *chunk;
3892 struct btrfs_stripe *stripe;
3893 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3894 int index = 0;
3895 int ret;
3896
3897 chunk = kzalloc(item_size, GFP_NOFS);
3898 if (!chunk)
3899 return -ENOMEM;
3900
3901 index = 0;
3902 while (index < map->num_stripes) {
3903 device = map->stripes[index].dev;
3904 device->bytes_used += stripe_size;
0b86a832 3905 ret = btrfs_update_device(trans, device);
3acd3953
MF		 3906 if (ret)
3907 goto out_free;
2b82032c
YZ		 3908 index++;
3909 }
3910
2bf64758
JB		 3911 spin_lock(&extent_root->fs_info->free_chunk_lock);
3912 extent_root->fs_info->free_chunk_space -= (stripe_size *
3913 map->num_stripes);
3914 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3915
2b82032c
YZ		 3916 index = 0;
3917 stripe = &chunk->stripe;
3918 while (index < map->num_stripes) {
3919 device = map->stripes[index].dev;
3920 dev_offset = map->stripes[index].physical;
0b86a832 3921
e17cade2
CM		 3922 btrfs_set_stack_stripe_devid(stripe, device->devid);
3923 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3924 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 3925 stripe++;
0b86a832
CM		 3926 index++;
3927 }
3928
2b82032c 3929 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 3930 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 3931 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3932 btrfs_set_stack_chunk_type(chunk, map->type);
3933 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3934 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3935 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b86a832 3936 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2b82032c 3937 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 3938
2b82032c
YZ		 3939 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3940 key.type = BTRFS_CHUNK_ITEM_KEY;
3941 key.offset = chunk_offset;
0b86a832 3942
2b82032c 3943 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
0b86a832 3944
4ed1d16e
MF		 3945 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3946 /*
3947 * TODO: Cleanup of inserted chunk root in case of
3948 * failure.
3949 */
125ccb0a 3950 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
2b82032c 3951 item_size);
8f18cf13 3952 }
1abe9b8a 3953
3acd3953 3954out_free:
0b86a832 3955 kfree(chunk);
4ed1d16e 3956 return ret;
2b82032c 3957}
0b86a832 3958
2b82032c
YZ		 3959/*
3960 * Chunk allocation falls into two parts. The first part does works
3961 * that make the new allocated chunk useable, but not do any operation
3962 * that modifies the chunk tree. The second part does the works that
3963 * require modifying the chunk tree. This division is important for the
3964 * bootstrap process of adding storage to a seed btrfs.
3965 */
3966int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3967 struct btrfs_root *extent_root, u64 type)
3968{
3969 u64 chunk_offset;
3970 u64 chunk_size;
3971 u64 stripe_size;
3972 struct map_lookup *map;
3973 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3974 int ret;
3975
3976 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3977 &chunk_offset);
3978 if (ret)
3979 return ret;
3980
3981 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3982 &stripe_size, chunk_offset, type);
3983 if (ret)
3984 return ret;
3985
3986 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3987 chunk_size, stripe_size);
79787eaa
JM		 3988 if (ret)
3989 return ret;
2b82032c
YZ		 3990 return 0;
3991}
3992
d397712b 3993static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2b82032c
YZ		 3994 struct btrfs_root *root,
3995 struct btrfs_device *device)
3996{
3997 u64 chunk_offset;
3998 u64 sys_chunk_offset;
3999 u64 chunk_size;
4000 u64 sys_chunk_size;
4001 u64 stripe_size;
4002 u64 sys_stripe_size;
4003 u64 alloc_profile;
4004 struct map_lookup *map;
4005 struct map_lookup *sys_map;
4006 struct btrfs_fs_info *fs_info = root->fs_info;
4007 struct btrfs_root *extent_root = fs_info->extent_root;
4008 int ret;
4009
4010 ret = find_next_chunk(fs_info->chunk_root,
4011 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
92b8e897
MF		 4012 if (ret)
4013 return ret;
2b82032c 4014
de98ced9 4015 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
2b82032c
YZ		 4016 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4017 &stripe_size, chunk_offset, alloc_profile);
79787eaa
JM		 4018 if (ret)
4019 return ret;
2b82032c
YZ		 4020
4021 sys_chunk_offset = chunk_offset + chunk_size;
4022
de98ced9 4023 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
2b82032c
YZ		 4024 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4025 &sys_chunk_size, &sys_stripe_size,
4026 sys_chunk_offset, alloc_profile);
005d6427
DS		 4027 if (ret) {
4028 btrfs_abort_transaction(trans, root, ret);
4029 goto out;
4030 }
2b82032c
YZ		 4031
4032 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
005d6427
DS		 4033 if (ret) {
4034 btrfs_abort_transaction(trans, root, ret);
4035 goto out;
4036 }
2b82032c
YZ		 4037
4038 /*
4039 * Modifying chunk tree needs allocating new blocks from both
4040 * system block group and metadata block group. So we only can
4041 * do operations require modifying the chunk tree after both
4042 * block groups were created.
4043 */
4044 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4045 chunk_size, stripe_size);
005d6427
DS		 4046 if (ret) {
4047 btrfs_abort_transaction(trans, root, ret);
4048 goto out;
4049 }
2b82032c
YZ		 4050
4051 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4052 sys_chunk_offset, sys_chunk_size,
4053 sys_stripe_size);
79787eaa 4054 if (ret)
005d6427 4055 btrfs_abort_transaction(trans, root, ret);
79787eaa 4056
005d6427 4057out:
79787eaa 4058
79787eaa 4059 return ret;
2b82032c
YZ		 4060}
4061
4062int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4063{
4064 struct extent_map *em;
4065 struct map_lookup *map;
4066 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4067 int readonly = 0;
4068 int i;
4069
890871be 4070 read_lock(&map_tree->map_tree.lock);
2b82032c 4071 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
890871be 4072 read_unlock(&map_tree->map_tree.lock);
2b82032c
YZ		 4073 if (!em)
4074 return 1;
4075
f48b9075
JB		 4076 if (btrfs_test_opt(root, DEGRADED)) {
4077 free_extent_map(em);
4078 return 0;
4079 }
4080
2b82032c
YZ		 4081 map = (struct map_lookup *)em->bdev;
4082 for (i = 0; i < map->num_stripes; i++) {
4083 if (!map->stripes[i].dev->writeable) {
4084 readonly = 1;
4085 break;
4086 }
4087 }
0b86a832 4088 free_extent_map(em);
2b82032c 4089 return readonly;
0b86a832
CM		 4090}
4091
4092void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4093{
a8067e02 4094 extent_map_tree_init(&tree->map_tree);
0b86a832
CM		 4095}
4096
4097void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4098{
4099 struct extent_map *em;
4100
d397712b 4101 while (1) {
890871be 4102 write_lock(&tree->map_tree.lock);
0b86a832
CM		 4103 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4104 if (em)
4105 remove_extent_mapping(&tree->map_tree, em);
890871be 4106 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 4107 if (!em)
4108 break;
4109 kfree(em->bdev);
4110 /* once for us */
4111 free_extent_map(em);
4112 /* once for the tree */
4113 free_extent_map(em);
4114 }
4115}
4116
5d964051 4117int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
f188591e 4118{
5d964051 4119 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
f188591e
CM		 4120 struct extent_map *em;
4121 struct map_lookup *map;
4122 struct extent_map_tree *em_tree = &map_tree->map_tree;
4123 int ret;
4124
890871be 4125 read_lock(&em_tree->lock);
f188591e 4126 em = lookup_extent_mapping(em_tree, logical, len);
890871be 4127 read_unlock(&em_tree->lock);
f188591e
CM		 4128 BUG_ON(!em);
4129
4130 BUG_ON(em->start > logical || em->start + em->len < logical);
4131 map = (struct map_lookup *)em->bdev;
4132 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4133 ret = map->num_stripes;
321aecc6
CM		 4134 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4135 ret = map->sub_stripes;
f188591e
CM		 4136 else
4137 ret = 1;
4138 free_extent_map(em);
ad6d620e
SB		 4139
4140 btrfs_dev_replace_lock(&fs_info->dev_replace);
4141 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4142 ret++;
4143 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4144
f188591e
CM		 4145 return ret;
4146}
4147
30d9861f
SB		 4148static int find_live_mirror(struct btrfs_fs_info *fs_info,
4149 struct map_lookup *map, int first, int num,
4150 int optimal, int dev_replace_is_ongoing)
dfe25020
CM		 4151{
4152 int i;
30d9861f
SB		 4153 int tolerance;
4154 struct btrfs_device *srcdev;
4155
4156 if (dev_replace_is_ongoing &&
4157 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4158 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4159 srcdev = fs_info->dev_replace.srcdev;
4160 else
4161 srcdev = NULL;
4162
4163 /*
4164 * try to avoid the drive that is the source drive for a
4165 * dev-replace procedure, only choose it if no other non-missing
4166 * mirror is available
4167 */
4168 for (tolerance = 0; tolerance < 2; tolerance++) {
4169 if (map->stripes[optimal].dev->bdev &&
4170 (tolerance || map->stripes[optimal].dev != srcdev))
4171 return optimal;
4172 for (i = first; i < first + num; i++) {
4173 if (map->stripes[i].dev->bdev &&
4174 (tolerance || map->stripes[i].dev != srcdev))
4175 return i;
4176 }
dfe25020 4177 }
30d9861f 4178
dfe25020
CM		 4179 /* we couldn't find one that doesn't fail. Just return something
4180 * and the io error handling code will clean up eventually
4181 */
4182 return optimal;
4183}
4184
3ec706c8 4185static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
f2d8d74d 4186 u64 logical, u64 *length,
a1d3c478 4187 struct btrfs_bio **bbio_ret,
7eaceacc 4188 int mirror_num)
0b86a832
CM		 4189{
4190 struct extent_map *em;
4191 struct map_lookup *map;
3ec706c8 4192 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
0b86a832
CM		 4193 struct extent_map_tree *em_tree = &map_tree->map_tree;
4194 u64 offset;
593060d7 4195 u64 stripe_offset;
fce3bb9a 4196 u64 stripe_end_offset;
593060d7 4197 u64 stripe_nr;
fce3bb9a
LD		 4198 u64 stripe_nr_orig;
4199 u64 stripe_nr_end;
593060d7 4200 int stripe_index;
cea9e445 4201 int i;
de11cc12 4202 int ret = 0;
f2d8d74d 4203 int num_stripes;
a236aed1 4204 int max_errors = 0;
a1d3c478 4205 struct btrfs_bio *bbio = NULL;
472262f3
SB		 4206 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4207 int dev_replace_is_ongoing = 0;
4208 int num_alloc_stripes;
ad6d620e
SB		 4209 int patch_the_first_stripe_for_dev_replace = 0;
4210 u64 physical_to_patch_in_first_stripe = 0;
0b86a832 4211
890871be 4212 read_lock(&em_tree->lock);
0b86a832 4213 em = lookup_extent_mapping(em_tree, logical, *length);
890871be 4214 read_unlock(&em_tree->lock);
f2d8d74d 4215
3b951516 4216 if (!em) {
48940662 4217 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
d397712b
CM		 4218 (unsigned long long)logical,
4219 (unsigned long long)*length);
f2d8d74d 4220 BUG();
3b951516 4221 }
0b86a832
CM		 4222
4223 BUG_ON(em->start > logical || em->start + em->len < logical);
4224 map = (struct map_lookup *)em->bdev;
4225 offset = logical - em->start;
593060d7
CM		 4226
4227 stripe_nr = offset;
4228 /*
4229 * stripe_nr counts the total number of stripes we have to stride
4230 * to get to this block
4231 */
4232 do_div(stripe_nr, map->stripe_len);
4233
4234 stripe_offset = stripe_nr * map->stripe_len;
4235 BUG_ON(offset < stripe_offset);
4236
4237 /* stripe_offset is the offset of this block in its stripe*/
4238 stripe_offset = offset - stripe_offset;
4239
fce3bb9a
LD		 4240 if (rw & REQ_DISCARD)
4241 *length = min_t(u64, em->len - offset, *length);
52ba6929 4242 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
cea9e445
CM		 4243 /* we limit the length of each bio to what fits in a stripe */
4244 *length = min_t(u64, em->len - offset,
fce3bb9a 4245 map->stripe_len - stripe_offset);
cea9e445
CM		 4246 } else {
4247 *length = em->len - offset;
4248 }
f2d8d74d 4249
a1d3c478 4250 if (!bbio_ret)
cea9e445
CM		 4251 goto out;
4252
472262f3
SB		 4253 btrfs_dev_replace_lock(dev_replace);
4254 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4255 if (!dev_replace_is_ongoing)
4256 btrfs_dev_replace_unlock(dev_replace);
4257
ad6d620e
SB		 4258 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4259 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4260 dev_replace->tgtdev != NULL) {
4261 /*
4262 * in dev-replace case, for repair case (that's the only
4263 * case where the mirror is selected explicitly when
4264 * calling btrfs_map_block), blocks left of the left cursor
4265 * can also be read from the target drive.
4266 * For REQ_GET_READ_MIRRORS, the target drive is added as
4267 * the last one to the array of stripes. For READ, it also
4268 * needs to be supported using the same mirror number.
4269 * If the requested block is not left of the left cursor,
4270 * EIO is returned. This can happen because btrfs_num_copies()
4271 * returns one more in the dev-replace case.
4272 */
4273 u64 tmp_length = *length;
4274 struct btrfs_bio *tmp_bbio = NULL;
4275 int tmp_num_stripes;
4276 u64 srcdev_devid = dev_replace->srcdev->devid;
4277 int index_srcdev = 0;
4278 int found = 0;
4279 u64 physical_of_found = 0;
4280
4281 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4282 logical, &tmp_length, &tmp_bbio, 0);
4283 if (ret) {
4284 WARN_ON(tmp_bbio != NULL);
4285 goto out;
4286 }
4287
4288 tmp_num_stripes = tmp_bbio->num_stripes;
4289 if (mirror_num > tmp_num_stripes) {
4290 /*
4291 * REQ_GET_READ_MIRRORS does not contain this
4292 * mirror, that means that the requested area
4293 * is not left of the left cursor
4294 */
4295 ret = -EIO;
4296 kfree(tmp_bbio);
4297 goto out;
4298 }
4299
4300 /*
4301 * process the rest of the function using the mirror_num
4302 * of the source drive. Therefore look it up first.
4303 * At the end, patch the device pointer to the one of the
4304 * target drive.
4305 */
4306 for (i = 0; i < tmp_num_stripes; i++) {
4307 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4308 /*
4309 * In case of DUP, in order to keep it
4310 * simple, only add the mirror with the
4311 * lowest physical address
4312 */
4313 if (found &&
4314 physical_of_found <=
4315 tmp_bbio->stripes[i].physical)
4316 continue;
4317 index_srcdev = i;
4318 found = 1;
4319 physical_of_found =
4320 tmp_bbio->stripes[i].physical;
4321 }
4322 }
4323
4324 if (found) {
4325 mirror_num = index_srcdev + 1;
4326 patch_the_first_stripe_for_dev_replace = 1;
4327 physical_to_patch_in_first_stripe = physical_of_found;
4328 } else {
4329 WARN_ON(1);
4330 ret = -EIO;
4331 kfree(tmp_bbio);
4332 goto out;
4333 }
4334
4335 kfree(tmp_bbio);
4336 } else if (mirror_num > map->num_stripes) {
4337 mirror_num = 0;
4338 }
4339
f2d8d74d 4340 num_stripes = 1;
cea9e445 4341 stripe_index = 0;
fce3bb9a
LD		 4342 stripe_nr_orig = stripe_nr;
4343 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4344 (~(map->stripe_len - 1));
4345 do_div(stripe_nr_end, map->stripe_len);
4346 stripe_end_offset = stripe_nr_end * map->stripe_len -
4347 (offset + *length);
4348 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4349 if (rw & REQ_DISCARD)
4350 num_stripes = min_t(u64, map->num_stripes,
4351 stripe_nr_end - stripe_nr_orig);
4352 stripe_index = do_div(stripe_nr, map->num_stripes);
4353 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
29a8d9a0 4354 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
f2d8d74d 4355 num_stripes = map->num_stripes;
2fff734f 4356 else if (mirror_num)
f188591e 4357 stripe_index = mirror_num - 1;
dfe25020 4358 else {
30d9861f 4359 stripe_index = find_live_mirror(fs_info, map, 0,
dfe25020 4360 map->num_stripes,
30d9861f
SB		 4361 current->pid % map->num_stripes,
4362 dev_replace_is_ongoing);
a1d3c478 4363 mirror_num = stripe_index + 1;
dfe25020 4364 }
2fff734f 4365
611f0e00 4366 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
29a8d9a0 4367 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
f2d8d74d 4368 num_stripes = map->num_stripes;
a1d3c478 4369 } else if (mirror_num) {
f188591e 4370 stripe_index = mirror_num - 1;
a1d3c478
JS		 4371 } else {
4372 mirror_num = 1;
4373 }
2fff734f 4374
321aecc6
CM		 4375 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4376 int factor = map->num_stripes / map->sub_stripes;
321aecc6
CM		 4377
4378 stripe_index = do_div(stripe_nr, factor);
4379 stripe_index *= map->sub_stripes;
4380
29a8d9a0 4381 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
f2d8d74d 4382 num_stripes = map->sub_stripes;
fce3bb9a
LD		 4383 else if (rw & REQ_DISCARD)
4384 num_stripes = min_t(u64, map->sub_stripes *
4385 (stripe_nr_end - stripe_nr_orig),
4386 map->num_stripes);
321aecc6
CM		 4387 else if (mirror_num)
4388 stripe_index += mirror_num - 1;
dfe25020 4389 else {
3e74317a 4390 int old_stripe_index = stripe_index;
30d9861f
SB		 4391 stripe_index = find_live_mirror(fs_info, map,
4392 stripe_index,
dfe25020 4393 map->sub_stripes, stripe_index +
30d9861f
SB		 4394 current->pid % map->sub_stripes,
4395 dev_replace_is_ongoing);
3e74317a 4396 mirror_num = stripe_index - old_stripe_index + 1;
dfe25020 4397 }
8790d502
CM		 4398 } else {
4399 /*
4400 * after this do_div call, stripe_nr is the number of stripes
4401 * on this device we have to walk to find the data, and
4402 * stripe_index is the number of our device in the stripe array
4403 */
4404 stripe_index = do_div(stripe_nr, map->num_stripes);
a1d3c478 4405 mirror_num = stripe_index + 1;
8790d502 4406 }
593060d7 4407 BUG_ON(stripe_index >= map->num_stripes);
cea9e445 4408
472262f3 4409 num_alloc_stripes = num_stripes;
ad6d620e
SB		 4410 if (dev_replace_is_ongoing) {
4411 if (rw & (REQ_WRITE | REQ_DISCARD))
4412 num_alloc_stripes <<= 1;
4413 if (rw & REQ_GET_READ_MIRRORS)
4414 num_alloc_stripes++;
4415 }
472262f3 4416 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
de11cc12
LZ		 4417 if (!bbio) {
4418 ret = -ENOMEM;
4419 goto out;
4420 }
4421 atomic_set(&bbio->error, 0);
4422
fce3bb9a 4423 if (rw & REQ_DISCARD) {
ec9ef7a1
LZ		 4424 int factor = 0;
4425 int sub_stripes = 0;
4426 u64 stripes_per_dev = 0;
4427 u32 remaining_stripes = 0;
b89203f7 4428 u32 last_stripe = 0;
ec9ef7a1
LZ		 4429
4430 if (map->type &
4431 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4432 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4433 sub_stripes = 1;
4434 else
4435 sub_stripes = map->sub_stripes;
4436
4437 factor = map->num_stripes / sub_stripes;
4438 stripes_per_dev = div_u64_rem(stripe_nr_end -
4439 stripe_nr_orig,
4440 factor,
4441 &remaining_stripes);
b89203f7
LB		 4442 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4443 last_stripe *= sub_stripes;
ec9ef7a1
LZ		 4444 }
4445
fce3bb9a 4446 for (i = 0; i < num_stripes; i++) {
a1d3c478 4447 bbio->stripes[i].physical =
f2d8d74d
CM		 4448 map->stripes[stripe_index].physical +
4449 stripe_offset + stripe_nr * map->stripe_len;
a1d3c478 4450 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
fce3bb9a 4451
ec9ef7a1
LZ		 4452 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4453 BTRFS_BLOCK_GROUP_RAID10)) {
4454 bbio->stripes[i].length = stripes_per_dev *
4455 map->stripe_len;
b89203f7 4456
ec9ef7a1
LZ		 4457 if (i / sub_stripes < remaining_stripes)
4458 bbio->stripes[i].length +=
4459 map->stripe_len;
b89203f7
LB		 4460
4461 /*
4462 * Special for the first stripe and
4463 * the last stripe:
4464 *
4465 * |-------|...|-------|
4466 * |----------|
4467 * off end_off
4468 */
ec9ef7a1 4469 if (i < sub_stripes)
a1d3c478 4470 bbio->stripes[i].length -=
fce3bb9a 4471 stripe_offset;
b89203f7
LB		 4472
4473 if (stripe_index >= last_stripe &&
4474 stripe_index <= (last_stripe +
4475 sub_stripes - 1))
a1d3c478 4476 bbio->stripes[i].length -=
fce3bb9a 4477 stripe_end_offset;
b89203f7 4478
ec9ef7a1
LZ		 4479 if (i == sub_stripes - 1)
4480 stripe_offset = 0;
fce3bb9a 4481 } else
a1d3c478 4482 bbio->stripes[i].length = *length;
fce3bb9a
LD		 4483
4484 stripe_index++;
4485 if (stripe_index == map->num_stripes) {
4486 /* This could only happen for RAID0/10 */
4487 stripe_index = 0;
4488 stripe_nr++;
4489 }
4490 }
4491 } else {
4492 for (i = 0; i < num_stripes; i++) {
a1d3c478 4493 bbio->stripes[i].physical =
212a17ab
LT		 4494 map->stripes[stripe_index].physical +
4495 stripe_offset +
4496 stripe_nr * map->stripe_len;
a1d3c478 4497 bbio->stripes[i].dev =
212a17ab 4498 map->stripes[stripe_index].dev;
fce3bb9a 4499 stripe_index++;
f2d8d74d 4500 }
593060d7 4501 }
de11cc12 4502
29a8d9a0 4503 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
de11cc12
LZ		 4504 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4505 BTRFS_BLOCK_GROUP_RAID10 |
4506 BTRFS_BLOCK_GROUP_DUP)) {
4507 max_errors = 1;
4508 }
f2d8d74d 4509 }
de11cc12 4510
472262f3
SB		 4511 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4512 dev_replace->tgtdev != NULL) {
4513 int index_where_to_add;
4514 u64 srcdev_devid = dev_replace->srcdev->devid;
4515
4516 /*
4517 * duplicate the write operations while the dev replace
4518 * procedure is running. Since the copying of the old disk
4519 * to the new disk takes place at run time while the
4520 * filesystem is mounted writable, the regular write
4521 * operations to the old disk have to be duplicated to go
4522 * to the new disk as well.
4523 * Note that device->missing is handled by the caller, and
4524 * that the write to the old disk is already set up in the
4525 * stripes array.
4526 */
4527 index_where_to_add = num_stripes;
4528 for (i = 0; i < num_stripes; i++) {
4529 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4530 /* write to new disk, too */
4531 struct btrfs_bio_stripe *new =
4532 bbio->stripes + index_where_to_add;
4533 struct btrfs_bio_stripe *old =
4534 bbio->stripes + i;
4535
4536 new->physical = old->physical;
4537 new->length = old->length;
4538 new->dev = dev_replace->tgtdev;
4539 index_where_to_add++;
4540 max_errors++;
4541 }
4542 }
4543 num_stripes = index_where_to_add;
ad6d620e
SB		 4544 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4545 dev_replace->tgtdev != NULL) {
4546 u64 srcdev_devid = dev_replace->srcdev->devid;
4547 int index_srcdev = 0;
4548 int found = 0;
4549 u64 physical_of_found = 0;
4550
4551 /*
4552 * During the dev-replace procedure, the target drive can
4553 * also be used to read data in case it is needed to repair
4554 * a corrupt block elsewhere. This is possible if the
4555 * requested area is left of the left cursor. In this area,
4556 * the target drive is a full copy of the source drive.
4557 */
4558 for (i = 0; i < num_stripes; i++) {
4559 if (bbio->stripes[i].dev->devid == srcdev_devid) {
4560 /*
4561 * In case of DUP, in order to keep it
4562 * simple, only add the mirror with the
4563 * lowest physical address
4564 */
4565 if (found &&
4566 physical_of_found <=
4567 bbio->stripes[i].physical)
4568 continue;
4569 index_srcdev = i;
4570 found = 1;
4571 physical_of_found = bbio->stripes[i].physical;
4572 }
4573 }
4574 if (found) {
4575 u64 length = map->stripe_len;
4576
4577 if (physical_of_found + length <=
4578 dev_replace->cursor_left) {
4579 struct btrfs_bio_stripe *tgtdev_stripe =
4580 bbio->stripes + num_stripes;
4581
4582 tgtdev_stripe->physical = physical_of_found;
4583 tgtdev_stripe->length =
4584 bbio->stripes[index_srcdev].length;
4585 tgtdev_stripe->dev = dev_replace->tgtdev;
4586
4587 num_stripes++;
4588 }
4589 }
472262f3
SB		 4590 }
4591
de11cc12
LZ		 4592 *bbio_ret = bbio;
4593 bbio->num_stripes = num_stripes;
4594 bbio->max_errors = max_errors;
4595 bbio->mirror_num = mirror_num;
ad6d620e
SB		 4596
4597 /*
4598 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4599 * mirror_num == num_stripes + 1 && dev_replace target drive is
4600 * available as a mirror
4601 */
4602 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4603 WARN_ON(num_stripes > 1);
4604 bbio->stripes[0].dev = dev_replace->tgtdev;
4605 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4606 bbio->mirror_num = map->num_stripes + 1;
4607 }
cea9e445 4608out:
472262f3
SB		 4609 if (dev_replace_is_ongoing)
4610 btrfs_dev_replace_unlock(dev_replace);
0b86a832 4611 free_extent_map(em);
de11cc12 4612 return ret;
0b86a832
CM		 4613}
4614
3ec706c8 4615int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
f2d8d74d 4616 u64 logical, u64 *length,
a1d3c478 4617 struct btrfs_bio **bbio_ret, int mirror_num)
f2d8d74d 4618{
3ec706c8 4619 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
7eaceacc 4620 mirror_num);
f2d8d74d
CM		 4621}
4622
a512bbf8
YZ		 4623int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4624 u64 chunk_start, u64 physical, u64 devid,
4625 u64 **logical, int *naddrs, int *stripe_len)
4626{
4627 struct extent_map_tree *em_tree = &map_tree->map_tree;
4628 struct extent_map *em;
4629 struct map_lookup *map;
4630 u64 *buf;
4631 u64 bytenr;
4632 u64 length;
4633 u64 stripe_nr;
4634 int i, j, nr = 0;
4635
890871be 4636 read_lock(&em_tree->lock);
a512bbf8 4637 em = lookup_extent_mapping(em_tree, chunk_start, 1);
890871be 4638 read_unlock(&em_tree->lock);
a512bbf8
YZ		 4639
4640 BUG_ON(!em || em->start != chunk_start);
4641 map = (struct map_lookup *)em->bdev;
4642
4643 length = em->len;
4644 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4645 do_div(length, map->num_stripes / map->sub_stripes);
4646 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4647 do_div(length, map->num_stripes);
4648
4649 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
79787eaa 4650 BUG_ON(!buf); /* -ENOMEM */
a512bbf8
YZ		 4651
4652 for (i = 0; i < map->num_stripes; i++) {
4653 if (devid && map->stripes[i].dev->devid != devid)
4654 continue;
4655 if (map->stripes[i].physical > physical ||
4656 map->stripes[i].physical + length <= physical)
4657 continue;
4658
4659 stripe_nr = physical - map->stripes[i].physical;
4660 do_div(stripe_nr, map->stripe_len);
4661
4662 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4663 stripe_nr = stripe_nr * map->num_stripes + i;
4664 do_div(stripe_nr, map->sub_stripes);
4665 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4666 stripe_nr = stripe_nr * map->num_stripes + i;
4667 }
4668 bytenr = chunk_start + stripe_nr * map->stripe_len;
934d375b 4669 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 4670 for (j = 0; j < nr; j++) {
4671 if (buf[j] == bytenr)
4672 break;
4673 }
934d375b
CM		 4674 if (j == nr) {
4675 WARN_ON(nr >= map->num_stripes);
a512bbf8 4676 buf[nr++] = bytenr;
934d375b 4677 }
a512bbf8
YZ		 4678 }
4679
a512bbf8
YZ		 4680 *logical = buf;
4681 *naddrs = nr;
4682 *stripe_len = map->stripe_len;
4683
4684 free_extent_map(em);
4685 return 0;
f2d8d74d
CM		 4686}
4687
442a4f63
SB		 4688static void *merge_stripe_index_into_bio_private(void *bi_private,
4689 unsigned int stripe_index)
4690{
4691 /*
4692 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4693 * at most 1.
4694 * The alternative solution (instead of stealing bits from the
4695 * pointer) would be to allocate an intermediate structure
4696 * that contains the old private pointer plus the stripe_index.
4697 */
4698 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4699 BUG_ON(stripe_index > 3);
4700 return (void *)(((uintptr_t)bi_private) | stripe_index);
4701}
4702
4703static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4704{
4705 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4706}
4707
4708static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4709{
4710 return (unsigned int)((uintptr_t)bi_private) & 3;
4711}
4712
a1d3c478 4713static void btrfs_end_bio(struct bio *bio, int err)
8790d502 4714{
442a4f63 4715 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
7d2b4daa 4716 int is_orig_bio = 0;
8790d502 4717
442a4f63 4718 if (err) {
a1d3c478 4719 atomic_inc(&bbio->error);
442a4f63
SB		 4720 if (err == -EIO || err == -EREMOTEIO) {
4721 unsigned int stripe_index =
4722 extract_stripe_index_from_bio_private(
4723 bio->bi_private);
4724 struct btrfs_device *dev;
4725
4726 BUG_ON(stripe_index >= bbio->num_stripes);
4727 dev = bbio->stripes[stripe_index].dev;
597a60fa
SB		 4728 if (dev->bdev) {
4729 if (bio->bi_rw & WRITE)
4730 btrfs_dev_stat_inc(dev,
4731 BTRFS_DEV_STAT_WRITE_ERRS);
4732 else
4733 btrfs_dev_stat_inc(dev,
4734 BTRFS_DEV_STAT_READ_ERRS);
4735 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4736 btrfs_dev_stat_inc(dev,
4737 BTRFS_DEV_STAT_FLUSH_ERRS);
4738 btrfs_dev_stat_print_on_error(dev);
4739 }
442a4f63
SB		 4740 }
4741 }
8790d502 4742
a1d3c478 4743 if (bio == bbio->orig_bio)
7d2b4daa
CM		 4744 is_orig_bio = 1;
4745
a1d3c478 4746 if (atomic_dec_and_test(&bbio->stripes_pending)) {
7d2b4daa
CM		 4747 if (!is_orig_bio) {
4748 bio_put(bio);
a1d3c478 4749 bio = bbio->orig_bio;
7d2b4daa 4750 }
a1d3c478
JS		 4751 bio->bi_private = bbio->private;
4752 bio->bi_end_io = bbio->end_io;
2774b2ca
JS		 4753 bio->bi_bdev = (struct block_device *)
4754 (unsigned long)bbio->mirror_num;
a236aed1
CM		 4755 /* only send an error to the higher layers if it is
4756 * beyond the tolerance of the multi-bio
4757 */
a1d3c478 4758 if (atomic_read(&bbio->error) > bbio->max_errors) {
a236aed1 4759 err = -EIO;
5dbc8fca 4760 } else {
1259ab75
CM		 4761 /*
4762 * this bio is actually up to date, we didn't
4763 * go over the max number of errors
4764 */
4765 set_bit(BIO_UPTODATE, &bio->bi_flags);
a236aed1 4766 err = 0;
1259ab75 4767 }
a1d3c478 4768 kfree(bbio);
8790d502
CM		 4769
4770 bio_endio(bio, err);
7d2b4daa 4771 } else if (!is_orig_bio) {
8790d502
CM		 4772 bio_put(bio);
4773 }
8790d502
CM		 4774}
4775
8b712842
CM		 4776struct async_sched {
4777 struct bio *bio;
4778 int rw;
4779 struct btrfs_fs_info *info;
4780 struct btrfs_work work;
4781};
4782
4783/*
4784 * see run_scheduled_bios for a description of why bios are collected for
4785 * async submit.
4786 *
4787 * This will add one bio to the pending list for a device and make sure
4788 * the work struct is scheduled.
4789 */
143bede5 4790static noinline void schedule_bio(struct btrfs_root *root,
a1b32a59
CM		 4791 struct btrfs_device *device,
4792 int rw, struct bio *bio)
8b712842
CM		 4793{
4794 int should_queue = 1;
ffbd517d 4795 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 4796
4797 /* don't bother with additional async steps for reads, right now */
7b6d91da 4798 if (!(rw & REQ_WRITE)) {
492bb6de 4799 bio_get(bio);
21adbd5c 4800 btrfsic_submit_bio(rw, bio);
492bb6de 4801 bio_put(bio);
143bede5 4802 return;
8b712842
CM		 4803 }
4804
4805 /*
0986fe9e 4806 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 4807 * higher layers. Otherwise, the async bio makes it appear we have
4808 * made progress against dirty pages when we've really just put it
4809 * on a queue for later
4810 */
0986fe9e 4811 atomic_inc(&root->fs_info->nr_async_bios);
492bb6de 4812 WARN_ON(bio->bi_next);
8b712842
CM		 4813 bio->bi_next = NULL;
4814 bio->bi_rw |= rw;
4815
4816 spin_lock(&device->io_lock);
7b6d91da 4817 if (bio->bi_rw & REQ_SYNC)
ffbd517d
CM		 4818 pending_bios = &device->pending_sync_bios;
4819 else
4820 pending_bios = &device->pending_bios;
8b712842 4821
ffbd517d
CM		 4822 if (pending_bios->tail)
4823 pending_bios->tail->bi_next = bio;
8b712842 4824
ffbd517d
CM		 4825 pending_bios->tail = bio;
4826 if (!pending_bios->head)
4827 pending_bios->head = bio;
8b712842
CM		 4828 if (device->running_pending)
4829 should_queue = 0;
4830
4831 spin_unlock(&device->io_lock);
4832
4833 if (should_queue)
1cc127b5
CM		 4834 btrfs_queue_worker(&root->fs_info->submit_workers,
4835 &device->work);
8b712842
CM		 4836}
4837
de1ee92a
JB		 4838static int bio_size_ok(struct block_device *bdev, struct bio *bio,
4839 sector_t sector)
4840{
4841 struct bio_vec *prev;
4842 struct request_queue *q = bdev_get_queue(bdev);
4843 unsigned short max_sectors = queue_max_sectors(q);
4844 struct bvec_merge_data bvm = {
4845 .bi_bdev = bdev,
4846 .bi_sector = sector,
4847 .bi_rw = bio->bi_rw,
4848 };
4849
4850 if (bio->bi_vcnt == 0) {
4851 WARN_ON(1);
4852 return 1;
4853 }
4854
4855 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
4856 if ((bio->bi_size >> 9) > max_sectors)
4857 return 0;
4858
4859 if (!q->merge_bvec_fn)
4860 return 1;
4861
4862 bvm.bi_size = bio->bi_size - prev->bv_len;
4863 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
4864 return 0;
4865 return 1;
4866}
4867
4868static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4869 struct bio *bio, u64 physical, int dev_nr,
4870 int rw, int async)
4871{
4872 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
4873
4874 bio->bi_private = bbio;
4875 bio->bi_private = merge_stripe_index_into_bio_private(
4876 bio->bi_private, (unsigned int)dev_nr);
4877 bio->bi_end_io = btrfs_end_bio;
4878 bio->bi_sector = physical >> 9;
4879#ifdef DEBUG
4880 {
4881 struct rcu_string *name;
4882
4883 rcu_read_lock();
4884 name = rcu_dereference(dev->name);
d1423248 4885 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
de1ee92a
JB		 4886 "(%s id %llu), size=%u\n", rw,
4887 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4888 name->str, dev->devid, bio->bi_size);
4889 rcu_read_unlock();
4890 }
4891#endif
4892 bio->bi_bdev = dev->bdev;
4893 if (async)
4894 schedule_bio(root, dev, rw, bio);
4895 else
4896 btrfsic_submit_bio(rw, bio);
4897}
4898
4899static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
4900 struct bio *first_bio, struct btrfs_device *dev,
4901 int dev_nr, int rw, int async)
4902{
4903 struct bio_vec *bvec = first_bio->bi_io_vec;
4904 struct bio *bio;
4905 int nr_vecs = bio_get_nr_vecs(dev->bdev);
4906 u64 physical = bbio->stripes[dev_nr].physical;
4907
4908again:
4909 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
4910 if (!bio)
4911 return -ENOMEM;
4912
4913 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
4914 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
4915 bvec->bv_offset) < bvec->bv_len) {
4916 u64 len = bio->bi_size;
4917
4918 atomic_inc(&bbio->stripes_pending);
4919 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
4920 rw, async);
4921 physical += len;
4922 goto again;
4923 }
4924 bvec++;
4925 }
4926
4927 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
4928 return 0;
4929}
4930
4931static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
4932{
4933 atomic_inc(&bbio->error);
4934 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4935 bio->bi_private = bbio->private;
4936 bio->bi_end_io = bbio->end_io;
4937 bio->bi_bdev = (struct block_device *)
4938 (unsigned long)bbio->mirror_num;
4939 bio->bi_sector = logical >> 9;
4940 kfree(bbio);
4941 bio_endio(bio, -EIO);
4942 }
4943}
4944
f188591e 4945int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
8b712842 4946 int mirror_num, int async_submit)
0b86a832 4947{
0b86a832 4948 struct btrfs_device *dev;
8790d502 4949 struct bio *first_bio = bio;
a62b9401 4950 u64 logical = (u64)bio->bi_sector << 9;
0b86a832
CM		 4951 u64 length = 0;
4952 u64 map_length;
0b86a832 4953 int ret;
8790d502
CM		 4954 int dev_nr = 0;
4955 int total_devs = 1;
a1d3c478 4956 struct btrfs_bio *bbio = NULL;
0b86a832 4957
f2d8d74d 4958 length = bio->bi_size;
0b86a832 4959 map_length = length;
cea9e445 4960
3ec706c8 4961 ret = btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
f188591e 4962 mirror_num);
61891923 4963 if (ret)
79787eaa 4964 return ret;
cea9e445 4965
a1d3c478 4966 total_devs = bbio->num_stripes;
cea9e445 4967 if (map_length < length) {
48940662 4968 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
d397712b
CM		 4969 "len %llu\n", (unsigned long long)logical,
4970 (unsigned long long)length,
4971 (unsigned long long)map_length);
cea9e445
CM		 4972 BUG();
4973 }
a1d3c478
JS		 4974
4975 bbio->orig_bio = first_bio;
4976 bbio->private = first_bio->bi_private;
4977 bbio->end_io = first_bio->bi_end_io;
4978 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
cea9e445 4979
d397712b 4980 while (dev_nr < total_devs) {
de1ee92a
JB		 4981 dev = bbio->stripes[dev_nr].dev;
4982 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
4983 bbio_error(bbio, first_bio, logical);
4984 dev_nr++;
4985 continue;
4986 }
4987
4988 /*
4989 * Check and see if we're ok with this bio based on it's size
4990 * and offset with the given device.
4991 */
4992 if (!bio_size_ok(dev->bdev, first_bio,
4993 bbio->stripes[dev_nr].physical >> 9)) {
4994 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
4995 dev_nr, rw, async_submit);
4996 BUG_ON(ret);
4997 dev_nr++;
4998 continue;
4999 }
5000
a1d3c478
JS		 5001 if (dev_nr < total_devs - 1) {
5002 bio = bio_clone(first_bio, GFP_NOFS);
79787eaa 5003 BUG_ON(!bio); /* -ENOMEM */
a1d3c478
JS		 5004 } else {
5005 bio = first_bio;
8790d502 5006 }
de1ee92a
JB		 5007
5008 submit_stripe_bio(root, bbio, bio,
5009 bbio->stripes[dev_nr].physical, dev_nr, rw,
5010 async_submit);
8790d502
CM		 5011 dev_nr++;
5012 }
0b86a832
CM		 5013 return 0;
5014}
5015
aa1b8cd4 5016struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
2b82032c 5017 u8 *uuid, u8 *fsid)
0b86a832 5018{
2b82032c
YZ		 5019 struct btrfs_device *device;
5020 struct btrfs_fs_devices *cur_devices;
5021
aa1b8cd4 5022 cur_devices = fs_info->fs_devices;
2b82032c
YZ		 5023 while (cur_devices) {
5024 if (!fsid ||
5025 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5026 device = __find_device(&cur_devices->devices,
5027 devid, uuid);
5028 if (device)
5029 return device;
5030 }
5031 cur_devices = cur_devices->seed;
5032 }
5033 return NULL;
0b86a832
CM		 5034}
5035
dfe25020
CM		 5036static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5037 u64 devid, u8 *dev_uuid)
5038{
5039 struct btrfs_device *device;
5040 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5041
5042 device = kzalloc(sizeof(*device), GFP_NOFS);
7cbd8a83 5043 if (!device)
5044 return NULL;
dfe25020
CM		 5045 list_add(&device->dev_list,
5046 &fs_devices->devices);
dfe25020
CM		 5047 device->dev_root = root->fs_info->dev_root;
5048 device->devid = devid;
8b712842 5049 device->work.func = pending_bios_fn;
e4404d6e 5050 device->fs_devices = fs_devices;
cd02dca5 5051 device->missing = 1;
dfe25020 5052 fs_devices->num_devices++;
cd02dca5 5053 fs_devices->missing_devices++;
dfe25020 5054 spin_lock_init(&device->io_lock);
d20f7043 5055 INIT_LIST_HEAD(&device->dev_alloc_list);
dfe25020
CM		 5056 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5057 return device;
5058}
5059
0b86a832
CM		 5060static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5061 struct extent_buffer *leaf,
5062 struct btrfs_chunk *chunk)
5063{
5064 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5065 struct map_lookup *map;
5066 struct extent_map *em;
5067 u64 logical;
5068 u64 length;
5069 u64 devid;
a443755f 5070 u8 uuid[BTRFS_UUID_SIZE];
593060d7 5071 int num_stripes;
0b86a832 5072 int ret;
593060d7 5073 int i;
0b86a832 5074
e17cade2
CM		 5075 logical = key->offset;
5076 length = btrfs_chunk_length(leaf, chunk);
a061fc8d 5077
890871be 5078 read_lock(&map_tree->map_tree.lock);
0b86a832 5079 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 5080 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 5081
5082 /* already mapped? */
5083 if (em && em->start <= logical && em->start + em->len > logical) {
5084 free_extent_map(em);
0b86a832
CM		 5085 return 0;
5086 } else if (em) {
5087 free_extent_map(em);
5088 }
0b86a832 5089
172ddd60 5090 em = alloc_extent_map();
0b86a832
CM		 5091 if (!em)
5092 return -ENOMEM;
593060d7
CM		 5093 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5094 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 5095 if (!map) {
5096 free_extent_map(em);
5097 return -ENOMEM;
5098 }
5099
5100 em->bdev = (struct block_device *)map;
5101 em->start = logical;
5102 em->len = length;
70c8a91c 5103 em->orig_start = 0;
0b86a832 5104 em->block_start = 0;
c8b97818 5105 em->block_len = em->len;
0b86a832 5106
593060d7
CM		 5107 map->num_stripes = num_stripes;
5108 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5109 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5110 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5111 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5112 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 5113 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 5114 for (i = 0; i < num_stripes; i++) {
5115 map->stripes[i].physical =
5116 btrfs_stripe_offset_nr(leaf, chunk, i);
5117 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 5118 read_extent_buffer(leaf, uuid, (unsigned long)
5119 btrfs_stripe_dev_uuid_nr(chunk, i),
5120 BTRFS_UUID_SIZE);
aa1b8cd4
SB		 5121 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5122 uuid, NULL);
dfe25020 5123 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
593060d7
CM		 5124 kfree(map);
5125 free_extent_map(em);
5126 return -EIO;
5127 }
dfe25020
CM		 5128 if (!map->stripes[i].dev) {
5129 map->stripes[i].dev =
5130 add_missing_dev(root, devid, uuid);
5131 if (!map->stripes[i].dev) {
5132 kfree(map);
5133 free_extent_map(em);
5134 return -EIO;
5135 }
5136 }
5137 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 5138 }
5139
890871be 5140 write_lock(&map_tree->map_tree.lock);
0b86a832 5141 ret = add_extent_mapping(&map_tree->map_tree, em);
890871be 5142 write_unlock(&map_tree->map_tree.lock);
79787eaa 5143 BUG_ON(ret); /* Tree corruption */
0b86a832
CM		 5144 free_extent_map(em);
5145
5146 return 0;
5147}
5148
143bede5 5149static void fill_device_from_item(struct extent_buffer *leaf,
0b86a832
CM		 5150 struct btrfs_dev_item *dev_item,
5151 struct btrfs_device *device)
5152{
5153 unsigned long ptr;
0b86a832
CM		 5154
5155 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 5156 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5157 device->total_bytes = device->disk_total_bytes;
0b86a832
CM		 5158 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5159 device->type = btrfs_device_type(leaf, dev_item);
5160 device->io_align = btrfs_device_io_align(leaf, dev_item);
5161 device->io_width = btrfs_device_io_width(leaf, dev_item);
5162 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
8dabb742 5163 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
63a212ab 5164 device->is_tgtdev_for_dev_replace = 0;
0b86a832
CM		 5165
5166 ptr = (unsigned long)btrfs_device_uuid(dev_item);
e17cade2 5167 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832
CM		 5168}
5169
2b82032c
YZ		 5170static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5171{
5172 struct btrfs_fs_devices *fs_devices;
5173 int ret;
5174
b367e47f 5175 BUG_ON(!mutex_is_locked(&uuid_mutex));
2b82032c
YZ		 5176
5177 fs_devices = root->fs_info->fs_devices->seed;
5178 while (fs_devices) {
5179 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5180 ret = 0;
5181 goto out;
5182 }
5183 fs_devices = fs_devices->seed;
5184 }
5185
5186 fs_devices = find_fsid(fsid);
5187 if (!fs_devices) {
5188 ret = -ENOENT;
5189 goto out;
5190 }
e4404d6e
YZ		 5191
5192 fs_devices = clone_fs_devices(fs_devices);
5193 if (IS_ERR(fs_devices)) {
5194 ret = PTR_ERR(fs_devices);
2b82032c
YZ		 5195 goto out;
5196 }
5197
97288f2c 5198 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
15916de8 5199 root->fs_info->bdev_holder);
48d28232
JL		 5200 if (ret) {
5201 free_fs_devices(fs_devices);
2b82032c 5202 goto out;
48d28232 5203 }
2b82032c
YZ		 5204
5205 if (!fs_devices->seeding) {
5206 __btrfs_close_devices(fs_devices);
e4404d6e 5207 free_fs_devices(fs_devices);
2b82032c
YZ		 5208 ret = -EINVAL;
5209 goto out;
5210 }
5211
5212 fs_devices->seed = root->fs_info->fs_devices->seed;
5213 root->fs_info->fs_devices->seed = fs_devices;
2b82032c 5214out:
2b82032c
YZ		 5215 return ret;
5216}
5217
0d81ba5d 5218static int read_one_dev(struct btrfs_root *root,
0b86a832
CM		 5219 struct extent_buffer *leaf,
5220 struct btrfs_dev_item *dev_item)
5221{
5222 struct btrfs_device *device;
5223 u64 devid;
5224 int ret;
2b82032c 5225 u8 fs_uuid[BTRFS_UUID_SIZE];
a443755f
CM		 5226 u8 dev_uuid[BTRFS_UUID_SIZE];
5227
0b86a832 5228 devid = btrfs_device_id(leaf, dev_item);
a443755f
CM		 5229 read_extent_buffer(leaf, dev_uuid,
5230 (unsigned long)btrfs_device_uuid(dev_item),
5231 BTRFS_UUID_SIZE);
2b82032c
YZ		 5232 read_extent_buffer(leaf, fs_uuid,
5233 (unsigned long)btrfs_device_fsid(dev_item),
5234 BTRFS_UUID_SIZE);
5235
5236 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5237 ret = open_seed_devices(root, fs_uuid);
e4404d6e 5238 if (ret && !btrfs_test_opt(root, DEGRADED))
2b82032c 5239 return ret;
2b82032c
YZ		 5240 }
5241
aa1b8cd4 5242 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
2b82032c 5243 if (!device || !device->bdev) {
e4404d6e 5244 if (!btrfs_test_opt(root, DEGRADED))
2b82032c
YZ		 5245 return -EIO;
5246
5247 if (!device) {
d397712b
CM		 5248 printk(KERN_WARNING "warning devid %llu missing\n",
5249 (unsigned long long)devid);
2b82032c
YZ		 5250 device = add_missing_dev(root, devid, dev_uuid);
5251 if (!device)
5252 return -ENOMEM;
cd02dca5
CM		 5253 } else if (!device->missing) {
5254 /*
5255 * this happens when a device that was properly setup
5256 * in the device info lists suddenly goes bad.
5257 * device->bdev is NULL, and so we have to set
5258 * device->missing to one here
5259 */
5260 root->fs_info->fs_devices->missing_devices++;
5261 device->missing = 1;
2b82032c
YZ		 5262 }
5263 }
5264
5265 if (device->fs_devices != root->fs_info->fs_devices) {
5266 BUG_ON(device->writeable);
5267 if (device->generation !=
5268 btrfs_device_generation(leaf, dev_item))
5269 return -EINVAL;
6324fbf3 5270 }
0b86a832
CM		 5271
5272 fill_device_from_item(leaf, dev_item, device);
5273 device->dev_root = root->fs_info->dev_root;
dfe25020 5274 device->in_fs_metadata = 1;
63a212ab 5275 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
2b82032c 5276 device->fs_devices->total_rw_bytes += device->total_bytes;
2bf64758
JB		 5277 spin_lock(&root->fs_info->free_chunk_lock);
5278 root->fs_info->free_chunk_space += device->total_bytes -
5279 device->bytes_used;
5280 spin_unlock(&root->fs_info->free_chunk_lock);
5281 }
0b86a832 5282 ret = 0;
0b86a832
CM		 5283 return ret;
5284}
5285
e4404d6e 5286int btrfs_read_sys_array(struct btrfs_root *root)
0b86a832 5287{
6c41761f 5288 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
a061fc8d 5289 struct extent_buffer *sb;
0b86a832 5290 struct btrfs_disk_key *disk_key;
0b86a832 5291 struct btrfs_chunk *chunk;
84eed90f
CM		 5292 u8 *ptr;
5293 unsigned long sb_ptr;
5294 int ret = 0;
0b86a832
CM		 5295 u32 num_stripes;
5296 u32 array_size;
5297 u32 len = 0;
0b86a832 5298 u32 cur;
84eed90f 5299 struct btrfs_key key;
0b86a832 5300
e4404d6e 5301 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
a061fc8d
CM		 5302 BTRFS_SUPER_INFO_SIZE);
5303 if (!sb)
5304 return -ENOMEM;
5305 btrfs_set_buffer_uptodate(sb);
85d4e461 5306 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
8a334426
DS		 5307 /*
5308 * The sb extent buffer is artifical and just used to read the system array.
5309 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5310 * pages up-to-date when the page is larger: extent does not cover the
5311 * whole page and consequently check_page_uptodate does not find all
5312 * the page's extents up-to-date (the hole beyond sb),
5313 * write_extent_buffer then triggers a WARN_ON.
5314 *
5315 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5316 * but sb spans only this function. Add an explicit SetPageUptodate call
5317 * to silence the warning eg. on PowerPC 64.
5318 */
5319 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
727011e0 5320 SetPageUptodate(sb->pages[0]);
4008c04a 5321
a061fc8d 5322 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 5323 array_size = btrfs_super_sys_array_size(super_copy);
5324
0b86a832
CM		 5325 ptr = super_copy->sys_chunk_array;
5326 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5327 cur = 0;
5328
5329 while (cur < array_size) {
5330 disk_key = (struct btrfs_disk_key *)ptr;
5331 btrfs_disk_key_to_cpu(&key, disk_key);
5332
a061fc8d 5333 len = sizeof(*disk_key); ptr += len;
0b86a832
CM		 5334 sb_ptr += len;
5335 cur += len;
5336
0d81ba5d 5337 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
0b86a832 5338 chunk = (struct btrfs_chunk *)sb_ptr;
0d81ba5d 5339 ret = read_one_chunk(root, &key, sb, chunk);
84eed90f
CM		 5340 if (ret)
5341 break;
0b86a832
CM		 5342 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5343 len = btrfs_chunk_item_size(num_stripes);
5344 } else {
84eed90f
CM		 5345 ret = -EIO;
5346 break;
0b86a832
CM		 5347 }
5348 ptr += len;
5349 sb_ptr += len;
5350 cur += len;
5351 }
a061fc8d 5352 free_extent_buffer(sb);
84eed90f 5353 return ret;
0b86a832
CM		 5354}
5355
5356int btrfs_read_chunk_tree(struct btrfs_root *root)
5357{
5358 struct btrfs_path *path;
5359 struct extent_buffer *leaf;
5360 struct btrfs_key key;
5361 struct btrfs_key found_key;
5362 int ret;
5363 int slot;
5364
5365 root = root->fs_info->chunk_root;
5366
5367 path = btrfs_alloc_path();
5368 if (!path)
5369 return -ENOMEM;
5370
b367e47f
LZ		 5371 mutex_lock(&uuid_mutex);
5372 lock_chunks(root);
5373
0b86a832
CM		 5374 /* first we search for all of the device items, and then we
5375 * read in all of the chunk items. This way we can create chunk
5376 * mappings that reference all of the devices that are afound
5377 */
5378 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5379 key.offset = 0;
5380 key.type = 0;
5381again:
5382 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 5383 if (ret < 0)
5384 goto error;
d397712b 5385 while (1) {
0b86a832
CM		 5386 leaf = path->nodes[0];
5387 slot = path->slots[0];
5388 if (slot >= btrfs_header_nritems(leaf)) {
5389 ret = btrfs_next_leaf(root, path);
5390 if (ret == 0)
5391 continue;
5392 if (ret < 0)
5393 goto error;
5394 break;
5395 }
5396 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5397 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5398 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5399 break;
5400 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5401 struct btrfs_dev_item *dev_item;
5402 dev_item = btrfs_item_ptr(leaf, slot,
5403 struct btrfs_dev_item);
0d81ba5d 5404 ret = read_one_dev(root, leaf, dev_item);
2b82032c
YZ		 5405 if (ret)
5406 goto error;
0b86a832
CM		 5407 }
5408 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5409 struct btrfs_chunk *chunk;
5410 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5411 ret = read_one_chunk(root, &found_key, leaf, chunk);
2b82032c
YZ		 5412 if (ret)
5413 goto error;
0b86a832
CM		 5414 }
5415 path->slots[0]++;
5416 }
5417 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5418 key.objectid = 0;
b3b4aa74 5419 btrfs_release_path(path);
0b86a832
CM		 5420 goto again;
5421 }
0b86a832
CM		 5422 ret = 0;
5423error:
b367e47f
LZ		 5424 unlock_chunks(root);
5425 mutex_unlock(&uuid_mutex);
5426
2b82032c 5427 btrfs_free_path(path);
0b86a832
CM		 5428 return ret;
5429}
442a4f63 5430
733f4fbb
SB		 5431static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5432{
5433 int i;
5434
5435 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5436 btrfs_dev_stat_reset(dev, i);
5437}
5438
5439int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5440{
5441 struct btrfs_key key;
5442 struct btrfs_key found_key;
5443 struct btrfs_root *dev_root = fs_info->dev_root;
5444 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5445 struct extent_buffer *eb;
5446 int slot;
5447 int ret = 0;
5448 struct btrfs_device *device;
5449 struct btrfs_path *path = NULL;
5450 int i;
5451
5452 path = btrfs_alloc_path();
5453 if (!path) {
5454 ret = -ENOMEM;
5455 goto out;
5456 }
5457
5458 mutex_lock(&fs_devices->device_list_mutex);
5459 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5460 int item_size;
5461 struct btrfs_dev_stats_item *ptr;
5462
5463 key.objectid = 0;
5464 key.type = BTRFS_DEV_STATS_KEY;
5465 key.offset = device->devid;
5466 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5467 if (ret) {
733f4fbb
SB		 5468 __btrfs_reset_dev_stats(device);
5469 device->dev_stats_valid = 1;
5470 btrfs_release_path(path);
5471 continue;
5472 }
5473 slot = path->slots[0];
5474 eb = path->nodes[0];
5475 btrfs_item_key_to_cpu(eb, &found_key, slot);
5476 item_size = btrfs_item_size_nr(eb, slot);
5477
5478 ptr = btrfs_item_ptr(eb, slot,
5479 struct btrfs_dev_stats_item);
5480
5481 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5482 if (item_size >= (1 + i) * sizeof(__le64))
5483 btrfs_dev_stat_set(device, i,
5484 btrfs_dev_stats_value(eb, ptr, i));
5485 else
5486 btrfs_dev_stat_reset(device, i);
5487 }
5488
5489 device->dev_stats_valid = 1;
5490 btrfs_dev_stat_print_on_load(device);
5491 btrfs_release_path(path);
5492 }
5493 mutex_unlock(&fs_devices->device_list_mutex);
5494
5495out:
5496 btrfs_free_path(path);
5497 return ret < 0 ? ret : 0;
5498}
5499
5500static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5501 struct btrfs_root *dev_root,
5502 struct btrfs_device *device)
5503{
5504 struct btrfs_path *path;
5505 struct btrfs_key key;
5506 struct extent_buffer *eb;
5507 struct btrfs_dev_stats_item *ptr;
5508 int ret;
5509 int i;
5510
5511 key.objectid = 0;
5512 key.type = BTRFS_DEV_STATS_KEY;
5513 key.offset = device->devid;
5514
5515 path = btrfs_alloc_path();
5516 BUG_ON(!path);
5517 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5518 if (ret < 0) {
606686ee
JB		 5519 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5520 ret, rcu_str_deref(device->name));
733f4fbb
SB		 5521 goto out;
5522 }
5523
5524 if (ret == 0 &&
5525 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5526 /* need to delete old one and insert a new one */
5527 ret = btrfs_del_item(trans, dev_root, path);
5528 if (ret != 0) {
606686ee
JB		 5529 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5530 rcu_str_deref(device->name), ret);
733f4fbb
SB		 5531 goto out;
5532 }
5533 ret = 1;
5534 }
5535
5536 if (ret == 1) {
5537 /* need to insert a new item */
5538 btrfs_release_path(path);
5539 ret = btrfs_insert_empty_item(trans, dev_root, path,
5540 &key, sizeof(*ptr));
5541 if (ret < 0) {
606686ee
JB		 5542 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5543 rcu_str_deref(device->name), ret);
733f4fbb
SB		 5544 goto out;
5545 }
5546 }
5547
5548 eb = path->nodes[0];
5549 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5550 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5551 btrfs_set_dev_stats_value(eb, ptr, i,
5552 btrfs_dev_stat_read(device, i));
5553 btrfs_mark_buffer_dirty(eb);
5554
5555out:
5556 btrfs_free_path(path);
5557 return ret;
5558}
5559
5560/*
5561 * called from commit_transaction. Writes all changed device stats to disk.
5562 */
5563int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5564 struct btrfs_fs_info *fs_info)
5565{
5566 struct btrfs_root *dev_root = fs_info->dev_root;
5567 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5568 struct btrfs_device *device;
5569 int ret = 0;
5570
5571 mutex_lock(&fs_devices->device_list_mutex);
5572 list_for_each_entry(device, &fs_devices->devices, dev_list) {
5573 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5574 continue;
5575
5576 ret = update_dev_stat_item(trans, dev_root, device);
5577 if (!ret)
5578 device->dev_stats_dirty = 0;
5579 }
5580 mutex_unlock(&fs_devices->device_list_mutex);
5581
5582 return ret;
5583}
5584
442a4f63
SB		 5585void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5586{
5587 btrfs_dev_stat_inc(dev, index);
5588 btrfs_dev_stat_print_on_error(dev);
5589}
5590
5591void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5592{
733f4fbb
SB		 5593 if (!dev->dev_stats_valid)
5594 return;
606686ee 5595 printk_ratelimited_in_rcu(KERN_ERR
442a4f63 5596 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
606686ee 5597 rcu_str_deref(dev->name),
442a4f63
SB		 5598 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5599 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5600 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5601 btrfs_dev_stat_read(dev,
5602 BTRFS_DEV_STAT_CORRUPTION_ERRS),
5603 btrfs_dev_stat_read(dev,
5604 BTRFS_DEV_STAT_GENERATION_ERRS));
5605}
c11d2c23 5606
733f4fbb
SB		 5607static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5608{
a98cdb85
SB		 5609 int i;
5610
5611 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5612 if (btrfs_dev_stat_read(dev, i) != 0)
5613 break;
5614 if (i == BTRFS_DEV_STAT_VALUES_MAX)
5615 return; /* all values == 0, suppress message */
5616
606686ee
JB		 5617 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5618 rcu_str_deref(dev->name),
733f4fbb
SB		 5619 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5620 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5621 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5622 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5623 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5624}
5625
c11d2c23 5626int btrfs_get_dev_stats(struct btrfs_root *root,
b27f7c0c 5627 struct btrfs_ioctl_get_dev_stats *stats)
c11d2c23
SB		 5628{
5629 struct btrfs_device *dev;
5630 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5631 int i;
5632
5633 mutex_lock(&fs_devices->device_list_mutex);
aa1b8cd4 5634 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
c11d2c23
SB		 5635 mutex_unlock(&fs_devices->device_list_mutex);
5636
5637 if (!dev) {
5638 printk(KERN_WARNING
5639 "btrfs: get dev_stats failed, device not found\n");
5640 return -ENODEV;
733f4fbb
SB		 5641 } else if (!dev->dev_stats_valid) {
5642 printk(KERN_WARNING
5643 "btrfs: get dev_stats failed, not yet valid\n");
5644 return -ENODEV;
b27f7c0c 5645 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
c11d2c23
SB		 5646 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5647 if (stats->nr_items > i)
5648 stats->values[i] =
5649 btrfs_dev_stat_read_and_reset(dev, i);
5650 else
5651 btrfs_dev_stat_reset(dev, i);
5652 }
5653 } else {
5654 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5655 if (stats->nr_items > i)
5656 stats->values[i] = btrfs_dev_stat_read(dev, i);
5657 }
5658 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5659 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5660 return 0;
5661}
a8a6dab7
SB		 5662
5663int btrfs_scratch_superblock(struct btrfs_device *device)
5664{
5665 struct buffer_head *bh;
5666 struct btrfs_super_block *disk_super;
5667
5668 bh = btrfs_read_dev_super(device->bdev);
5669 if (!bh)
5670 return -EINVAL;
5671 disk_super = (struct btrfs_super_block *)bh->b_data;
5672
5673 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5674 set_buffer_dirty(bh);
5675 sync_dirty_buffer(bh);
5676 brelse(bh);
5677
5678 return 0;
5679}
Linux 6.x block layer and io_uring tree(s)