projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Add checker for EXTENT_CSUM
[linux-block.git] / fs / btrfs / volumes.c
CommitLineData
0b86a832
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
5a0e3ad6 20#include <linux/slab.h>
8a4b83cc 21#include <linux/buffer_head.h>
f2d8d74d 22#include <linux/blkdev.h>
b765ead5 23#include <linux/iocontext.h>
6f88a440 24#include <linux/capability.h>
442a4f63 25#include <linux/ratelimit.h>
59641015 26#include <linux/kthread.h>
53b381b3 27#include <linux/raid/pq.h>
803b2f54 28#include <linux/semaphore.h>
8da4b8c4 29#include <linux/uuid.h>
53b381b3 30#include <asm/div64.h>
0b86a832
CM		 31#include "ctree.h"
32#include "extent_map.h"
33#include "disk-io.h"
34#include "transaction.h"
35#include "print-tree.h"
36#include "volumes.h"
53b381b3 37#include "raid56.h"
8b712842 38#include "async-thread.h"
21adbd5c 39#include "check-integrity.h"
606686ee 40#include "rcu-string.h"
3fed40cc 41#include "math.h"
8dabb742 42#include "dev-replace.h"
99994cde 43#include "sysfs.h"
0b86a832 44
af902047
ZL		 45const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
47 .sub_stripes = 2,
48 .dev_stripes = 1,
49 .devs_max = 0, /* 0 == as many as possible */
50 .devs_min = 4,
8789f4fe 51 .tolerated_failures = 1,
af902047
ZL		 52 .devs_increment = 2,
53 .ncopies = 2,
54 },
55 [BTRFS_RAID_RAID1] = {
56 .sub_stripes = 1,
57 .dev_stripes = 1,
58 .devs_max = 2,
59 .devs_min = 2,
8789f4fe 60 .tolerated_failures = 1,
af902047
ZL		 61 .devs_increment = 2,
62 .ncopies = 2,
63 },
64 [BTRFS_RAID_DUP] = {
65 .sub_stripes = 1,
66 .dev_stripes = 2,
67 .devs_max = 1,
68 .devs_min = 1,
8789f4fe 69 .tolerated_failures = 0,
af902047
ZL		 70 .devs_increment = 1,
71 .ncopies = 2,
72 },
73 [BTRFS_RAID_RAID0] = {
74 .sub_stripes = 1,
75 .dev_stripes = 1,
76 .devs_max = 0,
77 .devs_min = 2,
8789f4fe 78 .tolerated_failures = 0,
af902047
ZL		 79 .devs_increment = 1,
80 .ncopies = 1,
81 },
82 [BTRFS_RAID_SINGLE] = {
83 .sub_stripes = 1,
84 .dev_stripes = 1,
85 .devs_max = 1,
86 .devs_min = 1,
8789f4fe 87 .tolerated_failures = 0,
af902047
ZL		 88 .devs_increment = 1,
89 .ncopies = 1,
90 },
91 [BTRFS_RAID_RAID5] = {
92 .sub_stripes = 1,
93 .dev_stripes = 1,
94 .devs_max = 0,
95 .devs_min = 2,
8789f4fe 96 .tolerated_failures = 1,
af902047
ZL		 97 .devs_increment = 1,
98 .ncopies = 2,
99 },
100 [BTRFS_RAID_RAID6] = {
101 .sub_stripes = 1,
102 .dev_stripes = 1,
103 .devs_max = 0,
104 .devs_min = 3,
8789f4fe 105 .tolerated_failures = 2,
af902047
ZL		 106 .devs_increment = 1,
107 .ncopies = 3,
108 },
109};
110
fb75d857 111const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
af902047
ZL		 112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
119};
120
621292ba
DS		 121/*
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
125 */
126const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134};
135
2b82032c 136static int init_first_rw_device(struct btrfs_trans_handle *trans,
e4a4dce7 137 struct btrfs_fs_info *fs_info);
2ff7e61e 138static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
733f4fbb 139static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
48a3b636 140static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
733f4fbb 141static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
5ab56090
LB		 142static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
143 enum btrfs_map_op op,
144 u64 logical, u64 *length,
145 struct btrfs_bio **bbio_ret,
146 int mirror_num, int need_raid_map);
2b82032c 147
67a2c45e 148DEFINE_MUTEX(uuid_mutex);
8a4b83cc 149static LIST_HEAD(fs_uuids);
c73eccf7
AJ		 150struct list_head *btrfs_get_fs_uuids(void)
151{
152 return &fs_uuids;
153}
8a4b83cc 154
2dfeca9b
DS		 155/*
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
158 *
159 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160 * The returned struct is not linked onto any lists and can be destroyed with
161 * kfree() right away.
162 */
163static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
2208a378
ID		 164{
165 struct btrfs_fs_devices *fs_devs;
166
78f2c9e6 167 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
2208a378
ID		 168 if (!fs_devs)
169 return ERR_PTR(-ENOMEM);
170
171 mutex_init(&fs_devs->device_list_mutex);
172
173 INIT_LIST_HEAD(&fs_devs->devices);
935e5cc9 174 INIT_LIST_HEAD(&fs_devs->resized_devices);
2208a378
ID		 175 INIT_LIST_HEAD(&fs_devs->alloc_list);
176 INIT_LIST_HEAD(&fs_devs->list);
2208a378
ID		 177 if (fsid)
178 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
2208a378
ID		 179
180 return fs_devs;
181}
182
e4404d6e
YZ		 183static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
184{
185 struct btrfs_device *device;
186 WARN_ON(fs_devices->opened);
187 while (!list_empty(&fs_devices->devices)) {
188 device = list_entry(fs_devices->devices.next,
189 struct btrfs_device, dev_list);
190 list_del(&device->dev_list);
606686ee 191 rcu_string_free(device->name);
e4404d6e
YZ		 192 kfree(device);
193 }
194 kfree(fs_devices);
195}
196
b8b8ff59
LC		 197static void btrfs_kobject_uevent(struct block_device *bdev,
198 enum kobject_action action)
199{
200 int ret;
201
202 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
203 if (ret)
efe120a0 204 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
b8b8ff59
LC		 205 action,
206 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
207 &disk_to_dev(bdev->bd_disk)->kobj);
208}
209
143bede5 210void btrfs_cleanup_fs_uuids(void)
8a4b83cc
CM		 211{
212 struct btrfs_fs_devices *fs_devices;
8a4b83cc 213
2b82032c
YZ		 214 while (!list_empty(&fs_uuids)) {
215 fs_devices = list_entry(fs_uuids.next,
216 struct btrfs_fs_devices, list);
217 list_del(&fs_devices->list);
e4404d6e 218 free_fs_devices(fs_devices);
8a4b83cc 219 }
8a4b83cc
CM		 220}
221
12bd2fc0
ID		 222static struct btrfs_device *__alloc_device(void)
223{
224 struct btrfs_device *dev;
225
78f2c9e6 226 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
12bd2fc0
ID		 227 if (!dev)
228 return ERR_PTR(-ENOMEM);
229
e0ae9994
DS		 230 /*
231 * Preallocate a bio that's always going to be used for flushing device
232 * barriers and matches the device lifespan
233 */
234 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
235 if (!dev->flush_bio) {
236 kfree(dev);
237 return ERR_PTR(-ENOMEM);
238 }
239 bio_get(dev->flush_bio);
240
12bd2fc0
ID		 241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
935e5cc9 243 INIT_LIST_HEAD(&dev->resized_list);
12bd2fc0
ID		 244
245 spin_lock_init(&dev->io_lock);
246
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
addc3fa7 249 atomic_set(&dev->dev_stats_ccnt, 0);
546bed63 250 btrfs_device_data_ordered_init(dev);
9bcaaea7 251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
d0164adc 252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
12bd2fc0
ID		 253
254 return dev;
255}
256
35c70103
DS		 257/*
258 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
259 * return NULL.
260 *
261 * If devid and uuid are both specified, the match must be exact, otherwise
262 * only devid is used.
263 */
264static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
265 u64 devid, const u8 *uuid)
8a4b83cc 266{
35c70103 267 struct list_head *head = &fs_devices->devices;
8a4b83cc 268 struct btrfs_device *dev;
8a4b83cc 269
c6e30871 270 list_for_each_entry(dev, head, dev_list) {
a443755f 271 if (dev->devid == devid &&
8f18cf13 272 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
8a4b83cc 273 return dev;
a443755f 274 }
8a4b83cc
CM		 275 }
276 return NULL;
277}
278
a1b32a59 279static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
8a4b83cc 280{
8a4b83cc
CM		 281 struct btrfs_fs_devices *fs_devices;
282
c6e30871 283 list_for_each_entry(fs_devices, &fs_uuids, list) {
8a4b83cc
CM		 284 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
285 return fs_devices;
286 }
287 return NULL;
288}
289
beaf8ab3
SB		 290static int
291btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
292 int flush, struct block_device **bdev,
293 struct buffer_head **bh)
294{
295 int ret;
296
297 *bdev = blkdev_get_by_path(device_path, flags, holder);
298
299 if (IS_ERR(*bdev)) {
300 ret = PTR_ERR(*bdev);
beaf8ab3
SB		 301 goto error;
302 }
303
304 if (flush)
305 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
9f6d2510 306 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
beaf8ab3
SB		 307 if (ret) {
308 blkdev_put(*bdev, flags);
309 goto error;
310 }
311 invalidate_bdev(*bdev);
312 *bh = btrfs_read_dev_super(*bdev);
92fc03fb
AJ		 313 if (IS_ERR(*bh)) {
314 ret = PTR_ERR(*bh);
beaf8ab3
SB		 315 blkdev_put(*bdev, flags);
316 goto error;
317 }
318
319 return 0;
320
321error:
322 *bdev = NULL;
323 *bh = NULL;
324 return ret;
325}
326
ffbd517d
CM		 327static void requeue_list(struct btrfs_pending_bios *pending_bios,
328 struct bio *head, struct bio *tail)
329{
330
331 struct bio *old_head;
332
333 old_head = pending_bios->head;
334 pending_bios->head = head;
335 if (pending_bios->tail)
336 tail->bi_next = old_head;
337 else
338 pending_bios->tail = tail;
339}
340
8b712842
CM		 341/*
342 * we try to collect pending bios for a device so we don't get a large
343 * number of procs sending bios down to the same device. This greatly
344 * improves the schedulers ability to collect and merge the bios.
345 *
346 * But, it also turns into a long list of bios to process and that is sure
347 * to eventually make the worker thread block. The solution here is to
348 * make some progress and then put this work struct back at the end of
349 * the list if the block device is congested. This way, multiple devices
350 * can make progress from a single worker thread.
351 */
143bede5 352static noinline void run_scheduled_bios(struct btrfs_device *device)
8b712842 353{
0b246afa 354 struct btrfs_fs_info *fs_info = device->fs_info;
8b712842
CM		 355 struct bio *pending;
356 struct backing_dev_info *bdi;
ffbd517d 357 struct btrfs_pending_bios *pending_bios;
8b712842
CM		 358 struct bio *tail;
359 struct bio *cur;
360 int again = 0;
ffbd517d 361 unsigned long num_run;
d644d8a1 362 unsigned long batch_run = 0;
b64a2851 363 unsigned long limit;
b765ead5 364 unsigned long last_waited = 0;
d84275c9 365 int force_reg = 0;
0e588859 366 int sync_pending = 0;
211588ad
CM		 367 struct blk_plug plug;
368
369 /*
370 * this function runs all the bios we've collected for
371 * a particular device. We don't want to wander off to
372 * another device without first sending all of these down.
373 * So, setup a plug here and finish it off before we return
374 */
375 blk_start_plug(&plug);
8b712842 376
efa7c9f9 377 bdi = device->bdev->bd_bdi;
b64a2851
CM		 378 limit = btrfs_async_submit_limit(fs_info);
379 limit = limit * 2 / 3;
380
8b712842
CM		 381loop:
382 spin_lock(&device->io_lock);
383
a6837051 384loop_lock:
d84275c9 385 num_run = 0;
ffbd517d 386
8b712842
CM		 387 /* take all the bios off the list at once and process them
388 * later on (without the lock held). But, remember the
389 * tail and other pointers so the bios can be properly reinserted
390 * into the list if we hit congestion
391 */
d84275c9 392 if (!force_reg && device->pending_sync_bios.head) {
ffbd517d 393 pending_bios = &device->pending_sync_bios;
d84275c9
CM		 394 force_reg = 1;
395 } else {
ffbd517d 396 pending_bios = &device->pending_bios;
d84275c9
CM		 397 force_reg = 0;
398 }
ffbd517d
CM		 399
400 pending = pending_bios->head;
401 tail = pending_bios->tail;
8b712842 402 WARN_ON(pending && !tail);
8b712842
CM		 403
404 /*
405 * if pending was null this time around, no bios need processing
406 * at all and we can stop. Otherwise it'll loop back up again
407 * and do an additional check so no bios are missed.
408 *
409 * device->running_pending is used to synchronize with the
410 * schedule_bio code.
411 */
ffbd517d
CM		 412 if (device->pending_sync_bios.head == NULL &&
413 device->pending_bios.head == NULL) {
8b712842
CM		 414 again = 0;
415 device->running_pending = 0;
ffbd517d
CM		 416 } else {
417 again = 1;
418 device->running_pending = 1;
8b712842 419 }
ffbd517d
CM		 420
421 pending_bios->head = NULL;
422 pending_bios->tail = NULL;
423
8b712842
CM		 424 spin_unlock(&device->io_lock);
425
d397712b 426 while (pending) {
ffbd517d
CM		 427
428 rmb();
d84275c9
CM		 429 /* we want to work on both lists, but do more bios on the
430 * sync list than the regular list
431 */
432 if ((num_run > 32 &&
433 pending_bios != &device->pending_sync_bios &&
434 device->pending_sync_bios.head) ||
435 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
436 device->pending_bios.head)) {
ffbd517d
CM		 437 spin_lock(&device->io_lock);
438 requeue_list(pending_bios, pending, tail);
439 goto loop_lock;
440 }
441
8b712842
CM		 442 cur = pending;
443 pending = pending->bi_next;
444 cur->bi_next = NULL;
b64a2851 445
ee863954
DS		 446 /*
447 * atomic_dec_return implies a barrier for waitqueue_active
448 */
66657b31 449 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
b64a2851
CM		 450 waitqueue_active(&fs_info->async_submit_wait))
451 wake_up(&fs_info->async_submit_wait);
492bb6de 452
dac56212 453 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
d644d8a1 454
2ab1ba68
CM		 455 /*
456 * if we're doing the sync list, record that our
457 * plug has some sync requests on it
458 *
459 * If we're doing the regular list and there are
460 * sync requests sitting around, unplug before
461 * we add more
462 */
463 if (pending_bios == &device->pending_sync_bios) {
464 sync_pending = 1;
465 } else if (sync_pending) {
466 blk_finish_plug(&plug);
467 blk_start_plug(&plug);
468 sync_pending = 0;
469 }
470
4e49ea4a 471 btrfsic_submit_bio(cur);
5ff7ba3a
CM		 472 num_run++;
473 batch_run++;
853d8ec4
DS		 474
475 cond_resched();
8b712842
CM		 476
477 /*
478 * we made progress, there is more work to do and the bdi
479 * is now congested. Back off and let other work structs
480 * run instead
481 */
57fd5a5f 482 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
5f2cc086 483 fs_info->fs_devices->open_devices > 1) {
b765ead5 484 struct io_context *ioc;
8b712842 485
b765ead5
CM		 486 ioc = current->io_context;
487
488 /*
489 * the main goal here is that we don't want to
490 * block if we're going to be able to submit
491 * more requests without blocking.
492 *
493 * This code does two great things, it pokes into
494 * the elevator code from a filesystem _and_
495 * it makes assumptions about how batching works.
496 */
497 if (ioc && ioc->nr_batch_requests > 0 &&
498 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
499 (last_waited == 0 ||
500 ioc->last_waited == last_waited)) {
501 /*
502 * we want to go through our batch of
503 * requests and stop. So, we copy out
504 * the ioc->last_waited time and test
505 * against it before looping
506 */
507 last_waited = ioc->last_waited;
853d8ec4 508 cond_resched();
b765ead5
CM		 509 continue;
510 }
8b712842 511 spin_lock(&device->io_lock);
ffbd517d 512 requeue_list(pending_bios, pending, tail);
a6837051 513 device->running_pending = 1;
8b712842
CM		 514
515 spin_unlock(&device->io_lock);
a8c93d4e
QW		 516 btrfs_queue_work(fs_info->submit_workers,
517 &device->work);
8b712842
CM		 518 goto done;
519 }
520 }
ffbd517d 521
51684082
CM		 522 cond_resched();
523 if (again)
524 goto loop;
525
526 spin_lock(&device->io_lock);
527 if (device->pending_bios.head || device->pending_sync_bios.head)
528 goto loop_lock;
529 spin_unlock(&device->io_lock);
530
8b712842 531done:
211588ad 532 blk_finish_plug(&plug);
8b712842
CM		 533}
534
b2950863 535static void pending_bios_fn(struct btrfs_work *work)
8b712842
CM		 536{
537 struct btrfs_device *device;
538
539 device = container_of(work, struct btrfs_device, work);
540 run_scheduled_bios(device);
541}
542
4fde46f0 543
c9162bdf 544static void btrfs_free_stale_device(struct btrfs_device *cur_dev)
4fde46f0
AJ		 545{
546 struct btrfs_fs_devices *fs_devs;
547 struct btrfs_device *dev;
548
549 if (!cur_dev->name)
550 return;
551
552 list_for_each_entry(fs_devs, &fs_uuids, list) {
553 int del = 1;
554
555 if (fs_devs->opened)
556 continue;
557 if (fs_devs->seeding)
558 continue;
559
560 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
561
562 if (dev == cur_dev)
563 continue;
564 if (!dev->name)
565 continue;
566
567 /*
568 * Todo: This won't be enough. What if the same device
569 * comes back (with new uuid and) with its mapper path?
570 * But for now, this does help as mostly an admin will
571 * either use mapper or non mapper path throughout.
572 */
573 rcu_read_lock();
574 del = strcmp(rcu_str_deref(dev->name),
575 rcu_str_deref(cur_dev->name));
576 rcu_read_unlock();
577 if (!del)
578 break;
579 }
580
581 if (!del) {
582 /* delete the stale device */
583 if (fs_devs->num_devices == 1) {
584 btrfs_sysfs_remove_fsid(fs_devs);
585 list_del(&fs_devs->list);
586 free_fs_devices(fs_devs);
587 } else {
588 fs_devs->num_devices--;
589 list_del(&dev->dev_list);
590 rcu_string_free(dev->name);
591 kfree(dev);
592 }
593 break;
594 }
595 }
596}
597
60999ca4
DS		 598/*
599 * Add new device to list of registered devices
600 *
601 * Returns:
602 * 1 - first time device is seen
603 * 0 - device already known
604 * < 0 - error
605 */
a1b32a59 606static noinline int device_list_add(const char *path,
8a4b83cc
CM		 607 struct btrfs_super_block *disk_super,
608 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
609{
610 struct btrfs_device *device;
611 struct btrfs_fs_devices *fs_devices;
606686ee 612 struct rcu_string *name;
60999ca4 613 int ret = 0;
8a4b83cc
CM		 614 u64 found_transid = btrfs_super_generation(disk_super);
615
616 fs_devices = find_fsid(disk_super->fsid);
617 if (!fs_devices) {
2208a378
ID		 618 fs_devices = alloc_fs_devices(disk_super->fsid);
619 if (IS_ERR(fs_devices))
620 return PTR_ERR(fs_devices);
621
8a4b83cc 622 list_add(&fs_devices->list, &fs_uuids);
2208a378 623
8a4b83cc
CM		 624 device = NULL;
625 } else {
35c70103
DS		 626 device = find_device(fs_devices, devid,
627 disk_super->dev_item.uuid);
8a4b83cc 628 }
443f24fe 629
8a4b83cc 630 if (!device) {
2b82032c
YZ		 631 if (fs_devices->opened)
632 return -EBUSY;
633
12bd2fc0
ID		 634 device = btrfs_alloc_device(NULL, &devid,
635 disk_super->dev_item.uuid);
636 if (IS_ERR(device)) {
8a4b83cc 637 /* we can safely leave the fs_devices entry around */
12bd2fc0 638 return PTR_ERR(device);
8a4b83cc 639 }
606686ee
JB		 640
641 name = rcu_string_strdup(path, GFP_NOFS);
642 if (!name) {
8a4b83cc
CM		 643 kfree(device);
644 return -ENOMEM;
645 }
606686ee 646 rcu_assign_pointer(device->name, name);
90519d66 647
e5e9a520 648 mutex_lock(&fs_devices->device_list_mutex);
1f78160c 649 list_add_rcu(&device->dev_list, &fs_devices->devices);
f7171750 650 fs_devices->num_devices++;
e5e9a520
CM		 651 mutex_unlock(&fs_devices->device_list_mutex);
652
60999ca4 653 ret = 1;
2b82032c 654 device->fs_devices = fs_devices;
606686ee 655 } else if (!device->name || strcmp(device->name->str, path)) {
b96de000
AJ		 656 /*
657 * When FS is already mounted.
658 * 1. If you are here and if the device->name is NULL that
659 * means this device was missing at time of FS mount.
660 * 2. If you are here and if the device->name is different
661 * from 'path' that means either
662 * a. The same device disappeared and reappeared with
663 * different name. or
664 * b. The missing-disk-which-was-replaced, has
665 * reappeared now.
666 *
667 * We must allow 1 and 2a above. But 2b would be a spurious
668 * and unintentional.
669 *
670 * Further in case of 1 and 2a above, the disk at 'path'
671 * would have missed some transaction when it was away and
672 * in case of 2a the stale bdev has to be updated as well.
673 * 2b must not be allowed at all time.
674 */
675
676 /*
0f23ae74
CM		 677 * For now, we do allow update to btrfs_fs_device through the
678 * btrfs dev scan cli after FS has been mounted. We're still
679 * tracking a problem where systems fail mount by subvolume id
680 * when we reject replacement on a mounted FS.
b96de000 681 */
0f23ae74 682 if (!fs_devices->opened && found_transid < device->generation) {
77bdae4d
AJ		 683 /*
684 * That is if the FS is _not_ mounted and if you
685 * are here, that means there is more than one
686 * disk with same uuid and devid.We keep the one
687 * with larger generation number or the last-in if
688 * generation are equal.
689 */
0f23ae74 690 return -EEXIST;
77bdae4d 691 }
b96de000 692
606686ee 693 name = rcu_string_strdup(path, GFP_NOFS);
3a0524dc
TH		 694 if (!name)
695 return -ENOMEM;
606686ee
JB		 696 rcu_string_free(device->name);
697 rcu_assign_pointer(device->name, name);
cd02dca5
CM		 698 if (device->missing) {
699 fs_devices->missing_devices--;
700 device->missing = 0;
701 }
8a4b83cc
CM		 702 }
703
77bdae4d
AJ		 704 /*
705 * Unmount does not free the btrfs_device struct but would zero
706 * generation along with most of the other members. So just update
707 * it back. We need it to pick the disk with largest generation
708 * (as above).
709 */
710 if (!fs_devices->opened)
711 device->generation = found_transid;
712
4fde46f0
AJ		 713 /*
714 * if there is new btrfs on an already registered device,
715 * then remove the stale device entry.
716 */
02feae3c
AJ		 717 if (ret > 0)
718 btrfs_free_stale_device(device);
4fde46f0 719
8a4b83cc 720 *fs_devices_ret = fs_devices;
60999ca4
DS		 721
722 return ret;
8a4b83cc
CM		 723}
724
e4404d6e
YZ		 725static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
726{
727 struct btrfs_fs_devices *fs_devices;
728 struct btrfs_device *device;
729 struct btrfs_device *orig_dev;
730
2208a378
ID		 731 fs_devices = alloc_fs_devices(orig->fsid);
732 if (IS_ERR(fs_devices))
733 return fs_devices;
e4404d6e 734
adbbb863 735 mutex_lock(&orig->device_list_mutex);
02db0844 736 fs_devices->total_devices = orig->total_devices;
e4404d6e 737
46224705 738 /* We have held the volume lock, it is safe to get the devices. */
e4404d6e 739 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
606686ee
JB		 740 struct rcu_string *name;
741
12bd2fc0
ID		 742 device = btrfs_alloc_device(NULL, &orig_dev->devid,
743 orig_dev->uuid);
744 if (IS_ERR(device))
e4404d6e
YZ		 745 goto error;
746
606686ee
JB		 747 /*
748 * This is ok to do without rcu read locked because we hold the
749 * uuid mutex so nothing we touch in here is going to disappear.
750 */
e755f780 751 if (orig_dev->name) {
78f2c9e6
DS		 752 name = rcu_string_strdup(orig_dev->name->str,
753 GFP_KERNEL);
e755f780
AJ		 754 if (!name) {
755 kfree(device);
756 goto error;
757 }
758 rcu_assign_pointer(device->name, name);
fd2696f3 759 }
e4404d6e 760
e4404d6e
YZ		 761 list_add(&device->dev_list, &fs_devices->devices);
762 device->fs_devices = fs_devices;
763 fs_devices->num_devices++;
764 }
adbbb863 765 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 766 return fs_devices;
767error:
adbbb863 768 mutex_unlock(&orig->device_list_mutex);
e4404d6e
YZ		 769 free_fs_devices(fs_devices);
770 return ERR_PTR(-ENOMEM);
771}
772
9eaed21e 773void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
dfe25020 774{
c6e30871 775 struct btrfs_device *device, *next;
443f24fe 776 struct btrfs_device *latest_dev = NULL;
a6b0d5c8 777
dfe25020
CM		 778 mutex_lock(&uuid_mutex);
779again:
46224705 780 /* This is the initialized path, it is safe to release the devices. */
c6e30871 781 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
a6b0d5c8 782 if (device->in_fs_metadata) {
63a212ab 783 if (!device->is_tgtdev_for_dev_replace &&
443f24fe
MX		 784 (!latest_dev ||
785 device->generation > latest_dev->generation)) {
786 latest_dev = device;
a6b0d5c8 787 }
2b82032c 788 continue;
a6b0d5c8 789 }
2b82032c 790
8dabb742
SB		 791 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
792 /*
793 * In the first step, keep the device which has
794 * the correct fsid and the devid that is used
795 * for the dev_replace procedure.
796 * In the second step, the dev_replace state is
797 * read from the device tree and it is known
798 * whether the procedure is really active or
799 * not, which means whether this device is
800 * used or whether it should be removed.
801 */
802 if (step == 0 || device->is_tgtdev_for_dev_replace) {
803 continue;
804 }
805 }
2b82032c 806 if (device->bdev) {
d4d77629 807 blkdev_put(device->bdev, device->mode);
2b82032c
YZ		 808 device->bdev = NULL;
809 fs_devices->open_devices--;
810 }
811 if (device->writeable) {
812 list_del_init(&device->dev_alloc_list);
813 device->writeable = 0;
8dabb742
SB		 814 if (!device->is_tgtdev_for_dev_replace)
815 fs_devices->rw_devices--;
2b82032c 816 }
e4404d6e
YZ		 817 list_del_init(&device->dev_list);
818 fs_devices->num_devices--;
606686ee 819 rcu_string_free(device->name);
e4404d6e 820 kfree(device);
dfe25020 821 }
2b82032c
YZ		 822
823 if (fs_devices->seed) {
824 fs_devices = fs_devices->seed;
2b82032c
YZ		 825 goto again;
826 }
827
443f24fe 828 fs_devices->latest_bdev = latest_dev->bdev;
a6b0d5c8 829
dfe25020 830 mutex_unlock(&uuid_mutex);
dfe25020 831}
a0af469b 832
1f78160c
XG		 833static void __free_device(struct work_struct *work)
834{
835 struct btrfs_device *device;
836
837 device = container_of(work, struct btrfs_device, rcu_work);
606686ee 838 rcu_string_free(device->name);
e0ae9994 839 bio_put(device->flush_bio);
1f78160c
XG		 840 kfree(device);
841}
842
843static void free_device(struct rcu_head *head)
844{
845 struct btrfs_device *device;
846
847 device = container_of(head, struct btrfs_device, rcu);
848
849 INIT_WORK(&device->rcu_work, __free_device);
850 schedule_work(&device->rcu_work);
851}
852
14238819
AJ		 853static void btrfs_close_bdev(struct btrfs_device *device)
854{
855 if (device->bdev && device->writeable) {
856 sync_blockdev(device->bdev);
857 invalidate_bdev(device->bdev);
858 }
859
860 if (device->bdev)
861 blkdev_put(device->bdev, device->mode);
862}
863
0ccd0528 864static void btrfs_prepare_close_one_device(struct btrfs_device *device)
f448341a
AJ		 865{
866 struct btrfs_fs_devices *fs_devices = device->fs_devices;
867 struct btrfs_device *new_device;
868 struct rcu_string *name;
869
870 if (device->bdev)
871 fs_devices->open_devices--;
872
873 if (device->writeable &&
874 device->devid != BTRFS_DEV_REPLACE_DEVID) {
875 list_del_init(&device->dev_alloc_list);
876 fs_devices->rw_devices--;
877 }
878
879 if (device->missing)
880 fs_devices->missing_devices--;
881
882 new_device = btrfs_alloc_device(NULL, &device->devid,
883 device->uuid);
884 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
885
886 /* Safe because we are under uuid_mutex */
887 if (device->name) {
888 name = rcu_string_strdup(device->name->str, GFP_NOFS);
889 BUG_ON(!name); /* -ENOMEM */
890 rcu_assign_pointer(new_device->name, name);
891 }
892
893 list_replace_rcu(&device->dev_list, &new_device->dev_list);
894 new_device->fs_devices = device->fs_devices;
f448341a
AJ		 895}
896
2b82032c 897static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
8a4b83cc 898{
2037a093 899 struct btrfs_device *device, *tmp;
0ccd0528
AJ		 900 struct list_head pending_put;
901
902 INIT_LIST_HEAD(&pending_put);
e4404d6e 903
2b82032c
YZ		 904 if (--fs_devices->opened > 0)
905 return 0;
8a4b83cc 906
c9513edb 907 mutex_lock(&fs_devices->device_list_mutex);
2037a093 908 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
0ccd0528
AJ		 909 btrfs_prepare_close_one_device(device);
910 list_add(&device->dev_list, &pending_put);
8a4b83cc 911 }
c9513edb
XG		 912 mutex_unlock(&fs_devices->device_list_mutex);
913
0ccd0528
AJ		 914 /*
915 * btrfs_show_devname() is using the device_list_mutex,
916 * sometimes call to blkdev_put() leads vfs calling
917 * into this func. So do put outside of device_list_mutex,
918 * as of now.
919 */
920 while (!list_empty(&pending_put)) {
921 device = list_first_entry(&pending_put,
922 struct btrfs_device, dev_list);
923 list_del(&device->dev_list);
924 btrfs_close_bdev(device);
925 call_rcu(&device->rcu, free_device);
926 }
927
e4404d6e
YZ		 928 WARN_ON(fs_devices->open_devices);
929 WARN_ON(fs_devices->rw_devices);
2b82032c
YZ		 930 fs_devices->opened = 0;
931 fs_devices->seeding = 0;
2b82032c 932
8a4b83cc
CM		 933 return 0;
934}
935
2b82032c
YZ		 936int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
937{
e4404d6e 938 struct btrfs_fs_devices *seed_devices = NULL;
2b82032c
YZ		 939 int ret;
940
941 mutex_lock(&uuid_mutex);
942 ret = __btrfs_close_devices(fs_devices);
e4404d6e
YZ		 943 if (!fs_devices->opened) {
944 seed_devices = fs_devices->seed;
945 fs_devices->seed = NULL;
946 }
2b82032c 947 mutex_unlock(&uuid_mutex);
e4404d6e
YZ		 948
949 while (seed_devices) {
950 fs_devices = seed_devices;
951 seed_devices = fs_devices->seed;
952 __btrfs_close_devices(fs_devices);
953 free_fs_devices(fs_devices);
954 }
bc178622
ES		 955 /*
956 * Wait for rcu kworkers under __btrfs_close_devices
957 * to finish all blkdev_puts so device is really
958 * free when umount is done.
959 */
960 rcu_barrier();
2b82032c
YZ		 961 return ret;
962}
963
e4404d6e
YZ		 964static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
965 fmode_t flags, void *holder)
8a4b83cc 966{
d5e2003c 967 struct request_queue *q;
8a4b83cc
CM		 968 struct block_device *bdev;
969 struct list_head *head = &fs_devices->devices;
8a4b83cc 970 struct btrfs_device *device;
443f24fe 971 struct btrfs_device *latest_dev = NULL;
a0af469b
CM		 972 struct buffer_head *bh;
973 struct btrfs_super_block *disk_super;
a0af469b 974 u64 devid;
2b82032c 975 int seeding = 1;
a0af469b 976 int ret = 0;
8a4b83cc 977
d4d77629
TH		 978 flags |= FMODE_EXCL;
979
c6e30871 980 list_for_each_entry(device, head, dev_list) {
c1c4d91c
CM		 981 if (device->bdev)
982 continue;
dfe25020
CM		 983 if (!device->name)
984 continue;
985
f63e0cca
ES		 986 /* Just open everything we can; ignore failures here */
987 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
988 &bdev, &bh))
beaf8ab3 989 continue;
a0af469b
CM		 990
991 disk_super = (struct btrfs_super_block *)bh->b_data;
a343832f 992 devid = btrfs_stack_device_id(&disk_super->dev_item);
a0af469b
CM		 993 if (devid != device->devid)
994 goto error_brelse;
995
2b82032c
YZ		 996 if (memcmp(device->uuid, disk_super->dev_item.uuid,
997 BTRFS_UUID_SIZE))
998 goto error_brelse;
999
1000 device->generation = btrfs_super_generation(disk_super);
443f24fe
MX		 1001 if (!latest_dev ||
1002 device->generation > latest_dev->generation)
1003 latest_dev = device;
a0af469b 1004
2b82032c
YZ		 1005 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1006 device->writeable = 0;
1007 } else {
1008 device->writeable = !bdev_read_only(bdev);
1009 seeding = 0;
1010 }
1011
d5e2003c 1012 q = bdev_get_queue(bdev);
90180da4 1013 if (blk_queue_discard(q))
d5e2003c 1014 device->can_discard = 1;
e884f4f0
AJ		 1015 if (!blk_queue_nonrot(q))
1016 fs_devices->rotating = 1;
d5e2003c 1017
8a4b83cc 1018 device->bdev = bdev;
dfe25020 1019 device->in_fs_metadata = 0;
15916de8
CM		 1020 device->mode = flags;
1021
a0af469b 1022 fs_devices->open_devices++;
55e50e45
ID		 1023 if (device->writeable &&
1024 device->devid != BTRFS_DEV_REPLACE_DEVID) {
2b82032c
YZ		 1025 fs_devices->rw_devices++;
1026 list_add(&device->dev_alloc_list,
1027 &fs_devices->alloc_list);
1028 }
4f6c9328 1029 brelse(bh);
a0af469b 1030 continue;
a061fc8d 1031
a0af469b
CM		 1032error_brelse:
1033 brelse(bh);
d4d77629 1034 blkdev_put(bdev, flags);
a0af469b 1035 continue;
8a4b83cc 1036 }
a0af469b 1037 if (fs_devices->open_devices == 0) {
20bcd649 1038 ret = -EINVAL;
a0af469b
CM		 1039 goto out;
1040 }
2b82032c
YZ		 1041 fs_devices->seeding = seeding;
1042 fs_devices->opened = 1;
443f24fe 1043 fs_devices->latest_bdev = latest_dev->bdev;
2b82032c 1044 fs_devices->total_rw_bytes = 0;
a0af469b 1045out:
2b82032c
YZ		 1046 return ret;
1047}
1048
1049int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
97288f2c 1050 fmode_t flags, void *holder)
2b82032c
YZ		 1051{
1052 int ret;
1053
1054 mutex_lock(&uuid_mutex);
1055 if (fs_devices->opened) {
e4404d6e
YZ		 1056 fs_devices->opened++;
1057 ret = 0;
2b82032c 1058 } else {
15916de8 1059 ret = __btrfs_open_devices(fs_devices, flags, holder);
2b82032c 1060 }
8a4b83cc 1061 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 1062 return ret;
1063}
1064
c9162bdf 1065static void btrfs_release_disk_super(struct page *page)
6cf86a00
AJ		 1066{
1067 kunmap(page);
1068 put_page(page);
1069}
1070
c9162bdf
OS		 1071static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1072 struct page **page,
1073 struct btrfs_super_block **disk_super)
6cf86a00
AJ		 1074{
1075 void *p;
1076 pgoff_t index;
1077
1078 /* make sure our super fits in the device */
1079 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1080 return 1;
1081
1082 /* make sure our super fits in the page */
1083 if (sizeof(**disk_super) > PAGE_SIZE)
1084 return 1;
1085
1086 /* make sure our super doesn't straddle pages on disk */
1087 index = bytenr >> PAGE_SHIFT;
1088 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1089 return 1;
1090
1091 /* pull in the page with our super */
1092 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1093 index, GFP_KERNEL);
1094
1095 if (IS_ERR_OR_NULL(*page))
1096 return 1;
1097
1098 p = kmap(*page);
1099
1100 /* align our pointer to the offset of the super block */
1101 *disk_super = p + (bytenr & ~PAGE_MASK);
1102
1103 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1104 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1105 btrfs_release_disk_super(*page);
1106 return 1;
1107 }
1108
1109 if ((*disk_super)->label[0] &&
1110 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1111 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1112
1113 return 0;
1114}
1115
6f60cbd3
DS		 1116/*
1117 * Look for a btrfs signature on a device. This may be called out of the mount path
1118 * and we are not allowed to call set_blocksize during the scan. The superblock
1119 * is read via pagecache
1120 */
97288f2c 1121int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
8a4b83cc
CM		 1122 struct btrfs_fs_devices **fs_devices_ret)
1123{
1124 struct btrfs_super_block *disk_super;
1125 struct block_device *bdev;
6f60cbd3 1126 struct page *page;
6f60cbd3 1127 int ret = -EINVAL;
8a4b83cc 1128 u64 devid;
f2984462 1129 u64 transid;
02db0844 1130 u64 total_devices;
6f60cbd3 1131 u64 bytenr;
8a4b83cc 1132
6f60cbd3
DS		 1133 /*
1134 * we would like to check all the supers, but that would make
1135 * a btrfs mount succeed after a mkfs from a different FS.
1136 * So, we need to add a special mount option to scan for
1137 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1138 */
1139 bytenr = btrfs_sb_offset(0);
d4d77629 1140 flags |= FMODE_EXCL;
10f6327b 1141 mutex_lock(&uuid_mutex);
6f60cbd3
DS		 1142
1143 bdev = blkdev_get_by_path(path, flags, holder);
6f60cbd3
DS		 1144 if (IS_ERR(bdev)) {
1145 ret = PTR_ERR(bdev);
beaf8ab3 1146 goto error;
6f60cbd3
DS		 1147 }
1148
6cf86a00 1149 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
6f60cbd3
DS		 1150 goto error_bdev_put;
1151
a343832f 1152 devid = btrfs_stack_device_id(&disk_super->dev_item);
f2984462 1153 transid = btrfs_super_generation(disk_super);
02db0844 1154 total_devices = btrfs_super_num_devices(disk_super);
6f60cbd3 1155
8a4b83cc 1156 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
60999ca4
DS		 1157 if (ret > 0) {
1158 if (disk_super->label[0]) {
62e85577 1159 pr_info("BTRFS: device label %s ", disk_super->label);
60999ca4 1160 } else {
62e85577 1161 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
60999ca4
DS		 1162 }
1163
62e85577 1164 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
60999ca4
DS		 1165 ret = 0;
1166 }
02db0844
JB		 1167 if (!ret && fs_devices_ret)
1168 (*fs_devices_ret)->total_devices = total_devices;
6f60cbd3 1169
6cf86a00 1170 btrfs_release_disk_super(page);
6f60cbd3
DS		 1171
1172error_bdev_put:
d4d77629 1173 blkdev_put(bdev, flags);
8a4b83cc 1174error:
beaf8ab3 1175 mutex_unlock(&uuid_mutex);
8a4b83cc
CM		 1176 return ret;
1177}
0b86a832 1178
6d07bcec
MX		 1179/* helper to account the used device space in the range */
1180int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1181 u64 end, u64 *length)
1182{
1183 struct btrfs_key key;
fb456252 1184 struct btrfs_root *root = device->fs_info->dev_root;
6d07bcec
MX		 1185 struct btrfs_dev_extent *dev_extent;
1186 struct btrfs_path *path;
1187 u64 extent_end;
1188 int ret;
1189 int slot;
1190 struct extent_buffer *l;
1191
1192 *length = 0;
1193
63a212ab 1194 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
6d07bcec
MX		 1195 return 0;
1196
1197 path = btrfs_alloc_path();
1198 if (!path)
1199 return -ENOMEM;
e4058b54 1200 path->reada = READA_FORWARD;
6d07bcec
MX		 1201
1202 key.objectid = device->devid;
1203 key.offset = start;
1204 key.type = BTRFS_DEV_EXTENT_KEY;
1205
1206 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1207 if (ret < 0)
1208 goto out;
1209 if (ret > 0) {
1210 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1211 if (ret < 0)
1212 goto out;
1213 }
1214
1215 while (1) {
1216 l = path->nodes[0];
1217 slot = path->slots[0];
1218 if (slot >= btrfs_header_nritems(l)) {
1219 ret = btrfs_next_leaf(root, path);
1220 if (ret == 0)
1221 continue;
1222 if (ret < 0)
1223 goto out;
1224
1225 break;
1226 }
1227 btrfs_item_key_to_cpu(l, &key, slot);
1228
1229 if (key.objectid < device->devid)
1230 goto next;
1231
1232 if (key.objectid > device->devid)
1233 break;
1234
962a298f 1235 if (key.type != BTRFS_DEV_EXTENT_KEY)
6d07bcec
MX		 1236 goto next;
1237
1238 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1239 extent_end = key.offset + btrfs_dev_extent_length(l,
1240 dev_extent);
1241 if (key.offset <= start && extent_end > end) {
1242 *length = end - start + 1;
1243 break;
1244 } else if (key.offset <= start && extent_end > start)
1245 *length += extent_end - start;
1246 else if (key.offset > start && extent_end <= end)
1247 *length += extent_end - key.offset;
1248 else if (key.offset > start && key.offset <= end) {
1249 *length += end - key.offset + 1;
1250 break;
1251 } else if (key.offset > end)
1252 break;
1253
1254next:
1255 path->slots[0]++;
1256 }
1257 ret = 0;
1258out:
1259 btrfs_free_path(path);
1260 return ret;
1261}
1262
499f377f 1263static int contains_pending_extent(struct btrfs_transaction *transaction,
6df9a95e
JB		 1264 struct btrfs_device *device,
1265 u64 *start, u64 len)
1266{
fb456252 1267 struct btrfs_fs_info *fs_info = device->fs_info;
6df9a95e 1268 struct extent_map *em;
499f377f 1269 struct list_head *search_list = &fs_info->pinned_chunks;
6df9a95e 1270 int ret = 0;
1b984508 1271 u64 physical_start = *start;
6df9a95e 1272
499f377f
JM		 1273 if (transaction)
1274 search_list = &transaction->pending_chunks;
04216820
FM		 1275again:
1276 list_for_each_entry(em, search_list, list) {
6df9a95e
JB		 1277 struct map_lookup *map;
1278 int i;
1279
95617d69 1280 map = em->map_lookup;
6df9a95e 1281 for (i = 0; i < map->num_stripes; i++) {
c152b63e
FM		 1282 u64 end;
1283
6df9a95e
JB		 1284 if (map->stripes[i].dev != device)
1285 continue;
1b984508 1286 if (map->stripes[i].physical >= physical_start + len ||
6df9a95e 1287 map->stripes[i].physical + em->orig_block_len <=
1b984508 1288 physical_start)
6df9a95e 1289 continue;
c152b63e
FM		 1290 /*
1291 * Make sure that while processing the pinned list we do
1292 * not override our *start with a lower value, because
1293 * we can have pinned chunks that fall within this
1294 * device hole and that have lower physical addresses
1295 * than the pending chunks we processed before. If we
1296 * do not take this special care we can end up getting
1297 * 2 pending chunks that start at the same physical
1298 * device offsets because the end offset of a pinned
1299 * chunk can be equal to the start offset of some
1300 * pending chunk.
1301 */
1302 end = map->stripes[i].physical + em->orig_block_len;
1303 if (end > *start) {
1304 *start = end;
1305 ret = 1;
1306 }
6df9a95e
JB		 1307 }
1308 }
499f377f
JM		 1309 if (search_list != &fs_info->pinned_chunks) {
1310 search_list = &fs_info->pinned_chunks;
04216820
FM		 1311 goto again;
1312 }
6df9a95e
JB		 1313
1314 return ret;
1315}
1316
1317
0b86a832 1318/*
499f377f
JM		 1319 * find_free_dev_extent_start - find free space in the specified device
1320 * @device: the device which we search the free space in
1321 * @num_bytes: the size of the free space that we need
1322 * @search_start: the position from which to begin the search
1323 * @start: store the start of the free space.
1324 * @len: the size of the free space. that we find, or the size
1325 * of the max free space if we don't find suitable free space
7bfc837d 1326 *
0b86a832
CM		 1327 * this uses a pretty simple search, the expectation is that it is
1328 * called very infrequently and that a given device has a small number
1329 * of extents
7bfc837d
MX		 1330 *
1331 * @start is used to store the start of the free space if we find. But if we
1332 * don't find suitable free space, it will be used to store the start position
1333 * of the max free space.
1334 *
1335 * @len is used to store the size of the free space that we find.
1336 * But if we don't find suitable free space, it is used to store the size of
1337 * the max free space.
0b86a832 1338 */
499f377f
JM		 1339int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1340 struct btrfs_device *device, u64 num_bytes,
1341 u64 search_start, u64 *start, u64 *len)
0b86a832 1342{
0b246afa
JM		 1343 struct btrfs_fs_info *fs_info = device->fs_info;
1344 struct btrfs_root *root = fs_info->dev_root;
0b86a832 1345 struct btrfs_key key;
7bfc837d 1346 struct btrfs_dev_extent *dev_extent;
2b82032c 1347 struct btrfs_path *path;
7bfc837d
MX		 1348 u64 hole_size;
1349 u64 max_hole_start;
1350 u64 max_hole_size;
1351 u64 extent_end;
0b86a832
CM		 1352 u64 search_end = device->total_bytes;
1353 int ret;
7bfc837d 1354 int slot;
0b86a832 1355 struct extent_buffer *l;
8cdc7c5b
FM		 1356
1357 /*
1358 * We don't want to overwrite the superblock on the drive nor any area
1359 * used by the boot loader (grub for example), so we make sure to start
1360 * at an offset of at least 1MB.
1361 */
0d0c71b3 1362 search_start = max_t(u64, search_start, SZ_1M);
0b86a832 1363
6df9a95e
JB		 1364 path = btrfs_alloc_path();
1365 if (!path)
1366 return -ENOMEM;
f2ab7618 1367
7bfc837d
MX		 1368 max_hole_start = search_start;
1369 max_hole_size = 0;
1370
f2ab7618 1371again:
63a212ab 1372 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
7bfc837d 1373 ret = -ENOSPC;
6df9a95e 1374 goto out;
7bfc837d
MX		 1375 }
1376
e4058b54 1377 path->reada = READA_FORWARD;
6df9a95e
JB		 1378 path->search_commit_root = 1;
1379 path->skip_locking = 1;
7bfc837d 1380
0b86a832
CM		 1381 key.objectid = device->devid;
1382 key.offset = search_start;
1383 key.type = BTRFS_DEV_EXTENT_KEY;
7bfc837d 1384
125ccb0a 1385 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
0b86a832 1386 if (ret < 0)
7bfc837d 1387 goto out;
1fcbac58
YZ		 1388 if (ret > 0) {
1389 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1390 if (ret < 0)
7bfc837d 1391 goto out;
1fcbac58 1392 }
7bfc837d 1393
0b86a832
CM		 1394 while (1) {
1395 l = path->nodes[0];
1396 slot = path->slots[0];
1397 if (slot >= btrfs_header_nritems(l)) {
1398 ret = btrfs_next_leaf(root, path);
1399 if (ret == 0)
1400 continue;
1401 if (ret < 0)
7bfc837d
MX		 1402 goto out;
1403
1404 break;
0b86a832
CM		 1405 }
1406 btrfs_item_key_to_cpu(l, &key, slot);
1407
1408 if (key.objectid < device->devid)
1409 goto next;
1410
1411 if (key.objectid > device->devid)
7bfc837d 1412 break;
0b86a832 1413
962a298f 1414 if (key.type != BTRFS_DEV_EXTENT_KEY)
7bfc837d 1415 goto next;
9779b72f 1416
7bfc837d
MX		 1417 if (key.offset > search_start) {
1418 hole_size = key.offset - search_start;
9779b72f 1419
6df9a95e
JB		 1420 /*
1421 * Have to check before we set max_hole_start, otherwise
1422 * we could end up sending back this offset anyway.
1423 */
499f377f 1424 if (contains_pending_extent(transaction, device,
6df9a95e 1425 &search_start,
1b984508
FL		 1426 hole_size)) {
1427 if (key.offset >= search_start) {
1428 hole_size = key.offset - search_start;
1429 } else {
1430 WARN_ON_ONCE(1);
1431 hole_size = 0;
1432 }
1433 }
6df9a95e 1434
7bfc837d
MX		 1435 if (hole_size > max_hole_size) {
1436 max_hole_start = search_start;
1437 max_hole_size = hole_size;
1438 }
9779b72f 1439
7bfc837d
MX		 1440 /*
1441 * If this free space is greater than which we need,
1442 * it must be the max free space that we have found
1443 * until now, so max_hole_start must point to the start
1444 * of this free space and the length of this free space
1445 * is stored in max_hole_size. Thus, we return
1446 * max_hole_start and max_hole_size and go back to the
1447 * caller.
1448 */
1449 if (hole_size >= num_bytes) {
1450 ret = 0;
1451 goto out;
0b86a832
CM		 1452 }
1453 }
0b86a832 1454
0b86a832 1455 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
7bfc837d
MX		 1456 extent_end = key.offset + btrfs_dev_extent_length(l,
1457 dev_extent);
1458 if (extent_end > search_start)
1459 search_start = extent_end;
0b86a832
CM		 1460next:
1461 path->slots[0]++;
1462 cond_resched();
1463 }
0b86a832 1464
38c01b96 1465 /*
1466 * At this point, search_start should be the end of
1467 * allocated dev extents, and when shrinking the device,
1468 * search_end may be smaller than search_start.
1469 */
f2ab7618 1470 if (search_end > search_start) {
38c01b96 1471 hole_size = search_end - search_start;
1472
499f377f 1473 if (contains_pending_extent(transaction, device, &search_start,
f2ab7618
ZL		 1474 hole_size)) {
1475 btrfs_release_path(path);
1476 goto again;
1477 }
0b86a832 1478
f2ab7618
ZL		 1479 if (hole_size > max_hole_size) {
1480 max_hole_start = search_start;
1481 max_hole_size = hole_size;
1482 }
6df9a95e
JB		 1483 }
1484
7bfc837d 1485 /* See above. */
f2ab7618 1486 if (max_hole_size < num_bytes)
7bfc837d
MX		 1487 ret = -ENOSPC;
1488 else
1489 ret = 0;
1490
1491out:
2b82032c 1492 btrfs_free_path(path);
7bfc837d 1493 *start = max_hole_start;
b2117a39 1494 if (len)
7bfc837d 1495 *len = max_hole_size;
0b86a832
CM		 1496 return ret;
1497}
1498
499f377f
JM		 1499int find_free_dev_extent(struct btrfs_trans_handle *trans,
1500 struct btrfs_device *device, u64 num_bytes,
1501 u64 *start, u64 *len)
1502{
499f377f 1503 /* FIXME use last free of some kind */
499f377f 1504 return find_free_dev_extent_start(trans->transaction, device,
8cdc7c5b 1505 num_bytes, 0, start, len);
499f377f
JM		 1506}
1507
b2950863 1508static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
8f18cf13 1509 struct btrfs_device *device,
2196d6e8 1510 u64 start, u64 *dev_extent_len)
8f18cf13 1511{
0b246afa
JM		 1512 struct btrfs_fs_info *fs_info = device->fs_info;
1513 struct btrfs_root *root = fs_info->dev_root;
8f18cf13
CM		 1514 int ret;
1515 struct btrfs_path *path;
8f18cf13 1516 struct btrfs_key key;
a061fc8d
CM		 1517 struct btrfs_key found_key;
1518 struct extent_buffer *leaf = NULL;
1519 struct btrfs_dev_extent *extent = NULL;
8f18cf13
CM		 1520
1521 path = btrfs_alloc_path();
1522 if (!path)
1523 return -ENOMEM;
1524
1525 key.objectid = device->devid;
1526 key.offset = start;
1527 key.type = BTRFS_DEV_EXTENT_KEY;
924cd8fb 1528again:
8f18cf13 1529 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
a061fc8d
CM		 1530 if (ret > 0) {
1531 ret = btrfs_previous_item(root, path, key.objectid,
1532 BTRFS_DEV_EXTENT_KEY);
b0b802d7
TI		 1533 if (ret)
1534 goto out;
a061fc8d
CM		 1535 leaf = path->nodes[0];
1536 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1537 extent = btrfs_item_ptr(leaf, path->slots[0],
1538 struct btrfs_dev_extent);
1539 BUG_ON(found_key.offset > start || found_key.offset +
1540 btrfs_dev_extent_length(leaf, extent) < start);
924cd8fb
MX		 1541 key = found_key;
1542 btrfs_release_path(path);
1543 goto again;
a061fc8d
CM		 1544 } else if (ret == 0) {
1545 leaf = path->nodes[0];
1546 extent = btrfs_item_ptr(leaf, path->slots[0],
1547 struct btrfs_dev_extent);
79787eaa 1548 } else {
0b246afa 1549 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
79787eaa 1550 goto out;
a061fc8d 1551 }
8f18cf13 1552
2196d6e8
MX		 1553 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1554
8f18cf13 1555 ret = btrfs_del_item(trans, root, path);
79787eaa 1556 if (ret) {
0b246afa
JM		 1557 btrfs_handle_fs_error(fs_info, ret,
1558 "Failed to remove dev extent item");
13212b54 1559 } else {
3204d33c 1560 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
79787eaa 1561 }
b0b802d7 1562out:
8f18cf13
CM		 1563 btrfs_free_path(path);
1564 return ret;
1565}
1566
48a3b636
ES		 1567static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1568 struct btrfs_device *device,
48a3b636 1569 u64 chunk_offset, u64 start, u64 num_bytes)
0b86a832
CM		 1570{
1571 int ret;
1572 struct btrfs_path *path;
0b246afa
JM		 1573 struct btrfs_fs_info *fs_info = device->fs_info;
1574 struct btrfs_root *root = fs_info->dev_root;
0b86a832
CM		 1575 struct btrfs_dev_extent *extent;
1576 struct extent_buffer *leaf;
1577 struct btrfs_key key;
1578
dfe25020 1579 WARN_ON(!device->in_fs_metadata);
63a212ab 1580 WARN_ON(device->is_tgtdev_for_dev_replace);
0b86a832
CM		 1581 path = btrfs_alloc_path();
1582 if (!path)
1583 return -ENOMEM;
1584
0b86a832 1585 key.objectid = device->devid;
2b82032c 1586 key.offset = start;
0b86a832
CM		 1587 key.type = BTRFS_DEV_EXTENT_KEY;
1588 ret = btrfs_insert_empty_item(trans, root, path, &key,
1589 sizeof(*extent));
2cdcecbc
MF		 1590 if (ret)
1591 goto out;
0b86a832
CM		 1592
1593 leaf = path->nodes[0];
1594 extent = btrfs_item_ptr(leaf, path->slots[0],
1595 struct btrfs_dev_extent);
b5d9071c
NB		 1596 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1597 BTRFS_CHUNK_TREE_OBJECTID);
0ca00afb
NB		 1598 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1599 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
e17cade2
CM		 1600 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1601
0b86a832
CM		 1602 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1603 btrfs_mark_buffer_dirty(leaf);
2cdcecbc 1604out:
0b86a832
CM		 1605 btrfs_free_path(path);
1606 return ret;
1607}
1608
6df9a95e 1609static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
0b86a832 1610{
6df9a95e
JB		 1611 struct extent_map_tree *em_tree;
1612 struct extent_map *em;
1613 struct rb_node *n;
1614 u64 ret = 0;
0b86a832 1615
6df9a95e
JB		 1616 em_tree = &fs_info->mapping_tree.map_tree;
1617 read_lock(&em_tree->lock);
1618 n = rb_last(&em_tree->map);
1619 if (n) {
1620 em = rb_entry(n, struct extent_map, rb_node);
1621 ret = em->start + em->len;
0b86a832 1622 }
6df9a95e
JB		 1623 read_unlock(&em_tree->lock);
1624
0b86a832
CM		 1625 return ret;
1626}
1627
53f10659
ID		 1628static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1629 u64 *devid_ret)
0b86a832
CM		 1630{
1631 int ret;
1632 struct btrfs_key key;
1633 struct btrfs_key found_key;
2b82032c
YZ		 1634 struct btrfs_path *path;
1635
2b82032c
YZ		 1636 path = btrfs_alloc_path();
1637 if (!path)
1638 return -ENOMEM;
0b86a832
CM		 1639
1640 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1641 key.type = BTRFS_DEV_ITEM_KEY;
1642 key.offset = (u64)-1;
1643
53f10659 1644 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
0b86a832
CM		 1645 if (ret < 0)
1646 goto error;
1647
79787eaa 1648 BUG_ON(ret == 0); /* Corruption */
0b86a832 1649
53f10659
ID		 1650 ret = btrfs_previous_item(fs_info->chunk_root, path,
1651 BTRFS_DEV_ITEMS_OBJECTID,
0b86a832
CM		 1652 BTRFS_DEV_ITEM_KEY);
1653 if (ret) {
53f10659 1654 *devid_ret = 1;
0b86a832
CM		 1655 } else {
1656 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1657 path->slots[0]);
53f10659 1658 *devid_ret = found_key.offset + 1;
0b86a832
CM		 1659 }
1660 ret = 0;
1661error:
2b82032c 1662 btrfs_free_path(path);
0b86a832
CM		 1663 return ret;
1664}
1665
1666/*
1667 * the device information is stored in the chunk root
1668 * the btrfs_device struct should be fully filled in
1669 */
48a3b636 1670static int btrfs_add_device(struct btrfs_trans_handle *trans,
5b4aacef 1671 struct btrfs_fs_info *fs_info,
48a3b636 1672 struct btrfs_device *device)
0b86a832 1673{
5b4aacef 1674 struct btrfs_root *root = fs_info->chunk_root;
0b86a832
CM		 1675 int ret;
1676 struct btrfs_path *path;
1677 struct btrfs_dev_item *dev_item;
1678 struct extent_buffer *leaf;
1679 struct btrfs_key key;
1680 unsigned long ptr;
0b86a832 1681
0b86a832
CM		 1682 path = btrfs_alloc_path();
1683 if (!path)
1684 return -ENOMEM;
1685
0b86a832
CM		 1686 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687 key.type = BTRFS_DEV_ITEM_KEY;
2b82032c 1688 key.offset = device->devid;
0b86a832
CM		 1689
1690 ret = btrfs_insert_empty_item(trans, root, path, &key,
0d81ba5d 1691 sizeof(*dev_item));
0b86a832
CM		 1692 if (ret)
1693 goto out;
1694
1695 leaf = path->nodes[0];
1696 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1697
1698 btrfs_set_device_id(leaf, dev_item, device->devid);
2b82032c 1699 btrfs_set_device_generation(leaf, dev_item, 0);
0b86a832
CM		 1700 btrfs_set_device_type(leaf, dev_item, device->type);
1701 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1702 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1703 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
7cc8e58d
MX		 1704 btrfs_set_device_total_bytes(leaf, dev_item,
1705 btrfs_device_get_disk_total_bytes(device));
1706 btrfs_set_device_bytes_used(leaf, dev_item,
1707 btrfs_device_get_bytes_used(device));
e17cade2
CM		 1708 btrfs_set_device_group(leaf, dev_item, 0);
1709 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1710 btrfs_set_device_bandwidth(leaf, dev_item, 0);
c3027eb5 1711 btrfs_set_device_start_offset(leaf, dev_item, 0);
0b86a832 1712
410ba3a2 1713 ptr = btrfs_device_uuid(dev_item);
e17cade2 1714 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1473b24e 1715 ptr = btrfs_device_fsid(dev_item);
44880fdc 1716 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
0b86a832 1717 btrfs_mark_buffer_dirty(leaf);
0b86a832 1718
2b82032c 1719 ret = 0;
0b86a832
CM		 1720out:
1721 btrfs_free_path(path);
1722 return ret;
1723}
8f18cf13 1724
5a1972bd
QW		 1725/*
1726 * Function to update ctime/mtime for a given device path.
1727 * Mainly used for ctime/mtime based probe like libblkid.
1728 */
da353f6b 1729static void update_dev_time(const char *path_name)
5a1972bd
QW		 1730{
1731 struct file *filp;
1732
1733 filp = filp_open(path_name, O_RDWR, 0);
98af592f 1734 if (IS_ERR(filp))
5a1972bd
QW		 1735 return;
1736 file_update_time(filp);
1737 filp_close(filp, NULL);
5a1972bd
QW		 1738}
1739
5b4aacef 1740static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
a061fc8d
CM		 1741 struct btrfs_device *device)
1742{
5b4aacef 1743 struct btrfs_root *root = fs_info->chunk_root;
a061fc8d
CM		 1744 int ret;
1745 struct btrfs_path *path;
a061fc8d 1746 struct btrfs_key key;
a061fc8d
CM		 1747 struct btrfs_trans_handle *trans;
1748
a061fc8d
CM		 1749 path = btrfs_alloc_path();
1750 if (!path)
1751 return -ENOMEM;
1752
a22285a6 1753 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 1754 if (IS_ERR(trans)) {
1755 btrfs_free_path(path);
1756 return PTR_ERR(trans);
1757 }
a061fc8d
CM		 1758 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1759 key.type = BTRFS_DEV_ITEM_KEY;
1760 key.offset = device->devid;
1761
1762 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1763 if (ret < 0)
1764 goto out;
1765
1766 if (ret > 0) {
1767 ret = -ENOENT;
1768 goto out;
1769 }
1770
1771 ret = btrfs_del_item(trans, root, path);
1772 if (ret)
1773 goto out;
a061fc8d
CM		 1774out:
1775 btrfs_free_path(path);
3a45bb20 1776 btrfs_commit_transaction(trans);
a061fc8d
CM		 1777 return ret;
1778}
1779
3cc31a0d
DS		 1780/*
1781 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1782 * filesystem. It's up to the caller to adjust that number regarding eg. device
1783 * replace.
1784 */
1785static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1786 u64 num_devices)
a061fc8d 1787{
a061fc8d 1788 u64 all_avail;
de98ced9 1789 unsigned seq;
418775a2 1790 int i;
a061fc8d 1791
de98ced9 1792 do {
bd45ffbc 1793 seq = read_seqbegin(&fs_info->profiles_lock);
de98ced9 1794
bd45ffbc
AJ		 1795 all_avail = fs_info->avail_data_alloc_bits |
1796 fs_info->avail_system_alloc_bits |
1797 fs_info->avail_metadata_alloc_bits;
1798 } while (read_seqretry(&fs_info->profiles_lock, seq));
a061fc8d 1799
418775a2
DS		 1800 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1801 if (!(all_avail & btrfs_raid_group[i]))
1802 continue;
a061fc8d 1803
418775a2
DS		 1804 if (num_devices < btrfs_raid_array[i].devs_min) {
1805 int ret = btrfs_raid_mindev_error[i];
bd45ffbc 1806
418775a2
DS		 1807 if (ret)
1808 return ret;
1809 }
53b381b3
DW		 1810 }
1811
bd45ffbc 1812 return 0;
f1fa7f26
AJ		 1813}
1814
c9162bdf
OS		 1815static struct btrfs_device * btrfs_find_next_active_device(
1816 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
a061fc8d 1817{
2b82032c 1818 struct btrfs_device *next_device;
88acff64
AJ		 1819
1820 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1821 if (next_device != device &&
1822 !next_device->missing && next_device->bdev)
1823 return next_device;
1824 }
1825
1826 return NULL;
1827}
1828
1829/*
1830 * Helper function to check if the given device is part of s_bdev / latest_bdev
1831 * and replace it with the provided or the next active device, in the context
1832 * where this function called, there should be always be another device (or
1833 * this_dev) which is active.
1834 */
1835void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1836 struct btrfs_device *device, struct btrfs_device *this_dev)
1837{
1838 struct btrfs_device *next_device;
1839
1840 if (this_dev)
1841 next_device = this_dev;
1842 else
1843 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1844 device);
1845 ASSERT(next_device);
1846
1847 if (fs_info->sb->s_bdev &&
1848 (fs_info->sb->s_bdev == device->bdev))
1849 fs_info->sb->s_bdev = next_device->bdev;
1850
1851 if (fs_info->fs_devices->latest_bdev == device->bdev)
1852 fs_info->fs_devices->latest_bdev = next_device->bdev;
1853}
1854
da353f6b
DS		 1855int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1856 u64 devid)
f1fa7f26
AJ		 1857{
1858 struct btrfs_device *device;
1f78160c 1859 struct btrfs_fs_devices *cur_devices;
2b82032c 1860 u64 num_devices;
a061fc8d
CM		 1861 int ret = 0;
1862
a061fc8d
CM		 1863 mutex_lock(&uuid_mutex);
1864
0b246afa
JM		 1865 num_devices = fs_info->fs_devices->num_devices;
1866 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1867 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
8dabb742
SB		 1868 WARN_ON(num_devices < 1);
1869 num_devices--;
1870 }
0b246afa 1871 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
8dabb742 1872
0b246afa 1873 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
f1fa7f26 1874 if (ret)
a061fc8d 1875 goto out;
a061fc8d 1876
2ff7e61e
JM		 1877 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1878 &device);
24fc572f 1879 if (ret)
53b381b3 1880 goto out;
dfe25020 1881
63a212ab 1882 if (device->is_tgtdev_for_dev_replace) {
183860f6 1883 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
24fc572f 1884 goto out;
63a212ab
SB		 1885 }
1886
0b246afa 1887 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
183860f6 1888 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
24fc572f 1889 goto out;
2b82032c
YZ		 1890 }
1891
1892 if (device->writeable) {
34441361 1893 mutex_lock(&fs_info->chunk_mutex);
2b82032c 1894 list_del_init(&device->dev_alloc_list);
c3929c36 1895 device->fs_devices->rw_devices--;
34441361 1896 mutex_unlock(&fs_info->chunk_mutex);
dfe25020 1897 }
a061fc8d 1898
d7901554 1899 mutex_unlock(&uuid_mutex);
a061fc8d 1900 ret = btrfs_shrink_device(device, 0);
d7901554 1901 mutex_lock(&uuid_mutex);
a061fc8d 1902 if (ret)
9b3517e9 1903 goto error_undo;
a061fc8d 1904
63a212ab
SB		 1905 /*
1906 * TODO: the superblock still includes this device in its num_devices
1907 * counter although write_all_supers() is not locked out. This
1908 * could give a filesystem state which requires a degraded mount.
1909 */
0b246afa 1910 ret = btrfs_rm_dev_item(fs_info, device);
a061fc8d 1911 if (ret)
9b3517e9 1912 goto error_undo;
a061fc8d 1913
2b82032c 1914 device->in_fs_metadata = 0;
0b246afa 1915 btrfs_scrub_cancel_dev(fs_info, device);
e5e9a520
CM		 1916
1917 /*
1918 * the device list mutex makes sure that we don't change
1919 * the device list while someone else is writing out all
d7306801
FDBM		 1920 * the device supers. Whoever is writing all supers, should
1921 * lock the device list mutex before getting the number of
1922 * devices in the super block (super_copy). Conversely,
1923 * whoever updates the number of devices in the super block
1924 * (super_copy) should hold the device list mutex.
e5e9a520 1925 */
1f78160c
XG		 1926
1927 cur_devices = device->fs_devices;
0b246afa 1928 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1f78160c 1929 list_del_rcu(&device->dev_list);
e5e9a520 1930
e4404d6e 1931 device->fs_devices->num_devices--;
02db0844 1932 device->fs_devices->total_devices--;
2b82032c 1933
cd02dca5 1934 if (device->missing)
3a7d55c8 1935 device->fs_devices->missing_devices--;
cd02dca5 1936
0b246afa 1937 btrfs_assign_next_active_device(fs_info, device, NULL);
2b82032c 1938
0bfaa9c5 1939 if (device->bdev) {
e4404d6e 1940 device->fs_devices->open_devices--;
0bfaa9c5 1941 /* remove sysfs entry */
0b246afa 1942 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
0bfaa9c5 1943 }
99994cde 1944
0b246afa
JM		 1945 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1946 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1947 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2b82032c 1948
cea67ab9
JM		 1949 /*
1950 * at this point, the device is zero sized and detached from
1951 * the devices list. All that's left is to zero out the old
1952 * supers and free the device.
1953 */
1954 if (device->writeable)
1955 btrfs_scratch_superblocks(device->bdev, device->name->str);
1956
1957 btrfs_close_bdev(device);
1958 call_rcu(&device->rcu, free_device);
1959
1f78160c 1960 if (cur_devices->open_devices == 0) {
e4404d6e 1961 struct btrfs_fs_devices *fs_devices;
0b246afa 1962 fs_devices = fs_info->fs_devices;
e4404d6e 1963 while (fs_devices) {
8321cf25
RS		 1964 if (fs_devices->seed == cur_devices) {
1965 fs_devices->seed = cur_devices->seed;
e4404d6e 1966 break;
8321cf25 1967 }
e4404d6e 1968 fs_devices = fs_devices->seed;
2b82032c 1969 }
1f78160c 1970 cur_devices->seed = NULL;
1f78160c 1971 __btrfs_close_devices(cur_devices);
1f78160c 1972 free_fs_devices(cur_devices);
2b82032c
YZ		 1973 }
1974
a061fc8d
CM		 1975out:
1976 mutex_unlock(&uuid_mutex);
a061fc8d 1977 return ret;
24fc572f 1978
9b3517e9
ID		 1979error_undo:
1980 if (device->writeable) {
34441361 1981 mutex_lock(&fs_info->chunk_mutex);
9b3517e9 1982 list_add(&device->dev_alloc_list,
0b246afa 1983 &fs_info->fs_devices->alloc_list);
c3929c36 1984 device->fs_devices->rw_devices++;
34441361 1985 mutex_unlock(&fs_info->chunk_mutex);
9b3517e9 1986 }
24fc572f 1987 goto out;
a061fc8d
CM		 1988}
1989
084b6e7c
QW		 1990void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1991 struct btrfs_device *srcdev)
e93c89c1 1992{
d51908ce
AJ		 1993 struct btrfs_fs_devices *fs_devices;
1994
e93c89c1 1995 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1357272f 1996
25e8e911
AJ		 1997 /*
1998 * in case of fs with no seed, srcdev->fs_devices will point
1999 * to fs_devices of fs_info. However when the dev being replaced is
2000 * a seed dev it will point to the seed's local fs_devices. In short
2001 * srcdev will have its correct fs_devices in both the cases.
2002 */
2003 fs_devices = srcdev->fs_devices;
d51908ce 2004
e93c89c1
SB		 2005 list_del_rcu(&srcdev->dev_list);
2006 list_del_rcu(&srcdev->dev_alloc_list);
d51908ce 2007 fs_devices->num_devices--;
82372bc8 2008 if (srcdev->missing)
d51908ce 2009 fs_devices->missing_devices--;
e93c89c1 2010
48b3b9d4 2011 if (srcdev->writeable)
82372bc8 2012 fs_devices->rw_devices--;
1357272f 2013
82372bc8 2014 if (srcdev->bdev)
d51908ce 2015 fs_devices->open_devices--;
084b6e7c
QW		 2016}
2017
2018void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2019 struct btrfs_device *srcdev)
2020{
2021 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
e93c89c1 2022
48b3b9d4
AJ		 2023 if (srcdev->writeable) {
2024 /* zero out the old super if it is writable */
2025 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2026 }
14238819
AJ		 2027
2028 btrfs_close_bdev(srcdev);
2029
e93c89c1 2030 call_rcu(&srcdev->rcu, free_device);
94d5f0c2
AJ		 2031
2032 /*
2033 * unless fs_devices is seed fs, num_devices shouldn't go
2034 * zero
2035 */
2036 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2037
2038 /* if this is no devs we rather delete the fs_devices */
2039 if (!fs_devices->num_devices) {
2040 struct btrfs_fs_devices *tmp_fs_devices;
2041
2042 tmp_fs_devices = fs_info->fs_devices;
2043 while (tmp_fs_devices) {
2044 if (tmp_fs_devices->seed == fs_devices) {
2045 tmp_fs_devices->seed = fs_devices->seed;
2046 break;
2047 }
2048 tmp_fs_devices = tmp_fs_devices->seed;
2049 }
2050 fs_devices->seed = NULL;
8bef8401
AJ		 2051 __btrfs_close_devices(fs_devices);
2052 free_fs_devices(fs_devices);
94d5f0c2 2053 }
e93c89c1
SB		 2054}
2055
2056void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2057 struct btrfs_device *tgtdev)
2058{
67a2c45e 2059 mutex_lock(&uuid_mutex);
e93c89c1
SB		 2060 WARN_ON(!tgtdev);
2061 mutex_lock(&fs_info->fs_devices->device_list_mutex);
d2ff1b20 2062
32576040 2063 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
d2ff1b20 2064
779bf3fe 2065 if (tgtdev->bdev)
e93c89c1 2066 fs_info->fs_devices->open_devices--;
779bf3fe 2067
e93c89c1 2068 fs_info->fs_devices->num_devices--;
e93c89c1 2069
88acff64 2070 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
e93c89c1 2071
e93c89c1 2072 list_del_rcu(&tgtdev->dev_list);
e93c89c1
SB		 2073
2074 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
67a2c45e 2075 mutex_unlock(&uuid_mutex);
779bf3fe
AJ		 2076
2077 /*
2078 * The update_dev_time() with in btrfs_scratch_superblocks()
2079 * may lead to a call to btrfs_show_devname() which will try
2080 * to hold device_list_mutex. And here this device
2081 * is already out of device list, so we don't have to hold
2082 * the device_list_mutex lock.
2083 */
2084 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
14238819
AJ		 2085
2086 btrfs_close_bdev(tgtdev);
779bf3fe 2087 call_rcu(&tgtdev->rcu, free_device);
e93c89c1
SB		 2088}
2089
2ff7e61e 2090static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
da353f6b 2091 const char *device_path,
48a3b636 2092 struct btrfs_device **device)
7ba15b7d
SB		 2093{
2094 int ret = 0;
2095 struct btrfs_super_block *disk_super;
2096 u64 devid;
2097 u8 *dev_uuid;
2098 struct block_device *bdev;
2099 struct buffer_head *bh;
2100
2101 *device = NULL;
2102 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
0b246afa 2103 fs_info->bdev_holder, 0, &bdev, &bh);
7ba15b7d
SB		 2104 if (ret)
2105 return ret;
2106 disk_super = (struct btrfs_super_block *)bh->b_data;
2107 devid = btrfs_stack_device_id(&disk_super->dev_item);
2108 dev_uuid = disk_super->dev_item.uuid;
0b246afa 2109 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
7ba15b7d
SB		 2110 brelse(bh);
2111 if (!*device)
2112 ret = -ENOENT;
2113 blkdev_put(bdev, FMODE_READ);
2114 return ret;
2115}
2116
2ff7e61e 2117int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
da353f6b 2118 const char *device_path,
7ba15b7d
SB		 2119 struct btrfs_device **device)
2120{
2121 *device = NULL;
2122 if (strcmp(device_path, "missing") == 0) {
2123 struct list_head *devices;
2124 struct btrfs_device *tmp;
2125
0b246afa 2126 devices = &fs_info->fs_devices->devices;
7ba15b7d
SB		 2127 /*
2128 * It is safe to read the devices since the volume_mutex
2129 * is held by the caller.
2130 */
2131 list_for_each_entry(tmp, devices, dev_list) {
2132 if (tmp->in_fs_metadata && !tmp->bdev) {
2133 *device = tmp;
2134 break;
2135 }
2136 }
2137
d74a6259
AJ		 2138 if (!*device)
2139 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
7ba15b7d
SB		 2140
2141 return 0;
2142 } else {
2ff7e61e 2143 return btrfs_find_device_by_path(fs_info, device_path, device);
7ba15b7d
SB		 2144 }
2145}
2146
5c5c0df0
DS		 2147/*
2148 * Lookup a device given by device id, or the path if the id is 0.
2149 */
2ff7e61e 2150int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
da353f6b
DS		 2151 const char *devpath,
2152 struct btrfs_device **device)
24e0474b
AJ		 2153{
2154 int ret;
2155
5c5c0df0 2156 if (devid) {
24e0474b 2157 ret = 0;
0b246afa 2158 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
24e0474b
AJ		 2159 if (!*device)
2160 ret = -ENOENT;
2161 } else {
5c5c0df0 2162 if (!devpath || !devpath[0])
b3d1b153
AJ		 2163 return -EINVAL;
2164
2ff7e61e 2165 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
24e0474b
AJ		 2166 device);
2167 }
2168 return ret;
2169}
2170
2b82032c
YZ		 2171/*
2172 * does all the dirty work required for changing file system's UUID.
2173 */
2ff7e61e 2174static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2b82032c 2175{
0b246afa 2176 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2b82032c 2177 struct btrfs_fs_devices *old_devices;
e4404d6e 2178 struct btrfs_fs_devices *seed_devices;
0b246afa 2179 struct btrfs_super_block *disk_super = fs_info->super_copy;
2b82032c
YZ		 2180 struct btrfs_device *device;
2181 u64 super_flags;
2182
2183 BUG_ON(!mutex_is_locked(&uuid_mutex));
e4404d6e 2184 if (!fs_devices->seeding)
2b82032c
YZ		 2185 return -EINVAL;
2186
2dfeca9b 2187 seed_devices = alloc_fs_devices(NULL);
2208a378
ID		 2188 if (IS_ERR(seed_devices))
2189 return PTR_ERR(seed_devices);
2b82032c 2190
e4404d6e
YZ		 2191 old_devices = clone_fs_devices(fs_devices);
2192 if (IS_ERR(old_devices)) {
2193 kfree(seed_devices);
2194 return PTR_ERR(old_devices);
2b82032c 2195 }
e4404d6e 2196
2b82032c
YZ		 2197 list_add(&old_devices->list, &fs_uuids);
2198
e4404d6e
YZ		 2199 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2200 seed_devices->opened = 1;
2201 INIT_LIST_HEAD(&seed_devices->devices);
2202 INIT_LIST_HEAD(&seed_devices->alloc_list);
e5e9a520 2203 mutex_init(&seed_devices->device_list_mutex);
c9513edb 2204
0b246afa 2205 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1f78160c
XG		 2206 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2207 synchronize_rcu);
2196d6e8
MX		 2208 list_for_each_entry(device, &seed_devices->devices, dev_list)
2209 device->fs_devices = seed_devices;
c9513edb 2210
34441361 2211 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2212 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
34441361 2213 mutex_unlock(&fs_info->chunk_mutex);
e4404d6e 2214
2b82032c
YZ		 2215 fs_devices->seeding = 0;
2216 fs_devices->num_devices = 0;
2217 fs_devices->open_devices = 0;
69611ac8 2218 fs_devices->missing_devices = 0;
69611ac8 2219 fs_devices->rotating = 0;
e4404d6e 2220 fs_devices->seed = seed_devices;
2b82032c
YZ		 2221
2222 generate_random_uuid(fs_devices->fsid);
0b246afa 2223 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2b82032c 2224 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
0b246afa 2225 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
f7171750 2226
2b82032c
YZ		 2227 super_flags = btrfs_super_flags(disk_super) &
2228 ~BTRFS_SUPER_FLAG_SEEDING;
2229 btrfs_set_super_flags(disk_super, super_flags);
2230
2231 return 0;
2232}
2233
2234/*
01327610 2235 * Store the expected generation for seed devices in device items.
2b82032c
YZ		 2236 */
2237static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
5b4aacef 2238 struct btrfs_fs_info *fs_info)
2b82032c 2239{
5b4aacef 2240 struct btrfs_root *root = fs_info->chunk_root;
2b82032c
YZ		 2241 struct btrfs_path *path;
2242 struct extent_buffer *leaf;
2243 struct btrfs_dev_item *dev_item;
2244 struct btrfs_device *device;
2245 struct btrfs_key key;
44880fdc 2246 u8 fs_uuid[BTRFS_FSID_SIZE];
2b82032c
YZ		 2247 u8 dev_uuid[BTRFS_UUID_SIZE];
2248 u64 devid;
2249 int ret;
2250
2251 path = btrfs_alloc_path();
2252 if (!path)
2253 return -ENOMEM;
2254
2b82032c
YZ		 2255 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2256 key.offset = 0;
2257 key.type = BTRFS_DEV_ITEM_KEY;
2258
2259 while (1) {
2260 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2261 if (ret < 0)
2262 goto error;
2263
2264 leaf = path->nodes[0];
2265next_slot:
2266 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2267 ret = btrfs_next_leaf(root, path);
2268 if (ret > 0)
2269 break;
2270 if (ret < 0)
2271 goto error;
2272 leaf = path->nodes[0];
2273 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
b3b4aa74 2274 btrfs_release_path(path);
2b82032c
YZ		 2275 continue;
2276 }
2277
2278 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2279 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2280 key.type != BTRFS_DEV_ITEM_KEY)
2281 break;
2282
2283 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2284 struct btrfs_dev_item);
2285 devid = btrfs_device_id(leaf, dev_item);
410ba3a2 2286 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2b82032c 2287 BTRFS_UUID_SIZE);
1473b24e 2288 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
44880fdc 2289 BTRFS_FSID_SIZE);
0b246afa 2290 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
79787eaa 2291 BUG_ON(!device); /* Logic error */
2b82032c
YZ		 2292
2293 if (device->fs_devices->seeding) {
2294 btrfs_set_device_generation(leaf, dev_item,
2295 device->generation);
2296 btrfs_mark_buffer_dirty(leaf);
2297 }
2298
2299 path->slots[0]++;
2300 goto next_slot;
2301 }
2302 ret = 0;
2303error:
2304 btrfs_free_path(path);
2305 return ret;
2306}
2307
da353f6b 2308int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
788f20eb 2309{
5112febb 2310 struct btrfs_root *root = fs_info->dev_root;
d5e2003c 2311 struct request_queue *q;
788f20eb
CM		 2312 struct btrfs_trans_handle *trans;
2313 struct btrfs_device *device;
2314 struct block_device *bdev;
788f20eb 2315 struct list_head *devices;
0b246afa 2316 struct super_block *sb = fs_info->sb;
606686ee 2317 struct rcu_string *name;
3c1dbdf5 2318 u64 tmp;
2b82032c 2319 int seeding_dev = 0;
788f20eb
CM		 2320 int ret = 0;
2321
bc98a42c 2322 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
f8c5d0b4 2323 return -EROFS;
788f20eb 2324
a5d16333 2325 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
0b246afa 2326 fs_info->bdev_holder);
7f59203a
JB		 2327 if (IS_ERR(bdev))
2328 return PTR_ERR(bdev);
a2135011 2329
0b246afa 2330 if (fs_info->fs_devices->seeding) {
2b82032c
YZ		 2331 seeding_dev = 1;
2332 down_write(&sb->s_umount);
2333 mutex_lock(&uuid_mutex);
2334 }
2335
8c8bee1d 2336 filemap_write_and_wait(bdev->bd_inode->i_mapping);
a2135011 2337
0b246afa 2338 devices = &fs_info->fs_devices->devices;
d25628bd 2339
0b246afa 2340 mutex_lock(&fs_info->fs_devices->device_list_mutex);
c6e30871 2341 list_for_each_entry(device, devices, dev_list) {
788f20eb
CM		 2342 if (device->bdev == bdev) {
2343 ret = -EEXIST;
d25628bd 2344 mutex_unlock(
0b246afa 2345 &fs_info->fs_devices->device_list_mutex);
2b82032c 2346 goto error;
788f20eb
CM		 2347 }
2348 }
0b246afa 2349 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
788f20eb 2350
0b246afa 2351 device = btrfs_alloc_device(fs_info, NULL, NULL);
12bd2fc0 2352 if (IS_ERR(device)) {
788f20eb 2353 /* we can safely leave the fs_devices entry around */
12bd2fc0 2354 ret = PTR_ERR(device);
2b82032c 2355 goto error;
788f20eb
CM		 2356 }
2357
78f2c9e6 2358 name = rcu_string_strdup(device_path, GFP_KERNEL);
606686ee 2359 if (!name) {
788f20eb 2360 kfree(device);
2b82032c
YZ		 2361 ret = -ENOMEM;
2362 goto error;
788f20eb 2363 }
606686ee 2364 rcu_assign_pointer(device->name, name);
2b82032c 2365
a22285a6 2366 trans = btrfs_start_transaction(root, 0);
98d5dc13 2367 if (IS_ERR(trans)) {
606686ee 2368 rcu_string_free(device->name);
98d5dc13
TI		 2369 kfree(device);
2370 ret = PTR_ERR(trans);
2371 goto error;
2372 }
2373
d5e2003c
JB		 2374 q = bdev_get_queue(bdev);
2375 if (blk_queue_discard(q))
2376 device->can_discard = 1;
2b82032c 2377 device->writeable = 1;
2b82032c 2378 device->generation = trans->transid;
0b246afa
JM		 2379 device->io_width = fs_info->sectorsize;
2380 device->io_align = fs_info->sectorsize;
2381 device->sector_size = fs_info->sectorsize;
7dfb8be1
NB		 2382 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2383 fs_info->sectorsize);
2cc3c559 2384 device->disk_total_bytes = device->total_bytes;
935e5cc9 2385 device->commit_total_bytes = device->total_bytes;
fb456252 2386 device->fs_info = fs_info;
788f20eb 2387 device->bdev = bdev;
dfe25020 2388 device->in_fs_metadata = 1;
63a212ab 2389 device->is_tgtdev_for_dev_replace = 0;
fb01aa85 2390 device->mode = FMODE_EXCL;
27087f37 2391 device->dev_stats_valid = 1;
9f6d2510 2392 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
788f20eb 2393
2b82032c
YZ		 2394 if (seeding_dev) {
2395 sb->s_flags &= ~MS_RDONLY;
2ff7e61e 2396 ret = btrfs_prepare_sprout(fs_info);
79787eaa 2397 BUG_ON(ret); /* -ENOMEM */
2b82032c 2398 }
788f20eb 2399
0b246afa 2400 device->fs_devices = fs_info->fs_devices;
e5e9a520 2401
0b246afa 2402 mutex_lock(&fs_info->fs_devices->device_list_mutex);
34441361 2403 mutex_lock(&fs_info->chunk_mutex);
0b246afa 2404 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2b82032c 2405 list_add(&device->dev_alloc_list,
0b246afa
JM		 2406 &fs_info->fs_devices->alloc_list);
2407 fs_info->fs_devices->num_devices++;
2408 fs_info->fs_devices->open_devices++;
2409 fs_info->fs_devices->rw_devices++;
2410 fs_info->fs_devices->total_devices++;
2411 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
325cd4ba 2412
a5ed45f8 2413 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2bf64758 2414
e884f4f0 2415 if (!blk_queue_nonrot(q))
0b246afa 2416 fs_info->fs_devices->rotating = 1;
c289811c 2417
0b246afa
JM		 2418 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2419 btrfs_set_super_total_bytes(fs_info->super_copy,
7dfb8be1 2420 round_down(tmp + device->total_bytes, fs_info->sectorsize));
788f20eb 2421
0b246afa
JM		 2422 tmp = btrfs_super_num_devices(fs_info->super_copy);
2423 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
0d39376a
AJ		 2424
2425 /* add sysfs device entry */
0b246afa 2426 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
0d39376a 2427
2196d6e8
MX		 2428 /*
2429 * we've got more storage, clear any full flags on the space
2430 * infos
2431 */
0b246afa 2432 btrfs_clear_space_info_full(fs_info);
2196d6e8 2433
34441361 2434 mutex_unlock(&fs_info->chunk_mutex);
0b246afa 2435 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
788f20eb 2436
2b82032c 2437 if (seeding_dev) {
34441361 2438 mutex_lock(&fs_info->chunk_mutex);
e4a4dce7 2439 ret = init_first_rw_device(trans, fs_info);
34441361 2440 mutex_unlock(&fs_info->chunk_mutex);
005d6427 2441 if (ret) {
66642832 2442 btrfs_abort_transaction(trans, ret);
79787eaa 2443 goto error_trans;
005d6427 2444 }
2196d6e8
MX		 2445 }
2446
0b246afa 2447 ret = btrfs_add_device(trans, fs_info, device);
2196d6e8 2448 if (ret) {
66642832 2449 btrfs_abort_transaction(trans, ret);
2196d6e8
MX		 2450 goto error_trans;
2451 }
2452
2453 if (seeding_dev) {
2454 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2455
0b246afa 2456 ret = btrfs_finish_sprout(trans, fs_info);
005d6427 2457 if (ret) {
66642832 2458 btrfs_abort_transaction(trans, ret);
79787eaa 2459 goto error_trans;
005d6427 2460 }
b2373f25
AJ		 2461
2462 /* Sprouting would change fsid of the mounted root,
2463 * so rename the fsid on the sysfs
2464 */
2465 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
0b246afa
JM		 2466 fs_info->fsid);
2467 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2468 btrfs_warn(fs_info,
2469 "sysfs: failed to create fsid for sprout");
2b82032c
YZ		 2470 }
2471
3a45bb20 2472 ret = btrfs_commit_transaction(trans);
a2135011 2473
2b82032c
YZ		 2474 if (seeding_dev) {
2475 mutex_unlock(&uuid_mutex);
2476 up_write(&sb->s_umount);
788f20eb 2477
79787eaa
JM		 2478 if (ret) /* transaction commit */
2479 return ret;
2480
2ff7e61e 2481 ret = btrfs_relocate_sys_chunks(fs_info);
79787eaa 2482 if (ret < 0)
0b246afa 2483 btrfs_handle_fs_error(fs_info, ret,
5d163e0e 2484 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
671415b7
MX		 2485 trans = btrfs_attach_transaction(root);
2486 if (IS_ERR(trans)) {
2487 if (PTR_ERR(trans) == -ENOENT)
2488 return 0;
2489 return PTR_ERR(trans);
2490 }
3a45bb20 2491 ret = btrfs_commit_transaction(trans);
2b82032c 2492 }
c9e9f97b 2493
5a1972bd
QW		 2494 /* Update ctime/mtime for libblkid */
2495 update_dev_time(device_path);
2b82032c 2496 return ret;
79787eaa
JM		 2497
2498error_trans:
3a45bb20 2499 btrfs_end_transaction(trans);
606686ee 2500 rcu_string_free(device->name);
0b246afa 2501 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
79787eaa 2502 kfree(device);
2b82032c 2503error:
e525fd89 2504 blkdev_put(bdev, FMODE_EXCL);
2b82032c
YZ		 2505 if (seeding_dev) {
2506 mutex_unlock(&uuid_mutex);
2507 up_write(&sb->s_umount);
2508 }
c9e9f97b 2509 return ret;
788f20eb
CM		 2510}
2511
2ff7e61e 2512int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
da353f6b 2513 const char *device_path,
1c43366d 2514 struct btrfs_device *srcdev,
e93c89c1
SB		 2515 struct btrfs_device **device_out)
2516{
2517 struct request_queue *q;
2518 struct btrfs_device *device;
2519 struct block_device *bdev;
e93c89c1
SB		 2520 struct list_head *devices;
2521 struct rcu_string *name;
12bd2fc0 2522 u64 devid = BTRFS_DEV_REPLACE_DEVID;
e93c89c1
SB		 2523 int ret = 0;
2524
2525 *device_out = NULL;
1c43366d
MX		 2526 if (fs_info->fs_devices->seeding) {
2527 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
e93c89c1 2528 return -EINVAL;
1c43366d 2529 }
e93c89c1
SB		 2530
2531 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2532 fs_info->bdev_holder);
1c43366d
MX		 2533 if (IS_ERR(bdev)) {
2534 btrfs_err(fs_info, "target device %s is invalid!", device_path);
e93c89c1 2535 return PTR_ERR(bdev);
1c43366d 2536 }
e93c89c1
SB		 2537
2538 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2539
2540 devices = &fs_info->fs_devices->devices;
2541 list_for_each_entry(device, devices, dev_list) {
2542 if (device->bdev == bdev) {
5d163e0e
JM		 2543 btrfs_err(fs_info,
2544 "target device is in the filesystem!");
e93c89c1
SB		 2545 ret = -EEXIST;
2546 goto error;
2547 }
2548 }
2549
1c43366d 2550
7cc8e58d
MX		 2551 if (i_size_read(bdev->bd_inode) <
2552 btrfs_device_get_total_bytes(srcdev)) {
5d163e0e
JM		 2553 btrfs_err(fs_info,
2554 "target device is smaller than source device!");
1c43366d
MX		 2555 ret = -EINVAL;
2556 goto error;
2557 }
2558
2559
12bd2fc0
ID		 2560 device = btrfs_alloc_device(NULL, &devid, NULL);
2561 if (IS_ERR(device)) {
2562 ret = PTR_ERR(device);
e93c89c1
SB		 2563 goto error;
2564 }
2565
6165572c 2566 name = rcu_string_strdup(device_path, GFP_KERNEL);
e93c89c1
SB		 2567 if (!name) {
2568 kfree(device);
2569 ret = -ENOMEM;
2570 goto error;
2571 }
2572 rcu_assign_pointer(device->name, name);
2573
2574 q = bdev_get_queue(bdev);
2575 if (blk_queue_discard(q))
2576 device->can_discard = 1;
0b246afa 2577 mutex_lock(&fs_info->fs_devices->device_list_mutex);
e93c89c1 2578 device->writeable = 1;
e93c89c1 2579 device->generation = 0;
0b246afa
JM		 2580 device->io_width = fs_info->sectorsize;
2581 device->io_align = fs_info->sectorsize;
2582 device->sector_size = fs_info->sectorsize;
7cc8e58d
MX		 2583 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2584 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2585 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
935e5cc9
MX		 2586 ASSERT(list_empty(&srcdev->resized_list));
2587 device->commit_total_bytes = srcdev->commit_total_bytes;
ce7213c7 2588 device->commit_bytes_used = device->bytes_used;
fb456252 2589 device->fs_info = fs_info;
e93c89c1
SB		 2590 device->bdev = bdev;
2591 device->in_fs_metadata = 1;
2592 device->is_tgtdev_for_dev_replace = 1;
2593 device->mode = FMODE_EXCL;
27087f37 2594 device->dev_stats_valid = 1;
9f6d2510 2595 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
e93c89c1
SB		 2596 device->fs_devices = fs_info->fs_devices;
2597 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2598 fs_info->fs_devices->num_devices++;
2599 fs_info->fs_devices->open_devices++;
0b246afa 2600 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
e93c89c1
SB		 2601
2602 *device_out = device;
2603 return ret;
2604
2605error:
2606 blkdev_put(bdev, FMODE_EXCL);
2607 return ret;
2608}
2609
2610void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2611 struct btrfs_device *tgtdev)
2612{
da17066c
JM		 2613 u32 sectorsize = fs_info->sectorsize;
2614
e93c89c1 2615 WARN_ON(fs_info->fs_devices->rw_devices == 0);
da17066c
JM		 2616 tgtdev->io_width = sectorsize;
2617 tgtdev->io_align = sectorsize;
2618 tgtdev->sector_size = sectorsize;
fb456252 2619 tgtdev->fs_info = fs_info;
e93c89c1
SB		 2620 tgtdev->in_fs_metadata = 1;
2621}
2622
d397712b
CM		 2623static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2624 struct btrfs_device *device)
0b86a832
CM		 2625{
2626 int ret;
2627 struct btrfs_path *path;
0b246afa 2628 struct btrfs_root *root = device->fs_info->chunk_root;
0b86a832
CM		 2629 struct btrfs_dev_item *dev_item;
2630 struct extent_buffer *leaf;
2631 struct btrfs_key key;
2632
0b86a832
CM		 2633 path = btrfs_alloc_path();
2634 if (!path)
2635 return -ENOMEM;
2636
2637 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2638 key.type = BTRFS_DEV_ITEM_KEY;
2639 key.offset = device->devid;
2640
2641 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2642 if (ret < 0)
2643 goto out;
2644
2645 if (ret > 0) {
2646 ret = -ENOENT;
2647 goto out;
2648 }
2649
2650 leaf = path->nodes[0];
2651 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2652
2653 btrfs_set_device_id(leaf, dev_item, device->devid);
2654 btrfs_set_device_type(leaf, dev_item, device->type);
2655 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2656 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2657 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
7cc8e58d
MX		 2658 btrfs_set_device_total_bytes(leaf, dev_item,
2659 btrfs_device_get_disk_total_bytes(device));
2660 btrfs_set_device_bytes_used(leaf, dev_item,
2661 btrfs_device_get_bytes_used(device));
0b86a832
CM		 2662 btrfs_mark_buffer_dirty(leaf);
2663
2664out:
2665 btrfs_free_path(path);
2666 return ret;
2667}
2668
2196d6e8 2669int btrfs_grow_device(struct btrfs_trans_handle *trans,
8f18cf13
CM		 2670 struct btrfs_device *device, u64 new_size)
2671{
0b246afa
JM		 2672 struct btrfs_fs_info *fs_info = device->fs_info;
2673 struct btrfs_super_block *super_copy = fs_info->super_copy;
935e5cc9 2674 struct btrfs_fs_devices *fs_devices;
2196d6e8
MX		 2675 u64 old_total;
2676 u64 diff;
8f18cf13 2677
2b82032c
YZ		 2678 if (!device->writeable)
2679 return -EACCES;
2196d6e8 2680
7dfb8be1
NB		 2681 new_size = round_down(new_size, fs_info->sectorsize);
2682
34441361 2683 mutex_lock(&fs_info->chunk_mutex);
2196d6e8 2684 old_total = btrfs_super_total_bytes(super_copy);
0e4324a4 2685 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2196d6e8 2686
63a212ab 2687 if (new_size <= device->total_bytes ||
2196d6e8 2688 device->is_tgtdev_for_dev_replace) {
34441361 2689 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 2690 return -EINVAL;
2196d6e8 2691 }
2b82032c 2692
0b246afa 2693 fs_devices = fs_info->fs_devices;
2b82032c 2694
7dfb8be1
NB		 2695 btrfs_set_super_total_bytes(super_copy,
2696 round_down(old_total + diff, fs_info->sectorsize));
2b82032c
YZ		 2697 device->fs_devices->total_rw_bytes += diff;
2698
7cc8e58d
MX		 2699 btrfs_device_set_total_bytes(device, new_size);
2700 btrfs_device_set_disk_total_bytes(device, new_size);
fb456252 2701 btrfs_clear_space_info_full(device->fs_info);
935e5cc9
MX		 2702 if (list_empty(&device->resized_list))
2703 list_add_tail(&device->resized_list,
2704 &fs_devices->resized_devices);
34441361 2705 mutex_unlock(&fs_info->chunk_mutex);
4184ea7f 2706
8f18cf13
CM		 2707 return btrfs_update_device(trans, device);
2708}
2709
2710static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
408fbf19 2711 struct btrfs_fs_info *fs_info, u64 chunk_offset)
8f18cf13 2712{
5b4aacef 2713 struct btrfs_root *root = fs_info->chunk_root;
8f18cf13
CM		 2714 int ret;
2715 struct btrfs_path *path;
2716 struct btrfs_key key;
2717
8f18cf13
CM		 2718 path = btrfs_alloc_path();
2719 if (!path)
2720 return -ENOMEM;
2721
408fbf19 2722 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
8f18cf13
CM		 2723 key.offset = chunk_offset;
2724 key.type = BTRFS_CHUNK_ITEM_KEY;
2725
2726 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
79787eaa
JM		 2727 if (ret < 0)
2728 goto out;
2729 else if (ret > 0) { /* Logic error or corruption */
0b246afa
JM		 2730 btrfs_handle_fs_error(fs_info, -ENOENT,
2731 "Failed lookup while freeing chunk.");
79787eaa
JM		 2732 ret = -ENOENT;
2733 goto out;
2734 }
8f18cf13
CM		 2735
2736 ret = btrfs_del_item(trans, root, path);
79787eaa 2737 if (ret < 0)
0b246afa
JM		 2738 btrfs_handle_fs_error(fs_info, ret,
2739 "Failed to delete chunk item.");
79787eaa 2740out:
8f18cf13 2741 btrfs_free_path(path);
65a246c5 2742 return ret;
8f18cf13
CM		 2743}
2744
408fbf19 2745static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
8f18cf13 2746{
0b246afa 2747 struct btrfs_super_block *super_copy = fs_info->super_copy;
8f18cf13
CM		 2748 struct btrfs_disk_key *disk_key;
2749 struct btrfs_chunk *chunk;
2750 u8 *ptr;
2751 int ret = 0;
2752 u32 num_stripes;
2753 u32 array_size;
2754 u32 len = 0;
2755 u32 cur;
2756 struct btrfs_key key;
2757
34441361 2758 mutex_lock(&fs_info->chunk_mutex);
8f18cf13
CM		 2759 array_size = btrfs_super_sys_array_size(super_copy);
2760
2761 ptr = super_copy->sys_chunk_array;
2762 cur = 0;
2763
2764 while (cur < array_size) {
2765 disk_key = (struct btrfs_disk_key *)ptr;
2766 btrfs_disk_key_to_cpu(&key, disk_key);
2767
2768 len = sizeof(*disk_key);
2769
2770 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2771 chunk = (struct btrfs_chunk *)(ptr + len);
2772 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2773 len += btrfs_chunk_item_size(num_stripes);
2774 } else {
2775 ret = -EIO;
2776 break;
2777 }
408fbf19 2778 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
8f18cf13
CM		 2779 key.offset == chunk_offset) {
2780 memmove(ptr, ptr + len, array_size - (cur + len));
2781 array_size -= len;
2782 btrfs_set_super_sys_array_size(super_copy, array_size);
2783 } else {
2784 ptr += len;
2785 cur += len;
2786 }
2787 }
34441361 2788 mutex_unlock(&fs_info->chunk_mutex);
8f18cf13
CM		 2789 return ret;
2790}
2791
592d92ee
LB		 2792static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2793 u64 logical, u64 length)
2794{
2795 struct extent_map_tree *em_tree;
2796 struct extent_map *em;
2797
2798 em_tree = &fs_info->mapping_tree.map_tree;
2799 read_lock(&em_tree->lock);
2800 em = lookup_extent_mapping(em_tree, logical, length);
2801 read_unlock(&em_tree->lock);
2802
2803 if (!em) {
2804 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2805 logical, length);
2806 return ERR_PTR(-EINVAL);
2807 }
2808
2809 if (em->start > logical || em->start + em->len < logical) {
2810 btrfs_crit(fs_info,
2811 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2812 logical, length, em->start, em->start + em->len);
2813 free_extent_map(em);
2814 return ERR_PTR(-EINVAL);
2815 }
2816
2817 /* callers are responsible for dropping em's ref. */
2818 return em;
2819}
2820
47ab2a6c 2821int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
5b4aacef 2822 struct btrfs_fs_info *fs_info, u64 chunk_offset)
8f18cf13 2823{
8f18cf13
CM		 2824 struct extent_map *em;
2825 struct map_lookup *map;
2196d6e8 2826 u64 dev_extent_len = 0;
47ab2a6c 2827 int i, ret = 0;
0b246afa 2828 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
8f18cf13 2829
592d92ee
LB		 2830 em = get_chunk_map(fs_info, chunk_offset, 1);
2831 if (IS_ERR(em)) {
47ab2a6c
JB		 2832 /*
2833 * This is a logic error, but we don't want to just rely on the
bb7ab3b9 2834 * user having built with ASSERT enabled, so if ASSERT doesn't
47ab2a6c
JB		 2835 * do anything we still error out.
2836 */
2837 ASSERT(0);
592d92ee 2838 return PTR_ERR(em);
47ab2a6c 2839 }
95617d69 2840 map = em->map_lookup;
34441361 2841 mutex_lock(&fs_info->chunk_mutex);
2ff7e61e 2842 check_system_chunk(trans, fs_info, map->type);
34441361 2843 mutex_unlock(&fs_info->chunk_mutex);
8f18cf13 2844
57ba4cb8
FM		 2845 /*
2846 * Take the device list mutex to prevent races with the final phase of
2847 * a device replace operation that replaces the device object associated
2848 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2849 */
2850 mutex_lock(&fs_devices->device_list_mutex);
8f18cf13 2851 for (i = 0; i < map->num_stripes; i++) {
47ab2a6c 2852 struct btrfs_device *device = map->stripes[i].dev;
2196d6e8
MX		 2853 ret = btrfs_free_dev_extent(trans, device,
2854 map->stripes[i].physical,
2855 &dev_extent_len);
47ab2a6c 2856 if (ret) {
57ba4cb8 2857 mutex_unlock(&fs_devices->device_list_mutex);
66642832 2858 btrfs_abort_transaction(trans, ret);
47ab2a6c
JB		 2859 goto out;
2860 }
a061fc8d 2861
2196d6e8 2862 if (device->bytes_used > 0) {
34441361 2863 mutex_lock(&fs_info->chunk_mutex);
2196d6e8
MX		 2864 btrfs_device_set_bytes_used(device,
2865 device->bytes_used - dev_extent_len);
a5ed45f8 2866 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
0b246afa 2867 btrfs_clear_space_info_full(fs_info);
34441361 2868 mutex_unlock(&fs_info->chunk_mutex);
2196d6e8 2869 }
a061fc8d 2870
dfe25020
CM		 2871 if (map->stripes[i].dev) {
2872 ret = btrfs_update_device(trans, map->stripes[i].dev);
47ab2a6c 2873 if (ret) {
57ba4cb8 2874 mutex_unlock(&fs_devices->device_list_mutex);
66642832 2875 btrfs_abort_transaction(trans, ret);
47ab2a6c
JB		 2876 goto out;
2877 }
dfe25020 2878 }
8f18cf13 2879 }
57ba4cb8
FM		 2880 mutex_unlock(&fs_devices->device_list_mutex);
2881
408fbf19 2882 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
47ab2a6c 2883 if (ret) {
66642832 2884 btrfs_abort_transaction(trans, ret);
47ab2a6c
JB		 2885 goto out;
2886 }
8f18cf13 2887
6bccf3ab 2888 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
1abe9b8a 2889
8f18cf13 2890 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
408fbf19 2891 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
47ab2a6c 2892 if (ret) {
66642832 2893 btrfs_abort_transaction(trans, ret);
47ab2a6c
JB		 2894 goto out;
2895 }
8f18cf13
CM		 2896 }
2897
6bccf3ab 2898 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
47ab2a6c 2899 if (ret) {
66642832 2900 btrfs_abort_transaction(trans, ret);
47ab2a6c
JB		 2901 goto out;
2902 }
2b82032c 2903
47ab2a6c 2904out:
2b82032c
YZ		 2905 /* once for us */
2906 free_extent_map(em);
47ab2a6c
JB		 2907 return ret;
2908}
2b82032c 2909
5b4aacef 2910static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
47ab2a6c 2911{
5b4aacef 2912 struct btrfs_root *root = fs_info->chunk_root;
19c4d2f9 2913 struct btrfs_trans_handle *trans;
47ab2a6c 2914 int ret;
2b82032c 2915
67c5e7d4
FM		 2916 /*
2917 * Prevent races with automatic removal of unused block groups.
2918 * After we relocate and before we remove the chunk with offset
2919 * chunk_offset, automatic removal of the block group can kick in,
2920 * resulting in a failure when calling btrfs_remove_chunk() below.
2921 *
2922 * Make sure to acquire this mutex before doing a tree search (dev
2923 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2924 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2925 * we release the path used to search the chunk/dev tree and before
2926 * the current task acquires this mutex and calls us.
2927 */
0b246afa 2928 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
67c5e7d4 2929
0b246afa 2930 ret = btrfs_can_relocate(fs_info, chunk_offset);
47ab2a6c
JB		 2931 if (ret)
2932 return -ENOSPC;
2933
2934 /* step one, relocate all the extents inside this chunk */
2ff7e61e 2935 btrfs_scrub_pause(fs_info);
0b246afa 2936 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2ff7e61e 2937 btrfs_scrub_continue(fs_info);
47ab2a6c
JB		 2938 if (ret)
2939 return ret;
2940
19c4d2f9
CM		 2941 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2942 chunk_offset);
2943 if (IS_ERR(trans)) {
2944 ret = PTR_ERR(trans);
2945 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2946 return ret;
2947 }
2948
47ab2a6c 2949 /*
19c4d2f9
CM		 2950 * step two, delete the device extents and the
2951 * chunk tree entries
47ab2a6c 2952 */
5b4aacef 2953 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
3a45bb20 2954 btrfs_end_transaction(trans);
19c4d2f9 2955 return ret;
2b82032c
YZ		 2956}
2957
2ff7e61e 2958static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2b82032c 2959{
0b246afa 2960 struct btrfs_root *chunk_root = fs_info->chunk_root;
2b82032c
YZ		 2961 struct btrfs_path *path;
2962 struct extent_buffer *leaf;
2963 struct btrfs_chunk *chunk;
2964 struct btrfs_key key;
2965 struct btrfs_key found_key;
2b82032c 2966 u64 chunk_type;
ba1bf481
JB		 2967 bool retried = false;
2968 int failed = 0;
2b82032c
YZ		 2969 int ret;
2970
2971 path = btrfs_alloc_path();
2972 if (!path)
2973 return -ENOMEM;
2974
ba1bf481 2975again:
2b82032c
YZ		 2976 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2977 key.offset = (u64)-1;
2978 key.type = BTRFS_CHUNK_ITEM_KEY;
2979
2980 while (1) {
0b246afa 2981 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2b82032c 2982 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
67c5e7d4 2983 if (ret < 0) {
0b246afa 2984 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2b82032c 2985 goto error;
67c5e7d4 2986 }
79787eaa 2987 BUG_ON(ret == 0); /* Corruption */
2b82032c
YZ		 2988
2989 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2990 key.type);
67c5e7d4 2991 if (ret)
0b246afa 2992 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2b82032c
YZ		 2993 if (ret < 0)
2994 goto error;
2995 if (ret > 0)
2996 break;
1a40e23b 2997
2b82032c
YZ		 2998 leaf = path->nodes[0];
2999 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b 3000
2b82032c
YZ		 3001 chunk = btrfs_item_ptr(leaf, path->slots[0],
3002 struct btrfs_chunk);
3003 chunk_type = btrfs_chunk_type(leaf, chunk);
b3b4aa74 3004 btrfs_release_path(path);
8f18cf13 3005
2b82032c 3006 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
0b246afa 3007 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
ba1bf481
JB		 3008 if (ret == -ENOSPC)
3009 failed++;
14586651
HS		 3010 else
3011 BUG_ON(ret);
2b82032c 3012 }
0b246afa 3013 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
8f18cf13 3014
2b82032c
YZ		 3015 if (found_key.offset == 0)
3016 break;
3017 key.offset = found_key.offset - 1;
3018 }
3019 ret = 0;
ba1bf481
JB		 3020 if (failed && !retried) {
3021 failed = 0;
3022 retried = true;
3023 goto again;
fae7f21c 3024 } else if (WARN_ON(failed && retried)) {
ba1bf481
JB		 3025 ret = -ENOSPC;
3026 }
2b82032c
YZ		 3027error:
3028 btrfs_free_path(path);
3029 return ret;
8f18cf13
CM		 3030}
3031
6bccf3ab 3032static int insert_balance_item(struct btrfs_fs_info *fs_info,
0940ebf6
ID		 3033 struct btrfs_balance_control *bctl)
3034{
6bccf3ab 3035 struct btrfs_root *root = fs_info->tree_root;
0940ebf6
ID		 3036 struct btrfs_trans_handle *trans;
3037 struct btrfs_balance_item *item;
3038 struct btrfs_disk_balance_args disk_bargs;
3039 struct btrfs_path *path;
3040 struct extent_buffer *leaf;
3041 struct btrfs_key key;
3042 int ret, err;
3043
3044 path = btrfs_alloc_path();
3045 if (!path)
3046 return -ENOMEM;
3047
3048 trans = btrfs_start_transaction(root, 0);
3049 if (IS_ERR(trans)) {
3050 btrfs_free_path(path);
3051 return PTR_ERR(trans);
3052 }
3053
3054 key.objectid = BTRFS_BALANCE_OBJECTID;
c479cb4f 3055 key.type = BTRFS_TEMPORARY_ITEM_KEY;
0940ebf6
ID		 3056 key.offset = 0;
3057
3058 ret = btrfs_insert_empty_item(trans, root, path, &key,
3059 sizeof(*item));
3060 if (ret)
3061 goto out;
3062
3063 leaf = path->nodes[0];
3064 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3065
b159fa28 3066 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
0940ebf6
ID		 3067
3068 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3069 btrfs_set_balance_data(leaf, item, &disk_bargs);
3070 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3071 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3072 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3073 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3074
3075 btrfs_set_balance_flags(leaf, item, bctl->flags);
3076
3077 btrfs_mark_buffer_dirty(leaf);
3078out:
3079 btrfs_free_path(path);
3a45bb20 3080 err = btrfs_commit_transaction(trans);
0940ebf6
ID		 3081 if (err && !ret)
3082 ret = err;
3083 return ret;
3084}
3085
6bccf3ab 3086static int del_balance_item(struct btrfs_fs_info *fs_info)
0940ebf6 3087{
6bccf3ab 3088 struct btrfs_root *root = fs_info->tree_root;
0940ebf6
ID		 3089 struct btrfs_trans_handle *trans;
3090 struct btrfs_path *path;
3091 struct btrfs_key key;
3092 int ret, err;
3093
3094 path = btrfs_alloc_path();
3095 if (!path)
3096 return -ENOMEM;
3097
3098 trans = btrfs_start_transaction(root, 0);
3099 if (IS_ERR(trans)) {
3100 btrfs_free_path(path);
3101 return PTR_ERR(trans);
3102 }
3103
3104 key.objectid = BTRFS_BALANCE_OBJECTID;
c479cb4f 3105 key.type = BTRFS_TEMPORARY_ITEM_KEY;
0940ebf6
ID		 3106 key.offset = 0;
3107
3108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3109 if (ret < 0)
3110 goto out;
3111 if (ret > 0) {
3112 ret = -ENOENT;
3113 goto out;
3114 }
3115
3116 ret = btrfs_del_item(trans, root, path);
3117out:
3118 btrfs_free_path(path);
3a45bb20 3119 err = btrfs_commit_transaction(trans);
0940ebf6
ID		 3120 if (err && !ret)
3121 ret = err;
3122 return ret;
3123}
3124
59641015
ID		 3125/*
3126 * This is a heuristic used to reduce the number of chunks balanced on
3127 * resume after balance was interrupted.
3128 */
3129static void update_balance_args(struct btrfs_balance_control *bctl)
3130{
3131 /*
3132 * Turn on soft mode for chunk types that were being converted.
3133 */
3134 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3140
3141 /*
3142 * Turn on usage filter if is not already used. The idea is
3143 * that chunks that we have already balanced should be
3144 * reasonably full. Don't do it for chunks that are being
3145 * converted - that will keep us from relocating unconverted
3146 * (albeit full) chunks.
3147 */
3148 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
bc309467 3149 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
59641015
ID		 3150 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152 bctl->data.usage = 90;
3153 }
3154 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
bc309467 3155 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
59641015
ID		 3156 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158 bctl->sys.usage = 90;
3159 }
3160 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
bc309467 3161 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
59641015
ID		 3162 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3163 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3164 bctl->meta.usage = 90;
3165 }
3166}
3167
c9e9f97b
ID		 3168/*
3169 * Should be called with both balance and volume mutexes held to
3170 * serialize other volume operations (add_dev/rm_dev/resize) with
3171 * restriper. Same goes for unset_balance_control.
3172 */
3173static void set_balance_control(struct btrfs_balance_control *bctl)
3174{
3175 struct btrfs_fs_info *fs_info = bctl->fs_info;
3176
3177 BUG_ON(fs_info->balance_ctl);
3178
3179 spin_lock(&fs_info->balance_lock);
3180 fs_info->balance_ctl = bctl;
3181 spin_unlock(&fs_info->balance_lock);
3182}
3183
3184static void unset_balance_control(struct btrfs_fs_info *fs_info)
3185{
3186 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3187
3188 BUG_ON(!fs_info->balance_ctl);
3189
3190 spin_lock(&fs_info->balance_lock);
3191 fs_info->balance_ctl = NULL;
3192 spin_unlock(&fs_info->balance_lock);
3193
3194 kfree(bctl);
3195}
3196
ed25e9b2
ID		 3197/*
3198 * Balance filters. Return 1 if chunk should be filtered out
3199 * (should not be balanced).
3200 */
899c81ea 3201static int chunk_profiles_filter(u64 chunk_type,
ed25e9b2
ID		 3202 struct btrfs_balance_args *bargs)
3203{
899c81ea
ID		 3204 chunk_type = chunk_to_extended(chunk_type) &
3205 BTRFS_EXTENDED_PROFILE_MASK;
ed25e9b2 3206
899c81ea 3207 if (bargs->profiles & chunk_type)
ed25e9b2
ID		 3208 return 0;
3209
3210 return 1;
3211}
3212
dba72cb3 3213static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
5ce5b3c0 3214 struct btrfs_balance_args *bargs)
bc309467
DS		 3215{
3216 struct btrfs_block_group_cache *cache;
3217 u64 chunk_used;
3218 u64 user_thresh_min;
3219 u64 user_thresh_max;
3220 int ret = 1;
3221
3222 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3223 chunk_used = btrfs_block_group_used(&cache->item);
3224
3225 if (bargs->usage_min == 0)
3226 user_thresh_min = 0;
3227 else
3228 user_thresh_min = div_factor_fine(cache->key.offset,
3229 bargs->usage_min);
3230
3231 if (bargs->usage_max == 0)
3232 user_thresh_max = 1;
3233 else if (bargs->usage_max > 100)
3234 user_thresh_max = cache->key.offset;
3235 else
3236 user_thresh_max = div_factor_fine(cache->key.offset,
3237 bargs->usage_max);
3238
3239 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3240 ret = 0;
3241
3242 btrfs_put_block_group(cache);
3243 return ret;
3244}
3245
dba72cb3 3246static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
bc309467 3247 u64 chunk_offset, struct btrfs_balance_args *bargs)
5ce5b3c0
ID		 3248{
3249 struct btrfs_block_group_cache *cache;
3250 u64 chunk_used, user_thresh;
3251 int ret = 1;
3252
3253 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254 chunk_used = btrfs_block_group_used(&cache->item);
3255
bc309467 3256 if (bargs->usage_min == 0)
3e39cea6 3257 user_thresh = 1;
a105bb88
ID		 3258 else if (bargs->usage > 100)
3259 user_thresh = cache->key.offset;
3260 else
3261 user_thresh = div_factor_fine(cache->key.offset,
3262 bargs->usage);
3263
5ce5b3c0
ID		 3264 if (chunk_used < user_thresh)
3265 ret = 0;
3266
3267 btrfs_put_block_group(cache);
3268 return ret;
3269}
3270
409d404b
ID		 3271static int chunk_devid_filter(struct extent_buffer *leaf,
3272 struct btrfs_chunk *chunk,
3273 struct btrfs_balance_args *bargs)
3274{
3275 struct btrfs_stripe *stripe;
3276 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3277 int i;
3278
3279 for (i = 0; i < num_stripes; i++) {
3280 stripe = btrfs_stripe_nr(chunk, i);
3281 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3282 return 0;
3283 }
3284
3285 return 1;
3286}
3287
94e60d5a
ID		 3288/* [pstart, pend) */
3289static int chunk_drange_filter(struct extent_buffer *leaf,
3290 struct btrfs_chunk *chunk,
94e60d5a
ID		 3291 struct btrfs_balance_args *bargs)
3292{
3293 struct btrfs_stripe *stripe;
3294 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3295 u64 stripe_offset;
3296 u64 stripe_length;
3297 int factor;
3298 int i;
3299
3300 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3301 return 0;
3302
3303 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
53b381b3
DW		 3304 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3305 factor = num_stripes / 2;
3306 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3307 factor = num_stripes - 1;
3308 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3309 factor = num_stripes - 2;
3310 } else {
3311 factor = num_stripes;
3312 }
94e60d5a
ID		 3313
3314 for (i = 0; i < num_stripes; i++) {
3315 stripe = btrfs_stripe_nr(chunk, i);
3316 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3317 continue;
3318
3319 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3320 stripe_length = btrfs_chunk_length(leaf, chunk);
b8b93add 3321 stripe_length = div_u64(stripe_length, factor);
94e60d5a
ID		 3322
3323 if (stripe_offset < bargs->pend &&
3324 stripe_offset + stripe_length > bargs->pstart)
3325 return 0;
3326 }
3327
3328 return 1;
3329}
3330
ea67176a
ID		 3331/* [vstart, vend) */
3332static int chunk_vrange_filter(struct extent_buffer *leaf,
3333 struct btrfs_chunk *chunk,
3334 u64 chunk_offset,
3335 struct btrfs_balance_args *bargs)
3336{
3337 if (chunk_offset < bargs->vend &&
3338 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3339 /* at least part of the chunk is inside this vrange */
3340 return 0;
3341
3342 return 1;
3343}
3344
dee32d0a
GAP		 3345static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3346 struct btrfs_chunk *chunk,
3347 struct btrfs_balance_args *bargs)
3348{
3349 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3350
3351 if (bargs->stripes_min <= num_stripes
3352 && num_stripes <= bargs->stripes_max)
3353 return 0;
3354
3355 return 1;
3356}
3357
899c81ea 3358static int chunk_soft_convert_filter(u64 chunk_type,
cfa4c961
ID		 3359 struct btrfs_balance_args *bargs)
3360{
3361 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3362 return 0;
3363
899c81ea
ID		 3364 chunk_type = chunk_to_extended(chunk_type) &
3365 BTRFS_EXTENDED_PROFILE_MASK;
cfa4c961 3366
899c81ea 3367 if (bargs->target == chunk_type)
cfa4c961
ID		 3368 return 1;
3369
3370 return 0;
3371}
3372
2ff7e61e 3373static int should_balance_chunk(struct btrfs_fs_info *fs_info,
f43ffb60
ID		 3374 struct extent_buffer *leaf,
3375 struct btrfs_chunk *chunk, u64 chunk_offset)
3376{
0b246afa 3377 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
f43ffb60
ID		 3378 struct btrfs_balance_args *bargs = NULL;
3379 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3380
3381 /* type filter */
3382 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3383 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3384 return 0;
3385 }
3386
3387 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3388 bargs = &bctl->data;
3389 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3390 bargs = &bctl->sys;
3391 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3392 bargs = &bctl->meta;
3393
ed25e9b2
ID		 3394 /* profiles filter */
3395 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3396 chunk_profiles_filter(chunk_type, bargs)) {
3397 return 0;
5ce5b3c0
ID		 3398 }
3399
3400 /* usage filter */
3401 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
0b246afa 3402 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
5ce5b3c0 3403 return 0;
bc309467 3404 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
0b246afa 3405 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
bc309467 3406 return 0;
409d404b
ID		 3407 }
3408
3409 /* devid filter */
3410 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3411 chunk_devid_filter(leaf, chunk, bargs)) {
3412 return 0;
94e60d5a
ID		 3413 }
3414
3415 /* drange filter, makes sense only with devid filter */
3416 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
e4ff5fb5 3417 chunk_drange_filter(leaf, chunk, bargs)) {
94e60d5a 3418 return 0;
ea67176a
ID		 3419 }
3420
3421 /* vrange filter */
3422 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3423 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3424 return 0;
ed25e9b2
ID		 3425 }
3426
dee32d0a
GAP		 3427 /* stripes filter */
3428 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3429 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3430 return 0;
3431 }
3432
cfa4c961
ID		 3433 /* soft profile changing mode */
3434 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3435 chunk_soft_convert_filter(chunk_type, bargs)) {
3436 return 0;
3437 }
3438
7d824b6f
DS		 3439 /*
3440 * limited by count, must be the last filter
3441 */
3442 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3443 if (bargs->limit == 0)
3444 return 0;
3445 else
3446 bargs->limit--;
12907fc7
DS		 3447 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3448 /*
3449 * Same logic as the 'limit' filter; the minimum cannot be
01327610 3450 * determined here because we do not have the global information
12907fc7
DS		 3451 * about the count of all chunks that satisfy the filters.
3452 */
3453 if (bargs->limit_max == 0)
3454 return 0;
3455 else
3456 bargs->limit_max--;
7d824b6f
DS		 3457 }
3458
f43ffb60
ID		 3459 return 1;
3460}
3461
c9e9f97b 3462static int __btrfs_balance(struct btrfs_fs_info *fs_info)
ec44a35c 3463{
19a39dce 3464 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
c9e9f97b
ID		 3465 struct btrfs_root *chunk_root = fs_info->chunk_root;
3466 struct btrfs_root *dev_root = fs_info->dev_root;
3467 struct list_head *devices;
ec44a35c
CM		 3468 struct btrfs_device *device;
3469 u64 old_size;
3470 u64 size_to_free;
12907fc7 3471 u64 chunk_type;
f43ffb60 3472 struct btrfs_chunk *chunk;
5a488b9d 3473 struct btrfs_path *path = NULL;
ec44a35c 3474 struct btrfs_key key;
ec44a35c 3475 struct btrfs_key found_key;
c9e9f97b 3476 struct btrfs_trans_handle *trans;
f43ffb60
ID		 3477 struct extent_buffer *leaf;
3478 int slot;
c9e9f97b
ID		 3479 int ret;
3480 int enospc_errors = 0;
19a39dce 3481 bool counting = true;
12907fc7 3482 /* The single value limit and min/max limits use the same bytes in the */
7d824b6f
DS		 3483 u64 limit_data = bctl->data.limit;
3484 u64 limit_meta = bctl->meta.limit;
3485 u64 limit_sys = bctl->sys.limit;
12907fc7
DS		 3486 u32 count_data = 0;
3487 u32 count_meta = 0;
3488 u32 count_sys = 0;
2c9fe835 3489 int chunk_reserved = 0;
cf25ce51 3490 u64 bytes_used = 0;
ec44a35c 3491
ec44a35c 3492 /* step one make some room on all the devices */
c9e9f97b 3493 devices = &fs_info->fs_devices->devices;
c6e30871 3494 list_for_each_entry(device, devices, dev_list) {
7cc8e58d 3495 old_size = btrfs_device_get_total_bytes(device);
ec44a35c 3496 size_to_free = div_factor(old_size, 1);
ee22184b 3497 size_to_free = min_t(u64, size_to_free, SZ_1M);
2b82032c 3498 if (!device->writeable ||
7cc8e58d
MX		 3499 btrfs_device_get_total_bytes(device) -
3500 btrfs_device_get_bytes_used(device) > size_to_free ||
63a212ab 3501 device->is_tgtdev_for_dev_replace)
ec44a35c
CM		 3502 continue;
3503
3504 ret = btrfs_shrink_device(device, old_size - size_to_free);
ba1bf481
JB		 3505 if (ret == -ENOSPC)
3506 break;
5a488b9d
LB		 3507 if (ret) {
3508 /* btrfs_shrink_device never returns ret > 0 */
3509 WARN_ON(ret > 0);
3510 goto error;
3511 }
ec44a35c 3512
a22285a6 3513 trans = btrfs_start_transaction(dev_root, 0);
5a488b9d
LB		 3514 if (IS_ERR(trans)) {
3515 ret = PTR_ERR(trans);
3516 btrfs_info_in_rcu(fs_info,
3517 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3518 rcu_str_deref(device->name), ret,
3519 old_size, old_size - size_to_free);
3520 goto error;
3521 }
ec44a35c
CM		 3522
3523 ret = btrfs_grow_device(trans, device, old_size);
5a488b9d 3524 if (ret) {
3a45bb20 3525 btrfs_end_transaction(trans);
5a488b9d
LB		 3526 /* btrfs_grow_device never returns ret > 0 */
3527 WARN_ON(ret > 0);
3528 btrfs_info_in_rcu(fs_info,
3529 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3530 rcu_str_deref(device->name), ret,
3531 old_size, old_size - size_to_free);
3532 goto error;
3533 }
ec44a35c 3534
3a45bb20 3535 btrfs_end_transaction(trans);
ec44a35c
CM		 3536 }
3537
3538 /* step two, relocate all the chunks */
3539 path = btrfs_alloc_path();
17e9f796
MF		 3540 if (!path) {
3541 ret = -ENOMEM;
3542 goto error;
3543 }
19a39dce
ID		 3544
3545 /* zero out stat counters */
3546 spin_lock(&fs_info->balance_lock);
3547 memset(&bctl->stat, 0, sizeof(bctl->stat));
3548 spin_unlock(&fs_info->balance_lock);
3549again:
7d824b6f 3550 if (!counting) {
12907fc7
DS		 3551 /*
3552 * The single value limit and min/max limits use the same bytes
3553 * in the
3554 */
7d824b6f
DS		 3555 bctl->data.limit = limit_data;
3556 bctl->meta.limit = limit_meta;
3557 bctl->sys.limit = limit_sys;
3558 }
ec44a35c
CM		 3559 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3560 key.offset = (u64)-1;
3561 key.type = BTRFS_CHUNK_ITEM_KEY;
3562
d397712b 3563 while (1) {
19a39dce 3564 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
a7e99c69 3565 atomic_read(&fs_info->balance_cancel_req)) {
837d5b6e
ID		 3566 ret = -ECANCELED;
3567 goto error;
3568 }
3569
67c5e7d4 3570 mutex_lock(&fs_info->delete_unused_bgs_mutex);
ec44a35c 3571 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
67c5e7d4
FM		 3572 if (ret < 0) {
3573 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
ec44a35c 3574 goto error;
67c5e7d4 3575 }
ec44a35c
CM		 3576
3577 /*
3578 * this shouldn't happen, it means the last relocate
3579 * failed
3580 */
3581 if (ret == 0)
c9e9f97b 3582 BUG(); /* FIXME break ? */
ec44a35c
CM		 3583
3584 ret = btrfs_previous_item(chunk_root, path, 0,
3585 BTRFS_CHUNK_ITEM_KEY);
c9e9f97b 3586 if (ret) {
67c5e7d4 3587 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
c9e9f97b 3588 ret = 0;
ec44a35c 3589 break;
c9e9f97b 3590 }
7d9eb12c 3591
f43ffb60
ID		 3592 leaf = path->nodes[0];
3593 slot = path->slots[0];
3594 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7d9eb12c 3595
67c5e7d4
FM		 3596 if (found_key.objectid != key.objectid) {
3597 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
ec44a35c 3598 break;
67c5e7d4 3599 }
7d9eb12c 3600
f43ffb60 3601 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
12907fc7 3602 chunk_type = btrfs_chunk_type(leaf, chunk);
f43ffb60 3603
19a39dce
ID		 3604 if (!counting) {
3605 spin_lock(&fs_info->balance_lock);
3606 bctl->stat.considered++;
3607 spin_unlock(&fs_info->balance_lock);
3608 }
3609
2ff7e61e 3610 ret = should_balance_chunk(fs_info, leaf, chunk,
f43ffb60 3611 found_key.offset);
2c9fe835 3612
b3b4aa74 3613 btrfs_release_path(path);
67c5e7d4
FM		 3614 if (!ret) {
3615 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
f43ffb60 3616 goto loop;
67c5e7d4 3617 }
f43ffb60 3618
19a39dce 3619 if (counting) {
67c5e7d4 3620 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
19a39dce
ID		 3621 spin_lock(&fs_info->balance_lock);
3622 bctl->stat.expected++;
3623 spin_unlock(&fs_info->balance_lock);
12907fc7
DS		 3624
3625 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3626 count_data++;
3627 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3628 count_sys++;
3629 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3630 count_meta++;
3631
3632 goto loop;
3633 }
3634
3635 /*
3636 * Apply limit_min filter, no need to check if the LIMITS
3637 * filter is used, limit_min is 0 by default
3638 */
3639 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3640 count_data < bctl->data.limit_min)
3641 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3642 count_meta < bctl->meta.limit_min)
3643 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3644 count_sys < bctl->sys.limit_min)) {
3645 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
19a39dce
ID		 3646 goto loop;
3647 }
3648
cf25ce51
LB		 3649 ASSERT(fs_info->data_sinfo);
3650 spin_lock(&fs_info->data_sinfo->lock);
3651 bytes_used = fs_info->data_sinfo->bytes_used;
3652 spin_unlock(&fs_info->data_sinfo->lock);
3653
3654 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3655 !chunk_reserved && !bytes_used) {
2c9fe835
ZL		 3656 trans = btrfs_start_transaction(chunk_root, 0);
3657 if (IS_ERR(trans)) {
3658 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3659 ret = PTR_ERR(trans);
3660 goto error;
3661 }
3662
2ff7e61e 3663 ret = btrfs_force_chunk_alloc(trans, fs_info,
2c9fe835 3664 BTRFS_BLOCK_GROUP_DATA);
3a45bb20 3665 btrfs_end_transaction(trans);
2c9fe835
ZL		 3666 if (ret < 0) {
3667 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3668 goto error;
3669 }
2c9fe835
ZL		 3670 chunk_reserved = 1;
3671 }
3672
5b4aacef 3673 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
67c5e7d4 3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
508794eb
JB		 3675 if (ret && ret != -ENOSPC)
3676 goto error;
19a39dce 3677 if (ret == -ENOSPC) {
c9e9f97b 3678 enospc_errors++;
19a39dce
ID		 3679 } else {
3680 spin_lock(&fs_info->balance_lock);
3681 bctl->stat.completed++;
3682 spin_unlock(&fs_info->balance_lock);
3683 }
f43ffb60 3684loop:
795a3321
ID		 3685 if (found_key.offset == 0)
3686 break;
ba1bf481 3687 key.offset = found_key.offset - 1;
ec44a35c 3688 }
c9e9f97b 3689
19a39dce
ID		 3690 if (counting) {
3691 btrfs_release_path(path);
3692 counting = false;
3693 goto again;
3694 }
ec44a35c
CM		 3695error:
3696 btrfs_free_path(path);
c9e9f97b 3697 if (enospc_errors) {
efe120a0 3698 btrfs_info(fs_info, "%d enospc errors during balance",
5d163e0e 3699 enospc_errors);
c9e9f97b
ID		 3700 if (!ret)
3701 ret = -ENOSPC;
3702 }
3703
ec44a35c
CM		 3704 return ret;
3705}
3706
0c460c0d
ID		 3707/**
3708 * alloc_profile_is_valid - see if a given profile is valid and reduced
3709 * @flags: profile to validate
3710 * @extended: if true @flags is treated as an extended profile
3711 */
3712static int alloc_profile_is_valid(u64 flags, int extended)
3713{
3714 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3715 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3716
3717 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3718
3719 /* 1) check that all other bits are zeroed */
3720 if (flags & ~mask)
3721 return 0;
3722
3723 /* 2) see if profile is reduced */
3724 if (flags == 0)
3725 return !extended; /* "0" is valid for usual profiles */
3726
3727 /* true if exactly one bit set */
3728 return (flags & (flags - 1)) == 0;
3729}
3730
837d5b6e
ID		 3731static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3732{
a7e99c69
ID		 3733 /* cancel requested || normal exit path */
3734 return atomic_read(&fs_info->balance_cancel_req) ||
3735 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3736 atomic_read(&fs_info->balance_cancel_req) == 0);
837d5b6e
ID		 3737}
3738
c9e9f97b
ID		 3739static void __cancel_balance(struct btrfs_fs_info *fs_info)
3740{
0940ebf6
ID		 3741 int ret;
3742
c9e9f97b 3743 unset_balance_control(fs_info);
6bccf3ab 3744 ret = del_balance_item(fs_info);
0f788c58 3745 if (ret)
34d97007 3746 btrfs_handle_fs_error(fs_info, ret, NULL);
ed0fb78f 3747
171938e5 3748 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
c9e9f97b
ID		 3749}
3750
bdcd3c97
AM		 3751/* Non-zero return value signifies invalidity */
3752static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3753 u64 allowed)
3754{
3755 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3756 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3757 (bctl_arg->target & ~allowed)));
3758}
3759
c9e9f97b
ID		 3760/*
3761 * Should be called with both balance and volume mutexes held
3762 */
3763int btrfs_balance(struct btrfs_balance_control *bctl,
3764 struct btrfs_ioctl_balance_args *bargs)
3765{
3766 struct btrfs_fs_info *fs_info = bctl->fs_info;
14506127 3767 u64 meta_target, data_target;
f43ffb60 3768 u64 allowed;
e4837f8f 3769 int mixed = 0;
c9e9f97b 3770 int ret;
8dabb742 3771 u64 num_devices;
de98ced9 3772 unsigned seq;
c9e9f97b 3773
837d5b6e 3774 if (btrfs_fs_closing(fs_info) ||
a7e99c69
ID		 3775 atomic_read(&fs_info->balance_pause_req) ||
3776 atomic_read(&fs_info->balance_cancel_req)) {
c9e9f97b
ID		 3777 ret = -EINVAL;
3778 goto out;
3779 }
3780
e4837f8f
ID		 3781 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3782 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3783 mixed = 1;
3784
f43ffb60
ID		 3785 /*
3786 * In case of mixed groups both data and meta should be picked,
3787 * and identical options should be given for both of them.
3788 */
e4837f8f
ID		 3789 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3790 if (mixed && (bctl->flags & allowed)) {
f43ffb60
ID		 3791 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3792 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3793 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
5d163e0e
JM		 3794 btrfs_err(fs_info,
3795 "with mixed groups data and metadata balance options must be the same");
f43ffb60
ID		 3796 ret = -EINVAL;
3797 goto out;
3798 }
3799 }
3800
8dabb742 3801 num_devices = fs_info->fs_devices->num_devices;
73beece9 3802 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
8dabb742
SB		 3803 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3804 BUG_ON(num_devices < 1);
3805 num_devices--;
3806 }
73beece9 3807 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
88be159c
AH		 3808 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3809 if (num_devices > 1)
e4d8ec0f 3810 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
8250dabe
AP		 3811 if (num_devices > 2)
3812 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3813 if (num_devices > 3)
3814 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3815 BTRFS_BLOCK_GROUP_RAID6);
bdcd3c97 3816 if (validate_convert_profile(&bctl->data, allowed)) {
5d163e0e
JM		 3817 btrfs_err(fs_info,
3818 "unable to start balance with target data profile %llu",
3819 bctl->data.target);
e4d8ec0f
ID		 3820 ret = -EINVAL;
3821 goto out;
3822 }
bdcd3c97 3823 if (validate_convert_profile(&bctl->meta, allowed)) {
efe120a0 3824 btrfs_err(fs_info,
5d163e0e
JM		 3825 "unable to start balance with target metadata profile %llu",
3826 bctl->meta.target);
e4d8ec0f
ID		 3827 ret = -EINVAL;
3828 goto out;
3829 }
bdcd3c97 3830 if (validate_convert_profile(&bctl->sys, allowed)) {
efe120a0 3831 btrfs_err(fs_info,
5d163e0e
JM		 3832 "unable to start balance with target system profile %llu",
3833 bctl->sys.target);
e4d8ec0f
ID		 3834 ret = -EINVAL;
3835 goto out;
3836 }
3837
e4d8ec0f
ID		 3838 /* allow to reduce meta or sys integrity only if force set */
3839 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
53b381b3
DW		 3840 BTRFS_BLOCK_GROUP_RAID10 |
3841 BTRFS_BLOCK_GROUP_RAID5 |
3842 BTRFS_BLOCK_GROUP_RAID6;
de98ced9
MX		 3843 do {
3844 seq = read_seqbegin(&fs_info->profiles_lock);
3845
3846 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3847 (fs_info->avail_system_alloc_bits & allowed) &&
3848 !(bctl->sys.target & allowed)) ||
3849 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3850 (fs_info->avail_metadata_alloc_bits & allowed) &&
3851 !(bctl->meta.target & allowed))) {
3852 if (bctl->flags & BTRFS_BALANCE_FORCE) {
5d163e0e
JM		 3853 btrfs_info(fs_info,
3854 "force reducing metadata integrity");
de98ced9 3855 } else {
5d163e0e
JM		 3856 btrfs_err(fs_info,
3857 "balance will reduce metadata integrity, use force if you want this");
de98ced9
MX		 3858 ret = -EINVAL;
3859 goto out;
3860 }
e4d8ec0f 3861 }
de98ced9 3862 } while (read_seqretry(&fs_info->profiles_lock, seq));
e4d8ec0f 3863
14506127
AB		 3864 /* if we're not converting, the target field is uninitialized */
3865 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3866 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3867 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3868 bctl->data.target : fs_info->avail_data_alloc_bits;
3869 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3870 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
ee592d07 3871 btrfs_warn(fs_info,
5d163e0e 3872 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
14506127 3873 meta_target, data_target);
ee592d07
ST		 3874 }
3875
6bccf3ab 3876 ret = insert_balance_item(fs_info, bctl);
59641015 3877 if (ret && ret != -EEXIST)
0940ebf6
ID		 3878 goto out;
3879
59641015
ID		 3880 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3881 BUG_ON(ret == -EEXIST);
3882 set_balance_control(bctl);
3883 } else {
3884 BUG_ON(ret != -EEXIST);
3885 spin_lock(&fs_info->balance_lock);
3886 update_balance_args(bctl);
3887 spin_unlock(&fs_info->balance_lock);
3888 }
c9e9f97b 3889
837d5b6e 3890 atomic_inc(&fs_info->balance_running);
c9e9f97b
ID		 3891 mutex_unlock(&fs_info->balance_mutex);
3892
3893 ret = __btrfs_balance(fs_info);
3894
3895 mutex_lock(&fs_info->balance_mutex);
837d5b6e 3896 atomic_dec(&fs_info->balance_running);
c9e9f97b
ID		 3897
3898 if (bargs) {
3899 memset(bargs, 0, sizeof(*bargs));
19a39dce 3900 update_ioctl_balance_args(fs_info, 0, bargs);
c9e9f97b
ID		 3901 }
3902
3a01aa7a
ID		 3903 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3904 balance_need_close(fs_info)) {
3905 __cancel_balance(fs_info);
3906 }
3907
837d5b6e 3908 wake_up(&fs_info->balance_wait_q);
c9e9f97b
ID		 3909
3910 return ret;
3911out:
59641015
ID		 3912 if (bctl->flags & BTRFS_BALANCE_RESUME)
3913 __cancel_balance(fs_info);
ed0fb78f 3914 else {
59641015 3915 kfree(bctl);
171938e5 3916 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
ed0fb78f 3917 }
59641015
ID		 3918 return ret;
3919}
3920
3921static int balance_kthread(void *data)
3922{
2b6ba629 3923 struct btrfs_fs_info *fs_info = data;
9555c6c1 3924 int ret = 0;
59641015
ID		 3925
3926 mutex_lock(&fs_info->volume_mutex);
3927 mutex_lock(&fs_info->balance_mutex);
3928
2b6ba629 3929 if (fs_info->balance_ctl) {
efe120a0 3930 btrfs_info(fs_info, "continuing balance");
2b6ba629 3931 ret = btrfs_balance(fs_info->balance_ctl, NULL);
9555c6c1 3932 }
59641015
ID		 3933
3934 mutex_unlock(&fs_info->balance_mutex);
3935 mutex_unlock(&fs_info->volume_mutex);
2b6ba629 3936
59641015
ID		 3937 return ret;
3938}
3939
2b6ba629
ID		 3940int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3941{
3942 struct task_struct *tsk;
3943
3944 spin_lock(&fs_info->balance_lock);
3945 if (!fs_info->balance_ctl) {
3946 spin_unlock(&fs_info->balance_lock);
3947 return 0;
3948 }
3949 spin_unlock(&fs_info->balance_lock);
3950
3cdde224 3951 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
efe120a0 3952 btrfs_info(fs_info, "force skipping balance");
2b6ba629
ID		 3953 return 0;
3954 }
3955
3956 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
cd633972 3957 return PTR_ERR_OR_ZERO(tsk);
2b6ba629
ID		 3958}
3959
68310a5e 3960int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
59641015 3961{
59641015
ID		 3962 struct btrfs_balance_control *bctl;
3963 struct btrfs_balance_item *item;
3964 struct btrfs_disk_balance_args disk_bargs;
3965 struct btrfs_path *path;
3966 struct extent_buffer *leaf;
3967 struct btrfs_key key;
3968 int ret;
3969
3970 path = btrfs_alloc_path();
3971 if (!path)
3972 return -ENOMEM;
3973
59641015 3974 key.objectid = BTRFS_BALANCE_OBJECTID;
c479cb4f 3975 key.type = BTRFS_TEMPORARY_ITEM_KEY;
59641015
ID		 3976 key.offset = 0;
3977
68310a5e 3978 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
59641015 3979 if (ret < 0)
68310a5e 3980 goto out;
59641015
ID		 3981 if (ret > 0) { /* ret = -ENOENT; */
3982 ret = 0;
68310a5e
ID		 3983 goto out;
3984 }
3985
3986 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3987 if (!bctl) {
3988 ret = -ENOMEM;
3989 goto out;
59641015
ID		 3990 }
3991
3992 leaf = path->nodes[0];
3993 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3994
68310a5e
ID		 3995 bctl->fs_info = fs_info;
3996 bctl->flags = btrfs_balance_flags(leaf, item);
3997 bctl->flags |= BTRFS_BALANCE_RESUME;
59641015
ID		 3998
3999 btrfs_balance_data(leaf, item, &disk_bargs);
4000 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4001 btrfs_balance_meta(leaf, item, &disk_bargs);
4002 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4003 btrfs_balance_sys(leaf, item, &disk_bargs);
4004 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4005
171938e5 4006 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
ed0fb78f 4007
68310a5e
ID		 4008 mutex_lock(&fs_info->volume_mutex);
4009 mutex_lock(&fs_info->balance_mutex);
59641015 4010
68310a5e
ID		 4011 set_balance_control(bctl);
4012
4013 mutex_unlock(&fs_info->balance_mutex);
4014 mutex_unlock(&fs_info->volume_mutex);
59641015
ID		 4015out:
4016 btrfs_free_path(path);
ec44a35c
CM		 4017 return ret;
4018}
4019
837d5b6e
ID		 4020int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4021{
4022 int ret = 0;
4023
4024 mutex_lock(&fs_info->balance_mutex);
4025 if (!fs_info->balance_ctl) {
4026 mutex_unlock(&fs_info->balance_mutex);
4027 return -ENOTCONN;
4028 }
4029
4030 if (atomic_read(&fs_info->balance_running)) {
4031 atomic_inc(&fs_info->balance_pause_req);
4032 mutex_unlock(&fs_info->balance_mutex);
4033
4034 wait_event(fs_info->balance_wait_q,
4035 atomic_read(&fs_info->balance_running) == 0);
4036
4037 mutex_lock(&fs_info->balance_mutex);
4038 /* we are good with balance_ctl ripped off from under us */
4039 BUG_ON(atomic_read(&fs_info->balance_running));
4040 atomic_dec(&fs_info->balance_pause_req);
4041 } else {
4042 ret = -ENOTCONN;
4043 }
4044
4045 mutex_unlock(&fs_info->balance_mutex);
4046 return ret;
4047}
4048
a7e99c69
ID		 4049int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4050{
bc98a42c 4051 if (sb_rdonly(fs_info->sb))
e649e587
ID		 4052 return -EROFS;
4053
a7e99c69
ID		 4054 mutex_lock(&fs_info->balance_mutex);
4055 if (!fs_info->balance_ctl) {
4056 mutex_unlock(&fs_info->balance_mutex);
4057 return -ENOTCONN;
4058 }
4059
4060 atomic_inc(&fs_info->balance_cancel_req);
4061 /*
4062 * if we are running just wait and return, balance item is
4063 * deleted in btrfs_balance in this case
4064 */
4065 if (atomic_read(&fs_info->balance_running)) {
4066 mutex_unlock(&fs_info->balance_mutex);
4067 wait_event(fs_info->balance_wait_q,
4068 atomic_read(&fs_info->balance_running) == 0);
4069 mutex_lock(&fs_info->balance_mutex);
4070 } else {
4071 /* __cancel_balance needs volume_mutex */
4072 mutex_unlock(&fs_info->balance_mutex);
4073 mutex_lock(&fs_info->volume_mutex);
4074 mutex_lock(&fs_info->balance_mutex);
4075
4076 if (fs_info->balance_ctl)
4077 __cancel_balance(fs_info);
4078
4079 mutex_unlock(&fs_info->volume_mutex);
4080 }
4081
4082 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4083 atomic_dec(&fs_info->balance_cancel_req);
4084 mutex_unlock(&fs_info->balance_mutex);
4085 return 0;
4086}
4087
803b2f54
SB		 4088static int btrfs_uuid_scan_kthread(void *data)
4089{
4090 struct btrfs_fs_info *fs_info = data;
4091 struct btrfs_root *root = fs_info->tree_root;
4092 struct btrfs_key key;
803b2f54
SB		 4093 struct btrfs_path *path = NULL;
4094 int ret = 0;
4095 struct extent_buffer *eb;
4096 int slot;
4097 struct btrfs_root_item root_item;
4098 u32 item_size;
f45388f3 4099 struct btrfs_trans_handle *trans = NULL;
803b2f54
SB		 4100
4101 path = btrfs_alloc_path();
4102 if (!path) {
4103 ret = -ENOMEM;
4104 goto out;
4105 }
4106
4107 key.objectid = 0;
4108 key.type = BTRFS_ROOT_ITEM_KEY;
4109 key.offset = 0;
4110
803b2f54 4111 while (1) {
6174d3cb 4112 ret = btrfs_search_forward(root, &key, path, 0);
803b2f54
SB		 4113 if (ret) {
4114 if (ret > 0)
4115 ret = 0;
4116 break;
4117 }
4118
4119 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4120 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4121 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4122 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4123 goto skip;
4124
4125 eb = path->nodes[0];
4126 slot = path->slots[0];
4127 item_size = btrfs_item_size_nr(eb, slot);
4128 if (item_size < sizeof(root_item))
4129 goto skip;
4130
803b2f54
SB		 4131 read_extent_buffer(eb, &root_item,
4132 btrfs_item_ptr_offset(eb, slot),
4133 (int)sizeof(root_item));
4134 if (btrfs_root_refs(&root_item) == 0)
4135 goto skip;
f45388f3
FDBM		 4136
4137 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4138 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4139 if (trans)
4140 goto update_tree;
4141
4142 btrfs_release_path(path);
803b2f54
SB		 4143 /*
4144 * 1 - subvol uuid item
4145 * 1 - received_subvol uuid item
4146 */
4147 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4148 if (IS_ERR(trans)) {
4149 ret = PTR_ERR(trans);
4150 break;
4151 }
f45388f3
FDBM		 4152 continue;
4153 } else {
4154 goto skip;
4155 }
4156update_tree:
4157 if (!btrfs_is_empty_uuid(root_item.uuid)) {
6bccf3ab 4158 ret = btrfs_uuid_tree_add(trans, fs_info,
803b2f54
SB		 4159 root_item.uuid,
4160 BTRFS_UUID_KEY_SUBVOL,
4161 key.objectid);
4162 if (ret < 0) {
efe120a0 4163 btrfs_warn(fs_info, "uuid_tree_add failed %d",
803b2f54 4164 ret);
803b2f54
SB		 4165 break;
4166 }
4167 }
4168
4169 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
6bccf3ab 4170 ret = btrfs_uuid_tree_add(trans, fs_info,
803b2f54
SB		 4171 root_item.received_uuid,
4172 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4173 key.objectid);
4174 if (ret < 0) {
efe120a0 4175 btrfs_warn(fs_info, "uuid_tree_add failed %d",
803b2f54 4176 ret);
803b2f54
SB		 4177 break;
4178 }
4179 }
4180
f45388f3 4181skip:
803b2f54 4182 if (trans) {
3a45bb20 4183 ret = btrfs_end_transaction(trans);
f45388f3 4184 trans = NULL;
803b2f54
SB		 4185 if (ret)
4186 break;
4187 }
4188
803b2f54
SB		 4189 btrfs_release_path(path);
4190 if (key.offset < (u64)-1) {
4191 key.offset++;
4192 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4193 key.offset = 0;
4194 key.type = BTRFS_ROOT_ITEM_KEY;
4195 } else if (key.objectid < (u64)-1) {
4196 key.offset = 0;
4197 key.type = BTRFS_ROOT_ITEM_KEY;
4198 key.objectid++;
4199 } else {
4200 break;
4201 }
4202 cond_resched();
4203 }
4204
4205out:
4206 btrfs_free_path(path);
f45388f3 4207 if (trans && !IS_ERR(trans))
3a45bb20 4208 btrfs_end_transaction(trans);
803b2f54 4209 if (ret)
efe120a0 4210 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
70f80175 4211 else
afcdd129 4212 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
803b2f54
SB		 4213 up(&fs_info->uuid_tree_rescan_sem);
4214 return 0;
4215}
4216
70f80175
SB		 4217/*
4218 * Callback for btrfs_uuid_tree_iterate().
4219 * returns:
4220 * 0 check succeeded, the entry is not outdated.
bb7ab3b9 4221 * < 0 if an error occurred.
70f80175
SB		 4222 * > 0 if the check failed, which means the caller shall remove the entry.
4223 */
4224static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4225 u8 *uuid, u8 type, u64 subid)
4226{
4227 struct btrfs_key key;
4228 int ret = 0;
4229 struct btrfs_root *subvol_root;
4230
4231 if (type != BTRFS_UUID_KEY_SUBVOL &&
4232 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4233 goto out;
4234
4235 key.objectid = subid;
4236 key.type = BTRFS_ROOT_ITEM_KEY;
4237 key.offset = (u64)-1;
4238 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4239 if (IS_ERR(subvol_root)) {
4240 ret = PTR_ERR(subvol_root);
4241 if (ret == -ENOENT)
4242 ret = 1;
4243 goto out;
4244 }
4245
4246 switch (type) {
4247 case BTRFS_UUID_KEY_SUBVOL:
4248 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4249 ret = 1;
4250 break;
4251 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4252 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4253 BTRFS_UUID_SIZE))
4254 ret = 1;
4255 break;
4256 }
4257
4258out:
4259 return ret;
4260}
4261
4262static int btrfs_uuid_rescan_kthread(void *data)
4263{
4264 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4265 int ret;
4266
4267 /*
4268 * 1st step is to iterate through the existing UUID tree and
4269 * to delete all entries that contain outdated data.
4270 * 2nd step is to add all missing entries to the UUID tree.
4271 */
4272 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4273 if (ret < 0) {
efe120a0 4274 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
70f80175
SB		 4275 up(&fs_info->uuid_tree_rescan_sem);
4276 return ret;
4277 }
4278 return btrfs_uuid_scan_kthread(data);
4279}
4280
f7a81ea4
SB		 4281int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4282{
4283 struct btrfs_trans_handle *trans;
4284 struct btrfs_root *tree_root = fs_info->tree_root;
4285 struct btrfs_root *uuid_root;
803b2f54
SB		 4286 struct task_struct *task;
4287 int ret;
f7a81ea4
SB		 4288
4289 /*
4290 * 1 - root node
4291 * 1 - root item
4292 */
4293 trans = btrfs_start_transaction(tree_root, 2);
4294 if (IS_ERR(trans))
4295 return PTR_ERR(trans);
4296
4297 uuid_root = btrfs_create_tree(trans, fs_info,
4298 BTRFS_UUID_TREE_OBJECTID);
4299 if (IS_ERR(uuid_root)) {
6d13f549 4300 ret = PTR_ERR(uuid_root);
66642832 4301 btrfs_abort_transaction(trans, ret);
3a45bb20 4302 btrfs_end_transaction(trans);
6d13f549 4303 return ret;
f7a81ea4
SB		 4304 }
4305
4306 fs_info->uuid_root = uuid_root;
4307
3a45bb20 4308 ret = btrfs_commit_transaction(trans);
803b2f54
SB		 4309 if (ret)
4310 return ret;
4311
4312 down(&fs_info->uuid_tree_rescan_sem);
4313 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4314 if (IS_ERR(task)) {
70f80175 4315 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
efe120a0 4316 btrfs_warn(fs_info, "failed to start uuid_scan task");
803b2f54
SB		 4317 up(&fs_info->uuid_tree_rescan_sem);
4318 return PTR_ERR(task);
4319 }
4320
4321 return 0;
f7a81ea4 4322}
803b2f54 4323
70f80175
SB		 4324int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4325{
4326 struct task_struct *task;
4327
4328 down(&fs_info->uuid_tree_rescan_sem);
4329 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4330 if (IS_ERR(task)) {
4331 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
efe120a0 4332 btrfs_warn(fs_info, "failed to start uuid_rescan task");
70f80175
SB		 4333 up(&fs_info->uuid_tree_rescan_sem);
4334 return PTR_ERR(task);
4335 }
4336
4337 return 0;
4338}
4339
8f18cf13
CM		 4340/*
4341 * shrinking a device means finding all of the device extents past
4342 * the new size, and then following the back refs to the chunks.
4343 * The chunk relocation code actually frees the device extent
4344 */
4345int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4346{
0b246afa
JM		 4347 struct btrfs_fs_info *fs_info = device->fs_info;
4348 struct btrfs_root *root = fs_info->dev_root;
8f18cf13 4349 struct btrfs_trans_handle *trans;
8f18cf13
CM		 4350 struct btrfs_dev_extent *dev_extent = NULL;
4351 struct btrfs_path *path;
4352 u64 length;
8f18cf13
CM		 4353 u64 chunk_offset;
4354 int ret;
4355 int slot;
ba1bf481
JB		 4356 int failed = 0;
4357 bool retried = false;
53e489bc 4358 bool checked_pending_chunks = false;
8f18cf13
CM		 4359 struct extent_buffer *l;
4360 struct btrfs_key key;
0b246afa 4361 struct btrfs_super_block *super_copy = fs_info->super_copy;
8f18cf13 4362 u64 old_total = btrfs_super_total_bytes(super_copy);
7cc8e58d 4363 u64 old_size = btrfs_device_get_total_bytes(device);
7dfb8be1
NB		 4364 u64 diff;
4365
4366 new_size = round_down(new_size, fs_info->sectorsize);
0e4324a4 4367 diff = round_down(old_size - new_size, fs_info->sectorsize);
8f18cf13 4368
63a212ab
SB		 4369 if (device->is_tgtdev_for_dev_replace)
4370 return -EINVAL;
4371
8f18cf13
CM		 4372 path = btrfs_alloc_path();
4373 if (!path)
4374 return -ENOMEM;
4375
e4058b54 4376 path->reada = READA_FORWARD;
8f18cf13 4377
34441361 4378 mutex_lock(&fs_info->chunk_mutex);
7d9eb12c 4379
7cc8e58d 4380 btrfs_device_set_total_bytes(device, new_size);
2bf64758 4381 if (device->writeable) {
2b82032c 4382 device->fs_devices->total_rw_bytes -= diff;
a5ed45f8 4383 atomic64_sub(diff, &fs_info->free_chunk_space);
2bf64758 4384 }
34441361 4385 mutex_unlock(&fs_info->chunk_mutex);
8f18cf13 4386
ba1bf481 4387again:
8f18cf13
CM		 4388 key.objectid = device->devid;
4389 key.offset = (u64)-1;
4390 key.type = BTRFS_DEV_EXTENT_KEY;
4391
213e64da 4392 do {
0b246afa 4393 mutex_lock(&fs_info->delete_unused_bgs_mutex);
8f18cf13 4394 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
67c5e7d4 4395 if (ret < 0) {
0b246afa 4396 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
8f18cf13 4397 goto done;
67c5e7d4 4398 }
8f18cf13
CM		 4399
4400 ret = btrfs_previous_item(root, path, 0, key.type);
67c5e7d4 4401 if (ret)
0b246afa 4402 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
8f18cf13
CM		 4403 if (ret < 0)
4404 goto done;
4405 if (ret) {
4406 ret = 0;
b3b4aa74 4407 btrfs_release_path(path);
bf1fb512 4408 break;
8f18cf13
CM		 4409 }
4410
4411 l = path->nodes[0];
4412 slot = path->slots[0];
4413 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4414
ba1bf481 4415 if (key.objectid != device->devid) {
0b246afa 4416 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
b3b4aa74 4417 btrfs_release_path(path);
bf1fb512 4418 break;
ba1bf481 4419 }
8f18cf13
CM		 4420
4421 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4422 length = btrfs_dev_extent_length(l, dev_extent);
4423
ba1bf481 4424 if (key.offset + length <= new_size) {
0b246afa 4425 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
b3b4aa74 4426 btrfs_release_path(path);
d6397bae 4427 break;
ba1bf481 4428 }
8f18cf13 4429
8f18cf13 4430 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
b3b4aa74 4431 btrfs_release_path(path);
8f18cf13 4432
0b246afa
JM		 4433 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4434 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
ba1bf481 4435 if (ret && ret != -ENOSPC)
8f18cf13 4436 goto done;
ba1bf481
JB		 4437 if (ret == -ENOSPC)
4438 failed++;
213e64da 4439 } while (key.offset-- > 0);
ba1bf481
JB		 4440
4441 if (failed && !retried) {
4442 failed = 0;
4443 retried = true;
4444 goto again;
4445 } else if (failed && retried) {
4446 ret = -ENOSPC;
ba1bf481 4447 goto done;
8f18cf13
CM		 4448 }
4449
d6397bae 4450 /* Shrinking succeeded, else we would be at "done". */
a22285a6 4451 trans = btrfs_start_transaction(root, 0);
98d5dc13
TI		 4452 if (IS_ERR(trans)) {
4453 ret = PTR_ERR(trans);
4454 goto done;
4455 }
4456
34441361 4457 mutex_lock(&fs_info->chunk_mutex);
53e489bc
FM		 4458
4459 /*
4460 * We checked in the above loop all device extents that were already in
4461 * the device tree. However before we have updated the device's
4462 * total_bytes to the new size, we might have had chunk allocations that
4463 * have not complete yet (new block groups attached to transaction
4464 * handles), and therefore their device extents were not yet in the
4465 * device tree and we missed them in the loop above. So if we have any
4466 * pending chunk using a device extent that overlaps the device range
4467 * that we can not use anymore, commit the current transaction and
4468 * repeat the search on the device tree - this way we guarantee we will
4469 * not have chunks using device extents that end beyond 'new_size'.
4470 */
4471 if (!checked_pending_chunks) {
4472 u64 start = new_size;
4473 u64 len = old_size - new_size;
4474
499f377f
JM		 4475 if (contains_pending_extent(trans->transaction, device,
4476 &start, len)) {
34441361 4477 mutex_unlock(&fs_info->chunk_mutex);
53e489bc
FM		 4478 checked_pending_chunks = true;
4479 failed = 0;
4480 retried = false;
3a45bb20 4481 ret = btrfs_commit_transaction(trans);
53e489bc
FM		 4482 if (ret)
4483 goto done;
4484 goto again;
4485 }
4486 }
4487
7cc8e58d 4488 btrfs_device_set_disk_total_bytes(device, new_size);
935e5cc9
MX		 4489 if (list_empty(&device->resized_list))
4490 list_add_tail(&device->resized_list,
0b246afa 4491 &fs_info->fs_devices->resized_devices);
d6397bae 4492
d6397bae 4493 WARN_ON(diff > old_total);
7dfb8be1
NB		 4494 btrfs_set_super_total_bytes(super_copy,
4495 round_down(old_total - diff, fs_info->sectorsize));
34441361 4496 mutex_unlock(&fs_info->chunk_mutex);
2196d6e8
MX		 4497
4498 /* Now btrfs_update_device() will change the on-disk size. */
4499 ret = btrfs_update_device(trans, device);
3a45bb20 4500 btrfs_end_transaction(trans);
8f18cf13
CM		 4501done:
4502 btrfs_free_path(path);
53e489bc 4503 if (ret) {
34441361 4504 mutex_lock(&fs_info->chunk_mutex);
53e489bc
FM		 4505 btrfs_device_set_total_bytes(device, old_size);
4506 if (device->writeable)
4507 device->fs_devices->total_rw_bytes += diff;
a5ed45f8 4508 atomic64_add(diff, &fs_info->free_chunk_space);
34441361 4509 mutex_unlock(&fs_info->chunk_mutex);
53e489bc 4510 }
8f18cf13
CM		 4511 return ret;
4512}
4513
2ff7e61e 4514static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
0b86a832
CM		 4515 struct btrfs_key *key,
4516 struct btrfs_chunk *chunk, int item_size)
4517{
0b246afa 4518 struct btrfs_super_block *super_copy = fs_info->super_copy;
0b86a832
CM		 4519 struct btrfs_disk_key disk_key;
4520 u32 array_size;
4521 u8 *ptr;
4522
34441361 4523 mutex_lock(&fs_info->chunk_mutex);
0b86a832 4524 array_size = btrfs_super_sys_array_size(super_copy);
5f43f86e 4525 if (array_size + item_size + sizeof(disk_key)
fe48a5c0 4526 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
34441361 4527 mutex_unlock(&fs_info->chunk_mutex);
0b86a832 4528 return -EFBIG;
fe48a5c0 4529 }
0b86a832
CM		 4530
4531 ptr = super_copy->sys_chunk_array + array_size;
4532 btrfs_cpu_key_to_disk(&disk_key, key);
4533 memcpy(ptr, &disk_key, sizeof(disk_key));
4534 ptr += sizeof(disk_key);
4535 memcpy(ptr, chunk, item_size);
4536 item_size += sizeof(disk_key);
4537 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
34441361 4538 mutex_unlock(&fs_info->chunk_mutex);
fe48a5c0 4539
0b86a832
CM		 4540 return 0;
4541}
4542
73c5de00
AJ		 4543/*
4544 * sort the devices in descending order by max_avail, total_avail
4545 */
4546static int btrfs_cmp_device_info(const void *a, const void *b)
9b3f68b9 4547{
73c5de00
AJ		 4548 const struct btrfs_device_info *di_a = a;
4549 const struct btrfs_device_info *di_b = b;
9b3f68b9 4550
73c5de00 4551 if (di_a->max_avail > di_b->max_avail)
b2117a39 4552 return -1;
73c5de00 4553 if (di_a->max_avail < di_b->max_avail)
b2117a39 4554 return 1;
73c5de00
AJ		 4555 if (di_a->total_avail > di_b->total_avail)
4556 return -1;
4557 if (di_a->total_avail < di_b->total_avail)
4558 return 1;
4559 return 0;
b2117a39 4560}
0b86a832 4561
53b381b3
DW		 4562static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4563{
ffe2d203 4564 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
53b381b3
DW		 4565 return;
4566
ceda0864 4567 btrfs_set_fs_incompat(info, RAID56);
53b381b3
DW		 4568}
4569
da17066c 4570#define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
23f8f9b7
GH		 4571 - sizeof(struct btrfs_chunk)) \
4572 / sizeof(struct btrfs_stripe) + 1)
4573
4574#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4575 - 2 * sizeof(struct btrfs_disk_key) \
4576 - 2 * sizeof(struct btrfs_chunk)) \
4577 / sizeof(struct btrfs_stripe) + 1)
4578
73c5de00 4579static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
72b468c8 4580 u64 start, u64 type)
b2117a39 4581{
2ff7e61e 4582 struct btrfs_fs_info *info = trans->fs_info;
73c5de00 4583 struct btrfs_fs_devices *fs_devices = info->fs_devices;
ebcc9301 4584 struct btrfs_device *device;
73c5de00
AJ		 4585 struct map_lookup *map = NULL;
4586 struct extent_map_tree *em_tree;
4587 struct extent_map *em;
4588 struct btrfs_device_info *devices_info = NULL;
4589 u64 total_avail;
4590 int num_stripes; /* total number of stripes to allocate */
53b381b3
DW		 4591 int data_stripes; /* number of stripes that count for
4592 block group size */
73c5de00
AJ		 4593 int sub_stripes; /* sub_stripes info for map */
4594 int dev_stripes; /* stripes per dev */
4595 int devs_max; /* max devs to use */
4596 int devs_min; /* min devs needed */
4597 int devs_increment; /* ndevs has to be a multiple of this */
4598 int ncopies; /* how many copies to data has */
4599 int ret;
4600 u64 max_stripe_size;
4601 u64 max_chunk_size;
4602 u64 stripe_size;
4603 u64 num_bytes;
4604 int ndevs;
4605 int i;
4606 int j;
31e50229 4607 int index;
593060d7 4608
0c460c0d 4609 BUG_ON(!alloc_profile_is_valid(type, 0));
9b3f68b9 4610
73c5de00
AJ		 4611 if (list_empty(&fs_devices->alloc_list))
4612 return -ENOSPC;
b2117a39 4613
31e50229 4614 index = __get_raid_index(type);
73c5de00 4615
31e50229
LB		 4616 sub_stripes = btrfs_raid_array[index].sub_stripes;
4617 dev_stripes = btrfs_raid_array[index].dev_stripes;
4618 devs_max = btrfs_raid_array[index].devs_max;
4619 devs_min = btrfs_raid_array[index].devs_min;
4620 devs_increment = btrfs_raid_array[index].devs_increment;
4621 ncopies = btrfs_raid_array[index].ncopies;
b2117a39 4622
9b3f68b9 4623 if (type & BTRFS_BLOCK_GROUP_DATA) {
ee22184b 4624 max_stripe_size = SZ_1G;
73c5de00 4625 max_chunk_size = 10 * max_stripe_size;
23f8f9b7
GH		 4626 if (!devs_max)
4627 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
9b3f68b9 4628 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1100373f 4629 /* for larger filesystems, use larger metadata chunks */
ee22184b
BL		 4630 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4631 max_stripe_size = SZ_1G;
1100373f 4632 else
ee22184b 4633 max_stripe_size = SZ_256M;
73c5de00 4634 max_chunk_size = max_stripe_size;
23f8f9b7
GH		 4635 if (!devs_max)
4636 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
a40a90a0 4637 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
ee22184b 4638 max_stripe_size = SZ_32M;
73c5de00 4639 max_chunk_size = 2 * max_stripe_size;
23f8f9b7
GH		 4640 if (!devs_max)
4641 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
73c5de00 4642 } else {
351fd353 4643 btrfs_err(info, "invalid chunk type 0x%llx requested",
73c5de00
AJ		 4644 type);
4645 BUG_ON(1);
9b3f68b9
CM		 4646 }
4647
2b82032c
YZ		 4648 /* we don't want a chunk larger than 10% of writeable space */
4649 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4650 max_chunk_size);
9b3f68b9 4651
31e818fe 4652 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
73c5de00
AJ		 4653 GFP_NOFS);
4654 if (!devices_info)
4655 return -ENOMEM;
0cad8a11 4656
9f680ce0 4657 /*
73c5de00
AJ		 4658 * in the first pass through the devices list, we gather information
4659 * about the available holes on each device.
9f680ce0 4660 */
73c5de00 4661 ndevs = 0;
ebcc9301 4662 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
73c5de00
AJ		 4663 u64 max_avail;
4664 u64 dev_offset;
b2117a39 4665
73c5de00 4666 if (!device->writeable) {
31b1a2bd 4667 WARN(1, KERN_ERR
efe120a0 4668 "BTRFS: read-only device in alloc_list\n");
73c5de00
AJ		 4669 continue;
4670 }
b2117a39 4671
63a212ab
SB		 4672 if (!device->in_fs_metadata ||
4673 device->is_tgtdev_for_dev_replace)
73c5de00 4674 continue;
b2117a39 4675
73c5de00
AJ		 4676 if (device->total_bytes > device->bytes_used)
4677 total_avail = device->total_bytes - device->bytes_used;
4678 else
4679 total_avail = 0;
38c01b96 4680
4681 /* If there is no space on this device, skip it. */
4682 if (total_avail == 0)
4683 continue;
b2117a39 4684
6df9a95e 4685 ret = find_free_dev_extent(trans, device,
73c5de00
AJ		 4686 max_stripe_size * dev_stripes,
4687 &dev_offset, &max_avail);
4688 if (ret && ret != -ENOSPC)
4689 goto error;
b2117a39 4690
73c5de00
AJ		 4691 if (ret == 0)
4692 max_avail = max_stripe_size * dev_stripes;
b2117a39 4693
73c5de00
AJ		 4694 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4695 continue;
b2117a39 4696
063d006f
ES		 4697 if (ndevs == fs_devices->rw_devices) {
4698 WARN(1, "%s: found more than %llu devices\n",
4699 __func__, fs_devices->rw_devices);
4700 break;
4701 }
73c5de00
AJ		 4702 devices_info[ndevs].dev_offset = dev_offset;
4703 devices_info[ndevs].max_avail = max_avail;
4704 devices_info[ndevs].total_avail = total_avail;
4705 devices_info[ndevs].dev = device;
4706 ++ndevs;
4707 }
b2117a39 4708
73c5de00
AJ		 4709 /*
4710 * now sort the devices by hole size / available space
4711 */
4712 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4713 btrfs_cmp_device_info, NULL);
b2117a39 4714
73c5de00 4715 /* round down to number of usable stripes */
e5600fd6 4716 ndevs = round_down(ndevs, devs_increment);
b2117a39 4717
73c5de00
AJ		 4718 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4719 ret = -ENOSPC;
4720 goto error;
b2117a39 4721 }
9f680ce0 4722
f148ef4d
NB		 4723 ndevs = min(ndevs, devs_max);
4724
73c5de00
AJ		 4725 /*
4726 * the primary goal is to maximize the number of stripes, so use as many
4727 * devices as possible, even if the stripes are not maximum sized.
4728 */
4729 stripe_size = devices_info[ndevs-1].max_avail;
4730 num_stripes = ndevs * dev_stripes;
b2117a39 4731
53b381b3
DW		 4732 /*
4733 * this will have to be fixed for RAID1 and RAID10 over
4734 * more drives
4735 */
4736 data_stripes = num_stripes / ncopies;
4737
500ceed8 4738 if (type & BTRFS_BLOCK_GROUP_RAID5)
53b381b3 4739 data_stripes = num_stripes - 1;
500ceed8
NB		 4740
4741 if (type & BTRFS_BLOCK_GROUP_RAID6)
53b381b3 4742 data_stripes = num_stripes - 2;
86db2578
CM		 4743
4744 /*
4745 * Use the number of data stripes to figure out how big this chunk
4746 * is really going to be in terms of logical address space,
4747 * and compare that answer with the max chunk size
4748 */
4749 if (stripe_size * data_stripes > max_chunk_size) {
4750 u64 mask = (1ULL << 24) - 1;
b8b93add
DS		 4751
4752 stripe_size = div_u64(max_chunk_size, data_stripes);
86db2578
CM		 4753
4754 /* bump the answer up to a 16MB boundary */
4755 stripe_size = (stripe_size + mask) & ~mask;
4756
4757 /* but don't go higher than the limits we found
4758 * while searching for free extents
4759 */
4760 if (stripe_size > devices_info[ndevs-1].max_avail)
4761 stripe_size = devices_info[ndevs-1].max_avail;
4762 }
4763
b8b93add 4764 stripe_size = div_u64(stripe_size, dev_stripes);
37db63a4
ID		 4765
4766 /* align to BTRFS_STRIPE_LEN */
500ceed8 4767 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
b2117a39
MX		 4768
4769 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4770 if (!map) {
4771 ret = -ENOMEM;
4772 goto error;
4773 }
4774 map->num_stripes = num_stripes;
9b3f68b9 4775
73c5de00
AJ		 4776 for (i = 0; i < ndevs; ++i) {
4777 for (j = 0; j < dev_stripes; ++j) {
4778 int s = i * dev_stripes + j;
4779 map->stripes[s].dev = devices_info[i].dev;
4780 map->stripes[s].physical = devices_info[i].dev_offset +
4781 j * stripe_size;
6324fbf3 4782 }
6324fbf3 4783 }
500ceed8
NB		 4784 map->stripe_len = BTRFS_STRIPE_LEN;
4785 map->io_align = BTRFS_STRIPE_LEN;
4786 map->io_width = BTRFS_STRIPE_LEN;
2b82032c 4787 map->type = type;
2b82032c 4788 map->sub_stripes = sub_stripes;
0b86a832 4789
53b381b3 4790 num_bytes = stripe_size * data_stripes;
0b86a832 4791
6bccf3ab 4792 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
1abe9b8a 4793
172ddd60 4794 em = alloc_extent_map();
2b82032c 4795 if (!em) {
298a8f9c 4796 kfree(map);
b2117a39
MX		 4797 ret = -ENOMEM;
4798 goto error;
593060d7 4799 }
298a8f9c 4800 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
95617d69 4801 em->map_lookup = map;
2b82032c 4802 em->start = start;
73c5de00 4803 em->len = num_bytes;
2b82032c
YZ		 4804 em->block_start = 0;
4805 em->block_len = em->len;
6df9a95e 4806 em->orig_block_len = stripe_size;
593060d7 4807
0b246afa 4808 em_tree = &info->mapping_tree.map_tree;
890871be 4809 write_lock(&em_tree->lock);
09a2a8f9 4810 ret = add_extent_mapping(em_tree, em, 0);
0f5d42b2 4811 if (ret) {
1efb72a3 4812 write_unlock(&em_tree->lock);
0f5d42b2 4813 free_extent_map(em);
1dd4602f 4814 goto error;
0f5d42b2 4815 }
0b86a832 4816
1efb72a3
NB		 4817 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4818 refcount_inc(&em->refs);
4819 write_unlock(&em_tree->lock);
4820
0174484d 4821 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
6df9a95e
JB		 4822 if (ret)
4823 goto error_del_extent;
2b82032c 4824
7cc8e58d
MX		 4825 for (i = 0; i < map->num_stripes; i++) {
4826 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4827 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4828 }
43530c46 4829
a5ed45f8 4830 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
1c116187 4831
0f5d42b2 4832 free_extent_map(em);
0b246afa 4833 check_raid56_incompat_flag(info, type);
53b381b3 4834
b2117a39 4835 kfree(devices_info);
2b82032c 4836 return 0;
b2117a39 4837
6df9a95e 4838error_del_extent:
0f5d42b2
JB		 4839 write_lock(&em_tree->lock);
4840 remove_extent_mapping(em_tree, em);
4841 write_unlock(&em_tree->lock);
4842
4843 /* One for our allocation */
4844 free_extent_map(em);
4845 /* One for the tree reference */
4846 free_extent_map(em);
495e64f4
FM		 4847 /* One for the pending_chunks list reference */
4848 free_extent_map(em);
b2117a39 4849error:
b2117a39
MX		 4850 kfree(devices_info);
4851 return ret;
2b82032c
YZ		 4852}
4853
6df9a95e 4854int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
6bccf3ab 4855 struct btrfs_fs_info *fs_info,
6df9a95e 4856 u64 chunk_offset, u64 chunk_size)
2b82032c 4857{
6bccf3ab
JM		 4858 struct btrfs_root *extent_root = fs_info->extent_root;
4859 struct btrfs_root *chunk_root = fs_info->chunk_root;
2b82032c 4860 struct btrfs_key key;
2b82032c
YZ		 4861 struct btrfs_device *device;
4862 struct btrfs_chunk *chunk;
4863 struct btrfs_stripe *stripe;
6df9a95e
JB		 4864 struct extent_map *em;
4865 struct map_lookup *map;
4866 size_t item_size;
4867 u64 dev_offset;
4868 u64 stripe_size;
4869 int i = 0;
140e639f 4870 int ret = 0;
2b82032c 4871
592d92ee
LB		 4872 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4873 if (IS_ERR(em))
4874 return PTR_ERR(em);
6df9a95e 4875
95617d69 4876 map = em->map_lookup;
6df9a95e
JB		 4877 item_size = btrfs_chunk_item_size(map->num_stripes);
4878 stripe_size = em->orig_block_len;
4879
2b82032c 4880 chunk = kzalloc(item_size, GFP_NOFS);
6df9a95e
JB		 4881 if (!chunk) {
4882 ret = -ENOMEM;
4883 goto out;
4884 }
4885
50460e37
FM		 4886 /*
4887 * Take the device list mutex to prevent races with the final phase of
4888 * a device replace operation that replaces the device object associated
4889 * with the map's stripes, because the device object's id can change
4890 * at any time during that final phase of the device replace operation
4891 * (dev-replace.c:btrfs_dev_replace_finishing()).
4892 */
0b246afa 4893 mutex_lock(&fs_info->fs_devices->device_list_mutex);
6df9a95e
JB		 4894 for (i = 0; i < map->num_stripes; i++) {
4895 device = map->stripes[i].dev;
4896 dev_offset = map->stripes[i].physical;
2b82032c 4897
0b86a832 4898 ret = btrfs_update_device(trans, device);
3acd3953 4899 if (ret)
50460e37 4900 break;
b5d9071c
NB		 4901 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4902 dev_offset, stripe_size);
6df9a95e 4903 if (ret)
50460e37
FM		 4904 break;
4905 }
4906 if (ret) {
0b246afa 4907 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
50460e37 4908 goto out;
2b82032c
YZ		 4909 }
4910
2b82032c 4911 stripe = &chunk->stripe;
6df9a95e
JB		 4912 for (i = 0; i < map->num_stripes; i++) {
4913 device = map->stripes[i].dev;
4914 dev_offset = map->stripes[i].physical;
0b86a832 4915
e17cade2
CM		 4916 btrfs_set_stack_stripe_devid(stripe, device->devid);
4917 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4918 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2b82032c 4919 stripe++;
0b86a832 4920 }
0b246afa 4921 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
0b86a832 4922
2b82032c 4923 btrfs_set_stack_chunk_length(chunk, chunk_size);
0b86a832 4924 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2b82032c
YZ		 4925 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4926 btrfs_set_stack_chunk_type(chunk, map->type);
4927 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4928 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4929 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
0b246afa 4930 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
2b82032c 4931 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
0b86a832 4932
2b82032c
YZ		 4933 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4934 key.type = BTRFS_CHUNK_ITEM_KEY;
4935 key.offset = chunk_offset;
0b86a832 4936
2b82032c 4937 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4ed1d16e
MF		 4938 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4939 /*
4940 * TODO: Cleanup of inserted chunk root in case of
4941 * failure.
4942 */
2ff7e61e 4943 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
8f18cf13 4944 }
1abe9b8a 4945
6df9a95e 4946out:
0b86a832 4947 kfree(chunk);
6df9a95e 4948 free_extent_map(em);
4ed1d16e 4949 return ret;
2b82032c 4950}
0b86a832 4951
2b82032c
YZ		 4952/*
4953 * Chunk allocation falls into two parts. The first part does works
4954 * that make the new allocated chunk useable, but not do any operation
4955 * that modifies the chunk tree. The second part does the works that
4956 * require modifying the chunk tree. This division is important for the
4957 * bootstrap process of adding storage to a seed btrfs.
4958 */
4959int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2ff7e61e 4960 struct btrfs_fs_info *fs_info, u64 type)
2b82032c
YZ		 4961{
4962 u64 chunk_offset;
2b82032c 4963
0b246afa
JM		 4964 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4965 chunk_offset = find_next_chunk(fs_info);
72b468c8 4966 return __btrfs_alloc_chunk(trans, chunk_offset, type);
2b82032c
YZ		 4967}
4968
d397712b 4969static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
e4a4dce7 4970 struct btrfs_fs_info *fs_info)
2b82032c
YZ		 4971{
4972 u64 chunk_offset;
4973 u64 sys_chunk_offset;
2b82032c 4974 u64 alloc_profile;
2b82032c
YZ		 4975 int ret;
4976
6df9a95e 4977 chunk_offset = find_next_chunk(fs_info);
1b86826d 4978 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
72b468c8 4979 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
79787eaa
JM		 4980 if (ret)
4981 return ret;
2b82032c 4982
0b246afa 4983 sys_chunk_offset = find_next_chunk(fs_info);
1b86826d 4984 alloc_profile = btrfs_system_alloc_profile(fs_info);
72b468c8 4985 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
79787eaa 4986 return ret;
2b82032c
YZ		 4987}
4988
d20983b4
MX		 4989static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4990{
4991 int max_errors;
4992
4993 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4994 BTRFS_BLOCK_GROUP_RAID10 |
4995 BTRFS_BLOCK_GROUP_RAID5 |
4996 BTRFS_BLOCK_GROUP_DUP)) {
4997 max_errors = 1;
4998 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4999 max_errors = 2;
5000 } else {
5001 max_errors = 0;
005d6427 5002 }
2b82032c 5003
d20983b4 5004 return max_errors;
2b82032c
YZ		 5005}
5006
2ff7e61e 5007int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2b82032c
YZ		 5008{
5009 struct extent_map *em;
5010 struct map_lookup *map;
2b82032c 5011 int readonly = 0;
d20983b4 5012 int miss_ndevs = 0;
2b82032c
YZ		 5013 int i;
5014
592d92ee
LB		 5015 em = get_chunk_map(fs_info, chunk_offset, 1);
5016 if (IS_ERR(em))
2b82032c
YZ		 5017 return 1;
5018
95617d69 5019 map = em->map_lookup;
2b82032c 5020 for (i = 0; i < map->num_stripes; i++) {
d20983b4
MX		 5021 if (map->stripes[i].dev->missing) {
5022 miss_ndevs++;
5023 continue;
5024 }
5025
2b82032c
YZ		 5026 if (!map->stripes[i].dev->writeable) {
5027 readonly = 1;
d20983b4 5028 goto end;
2b82032c
YZ		 5029 }
5030 }
d20983b4
MX		 5031
5032 /*
5033 * If the number of missing devices is larger than max errors,
5034 * we can not write the data into that chunk successfully, so
5035 * set it readonly.
5036 */
5037 if (miss_ndevs > btrfs_chunk_max_errors(map))
5038 readonly = 1;
5039end:
0b86a832 5040 free_extent_map(em);
2b82032c 5041 return readonly;
0b86a832
CM		 5042}
5043
5044void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5045{
a8067e02 5046 extent_map_tree_init(&tree->map_tree);
0b86a832
CM		 5047}
5048
5049void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5050{
5051 struct extent_map *em;
5052
d397712b 5053 while (1) {
890871be 5054 write_lock(&tree->map_tree.lock);
0b86a832
CM		 5055 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5056 if (em)
5057 remove_extent_mapping(&tree->map_tree, em);
890871be 5058 write_unlock(&tree->map_tree.lock);
0b86a832
CM		 5059 if (!em)
5060 break;
0b86a832
CM		 5061 /* once for us */
5062 free_extent_map(em);
5063 /* once for the tree */
5064 free_extent_map(em);
5065 }
5066}
5067
5d964051 5068int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
f188591e
CM		 5069{
5070 struct extent_map *em;
5071 struct map_lookup *map;
f188591e
CM		 5072 int ret;
5073
592d92ee
LB		 5074 em = get_chunk_map(fs_info, logical, len);
5075 if (IS_ERR(em))
5076 /*
5077 * We could return errors for these cases, but that could get
5078 * ugly and we'd probably do the same thing which is just not do
5079 * anything else and exit, so return 1 so the callers don't try
5080 * to use other copies.
5081 */
fb7669b5 5082 return 1;
fb7669b5 5083
95617d69 5084 map = em->map_lookup;
f188591e
CM		 5085 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5086 ret = map->num_stripes;
321aecc6
CM		 5087 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5088 ret = map->sub_stripes;
53b381b3
DW		 5089 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5090 ret = 2;
5091 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5092 ret = 3;
f188591e
CM		 5093 else
5094 ret = 1;
5095 free_extent_map(em);
ad6d620e 5096
73beece9 5097 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
6fad823f
LB		 5098 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5099 fs_info->dev_replace.tgtdev)
ad6d620e 5100 ret++;
73beece9 5101 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
ad6d620e 5102
f188591e
CM		 5103 return ret;
5104}
5105
2ff7e61e 5106unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
53b381b3
DW		 5107 u64 logical)
5108{
5109 struct extent_map *em;
5110 struct map_lookup *map;
0b246afa 5111 unsigned long len = fs_info->sectorsize;
53b381b3 5112
592d92ee 5113 em = get_chunk_map(fs_info, logical, len);
53b381b3 5114
69f03f13
NB		 5115 if (!WARN_ON(IS_ERR(em))) {
5116 map = em->map_lookup;
5117 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5118 len = map->stripe_len * nr_data_stripes(map);
5119 free_extent_map(em);
5120 }
53b381b3
DW		 5121 return len;
5122}
5123
e4ff5fb5 5124int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
53b381b3
DW		 5125{
5126 struct extent_map *em;
5127 struct map_lookup *map;
53b381b3
DW		 5128 int ret = 0;
5129
592d92ee 5130 em = get_chunk_map(fs_info, logical, len);
53b381b3 5131
69f03f13
NB		 5132 if(!WARN_ON(IS_ERR(em))) {
5133 map = em->map_lookup;
5134 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5135 ret = 1;
5136 free_extent_map(em);
5137 }
53b381b3
DW		 5138 return ret;
5139}
5140
30d9861f
SB		 5141static int find_live_mirror(struct btrfs_fs_info *fs_info,
5142 struct map_lookup *map, int first, int num,
5143 int optimal, int dev_replace_is_ongoing)
dfe25020
CM		 5144{
5145 int i;
30d9861f
SB		 5146 int tolerance;
5147 struct btrfs_device *srcdev;
5148
5149 if (dev_replace_is_ongoing &&
5150 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5151 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5152 srcdev = fs_info->dev_replace.srcdev;
5153 else
5154 srcdev = NULL;
5155
5156 /*
5157 * try to avoid the drive that is the source drive for a
5158 * dev-replace procedure, only choose it if no other non-missing
5159 * mirror is available
5160 */
5161 for (tolerance = 0; tolerance < 2; tolerance++) {
5162 if (map->stripes[optimal].dev->bdev &&
5163 (tolerance || map->stripes[optimal].dev != srcdev))
5164 return optimal;
5165 for (i = first; i < first + num; i++) {
5166 if (map->stripes[i].dev->bdev &&
5167 (tolerance || map->stripes[i].dev != srcdev))
5168 return i;
5169 }
dfe25020 5170 }
30d9861f 5171
dfe25020
CM		 5172 /* we couldn't find one that doesn't fail. Just return something
5173 * and the io error handling code will clean up eventually
5174 */
5175 return optimal;
5176}
5177
53b381b3
DW		 5178static inline int parity_smaller(u64 a, u64 b)
5179{
5180 return a > b;
5181}
5182
5183/* Bubble-sort the stripe set to put the parity/syndrome stripes last */
8e5cfb55 5184static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
53b381b3
DW		 5185{
5186 struct btrfs_bio_stripe s;
5187 int i;
5188 u64 l;
5189 int again = 1;
5190
5191 while (again) {
5192 again = 0;
cc7539ed 5193 for (i = 0; i < num_stripes - 1; i++) {
8e5cfb55
ZL		 5194 if (parity_smaller(bbio->raid_map[i],
5195 bbio->raid_map[i+1])) {
53b381b3 5196 s = bbio->stripes[i];
8e5cfb55 5197 l = bbio->raid_map[i];
53b381b3 5198 bbio->stripes[i] = bbio->stripes[i+1];
8e5cfb55 5199 bbio->raid_map[i] = bbio->raid_map[i+1];
53b381b3 5200 bbio->stripes[i+1] = s;
8e5cfb55 5201 bbio->raid_map[i+1] = l;
2c8cdd6e 5202
53b381b3
DW		 5203 again = 1;
5204 }
5205 }
5206 }
5207}
5208
6e9606d2
ZL		 5209static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5210{
5211 struct btrfs_bio *bbio = kzalloc(
e57cf21e 5212 /* the size of the btrfs_bio */
6e9606d2 5213 sizeof(struct btrfs_bio) +
e57cf21e 5214 /* plus the variable array for the stripes */
6e9606d2 5215 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
e57cf21e 5216 /* plus the variable array for the tgt dev */
6e9606d2 5217 sizeof(int) * (real_stripes) +
e57cf21e
CM		 5218 /*
5219 * plus the raid_map, which includes both the tgt dev
5220 * and the stripes
5221 */
5222 sizeof(u64) * (total_stripes),
277fb5fc 5223 GFP_NOFS|__GFP_NOFAIL);
6e9606d2
ZL		 5224
5225 atomic_set(&bbio->error, 0);
140475ae 5226 refcount_set(&bbio->refs, 1);
6e9606d2
ZL		 5227
5228 return bbio;
5229}
5230
5231void btrfs_get_bbio(struct btrfs_bio *bbio)
5232{
140475ae
ER		 5233 WARN_ON(!refcount_read(&bbio->refs));
5234 refcount_inc(&bbio->refs);
6e9606d2
ZL		 5235}
5236
5237void btrfs_put_bbio(struct btrfs_bio *bbio)
5238{
5239 if (!bbio)
5240 return;
140475ae 5241 if (refcount_dec_and_test(&bbio->refs))
6e9606d2
ZL		 5242 kfree(bbio);
5243}
5244
0b3d4cd3
LB		 5245/* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5246/*
5247 * Please note that, discard won't be sent to target device of device
5248 * replace.
5249 */
5250static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5251 u64 logical, u64 length,
5252 struct btrfs_bio **bbio_ret)
5253{
5254 struct extent_map *em;
5255 struct map_lookup *map;
5256 struct btrfs_bio *bbio;
5257 u64 offset;
5258 u64 stripe_nr;
5259 u64 stripe_nr_end;
5260 u64 stripe_end_offset;
5261 u64 stripe_cnt;
5262 u64 stripe_len;
5263 u64 stripe_offset;
5264 u64 num_stripes;
5265 u32 stripe_index;
5266 u32 factor = 0;
5267 u32 sub_stripes = 0;
5268 u64 stripes_per_dev = 0;
5269 u32 remaining_stripes = 0;
5270 u32 last_stripe = 0;
5271 int ret = 0;
5272 int i;
5273
5274 /* discard always return a bbio */
5275 ASSERT(bbio_ret);
5276
5277 em = get_chunk_map(fs_info, logical, length);
5278 if (IS_ERR(em))
5279 return PTR_ERR(em);
5280
5281 map = em->map_lookup;
5282 /* we don't discard raid56 yet */
5283 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5284 ret = -EOPNOTSUPP;
5285 goto out;
5286 }
5287
5288 offset = logical - em->start;
5289 length = min_t(u64, em->len - offset, length);
5290
5291 stripe_len = map->stripe_len;
5292 /*
5293 * stripe_nr counts the total number of stripes we have to stride
5294 * to get to this block
5295 */
5296 stripe_nr = div64_u64(offset, stripe_len);
5297
5298 /* stripe_offset is the offset of this block in its stripe */
5299 stripe_offset = offset - stripe_nr * stripe_len;
5300
5301 stripe_nr_end = round_up(offset + length, map->stripe_len);
42c61ab6 5302 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
0b3d4cd3
LB		 5303 stripe_cnt = stripe_nr_end - stripe_nr;
5304 stripe_end_offset = stripe_nr_end * map->stripe_len -
5305 (offset + length);
5306 /*
5307 * after this, stripe_nr is the number of stripes on this
5308 * device we have to walk to find the data, and stripe_index is
5309 * the number of our device in the stripe array
5310 */
5311 num_stripes = 1;
5312 stripe_index = 0;
5313 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5314 BTRFS_BLOCK_GROUP_RAID10)) {
5315 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5316 sub_stripes = 1;
5317 else
5318 sub_stripes = map->sub_stripes;
5319
5320 factor = map->num_stripes / sub_stripes;
5321 num_stripes = min_t(u64, map->num_stripes,
5322 sub_stripes * stripe_cnt);
5323 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5324 stripe_index *= sub_stripes;
5325 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5326 &remaining_stripes);
5327 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5328 last_stripe *= sub_stripes;
5329 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5330 BTRFS_BLOCK_GROUP_DUP)) {
5331 num_stripes = map->num_stripes;
5332 } else {
5333 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5334 &stripe_index);
5335 }
5336
5337 bbio = alloc_btrfs_bio(num_stripes, 0);
5338 if (!bbio) {
5339 ret = -ENOMEM;
5340 goto out;
5341 }
5342
5343 for (i = 0; i < num_stripes; i++) {
5344 bbio->stripes[i].physical =
5345 map->stripes[stripe_index].physical +
5346 stripe_offset + stripe_nr * map->stripe_len;
5347 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5348
5349 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5350 BTRFS_BLOCK_GROUP_RAID10)) {
5351 bbio->stripes[i].length = stripes_per_dev *
5352 map->stripe_len;
5353
5354 if (i / sub_stripes < remaining_stripes)
5355 bbio->stripes[i].length +=
5356 map->stripe_len;
5357
5358 /*
5359 * Special for the first stripe and
5360 * the last stripe:
5361 *
5362 * |-------|...|-------|
5363 * |----------|
5364 * off end_off
5365 */
5366 if (i < sub_stripes)
5367 bbio->stripes[i].length -=
5368 stripe_offset;
5369
5370 if (stripe_index >= last_stripe &&
5371 stripe_index <= (last_stripe +
5372 sub_stripes - 1))
5373 bbio->stripes[i].length -=
5374 stripe_end_offset;
5375
5376 if (i == sub_stripes - 1)
5377 stripe_offset = 0;
5378 } else {
5379 bbio->stripes[i].length = length;
5380 }
5381
5382 stripe_index++;
5383 if (stripe_index == map->num_stripes) {
5384 stripe_index = 0;
5385 stripe_nr++;
5386 }
5387 }
5388
5389 *bbio_ret = bbio;
5390 bbio->map_type = map->type;
5391 bbio->num_stripes = num_stripes;
5392out:
5393 free_extent_map(em);
5394 return ret;
5395}
5396
5ab56090
LB		 5397/*
5398 * In dev-replace case, for repair case (that's the only case where the mirror
5399 * is selected explicitly when calling btrfs_map_block), blocks left of the
5400 * left cursor can also be read from the target drive.
5401 *
5402 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5403 * array of stripes.
5404 * For READ, it also needs to be supported using the same mirror number.
5405 *
5406 * If the requested block is not left of the left cursor, EIO is returned. This
5407 * can happen because btrfs_num_copies() returns one more in the dev-replace
5408 * case.
5409 */
5410static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5411 u64 logical, u64 length,
5412 u64 srcdev_devid, int *mirror_num,
5413 u64 *physical)
5414{
5415 struct btrfs_bio *bbio = NULL;
5416 int num_stripes;
5417 int index_srcdev = 0;
5418 int found = 0;
5419 u64 physical_of_found = 0;
5420 int i;
5421 int ret = 0;
5422
5423 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5424 logical, &length, &bbio, 0, 0);
5425 if (ret) {
5426 ASSERT(bbio == NULL);
5427 return ret;
5428 }
5429
5430 num_stripes = bbio->num_stripes;
5431 if (*mirror_num > num_stripes) {
5432 /*
5433 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5434 * that means that the requested area is not left of the left
5435 * cursor
5436 */
5437 btrfs_put_bbio(bbio);
5438 return -EIO;
5439 }
5440
5441 /*
5442 * process the rest of the function using the mirror_num of the source
5443 * drive. Therefore look it up first. At the end, patch the device
5444 * pointer to the one of the target drive.
5445 */
5446 for (i = 0; i < num_stripes; i++) {
5447 if (bbio->stripes[i].dev->devid != srcdev_devid)
5448 continue;
5449
5450 /*
5451 * In case of DUP, in order to keep it simple, only add the
5452 * mirror with the lowest physical address
5453 */
5454 if (found &&
5455 physical_of_found <= bbio->stripes[i].physical)
5456 continue;
5457
5458 index_srcdev = i;
5459 found = 1;
5460 physical_of_found = bbio->stripes[i].physical;
5461 }
5462
5463 btrfs_put_bbio(bbio);
5464
5465 ASSERT(found);
5466 if (!found)
5467 return -EIO;
5468
5469 *mirror_num = index_srcdev + 1;
5470 *physical = physical_of_found;
5471 return ret;
5472}
5473
73c0f228
LB		 5474static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5475 struct btrfs_bio **bbio_ret,
5476 struct btrfs_dev_replace *dev_replace,
5477 int *num_stripes_ret, int *max_errors_ret)
5478{
5479 struct btrfs_bio *bbio = *bbio_ret;
5480 u64 srcdev_devid = dev_replace->srcdev->devid;
5481 int tgtdev_indexes = 0;
5482 int num_stripes = *num_stripes_ret;
5483 int max_errors = *max_errors_ret;
5484 int i;
5485
5486 if (op == BTRFS_MAP_WRITE) {
5487 int index_where_to_add;
5488
5489 /*
5490 * duplicate the write operations while the dev replace
5491 * procedure is running. Since the copying of the old disk to
5492 * the new disk takes place at run time while the filesystem is
5493 * mounted writable, the regular write operations to the old
5494 * disk have to be duplicated to go to the new disk as well.
5495 *
5496 * Note that device->missing is handled by the caller, and that
5497 * the write to the old disk is already set up in the stripes
5498 * array.
5499 */
5500 index_where_to_add = num_stripes;
5501 for (i = 0; i < num_stripes; i++) {
5502 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5503 /* write to new disk, too */
5504 struct btrfs_bio_stripe *new =
5505 bbio->stripes + index_where_to_add;
5506 struct btrfs_bio_stripe *old =
5507 bbio->stripes + i;
5508
5509 new->physical = old->physical;
5510 new->length = old->length;
5511 new->dev = dev_replace->tgtdev;
5512 bbio->tgtdev_map[i] = index_where_to_add;
5513 index_where_to_add++;
5514 max_errors++;
5515 tgtdev_indexes++;
5516 }
5517 }
5518 num_stripes = index_where_to_add;
5519 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5520 int index_srcdev = 0;
5521 int found = 0;
5522 u64 physical_of_found = 0;
5523
5524 /*
5525 * During the dev-replace procedure, the target drive can also
5526 * be used to read data in case it is needed to repair a corrupt
5527 * block elsewhere. This is possible if the requested area is
5528 * left of the left cursor. In this area, the target drive is a
5529 * full copy of the source drive.
5530 */
5531 for (i = 0; i < num_stripes; i++) {
5532 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5533 /*
5534 * In case of DUP, in order to keep it simple,
5535 * only add the mirror with the lowest physical
5536 * address
5537 */
5538 if (found &&
5539 physical_of_found <=
5540 bbio->stripes[i].physical)
5541 continue;
5542 index_srcdev = i;
5543 found = 1;
5544 physical_of_found = bbio->stripes[i].physical;
5545 }
5546 }
5547 if (found) {
5548 struct btrfs_bio_stripe *tgtdev_stripe =
5549 bbio->stripes + num_stripes;
5550
5551 tgtdev_stripe->physical = physical_of_found;
5552 tgtdev_stripe->length =
5553 bbio->stripes[index_srcdev].length;
5554 tgtdev_stripe->dev = dev_replace->tgtdev;
5555 bbio->tgtdev_map[index_srcdev] = num_stripes;
5556
5557 tgtdev_indexes++;
5558 num_stripes++;
5559 }
5560 }
5561
5562 *num_stripes_ret = num_stripes;
5563 *max_errors_ret = max_errors;
5564 bbio->num_tgtdevs = tgtdev_indexes;
5565 *bbio_ret = bbio;
5566}
5567
2b19a1fe
LB		 5568static bool need_full_stripe(enum btrfs_map_op op)
5569{
5570 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5571}
5572
cf8cddd3
CH		 5573static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5574 enum btrfs_map_op op,
f2d8d74d 5575 u64 logical, u64 *length,
a1d3c478 5576 struct btrfs_bio **bbio_ret,
8e5cfb55 5577 int mirror_num, int need_raid_map)
0b86a832
CM		 5578{
5579 struct extent_map *em;
5580 struct map_lookup *map;
0b86a832 5581 u64 offset;
593060d7
CM		 5582 u64 stripe_offset;
5583 u64 stripe_nr;
53b381b3 5584 u64 stripe_len;
9d644a62 5585 u32 stripe_index;
cea9e445 5586 int i;
de11cc12 5587 int ret = 0;
f2d8d74d 5588 int num_stripes;
a236aed1 5589 int max_errors = 0;
2c8cdd6e 5590 int tgtdev_indexes = 0;
a1d3c478 5591 struct btrfs_bio *bbio = NULL;
472262f3
SB		 5592 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5593 int dev_replace_is_ongoing = 0;
5594 int num_alloc_stripes;
ad6d620e
SB		 5595 int patch_the_first_stripe_for_dev_replace = 0;
5596 u64 physical_to_patch_in_first_stripe = 0;
53b381b3 5597 u64 raid56_full_stripe_start = (u64)-1;
0b86a832 5598
0b3d4cd3
LB		 5599 if (op == BTRFS_MAP_DISCARD)
5600 return __btrfs_map_block_for_discard(fs_info, logical,
5601 *length, bbio_ret);
5602
592d92ee
LB		 5603 em = get_chunk_map(fs_info, logical, *length);
5604 if (IS_ERR(em))
5605 return PTR_ERR(em);
0b86a832 5606
95617d69 5607 map = em->map_lookup;
0b86a832 5608 offset = logical - em->start;
593060d7 5609
53b381b3 5610 stripe_len = map->stripe_len;
593060d7
CM		 5611 stripe_nr = offset;
5612 /*
5613 * stripe_nr counts the total number of stripes we have to stride
5614 * to get to this block
5615 */
47c5713f 5616 stripe_nr = div64_u64(stripe_nr, stripe_len);
593060d7 5617
53b381b3 5618 stripe_offset = stripe_nr * stripe_len;
e042d1ec 5619 if (offset < stripe_offset) {
5d163e0e
JM		 5620 btrfs_crit(fs_info,
5621 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
e042d1ec
JB		 5622 stripe_offset, offset, em->start, logical,
5623 stripe_len);
5624 free_extent_map(em);
5625 return -EINVAL;
5626 }
593060d7
CM		 5627
5628 /* stripe_offset is the offset of this block in its stripe*/
5629 stripe_offset = offset - stripe_offset;
5630
53b381b3 5631 /* if we're here for raid56, we need to know the stripe aligned start */
ffe2d203 5632 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
53b381b3
DW		 5633 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5634 raid56_full_stripe_start = offset;
5635
5636 /* allow a write of a full stripe, but make sure we don't
5637 * allow straddling of stripes
5638 */
47c5713f
DS		 5639 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5640 full_stripe_len);
53b381b3
DW		 5641 raid56_full_stripe_start *= full_stripe_len;
5642 }
5643
0b3d4cd3 5644 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
53b381b3
DW		 5645 u64 max_len;
5646 /* For writes to RAID[56], allow a full stripeset across all disks.
5647 For other RAID types and for RAID[56] reads, just allow a single
5648 stripe (on a single disk). */
ffe2d203 5649 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
cf8cddd3 5650 (op == BTRFS_MAP_WRITE)) {
53b381b3
DW		 5651 max_len = stripe_len * nr_data_stripes(map) -
5652 (offset - raid56_full_stripe_start);
5653 } else {
5654 /* we limit the length of each bio to what fits in a stripe */
5655 max_len = stripe_len - stripe_offset;
5656 }
5657 *length = min_t(u64, em->len - offset, max_len);
cea9e445
CM		 5658 } else {
5659 *length = em->len - offset;
5660 }
f2d8d74d 5661
53b381b3
DW		 5662 /* This is for when we're called from btrfs_merge_bio_hook() and all
5663 it cares about is the length */
a1d3c478 5664 if (!bbio_ret)
cea9e445
CM		 5665 goto out;
5666
73beece9 5667 btrfs_dev_replace_lock(dev_replace, 0);
472262f3
SB		 5668 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5669 if (!dev_replace_is_ongoing)
73beece9
LB		 5670 btrfs_dev_replace_unlock(dev_replace, 0);
5671 else
5672 btrfs_dev_replace_set_lock_blocking(dev_replace);
472262f3 5673
ad6d620e 5674 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
2b19a1fe 5675 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5ab56090
LB		 5676 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5677 dev_replace->srcdev->devid,
5678 &mirror_num,
5679 &physical_to_patch_in_first_stripe);
5680 if (ret)
ad6d620e 5681 goto out;
5ab56090
LB		 5682 else
5683 patch_the_first_stripe_for_dev_replace = 1;
ad6d620e
SB		 5684 } else if (mirror_num > map->num_stripes) {
5685 mirror_num = 0;
5686 }
5687
f2d8d74d 5688 num_stripes = 1;
cea9e445 5689 stripe_index = 0;
fce3bb9a 5690 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
47c5713f
DS		 5691 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5692 &stripe_index);
0b3d4cd3 5693 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
28e1cc7d 5694 mirror_num = 1;
fce3bb9a 5695 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
0b3d4cd3 5696 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
f2d8d74d 5697 num_stripes = map->num_stripes;
2fff734f 5698 else if (mirror_num)
f188591e 5699 stripe_index = mirror_num - 1;
dfe25020 5700 else {
30d9861f 5701 stripe_index = find_live_mirror(fs_info, map, 0,
dfe25020 5702 map->num_stripes,
30d9861f
SB		 5703 current->pid % map->num_stripes,
5704 dev_replace_is_ongoing);
a1d3c478 5705 mirror_num = stripe_index + 1;
dfe25020 5706 }
2fff734f 5707
611f0e00 5708 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
0b3d4cd3 5709 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
f2d8d74d 5710 num_stripes = map->num_stripes;
a1d3c478 5711 } else if (mirror_num) {
f188591e 5712 stripe_index = mirror_num - 1;
a1d3c478
JS		 5713 } else {
5714 mirror_num = 1;
5715 }
2fff734f 5716
321aecc6 5717 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
9d644a62 5718 u32 factor = map->num_stripes / map->sub_stripes;
321aecc6 5719
47c5713f 5720 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
321aecc6
CM		 5721 stripe_index *= map->sub_stripes;
5722
cf8cddd3 5723 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
f2d8d74d 5724 num_stripes = map->sub_stripes;
321aecc6
CM		 5725 else if (mirror_num)
5726 stripe_index += mirror_num - 1;
dfe25020 5727 else {
3e74317a 5728 int old_stripe_index = stripe_index;
30d9861f
SB		 5729 stripe_index = find_live_mirror(fs_info, map,
5730 stripe_index,
dfe25020 5731 map->sub_stripes, stripe_index +
30d9861f
SB		 5732 current->pid % map->sub_stripes,
5733 dev_replace_is_ongoing);
3e74317a 5734 mirror_num = stripe_index - old_stripe_index + 1;
dfe25020 5735 }
53b381b3 5736
ffe2d203 5737 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8e5cfb55 5738 if (need_raid_map &&
cf8cddd3 5739 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
af8e2d1d 5740 mirror_num > 1)) {
53b381b3 5741 /* push stripe_nr back to the start of the full stripe */
42c61ab6 5742 stripe_nr = div64_u64(raid56_full_stripe_start,
b8b93add 5743 stripe_len * nr_data_stripes(map));
53b381b3
DW		 5744
5745 /* RAID[56] write or recovery. Return all stripes */
5746 num_stripes = map->num_stripes;
5747 max_errors = nr_parity_stripes(map);
5748
53b381b3
DW		 5749 *length = map->stripe_len;
5750 stripe_index = 0;
5751 stripe_offset = 0;
5752 } else {
5753 /*
5754 * Mirror #0 or #1 means the original data block.
5755 * Mirror #2 is RAID5 parity block.
5756 * Mirror #3 is RAID6 Q block.
5757 */
47c5713f
DS		 5758 stripe_nr = div_u64_rem(stripe_nr,
5759 nr_data_stripes(map), &stripe_index);
53b381b3
DW		 5760 if (mirror_num > 1)
5761 stripe_index = nr_data_stripes(map) +
5762 mirror_num - 2;
5763
5764 /* We distribute the parity blocks across stripes */
47c5713f
DS		 5765 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5766 &stripe_index);
0b3d4cd3
LB		 5767 if ((op != BTRFS_MAP_WRITE &&
5768 op != BTRFS_MAP_GET_READ_MIRRORS) &&
5769 mirror_num <= 1)
28e1cc7d 5770 mirror_num = 1;
53b381b3 5771 }
8790d502
CM		 5772 } else {
5773 /*
47c5713f
DS		 5774 * after this, stripe_nr is the number of stripes on this
5775 * device we have to walk to find the data, and stripe_index is
5776 * the number of our device in the stripe array
8790d502 5777 */
47c5713f
DS		 5778 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5779 &stripe_index);
a1d3c478 5780 mirror_num = stripe_index + 1;
8790d502 5781 }
e042d1ec 5782 if (stripe_index >= map->num_stripes) {
5d163e0e
JM		 5783 btrfs_crit(fs_info,
5784 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
e042d1ec
JB		 5785 stripe_index, map->num_stripes);
5786 ret = -EINVAL;
5787 goto out;
5788 }
cea9e445 5789
472262f3 5790 num_alloc_stripes = num_stripes;
6fad823f 5791 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
0b3d4cd3 5792 if (op == BTRFS_MAP_WRITE)
ad6d620e 5793 num_alloc_stripes <<= 1;
cf8cddd3 5794 if (op == BTRFS_MAP_GET_READ_MIRRORS)
ad6d620e 5795 num_alloc_stripes++;
2c8cdd6e 5796 tgtdev_indexes = num_stripes;
ad6d620e 5797 }
2c8cdd6e 5798
6e9606d2 5799 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
de11cc12
LZ		 5800 if (!bbio) {
5801 ret = -ENOMEM;
5802 goto out;
5803 }
6fad823f 5804 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
2c8cdd6e 5805 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
de11cc12 5806
8e5cfb55 5807 /* build raid_map */
2b19a1fe
LB		 5808 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5809 (need_full_stripe(op) || mirror_num > 1)) {
8e5cfb55 5810 u64 tmp;
9d644a62 5811 unsigned rot;
8e5cfb55
ZL		 5812
5813 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5814 sizeof(struct btrfs_bio_stripe) *
5815 num_alloc_stripes +
5816 sizeof(int) * tgtdev_indexes);
5817
5818 /* Work out the disk rotation on this stripe-set */
47c5713f 5819 div_u64_rem(stripe_nr, num_stripes, &rot);
8e5cfb55
ZL		 5820
5821 /* Fill in the logical address of each stripe */
5822 tmp = stripe_nr * nr_data_stripes(map);
5823 for (i = 0; i < nr_data_stripes(map); i++)
5824 bbio->raid_map[(i+rot) % num_stripes] =
5825 em->start + (tmp + i) * map->stripe_len;
5826
5827 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5828 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5829 bbio->raid_map[(i+rot+1) % num_stripes] =
5830 RAID6_Q_STRIPE;
5831 }
5832
b89203f7 5833
0b3d4cd3
LB		 5834 for (i = 0; i < num_stripes; i++) {
5835 bbio->stripes[i].physical =
5836 map->stripes[stripe_index].physical +
5837 stripe_offset +
5838 stripe_nr * map->stripe_len;
5839 bbio->stripes[i].dev =
5840 map->stripes[stripe_index].dev;
5841 stripe_index++;
593060d7 5842 }
de11cc12 5843
2b19a1fe 5844 if (need_full_stripe(op))
d20983b4 5845 max_errors = btrfs_chunk_max_errors(map);
de11cc12 5846
8e5cfb55
ZL		 5847 if (bbio->raid_map)
5848 sort_parity_stripes(bbio, num_stripes);
cc7539ed 5849
73c0f228 5850 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
2b19a1fe 5851 need_full_stripe(op)) {
73c0f228
LB		 5852 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5853 &max_errors);
472262f3
SB		 5854 }
5855
de11cc12 5856 *bbio_ret = bbio;
10f11900 5857 bbio->map_type = map->type;
de11cc12
LZ		 5858 bbio->num_stripes = num_stripes;
5859 bbio->max_errors = max_errors;
5860 bbio->mirror_num = mirror_num;
ad6d620e
SB		 5861
5862 /*
5863 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5864 * mirror_num == num_stripes + 1 && dev_replace target drive is
5865 * available as a mirror
5866 */
5867 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5868 WARN_ON(num_stripes > 1);
5869 bbio->stripes[0].dev = dev_replace->tgtdev;
5870 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5871 bbio->mirror_num = map->num_stripes + 1;
5872 }
cea9e445 5873out:
73beece9
LB		 5874 if (dev_replace_is_ongoing) {
5875 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5876 btrfs_dev_replace_unlock(dev_replace, 0);
5877 }
0b86a832 5878 free_extent_map(em);
de11cc12 5879 return ret;
0b86a832
CM		 5880}
5881
cf8cddd3 5882int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
f2d8d74d 5883 u64 logical, u64 *length,
a1d3c478 5884 struct btrfs_bio **bbio_ret, int mirror_num)
f2d8d74d 5885{
b3d3fa51 5886 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
8e5cfb55 5887 mirror_num, 0);
f2d8d74d
CM		 5888}
5889
af8e2d1d 5890/* For Scrub/replace */
cf8cddd3 5891int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
af8e2d1d 5892 u64 logical, u64 *length,
825ad4c9 5893 struct btrfs_bio **bbio_ret)
af8e2d1d 5894{
825ad4c9 5895 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
af8e2d1d
MX		 5896}
5897
ab8d0fc4 5898int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
a512bbf8
YZ		 5899 u64 chunk_start, u64 physical, u64 devid,
5900 u64 **logical, int *naddrs, int *stripe_len)
5901{
a512bbf8
YZ		 5902 struct extent_map *em;
5903 struct map_lookup *map;
5904 u64 *buf;
5905 u64 bytenr;
5906 u64 length;
5907 u64 stripe_nr;
53b381b3 5908 u64 rmap_len;
a512bbf8
YZ		 5909 int i, j, nr = 0;
5910
592d92ee
LB		 5911 em = get_chunk_map(fs_info, chunk_start, 1);
5912 if (IS_ERR(em))
835d974f 5913 return -EIO;
835d974f 5914
95617d69 5915 map = em->map_lookup;
a512bbf8 5916 length = em->len;
53b381b3
DW		 5917 rmap_len = map->stripe_len;
5918
a512bbf8 5919 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
b8b93add 5920 length = div_u64(length, map->num_stripes / map->sub_stripes);
a512bbf8 5921 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
b8b93add 5922 length = div_u64(length, map->num_stripes);
ffe2d203 5923 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
b8b93add 5924 length = div_u64(length, nr_data_stripes(map));
53b381b3
DW		 5925 rmap_len = map->stripe_len * nr_data_stripes(map);
5926 }
a512bbf8 5927
31e818fe 5928 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
79787eaa 5929 BUG_ON(!buf); /* -ENOMEM */
a512bbf8
YZ		 5930
5931 for (i = 0; i < map->num_stripes; i++) {
5932 if (devid && map->stripes[i].dev->devid != devid)
5933 continue;
5934 if (map->stripes[i].physical > physical ||
5935 map->stripes[i].physical + length <= physical)
5936 continue;
5937
5938 stripe_nr = physical - map->stripes[i].physical;
42c61ab6 5939 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
a512bbf8
YZ		 5940
5941 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5942 stripe_nr = stripe_nr * map->num_stripes + i;
b8b93add 5943 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
a512bbf8
YZ		 5944 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5945 stripe_nr = stripe_nr * map->num_stripes + i;
53b381b3
DW		 5946 } /* else if RAID[56], multiply by nr_data_stripes().
5947 * Alternatively, just use rmap_len below instead of
5948 * map->stripe_len */
5949
5950 bytenr = chunk_start + stripe_nr * rmap_len;
934d375b 5951 WARN_ON(nr >= map->num_stripes);
a512bbf8
YZ		 5952 for (j = 0; j < nr; j++) {
5953 if (buf[j] == bytenr)
5954 break;
5955 }
934d375b
CM		 5956 if (j == nr) {
5957 WARN_ON(nr >= map->num_stripes);
a512bbf8 5958 buf[nr++] = bytenr;
934d375b 5959 }
a512bbf8
YZ		 5960 }
5961
a512bbf8
YZ		 5962 *logical = buf;
5963 *naddrs = nr;
53b381b3 5964 *stripe_len = rmap_len;
a512bbf8
YZ		 5965
5966 free_extent_map(em);
5967 return 0;
f2d8d74d
CM		 5968}
5969
4246a0b6 5970static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
8408c716 5971{
326e1dbb
MS		 5972 bio->bi_private = bbio->private;
5973 bio->bi_end_io = bbio->end_io;
4246a0b6 5974 bio_endio(bio);
326e1dbb 5975
6e9606d2 5976 btrfs_put_bbio(bbio);
8408c716
MX		 5977}
5978
4246a0b6 5979static void btrfs_end_bio(struct bio *bio)
8790d502 5980{
9be3395b 5981 struct btrfs_bio *bbio = bio->bi_private;
7d2b4daa 5982 int is_orig_bio = 0;
8790d502 5983
4e4cbee9 5984 if (bio->bi_status) {
a1d3c478 5985 atomic_inc(&bbio->error);
4e4cbee9
CH		 5986 if (bio->bi_status == BLK_STS_IOERR ||
5987 bio->bi_status == BLK_STS_TARGET) {
442a4f63 5988 unsigned int stripe_index =
9be3395b 5989 btrfs_io_bio(bio)->stripe_index;
65f53338 5990 struct btrfs_device *dev;
442a4f63
SB		 5991
5992 BUG_ON(stripe_index >= bbio->num_stripes);
5993 dev = bbio->stripes[stripe_index].dev;
597a60fa 5994 if (dev->bdev) {
37226b21 5995 if (bio_op(bio) == REQ_OP_WRITE)
597a60fa
SB		 5996 btrfs_dev_stat_inc(dev,
5997 BTRFS_DEV_STAT_WRITE_ERRS);
5998 else
5999 btrfs_dev_stat_inc(dev,
6000 BTRFS_DEV_STAT_READ_ERRS);
70fd7614 6001 if (bio->bi_opf & REQ_PREFLUSH)
597a60fa
SB		 6002 btrfs_dev_stat_inc(dev,
6003 BTRFS_DEV_STAT_FLUSH_ERRS);
6004 btrfs_dev_stat_print_on_error(dev);
6005 }
442a4f63
SB		 6006 }
6007 }
8790d502 6008
a1d3c478 6009 if (bio == bbio->orig_bio)
7d2b4daa
CM		 6010 is_orig_bio = 1;
6011
c404e0dc
MX		 6012 btrfs_bio_counter_dec(bbio->fs_info);
6013
a1d3c478 6014 if (atomic_dec_and_test(&bbio->stripes_pending)) {
7d2b4daa
CM		 6015 if (!is_orig_bio) {
6016 bio_put(bio);
a1d3c478 6017 bio = bbio->orig_bio;
7d2b4daa 6018 }
c7b22bb1 6019
9be3395b 6020 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
a236aed1 6021 /* only send an error to the higher layers if it is
53b381b3 6022 * beyond the tolerance of the btrfs bio
a236aed1 6023 */
a1d3c478 6024 if (atomic_read(&bbio->error) > bbio->max_errors) {
4e4cbee9 6025 bio->bi_status = BLK_STS_IOERR;
5dbc8fca 6026 } else {
1259ab75
CM		 6027 /*
6028 * this bio is actually up to date, we didn't
6029 * go over the max number of errors
6030 */
4e4cbee9 6031 bio->bi_status = 0;
1259ab75 6032 }
c55f1396 6033
4246a0b6 6034 btrfs_end_bbio(bbio, bio);
7d2b4daa 6035 } else if (!is_orig_bio) {
8790d502
CM		 6036 bio_put(bio);
6037 }
8790d502
CM		 6038}
6039
8b712842
CM		 6040/*
6041 * see run_scheduled_bios for a description of why bios are collected for
6042 * async submit.
6043 *
6044 * This will add one bio to the pending list for a device and make sure
6045 * the work struct is scheduled.
6046 */
2ff7e61e 6047static noinline void btrfs_schedule_bio(struct btrfs_device *device,
4e49ea4a 6048 struct bio *bio)
8b712842 6049{
0b246afa 6050 struct btrfs_fs_info *fs_info = device->fs_info;
8b712842 6051 int should_queue = 1;
ffbd517d 6052 struct btrfs_pending_bios *pending_bios;
8b712842 6053
53b381b3 6054 if (device->missing || !device->bdev) {
4246a0b6 6055 bio_io_error(bio);
53b381b3
DW		 6056 return;
6057 }
6058
8b712842 6059 /* don't bother with additional async steps for reads, right now */
37226b21 6060 if (bio_op(bio) == REQ_OP_READ) {
492bb6de 6061 bio_get(bio);
4e49ea4a 6062 btrfsic_submit_bio(bio);
492bb6de 6063 bio_put(bio);
143bede5 6064 return;
8b712842
CM		 6065 }
6066
6067 /*
0986fe9e 6068 * nr_async_bios allows us to reliably return congestion to the
8b712842
CM		 6069 * higher layers. Otherwise, the async bio makes it appear we have
6070 * made progress against dirty pages when we've really just put it
6071 * on a queue for later
6072 */
0b246afa 6073 atomic_inc(&fs_info->nr_async_bios);
492bb6de 6074 WARN_ON(bio->bi_next);
8b712842 6075 bio->bi_next = NULL;
8b712842
CM		 6076
6077 spin_lock(&device->io_lock);
67f055c7 6078 if (op_is_sync(bio->bi_opf))
ffbd517d
CM		 6079 pending_bios = &device->pending_sync_bios;
6080 else
6081 pending_bios = &device->pending_bios;
8b712842 6082
ffbd517d
CM		 6083 if (pending_bios->tail)
6084 pending_bios->tail->bi_next = bio;
8b712842 6085
ffbd517d
CM		 6086 pending_bios->tail = bio;
6087 if (!pending_bios->head)
6088 pending_bios->head = bio;
8b712842
CM		 6089 if (device->running_pending)
6090 should_queue = 0;
6091
6092 spin_unlock(&device->io_lock);
6093
6094 if (should_queue)
0b246afa 6095 btrfs_queue_work(fs_info->submit_workers, &device->work);
8b712842
CM		 6096}
6097
2ff7e61e
JM		 6098static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6099 u64 physical, int dev_nr, int async)
de1ee92a
JB		 6100{
6101 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
2ff7e61e 6102 struct btrfs_fs_info *fs_info = bbio->fs_info;
de1ee92a
JB		 6103
6104 bio->bi_private = bbio;
9be3395b 6105 btrfs_io_bio(bio)->stripe_index = dev_nr;
de1ee92a 6106 bio->bi_end_io = btrfs_end_bio;
4f024f37 6107 bio->bi_iter.bi_sector = physical >> 9;
de1ee92a
JB		 6108#ifdef DEBUG
6109 {
6110 struct rcu_string *name;
6111
6112 rcu_read_lock();
6113 name = rcu_dereference(dev->name);
ab8d0fc4
JM		 6114 btrfs_debug(fs_info,
6115 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6116 bio_op(bio), bio->bi_opf,
6117 (u64)bio->bi_iter.bi_sector,
6118 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6119 bio->bi_iter.bi_size);
de1ee92a
JB		 6120 rcu_read_unlock();
6121 }
6122#endif
74d46992 6123 bio_set_dev(bio, dev->bdev);
c404e0dc 6124
2ff7e61e 6125 btrfs_bio_counter_inc_noblocked(fs_info);
c404e0dc 6126
de1ee92a 6127 if (async)
2ff7e61e 6128 btrfs_schedule_bio(dev, bio);
de1ee92a 6129 else
4e49ea4a 6130 btrfsic_submit_bio(bio);
de1ee92a
JB		 6131}
6132
de1ee92a
JB		 6133static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6134{
6135 atomic_inc(&bbio->error);
6136 if (atomic_dec_and_test(&bbio->stripes_pending)) {
01327610 6137 /* Should be the original bio. */
8408c716
MX		 6138 WARN_ON(bio != bbio->orig_bio);
6139
9be3395b 6140 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
4f024f37 6141 bio->bi_iter.bi_sector = logical >> 9;
4e4cbee9 6142 bio->bi_status = BLK_STS_IOERR;
4246a0b6 6143 btrfs_end_bbio(bbio, bio);
de1ee92a
JB		 6144 }
6145}
6146
58efbc9f
OS		 6147blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6148 int mirror_num, int async_submit)
0b86a832 6149{
0b86a832 6150 struct btrfs_device *dev;
8790d502 6151 struct bio *first_bio = bio;
4f024f37 6152 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
0b86a832
CM		 6153 u64 length = 0;
6154 u64 map_length;
0b86a832 6155 int ret;
08da757d
ZL		 6156 int dev_nr;
6157 int total_devs;
a1d3c478 6158 struct btrfs_bio *bbio = NULL;
0b86a832 6159
4f024f37 6160 length = bio->bi_iter.bi_size;
0b86a832 6161 map_length = length;
cea9e445 6162
0b246afa 6163 btrfs_bio_counter_inc_blocked(fs_info);
bd7d63c2 6164 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
37226b21 6165 &map_length, &bbio, mirror_num, 1);
c404e0dc 6166 if (ret) {
0b246afa 6167 btrfs_bio_counter_dec(fs_info);
58efbc9f 6168 return errno_to_blk_status(ret);
c404e0dc 6169 }
cea9e445 6170
a1d3c478 6171 total_devs = bbio->num_stripes;
53b381b3
DW		 6172 bbio->orig_bio = first_bio;
6173 bbio->private = first_bio->bi_private;
6174 bbio->end_io = first_bio->bi_end_io;
0b246afa 6175 bbio->fs_info = fs_info;
53b381b3
DW		 6176 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6177
ad1ba2a0 6178 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
37226b21 6179 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
53b381b3
DW		 6180 /* In this case, map_length has been set to the length of
6181 a single stripe; not the whole write */
37226b21 6182 if (bio_op(bio) == REQ_OP_WRITE) {
2ff7e61e
JM		 6183 ret = raid56_parity_write(fs_info, bio, bbio,
6184 map_length);
53b381b3 6185 } else {
2ff7e61e
JM		 6186 ret = raid56_parity_recover(fs_info, bio, bbio,
6187 map_length, mirror_num, 1);
53b381b3 6188 }
4245215d 6189
0b246afa 6190 btrfs_bio_counter_dec(fs_info);
58efbc9f 6191 return errno_to_blk_status(ret);
53b381b3
DW		 6192 }
6193
cea9e445 6194 if (map_length < length) {
0b246afa 6195 btrfs_crit(fs_info,
5d163e0e
JM		 6196 "mapping failed logical %llu bio len %llu len %llu",
6197 logical, length, map_length);
cea9e445
CM		 6198 BUG();
6199 }
a1d3c478 6200
08da757d 6201 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
de1ee92a 6202 dev = bbio->stripes[dev_nr].dev;
37226b21 6203 if (!dev || !dev->bdev ||
a967efb3 6204 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
de1ee92a 6205 bbio_error(bbio, first_bio, logical);
de1ee92a
JB		 6206 continue;
6207 }
6208
3aa8e074 6209 if (dev_nr < total_devs - 1)
8b6c1d56 6210 bio = btrfs_bio_clone(first_bio);
3aa8e074 6211 else
a1d3c478 6212 bio = first_bio;
de1ee92a 6213
2ff7e61e
JM		 6214 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6215 dev_nr, async_submit);
8790d502 6216 }
0b246afa 6217 btrfs_bio_counter_dec(fs_info);
58efbc9f 6218 return BLK_STS_OK;
0b86a832
CM		 6219}
6220
aa1b8cd4 6221struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
2b82032c 6222 u8 *uuid, u8 *fsid)
0b86a832 6223{
2b82032c
YZ		 6224 struct btrfs_device *device;
6225 struct btrfs_fs_devices *cur_devices;
6226
aa1b8cd4 6227 cur_devices = fs_info->fs_devices;
2b82032c
YZ		 6228 while (cur_devices) {
6229 if (!fsid ||
44880fdc 6230 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
35c70103 6231 device = find_device(cur_devices, devid, uuid);
2b82032c
YZ		 6232 if (device)
6233 return device;
6234 }
6235 cur_devices = cur_devices->seed;
6236 }
6237 return NULL;
0b86a832
CM		 6238}
6239
2ff7e61e 6240static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
dfe25020
CM		 6241 u64 devid, u8 *dev_uuid)
6242{
6243 struct btrfs_device *device;
dfe25020 6244
12bd2fc0
ID		 6245 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6246 if (IS_ERR(device))
7cbd8a83 6247 return NULL;
12bd2fc0
ID		 6248
6249 list_add(&device->dev_list, &fs_devices->devices);
e4404d6e 6250 device->fs_devices = fs_devices;
dfe25020 6251 fs_devices->num_devices++;
12bd2fc0
ID		 6252
6253 device->missing = 1;
cd02dca5 6254 fs_devices->missing_devices++;
12bd2fc0 6255
dfe25020
CM		 6256 return device;
6257}
6258
12bd2fc0
ID		 6259/**
6260 * btrfs_alloc_device - allocate struct btrfs_device
6261 * @fs_info: used only for generating a new devid, can be NULL if
6262 * devid is provided (i.e. @devid != NULL).
6263 * @devid: a pointer to devid for this device. If NULL a new devid
6264 * is generated.
6265 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6266 * is generated.
6267 *
6268 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6269 * on error. Returned struct is not linked onto any lists and can be
6270 * destroyed with kfree() right away.
6271 */
6272struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6273 const u64 *devid,
6274 const u8 *uuid)
6275{
6276 struct btrfs_device *dev;
6277 u64 tmp;
6278
fae7f21c 6279 if (WARN_ON(!devid && !fs_info))
12bd2fc0 6280 return ERR_PTR(-EINVAL);
12bd2fc0
ID		 6281
6282 dev = __alloc_device();
6283 if (IS_ERR(dev))
6284 return dev;
6285
6286 if (devid)
6287 tmp = *devid;
6288 else {
6289 int ret;
6290
6291 ret = find_next_devid(fs_info, &tmp);
6292 if (ret) {
6293 kfree(dev);
6294 return ERR_PTR(ret);
6295 }
6296 }
6297 dev->devid = tmp;
6298
6299 if (uuid)
6300 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6301 else
6302 generate_random_uuid(dev->uuid);
6303
9e0af237
LB		 6304 btrfs_init_work(&dev->work, btrfs_submit_helper,
6305 pending_bios_fn, NULL, NULL);
12bd2fc0
ID		 6306
6307 return dev;
6308}
6309
e06cd3dd 6310/* Return -EIO if any error, otherwise return 0. */
2ff7e61e 6311static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
e06cd3dd
LB		 6312 struct extent_buffer *leaf,
6313 struct btrfs_chunk *chunk, u64 logical)
0b86a832 6314{
0b86a832 6315 u64 length;
f04b772b 6316 u64 stripe_len;
e06cd3dd
LB		 6317 u16 num_stripes;
6318 u16 sub_stripes;
6319 u64 type;
0b86a832 6320
e17cade2 6321 length = btrfs_chunk_length(leaf, chunk);
f04b772b
QW		 6322 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6323 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
e06cd3dd
LB		 6324 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6325 type = btrfs_chunk_type(leaf, chunk);
6326
f04b772b 6327 if (!num_stripes) {
0b246afa 6328 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
f04b772b
QW		 6329 num_stripes);
6330 return -EIO;
6331 }
0b246afa
JM		 6332 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6333 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
f04b772b
QW		 6334 return -EIO;
6335 }
0b246afa
JM		 6336 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6337 btrfs_err(fs_info, "invalid chunk sectorsize %u",
e06cd3dd
LB		 6338 btrfs_chunk_sector_size(leaf, chunk));
6339 return -EIO;
6340 }
0b246afa
JM		 6341 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6342 btrfs_err(fs_info, "invalid chunk length %llu", length);
f04b772b
QW		 6343 return -EIO;
6344 }
3d8da678 6345 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
0b246afa 6346 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
f04b772b
QW		 6347 stripe_len);
6348 return -EIO;
6349 }
6350 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
e06cd3dd 6351 type) {
0b246afa 6352 btrfs_err(fs_info, "unrecognized chunk type: %llu",
f04b772b
QW		 6353 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6354 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6355 btrfs_chunk_type(leaf, chunk));
6356 return -EIO;
6357 }
e06cd3dd
LB		 6358 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6359 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6360 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6361 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6362 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6363 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6364 num_stripes != 1)) {
0b246afa 6365 btrfs_err(fs_info,
e06cd3dd
LB		 6366 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6367 num_stripes, sub_stripes,
6368 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6369 return -EIO;
6370 }
6371
6372 return 0;
6373}
6374
2ff7e61e 6375static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
e06cd3dd
LB		 6376 struct extent_buffer *leaf,
6377 struct btrfs_chunk *chunk)
6378{
0b246afa 6379 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
e06cd3dd
LB		 6380 struct map_lookup *map;
6381 struct extent_map *em;
6382 u64 logical;
6383 u64 length;
e06cd3dd
LB		 6384 u64 devid;
6385 u8 uuid[BTRFS_UUID_SIZE];
6386 int num_stripes;
6387 int ret;
6388 int i;
6389
6390 logical = key->offset;
6391 length = btrfs_chunk_length(leaf, chunk);
e06cd3dd
LB		 6392 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6393
2ff7e61e 6394 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
e06cd3dd
LB		 6395 if (ret)
6396 return ret;
a061fc8d 6397
890871be 6398 read_lock(&map_tree->map_tree.lock);
0b86a832 6399 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
890871be 6400 read_unlock(&map_tree->map_tree.lock);
0b86a832
CM		 6401
6402 /* already mapped? */
6403 if (em && em->start <= logical && em->start + em->len > logical) {
6404 free_extent_map(em);
0b86a832
CM		 6405 return 0;
6406 } else if (em) {
6407 free_extent_map(em);
6408 }
0b86a832 6409
172ddd60 6410 em = alloc_extent_map();
0b86a832
CM		 6411 if (!em)
6412 return -ENOMEM;
593060d7 6413 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
0b86a832
CM		 6414 if (!map) {
6415 free_extent_map(em);
6416 return -ENOMEM;
6417 }
6418
298a8f9c 6419 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
95617d69 6420 em->map_lookup = map;
0b86a832
CM		 6421 em->start = logical;
6422 em->len = length;
70c8a91c 6423 em->orig_start = 0;
0b86a832 6424 em->block_start = 0;
c8b97818 6425 em->block_len = em->len;
0b86a832 6426
593060d7
CM		 6427 map->num_stripes = num_stripes;
6428 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6429 map->io_align = btrfs_chunk_io_align(leaf, chunk);
593060d7
CM		 6430 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6431 map->type = btrfs_chunk_type(leaf, chunk);
321aecc6 6432 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
593060d7
CM		 6433 for (i = 0; i < num_stripes; i++) {
6434 map->stripes[i].physical =
6435 btrfs_stripe_offset_nr(leaf, chunk, i);
6436 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
a443755f
CM		 6437 read_extent_buffer(leaf, uuid, (unsigned long)
6438 btrfs_stripe_dev_uuid_nr(chunk, i),
6439 BTRFS_UUID_SIZE);
0b246afa 6440 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
aa1b8cd4 6441 uuid, NULL);
3cdde224 6442 if (!map->stripes[i].dev &&
0b246afa 6443 !btrfs_test_opt(fs_info, DEGRADED)) {
593060d7 6444 free_extent_map(em);
c5502451 6445 btrfs_report_missing_device(fs_info, devid, uuid);
593060d7
CM		 6446 return -EIO;
6447 }
dfe25020
CM		 6448 if (!map->stripes[i].dev) {
6449 map->stripes[i].dev =
2ff7e61e
JM		 6450 add_missing_dev(fs_info->fs_devices, devid,
6451 uuid);
dfe25020 6452 if (!map->stripes[i].dev) {
dfe25020
CM		 6453 free_extent_map(em);
6454 return -EIO;
6455 }
c5502451 6456 btrfs_report_missing_device(fs_info, devid, uuid);
dfe25020
CM		 6457 }
6458 map->stripes[i].dev->in_fs_metadata = 1;
0b86a832
CM		 6459 }
6460
890871be 6461 write_lock(&map_tree->map_tree.lock);
09a2a8f9 6462 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
890871be 6463 write_unlock(&map_tree->map_tree.lock);
79787eaa 6464 BUG_ON(ret); /* Tree corruption */
0b86a832
CM		 6465 free_extent_map(em);
6466
6467 return 0;
6468}
6469
143bede5 6470static void fill_device_from_item(struct extent_buffer *leaf,
0b86a832
CM		 6471 struct btrfs_dev_item *dev_item,
6472 struct btrfs_device *device)
6473{
6474 unsigned long ptr;
0b86a832
CM		 6475
6476 device->devid = btrfs_device_id(leaf, dev_item);
d6397bae
CB		 6477 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6478 device->total_bytes = device->disk_total_bytes;
935e5cc9 6479 device->commit_total_bytes = device->disk_total_bytes;
0b86a832 6480 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
ce7213c7 6481 device->commit_bytes_used = device->bytes_used;
0b86a832
CM		 6482 device->type = btrfs_device_type(leaf, dev_item);
6483 device->io_align = btrfs_device_io_align(leaf, dev_item);
6484 device->io_width = btrfs_device_io_width(leaf, dev_item);
6485 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
8dabb742 6486 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
63a212ab 6487 device->is_tgtdev_for_dev_replace = 0;
0b86a832 6488
410ba3a2 6489 ptr = btrfs_device_uuid(dev_item);
e17cade2 6490 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
0b86a832
CM		 6491}
6492
2ff7e61e 6493static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
5f375835 6494 u8 *fsid)
2b82032c
YZ		 6495{
6496 struct btrfs_fs_devices *fs_devices;
6497 int ret;
6498
b367e47f 6499 BUG_ON(!mutex_is_locked(&uuid_mutex));
2dfeca9b 6500 ASSERT(fsid);
2b82032c 6501
0b246afa 6502 fs_devices = fs_info->fs_devices->seed;
2b82032c 6503 while (fs_devices) {
44880fdc 6504 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
5f375835
MX		 6505 return fs_devices;
6506
2b82032c
YZ		 6507 fs_devices = fs_devices->seed;
6508 }
6509
6510 fs_devices = find_fsid(fsid);
6511 if (!fs_devices) {
0b246afa 6512 if (!btrfs_test_opt(fs_info, DEGRADED))
5f375835
MX		 6513 return ERR_PTR(-ENOENT);
6514
6515 fs_devices = alloc_fs_devices(fsid);
6516 if (IS_ERR(fs_devices))
6517 return fs_devices;
6518
6519 fs_devices->seeding = 1;
6520 fs_devices->opened = 1;
6521 return fs_devices;
2b82032c 6522 }
e4404d6e
YZ		 6523
6524 fs_devices = clone_fs_devices(fs_devices);
5f375835
MX		 6525 if (IS_ERR(fs_devices))
6526 return fs_devices;
2b82032c 6527
97288f2c 6528 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
0b246afa 6529 fs_info->bdev_holder);
48d28232
JL		 6530 if (ret) {
6531 free_fs_devices(fs_devices);
5f375835 6532 fs_devices = ERR_PTR(ret);
2b82032c 6533 goto out;
48d28232 6534 }
2b82032c
YZ		 6535
6536 if (!fs_devices->seeding) {
6537 __btrfs_close_devices(fs_devices);
e4404d6e 6538 free_fs_devices(fs_devices);
5f375835 6539 fs_devices = ERR_PTR(-EINVAL);
2b82032c
YZ		 6540 goto out;
6541 }
6542
0b246afa
JM		 6543 fs_devices->seed = fs_info->fs_devices->seed;
6544 fs_info->fs_devices->seed = fs_devices;
2b82032c 6545out:
5f375835 6546 return fs_devices;
2b82032c
YZ		 6547}
6548
2ff7e61e 6549static int read_one_dev(struct btrfs_fs_info *fs_info,
0b86a832
CM		 6550 struct extent_buffer *leaf,
6551 struct btrfs_dev_item *dev_item)
6552{
0b246afa 6553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
0b86a832
CM		 6554 struct btrfs_device *device;
6555 u64 devid;
6556 int ret;
44880fdc 6557 u8 fs_uuid[BTRFS_FSID_SIZE];
a443755f
CM		 6558 u8 dev_uuid[BTRFS_UUID_SIZE];
6559
0b86a832 6560 devid = btrfs_device_id(leaf, dev_item);
410ba3a2 6561 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
a443755f 6562 BTRFS_UUID_SIZE);
1473b24e 6563 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
44880fdc 6564 BTRFS_FSID_SIZE);
2b82032c 6565
44880fdc 6566 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
2ff7e61e 6567 fs_devices = open_seed_devices(fs_info, fs_uuid);
5f375835
MX		 6568 if (IS_ERR(fs_devices))
6569 return PTR_ERR(fs_devices);
2b82032c
YZ		 6570 }
6571
0b246afa 6572 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
5f375835 6573 if (!device) {
c5502451
QW		 6574 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6575 btrfs_report_missing_device(fs_info, devid, dev_uuid);
2b82032c 6576 return -EIO;
c5502451 6577 }
2b82032c 6578
2ff7e61e 6579 device = add_missing_dev(fs_devices, devid, dev_uuid);
5f375835
MX		 6580 if (!device)
6581 return -ENOMEM;
c5502451 6582 btrfs_report_missing_device(fs_info, devid, dev_uuid);
5f375835 6583 } else {
c5502451
QW		 6584 if (!device->bdev) {
6585 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6586 if (!btrfs_test_opt(fs_info, DEGRADED))
6587 return -EIO;
6588 }
5f375835
MX		 6589
6590 if(!device->bdev && !device->missing) {
cd02dca5
CM		 6591 /*
6592 * this happens when a device that was properly setup
6593 * in the device info lists suddenly goes bad.
6594 * device->bdev is NULL, and so we have to set
6595 * device->missing to one here
6596 */
5f375835 6597 device->fs_devices->missing_devices++;
cd02dca5 6598 device->missing = 1;
2b82032c 6599 }
5f375835
MX		 6600
6601 /* Move the device to its own fs_devices */
6602 if (device->fs_devices != fs_devices) {
6603 ASSERT(device->missing);
6604
6605 list_move(&device->dev_list, &fs_devices->devices);
6606 device->fs_devices->num_devices--;
6607 fs_devices->num_devices++;
6608
6609 device->fs_devices->missing_devices--;
6610 fs_devices->missing_devices++;
6611
6612 device->fs_devices = fs_devices;
6613 }
2b82032c
YZ		 6614 }
6615
0b246afa 6616 if (device->fs_devices != fs_info->fs_devices) {
2b82032c
YZ		 6617 BUG_ON(device->writeable);
6618 if (device->generation !=
6619 btrfs_device_generation(leaf, dev_item))
6620 return -EINVAL;
6324fbf3 6621 }
0b86a832
CM		 6622
6623 fill_device_from_item(leaf, dev_item, device);
dfe25020 6624 device->in_fs_metadata = 1;
63a212ab 6625 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
2b82032c 6626 device->fs_devices->total_rw_bytes += device->total_bytes;
a5ed45f8
NB		 6627 atomic64_add(device->total_bytes - device->bytes_used,
6628 &fs_info->free_chunk_space);
2bf64758 6629 }
0b86a832 6630 ret = 0;
0b86a832
CM		 6631 return ret;
6632}
6633
6bccf3ab 6634int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
0b86a832 6635{
6bccf3ab 6636 struct btrfs_root *root = fs_info->tree_root;
ab8d0fc4 6637 struct btrfs_super_block *super_copy = fs_info->super_copy;
a061fc8d 6638 struct extent_buffer *sb;
0b86a832 6639 struct btrfs_disk_key *disk_key;
0b86a832 6640 struct btrfs_chunk *chunk;
1ffb22cf
DS		 6641 u8 *array_ptr;
6642 unsigned long sb_array_offset;
84eed90f 6643 int ret = 0;
0b86a832
CM		 6644 u32 num_stripes;
6645 u32 array_size;
6646 u32 len = 0;
1ffb22cf 6647 u32 cur_offset;
e06cd3dd 6648 u64 type;
84eed90f 6649 struct btrfs_key key;
0b86a832 6650
0b246afa 6651 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
a83fffb7
DS		 6652 /*
6653 * This will create extent buffer of nodesize, superblock size is
6654 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6655 * overallocate but we can keep it as-is, only the first page is used.
6656 */
2ff7e61e 6657 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
c871b0f2
LB		 6658 if (IS_ERR(sb))
6659 return PTR_ERR(sb);
4db8c528 6660 set_extent_buffer_uptodate(sb);
85d4e461 6661 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
8a334426 6662 /*
01327610 6663 * The sb extent buffer is artificial and just used to read the system array.
4db8c528 6664 * set_extent_buffer_uptodate() call does not properly mark all it's
8a334426
DS		 6665 * pages up-to-date when the page is larger: extent does not cover the
6666 * whole page and consequently check_page_uptodate does not find all
6667 * the page's extents up-to-date (the hole beyond sb),
6668 * write_extent_buffer then triggers a WARN_ON.
6669 *
6670 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6671 * but sb spans only this function. Add an explicit SetPageUptodate call
6672 * to silence the warning eg. on PowerPC 64.
6673 */
09cbfeaf 6674 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
727011e0 6675 SetPageUptodate(sb->pages[0]);
4008c04a 6676
a061fc8d 6677 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
0b86a832
CM		 6678 array_size = btrfs_super_sys_array_size(super_copy);
6679
1ffb22cf
DS		 6680 array_ptr = super_copy->sys_chunk_array;
6681 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6682 cur_offset = 0;
0b86a832 6683
1ffb22cf
DS		 6684 while (cur_offset < array_size) {
6685 disk_key = (struct btrfs_disk_key *)array_ptr;
e3540eab
DS		 6686 len = sizeof(*disk_key);
6687 if (cur_offset + len > array_size)
6688 goto out_short_read;
6689
0b86a832
CM		 6690 btrfs_disk_key_to_cpu(&key, disk_key);
6691
1ffb22cf
DS		 6692 array_ptr += len;
6693 sb_array_offset += len;
6694 cur_offset += len;
0b86a832 6695
0d81ba5d 6696 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1ffb22cf 6697 chunk = (struct btrfs_chunk *)sb_array_offset;
e3540eab
DS		 6698 /*
6699 * At least one btrfs_chunk with one stripe must be
6700 * present, exact stripe count check comes afterwards
6701 */
6702 len = btrfs_chunk_item_size(1);
6703 if (cur_offset + len > array_size)
6704 goto out_short_read;
6705
6706 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
f5cdedd7 6707 if (!num_stripes) {
ab8d0fc4
JM		 6708 btrfs_err(fs_info,
6709 "invalid number of stripes %u in sys_array at offset %u",
f5cdedd7
DS		 6710 num_stripes, cur_offset);
6711 ret = -EIO;
6712 break;
6713 }
6714
e06cd3dd
LB		 6715 type = btrfs_chunk_type(sb, chunk);
6716 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
ab8d0fc4 6717 btrfs_err(fs_info,
e06cd3dd
LB		 6718 "invalid chunk type %llu in sys_array at offset %u",
6719 type, cur_offset);
6720 ret = -EIO;
6721 break;
6722 }
6723
e3540eab
DS		 6724 len = btrfs_chunk_item_size(num_stripes);
6725 if (cur_offset + len > array_size)
6726 goto out_short_read;
6727
2ff7e61e 6728 ret = read_one_chunk(fs_info, &key, sb, chunk);
84eed90f
CM		 6729 if (ret)
6730 break;
0b86a832 6731 } else {
ab8d0fc4
JM		 6732 btrfs_err(fs_info,
6733 "unexpected item type %u in sys_array at offset %u",
6734 (u32)key.type, cur_offset);
84eed90f
CM		 6735 ret = -EIO;
6736 break;
0b86a832 6737 }
1ffb22cf
DS		 6738 array_ptr += len;
6739 sb_array_offset += len;
6740 cur_offset += len;
0b86a832 6741 }
d865177a 6742 clear_extent_buffer_uptodate(sb);
1c8b5b6e 6743 free_extent_buffer_stale(sb);
84eed90f 6744 return ret;
e3540eab
DS		 6745
6746out_short_read:
ab8d0fc4 6747 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
e3540eab 6748 len, cur_offset);
d865177a 6749 clear_extent_buffer_uptodate(sb);
1c8b5b6e 6750 free_extent_buffer_stale(sb);
e3540eab 6751 return -EIO;
0b86a832
CM		 6752}
6753
c5502451
QW		 6754void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, u64 devid,
6755 u8 *uuid)
6756{
6757 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", devid, uuid);
6758}
6759
21634a19
QW		 6760/*
6761 * Check if all chunks in the fs are OK for read-write degraded mount
6762 *
6763 * Return true if all chunks meet the minimal RW mount requirements.
6764 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6765 */
6766bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6767{
6768 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6769 struct extent_map *em;
6770 u64 next_start = 0;
6771 bool ret = true;
6772
6773 read_lock(&map_tree->map_tree.lock);
6774 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6775 read_unlock(&map_tree->map_tree.lock);
6776 /* No chunk at all? Return false anyway */
6777 if (!em) {
6778 ret = false;
6779 goto out;
6780 }
6781 while (em) {
6782 struct map_lookup *map;
6783 int missing = 0;
6784 int max_tolerated;
6785 int i;
6786
6787 map = em->map_lookup;
6788 max_tolerated =
6789 btrfs_get_num_tolerated_disk_barrier_failures(
6790 map->type);
6791 for (i = 0; i < map->num_stripes; i++) {
6792 struct btrfs_device *dev = map->stripes[i].dev;
6793
6794 if (!dev || !dev->bdev || dev->missing ||
6795 dev->last_flush_error)
6796 missing++;
6797 }
6798 if (missing > max_tolerated) {
6799 btrfs_warn(fs_info,
6800 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6801 em->start, missing, max_tolerated);
6802 free_extent_map(em);
6803 ret = false;
6804 goto out;
6805 }
6806 next_start = extent_map_end(em);
6807 free_extent_map(em);
6808
6809 read_lock(&map_tree->map_tree.lock);
6810 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6811 (u64)(-1) - next_start);
6812 read_unlock(&map_tree->map_tree.lock);
6813 }
6814out:
6815 return ret;
6816}
6817
5b4aacef 6818int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
0b86a832 6819{
5b4aacef 6820 struct btrfs_root *root = fs_info->chunk_root;
0b86a832
CM		 6821 struct btrfs_path *path;
6822 struct extent_buffer *leaf;
6823 struct btrfs_key key;
6824 struct btrfs_key found_key;
6825 int ret;
6826 int slot;
99e3ecfc 6827 u64 total_dev = 0;
0b86a832 6828
0b86a832
CM		 6829 path = btrfs_alloc_path();
6830 if (!path)
6831 return -ENOMEM;
6832
b367e47f 6833 mutex_lock(&uuid_mutex);
34441361 6834 mutex_lock(&fs_info->chunk_mutex);
b367e47f 6835
395927a9
FDBM		 6836 /*
6837 * Read all device items, and then all the chunk items. All
6838 * device items are found before any chunk item (their object id
6839 * is smaller than the lowest possible object id for a chunk
6840 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
0b86a832
CM		 6841 */
6842 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6843 key.offset = 0;
6844 key.type = 0;
0b86a832 6845 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
ab59381e
ZL		 6846 if (ret < 0)
6847 goto error;
d397712b 6848 while (1) {
0b86a832
CM		 6849 leaf = path->nodes[0];
6850 slot = path->slots[0];
6851 if (slot >= btrfs_header_nritems(leaf)) {
6852 ret = btrfs_next_leaf(root, path);
6853 if (ret == 0)
6854 continue;
6855 if (ret < 0)
6856 goto error;
6857 break;
6858 }
6859 btrfs_item_key_to_cpu(leaf, &found_key, slot);
395927a9
FDBM		 6860 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6861 struct btrfs_dev_item *dev_item;
6862 dev_item = btrfs_item_ptr(leaf, slot,
0b86a832 6863 struct btrfs_dev_item);
2ff7e61e 6864 ret = read_one_dev(fs_info, leaf, dev_item);
395927a9
FDBM		 6865 if (ret)
6866 goto error;
99e3ecfc 6867 total_dev++;
0b86a832
CM		 6868 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6869 struct btrfs_chunk *chunk;
6870 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2ff7e61e 6871 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
2b82032c
YZ		 6872 if (ret)
6873 goto error;
0b86a832
CM		 6874 }
6875 path->slots[0]++;
6876 }
99e3ecfc
LB		 6877
6878 /*
6879 * After loading chunk tree, we've got all device information,
6880 * do another round of validation checks.
6881 */
0b246afa
JM		 6882 if (total_dev != fs_info->fs_devices->total_devices) {
6883 btrfs_err(fs_info,
99e3ecfc 6884 "super_num_devices %llu mismatch with num_devices %llu found here",
0b246afa 6885 btrfs_super_num_devices(fs_info->super_copy),
99e3ecfc
LB		 6886 total_dev);
6887 ret = -EINVAL;
6888 goto error;
6889 }
0b246afa
JM		 6890 if (btrfs_super_total_bytes(fs_info->super_copy) <
6891 fs_info->fs_devices->total_rw_bytes) {
6892 btrfs_err(fs_info,
99e3ecfc 6893 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
0b246afa
JM		 6894 btrfs_super_total_bytes(fs_info->super_copy),
6895 fs_info->fs_devices->total_rw_bytes);
99e3ecfc
LB		 6896 ret = -EINVAL;
6897 goto error;
6898 }
0b86a832
CM		 6899 ret = 0;
6900error:
34441361 6901 mutex_unlock(&fs_info->chunk_mutex);
b367e47f
LZ		 6902 mutex_unlock(&uuid_mutex);
6903
2b82032c 6904 btrfs_free_path(path);
0b86a832
CM		 6905 return ret;
6906}
442a4f63 6907
cb517eab
MX		 6908void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6909{
6910 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6911 struct btrfs_device *device;
6912
29cc83f6
LB		 6913 while (fs_devices) {
6914 mutex_lock(&fs_devices->device_list_mutex);
6915 list_for_each_entry(device, &fs_devices->devices, dev_list)
fb456252 6916 device->fs_info = fs_info;
29cc83f6
LB		 6917 mutex_unlock(&fs_devices->device_list_mutex);
6918
6919 fs_devices = fs_devices->seed;
6920 }
cb517eab
MX		 6921}
6922
733f4fbb
SB		 6923static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6924{
6925 int i;
6926
6927 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6928 btrfs_dev_stat_reset(dev, i);
6929}
6930
6931int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6932{
6933 struct btrfs_key key;
6934 struct btrfs_key found_key;
6935 struct btrfs_root *dev_root = fs_info->dev_root;
6936 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6937 struct extent_buffer *eb;
6938 int slot;
6939 int ret = 0;
6940 struct btrfs_device *device;
6941 struct btrfs_path *path = NULL;
6942 int i;
6943
6944 path = btrfs_alloc_path();
6945 if (!path) {
6946 ret = -ENOMEM;
6947 goto out;
6948 }
6949
6950 mutex_lock(&fs_devices->device_list_mutex);
6951 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6952 int item_size;
6953 struct btrfs_dev_stats_item *ptr;
6954
242e2956
DS		 6955 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6956 key.type = BTRFS_PERSISTENT_ITEM_KEY;
733f4fbb
SB		 6957 key.offset = device->devid;
6958 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6959 if (ret) {
733f4fbb
SB		 6960 __btrfs_reset_dev_stats(device);
6961 device->dev_stats_valid = 1;
6962 btrfs_release_path(path);
6963 continue;
6964 }
6965 slot = path->slots[0];
6966 eb = path->nodes[0];
6967 btrfs_item_key_to_cpu(eb, &found_key, slot);
6968 item_size = btrfs_item_size_nr(eb, slot);
6969
6970 ptr = btrfs_item_ptr(eb, slot,
6971 struct btrfs_dev_stats_item);
6972
6973 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6974 if (item_size >= (1 + i) * sizeof(__le64))
6975 btrfs_dev_stat_set(device, i,
6976 btrfs_dev_stats_value(eb, ptr, i));
6977 else
6978 btrfs_dev_stat_reset(device, i);
6979 }
6980
6981 device->dev_stats_valid = 1;
6982 btrfs_dev_stat_print_on_load(device);
6983 btrfs_release_path(path);
6984 }
6985 mutex_unlock(&fs_devices->device_list_mutex);
6986
6987out:
6988 btrfs_free_path(path);
6989 return ret < 0 ? ret : 0;
6990}
6991
6992static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6bccf3ab 6993 struct btrfs_fs_info *fs_info,
733f4fbb
SB		 6994 struct btrfs_device *device)
6995{
6bccf3ab 6996 struct btrfs_root *dev_root = fs_info->dev_root;
733f4fbb
SB		 6997 struct btrfs_path *path;
6998 struct btrfs_key key;
6999 struct extent_buffer *eb;
7000 struct btrfs_dev_stats_item *ptr;
7001 int ret;
7002 int i;
7003
242e2956
DS		 7004 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7005 key.type = BTRFS_PERSISTENT_ITEM_KEY;
733f4fbb
SB		 7006 key.offset = device->devid;
7007
7008 path = btrfs_alloc_path();
fa252992
DS		 7009 if (!path)
7010 return -ENOMEM;
733f4fbb
SB		 7011 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7012 if (ret < 0) {
0b246afa 7013 btrfs_warn_in_rcu(fs_info,
ecaeb14b 7014 "error %d while searching for dev_stats item for device %s",
606686ee 7015 ret, rcu_str_deref(device->name));
733f4fbb
SB		 7016 goto out;
7017 }
7018
7019 if (ret == 0 &&
7020 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7021 /* need to delete old one and insert a new one */
7022 ret = btrfs_del_item(trans, dev_root, path);
7023 if (ret != 0) {
0b246afa 7024 btrfs_warn_in_rcu(fs_info,
ecaeb14b 7025 "delete too small dev_stats item for device %s failed %d",
606686ee 7026 rcu_str_deref(device->name), ret);
733f4fbb
SB		 7027 goto out;
7028 }
7029 ret = 1;
7030 }
7031
7032 if (ret == 1) {
7033 /* need to insert a new item */
7034 btrfs_release_path(path);
7035 ret = btrfs_insert_empty_item(trans, dev_root, path,
7036 &key, sizeof(*ptr));
7037 if (ret < 0) {
0b246afa 7038 btrfs_warn_in_rcu(fs_info,
ecaeb14b
DS		 7039 "insert dev_stats item for device %s failed %d",
7040 rcu_str_deref(device->name), ret);
733f4fbb
SB		 7041 goto out;
7042 }
7043 }
7044
7045 eb = path->nodes[0];
7046 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7047 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7048 btrfs_set_dev_stats_value(eb, ptr, i,
7049 btrfs_dev_stat_read(device, i));
7050 btrfs_mark_buffer_dirty(eb);
7051
7052out:
7053 btrfs_free_path(path);
7054 return ret;
7055}
7056
7057/*
7058 * called from commit_transaction. Writes all changed device stats to disk.
7059 */
7060int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7061 struct btrfs_fs_info *fs_info)
7062{
733f4fbb
SB		 7063 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7064 struct btrfs_device *device;
addc3fa7 7065 int stats_cnt;
733f4fbb
SB		 7066 int ret = 0;
7067
7068 mutex_lock(&fs_devices->device_list_mutex);
7069 list_for_each_entry(device, &fs_devices->devices, dev_list) {
addc3fa7 7070 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
733f4fbb
SB		 7071 continue;
7072
addc3fa7 7073 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6bccf3ab 7074 ret = update_dev_stat_item(trans, fs_info, device);
733f4fbb 7075 if (!ret)
addc3fa7 7076 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
733f4fbb
SB		 7077 }
7078 mutex_unlock(&fs_devices->device_list_mutex);
7079
7080 return ret;
7081}
7082
442a4f63
SB		 7083void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7084{
7085 btrfs_dev_stat_inc(dev, index);
7086 btrfs_dev_stat_print_on_error(dev);
7087}
7088
48a3b636 7089static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
442a4f63 7090{
733f4fbb
SB		 7091 if (!dev->dev_stats_valid)
7092 return;
fb456252 7093 btrfs_err_rl_in_rcu(dev->fs_info,
b14af3b4 7094 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
606686ee 7095 rcu_str_deref(dev->name),
442a4f63
SB		 7096 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7097 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7098 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
efe120a0
FH		 7099 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7100 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
442a4f63 7101}
c11d2c23 7102
733f4fbb
SB		 7103static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7104{
a98cdb85
SB		 7105 int i;
7106
7107 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7108 if (btrfs_dev_stat_read(dev, i) != 0)
7109 break;
7110 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7111 return; /* all values == 0, suppress message */
7112
fb456252 7113 btrfs_info_in_rcu(dev->fs_info,
ecaeb14b 7114 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
606686ee 7115 rcu_str_deref(dev->name),
733f4fbb
SB		 7116 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7117 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7118 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7119 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7120 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7121}
7122
2ff7e61e 7123int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
b27f7c0c 7124 struct btrfs_ioctl_get_dev_stats *stats)
c11d2c23
SB		 7125{
7126 struct btrfs_device *dev;
0b246afa 7127 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
c11d2c23
SB		 7128 int i;
7129
7130 mutex_lock(&fs_devices->device_list_mutex);
0b246afa 7131 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
c11d2c23
SB		 7132 mutex_unlock(&fs_devices->device_list_mutex);
7133
7134 if (!dev) {
0b246afa 7135 btrfs_warn(fs_info, "get dev_stats failed, device not found");
c11d2c23 7136 return -ENODEV;
733f4fbb 7137 } else if (!dev->dev_stats_valid) {
0b246afa 7138 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
733f4fbb 7139 return -ENODEV;
b27f7c0c 7140 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
c11d2c23
SB		 7141 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7142 if (stats->nr_items > i)
7143 stats->values[i] =
7144 btrfs_dev_stat_read_and_reset(dev, i);
7145 else
7146 btrfs_dev_stat_reset(dev, i);
7147 }
7148 } else {
7149 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7150 if (stats->nr_items > i)
7151 stats->values[i] = btrfs_dev_stat_read(dev, i);
7152 }
7153 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7154 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7155 return 0;
7156}
a8a6dab7 7157
da353f6b 7158void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
a8a6dab7
SB		 7159{
7160 struct buffer_head *bh;
7161 struct btrfs_super_block *disk_super;
12b1c263 7162 int copy_num;
a8a6dab7 7163
12b1c263
AJ		 7164 if (!bdev)
7165 return;
a8a6dab7 7166
12b1c263
AJ		 7167 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7168 copy_num++) {
a8a6dab7 7169
12b1c263
AJ		 7170 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7171 continue;
7172
7173 disk_super = (struct btrfs_super_block *)bh->b_data;
7174
7175 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7176 set_buffer_dirty(bh);
7177 sync_dirty_buffer(bh);
7178 brelse(bh);
7179 }
7180
7181 /* Notify udev that device has changed */
7182 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7183
7184 /* Update ctime/mtime for device path for libblkid */
7185 update_dev_time(device_path);
a8a6dab7 7186}
935e5cc9
MX		 7187
7188/*
7189 * Update the size of all devices, which is used for writing out the
7190 * super blocks.
7191 */
7192void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7193{
7194 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7195 struct btrfs_device *curr, *next;
7196
7197 if (list_empty(&fs_devices->resized_devices))
7198 return;
7199
7200 mutex_lock(&fs_devices->device_list_mutex);
34441361 7201 mutex_lock(&fs_info->chunk_mutex);
935e5cc9
MX		 7202 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7203 resized_list) {
7204 list_del_init(&curr->resized_list);
7205 curr->commit_total_bytes = curr->disk_total_bytes;
7206 }
34441361 7207 mutex_unlock(&fs_info->chunk_mutex);
935e5cc9
MX		 7208 mutex_unlock(&fs_devices->device_list_mutex);
7209}
ce7213c7
MX		 7210
7211/* Must be invoked during the transaction commit */
2ff7e61e 7212void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
ce7213c7
MX		 7213 struct btrfs_transaction *transaction)
7214{
7215 struct extent_map *em;
7216 struct map_lookup *map;
7217 struct btrfs_device *dev;
7218 int i;
7219
7220 if (list_empty(&transaction->pending_chunks))
7221 return;
7222
7223 /* In order to kick the device replace finish process */
34441361 7224 mutex_lock(&fs_info->chunk_mutex);
ce7213c7 7225 list_for_each_entry(em, &transaction->pending_chunks, list) {
95617d69 7226 map = em->map_lookup;
ce7213c7
MX		 7227
7228 for (i = 0; i < map->num_stripes; i++) {
7229 dev = map->stripes[i].dev;
7230 dev->commit_bytes_used = dev->bytes_used;
7231 }
7232 }
34441361 7233 mutex_unlock(&fs_info->chunk_mutex);
ce7213c7 7234}
5a13f430
AJ		 7235
7236void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7237{
7238 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7239 while (fs_devices) {
7240 fs_devices->fs_info = fs_info;
7241 fs_devices = fs_devices->seed;
7242 }
7243}
7244
7245void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7246{
7247 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7248 while (fs_devices) {
7249 fs_devices->fs_info = NULL;
7250 fs_devices = fs_devices->seed;
7251 }
7252}
Linux 6.x block layer and io_uring tree(s)