xref: /openbmc/linux/fs/btrfs/volumes.c (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
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>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
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"
37 #include "raid56.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
41 #include "math.h"
42 #include "dev-replace.h"
43 #include "sysfs.h"
44 
45 const 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,
51 		.tolerated_failures = 1,
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,
60 		.tolerated_failures = 1,
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,
69 		.tolerated_failures = 0,
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,
78 		.tolerated_failures = 0,
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,
87 		.tolerated_failures = 0,
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,
96 		.tolerated_failures = 1,
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,
105 		.tolerated_failures = 2,
106 		.devs_increment	= 1,
107 		.ncopies	= 3,
108 	},
109 };
110 
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
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 
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  */
126 const 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 
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 				struct btrfs_fs_info *fs_info,
138 				struct btrfs_device *device);
139 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143 
144 DEFINE_MUTEX(uuid_mutex);
145 static LIST_HEAD(fs_uuids);
146 struct list_head *btrfs_get_fs_uuids(void)
147 {
148 	return &fs_uuids;
149 }
150 
151 static struct btrfs_fs_devices *__alloc_fs_devices(void)
152 {
153 	struct btrfs_fs_devices *fs_devs;
154 
155 	fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
156 	if (!fs_devs)
157 		return ERR_PTR(-ENOMEM);
158 
159 	mutex_init(&fs_devs->device_list_mutex);
160 
161 	INIT_LIST_HEAD(&fs_devs->devices);
162 	INIT_LIST_HEAD(&fs_devs->resized_devices);
163 	INIT_LIST_HEAD(&fs_devs->alloc_list);
164 	INIT_LIST_HEAD(&fs_devs->list);
165 
166 	return fs_devs;
167 }
168 
169 /**
170  * alloc_fs_devices - allocate struct btrfs_fs_devices
171  * @fsid:	a pointer to UUID for this FS.  If NULL a new UUID is
172  *		generated.
173  *
174  * Return: a pointer to a new &struct btrfs_fs_devices on success;
175  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
176  * can be destroyed with kfree() right away.
177  */
178 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
179 {
180 	struct btrfs_fs_devices *fs_devs;
181 
182 	fs_devs = __alloc_fs_devices();
183 	if (IS_ERR(fs_devs))
184 		return fs_devs;
185 
186 	if (fsid)
187 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
188 	else
189 		generate_random_uuid(fs_devs->fsid);
190 
191 	return fs_devs;
192 }
193 
194 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
195 {
196 	struct btrfs_device *device;
197 	WARN_ON(fs_devices->opened);
198 	while (!list_empty(&fs_devices->devices)) {
199 		device = list_entry(fs_devices->devices.next,
200 				    struct btrfs_device, dev_list);
201 		list_del(&device->dev_list);
202 		rcu_string_free(device->name);
203 		kfree(device);
204 	}
205 	kfree(fs_devices);
206 }
207 
208 static void btrfs_kobject_uevent(struct block_device *bdev,
209 				 enum kobject_action action)
210 {
211 	int ret;
212 
213 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
214 	if (ret)
215 		pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
216 			action,
217 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218 			&disk_to_dev(bdev->bd_disk)->kobj);
219 }
220 
221 void btrfs_cleanup_fs_uuids(void)
222 {
223 	struct btrfs_fs_devices *fs_devices;
224 
225 	while (!list_empty(&fs_uuids)) {
226 		fs_devices = list_entry(fs_uuids.next,
227 					struct btrfs_fs_devices, list);
228 		list_del(&fs_devices->list);
229 		free_fs_devices(fs_devices);
230 	}
231 }
232 
233 static struct btrfs_device *__alloc_device(void)
234 {
235 	struct btrfs_device *dev;
236 
237 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
238 	if (!dev)
239 		return ERR_PTR(-ENOMEM);
240 
241 	INIT_LIST_HEAD(&dev->dev_list);
242 	INIT_LIST_HEAD(&dev->dev_alloc_list);
243 	INIT_LIST_HEAD(&dev->resized_list);
244 
245 	spin_lock_init(&dev->io_lock);
246 
247 	spin_lock_init(&dev->reada_lock);
248 	atomic_set(&dev->reada_in_flight, 0);
249 	atomic_set(&dev->dev_stats_ccnt, 0);
250 	btrfs_device_data_ordered_init(dev);
251 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253 
254 	return dev;
255 }
256 
257 static noinline struct btrfs_device *__find_device(struct list_head *head,
258 						   u64 devid, u8 *uuid)
259 {
260 	struct btrfs_device *dev;
261 
262 	list_for_each_entry(dev, head, dev_list) {
263 		if (dev->devid == devid &&
264 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
265 			return dev;
266 		}
267 	}
268 	return NULL;
269 }
270 
271 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
272 {
273 	struct btrfs_fs_devices *fs_devices;
274 
275 	list_for_each_entry(fs_devices, &fs_uuids, list) {
276 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
277 			return fs_devices;
278 	}
279 	return NULL;
280 }
281 
282 static int
283 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284 		      int flush, struct block_device **bdev,
285 		      struct buffer_head **bh)
286 {
287 	int ret;
288 
289 	*bdev = blkdev_get_by_path(device_path, flags, holder);
290 
291 	if (IS_ERR(*bdev)) {
292 		ret = PTR_ERR(*bdev);
293 		goto error;
294 	}
295 
296 	if (flush)
297 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298 	ret = set_blocksize(*bdev, 4096);
299 	if (ret) {
300 		blkdev_put(*bdev, flags);
301 		goto error;
302 	}
303 	invalidate_bdev(*bdev);
304 	*bh = btrfs_read_dev_super(*bdev);
305 	if (IS_ERR(*bh)) {
306 		ret = PTR_ERR(*bh);
307 		blkdev_put(*bdev, flags);
308 		goto error;
309 	}
310 
311 	return 0;
312 
313 error:
314 	*bdev = NULL;
315 	*bh = NULL;
316 	return ret;
317 }
318 
319 static void requeue_list(struct btrfs_pending_bios *pending_bios,
320 			struct bio *head, struct bio *tail)
321 {
322 
323 	struct bio *old_head;
324 
325 	old_head = pending_bios->head;
326 	pending_bios->head = head;
327 	if (pending_bios->tail)
328 		tail->bi_next = old_head;
329 	else
330 		pending_bios->tail = tail;
331 }
332 
333 /*
334  * we try to collect pending bios for a device so we don't get a large
335  * number of procs sending bios down to the same device.  This greatly
336  * improves the schedulers ability to collect and merge the bios.
337  *
338  * But, it also turns into a long list of bios to process and that is sure
339  * to eventually make the worker thread block.  The solution here is to
340  * make some progress and then put this work struct back at the end of
341  * the list if the block device is congested.  This way, multiple devices
342  * can make progress from a single worker thread.
343  */
344 static noinline void run_scheduled_bios(struct btrfs_device *device)
345 {
346 	struct btrfs_fs_info *fs_info = device->fs_info;
347 	struct bio *pending;
348 	struct backing_dev_info *bdi;
349 	struct btrfs_pending_bios *pending_bios;
350 	struct bio *tail;
351 	struct bio *cur;
352 	int again = 0;
353 	unsigned long num_run;
354 	unsigned long batch_run = 0;
355 	unsigned long limit;
356 	unsigned long last_waited = 0;
357 	int force_reg = 0;
358 	int sync_pending = 0;
359 	struct blk_plug plug;
360 
361 	/*
362 	 * this function runs all the bios we've collected for
363 	 * a particular device.  We don't want to wander off to
364 	 * another device without first sending all of these down.
365 	 * So, setup a plug here and finish it off before we return
366 	 */
367 	blk_start_plug(&plug);
368 
369 	bdi = blk_get_backing_dev_info(device->bdev);
370 	limit = btrfs_async_submit_limit(fs_info);
371 	limit = limit * 2 / 3;
372 
373 loop:
374 	spin_lock(&device->io_lock);
375 
376 loop_lock:
377 	num_run = 0;
378 
379 	/* take all the bios off the list at once and process them
380 	 * later on (without the lock held).  But, remember the
381 	 * tail and other pointers so the bios can be properly reinserted
382 	 * into the list if we hit congestion
383 	 */
384 	if (!force_reg && device->pending_sync_bios.head) {
385 		pending_bios = &device->pending_sync_bios;
386 		force_reg = 1;
387 	} else {
388 		pending_bios = &device->pending_bios;
389 		force_reg = 0;
390 	}
391 
392 	pending = pending_bios->head;
393 	tail = pending_bios->tail;
394 	WARN_ON(pending && !tail);
395 
396 	/*
397 	 * if pending was null this time around, no bios need processing
398 	 * at all and we can stop.  Otherwise it'll loop back up again
399 	 * and do an additional check so no bios are missed.
400 	 *
401 	 * device->running_pending is used to synchronize with the
402 	 * schedule_bio code.
403 	 */
404 	if (device->pending_sync_bios.head == NULL &&
405 	    device->pending_bios.head == NULL) {
406 		again = 0;
407 		device->running_pending = 0;
408 	} else {
409 		again = 1;
410 		device->running_pending = 1;
411 	}
412 
413 	pending_bios->head = NULL;
414 	pending_bios->tail = NULL;
415 
416 	spin_unlock(&device->io_lock);
417 
418 	while (pending) {
419 
420 		rmb();
421 		/* we want to work on both lists, but do more bios on the
422 		 * sync list than the regular list
423 		 */
424 		if ((num_run > 32 &&
425 		    pending_bios != &device->pending_sync_bios &&
426 		    device->pending_sync_bios.head) ||
427 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
428 		    device->pending_bios.head)) {
429 			spin_lock(&device->io_lock);
430 			requeue_list(pending_bios, pending, tail);
431 			goto loop_lock;
432 		}
433 
434 		cur = pending;
435 		pending = pending->bi_next;
436 		cur->bi_next = NULL;
437 
438 		/*
439 		 * atomic_dec_return implies a barrier for waitqueue_active
440 		 */
441 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
442 		    waitqueue_active(&fs_info->async_submit_wait))
443 			wake_up(&fs_info->async_submit_wait);
444 
445 		BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
446 
447 		/*
448 		 * if we're doing the sync list, record that our
449 		 * plug has some sync requests on it
450 		 *
451 		 * If we're doing the regular list and there are
452 		 * sync requests sitting around, unplug before
453 		 * we add more
454 		 */
455 		if (pending_bios == &device->pending_sync_bios) {
456 			sync_pending = 1;
457 		} else if (sync_pending) {
458 			blk_finish_plug(&plug);
459 			blk_start_plug(&plug);
460 			sync_pending = 0;
461 		}
462 
463 		btrfsic_submit_bio(cur);
464 		num_run++;
465 		batch_run++;
466 
467 		cond_resched();
468 
469 		/*
470 		 * we made progress, there is more work to do and the bdi
471 		 * is now congested.  Back off and let other work structs
472 		 * run instead
473 		 */
474 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
475 		    fs_info->fs_devices->open_devices > 1) {
476 			struct io_context *ioc;
477 
478 			ioc = current->io_context;
479 
480 			/*
481 			 * the main goal here is that we don't want to
482 			 * block if we're going to be able to submit
483 			 * more requests without blocking.
484 			 *
485 			 * This code does two great things, it pokes into
486 			 * the elevator code from a filesystem _and_
487 			 * it makes assumptions about how batching works.
488 			 */
489 			if (ioc && ioc->nr_batch_requests > 0 &&
490 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
491 			    (last_waited == 0 ||
492 			     ioc->last_waited == last_waited)) {
493 				/*
494 				 * we want to go through our batch of
495 				 * requests and stop.  So, we copy out
496 				 * the ioc->last_waited time and test
497 				 * against it before looping
498 				 */
499 				last_waited = ioc->last_waited;
500 				cond_resched();
501 				continue;
502 			}
503 			spin_lock(&device->io_lock);
504 			requeue_list(pending_bios, pending, tail);
505 			device->running_pending = 1;
506 
507 			spin_unlock(&device->io_lock);
508 			btrfs_queue_work(fs_info->submit_workers,
509 					 &device->work);
510 			goto done;
511 		}
512 		/* unplug every 64 requests just for good measure */
513 		if (batch_run % 64 == 0) {
514 			blk_finish_plug(&plug);
515 			blk_start_plug(&plug);
516 			sync_pending = 0;
517 		}
518 	}
519 
520 	cond_resched();
521 	if (again)
522 		goto loop;
523 
524 	spin_lock(&device->io_lock);
525 	if (device->pending_bios.head || device->pending_sync_bios.head)
526 		goto loop_lock;
527 	spin_unlock(&device->io_lock);
528 
529 done:
530 	blk_finish_plug(&plug);
531 }
532 
533 static void pending_bios_fn(struct btrfs_work *work)
534 {
535 	struct btrfs_device *device;
536 
537 	device = container_of(work, struct btrfs_device, work);
538 	run_scheduled_bios(device);
539 }
540 
541 
542 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
543 {
544 	struct btrfs_fs_devices *fs_devs;
545 	struct btrfs_device *dev;
546 
547 	if (!cur_dev->name)
548 		return;
549 
550 	list_for_each_entry(fs_devs, &fs_uuids, list) {
551 		int del = 1;
552 
553 		if (fs_devs->opened)
554 			continue;
555 		if (fs_devs->seeding)
556 			continue;
557 
558 		list_for_each_entry(dev, &fs_devs->devices, dev_list) {
559 
560 			if (dev == cur_dev)
561 				continue;
562 			if (!dev->name)
563 				continue;
564 
565 			/*
566 			 * Todo: This won't be enough. What if the same device
567 			 * comes back (with new uuid and) with its mapper path?
568 			 * But for now, this does help as mostly an admin will
569 			 * either use mapper or non mapper path throughout.
570 			 */
571 			rcu_read_lock();
572 			del = strcmp(rcu_str_deref(dev->name),
573 						rcu_str_deref(cur_dev->name));
574 			rcu_read_unlock();
575 			if (!del)
576 				break;
577 		}
578 
579 		if (!del) {
580 			/* delete the stale device */
581 			if (fs_devs->num_devices == 1) {
582 				btrfs_sysfs_remove_fsid(fs_devs);
583 				list_del(&fs_devs->list);
584 				free_fs_devices(fs_devs);
585 			} else {
586 				fs_devs->num_devices--;
587 				list_del(&dev->dev_list);
588 				rcu_string_free(dev->name);
589 				kfree(dev);
590 			}
591 			break;
592 		}
593 	}
594 }
595 
596 /*
597  * Add new device to list of registered devices
598  *
599  * Returns:
600  * 1   - first time device is seen
601  * 0   - device already known
602  * < 0 - error
603  */
604 static noinline int device_list_add(const char *path,
605 			   struct btrfs_super_block *disk_super,
606 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
607 {
608 	struct btrfs_device *device;
609 	struct btrfs_fs_devices *fs_devices;
610 	struct rcu_string *name;
611 	int ret = 0;
612 	u64 found_transid = btrfs_super_generation(disk_super);
613 
614 	fs_devices = find_fsid(disk_super->fsid);
615 	if (!fs_devices) {
616 		fs_devices = alloc_fs_devices(disk_super->fsid);
617 		if (IS_ERR(fs_devices))
618 			return PTR_ERR(fs_devices);
619 
620 		list_add(&fs_devices->list, &fs_uuids);
621 
622 		device = NULL;
623 	} else {
624 		device = __find_device(&fs_devices->devices, devid,
625 				       disk_super->dev_item.uuid);
626 	}
627 
628 	if (!device) {
629 		if (fs_devices->opened)
630 			return -EBUSY;
631 
632 		device = btrfs_alloc_device(NULL, &devid,
633 					    disk_super->dev_item.uuid);
634 		if (IS_ERR(device)) {
635 			/* we can safely leave the fs_devices entry around */
636 			return PTR_ERR(device);
637 		}
638 
639 		name = rcu_string_strdup(path, GFP_NOFS);
640 		if (!name) {
641 			kfree(device);
642 			return -ENOMEM;
643 		}
644 		rcu_assign_pointer(device->name, name);
645 
646 		mutex_lock(&fs_devices->device_list_mutex);
647 		list_add_rcu(&device->dev_list, &fs_devices->devices);
648 		fs_devices->num_devices++;
649 		mutex_unlock(&fs_devices->device_list_mutex);
650 
651 		ret = 1;
652 		device->fs_devices = fs_devices;
653 	} else if (!device->name || strcmp(device->name->str, path)) {
654 		/*
655 		 * When FS is already mounted.
656 		 * 1. If you are here and if the device->name is NULL that
657 		 *    means this device was missing at time of FS mount.
658 		 * 2. If you are here and if the device->name is different
659 		 *    from 'path' that means either
660 		 *      a. The same device disappeared and reappeared with
661 		 *         different name. or
662 		 *      b. The missing-disk-which-was-replaced, has
663 		 *         reappeared now.
664 		 *
665 		 * We must allow 1 and 2a above. But 2b would be a spurious
666 		 * and unintentional.
667 		 *
668 		 * Further in case of 1 and 2a above, the disk at 'path'
669 		 * would have missed some transaction when it was away and
670 		 * in case of 2a the stale bdev has to be updated as well.
671 		 * 2b must not be allowed at all time.
672 		 */
673 
674 		/*
675 		 * For now, we do allow update to btrfs_fs_device through the
676 		 * btrfs dev scan cli after FS has been mounted.  We're still
677 		 * tracking a problem where systems fail mount by subvolume id
678 		 * when we reject replacement on a mounted FS.
679 		 */
680 		if (!fs_devices->opened && found_transid < device->generation) {
681 			/*
682 			 * That is if the FS is _not_ mounted and if you
683 			 * are here, that means there is more than one
684 			 * disk with same uuid and devid.We keep the one
685 			 * with larger generation number or the last-in if
686 			 * generation are equal.
687 			 */
688 			return -EEXIST;
689 		}
690 
691 		name = rcu_string_strdup(path, GFP_NOFS);
692 		if (!name)
693 			return -ENOMEM;
694 		rcu_string_free(device->name);
695 		rcu_assign_pointer(device->name, name);
696 		if (device->missing) {
697 			fs_devices->missing_devices--;
698 			device->missing = 0;
699 		}
700 	}
701 
702 	/*
703 	 * Unmount does not free the btrfs_device struct but would zero
704 	 * generation along with most of the other members. So just update
705 	 * it back. We need it to pick the disk with largest generation
706 	 * (as above).
707 	 */
708 	if (!fs_devices->opened)
709 		device->generation = found_transid;
710 
711 	/*
712 	 * if there is new btrfs on an already registered device,
713 	 * then remove the stale device entry.
714 	 */
715 	if (ret > 0)
716 		btrfs_free_stale_device(device);
717 
718 	*fs_devices_ret = fs_devices;
719 
720 	return ret;
721 }
722 
723 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
724 {
725 	struct btrfs_fs_devices *fs_devices;
726 	struct btrfs_device *device;
727 	struct btrfs_device *orig_dev;
728 
729 	fs_devices = alloc_fs_devices(orig->fsid);
730 	if (IS_ERR(fs_devices))
731 		return fs_devices;
732 
733 	mutex_lock(&orig->device_list_mutex);
734 	fs_devices->total_devices = orig->total_devices;
735 
736 	/* We have held the volume lock, it is safe to get the devices. */
737 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
738 		struct rcu_string *name;
739 
740 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
741 					    orig_dev->uuid);
742 		if (IS_ERR(device))
743 			goto error;
744 
745 		/*
746 		 * This is ok to do without rcu read locked because we hold the
747 		 * uuid mutex so nothing we touch in here is going to disappear.
748 		 */
749 		if (orig_dev->name) {
750 			name = rcu_string_strdup(orig_dev->name->str,
751 					GFP_KERNEL);
752 			if (!name) {
753 				kfree(device);
754 				goto error;
755 			}
756 			rcu_assign_pointer(device->name, name);
757 		}
758 
759 		list_add(&device->dev_list, &fs_devices->devices);
760 		device->fs_devices = fs_devices;
761 		fs_devices->num_devices++;
762 	}
763 	mutex_unlock(&orig->device_list_mutex);
764 	return fs_devices;
765 error:
766 	mutex_unlock(&orig->device_list_mutex);
767 	free_fs_devices(fs_devices);
768 	return ERR_PTR(-ENOMEM);
769 }
770 
771 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
772 {
773 	struct btrfs_device *device, *next;
774 	struct btrfs_device *latest_dev = NULL;
775 
776 	mutex_lock(&uuid_mutex);
777 again:
778 	/* This is the initialized path, it is safe to release the devices. */
779 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
780 		if (device->in_fs_metadata) {
781 			if (!device->is_tgtdev_for_dev_replace &&
782 			    (!latest_dev ||
783 			     device->generation > latest_dev->generation)) {
784 				latest_dev = device;
785 			}
786 			continue;
787 		}
788 
789 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
790 			/*
791 			 * In the first step, keep the device which has
792 			 * the correct fsid and the devid that is used
793 			 * for the dev_replace procedure.
794 			 * In the second step, the dev_replace state is
795 			 * read from the device tree and it is known
796 			 * whether the procedure is really active or
797 			 * not, which means whether this device is
798 			 * used or whether it should be removed.
799 			 */
800 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
801 				continue;
802 			}
803 		}
804 		if (device->bdev) {
805 			blkdev_put(device->bdev, device->mode);
806 			device->bdev = NULL;
807 			fs_devices->open_devices--;
808 		}
809 		if (device->writeable) {
810 			list_del_init(&device->dev_alloc_list);
811 			device->writeable = 0;
812 			if (!device->is_tgtdev_for_dev_replace)
813 				fs_devices->rw_devices--;
814 		}
815 		list_del_init(&device->dev_list);
816 		fs_devices->num_devices--;
817 		rcu_string_free(device->name);
818 		kfree(device);
819 	}
820 
821 	if (fs_devices->seed) {
822 		fs_devices = fs_devices->seed;
823 		goto again;
824 	}
825 
826 	fs_devices->latest_bdev = latest_dev->bdev;
827 
828 	mutex_unlock(&uuid_mutex);
829 }
830 
831 static void __free_device(struct work_struct *work)
832 {
833 	struct btrfs_device *device;
834 
835 	device = container_of(work, struct btrfs_device, rcu_work);
836 	rcu_string_free(device->name);
837 	kfree(device);
838 }
839 
840 static void free_device(struct rcu_head *head)
841 {
842 	struct btrfs_device *device;
843 
844 	device = container_of(head, struct btrfs_device, rcu);
845 
846 	INIT_WORK(&device->rcu_work, __free_device);
847 	schedule_work(&device->rcu_work);
848 }
849 
850 static void btrfs_close_bdev(struct btrfs_device *device)
851 {
852 	if (device->bdev && device->writeable) {
853 		sync_blockdev(device->bdev);
854 		invalidate_bdev(device->bdev);
855 	}
856 
857 	if (device->bdev)
858 		blkdev_put(device->bdev, device->mode);
859 }
860 
861 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
862 {
863 	struct btrfs_fs_devices *fs_devices = device->fs_devices;
864 	struct btrfs_device *new_device;
865 	struct rcu_string *name;
866 
867 	if (device->bdev)
868 		fs_devices->open_devices--;
869 
870 	if (device->writeable &&
871 	    device->devid != BTRFS_DEV_REPLACE_DEVID) {
872 		list_del_init(&device->dev_alloc_list);
873 		fs_devices->rw_devices--;
874 	}
875 
876 	if (device->missing)
877 		fs_devices->missing_devices--;
878 
879 	new_device = btrfs_alloc_device(NULL, &device->devid,
880 					device->uuid);
881 	BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
882 
883 	/* Safe because we are under uuid_mutex */
884 	if (device->name) {
885 		name = rcu_string_strdup(device->name->str, GFP_NOFS);
886 		BUG_ON(!name); /* -ENOMEM */
887 		rcu_assign_pointer(new_device->name, name);
888 	}
889 
890 	list_replace_rcu(&device->dev_list, &new_device->dev_list);
891 	new_device->fs_devices = device->fs_devices;
892 }
893 
894 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
895 {
896 	struct btrfs_device *device, *tmp;
897 	struct list_head pending_put;
898 
899 	INIT_LIST_HEAD(&pending_put);
900 
901 	if (--fs_devices->opened > 0)
902 		return 0;
903 
904 	mutex_lock(&fs_devices->device_list_mutex);
905 	list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
906 		btrfs_prepare_close_one_device(device);
907 		list_add(&device->dev_list, &pending_put);
908 	}
909 	mutex_unlock(&fs_devices->device_list_mutex);
910 
911 	/*
912 	 * btrfs_show_devname() is using the device_list_mutex,
913 	 * sometimes call to blkdev_put() leads vfs calling
914 	 * into this func. So do put outside of device_list_mutex,
915 	 * as of now.
916 	 */
917 	while (!list_empty(&pending_put)) {
918 		device = list_first_entry(&pending_put,
919 				struct btrfs_device, dev_list);
920 		list_del(&device->dev_list);
921 		btrfs_close_bdev(device);
922 		call_rcu(&device->rcu, free_device);
923 	}
924 
925 	WARN_ON(fs_devices->open_devices);
926 	WARN_ON(fs_devices->rw_devices);
927 	fs_devices->opened = 0;
928 	fs_devices->seeding = 0;
929 
930 	return 0;
931 }
932 
933 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
934 {
935 	struct btrfs_fs_devices *seed_devices = NULL;
936 	int ret;
937 
938 	mutex_lock(&uuid_mutex);
939 	ret = __btrfs_close_devices(fs_devices);
940 	if (!fs_devices->opened) {
941 		seed_devices = fs_devices->seed;
942 		fs_devices->seed = NULL;
943 	}
944 	mutex_unlock(&uuid_mutex);
945 
946 	while (seed_devices) {
947 		fs_devices = seed_devices;
948 		seed_devices = fs_devices->seed;
949 		__btrfs_close_devices(fs_devices);
950 		free_fs_devices(fs_devices);
951 	}
952 	/*
953 	 * Wait for rcu kworkers under __btrfs_close_devices
954 	 * to finish all blkdev_puts so device is really
955 	 * free when umount is done.
956 	 */
957 	rcu_barrier();
958 	return ret;
959 }
960 
961 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
962 				fmode_t flags, void *holder)
963 {
964 	struct request_queue *q;
965 	struct block_device *bdev;
966 	struct list_head *head = &fs_devices->devices;
967 	struct btrfs_device *device;
968 	struct btrfs_device *latest_dev = NULL;
969 	struct buffer_head *bh;
970 	struct btrfs_super_block *disk_super;
971 	u64 devid;
972 	int seeding = 1;
973 	int ret = 0;
974 
975 	flags |= FMODE_EXCL;
976 
977 	list_for_each_entry(device, head, dev_list) {
978 		if (device->bdev)
979 			continue;
980 		if (!device->name)
981 			continue;
982 
983 		/* Just open everything we can; ignore failures here */
984 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
985 					    &bdev, &bh))
986 			continue;
987 
988 		disk_super = (struct btrfs_super_block *)bh->b_data;
989 		devid = btrfs_stack_device_id(&disk_super->dev_item);
990 		if (devid != device->devid)
991 			goto error_brelse;
992 
993 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
994 			   BTRFS_UUID_SIZE))
995 			goto error_brelse;
996 
997 		device->generation = btrfs_super_generation(disk_super);
998 		if (!latest_dev ||
999 		    device->generation > latest_dev->generation)
1000 			latest_dev = device;
1001 
1002 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1003 			device->writeable = 0;
1004 		} else {
1005 			device->writeable = !bdev_read_only(bdev);
1006 			seeding = 0;
1007 		}
1008 
1009 		q = bdev_get_queue(bdev);
1010 		if (blk_queue_discard(q))
1011 			device->can_discard = 1;
1012 
1013 		device->bdev = bdev;
1014 		device->in_fs_metadata = 0;
1015 		device->mode = flags;
1016 
1017 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1018 			fs_devices->rotating = 1;
1019 
1020 		fs_devices->open_devices++;
1021 		if (device->writeable &&
1022 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
1023 			fs_devices->rw_devices++;
1024 			list_add(&device->dev_alloc_list,
1025 				 &fs_devices->alloc_list);
1026 		}
1027 		brelse(bh);
1028 		continue;
1029 
1030 error_brelse:
1031 		brelse(bh);
1032 		blkdev_put(bdev, flags);
1033 		continue;
1034 	}
1035 	if (fs_devices->open_devices == 0) {
1036 		ret = -EINVAL;
1037 		goto out;
1038 	}
1039 	fs_devices->seeding = seeding;
1040 	fs_devices->opened = 1;
1041 	fs_devices->latest_bdev = latest_dev->bdev;
1042 	fs_devices->total_rw_bytes = 0;
1043 out:
1044 	return ret;
1045 }
1046 
1047 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1048 		       fmode_t flags, void *holder)
1049 {
1050 	int ret;
1051 
1052 	mutex_lock(&uuid_mutex);
1053 	if (fs_devices->opened) {
1054 		fs_devices->opened++;
1055 		ret = 0;
1056 	} else {
1057 		ret = __btrfs_open_devices(fs_devices, flags, holder);
1058 	}
1059 	mutex_unlock(&uuid_mutex);
1060 	return ret;
1061 }
1062 
1063 void btrfs_release_disk_super(struct page *page)
1064 {
1065 	kunmap(page);
1066 	put_page(page);
1067 }
1068 
1069 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1070 		struct page **page, struct btrfs_super_block **disk_super)
1071 {
1072 	void *p;
1073 	pgoff_t index;
1074 
1075 	/* make sure our super fits in the device */
1076 	if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1077 		return 1;
1078 
1079 	/* make sure our super fits in the page */
1080 	if (sizeof(**disk_super) > PAGE_SIZE)
1081 		return 1;
1082 
1083 	/* make sure our super doesn't straddle pages on disk */
1084 	index = bytenr >> PAGE_SHIFT;
1085 	if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1086 		return 1;
1087 
1088 	/* pull in the page with our super */
1089 	*page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1090 				   index, GFP_KERNEL);
1091 
1092 	if (IS_ERR_OR_NULL(*page))
1093 		return 1;
1094 
1095 	p = kmap(*page);
1096 
1097 	/* align our pointer to the offset of the super block */
1098 	*disk_super = p + (bytenr & ~PAGE_MASK);
1099 
1100 	if (btrfs_super_bytenr(*disk_super) != bytenr ||
1101 	    btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1102 		btrfs_release_disk_super(*page);
1103 		return 1;
1104 	}
1105 
1106 	if ((*disk_super)->label[0] &&
1107 		(*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1108 		(*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1109 
1110 	return 0;
1111 }
1112 
1113 /*
1114  * Look for a btrfs signature on a device. This may be called out of the mount path
1115  * and we are not allowed to call set_blocksize during the scan. The superblock
1116  * is read via pagecache
1117  */
1118 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1119 			  struct btrfs_fs_devices **fs_devices_ret)
1120 {
1121 	struct btrfs_super_block *disk_super;
1122 	struct block_device *bdev;
1123 	struct page *page;
1124 	int ret = -EINVAL;
1125 	u64 devid;
1126 	u64 transid;
1127 	u64 total_devices;
1128 	u64 bytenr;
1129 
1130 	/*
1131 	 * we would like to check all the supers, but that would make
1132 	 * a btrfs mount succeed after a mkfs from a different FS.
1133 	 * So, we need to add a special mount option to scan for
1134 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1135 	 */
1136 	bytenr = btrfs_sb_offset(0);
1137 	flags |= FMODE_EXCL;
1138 	mutex_lock(&uuid_mutex);
1139 
1140 	bdev = blkdev_get_by_path(path, flags, holder);
1141 	if (IS_ERR(bdev)) {
1142 		ret = PTR_ERR(bdev);
1143 		goto error;
1144 	}
1145 
1146 	if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1147 		goto error_bdev_put;
1148 
1149 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1150 	transid = btrfs_super_generation(disk_super);
1151 	total_devices = btrfs_super_num_devices(disk_super);
1152 
1153 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1154 	if (ret > 0) {
1155 		if (disk_super->label[0]) {
1156 			pr_info("BTRFS: device label %s ", disk_super->label);
1157 		} else {
1158 			pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1159 		}
1160 
1161 		pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1162 		ret = 0;
1163 	}
1164 	if (!ret && fs_devices_ret)
1165 		(*fs_devices_ret)->total_devices = total_devices;
1166 
1167 	btrfs_release_disk_super(page);
1168 
1169 error_bdev_put:
1170 	blkdev_put(bdev, flags);
1171 error:
1172 	mutex_unlock(&uuid_mutex);
1173 	return ret;
1174 }
1175 
1176 /* helper to account the used device space in the range */
1177 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1178 				   u64 end, u64 *length)
1179 {
1180 	struct btrfs_key key;
1181 	struct btrfs_root *root = device->fs_info->dev_root;
1182 	struct btrfs_dev_extent *dev_extent;
1183 	struct btrfs_path *path;
1184 	u64 extent_end;
1185 	int ret;
1186 	int slot;
1187 	struct extent_buffer *l;
1188 
1189 	*length = 0;
1190 
1191 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1192 		return 0;
1193 
1194 	path = btrfs_alloc_path();
1195 	if (!path)
1196 		return -ENOMEM;
1197 	path->reada = READA_FORWARD;
1198 
1199 	key.objectid = device->devid;
1200 	key.offset = start;
1201 	key.type = BTRFS_DEV_EXTENT_KEY;
1202 
1203 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1204 	if (ret < 0)
1205 		goto out;
1206 	if (ret > 0) {
1207 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1208 		if (ret < 0)
1209 			goto out;
1210 	}
1211 
1212 	while (1) {
1213 		l = path->nodes[0];
1214 		slot = path->slots[0];
1215 		if (slot >= btrfs_header_nritems(l)) {
1216 			ret = btrfs_next_leaf(root, path);
1217 			if (ret == 0)
1218 				continue;
1219 			if (ret < 0)
1220 				goto out;
1221 
1222 			break;
1223 		}
1224 		btrfs_item_key_to_cpu(l, &key, slot);
1225 
1226 		if (key.objectid < device->devid)
1227 			goto next;
1228 
1229 		if (key.objectid > device->devid)
1230 			break;
1231 
1232 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1233 			goto next;
1234 
1235 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1236 		extent_end = key.offset + btrfs_dev_extent_length(l,
1237 								  dev_extent);
1238 		if (key.offset <= start && extent_end > end) {
1239 			*length = end - start + 1;
1240 			break;
1241 		} else if (key.offset <= start && extent_end > start)
1242 			*length += extent_end - start;
1243 		else if (key.offset > start && extent_end <= end)
1244 			*length += extent_end - key.offset;
1245 		else if (key.offset > start && key.offset <= end) {
1246 			*length += end - key.offset + 1;
1247 			break;
1248 		} else if (key.offset > end)
1249 			break;
1250 
1251 next:
1252 		path->slots[0]++;
1253 	}
1254 	ret = 0;
1255 out:
1256 	btrfs_free_path(path);
1257 	return ret;
1258 }
1259 
1260 static int contains_pending_extent(struct btrfs_transaction *transaction,
1261 				   struct btrfs_device *device,
1262 				   u64 *start, u64 len)
1263 {
1264 	struct btrfs_fs_info *fs_info = device->fs_info;
1265 	struct extent_map *em;
1266 	struct list_head *search_list = &fs_info->pinned_chunks;
1267 	int ret = 0;
1268 	u64 physical_start = *start;
1269 
1270 	if (transaction)
1271 		search_list = &transaction->pending_chunks;
1272 again:
1273 	list_for_each_entry(em, search_list, list) {
1274 		struct map_lookup *map;
1275 		int i;
1276 
1277 		map = em->map_lookup;
1278 		for (i = 0; i < map->num_stripes; i++) {
1279 			u64 end;
1280 
1281 			if (map->stripes[i].dev != device)
1282 				continue;
1283 			if (map->stripes[i].physical >= physical_start + len ||
1284 			    map->stripes[i].physical + em->orig_block_len <=
1285 			    physical_start)
1286 				continue;
1287 			/*
1288 			 * Make sure that while processing the pinned list we do
1289 			 * not override our *start with a lower value, because
1290 			 * we can have pinned chunks that fall within this
1291 			 * device hole and that have lower physical addresses
1292 			 * than the pending chunks we processed before. If we
1293 			 * do not take this special care we can end up getting
1294 			 * 2 pending chunks that start at the same physical
1295 			 * device offsets because the end offset of a pinned
1296 			 * chunk can be equal to the start offset of some
1297 			 * pending chunk.
1298 			 */
1299 			end = map->stripes[i].physical + em->orig_block_len;
1300 			if (end > *start) {
1301 				*start = end;
1302 				ret = 1;
1303 			}
1304 		}
1305 	}
1306 	if (search_list != &fs_info->pinned_chunks) {
1307 		search_list = &fs_info->pinned_chunks;
1308 		goto again;
1309 	}
1310 
1311 	return ret;
1312 }
1313 
1314 
1315 /*
1316  * find_free_dev_extent_start - find free space in the specified device
1317  * @device:	  the device which we search the free space in
1318  * @num_bytes:	  the size of the free space that we need
1319  * @search_start: the position from which to begin the search
1320  * @start:	  store the start of the free space.
1321  * @len:	  the size of the free space. that we find, or the size
1322  *		  of the max free space if we don't find suitable free space
1323  *
1324  * this uses a pretty simple search, the expectation is that it is
1325  * called very infrequently and that a given device has a small number
1326  * of extents
1327  *
1328  * @start is used to store the start of the free space if we find. But if we
1329  * don't find suitable free space, it will be used to store the start position
1330  * of the max free space.
1331  *
1332  * @len is used to store the size of the free space that we find.
1333  * But if we don't find suitable free space, it is used to store the size of
1334  * the max free space.
1335  */
1336 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1337 			       struct btrfs_device *device, u64 num_bytes,
1338 			       u64 search_start, u64 *start, u64 *len)
1339 {
1340 	struct btrfs_fs_info *fs_info = device->fs_info;
1341 	struct btrfs_root *root = fs_info->dev_root;
1342 	struct btrfs_key key;
1343 	struct btrfs_dev_extent *dev_extent;
1344 	struct btrfs_path *path;
1345 	u64 hole_size;
1346 	u64 max_hole_start;
1347 	u64 max_hole_size;
1348 	u64 extent_end;
1349 	u64 search_end = device->total_bytes;
1350 	int ret;
1351 	int slot;
1352 	struct extent_buffer *l;
1353 	u64 min_search_start;
1354 
1355 	/*
1356 	 * We don't want to overwrite the superblock on the drive nor any area
1357 	 * used by the boot loader (grub for example), so we make sure to start
1358 	 * at an offset of at least 1MB.
1359 	 */
1360 	min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1361 	search_start = max(search_start, min_search_start);
1362 
1363 	path = btrfs_alloc_path();
1364 	if (!path)
1365 		return -ENOMEM;
1366 
1367 	max_hole_start = search_start;
1368 	max_hole_size = 0;
1369 
1370 again:
1371 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1372 		ret = -ENOSPC;
1373 		goto out;
1374 	}
1375 
1376 	path->reada = READA_FORWARD;
1377 	path->search_commit_root = 1;
1378 	path->skip_locking = 1;
1379 
1380 	key.objectid = device->devid;
1381 	key.offset = search_start;
1382 	key.type = BTRFS_DEV_EXTENT_KEY;
1383 
1384 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1385 	if (ret < 0)
1386 		goto out;
1387 	if (ret > 0) {
1388 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1389 		if (ret < 0)
1390 			goto out;
1391 	}
1392 
1393 	while (1) {
1394 		l = path->nodes[0];
1395 		slot = path->slots[0];
1396 		if (slot >= btrfs_header_nritems(l)) {
1397 			ret = btrfs_next_leaf(root, path);
1398 			if (ret == 0)
1399 				continue;
1400 			if (ret < 0)
1401 				goto out;
1402 
1403 			break;
1404 		}
1405 		btrfs_item_key_to_cpu(l, &key, slot);
1406 
1407 		if (key.objectid < device->devid)
1408 			goto next;
1409 
1410 		if (key.objectid > device->devid)
1411 			break;
1412 
1413 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1414 			goto next;
1415 
1416 		if (key.offset > search_start) {
1417 			hole_size = key.offset - search_start;
1418 
1419 			/*
1420 			 * Have to check before we set max_hole_start, otherwise
1421 			 * we could end up sending back this offset anyway.
1422 			 */
1423 			if (contains_pending_extent(transaction, device,
1424 						    &search_start,
1425 						    hole_size)) {
1426 				if (key.offset >= search_start) {
1427 					hole_size = key.offset - search_start;
1428 				} else {
1429 					WARN_ON_ONCE(1);
1430 					hole_size = 0;
1431 				}
1432 			}
1433 
1434 			if (hole_size > max_hole_size) {
1435 				max_hole_start = search_start;
1436 				max_hole_size = hole_size;
1437 			}
1438 
1439 			/*
1440 			 * If this free space is greater than which we need,
1441 			 * it must be the max free space that we have found
1442 			 * until now, so max_hole_start must point to the start
1443 			 * of this free space and the length of this free space
1444 			 * is stored in max_hole_size. Thus, we return
1445 			 * max_hole_start and max_hole_size and go back to the
1446 			 * caller.
1447 			 */
1448 			if (hole_size >= num_bytes) {
1449 				ret = 0;
1450 				goto out;
1451 			}
1452 		}
1453 
1454 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1455 		extent_end = key.offset + btrfs_dev_extent_length(l,
1456 								  dev_extent);
1457 		if (extent_end > search_start)
1458 			search_start = extent_end;
1459 next:
1460 		path->slots[0]++;
1461 		cond_resched();
1462 	}
1463 
1464 	/*
1465 	 * At this point, search_start should be the end of
1466 	 * allocated dev extents, and when shrinking the device,
1467 	 * search_end may be smaller than search_start.
1468 	 */
1469 	if (search_end > search_start) {
1470 		hole_size = search_end - search_start;
1471 
1472 		if (contains_pending_extent(transaction, device, &search_start,
1473 					    hole_size)) {
1474 			btrfs_release_path(path);
1475 			goto again;
1476 		}
1477 
1478 		if (hole_size > max_hole_size) {
1479 			max_hole_start = search_start;
1480 			max_hole_size = hole_size;
1481 		}
1482 	}
1483 
1484 	/* See above. */
1485 	if (max_hole_size < num_bytes)
1486 		ret = -ENOSPC;
1487 	else
1488 		ret = 0;
1489 
1490 out:
1491 	btrfs_free_path(path);
1492 	*start = max_hole_start;
1493 	if (len)
1494 		*len = max_hole_size;
1495 	return ret;
1496 }
1497 
1498 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1499 			 struct btrfs_device *device, u64 num_bytes,
1500 			 u64 *start, u64 *len)
1501 {
1502 	/* FIXME use last free of some kind */
1503 	return find_free_dev_extent_start(trans->transaction, device,
1504 					  num_bytes, 0, start, len);
1505 }
1506 
1507 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1508 			  struct btrfs_device *device,
1509 			  u64 start, u64 *dev_extent_len)
1510 {
1511 	struct btrfs_fs_info *fs_info = device->fs_info;
1512 	struct btrfs_root *root = fs_info->dev_root;
1513 	int ret;
1514 	struct btrfs_path *path;
1515 	struct btrfs_key key;
1516 	struct btrfs_key found_key;
1517 	struct extent_buffer *leaf = NULL;
1518 	struct btrfs_dev_extent *extent = NULL;
1519 
1520 	path = btrfs_alloc_path();
1521 	if (!path)
1522 		return -ENOMEM;
1523 
1524 	key.objectid = device->devid;
1525 	key.offset = start;
1526 	key.type = BTRFS_DEV_EXTENT_KEY;
1527 again:
1528 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1529 	if (ret > 0) {
1530 		ret = btrfs_previous_item(root, path, key.objectid,
1531 					  BTRFS_DEV_EXTENT_KEY);
1532 		if (ret)
1533 			goto out;
1534 		leaf = path->nodes[0];
1535 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1536 		extent = btrfs_item_ptr(leaf, path->slots[0],
1537 					struct btrfs_dev_extent);
1538 		BUG_ON(found_key.offset > start || found_key.offset +
1539 		       btrfs_dev_extent_length(leaf, extent) < start);
1540 		key = found_key;
1541 		btrfs_release_path(path);
1542 		goto again;
1543 	} else if (ret == 0) {
1544 		leaf = path->nodes[0];
1545 		extent = btrfs_item_ptr(leaf, path->slots[0],
1546 					struct btrfs_dev_extent);
1547 	} else {
1548 		btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1549 		goto out;
1550 	}
1551 
1552 	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1553 
1554 	ret = btrfs_del_item(trans, root, path);
1555 	if (ret) {
1556 		btrfs_handle_fs_error(fs_info, ret,
1557 				      "Failed to remove dev extent item");
1558 	} else {
1559 		set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1560 	}
1561 out:
1562 	btrfs_free_path(path);
1563 	return ret;
1564 }
1565 
1566 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1567 				  struct btrfs_device *device,
1568 				  u64 chunk_tree, u64 chunk_objectid,
1569 				  u64 chunk_offset, u64 start, u64 num_bytes)
1570 {
1571 	int ret;
1572 	struct btrfs_path *path;
1573 	struct btrfs_fs_info *fs_info = device->fs_info;
1574 	struct btrfs_root *root = fs_info->dev_root;
1575 	struct btrfs_dev_extent *extent;
1576 	struct extent_buffer *leaf;
1577 	struct btrfs_key key;
1578 
1579 	WARN_ON(!device->in_fs_metadata);
1580 	WARN_ON(device->is_tgtdev_for_dev_replace);
1581 	path = btrfs_alloc_path();
1582 	if (!path)
1583 		return -ENOMEM;
1584 
1585 	key.objectid = device->devid;
1586 	key.offset = start;
1587 	key.type = BTRFS_DEV_EXTENT_KEY;
1588 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1589 				      sizeof(*extent));
1590 	if (ret)
1591 		goto out;
1592 
1593 	leaf = path->nodes[0];
1594 	extent = btrfs_item_ptr(leaf, path->slots[0],
1595 				struct btrfs_dev_extent);
1596 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1597 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1598 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1599 
1600 	write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1601 
1602 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1603 	btrfs_mark_buffer_dirty(leaf);
1604 out:
1605 	btrfs_free_path(path);
1606 	return ret;
1607 }
1608 
1609 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1610 {
1611 	struct extent_map_tree *em_tree;
1612 	struct extent_map *em;
1613 	struct rb_node *n;
1614 	u64 ret = 0;
1615 
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;
1622 	}
1623 	read_unlock(&em_tree->lock);
1624 
1625 	return ret;
1626 }
1627 
1628 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1629 				    u64 *devid_ret)
1630 {
1631 	int ret;
1632 	struct btrfs_key key;
1633 	struct btrfs_key found_key;
1634 	struct btrfs_path *path;
1635 
1636 	path = btrfs_alloc_path();
1637 	if (!path)
1638 		return -ENOMEM;
1639 
1640 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1641 	key.type = BTRFS_DEV_ITEM_KEY;
1642 	key.offset = (u64)-1;
1643 
1644 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1645 	if (ret < 0)
1646 		goto error;
1647 
1648 	BUG_ON(ret == 0); /* Corruption */
1649 
1650 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1651 				  BTRFS_DEV_ITEMS_OBJECTID,
1652 				  BTRFS_DEV_ITEM_KEY);
1653 	if (ret) {
1654 		*devid_ret = 1;
1655 	} else {
1656 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1657 				      path->slots[0]);
1658 		*devid_ret = found_key.offset + 1;
1659 	}
1660 	ret = 0;
1661 error:
1662 	btrfs_free_path(path);
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  */
1670 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1671 			    struct btrfs_fs_info *fs_info,
1672 			    struct btrfs_device *device)
1673 {
1674 	struct btrfs_root *root = fs_info->chunk_root;
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;
1681 
1682 	path = btrfs_alloc_path();
1683 	if (!path)
1684 		return -ENOMEM;
1685 
1686 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687 	key.type = BTRFS_DEV_ITEM_KEY;
1688 	key.offset = device->devid;
1689 
1690 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1691 				      sizeof(*dev_item));
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);
1699 	btrfs_set_device_generation(leaf, dev_item, 0);
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);
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));
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);
1711 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1712 
1713 	ptr = btrfs_device_uuid(dev_item);
1714 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1715 	ptr = btrfs_device_fsid(dev_item);
1716 	write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1717 	btrfs_mark_buffer_dirty(leaf);
1718 
1719 	ret = 0;
1720 out:
1721 	btrfs_free_path(path);
1722 	return ret;
1723 }
1724 
1725 /*
1726  * Function to update ctime/mtime for a given device path.
1727  * Mainly used for ctime/mtime based probe like libblkid.
1728  */
1729 static void update_dev_time(char *path_name)
1730 {
1731 	struct file *filp;
1732 
1733 	filp = filp_open(path_name, O_RDWR, 0);
1734 	if (IS_ERR(filp))
1735 		return;
1736 	file_update_time(filp);
1737 	filp_close(filp, NULL);
1738 }
1739 
1740 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1741 			     struct btrfs_device *device)
1742 {
1743 	struct btrfs_root *root = fs_info->chunk_root;
1744 	int ret;
1745 	struct btrfs_path *path;
1746 	struct btrfs_key key;
1747 	struct btrfs_trans_handle *trans;
1748 
1749 	path = btrfs_alloc_path();
1750 	if (!path)
1751 		return -ENOMEM;
1752 
1753 	trans = btrfs_start_transaction(root, 0);
1754 	if (IS_ERR(trans)) {
1755 		btrfs_free_path(path);
1756 		return PTR_ERR(trans);
1757 	}
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;
1774 out:
1775 	btrfs_free_path(path);
1776 	btrfs_commit_transaction(trans);
1777 	return ret;
1778 }
1779 
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  */
1785 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1786 		u64 num_devices)
1787 {
1788 	u64 all_avail;
1789 	unsigned seq;
1790 	int i;
1791 
1792 	do {
1793 		seq = read_seqbegin(&fs_info->profiles_lock);
1794 
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));
1799 
1800 	for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1801 		if (!(all_avail & btrfs_raid_group[i]))
1802 			continue;
1803 
1804 		if (num_devices < btrfs_raid_array[i].devs_min) {
1805 			int ret = btrfs_raid_mindev_error[i];
1806 
1807 			if (ret)
1808 				return ret;
1809 		}
1810 	}
1811 
1812 	return 0;
1813 }
1814 
1815 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1816 					struct btrfs_device *device)
1817 {
1818 	struct btrfs_device *next_device;
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  */
1835 void 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 
1855 int btrfs_rm_device(struct btrfs_fs_info *fs_info, char *device_path, u64 devid)
1856 {
1857 	struct btrfs_device *device;
1858 	struct btrfs_fs_devices *cur_devices;
1859 	u64 num_devices;
1860 	int ret = 0;
1861 	bool clear_super = false;
1862 
1863 	mutex_lock(&uuid_mutex);
1864 
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)) {
1868 		WARN_ON(num_devices < 1);
1869 		num_devices--;
1870 	}
1871 	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1872 
1873 	ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1874 	if (ret)
1875 		goto out;
1876 
1877 	ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1878 					   &device);
1879 	if (ret)
1880 		goto out;
1881 
1882 	if (device->is_tgtdev_for_dev_replace) {
1883 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1884 		goto out;
1885 	}
1886 
1887 	if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1888 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1889 		goto out;
1890 	}
1891 
1892 	if (device->writeable) {
1893 		mutex_lock(&fs_info->chunk_mutex);
1894 		list_del_init(&device->dev_alloc_list);
1895 		device->fs_devices->rw_devices--;
1896 		mutex_unlock(&fs_info->chunk_mutex);
1897 		clear_super = true;
1898 	}
1899 
1900 	mutex_unlock(&uuid_mutex);
1901 	ret = btrfs_shrink_device(device, 0);
1902 	mutex_lock(&uuid_mutex);
1903 	if (ret)
1904 		goto error_undo;
1905 
1906 	/*
1907 	 * TODO: the superblock still includes this device in its num_devices
1908 	 * counter although write_all_supers() is not locked out. This
1909 	 * could give a filesystem state which requires a degraded mount.
1910 	 */
1911 	ret = btrfs_rm_dev_item(fs_info, device);
1912 	if (ret)
1913 		goto error_undo;
1914 
1915 	device->in_fs_metadata = 0;
1916 	btrfs_scrub_cancel_dev(fs_info, device);
1917 
1918 	/*
1919 	 * the device list mutex makes sure that we don't change
1920 	 * the device list while someone else is writing out all
1921 	 * the device supers. Whoever is writing all supers, should
1922 	 * lock the device list mutex before getting the number of
1923 	 * devices in the super block (super_copy). Conversely,
1924 	 * whoever updates the number of devices in the super block
1925 	 * (super_copy) should hold the device list mutex.
1926 	 */
1927 
1928 	cur_devices = device->fs_devices;
1929 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1930 	list_del_rcu(&device->dev_list);
1931 
1932 	device->fs_devices->num_devices--;
1933 	device->fs_devices->total_devices--;
1934 
1935 	if (device->missing)
1936 		device->fs_devices->missing_devices--;
1937 
1938 	btrfs_assign_next_active_device(fs_info, device, NULL);
1939 
1940 	if (device->bdev) {
1941 		device->fs_devices->open_devices--;
1942 		/* remove sysfs entry */
1943 		btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1944 	}
1945 
1946 	num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1947 	btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1948 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1949 
1950 	/*
1951 	 * at this point, the device is zero sized and detached from
1952 	 * the devices list.  All that's left is to zero out the old
1953 	 * supers and free the device.
1954 	 */
1955 	if (device->writeable)
1956 		btrfs_scratch_superblocks(device->bdev, device->name->str);
1957 
1958 	btrfs_close_bdev(device);
1959 	call_rcu(&device->rcu, free_device);
1960 
1961 	if (cur_devices->open_devices == 0) {
1962 		struct btrfs_fs_devices *fs_devices;
1963 		fs_devices = fs_info->fs_devices;
1964 		while (fs_devices) {
1965 			if (fs_devices->seed == cur_devices) {
1966 				fs_devices->seed = cur_devices->seed;
1967 				break;
1968 			}
1969 			fs_devices = fs_devices->seed;
1970 		}
1971 		cur_devices->seed = NULL;
1972 		__btrfs_close_devices(cur_devices);
1973 		free_fs_devices(cur_devices);
1974 	}
1975 
1976 	fs_info->num_tolerated_disk_barrier_failures =
1977 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1978 
1979 out:
1980 	mutex_unlock(&uuid_mutex);
1981 	return ret;
1982 
1983 error_undo:
1984 	if (device->writeable) {
1985 		mutex_lock(&fs_info->chunk_mutex);
1986 		list_add(&device->dev_alloc_list,
1987 			 &fs_info->fs_devices->alloc_list);
1988 		device->fs_devices->rw_devices++;
1989 		mutex_unlock(&fs_info->chunk_mutex);
1990 	}
1991 	goto out;
1992 }
1993 
1994 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1995 					struct btrfs_device *srcdev)
1996 {
1997 	struct btrfs_fs_devices *fs_devices;
1998 
1999 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2000 
2001 	/*
2002 	 * in case of fs with no seed, srcdev->fs_devices will point
2003 	 * to fs_devices of fs_info. However when the dev being replaced is
2004 	 * a seed dev it will point to the seed's local fs_devices. In short
2005 	 * srcdev will have its correct fs_devices in both the cases.
2006 	 */
2007 	fs_devices = srcdev->fs_devices;
2008 
2009 	list_del_rcu(&srcdev->dev_list);
2010 	list_del_rcu(&srcdev->dev_alloc_list);
2011 	fs_devices->num_devices--;
2012 	if (srcdev->missing)
2013 		fs_devices->missing_devices--;
2014 
2015 	if (srcdev->writeable)
2016 		fs_devices->rw_devices--;
2017 
2018 	if (srcdev->bdev)
2019 		fs_devices->open_devices--;
2020 }
2021 
2022 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2023 				      struct btrfs_device *srcdev)
2024 {
2025 	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2026 
2027 	if (srcdev->writeable) {
2028 		/* zero out the old super if it is writable */
2029 		btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2030 	}
2031 
2032 	btrfs_close_bdev(srcdev);
2033 
2034 	call_rcu(&srcdev->rcu, free_device);
2035 
2036 	/*
2037 	 * unless fs_devices is seed fs, num_devices shouldn't go
2038 	 * zero
2039 	 */
2040 	BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2041 
2042 	/* if this is no devs we rather delete the fs_devices */
2043 	if (!fs_devices->num_devices) {
2044 		struct btrfs_fs_devices *tmp_fs_devices;
2045 
2046 		tmp_fs_devices = fs_info->fs_devices;
2047 		while (tmp_fs_devices) {
2048 			if (tmp_fs_devices->seed == fs_devices) {
2049 				tmp_fs_devices->seed = fs_devices->seed;
2050 				break;
2051 			}
2052 			tmp_fs_devices = tmp_fs_devices->seed;
2053 		}
2054 		fs_devices->seed = NULL;
2055 		__btrfs_close_devices(fs_devices);
2056 		free_fs_devices(fs_devices);
2057 	}
2058 }
2059 
2060 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2061 				      struct btrfs_device *tgtdev)
2062 {
2063 	mutex_lock(&uuid_mutex);
2064 	WARN_ON(!tgtdev);
2065 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2066 
2067 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2068 
2069 	if (tgtdev->bdev)
2070 		fs_info->fs_devices->open_devices--;
2071 
2072 	fs_info->fs_devices->num_devices--;
2073 
2074 	btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2075 
2076 	list_del_rcu(&tgtdev->dev_list);
2077 
2078 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2079 	mutex_unlock(&uuid_mutex);
2080 
2081 	/*
2082 	 * The update_dev_time() with in btrfs_scratch_superblocks()
2083 	 * may lead to a call to btrfs_show_devname() which will try
2084 	 * to hold device_list_mutex. And here this device
2085 	 * is already out of device list, so we don't have to hold
2086 	 * the device_list_mutex lock.
2087 	 */
2088 	btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2089 
2090 	btrfs_close_bdev(tgtdev);
2091 	call_rcu(&tgtdev->rcu, free_device);
2092 }
2093 
2094 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2095 				     char *device_path,
2096 				     struct btrfs_device **device)
2097 {
2098 	int ret = 0;
2099 	struct btrfs_super_block *disk_super;
2100 	u64 devid;
2101 	u8 *dev_uuid;
2102 	struct block_device *bdev;
2103 	struct buffer_head *bh;
2104 
2105 	*device = NULL;
2106 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107 				    fs_info->bdev_holder, 0, &bdev, &bh);
2108 	if (ret)
2109 		return ret;
2110 	disk_super = (struct btrfs_super_block *)bh->b_data;
2111 	devid = btrfs_stack_device_id(&disk_super->dev_item);
2112 	dev_uuid = disk_super->dev_item.uuid;
2113 	*device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2114 	brelse(bh);
2115 	if (!*device)
2116 		ret = -ENOENT;
2117 	blkdev_put(bdev, FMODE_READ);
2118 	return ret;
2119 }
2120 
2121 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2122 					 char *device_path,
2123 					 struct btrfs_device **device)
2124 {
2125 	*device = NULL;
2126 	if (strcmp(device_path, "missing") == 0) {
2127 		struct list_head *devices;
2128 		struct btrfs_device *tmp;
2129 
2130 		devices = &fs_info->fs_devices->devices;
2131 		/*
2132 		 * It is safe to read the devices since the volume_mutex
2133 		 * is held by the caller.
2134 		 */
2135 		list_for_each_entry(tmp, devices, dev_list) {
2136 			if (tmp->in_fs_metadata && !tmp->bdev) {
2137 				*device = tmp;
2138 				break;
2139 			}
2140 		}
2141 
2142 		if (!*device)
2143 			return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2144 
2145 		return 0;
2146 	} else {
2147 		return btrfs_find_device_by_path(fs_info, device_path, device);
2148 	}
2149 }
2150 
2151 /*
2152  * Lookup a device given by device id, or the path if the id is 0.
2153  */
2154 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2155 				 char *devpath, struct btrfs_device **device)
2156 {
2157 	int ret;
2158 
2159 	if (devid) {
2160 		ret = 0;
2161 		*device = btrfs_find_device(fs_info, devid, NULL, NULL);
2162 		if (!*device)
2163 			ret = -ENOENT;
2164 	} else {
2165 		if (!devpath || !devpath[0])
2166 			return -EINVAL;
2167 
2168 		ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2169 							   device);
2170 	}
2171 	return ret;
2172 }
2173 
2174 /*
2175  * does all the dirty work required for changing file system's UUID.
2176  */
2177 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2178 {
2179 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2180 	struct btrfs_fs_devices *old_devices;
2181 	struct btrfs_fs_devices *seed_devices;
2182 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2183 	struct btrfs_device *device;
2184 	u64 super_flags;
2185 
2186 	BUG_ON(!mutex_is_locked(&uuid_mutex));
2187 	if (!fs_devices->seeding)
2188 		return -EINVAL;
2189 
2190 	seed_devices = __alloc_fs_devices();
2191 	if (IS_ERR(seed_devices))
2192 		return PTR_ERR(seed_devices);
2193 
2194 	old_devices = clone_fs_devices(fs_devices);
2195 	if (IS_ERR(old_devices)) {
2196 		kfree(seed_devices);
2197 		return PTR_ERR(old_devices);
2198 	}
2199 
2200 	list_add(&old_devices->list, &fs_uuids);
2201 
2202 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2203 	seed_devices->opened = 1;
2204 	INIT_LIST_HEAD(&seed_devices->devices);
2205 	INIT_LIST_HEAD(&seed_devices->alloc_list);
2206 	mutex_init(&seed_devices->device_list_mutex);
2207 
2208 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2209 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2210 			      synchronize_rcu);
2211 	list_for_each_entry(device, &seed_devices->devices, dev_list)
2212 		device->fs_devices = seed_devices;
2213 
2214 	mutex_lock(&fs_info->chunk_mutex);
2215 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2216 	mutex_unlock(&fs_info->chunk_mutex);
2217 
2218 	fs_devices->seeding = 0;
2219 	fs_devices->num_devices = 0;
2220 	fs_devices->open_devices = 0;
2221 	fs_devices->missing_devices = 0;
2222 	fs_devices->rotating = 0;
2223 	fs_devices->seed = seed_devices;
2224 
2225 	generate_random_uuid(fs_devices->fsid);
2226 	memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2227 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2228 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2229 
2230 	super_flags = btrfs_super_flags(disk_super) &
2231 		      ~BTRFS_SUPER_FLAG_SEEDING;
2232 	btrfs_set_super_flags(disk_super, super_flags);
2233 
2234 	return 0;
2235 }
2236 
2237 /*
2238  * Store the expected generation for seed devices in device items.
2239  */
2240 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2241 			       struct btrfs_fs_info *fs_info)
2242 {
2243 	struct btrfs_root *root = fs_info->chunk_root;
2244 	struct btrfs_path *path;
2245 	struct extent_buffer *leaf;
2246 	struct btrfs_dev_item *dev_item;
2247 	struct btrfs_device *device;
2248 	struct btrfs_key key;
2249 	u8 fs_uuid[BTRFS_UUID_SIZE];
2250 	u8 dev_uuid[BTRFS_UUID_SIZE];
2251 	u64 devid;
2252 	int ret;
2253 
2254 	path = btrfs_alloc_path();
2255 	if (!path)
2256 		return -ENOMEM;
2257 
2258 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2259 	key.offset = 0;
2260 	key.type = BTRFS_DEV_ITEM_KEY;
2261 
2262 	while (1) {
2263 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2264 		if (ret < 0)
2265 			goto error;
2266 
2267 		leaf = path->nodes[0];
2268 next_slot:
2269 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2270 			ret = btrfs_next_leaf(root, path);
2271 			if (ret > 0)
2272 				break;
2273 			if (ret < 0)
2274 				goto error;
2275 			leaf = path->nodes[0];
2276 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2277 			btrfs_release_path(path);
2278 			continue;
2279 		}
2280 
2281 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2283 		    key.type != BTRFS_DEV_ITEM_KEY)
2284 			break;
2285 
2286 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2287 					  struct btrfs_dev_item);
2288 		devid = btrfs_device_id(leaf, dev_item);
2289 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2290 				   BTRFS_UUID_SIZE);
2291 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2292 				   BTRFS_UUID_SIZE);
2293 		device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2294 		BUG_ON(!device); /* Logic error */
2295 
2296 		if (device->fs_devices->seeding) {
2297 			btrfs_set_device_generation(leaf, dev_item,
2298 						    device->generation);
2299 			btrfs_mark_buffer_dirty(leaf);
2300 		}
2301 
2302 		path->slots[0]++;
2303 		goto next_slot;
2304 	}
2305 	ret = 0;
2306 error:
2307 	btrfs_free_path(path);
2308 	return ret;
2309 }
2310 
2311 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, char *device_path)
2312 {
2313 	struct btrfs_root *root = fs_info->dev_root;
2314 	struct request_queue *q;
2315 	struct btrfs_trans_handle *trans;
2316 	struct btrfs_device *device;
2317 	struct block_device *bdev;
2318 	struct list_head *devices;
2319 	struct super_block *sb = fs_info->sb;
2320 	struct rcu_string *name;
2321 	u64 tmp;
2322 	int seeding_dev = 0;
2323 	int ret = 0;
2324 
2325 	if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2326 		return -EROFS;
2327 
2328 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2329 				  fs_info->bdev_holder);
2330 	if (IS_ERR(bdev))
2331 		return PTR_ERR(bdev);
2332 
2333 	if (fs_info->fs_devices->seeding) {
2334 		seeding_dev = 1;
2335 		down_write(&sb->s_umount);
2336 		mutex_lock(&uuid_mutex);
2337 	}
2338 
2339 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2340 
2341 	devices = &fs_info->fs_devices->devices;
2342 
2343 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2344 	list_for_each_entry(device, devices, dev_list) {
2345 		if (device->bdev == bdev) {
2346 			ret = -EEXIST;
2347 			mutex_unlock(
2348 				&fs_info->fs_devices->device_list_mutex);
2349 			goto error;
2350 		}
2351 	}
2352 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2353 
2354 	device = btrfs_alloc_device(fs_info, NULL, NULL);
2355 	if (IS_ERR(device)) {
2356 		/* we can safely leave the fs_devices entry around */
2357 		ret = PTR_ERR(device);
2358 		goto error;
2359 	}
2360 
2361 	name = rcu_string_strdup(device_path, GFP_KERNEL);
2362 	if (!name) {
2363 		kfree(device);
2364 		ret = -ENOMEM;
2365 		goto error;
2366 	}
2367 	rcu_assign_pointer(device->name, name);
2368 
2369 	trans = btrfs_start_transaction(root, 0);
2370 	if (IS_ERR(trans)) {
2371 		rcu_string_free(device->name);
2372 		kfree(device);
2373 		ret = PTR_ERR(trans);
2374 		goto error;
2375 	}
2376 
2377 	q = bdev_get_queue(bdev);
2378 	if (blk_queue_discard(q))
2379 		device->can_discard = 1;
2380 	device->writeable = 1;
2381 	device->generation = trans->transid;
2382 	device->io_width = fs_info->sectorsize;
2383 	device->io_align = fs_info->sectorsize;
2384 	device->sector_size = fs_info->sectorsize;
2385 	device->total_bytes = i_size_read(bdev->bd_inode);
2386 	device->disk_total_bytes = device->total_bytes;
2387 	device->commit_total_bytes = device->total_bytes;
2388 	device->fs_info = fs_info;
2389 	device->bdev = bdev;
2390 	device->in_fs_metadata = 1;
2391 	device->is_tgtdev_for_dev_replace = 0;
2392 	device->mode = FMODE_EXCL;
2393 	device->dev_stats_valid = 1;
2394 	set_blocksize(device->bdev, 4096);
2395 
2396 	if (seeding_dev) {
2397 		sb->s_flags &= ~MS_RDONLY;
2398 		ret = btrfs_prepare_sprout(fs_info);
2399 		BUG_ON(ret); /* -ENOMEM */
2400 	}
2401 
2402 	device->fs_devices = fs_info->fs_devices;
2403 
2404 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2405 	mutex_lock(&fs_info->chunk_mutex);
2406 	list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2407 	list_add(&device->dev_alloc_list,
2408 		 &fs_info->fs_devices->alloc_list);
2409 	fs_info->fs_devices->num_devices++;
2410 	fs_info->fs_devices->open_devices++;
2411 	fs_info->fs_devices->rw_devices++;
2412 	fs_info->fs_devices->total_devices++;
2413 	fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2414 
2415 	spin_lock(&fs_info->free_chunk_lock);
2416 	fs_info->free_chunk_space += device->total_bytes;
2417 	spin_unlock(&fs_info->free_chunk_lock);
2418 
2419 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2420 		fs_info->fs_devices->rotating = 1;
2421 
2422 	tmp = btrfs_super_total_bytes(fs_info->super_copy);
2423 	btrfs_set_super_total_bytes(fs_info->super_copy,
2424 				    tmp + device->total_bytes);
2425 
2426 	tmp = btrfs_super_num_devices(fs_info->super_copy);
2427 	btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2428 
2429 	/* add sysfs device entry */
2430 	btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2431 
2432 	/*
2433 	 * we've got more storage, clear any full flags on the space
2434 	 * infos
2435 	 */
2436 	btrfs_clear_space_info_full(fs_info);
2437 
2438 	mutex_unlock(&fs_info->chunk_mutex);
2439 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2440 
2441 	if (seeding_dev) {
2442 		mutex_lock(&fs_info->chunk_mutex);
2443 		ret = init_first_rw_device(trans, fs_info, device);
2444 		mutex_unlock(&fs_info->chunk_mutex);
2445 		if (ret) {
2446 			btrfs_abort_transaction(trans, ret);
2447 			goto error_trans;
2448 		}
2449 	}
2450 
2451 	ret = btrfs_add_device(trans, fs_info, device);
2452 	if (ret) {
2453 		btrfs_abort_transaction(trans, ret);
2454 		goto error_trans;
2455 	}
2456 
2457 	if (seeding_dev) {
2458 		char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2459 
2460 		ret = btrfs_finish_sprout(trans, fs_info);
2461 		if (ret) {
2462 			btrfs_abort_transaction(trans, ret);
2463 			goto error_trans;
2464 		}
2465 
2466 		/* Sprouting would change fsid of the mounted root,
2467 		 * so rename the fsid on the sysfs
2468 		 */
2469 		snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2470 						fs_info->fsid);
2471 		if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2472 			btrfs_warn(fs_info,
2473 				   "sysfs: failed to create fsid for sprout");
2474 	}
2475 
2476 	fs_info->num_tolerated_disk_barrier_failures =
2477 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2478 	ret = btrfs_commit_transaction(trans);
2479 
2480 	if (seeding_dev) {
2481 		mutex_unlock(&uuid_mutex);
2482 		up_write(&sb->s_umount);
2483 
2484 		if (ret) /* transaction commit */
2485 			return ret;
2486 
2487 		ret = btrfs_relocate_sys_chunks(fs_info);
2488 		if (ret < 0)
2489 			btrfs_handle_fs_error(fs_info, ret,
2490 				    "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2491 		trans = btrfs_attach_transaction(root);
2492 		if (IS_ERR(trans)) {
2493 			if (PTR_ERR(trans) == -ENOENT)
2494 				return 0;
2495 			return PTR_ERR(trans);
2496 		}
2497 		ret = btrfs_commit_transaction(trans);
2498 	}
2499 
2500 	/* Update ctime/mtime for libblkid */
2501 	update_dev_time(device_path);
2502 	return ret;
2503 
2504 error_trans:
2505 	btrfs_end_transaction(trans);
2506 	rcu_string_free(device->name);
2507 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2508 	kfree(device);
2509 error:
2510 	blkdev_put(bdev, FMODE_EXCL);
2511 	if (seeding_dev) {
2512 		mutex_unlock(&uuid_mutex);
2513 		up_write(&sb->s_umount);
2514 	}
2515 	return ret;
2516 }
2517 
2518 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2519 				  char *device_path,
2520 				  struct btrfs_device *srcdev,
2521 				  struct btrfs_device **device_out)
2522 {
2523 	struct request_queue *q;
2524 	struct btrfs_device *device;
2525 	struct block_device *bdev;
2526 	struct list_head *devices;
2527 	struct rcu_string *name;
2528 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
2529 	int ret = 0;
2530 
2531 	*device_out = NULL;
2532 	if (fs_info->fs_devices->seeding) {
2533 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2534 		return -EINVAL;
2535 	}
2536 
2537 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2538 				  fs_info->bdev_holder);
2539 	if (IS_ERR(bdev)) {
2540 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
2541 		return PTR_ERR(bdev);
2542 	}
2543 
2544 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2545 
2546 	devices = &fs_info->fs_devices->devices;
2547 	list_for_each_entry(device, devices, dev_list) {
2548 		if (device->bdev == bdev) {
2549 			btrfs_err(fs_info,
2550 				  "target device is in the filesystem!");
2551 			ret = -EEXIST;
2552 			goto error;
2553 		}
2554 	}
2555 
2556 
2557 	if (i_size_read(bdev->bd_inode) <
2558 	    btrfs_device_get_total_bytes(srcdev)) {
2559 		btrfs_err(fs_info,
2560 			  "target device is smaller than source device!");
2561 		ret = -EINVAL;
2562 		goto error;
2563 	}
2564 
2565 
2566 	device = btrfs_alloc_device(NULL, &devid, NULL);
2567 	if (IS_ERR(device)) {
2568 		ret = PTR_ERR(device);
2569 		goto error;
2570 	}
2571 
2572 	name = rcu_string_strdup(device_path, GFP_NOFS);
2573 	if (!name) {
2574 		kfree(device);
2575 		ret = -ENOMEM;
2576 		goto error;
2577 	}
2578 	rcu_assign_pointer(device->name, name);
2579 
2580 	q = bdev_get_queue(bdev);
2581 	if (blk_queue_discard(q))
2582 		device->can_discard = 1;
2583 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
2584 	device->writeable = 1;
2585 	device->generation = 0;
2586 	device->io_width = fs_info->sectorsize;
2587 	device->io_align = fs_info->sectorsize;
2588 	device->sector_size = fs_info->sectorsize;
2589 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2590 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2591 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2592 	ASSERT(list_empty(&srcdev->resized_list));
2593 	device->commit_total_bytes = srcdev->commit_total_bytes;
2594 	device->commit_bytes_used = device->bytes_used;
2595 	device->fs_info = fs_info;
2596 	device->bdev = bdev;
2597 	device->in_fs_metadata = 1;
2598 	device->is_tgtdev_for_dev_replace = 1;
2599 	device->mode = FMODE_EXCL;
2600 	device->dev_stats_valid = 1;
2601 	set_blocksize(device->bdev, 4096);
2602 	device->fs_devices = fs_info->fs_devices;
2603 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2604 	fs_info->fs_devices->num_devices++;
2605 	fs_info->fs_devices->open_devices++;
2606 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2607 
2608 	*device_out = device;
2609 	return ret;
2610 
2611 error:
2612 	blkdev_put(bdev, FMODE_EXCL);
2613 	return ret;
2614 }
2615 
2616 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2617 					      struct btrfs_device *tgtdev)
2618 {
2619 	u32 sectorsize = fs_info->sectorsize;
2620 
2621 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2622 	tgtdev->io_width = sectorsize;
2623 	tgtdev->io_align = sectorsize;
2624 	tgtdev->sector_size = sectorsize;
2625 	tgtdev->fs_info = fs_info;
2626 	tgtdev->in_fs_metadata = 1;
2627 }
2628 
2629 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2630 					struct btrfs_device *device)
2631 {
2632 	int ret;
2633 	struct btrfs_path *path;
2634 	struct btrfs_root *root = device->fs_info->chunk_root;
2635 	struct btrfs_dev_item *dev_item;
2636 	struct extent_buffer *leaf;
2637 	struct btrfs_key key;
2638 
2639 	path = btrfs_alloc_path();
2640 	if (!path)
2641 		return -ENOMEM;
2642 
2643 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2644 	key.type = BTRFS_DEV_ITEM_KEY;
2645 	key.offset = device->devid;
2646 
2647 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2648 	if (ret < 0)
2649 		goto out;
2650 
2651 	if (ret > 0) {
2652 		ret = -ENOENT;
2653 		goto out;
2654 	}
2655 
2656 	leaf = path->nodes[0];
2657 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2658 
2659 	btrfs_set_device_id(leaf, dev_item, device->devid);
2660 	btrfs_set_device_type(leaf, dev_item, device->type);
2661 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2662 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2663 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2664 	btrfs_set_device_total_bytes(leaf, dev_item,
2665 				     btrfs_device_get_disk_total_bytes(device));
2666 	btrfs_set_device_bytes_used(leaf, dev_item,
2667 				    btrfs_device_get_bytes_used(device));
2668 	btrfs_mark_buffer_dirty(leaf);
2669 
2670 out:
2671 	btrfs_free_path(path);
2672 	return ret;
2673 }
2674 
2675 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2676 		      struct btrfs_device *device, u64 new_size)
2677 {
2678 	struct btrfs_fs_info *fs_info = device->fs_info;
2679 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2680 	struct btrfs_fs_devices *fs_devices;
2681 	u64 old_total;
2682 	u64 diff;
2683 
2684 	if (!device->writeable)
2685 		return -EACCES;
2686 
2687 	mutex_lock(&fs_info->chunk_mutex);
2688 	old_total = btrfs_super_total_bytes(super_copy);
2689 	diff = new_size - device->total_bytes;
2690 
2691 	if (new_size <= device->total_bytes ||
2692 	    device->is_tgtdev_for_dev_replace) {
2693 		mutex_unlock(&fs_info->chunk_mutex);
2694 		return -EINVAL;
2695 	}
2696 
2697 	fs_devices = fs_info->fs_devices;
2698 
2699 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2700 	device->fs_devices->total_rw_bytes += diff;
2701 
2702 	btrfs_device_set_total_bytes(device, new_size);
2703 	btrfs_device_set_disk_total_bytes(device, new_size);
2704 	btrfs_clear_space_info_full(device->fs_info);
2705 	if (list_empty(&device->resized_list))
2706 		list_add_tail(&device->resized_list,
2707 			      &fs_devices->resized_devices);
2708 	mutex_unlock(&fs_info->chunk_mutex);
2709 
2710 	return btrfs_update_device(trans, device);
2711 }
2712 
2713 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2714 			    struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2715 			    u64 chunk_offset)
2716 {
2717 	struct btrfs_root *root = fs_info->chunk_root;
2718 	int ret;
2719 	struct btrfs_path *path;
2720 	struct btrfs_key key;
2721 
2722 	path = btrfs_alloc_path();
2723 	if (!path)
2724 		return -ENOMEM;
2725 
2726 	key.objectid = chunk_objectid;
2727 	key.offset = chunk_offset;
2728 	key.type = BTRFS_CHUNK_ITEM_KEY;
2729 
2730 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2731 	if (ret < 0)
2732 		goto out;
2733 	else if (ret > 0) { /* Logic error or corruption */
2734 		btrfs_handle_fs_error(fs_info, -ENOENT,
2735 				      "Failed lookup while freeing chunk.");
2736 		ret = -ENOENT;
2737 		goto out;
2738 	}
2739 
2740 	ret = btrfs_del_item(trans, root, path);
2741 	if (ret < 0)
2742 		btrfs_handle_fs_error(fs_info, ret,
2743 				      "Failed to delete chunk item.");
2744 out:
2745 	btrfs_free_path(path);
2746 	return ret;
2747 }
2748 
2749 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2750 			       u64 chunk_objectid, u64 chunk_offset)
2751 {
2752 	struct btrfs_super_block *super_copy = fs_info->super_copy;
2753 	struct btrfs_disk_key *disk_key;
2754 	struct btrfs_chunk *chunk;
2755 	u8 *ptr;
2756 	int ret = 0;
2757 	u32 num_stripes;
2758 	u32 array_size;
2759 	u32 len = 0;
2760 	u32 cur;
2761 	struct btrfs_key key;
2762 
2763 	mutex_lock(&fs_info->chunk_mutex);
2764 	array_size = btrfs_super_sys_array_size(super_copy);
2765 
2766 	ptr = super_copy->sys_chunk_array;
2767 	cur = 0;
2768 
2769 	while (cur < array_size) {
2770 		disk_key = (struct btrfs_disk_key *)ptr;
2771 		btrfs_disk_key_to_cpu(&key, disk_key);
2772 
2773 		len = sizeof(*disk_key);
2774 
2775 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2776 			chunk = (struct btrfs_chunk *)(ptr + len);
2777 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2778 			len += btrfs_chunk_item_size(num_stripes);
2779 		} else {
2780 			ret = -EIO;
2781 			break;
2782 		}
2783 		if (key.objectid == chunk_objectid &&
2784 		    key.offset == chunk_offset) {
2785 			memmove(ptr, ptr + len, array_size - (cur + len));
2786 			array_size -= len;
2787 			btrfs_set_super_sys_array_size(super_copy, array_size);
2788 		} else {
2789 			ptr += len;
2790 			cur += len;
2791 		}
2792 	}
2793 	mutex_unlock(&fs_info->chunk_mutex);
2794 	return ret;
2795 }
2796 
2797 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2798 		       struct btrfs_fs_info *fs_info, u64 chunk_offset)
2799 {
2800 	struct extent_map_tree *em_tree;
2801 	struct extent_map *em;
2802 	struct map_lookup *map;
2803 	u64 dev_extent_len = 0;
2804 	u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2805 	int i, ret = 0;
2806 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2807 
2808 	em_tree = &fs_info->mapping_tree.map_tree;
2809 
2810 	read_lock(&em_tree->lock);
2811 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2812 	read_unlock(&em_tree->lock);
2813 
2814 	if (!em || em->start > chunk_offset ||
2815 	    em->start + em->len < chunk_offset) {
2816 		/*
2817 		 * This is a logic error, but we don't want to just rely on the
2818 		 * user having built with ASSERT enabled, so if ASSERT doesn't
2819 		 * do anything we still error out.
2820 		 */
2821 		ASSERT(0);
2822 		if (em)
2823 			free_extent_map(em);
2824 		return -EINVAL;
2825 	}
2826 	map = em->map_lookup;
2827 	mutex_lock(&fs_info->chunk_mutex);
2828 	check_system_chunk(trans, fs_info, map->type);
2829 	mutex_unlock(&fs_info->chunk_mutex);
2830 
2831 	/*
2832 	 * Take the device list mutex to prevent races with the final phase of
2833 	 * a device replace operation that replaces the device object associated
2834 	 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2835 	 */
2836 	mutex_lock(&fs_devices->device_list_mutex);
2837 	for (i = 0; i < map->num_stripes; i++) {
2838 		struct btrfs_device *device = map->stripes[i].dev;
2839 		ret = btrfs_free_dev_extent(trans, device,
2840 					    map->stripes[i].physical,
2841 					    &dev_extent_len);
2842 		if (ret) {
2843 			mutex_unlock(&fs_devices->device_list_mutex);
2844 			btrfs_abort_transaction(trans, ret);
2845 			goto out;
2846 		}
2847 
2848 		if (device->bytes_used > 0) {
2849 			mutex_lock(&fs_info->chunk_mutex);
2850 			btrfs_device_set_bytes_used(device,
2851 					device->bytes_used - dev_extent_len);
2852 			spin_lock(&fs_info->free_chunk_lock);
2853 			fs_info->free_chunk_space += dev_extent_len;
2854 			spin_unlock(&fs_info->free_chunk_lock);
2855 			btrfs_clear_space_info_full(fs_info);
2856 			mutex_unlock(&fs_info->chunk_mutex);
2857 		}
2858 
2859 		if (map->stripes[i].dev) {
2860 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2861 			if (ret) {
2862 				mutex_unlock(&fs_devices->device_list_mutex);
2863 				btrfs_abort_transaction(trans, ret);
2864 				goto out;
2865 			}
2866 		}
2867 	}
2868 	mutex_unlock(&fs_devices->device_list_mutex);
2869 
2870 	ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2871 	if (ret) {
2872 		btrfs_abort_transaction(trans, ret);
2873 		goto out;
2874 	}
2875 
2876 	trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2877 
2878 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2879 		ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2880 					  chunk_offset);
2881 		if (ret) {
2882 			btrfs_abort_transaction(trans, ret);
2883 			goto out;
2884 		}
2885 	}
2886 
2887 	ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2888 	if (ret) {
2889 		btrfs_abort_transaction(trans, ret);
2890 		goto out;
2891 	}
2892 
2893 out:
2894 	/* once for us */
2895 	free_extent_map(em);
2896 	return ret;
2897 }
2898 
2899 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2900 {
2901 	struct btrfs_root *root = fs_info->chunk_root;
2902 	struct btrfs_trans_handle *trans;
2903 	int ret;
2904 
2905 	/*
2906 	 * Prevent races with automatic removal of unused block groups.
2907 	 * After we relocate and before we remove the chunk with offset
2908 	 * chunk_offset, automatic removal of the block group can kick in,
2909 	 * resulting in a failure when calling btrfs_remove_chunk() below.
2910 	 *
2911 	 * Make sure to acquire this mutex before doing a tree search (dev
2912 	 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2913 	 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2914 	 * we release the path used to search the chunk/dev tree and before
2915 	 * the current task acquires this mutex and calls us.
2916 	 */
2917 	ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2918 
2919 	ret = btrfs_can_relocate(fs_info, chunk_offset);
2920 	if (ret)
2921 		return -ENOSPC;
2922 
2923 	/* step one, relocate all the extents inside this chunk */
2924 	btrfs_scrub_pause(fs_info);
2925 	ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2926 	btrfs_scrub_continue(fs_info);
2927 	if (ret)
2928 		return ret;
2929 
2930 	trans = btrfs_start_trans_remove_block_group(root->fs_info,
2931 						     chunk_offset);
2932 	if (IS_ERR(trans)) {
2933 		ret = PTR_ERR(trans);
2934 		btrfs_handle_fs_error(root->fs_info, ret, NULL);
2935 		return ret;
2936 	}
2937 
2938 	/*
2939 	 * step two, delete the device extents and the
2940 	 * chunk tree entries
2941 	 */
2942 	ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2943 	btrfs_end_transaction(trans);
2944 	return ret;
2945 }
2946 
2947 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2948 {
2949 	struct btrfs_root *chunk_root = fs_info->chunk_root;
2950 	struct btrfs_path *path;
2951 	struct extent_buffer *leaf;
2952 	struct btrfs_chunk *chunk;
2953 	struct btrfs_key key;
2954 	struct btrfs_key found_key;
2955 	u64 chunk_type;
2956 	bool retried = false;
2957 	int failed = 0;
2958 	int ret;
2959 
2960 	path = btrfs_alloc_path();
2961 	if (!path)
2962 		return -ENOMEM;
2963 
2964 again:
2965 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2966 	key.offset = (u64)-1;
2967 	key.type = BTRFS_CHUNK_ITEM_KEY;
2968 
2969 	while (1) {
2970 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
2971 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2972 		if (ret < 0) {
2973 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2974 			goto error;
2975 		}
2976 		BUG_ON(ret == 0); /* Corruption */
2977 
2978 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2979 					  key.type);
2980 		if (ret)
2981 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2982 		if (ret < 0)
2983 			goto error;
2984 		if (ret > 0)
2985 			break;
2986 
2987 		leaf = path->nodes[0];
2988 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2989 
2990 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2991 				       struct btrfs_chunk);
2992 		chunk_type = btrfs_chunk_type(leaf, chunk);
2993 		btrfs_release_path(path);
2994 
2995 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2996 			ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2997 			if (ret == -ENOSPC)
2998 				failed++;
2999 			else
3000 				BUG_ON(ret);
3001 		}
3002 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3003 
3004 		if (found_key.offset == 0)
3005 			break;
3006 		key.offset = found_key.offset - 1;
3007 	}
3008 	ret = 0;
3009 	if (failed && !retried) {
3010 		failed = 0;
3011 		retried = true;
3012 		goto again;
3013 	} else if (WARN_ON(failed && retried)) {
3014 		ret = -ENOSPC;
3015 	}
3016 error:
3017 	btrfs_free_path(path);
3018 	return ret;
3019 }
3020 
3021 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3022 			       struct btrfs_balance_control *bctl)
3023 {
3024 	struct btrfs_root *root = fs_info->tree_root;
3025 	struct btrfs_trans_handle *trans;
3026 	struct btrfs_balance_item *item;
3027 	struct btrfs_disk_balance_args disk_bargs;
3028 	struct btrfs_path *path;
3029 	struct extent_buffer *leaf;
3030 	struct btrfs_key key;
3031 	int ret, err;
3032 
3033 	path = btrfs_alloc_path();
3034 	if (!path)
3035 		return -ENOMEM;
3036 
3037 	trans = btrfs_start_transaction(root, 0);
3038 	if (IS_ERR(trans)) {
3039 		btrfs_free_path(path);
3040 		return PTR_ERR(trans);
3041 	}
3042 
3043 	key.objectid = BTRFS_BALANCE_OBJECTID;
3044 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3045 	key.offset = 0;
3046 
3047 	ret = btrfs_insert_empty_item(trans, root, path, &key,
3048 				      sizeof(*item));
3049 	if (ret)
3050 		goto out;
3051 
3052 	leaf = path->nodes[0];
3053 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3054 
3055 	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3056 
3057 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3058 	btrfs_set_balance_data(leaf, item, &disk_bargs);
3059 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3060 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
3061 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3062 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
3063 
3064 	btrfs_set_balance_flags(leaf, item, bctl->flags);
3065 
3066 	btrfs_mark_buffer_dirty(leaf);
3067 out:
3068 	btrfs_free_path(path);
3069 	err = btrfs_commit_transaction(trans);
3070 	if (err && !ret)
3071 		ret = err;
3072 	return ret;
3073 }
3074 
3075 static int del_balance_item(struct btrfs_fs_info *fs_info)
3076 {
3077 	struct btrfs_root *root = fs_info->tree_root;
3078 	struct btrfs_trans_handle *trans;
3079 	struct btrfs_path *path;
3080 	struct btrfs_key key;
3081 	int ret, err;
3082 
3083 	path = btrfs_alloc_path();
3084 	if (!path)
3085 		return -ENOMEM;
3086 
3087 	trans = btrfs_start_transaction(root, 0);
3088 	if (IS_ERR(trans)) {
3089 		btrfs_free_path(path);
3090 		return PTR_ERR(trans);
3091 	}
3092 
3093 	key.objectid = BTRFS_BALANCE_OBJECTID;
3094 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3095 	key.offset = 0;
3096 
3097 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3098 	if (ret < 0)
3099 		goto out;
3100 	if (ret > 0) {
3101 		ret = -ENOENT;
3102 		goto out;
3103 	}
3104 
3105 	ret = btrfs_del_item(trans, root, path);
3106 out:
3107 	btrfs_free_path(path);
3108 	err = btrfs_commit_transaction(trans);
3109 	if (err && !ret)
3110 		ret = err;
3111 	return ret;
3112 }
3113 
3114 /*
3115  * This is a heuristic used to reduce the number of chunks balanced on
3116  * resume after balance was interrupted.
3117  */
3118 static void update_balance_args(struct btrfs_balance_control *bctl)
3119 {
3120 	/*
3121 	 * Turn on soft mode for chunk types that were being converted.
3122 	 */
3123 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3124 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3125 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3126 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3127 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3128 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3129 
3130 	/*
3131 	 * Turn on usage filter if is not already used.  The idea is
3132 	 * that chunks that we have already balanced should be
3133 	 * reasonably full.  Don't do it for chunks that are being
3134 	 * converted - that will keep us from relocating unconverted
3135 	 * (albeit full) chunks.
3136 	 */
3137 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3138 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3139 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3140 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3141 		bctl->data.usage = 90;
3142 	}
3143 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3144 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3145 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3146 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3147 		bctl->sys.usage = 90;
3148 	}
3149 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3150 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3151 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3152 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3153 		bctl->meta.usage = 90;
3154 	}
3155 }
3156 
3157 /*
3158  * Should be called with both balance and volume mutexes held to
3159  * serialize other volume operations (add_dev/rm_dev/resize) with
3160  * restriper.  Same goes for unset_balance_control.
3161  */
3162 static void set_balance_control(struct btrfs_balance_control *bctl)
3163 {
3164 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3165 
3166 	BUG_ON(fs_info->balance_ctl);
3167 
3168 	spin_lock(&fs_info->balance_lock);
3169 	fs_info->balance_ctl = bctl;
3170 	spin_unlock(&fs_info->balance_lock);
3171 }
3172 
3173 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3174 {
3175 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3176 
3177 	BUG_ON(!fs_info->balance_ctl);
3178 
3179 	spin_lock(&fs_info->balance_lock);
3180 	fs_info->balance_ctl = NULL;
3181 	spin_unlock(&fs_info->balance_lock);
3182 
3183 	kfree(bctl);
3184 }
3185 
3186 /*
3187  * Balance filters.  Return 1 if chunk should be filtered out
3188  * (should not be balanced).
3189  */
3190 static int chunk_profiles_filter(u64 chunk_type,
3191 				 struct btrfs_balance_args *bargs)
3192 {
3193 	chunk_type = chunk_to_extended(chunk_type) &
3194 				BTRFS_EXTENDED_PROFILE_MASK;
3195 
3196 	if (bargs->profiles & chunk_type)
3197 		return 0;
3198 
3199 	return 1;
3200 }
3201 
3202 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3203 			      struct btrfs_balance_args *bargs)
3204 {
3205 	struct btrfs_block_group_cache *cache;
3206 	u64 chunk_used;
3207 	u64 user_thresh_min;
3208 	u64 user_thresh_max;
3209 	int ret = 1;
3210 
3211 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3212 	chunk_used = btrfs_block_group_used(&cache->item);
3213 
3214 	if (bargs->usage_min == 0)
3215 		user_thresh_min = 0;
3216 	else
3217 		user_thresh_min = div_factor_fine(cache->key.offset,
3218 					bargs->usage_min);
3219 
3220 	if (bargs->usage_max == 0)
3221 		user_thresh_max = 1;
3222 	else if (bargs->usage_max > 100)
3223 		user_thresh_max = cache->key.offset;
3224 	else
3225 		user_thresh_max = div_factor_fine(cache->key.offset,
3226 					bargs->usage_max);
3227 
3228 	if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3229 		ret = 0;
3230 
3231 	btrfs_put_block_group(cache);
3232 	return ret;
3233 }
3234 
3235 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3236 		u64 chunk_offset, struct btrfs_balance_args *bargs)
3237 {
3238 	struct btrfs_block_group_cache *cache;
3239 	u64 chunk_used, user_thresh;
3240 	int ret = 1;
3241 
3242 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3243 	chunk_used = btrfs_block_group_used(&cache->item);
3244 
3245 	if (bargs->usage_min == 0)
3246 		user_thresh = 1;
3247 	else if (bargs->usage > 100)
3248 		user_thresh = cache->key.offset;
3249 	else
3250 		user_thresh = div_factor_fine(cache->key.offset,
3251 					      bargs->usage);
3252 
3253 	if (chunk_used < user_thresh)
3254 		ret = 0;
3255 
3256 	btrfs_put_block_group(cache);
3257 	return ret;
3258 }
3259 
3260 static int chunk_devid_filter(struct extent_buffer *leaf,
3261 			      struct btrfs_chunk *chunk,
3262 			      struct btrfs_balance_args *bargs)
3263 {
3264 	struct btrfs_stripe *stripe;
3265 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3266 	int i;
3267 
3268 	for (i = 0; i < num_stripes; i++) {
3269 		stripe = btrfs_stripe_nr(chunk, i);
3270 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3271 			return 0;
3272 	}
3273 
3274 	return 1;
3275 }
3276 
3277 /* [pstart, pend) */
3278 static int chunk_drange_filter(struct extent_buffer *leaf,
3279 			       struct btrfs_chunk *chunk,
3280 			       u64 chunk_offset,
3281 			       struct btrfs_balance_args *bargs)
3282 {
3283 	struct btrfs_stripe *stripe;
3284 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3285 	u64 stripe_offset;
3286 	u64 stripe_length;
3287 	int factor;
3288 	int i;
3289 
3290 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3291 		return 0;
3292 
3293 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3294 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3295 		factor = num_stripes / 2;
3296 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3297 		factor = num_stripes - 1;
3298 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3299 		factor = num_stripes - 2;
3300 	} else {
3301 		factor = num_stripes;
3302 	}
3303 
3304 	for (i = 0; i < num_stripes; i++) {
3305 		stripe = btrfs_stripe_nr(chunk, i);
3306 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3307 			continue;
3308 
3309 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3310 		stripe_length = btrfs_chunk_length(leaf, chunk);
3311 		stripe_length = div_u64(stripe_length, factor);
3312 
3313 		if (stripe_offset < bargs->pend &&
3314 		    stripe_offset + stripe_length > bargs->pstart)
3315 			return 0;
3316 	}
3317 
3318 	return 1;
3319 }
3320 
3321 /* [vstart, vend) */
3322 static int chunk_vrange_filter(struct extent_buffer *leaf,
3323 			       struct btrfs_chunk *chunk,
3324 			       u64 chunk_offset,
3325 			       struct btrfs_balance_args *bargs)
3326 {
3327 	if (chunk_offset < bargs->vend &&
3328 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3329 		/* at least part of the chunk is inside this vrange */
3330 		return 0;
3331 
3332 	return 1;
3333 }
3334 
3335 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3336 			       struct btrfs_chunk *chunk,
3337 			       struct btrfs_balance_args *bargs)
3338 {
3339 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3340 
3341 	if (bargs->stripes_min <= num_stripes
3342 			&& num_stripes <= bargs->stripes_max)
3343 		return 0;
3344 
3345 	return 1;
3346 }
3347 
3348 static int chunk_soft_convert_filter(u64 chunk_type,
3349 				     struct btrfs_balance_args *bargs)
3350 {
3351 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3352 		return 0;
3353 
3354 	chunk_type = chunk_to_extended(chunk_type) &
3355 				BTRFS_EXTENDED_PROFILE_MASK;
3356 
3357 	if (bargs->target == chunk_type)
3358 		return 1;
3359 
3360 	return 0;
3361 }
3362 
3363 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3364 				struct extent_buffer *leaf,
3365 				struct btrfs_chunk *chunk, u64 chunk_offset)
3366 {
3367 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3368 	struct btrfs_balance_args *bargs = NULL;
3369 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3370 
3371 	/* type filter */
3372 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3373 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3374 		return 0;
3375 	}
3376 
3377 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3378 		bargs = &bctl->data;
3379 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3380 		bargs = &bctl->sys;
3381 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3382 		bargs = &bctl->meta;
3383 
3384 	/* profiles filter */
3385 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3386 	    chunk_profiles_filter(chunk_type, bargs)) {
3387 		return 0;
3388 	}
3389 
3390 	/* usage filter */
3391 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3392 	    chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3393 		return 0;
3394 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3395 	    chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3396 		return 0;
3397 	}
3398 
3399 	/* devid filter */
3400 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3401 	    chunk_devid_filter(leaf, chunk, bargs)) {
3402 		return 0;
3403 	}
3404 
3405 	/* drange filter, makes sense only with devid filter */
3406 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3407 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3408 		return 0;
3409 	}
3410 
3411 	/* vrange filter */
3412 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3413 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3414 		return 0;
3415 	}
3416 
3417 	/* stripes filter */
3418 	if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3419 	    chunk_stripes_range_filter(leaf, chunk, bargs)) {
3420 		return 0;
3421 	}
3422 
3423 	/* soft profile changing mode */
3424 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3425 	    chunk_soft_convert_filter(chunk_type, bargs)) {
3426 		return 0;
3427 	}
3428 
3429 	/*
3430 	 * limited by count, must be the last filter
3431 	 */
3432 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3433 		if (bargs->limit == 0)
3434 			return 0;
3435 		else
3436 			bargs->limit--;
3437 	} else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3438 		/*
3439 		 * Same logic as the 'limit' filter; the minimum cannot be
3440 		 * determined here because we do not have the global information
3441 		 * about the count of all chunks that satisfy the filters.
3442 		 */
3443 		if (bargs->limit_max == 0)
3444 			return 0;
3445 		else
3446 			bargs->limit_max--;
3447 	}
3448 
3449 	return 1;
3450 }
3451 
3452 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3453 {
3454 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3455 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3456 	struct btrfs_root *dev_root = fs_info->dev_root;
3457 	struct list_head *devices;
3458 	struct btrfs_device *device;
3459 	u64 old_size;
3460 	u64 size_to_free;
3461 	u64 chunk_type;
3462 	struct btrfs_chunk *chunk;
3463 	struct btrfs_path *path = NULL;
3464 	struct btrfs_key key;
3465 	struct btrfs_key found_key;
3466 	struct btrfs_trans_handle *trans;
3467 	struct extent_buffer *leaf;
3468 	int slot;
3469 	int ret;
3470 	int enospc_errors = 0;
3471 	bool counting = true;
3472 	/* The single value limit and min/max limits use the same bytes in the */
3473 	u64 limit_data = bctl->data.limit;
3474 	u64 limit_meta = bctl->meta.limit;
3475 	u64 limit_sys = bctl->sys.limit;
3476 	u32 count_data = 0;
3477 	u32 count_meta = 0;
3478 	u32 count_sys = 0;
3479 	int chunk_reserved = 0;
3480 	u64 bytes_used = 0;
3481 
3482 	/* step one make some room on all the devices */
3483 	devices = &fs_info->fs_devices->devices;
3484 	list_for_each_entry(device, devices, dev_list) {
3485 		old_size = btrfs_device_get_total_bytes(device);
3486 		size_to_free = div_factor(old_size, 1);
3487 		size_to_free = min_t(u64, size_to_free, SZ_1M);
3488 		if (!device->writeable ||
3489 		    btrfs_device_get_total_bytes(device) -
3490 		    btrfs_device_get_bytes_used(device) > size_to_free ||
3491 		    device->is_tgtdev_for_dev_replace)
3492 			continue;
3493 
3494 		ret = btrfs_shrink_device(device, old_size - size_to_free);
3495 		if (ret == -ENOSPC)
3496 			break;
3497 		if (ret) {
3498 			/* btrfs_shrink_device never returns ret > 0 */
3499 			WARN_ON(ret > 0);
3500 			goto error;
3501 		}
3502 
3503 		trans = btrfs_start_transaction(dev_root, 0);
3504 		if (IS_ERR(trans)) {
3505 			ret = PTR_ERR(trans);
3506 			btrfs_info_in_rcu(fs_info,
3507 		 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3508 					  rcu_str_deref(device->name), ret,
3509 					  old_size, old_size - size_to_free);
3510 			goto error;
3511 		}
3512 
3513 		ret = btrfs_grow_device(trans, device, old_size);
3514 		if (ret) {
3515 			btrfs_end_transaction(trans);
3516 			/* btrfs_grow_device never returns ret > 0 */
3517 			WARN_ON(ret > 0);
3518 			btrfs_info_in_rcu(fs_info,
3519 		 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3520 					  rcu_str_deref(device->name), ret,
3521 					  old_size, old_size - size_to_free);
3522 			goto error;
3523 		}
3524 
3525 		btrfs_end_transaction(trans);
3526 	}
3527 
3528 	/* step two, relocate all the chunks */
3529 	path = btrfs_alloc_path();
3530 	if (!path) {
3531 		ret = -ENOMEM;
3532 		goto error;
3533 	}
3534 
3535 	/* zero out stat counters */
3536 	spin_lock(&fs_info->balance_lock);
3537 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3538 	spin_unlock(&fs_info->balance_lock);
3539 again:
3540 	if (!counting) {
3541 		/*
3542 		 * The single value limit and min/max limits use the same bytes
3543 		 * in the
3544 		 */
3545 		bctl->data.limit = limit_data;
3546 		bctl->meta.limit = limit_meta;
3547 		bctl->sys.limit = limit_sys;
3548 	}
3549 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3550 	key.offset = (u64)-1;
3551 	key.type = BTRFS_CHUNK_ITEM_KEY;
3552 
3553 	while (1) {
3554 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3555 		    atomic_read(&fs_info->balance_cancel_req)) {
3556 			ret = -ECANCELED;
3557 			goto error;
3558 		}
3559 
3560 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
3561 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3562 		if (ret < 0) {
3563 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3564 			goto error;
3565 		}
3566 
3567 		/*
3568 		 * this shouldn't happen, it means the last relocate
3569 		 * failed
3570 		 */
3571 		if (ret == 0)
3572 			BUG(); /* FIXME break ? */
3573 
3574 		ret = btrfs_previous_item(chunk_root, path, 0,
3575 					  BTRFS_CHUNK_ITEM_KEY);
3576 		if (ret) {
3577 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3578 			ret = 0;
3579 			break;
3580 		}
3581 
3582 		leaf = path->nodes[0];
3583 		slot = path->slots[0];
3584 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3585 
3586 		if (found_key.objectid != key.objectid) {
3587 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3588 			break;
3589 		}
3590 
3591 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3592 		chunk_type = btrfs_chunk_type(leaf, chunk);
3593 
3594 		if (!counting) {
3595 			spin_lock(&fs_info->balance_lock);
3596 			bctl->stat.considered++;
3597 			spin_unlock(&fs_info->balance_lock);
3598 		}
3599 
3600 		ret = should_balance_chunk(fs_info, leaf, chunk,
3601 					   found_key.offset);
3602 
3603 		btrfs_release_path(path);
3604 		if (!ret) {
3605 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3606 			goto loop;
3607 		}
3608 
3609 		if (counting) {
3610 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3611 			spin_lock(&fs_info->balance_lock);
3612 			bctl->stat.expected++;
3613 			spin_unlock(&fs_info->balance_lock);
3614 
3615 			if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3616 				count_data++;
3617 			else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3618 				count_sys++;
3619 			else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3620 				count_meta++;
3621 
3622 			goto loop;
3623 		}
3624 
3625 		/*
3626 		 * Apply limit_min filter, no need to check if the LIMITS
3627 		 * filter is used, limit_min is 0 by default
3628 		 */
3629 		if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3630 					count_data < bctl->data.limit_min)
3631 				|| ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3632 					count_meta < bctl->meta.limit_min)
3633 				|| ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3634 					count_sys < bctl->sys.limit_min)) {
3635 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3636 			goto loop;
3637 		}
3638 
3639 		ASSERT(fs_info->data_sinfo);
3640 		spin_lock(&fs_info->data_sinfo->lock);
3641 		bytes_used = fs_info->data_sinfo->bytes_used;
3642 		spin_unlock(&fs_info->data_sinfo->lock);
3643 
3644 		if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3645 		    !chunk_reserved && !bytes_used) {
3646 			trans = btrfs_start_transaction(chunk_root, 0);
3647 			if (IS_ERR(trans)) {
3648 				mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3649 				ret = PTR_ERR(trans);
3650 				goto error;
3651 			}
3652 
3653 			ret = btrfs_force_chunk_alloc(trans, fs_info,
3654 						      BTRFS_BLOCK_GROUP_DATA);
3655 			btrfs_end_transaction(trans);
3656 			if (ret < 0) {
3657 				mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3658 				goto error;
3659 			}
3660 			chunk_reserved = 1;
3661 		}
3662 
3663 		ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3664 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3665 		if (ret && ret != -ENOSPC)
3666 			goto error;
3667 		if (ret == -ENOSPC) {
3668 			enospc_errors++;
3669 		} else {
3670 			spin_lock(&fs_info->balance_lock);
3671 			bctl->stat.completed++;
3672 			spin_unlock(&fs_info->balance_lock);
3673 		}
3674 loop:
3675 		if (found_key.offset == 0)
3676 			break;
3677 		key.offset = found_key.offset - 1;
3678 	}
3679 
3680 	if (counting) {
3681 		btrfs_release_path(path);
3682 		counting = false;
3683 		goto again;
3684 	}
3685 error:
3686 	btrfs_free_path(path);
3687 	if (enospc_errors) {
3688 		btrfs_info(fs_info, "%d enospc errors during balance",
3689 			   enospc_errors);
3690 		if (!ret)
3691 			ret = -ENOSPC;
3692 	}
3693 
3694 	return ret;
3695 }
3696 
3697 /**
3698  * alloc_profile_is_valid - see if a given profile is valid and reduced
3699  * @flags: profile to validate
3700  * @extended: if true @flags is treated as an extended profile
3701  */
3702 static int alloc_profile_is_valid(u64 flags, int extended)
3703 {
3704 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3705 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3706 
3707 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3708 
3709 	/* 1) check that all other bits are zeroed */
3710 	if (flags & ~mask)
3711 		return 0;
3712 
3713 	/* 2) see if profile is reduced */
3714 	if (flags == 0)
3715 		return !extended; /* "0" is valid for usual profiles */
3716 
3717 	/* true if exactly one bit set */
3718 	return (flags & (flags - 1)) == 0;
3719 }
3720 
3721 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3722 {
3723 	/* cancel requested || normal exit path */
3724 	return atomic_read(&fs_info->balance_cancel_req) ||
3725 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3726 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3727 }
3728 
3729 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3730 {
3731 	int ret;
3732 
3733 	unset_balance_control(fs_info);
3734 	ret = del_balance_item(fs_info);
3735 	if (ret)
3736 		btrfs_handle_fs_error(fs_info, ret, NULL);
3737 
3738 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3739 }
3740 
3741 /* Non-zero return value signifies invalidity */
3742 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3743 		u64 allowed)
3744 {
3745 	return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3746 		(!alloc_profile_is_valid(bctl_arg->target, 1) ||
3747 		 (bctl_arg->target & ~allowed)));
3748 }
3749 
3750 /*
3751  * Should be called with both balance and volume mutexes held
3752  */
3753 int btrfs_balance(struct btrfs_balance_control *bctl,
3754 		  struct btrfs_ioctl_balance_args *bargs)
3755 {
3756 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3757 	u64 allowed;
3758 	int mixed = 0;
3759 	int ret;
3760 	u64 num_devices;
3761 	unsigned seq;
3762 
3763 	if (btrfs_fs_closing(fs_info) ||
3764 	    atomic_read(&fs_info->balance_pause_req) ||
3765 	    atomic_read(&fs_info->balance_cancel_req)) {
3766 		ret = -EINVAL;
3767 		goto out;
3768 	}
3769 
3770 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3771 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3772 		mixed = 1;
3773 
3774 	/*
3775 	 * In case of mixed groups both data and meta should be picked,
3776 	 * and identical options should be given for both of them.
3777 	 */
3778 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3779 	if (mixed && (bctl->flags & allowed)) {
3780 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3781 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3782 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3783 			btrfs_err(fs_info,
3784 				  "with mixed groups data and metadata balance options must be the same");
3785 			ret = -EINVAL;
3786 			goto out;
3787 		}
3788 	}
3789 
3790 	num_devices = fs_info->fs_devices->num_devices;
3791 	btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3792 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3793 		BUG_ON(num_devices < 1);
3794 		num_devices--;
3795 	}
3796 	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3797 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3798 	if (num_devices > 1)
3799 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3800 	if (num_devices > 2)
3801 		allowed |= BTRFS_BLOCK_GROUP_RAID5;
3802 	if (num_devices > 3)
3803 		allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3804 			    BTRFS_BLOCK_GROUP_RAID6);
3805 	if (validate_convert_profile(&bctl->data, allowed)) {
3806 		btrfs_err(fs_info,
3807 			  "unable to start balance with target data profile %llu",
3808 			  bctl->data.target);
3809 		ret = -EINVAL;
3810 		goto out;
3811 	}
3812 	if (validate_convert_profile(&bctl->meta, allowed)) {
3813 		btrfs_err(fs_info,
3814 			  "unable to start balance with target metadata profile %llu",
3815 			  bctl->meta.target);
3816 		ret = -EINVAL;
3817 		goto out;
3818 	}
3819 	if (validate_convert_profile(&bctl->sys, allowed)) {
3820 		btrfs_err(fs_info,
3821 			  "unable to start balance with target system profile %llu",
3822 			  bctl->sys.target);
3823 		ret = -EINVAL;
3824 		goto out;
3825 	}
3826 
3827 	/* allow to reduce meta or sys integrity only if force set */
3828 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3829 			BTRFS_BLOCK_GROUP_RAID10 |
3830 			BTRFS_BLOCK_GROUP_RAID5 |
3831 			BTRFS_BLOCK_GROUP_RAID6;
3832 	do {
3833 		seq = read_seqbegin(&fs_info->profiles_lock);
3834 
3835 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3836 		     (fs_info->avail_system_alloc_bits & allowed) &&
3837 		     !(bctl->sys.target & allowed)) ||
3838 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3839 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
3840 		     !(bctl->meta.target & allowed))) {
3841 			if (bctl->flags & BTRFS_BALANCE_FORCE) {
3842 				btrfs_info(fs_info,
3843 					   "force reducing metadata integrity");
3844 			} else {
3845 				btrfs_err(fs_info,
3846 					  "balance will reduce metadata integrity, use force if you want this");
3847 				ret = -EINVAL;
3848 				goto out;
3849 			}
3850 		}
3851 	} while (read_seqretry(&fs_info->profiles_lock, seq));
3852 
3853 	if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3854 		btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3855 		btrfs_warn(fs_info,
3856 			   "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3857 			   bctl->meta.target, bctl->data.target);
3858 	}
3859 
3860 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3861 		fs_info->num_tolerated_disk_barrier_failures = min(
3862 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3863 			btrfs_get_num_tolerated_disk_barrier_failures(
3864 				bctl->sys.target));
3865 	}
3866 
3867 	ret = insert_balance_item(fs_info, bctl);
3868 	if (ret && ret != -EEXIST)
3869 		goto out;
3870 
3871 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3872 		BUG_ON(ret == -EEXIST);
3873 		set_balance_control(bctl);
3874 	} else {
3875 		BUG_ON(ret != -EEXIST);
3876 		spin_lock(&fs_info->balance_lock);
3877 		update_balance_args(bctl);
3878 		spin_unlock(&fs_info->balance_lock);
3879 	}
3880 
3881 	atomic_inc(&fs_info->balance_running);
3882 	mutex_unlock(&fs_info->balance_mutex);
3883 
3884 	ret = __btrfs_balance(fs_info);
3885 
3886 	mutex_lock(&fs_info->balance_mutex);
3887 	atomic_dec(&fs_info->balance_running);
3888 
3889 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3890 		fs_info->num_tolerated_disk_barrier_failures =
3891 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3892 	}
3893 
3894 	if (bargs) {
3895 		memset(bargs, 0, sizeof(*bargs));
3896 		update_ioctl_balance_args(fs_info, 0, bargs);
3897 	}
3898 
3899 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3900 	    balance_need_close(fs_info)) {
3901 		__cancel_balance(fs_info);
3902 	}
3903 
3904 	wake_up(&fs_info->balance_wait_q);
3905 
3906 	return ret;
3907 out:
3908 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3909 		__cancel_balance(fs_info);
3910 	else {
3911 		kfree(bctl);
3912 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3913 	}
3914 	return ret;
3915 }
3916 
3917 static int balance_kthread(void *data)
3918 {
3919 	struct btrfs_fs_info *fs_info = data;
3920 	int ret = 0;
3921 
3922 	mutex_lock(&fs_info->volume_mutex);
3923 	mutex_lock(&fs_info->balance_mutex);
3924 
3925 	if (fs_info->balance_ctl) {
3926 		btrfs_info(fs_info, "continuing balance");
3927 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3928 	}
3929 
3930 	mutex_unlock(&fs_info->balance_mutex);
3931 	mutex_unlock(&fs_info->volume_mutex);
3932 
3933 	return ret;
3934 }
3935 
3936 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3937 {
3938 	struct task_struct *tsk;
3939 
3940 	spin_lock(&fs_info->balance_lock);
3941 	if (!fs_info->balance_ctl) {
3942 		spin_unlock(&fs_info->balance_lock);
3943 		return 0;
3944 	}
3945 	spin_unlock(&fs_info->balance_lock);
3946 
3947 	if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3948 		btrfs_info(fs_info, "force skipping balance");
3949 		return 0;
3950 	}
3951 
3952 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3953 	return PTR_ERR_OR_ZERO(tsk);
3954 }
3955 
3956 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3957 {
3958 	struct btrfs_balance_control *bctl;
3959 	struct btrfs_balance_item *item;
3960 	struct btrfs_disk_balance_args disk_bargs;
3961 	struct btrfs_path *path;
3962 	struct extent_buffer *leaf;
3963 	struct btrfs_key key;
3964 	int ret;
3965 
3966 	path = btrfs_alloc_path();
3967 	if (!path)
3968 		return -ENOMEM;
3969 
3970 	key.objectid = BTRFS_BALANCE_OBJECTID;
3971 	key.type = BTRFS_TEMPORARY_ITEM_KEY;
3972 	key.offset = 0;
3973 
3974 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3975 	if (ret < 0)
3976 		goto out;
3977 	if (ret > 0) { /* ret = -ENOENT; */
3978 		ret = 0;
3979 		goto out;
3980 	}
3981 
3982 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3983 	if (!bctl) {
3984 		ret = -ENOMEM;
3985 		goto out;
3986 	}
3987 
3988 	leaf = path->nodes[0];
3989 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3990 
3991 	bctl->fs_info = fs_info;
3992 	bctl->flags = btrfs_balance_flags(leaf, item);
3993 	bctl->flags |= BTRFS_BALANCE_RESUME;
3994 
3995 	btrfs_balance_data(leaf, item, &disk_bargs);
3996 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3997 	btrfs_balance_meta(leaf, item, &disk_bargs);
3998 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3999 	btrfs_balance_sys(leaf, item, &disk_bargs);
4000 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4001 
4002 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4003 
4004 	mutex_lock(&fs_info->volume_mutex);
4005 	mutex_lock(&fs_info->balance_mutex);
4006 
4007 	set_balance_control(bctl);
4008 
4009 	mutex_unlock(&fs_info->balance_mutex);
4010 	mutex_unlock(&fs_info->volume_mutex);
4011 out:
4012 	btrfs_free_path(path);
4013 	return ret;
4014 }
4015 
4016 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4017 {
4018 	int ret = 0;
4019 
4020 	mutex_lock(&fs_info->balance_mutex);
4021 	if (!fs_info->balance_ctl) {
4022 		mutex_unlock(&fs_info->balance_mutex);
4023 		return -ENOTCONN;
4024 	}
4025 
4026 	if (atomic_read(&fs_info->balance_running)) {
4027 		atomic_inc(&fs_info->balance_pause_req);
4028 		mutex_unlock(&fs_info->balance_mutex);
4029 
4030 		wait_event(fs_info->balance_wait_q,
4031 			   atomic_read(&fs_info->balance_running) == 0);
4032 
4033 		mutex_lock(&fs_info->balance_mutex);
4034 		/* we are good with balance_ctl ripped off from under us */
4035 		BUG_ON(atomic_read(&fs_info->balance_running));
4036 		atomic_dec(&fs_info->balance_pause_req);
4037 	} else {
4038 		ret = -ENOTCONN;
4039 	}
4040 
4041 	mutex_unlock(&fs_info->balance_mutex);
4042 	return ret;
4043 }
4044 
4045 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4046 {
4047 	if (fs_info->sb->s_flags & MS_RDONLY)
4048 		return -EROFS;
4049 
4050 	mutex_lock(&fs_info->balance_mutex);
4051 	if (!fs_info->balance_ctl) {
4052 		mutex_unlock(&fs_info->balance_mutex);
4053 		return -ENOTCONN;
4054 	}
4055 
4056 	atomic_inc(&fs_info->balance_cancel_req);
4057 	/*
4058 	 * if we are running just wait and return, balance item is
4059 	 * deleted in btrfs_balance in this case
4060 	 */
4061 	if (atomic_read(&fs_info->balance_running)) {
4062 		mutex_unlock(&fs_info->balance_mutex);
4063 		wait_event(fs_info->balance_wait_q,
4064 			   atomic_read(&fs_info->balance_running) == 0);
4065 		mutex_lock(&fs_info->balance_mutex);
4066 	} else {
4067 		/* __cancel_balance needs volume_mutex */
4068 		mutex_unlock(&fs_info->balance_mutex);
4069 		mutex_lock(&fs_info->volume_mutex);
4070 		mutex_lock(&fs_info->balance_mutex);
4071 
4072 		if (fs_info->balance_ctl)
4073 			__cancel_balance(fs_info);
4074 
4075 		mutex_unlock(&fs_info->volume_mutex);
4076 	}
4077 
4078 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4079 	atomic_dec(&fs_info->balance_cancel_req);
4080 	mutex_unlock(&fs_info->balance_mutex);
4081 	return 0;
4082 }
4083 
4084 static int btrfs_uuid_scan_kthread(void *data)
4085 {
4086 	struct btrfs_fs_info *fs_info = data;
4087 	struct btrfs_root *root = fs_info->tree_root;
4088 	struct btrfs_key key;
4089 	struct btrfs_key max_key;
4090 	struct btrfs_path *path = NULL;
4091 	int ret = 0;
4092 	struct extent_buffer *eb;
4093 	int slot;
4094 	struct btrfs_root_item root_item;
4095 	u32 item_size;
4096 	struct btrfs_trans_handle *trans = NULL;
4097 
4098 	path = btrfs_alloc_path();
4099 	if (!path) {
4100 		ret = -ENOMEM;
4101 		goto out;
4102 	}
4103 
4104 	key.objectid = 0;
4105 	key.type = BTRFS_ROOT_ITEM_KEY;
4106 	key.offset = 0;
4107 
4108 	max_key.objectid = (u64)-1;
4109 	max_key.type = BTRFS_ROOT_ITEM_KEY;
4110 	max_key.offset = (u64)-1;
4111 
4112 	while (1) {
4113 		ret = btrfs_search_forward(root, &key, path, 0);
4114 		if (ret) {
4115 			if (ret > 0)
4116 				ret = 0;
4117 			break;
4118 		}
4119 
4120 		if (key.type != BTRFS_ROOT_ITEM_KEY ||
4121 		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4122 		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4123 		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
4124 			goto skip;
4125 
4126 		eb = path->nodes[0];
4127 		slot = path->slots[0];
4128 		item_size = btrfs_item_size_nr(eb, slot);
4129 		if (item_size < sizeof(root_item))
4130 			goto skip;
4131 
4132 		read_extent_buffer(eb, &root_item,
4133 				   btrfs_item_ptr_offset(eb, slot),
4134 				   (int)sizeof(root_item));
4135 		if (btrfs_root_refs(&root_item) == 0)
4136 			goto skip;
4137 
4138 		if (!btrfs_is_empty_uuid(root_item.uuid) ||
4139 		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
4140 			if (trans)
4141 				goto update_tree;
4142 
4143 			btrfs_release_path(path);
4144 			/*
4145 			 * 1 - subvol uuid item
4146 			 * 1 - received_subvol uuid item
4147 			 */
4148 			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4149 			if (IS_ERR(trans)) {
4150 				ret = PTR_ERR(trans);
4151 				break;
4152 			}
4153 			continue;
4154 		} else {
4155 			goto skip;
4156 		}
4157 update_tree:
4158 		if (!btrfs_is_empty_uuid(root_item.uuid)) {
4159 			ret = btrfs_uuid_tree_add(trans, fs_info,
4160 						  root_item.uuid,
4161 						  BTRFS_UUID_KEY_SUBVOL,
4162 						  key.objectid);
4163 			if (ret < 0) {
4164 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4165 					ret);
4166 				break;
4167 			}
4168 		}
4169 
4170 		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4171 			ret = btrfs_uuid_tree_add(trans, fs_info,
4172 						  root_item.received_uuid,
4173 						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4174 						  key.objectid);
4175 			if (ret < 0) {
4176 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
4177 					ret);
4178 				break;
4179 			}
4180 		}
4181 
4182 skip:
4183 		if (trans) {
4184 			ret = btrfs_end_transaction(trans);
4185 			trans = NULL;
4186 			if (ret)
4187 				break;
4188 		}
4189 
4190 		btrfs_release_path(path);
4191 		if (key.offset < (u64)-1) {
4192 			key.offset++;
4193 		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4194 			key.offset = 0;
4195 			key.type = BTRFS_ROOT_ITEM_KEY;
4196 		} else if (key.objectid < (u64)-1) {
4197 			key.offset = 0;
4198 			key.type = BTRFS_ROOT_ITEM_KEY;
4199 			key.objectid++;
4200 		} else {
4201 			break;
4202 		}
4203 		cond_resched();
4204 	}
4205 
4206 out:
4207 	btrfs_free_path(path);
4208 	if (trans && !IS_ERR(trans))
4209 		btrfs_end_transaction(trans);
4210 	if (ret)
4211 		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4212 	else
4213 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4214 	up(&fs_info->uuid_tree_rescan_sem);
4215 	return 0;
4216 }
4217 
4218 /*
4219  * Callback for btrfs_uuid_tree_iterate().
4220  * returns:
4221  * 0	check succeeded, the entry is not outdated.
4222  * < 0	if an error occurred.
4223  * > 0	if the check failed, which means the caller shall remove the entry.
4224  */
4225 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4226 				       u8 *uuid, u8 type, u64 subid)
4227 {
4228 	struct btrfs_key key;
4229 	int ret = 0;
4230 	struct btrfs_root *subvol_root;
4231 
4232 	if (type != BTRFS_UUID_KEY_SUBVOL &&
4233 	    type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4234 		goto out;
4235 
4236 	key.objectid = subid;
4237 	key.type = BTRFS_ROOT_ITEM_KEY;
4238 	key.offset = (u64)-1;
4239 	subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4240 	if (IS_ERR(subvol_root)) {
4241 		ret = PTR_ERR(subvol_root);
4242 		if (ret == -ENOENT)
4243 			ret = 1;
4244 		goto out;
4245 	}
4246 
4247 	switch (type) {
4248 	case BTRFS_UUID_KEY_SUBVOL:
4249 		if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4250 			ret = 1;
4251 		break;
4252 	case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4253 		if (memcmp(uuid, subvol_root->root_item.received_uuid,
4254 			   BTRFS_UUID_SIZE))
4255 			ret = 1;
4256 		break;
4257 	}
4258 
4259 out:
4260 	return ret;
4261 }
4262 
4263 static int btrfs_uuid_rescan_kthread(void *data)
4264 {
4265 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4266 	int ret;
4267 
4268 	/*
4269 	 * 1st step is to iterate through the existing UUID tree and
4270 	 * to delete all entries that contain outdated data.
4271 	 * 2nd step is to add all missing entries to the UUID tree.
4272 	 */
4273 	ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4274 	if (ret < 0) {
4275 		btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4276 		up(&fs_info->uuid_tree_rescan_sem);
4277 		return ret;
4278 	}
4279 	return btrfs_uuid_scan_kthread(data);
4280 }
4281 
4282 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4283 {
4284 	struct btrfs_trans_handle *trans;
4285 	struct btrfs_root *tree_root = fs_info->tree_root;
4286 	struct btrfs_root *uuid_root;
4287 	struct task_struct *task;
4288 	int ret;
4289 
4290 	/*
4291 	 * 1 - root node
4292 	 * 1 - root item
4293 	 */
4294 	trans = btrfs_start_transaction(tree_root, 2);
4295 	if (IS_ERR(trans))
4296 		return PTR_ERR(trans);
4297 
4298 	uuid_root = btrfs_create_tree(trans, fs_info,
4299 				      BTRFS_UUID_TREE_OBJECTID);
4300 	if (IS_ERR(uuid_root)) {
4301 		ret = PTR_ERR(uuid_root);
4302 		btrfs_abort_transaction(trans, ret);
4303 		btrfs_end_transaction(trans);
4304 		return ret;
4305 	}
4306 
4307 	fs_info->uuid_root = uuid_root;
4308 
4309 	ret = btrfs_commit_transaction(trans);
4310 	if (ret)
4311 		return ret;
4312 
4313 	down(&fs_info->uuid_tree_rescan_sem);
4314 	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4315 	if (IS_ERR(task)) {
4316 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
4317 		btrfs_warn(fs_info, "failed to start uuid_scan task");
4318 		up(&fs_info->uuid_tree_rescan_sem);
4319 		return PTR_ERR(task);
4320 	}
4321 
4322 	return 0;
4323 }
4324 
4325 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4326 {
4327 	struct task_struct *task;
4328 
4329 	down(&fs_info->uuid_tree_rescan_sem);
4330 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4331 	if (IS_ERR(task)) {
4332 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
4333 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
4334 		up(&fs_info->uuid_tree_rescan_sem);
4335 		return PTR_ERR(task);
4336 	}
4337 
4338 	return 0;
4339 }
4340 
4341 /*
4342  * shrinking a device means finding all of the device extents past
4343  * the new size, and then following the back refs to the chunks.
4344  * The chunk relocation code actually frees the device extent
4345  */
4346 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4347 {
4348 	struct btrfs_fs_info *fs_info = device->fs_info;
4349 	struct btrfs_root *root = fs_info->dev_root;
4350 	struct btrfs_trans_handle *trans;
4351 	struct btrfs_dev_extent *dev_extent = NULL;
4352 	struct btrfs_path *path;
4353 	u64 length;
4354 	u64 chunk_offset;
4355 	int ret;
4356 	int slot;
4357 	int failed = 0;
4358 	bool retried = false;
4359 	bool checked_pending_chunks = false;
4360 	struct extent_buffer *l;
4361 	struct btrfs_key key;
4362 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4363 	u64 old_total = btrfs_super_total_bytes(super_copy);
4364 	u64 old_size = btrfs_device_get_total_bytes(device);
4365 	u64 diff = old_size - new_size;
4366 
4367 	if (device->is_tgtdev_for_dev_replace)
4368 		return -EINVAL;
4369 
4370 	path = btrfs_alloc_path();
4371 	if (!path)
4372 		return -ENOMEM;
4373 
4374 	path->reada = READA_FORWARD;
4375 
4376 	mutex_lock(&fs_info->chunk_mutex);
4377 
4378 	btrfs_device_set_total_bytes(device, new_size);
4379 	if (device->writeable) {
4380 		device->fs_devices->total_rw_bytes -= diff;
4381 		spin_lock(&fs_info->free_chunk_lock);
4382 		fs_info->free_chunk_space -= diff;
4383 		spin_unlock(&fs_info->free_chunk_lock);
4384 	}
4385 	mutex_unlock(&fs_info->chunk_mutex);
4386 
4387 again:
4388 	key.objectid = device->devid;
4389 	key.offset = (u64)-1;
4390 	key.type = BTRFS_DEV_EXTENT_KEY;
4391 
4392 	do {
4393 		mutex_lock(&fs_info->delete_unused_bgs_mutex);
4394 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4395 		if (ret < 0) {
4396 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4397 			goto done;
4398 		}
4399 
4400 		ret = btrfs_previous_item(root, path, 0, key.type);
4401 		if (ret)
4402 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4403 		if (ret < 0)
4404 			goto done;
4405 		if (ret) {
4406 			ret = 0;
4407 			btrfs_release_path(path);
4408 			break;
4409 		}
4410 
4411 		l = path->nodes[0];
4412 		slot = path->slots[0];
4413 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4414 
4415 		if (key.objectid != device->devid) {
4416 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4417 			btrfs_release_path(path);
4418 			break;
4419 		}
4420 
4421 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4422 		length = btrfs_dev_extent_length(l, dev_extent);
4423 
4424 		if (key.offset + length <= new_size) {
4425 			mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4426 			btrfs_release_path(path);
4427 			break;
4428 		}
4429 
4430 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4431 		btrfs_release_path(path);
4432 
4433 		ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4434 		mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4435 		if (ret && ret != -ENOSPC)
4436 			goto done;
4437 		if (ret == -ENOSPC)
4438 			failed++;
4439 	} while (key.offset-- > 0);
4440 
4441 	if (failed && !retried) {
4442 		failed = 0;
4443 		retried = true;
4444 		goto again;
4445 	} else if (failed && retried) {
4446 		ret = -ENOSPC;
4447 		goto done;
4448 	}
4449 
4450 	/* Shrinking succeeded, else we would be at "done". */
4451 	trans = btrfs_start_transaction(root, 0);
4452 	if (IS_ERR(trans)) {
4453 		ret = PTR_ERR(trans);
4454 		goto done;
4455 	}
4456 
4457 	mutex_lock(&fs_info->chunk_mutex);
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 
4475 		if (contains_pending_extent(trans->transaction, device,
4476 					    &start, len)) {
4477 			mutex_unlock(&fs_info->chunk_mutex);
4478 			checked_pending_chunks = true;
4479 			failed = 0;
4480 			retried = false;
4481 			ret = btrfs_commit_transaction(trans);
4482 			if (ret)
4483 				goto done;
4484 			goto again;
4485 		}
4486 	}
4487 
4488 	btrfs_device_set_disk_total_bytes(device, new_size);
4489 	if (list_empty(&device->resized_list))
4490 		list_add_tail(&device->resized_list,
4491 			      &fs_info->fs_devices->resized_devices);
4492 
4493 	WARN_ON(diff > old_total);
4494 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
4495 	mutex_unlock(&fs_info->chunk_mutex);
4496 
4497 	/* Now btrfs_update_device() will change the on-disk size. */
4498 	ret = btrfs_update_device(trans, device);
4499 	btrfs_end_transaction(trans);
4500 done:
4501 	btrfs_free_path(path);
4502 	if (ret) {
4503 		mutex_lock(&fs_info->chunk_mutex);
4504 		btrfs_device_set_total_bytes(device, old_size);
4505 		if (device->writeable)
4506 			device->fs_devices->total_rw_bytes += diff;
4507 		spin_lock(&fs_info->free_chunk_lock);
4508 		fs_info->free_chunk_space += diff;
4509 		spin_unlock(&fs_info->free_chunk_lock);
4510 		mutex_unlock(&fs_info->chunk_mutex);
4511 	}
4512 	return ret;
4513 }
4514 
4515 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4516 			   struct btrfs_key *key,
4517 			   struct btrfs_chunk *chunk, int item_size)
4518 {
4519 	struct btrfs_super_block *super_copy = fs_info->super_copy;
4520 	struct btrfs_disk_key disk_key;
4521 	u32 array_size;
4522 	u8 *ptr;
4523 
4524 	mutex_lock(&fs_info->chunk_mutex);
4525 	array_size = btrfs_super_sys_array_size(super_copy);
4526 	if (array_size + item_size + sizeof(disk_key)
4527 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4528 		mutex_unlock(&fs_info->chunk_mutex);
4529 		return -EFBIG;
4530 	}
4531 
4532 	ptr = super_copy->sys_chunk_array + array_size;
4533 	btrfs_cpu_key_to_disk(&disk_key, key);
4534 	memcpy(ptr, &disk_key, sizeof(disk_key));
4535 	ptr += sizeof(disk_key);
4536 	memcpy(ptr, chunk, item_size);
4537 	item_size += sizeof(disk_key);
4538 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4539 	mutex_unlock(&fs_info->chunk_mutex);
4540 
4541 	return 0;
4542 }
4543 
4544 /*
4545  * sort the devices in descending order by max_avail, total_avail
4546  */
4547 static int btrfs_cmp_device_info(const void *a, const void *b)
4548 {
4549 	const struct btrfs_device_info *di_a = a;
4550 	const struct btrfs_device_info *di_b = b;
4551 
4552 	if (di_a->max_avail > di_b->max_avail)
4553 		return -1;
4554 	if (di_a->max_avail < di_b->max_avail)
4555 		return 1;
4556 	if (di_a->total_avail > di_b->total_avail)
4557 		return -1;
4558 	if (di_a->total_avail < di_b->total_avail)
4559 		return 1;
4560 	return 0;
4561 }
4562 
4563 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4564 {
4565 	/* TODO allow them to set a preferred stripe size */
4566 	return SZ_64K;
4567 }
4568 
4569 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4570 {
4571 	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4572 		return;
4573 
4574 	btrfs_set_fs_incompat(info, RAID56);
4575 }
4576 
4577 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info)		\
4578 			- sizeof(struct btrfs_chunk))		\
4579 			/ sizeof(struct btrfs_stripe) + 1)
4580 
4581 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
4582 				- 2 * sizeof(struct btrfs_disk_key)	\
4583 				- 2 * sizeof(struct btrfs_chunk))	\
4584 				/ sizeof(struct btrfs_stripe) + 1)
4585 
4586 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4587 			       struct btrfs_fs_info *fs_info, u64 start,
4588 			       u64 type)
4589 {
4590 	struct btrfs_fs_info *info = trans->fs_info;
4591 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
4592 	struct list_head *cur;
4593 	struct map_lookup *map = NULL;
4594 	struct extent_map_tree *em_tree;
4595 	struct extent_map *em;
4596 	struct btrfs_device_info *devices_info = NULL;
4597 	u64 total_avail;
4598 	int num_stripes;	/* total number of stripes to allocate */
4599 	int data_stripes;	/* number of stripes that count for
4600 				   block group size */
4601 	int sub_stripes;	/* sub_stripes info for map */
4602 	int dev_stripes;	/* stripes per dev */
4603 	int devs_max;		/* max devs to use */
4604 	int devs_min;		/* min devs needed */
4605 	int devs_increment;	/* ndevs has to be a multiple of this */
4606 	int ncopies;		/* how many copies to data has */
4607 	int ret;
4608 	u64 max_stripe_size;
4609 	u64 max_chunk_size;
4610 	u64 stripe_size;
4611 	u64 num_bytes;
4612 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4613 	int ndevs;
4614 	int i;
4615 	int j;
4616 	int index;
4617 
4618 	BUG_ON(!alloc_profile_is_valid(type, 0));
4619 
4620 	if (list_empty(&fs_devices->alloc_list))
4621 		return -ENOSPC;
4622 
4623 	index = __get_raid_index(type);
4624 
4625 	sub_stripes = btrfs_raid_array[index].sub_stripes;
4626 	dev_stripes = btrfs_raid_array[index].dev_stripes;
4627 	devs_max = btrfs_raid_array[index].devs_max;
4628 	devs_min = btrfs_raid_array[index].devs_min;
4629 	devs_increment = btrfs_raid_array[index].devs_increment;
4630 	ncopies = btrfs_raid_array[index].ncopies;
4631 
4632 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4633 		max_stripe_size = SZ_1G;
4634 		max_chunk_size = 10 * max_stripe_size;
4635 		if (!devs_max)
4636 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4637 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4638 		/* for larger filesystems, use larger metadata chunks */
4639 		if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4640 			max_stripe_size = SZ_1G;
4641 		else
4642 			max_stripe_size = SZ_256M;
4643 		max_chunk_size = max_stripe_size;
4644 		if (!devs_max)
4645 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4646 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4647 		max_stripe_size = SZ_32M;
4648 		max_chunk_size = 2 * max_stripe_size;
4649 		if (!devs_max)
4650 			devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4651 	} else {
4652 		btrfs_err(info, "invalid chunk type 0x%llx requested",
4653 		       type);
4654 		BUG_ON(1);
4655 	}
4656 
4657 	/* we don't want a chunk larger than 10% of writeable space */
4658 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4659 			     max_chunk_size);
4660 
4661 	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4662 			       GFP_NOFS);
4663 	if (!devices_info)
4664 		return -ENOMEM;
4665 
4666 	cur = fs_devices->alloc_list.next;
4667 
4668 	/*
4669 	 * in the first pass through the devices list, we gather information
4670 	 * about the available holes on each device.
4671 	 */
4672 	ndevs = 0;
4673 	while (cur != &fs_devices->alloc_list) {
4674 		struct btrfs_device *device;
4675 		u64 max_avail;
4676 		u64 dev_offset;
4677 
4678 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4679 
4680 		cur = cur->next;
4681 
4682 		if (!device->writeable) {
4683 			WARN(1, KERN_ERR
4684 			       "BTRFS: read-only device in alloc_list\n");
4685 			continue;
4686 		}
4687 
4688 		if (!device->in_fs_metadata ||
4689 		    device->is_tgtdev_for_dev_replace)
4690 			continue;
4691 
4692 		if (device->total_bytes > device->bytes_used)
4693 			total_avail = device->total_bytes - device->bytes_used;
4694 		else
4695 			total_avail = 0;
4696 
4697 		/* If there is no space on this device, skip it. */
4698 		if (total_avail == 0)
4699 			continue;
4700 
4701 		ret = find_free_dev_extent(trans, device,
4702 					   max_stripe_size * dev_stripes,
4703 					   &dev_offset, &max_avail);
4704 		if (ret && ret != -ENOSPC)
4705 			goto error;
4706 
4707 		if (ret == 0)
4708 			max_avail = max_stripe_size * dev_stripes;
4709 
4710 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4711 			continue;
4712 
4713 		if (ndevs == fs_devices->rw_devices) {
4714 			WARN(1, "%s: found more than %llu devices\n",
4715 			     __func__, fs_devices->rw_devices);
4716 			break;
4717 		}
4718 		devices_info[ndevs].dev_offset = dev_offset;
4719 		devices_info[ndevs].max_avail = max_avail;
4720 		devices_info[ndevs].total_avail = total_avail;
4721 		devices_info[ndevs].dev = device;
4722 		++ndevs;
4723 	}
4724 
4725 	/*
4726 	 * now sort the devices by hole size / available space
4727 	 */
4728 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4729 	     btrfs_cmp_device_info, NULL);
4730 
4731 	/* round down to number of usable stripes */
4732 	ndevs -= ndevs % devs_increment;
4733 
4734 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4735 		ret = -ENOSPC;
4736 		goto error;
4737 	}
4738 
4739 	if (devs_max && ndevs > devs_max)
4740 		ndevs = devs_max;
4741 	/*
4742 	 * the primary goal is to maximize the number of stripes, so use as many
4743 	 * devices as possible, even if the stripes are not maximum sized.
4744 	 */
4745 	stripe_size = devices_info[ndevs-1].max_avail;
4746 	num_stripes = ndevs * dev_stripes;
4747 
4748 	/*
4749 	 * this will have to be fixed for RAID1 and RAID10 over
4750 	 * more drives
4751 	 */
4752 	data_stripes = num_stripes / ncopies;
4753 
4754 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
4755 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4756 							 info->stripesize);
4757 		data_stripes = num_stripes - 1;
4758 	}
4759 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
4760 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4761 							 info->stripesize);
4762 		data_stripes = num_stripes - 2;
4763 	}
4764 
4765 	/*
4766 	 * Use the number of data stripes to figure out how big this chunk
4767 	 * is really going to be in terms of logical address space,
4768 	 * and compare that answer with the max chunk size
4769 	 */
4770 	if (stripe_size * data_stripes > max_chunk_size) {
4771 		u64 mask = (1ULL << 24) - 1;
4772 
4773 		stripe_size = div_u64(max_chunk_size, data_stripes);
4774 
4775 		/* bump the answer up to a 16MB boundary */
4776 		stripe_size = (stripe_size + mask) & ~mask;
4777 
4778 		/* but don't go higher than the limits we found
4779 		 * while searching for free extents
4780 		 */
4781 		if (stripe_size > devices_info[ndevs-1].max_avail)
4782 			stripe_size = devices_info[ndevs-1].max_avail;
4783 	}
4784 
4785 	stripe_size = div_u64(stripe_size, dev_stripes);
4786 
4787 	/* align to BTRFS_STRIPE_LEN */
4788 	stripe_size = div_u64(stripe_size, raid_stripe_len);
4789 	stripe_size *= raid_stripe_len;
4790 
4791 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4792 	if (!map) {
4793 		ret = -ENOMEM;
4794 		goto error;
4795 	}
4796 	map->num_stripes = num_stripes;
4797 
4798 	for (i = 0; i < ndevs; ++i) {
4799 		for (j = 0; j < dev_stripes; ++j) {
4800 			int s = i * dev_stripes + j;
4801 			map->stripes[s].dev = devices_info[i].dev;
4802 			map->stripes[s].physical = devices_info[i].dev_offset +
4803 						   j * stripe_size;
4804 		}
4805 	}
4806 	map->sector_size = info->sectorsize;
4807 	map->stripe_len = raid_stripe_len;
4808 	map->io_align = raid_stripe_len;
4809 	map->io_width = raid_stripe_len;
4810 	map->type = type;
4811 	map->sub_stripes = sub_stripes;
4812 
4813 	num_bytes = stripe_size * data_stripes;
4814 
4815 	trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4816 
4817 	em = alloc_extent_map();
4818 	if (!em) {
4819 		kfree(map);
4820 		ret = -ENOMEM;
4821 		goto error;
4822 	}
4823 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4824 	em->map_lookup = map;
4825 	em->start = start;
4826 	em->len = num_bytes;
4827 	em->block_start = 0;
4828 	em->block_len = em->len;
4829 	em->orig_block_len = stripe_size;
4830 
4831 	em_tree = &info->mapping_tree.map_tree;
4832 	write_lock(&em_tree->lock);
4833 	ret = add_extent_mapping(em_tree, em, 0);
4834 	if (!ret) {
4835 		list_add_tail(&em->list, &trans->transaction->pending_chunks);
4836 		atomic_inc(&em->refs);
4837 	}
4838 	write_unlock(&em_tree->lock);
4839 	if (ret) {
4840 		free_extent_map(em);
4841 		goto error;
4842 	}
4843 
4844 	ret = btrfs_make_block_group(trans, info, 0, type,
4845 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4846 				     start, num_bytes);
4847 	if (ret)
4848 		goto error_del_extent;
4849 
4850 	for (i = 0; i < map->num_stripes; i++) {
4851 		num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4852 		btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4853 	}
4854 
4855 	spin_lock(&info->free_chunk_lock);
4856 	info->free_chunk_space -= (stripe_size * map->num_stripes);
4857 	spin_unlock(&info->free_chunk_lock);
4858 
4859 	free_extent_map(em);
4860 	check_raid56_incompat_flag(info, type);
4861 
4862 	kfree(devices_info);
4863 	return 0;
4864 
4865 error_del_extent:
4866 	write_lock(&em_tree->lock);
4867 	remove_extent_mapping(em_tree, em);
4868 	write_unlock(&em_tree->lock);
4869 
4870 	/* One for our allocation */
4871 	free_extent_map(em);
4872 	/* One for the tree reference */
4873 	free_extent_map(em);
4874 	/* One for the pending_chunks list reference */
4875 	free_extent_map(em);
4876 error:
4877 	kfree(devices_info);
4878 	return ret;
4879 }
4880 
4881 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4882 				struct btrfs_fs_info *fs_info,
4883 				u64 chunk_offset, u64 chunk_size)
4884 {
4885 	struct btrfs_root *extent_root = fs_info->extent_root;
4886 	struct btrfs_root *chunk_root = fs_info->chunk_root;
4887 	struct btrfs_key key;
4888 	struct btrfs_device *device;
4889 	struct btrfs_chunk *chunk;
4890 	struct btrfs_stripe *stripe;
4891 	struct extent_map_tree *em_tree;
4892 	struct extent_map *em;
4893 	struct map_lookup *map;
4894 	size_t item_size;
4895 	u64 dev_offset;
4896 	u64 stripe_size;
4897 	int i = 0;
4898 	int ret = 0;
4899 
4900 	em_tree = &fs_info->mapping_tree.map_tree;
4901 	read_lock(&em_tree->lock);
4902 	em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4903 	read_unlock(&em_tree->lock);
4904 
4905 	if (!em) {
4906 		btrfs_crit(fs_info, "unable to find logical %Lu len %Lu",
4907 			   chunk_offset, chunk_size);
4908 		return -EINVAL;
4909 	}
4910 
4911 	if (em->start != chunk_offset || em->len != chunk_size) {
4912 		btrfs_crit(fs_info,
4913 			   "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4914 			    chunk_offset, chunk_size, em->start, em->len);
4915 		free_extent_map(em);
4916 		return -EINVAL;
4917 	}
4918 
4919 	map = em->map_lookup;
4920 	item_size = btrfs_chunk_item_size(map->num_stripes);
4921 	stripe_size = em->orig_block_len;
4922 
4923 	chunk = kzalloc(item_size, GFP_NOFS);
4924 	if (!chunk) {
4925 		ret = -ENOMEM;
4926 		goto out;
4927 	}
4928 
4929 	/*
4930 	 * Take the device list mutex to prevent races with the final phase of
4931 	 * a device replace operation that replaces the device object associated
4932 	 * with the map's stripes, because the device object's id can change
4933 	 * at any time during that final phase of the device replace operation
4934 	 * (dev-replace.c:btrfs_dev_replace_finishing()).
4935 	 */
4936 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
4937 	for (i = 0; i < map->num_stripes; i++) {
4938 		device = map->stripes[i].dev;
4939 		dev_offset = map->stripes[i].physical;
4940 
4941 		ret = btrfs_update_device(trans, device);
4942 		if (ret)
4943 			break;
4944 		ret = btrfs_alloc_dev_extent(trans, device,
4945 					     chunk_root->root_key.objectid,
4946 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4947 					     chunk_offset, dev_offset,
4948 					     stripe_size);
4949 		if (ret)
4950 			break;
4951 	}
4952 	if (ret) {
4953 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4954 		goto out;
4955 	}
4956 
4957 	stripe = &chunk->stripe;
4958 	for (i = 0; i < map->num_stripes; i++) {
4959 		device = map->stripes[i].dev;
4960 		dev_offset = map->stripes[i].physical;
4961 
4962 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4963 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4964 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4965 		stripe++;
4966 	}
4967 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4968 
4969 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4970 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4971 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4972 	btrfs_set_stack_chunk_type(chunk, map->type);
4973 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4974 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4975 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4976 	btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4977 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4978 
4979 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4980 	key.type = BTRFS_CHUNK_ITEM_KEY;
4981 	key.offset = chunk_offset;
4982 
4983 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4984 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4985 		/*
4986 		 * TODO: Cleanup of inserted chunk root in case of
4987 		 * failure.
4988 		 */
4989 		ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4990 	}
4991 
4992 out:
4993 	kfree(chunk);
4994 	free_extent_map(em);
4995 	return ret;
4996 }
4997 
4998 /*
4999  * Chunk allocation falls into two parts. The first part does works
5000  * that make the new allocated chunk useable, but not do any operation
5001  * that modifies the chunk tree. The second part does the works that
5002  * require modifying the chunk tree. This division is important for the
5003  * bootstrap process of adding storage to a seed btrfs.
5004  */
5005 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5006 		      struct btrfs_fs_info *fs_info, u64 type)
5007 {
5008 	u64 chunk_offset;
5009 
5010 	ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5011 	chunk_offset = find_next_chunk(fs_info);
5012 	return __btrfs_alloc_chunk(trans, fs_info, chunk_offset, type);
5013 }
5014 
5015 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5016 					 struct btrfs_fs_info *fs_info,
5017 					 struct btrfs_device *device)
5018 {
5019 	struct btrfs_root *extent_root = fs_info->extent_root;
5020 	u64 chunk_offset;
5021 	u64 sys_chunk_offset;
5022 	u64 alloc_profile;
5023 	int ret;
5024 
5025 	chunk_offset = find_next_chunk(fs_info);
5026 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5027 	ret = __btrfs_alloc_chunk(trans, fs_info, chunk_offset, alloc_profile);
5028 	if (ret)
5029 		return ret;
5030 
5031 	sys_chunk_offset = find_next_chunk(fs_info);
5032 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5033 	ret = __btrfs_alloc_chunk(trans, fs_info, sys_chunk_offset,
5034 				  alloc_profile);
5035 	return ret;
5036 }
5037 
5038 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5039 {
5040 	int max_errors;
5041 
5042 	if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5043 			 BTRFS_BLOCK_GROUP_RAID10 |
5044 			 BTRFS_BLOCK_GROUP_RAID5 |
5045 			 BTRFS_BLOCK_GROUP_DUP)) {
5046 		max_errors = 1;
5047 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5048 		max_errors = 2;
5049 	} else {
5050 		max_errors = 0;
5051 	}
5052 
5053 	return max_errors;
5054 }
5055 
5056 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5057 {
5058 	struct extent_map *em;
5059 	struct map_lookup *map;
5060 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5061 	int readonly = 0;
5062 	int miss_ndevs = 0;
5063 	int i;
5064 
5065 	read_lock(&map_tree->map_tree.lock);
5066 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5067 	read_unlock(&map_tree->map_tree.lock);
5068 	if (!em)
5069 		return 1;
5070 
5071 	map = em->map_lookup;
5072 	for (i = 0; i < map->num_stripes; i++) {
5073 		if (map->stripes[i].dev->missing) {
5074 			miss_ndevs++;
5075 			continue;
5076 		}
5077 
5078 		if (!map->stripes[i].dev->writeable) {
5079 			readonly = 1;
5080 			goto end;
5081 		}
5082 	}
5083 
5084 	/*
5085 	 * If the number of missing devices is larger than max errors,
5086 	 * we can not write the data into that chunk successfully, so
5087 	 * set it readonly.
5088 	 */
5089 	if (miss_ndevs > btrfs_chunk_max_errors(map))
5090 		readonly = 1;
5091 end:
5092 	free_extent_map(em);
5093 	return readonly;
5094 }
5095 
5096 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5097 {
5098 	extent_map_tree_init(&tree->map_tree);
5099 }
5100 
5101 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5102 {
5103 	struct extent_map *em;
5104 
5105 	while (1) {
5106 		write_lock(&tree->map_tree.lock);
5107 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5108 		if (em)
5109 			remove_extent_mapping(&tree->map_tree, em);
5110 		write_unlock(&tree->map_tree.lock);
5111 		if (!em)
5112 			break;
5113 		/* once for us */
5114 		free_extent_map(em);
5115 		/* once for the tree */
5116 		free_extent_map(em);
5117 	}
5118 }
5119 
5120 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5121 {
5122 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5123 	struct extent_map *em;
5124 	struct map_lookup *map;
5125 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5126 	int ret;
5127 
5128 	read_lock(&em_tree->lock);
5129 	em = lookup_extent_mapping(em_tree, logical, len);
5130 	read_unlock(&em_tree->lock);
5131 
5132 	/*
5133 	 * We could return errors for these cases, but that could get ugly and
5134 	 * we'd probably do the same thing which is just not do anything else
5135 	 * and exit, so return 1 so the callers don't try to use other copies.
5136 	 */
5137 	if (!em) {
5138 		btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5139 			    logical+len);
5140 		return 1;
5141 	}
5142 
5143 	if (em->start > logical || em->start + em->len < logical) {
5144 		btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5145 			   logical, logical+len, em->start,
5146 			   em->start + em->len);
5147 		free_extent_map(em);
5148 		return 1;
5149 	}
5150 
5151 	map = em->map_lookup;
5152 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5153 		ret = map->num_stripes;
5154 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5155 		ret = map->sub_stripes;
5156 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5157 		ret = 2;
5158 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5159 		ret = 3;
5160 	else
5161 		ret = 1;
5162 	free_extent_map(em);
5163 
5164 	btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5165 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5166 		ret++;
5167 	btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5168 
5169 	return ret;
5170 }
5171 
5172 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5173 				    struct btrfs_mapping_tree *map_tree,
5174 				    u64 logical)
5175 {
5176 	struct extent_map *em;
5177 	struct map_lookup *map;
5178 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5179 	unsigned long len = fs_info->sectorsize;
5180 
5181 	read_lock(&em_tree->lock);
5182 	em = lookup_extent_mapping(em_tree, logical, len);
5183 	read_unlock(&em_tree->lock);
5184 	BUG_ON(!em);
5185 
5186 	BUG_ON(em->start > logical || em->start + em->len < logical);
5187 	map = em->map_lookup;
5188 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5189 		len = map->stripe_len * nr_data_stripes(map);
5190 	free_extent_map(em);
5191 	return len;
5192 }
5193 
5194 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5195 			   u64 logical, u64 len, int mirror_num)
5196 {
5197 	struct extent_map *em;
5198 	struct map_lookup *map;
5199 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5200 	int ret = 0;
5201 
5202 	read_lock(&em_tree->lock);
5203 	em = lookup_extent_mapping(em_tree, logical, len);
5204 	read_unlock(&em_tree->lock);
5205 	BUG_ON(!em);
5206 
5207 	BUG_ON(em->start > logical || em->start + em->len < logical);
5208 	map = em->map_lookup;
5209 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5210 		ret = 1;
5211 	free_extent_map(em);
5212 	return ret;
5213 }
5214 
5215 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5216 			    struct map_lookup *map, int first, int num,
5217 			    int optimal, int dev_replace_is_ongoing)
5218 {
5219 	int i;
5220 	int tolerance;
5221 	struct btrfs_device *srcdev;
5222 
5223 	if (dev_replace_is_ongoing &&
5224 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5225 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5226 		srcdev = fs_info->dev_replace.srcdev;
5227 	else
5228 		srcdev = NULL;
5229 
5230 	/*
5231 	 * try to avoid the drive that is the source drive for a
5232 	 * dev-replace procedure, only choose it if no other non-missing
5233 	 * mirror is available
5234 	 */
5235 	for (tolerance = 0; tolerance < 2; tolerance++) {
5236 		if (map->stripes[optimal].dev->bdev &&
5237 		    (tolerance || map->stripes[optimal].dev != srcdev))
5238 			return optimal;
5239 		for (i = first; i < first + num; i++) {
5240 			if (map->stripes[i].dev->bdev &&
5241 			    (tolerance || map->stripes[i].dev != srcdev))
5242 				return i;
5243 		}
5244 	}
5245 
5246 	/* we couldn't find one that doesn't fail.  Just return something
5247 	 * and the io error handling code will clean up eventually
5248 	 */
5249 	return optimal;
5250 }
5251 
5252 static inline int parity_smaller(u64 a, u64 b)
5253 {
5254 	return a > b;
5255 }
5256 
5257 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5258 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5259 {
5260 	struct btrfs_bio_stripe s;
5261 	int i;
5262 	u64 l;
5263 	int again = 1;
5264 
5265 	while (again) {
5266 		again = 0;
5267 		for (i = 0; i < num_stripes - 1; i++) {
5268 			if (parity_smaller(bbio->raid_map[i],
5269 					   bbio->raid_map[i+1])) {
5270 				s = bbio->stripes[i];
5271 				l = bbio->raid_map[i];
5272 				bbio->stripes[i] = bbio->stripes[i+1];
5273 				bbio->raid_map[i] = bbio->raid_map[i+1];
5274 				bbio->stripes[i+1] = s;
5275 				bbio->raid_map[i+1] = l;
5276 
5277 				again = 1;
5278 			}
5279 		}
5280 	}
5281 }
5282 
5283 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5284 {
5285 	struct btrfs_bio *bbio = kzalloc(
5286 		 /* the size of the btrfs_bio */
5287 		sizeof(struct btrfs_bio) +
5288 		/* plus the variable array for the stripes */
5289 		sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5290 		/* plus the variable array for the tgt dev */
5291 		sizeof(int) * (real_stripes) +
5292 		/*
5293 		 * plus the raid_map, which includes both the tgt dev
5294 		 * and the stripes
5295 		 */
5296 		sizeof(u64) * (total_stripes),
5297 		GFP_NOFS|__GFP_NOFAIL);
5298 
5299 	atomic_set(&bbio->error, 0);
5300 	atomic_set(&bbio->refs, 1);
5301 
5302 	return bbio;
5303 }
5304 
5305 void btrfs_get_bbio(struct btrfs_bio *bbio)
5306 {
5307 	WARN_ON(!atomic_read(&bbio->refs));
5308 	atomic_inc(&bbio->refs);
5309 }
5310 
5311 void btrfs_put_bbio(struct btrfs_bio *bbio)
5312 {
5313 	if (!bbio)
5314 		return;
5315 	if (atomic_dec_and_test(&bbio->refs))
5316 		kfree(bbio);
5317 }
5318 
5319 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5320 			     enum btrfs_map_op op,
5321 			     u64 logical, u64 *length,
5322 			     struct btrfs_bio **bbio_ret,
5323 			     int mirror_num, int need_raid_map)
5324 {
5325 	struct extent_map *em;
5326 	struct map_lookup *map;
5327 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5328 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5329 	u64 offset;
5330 	u64 stripe_offset;
5331 	u64 stripe_end_offset;
5332 	u64 stripe_nr;
5333 	u64 stripe_nr_orig;
5334 	u64 stripe_nr_end;
5335 	u64 stripe_len;
5336 	u32 stripe_index;
5337 	int i;
5338 	int ret = 0;
5339 	int num_stripes;
5340 	int max_errors = 0;
5341 	int tgtdev_indexes = 0;
5342 	struct btrfs_bio *bbio = NULL;
5343 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5344 	int dev_replace_is_ongoing = 0;
5345 	int num_alloc_stripes;
5346 	int patch_the_first_stripe_for_dev_replace = 0;
5347 	u64 physical_to_patch_in_first_stripe = 0;
5348 	u64 raid56_full_stripe_start = (u64)-1;
5349 
5350 	read_lock(&em_tree->lock);
5351 	em = lookup_extent_mapping(em_tree, logical, *length);
5352 	read_unlock(&em_tree->lock);
5353 
5354 	if (!em) {
5355 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5356 			logical, *length);
5357 		return -EINVAL;
5358 	}
5359 
5360 	if (em->start > logical || em->start + em->len < logical) {
5361 		btrfs_crit(fs_info,
5362 			   "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5363 			   logical, em->start, em->start + em->len);
5364 		free_extent_map(em);
5365 		return -EINVAL;
5366 	}
5367 
5368 	map = em->map_lookup;
5369 	offset = logical - em->start;
5370 
5371 	stripe_len = map->stripe_len;
5372 	stripe_nr = offset;
5373 	/*
5374 	 * stripe_nr counts the total number of stripes we have to stride
5375 	 * to get to this block
5376 	 */
5377 	stripe_nr = div64_u64(stripe_nr, stripe_len);
5378 
5379 	stripe_offset = stripe_nr * stripe_len;
5380 	if (offset < stripe_offset) {
5381 		btrfs_crit(fs_info,
5382 			   "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5383 			   stripe_offset, offset, em->start, logical,
5384 			   stripe_len);
5385 		free_extent_map(em);
5386 		return -EINVAL;
5387 	}
5388 
5389 	/* stripe_offset is the offset of this block in its stripe*/
5390 	stripe_offset = offset - stripe_offset;
5391 
5392 	/* if we're here for raid56, we need to know the stripe aligned start */
5393 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5394 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5395 		raid56_full_stripe_start = offset;
5396 
5397 		/* allow a write of a full stripe, but make sure we don't
5398 		 * allow straddling of stripes
5399 		 */
5400 		raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5401 				full_stripe_len);
5402 		raid56_full_stripe_start *= full_stripe_len;
5403 	}
5404 
5405 	if (op == BTRFS_MAP_DISCARD) {
5406 		/* we don't discard raid56 yet */
5407 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5408 			ret = -EOPNOTSUPP;
5409 			goto out;
5410 		}
5411 		*length = min_t(u64, em->len - offset, *length);
5412 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5413 		u64 max_len;
5414 		/* For writes to RAID[56], allow a full stripeset across all disks.
5415 		   For other RAID types and for RAID[56] reads, just allow a single
5416 		   stripe (on a single disk). */
5417 		if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5418 		    (op == BTRFS_MAP_WRITE)) {
5419 			max_len = stripe_len * nr_data_stripes(map) -
5420 				(offset - raid56_full_stripe_start);
5421 		} else {
5422 			/* we limit the length of each bio to what fits in a stripe */
5423 			max_len = stripe_len - stripe_offset;
5424 		}
5425 		*length = min_t(u64, em->len - offset, max_len);
5426 	} else {
5427 		*length = em->len - offset;
5428 	}
5429 
5430 	/* This is for when we're called from btrfs_merge_bio_hook() and all
5431 	   it cares about is the length */
5432 	if (!bbio_ret)
5433 		goto out;
5434 
5435 	btrfs_dev_replace_lock(dev_replace, 0);
5436 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5437 	if (!dev_replace_is_ongoing)
5438 		btrfs_dev_replace_unlock(dev_replace, 0);
5439 	else
5440 		btrfs_dev_replace_set_lock_blocking(dev_replace);
5441 
5442 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5443 	    op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5444 	    op != BTRFS_MAP_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5445 		/*
5446 		 * in dev-replace case, for repair case (that's the only
5447 		 * case where the mirror is selected explicitly when
5448 		 * calling btrfs_map_block), blocks left of the left cursor
5449 		 * can also be read from the target drive.
5450 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
5451 		 * the last one to the array of stripes. For READ, it also
5452 		 * needs to be supported using the same mirror number.
5453 		 * If the requested block is not left of the left cursor,
5454 		 * EIO is returned. This can happen because btrfs_num_copies()
5455 		 * returns one more in the dev-replace case.
5456 		 */
5457 		u64 tmp_length = *length;
5458 		struct btrfs_bio *tmp_bbio = NULL;
5459 		int tmp_num_stripes;
5460 		u64 srcdev_devid = dev_replace->srcdev->devid;
5461 		int index_srcdev = 0;
5462 		int found = 0;
5463 		u64 physical_of_found = 0;
5464 
5465 		ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5466 			     logical, &tmp_length, &tmp_bbio, 0, 0);
5467 		if (ret) {
5468 			WARN_ON(tmp_bbio != NULL);
5469 			goto out;
5470 		}
5471 
5472 		tmp_num_stripes = tmp_bbio->num_stripes;
5473 		if (mirror_num > tmp_num_stripes) {
5474 			/*
5475 			 * BTRFS_MAP_GET_READ_MIRRORS does not contain this
5476 			 * mirror, that means that the requested area
5477 			 * is not left of the left cursor
5478 			 */
5479 			ret = -EIO;
5480 			btrfs_put_bbio(tmp_bbio);
5481 			goto out;
5482 		}
5483 
5484 		/*
5485 		 * process the rest of the function using the mirror_num
5486 		 * of the source drive. Therefore look it up first.
5487 		 * At the end, patch the device pointer to the one of the
5488 		 * target drive.
5489 		 */
5490 		for (i = 0; i < tmp_num_stripes; i++) {
5491 			if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5492 				continue;
5493 
5494 			/*
5495 			 * In case of DUP, in order to keep it simple, only add
5496 			 * the mirror with the lowest physical address
5497 			 */
5498 			if (found &&
5499 			    physical_of_found <= tmp_bbio->stripes[i].physical)
5500 				continue;
5501 
5502 			index_srcdev = i;
5503 			found = 1;
5504 			physical_of_found = tmp_bbio->stripes[i].physical;
5505 		}
5506 
5507 		btrfs_put_bbio(tmp_bbio);
5508 
5509 		if (!found) {
5510 			WARN_ON(1);
5511 			ret = -EIO;
5512 			goto out;
5513 		}
5514 
5515 		mirror_num = index_srcdev + 1;
5516 		patch_the_first_stripe_for_dev_replace = 1;
5517 		physical_to_patch_in_first_stripe = physical_of_found;
5518 	} else if (mirror_num > map->num_stripes) {
5519 		mirror_num = 0;
5520 	}
5521 
5522 	num_stripes = 1;
5523 	stripe_index = 0;
5524 	stripe_nr_orig = stripe_nr;
5525 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5526 	stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5527 	stripe_end_offset = stripe_nr_end * map->stripe_len -
5528 			    (offset + *length);
5529 
5530 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5531 		if (op == BTRFS_MAP_DISCARD)
5532 			num_stripes = min_t(u64, map->num_stripes,
5533 					    stripe_nr_end - stripe_nr_orig);
5534 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5535 				&stripe_index);
5536 		if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5537 		    op != BTRFS_MAP_GET_READ_MIRRORS)
5538 			mirror_num = 1;
5539 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5540 		if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5541 		    op == BTRFS_MAP_GET_READ_MIRRORS)
5542 			num_stripes = map->num_stripes;
5543 		else if (mirror_num)
5544 			stripe_index = mirror_num - 1;
5545 		else {
5546 			stripe_index = find_live_mirror(fs_info, map, 0,
5547 					    map->num_stripes,
5548 					    current->pid % map->num_stripes,
5549 					    dev_replace_is_ongoing);
5550 			mirror_num = stripe_index + 1;
5551 		}
5552 
5553 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5554 		if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD ||
5555 		    op == BTRFS_MAP_GET_READ_MIRRORS) {
5556 			num_stripes = map->num_stripes;
5557 		} else if (mirror_num) {
5558 			stripe_index = mirror_num - 1;
5559 		} else {
5560 			mirror_num = 1;
5561 		}
5562 
5563 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5564 		u32 factor = map->num_stripes / map->sub_stripes;
5565 
5566 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5567 		stripe_index *= map->sub_stripes;
5568 
5569 		if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5570 			num_stripes = map->sub_stripes;
5571 		else if (op == BTRFS_MAP_DISCARD)
5572 			num_stripes = min_t(u64, map->sub_stripes *
5573 					    (stripe_nr_end - stripe_nr_orig),
5574 					    map->num_stripes);
5575 		else if (mirror_num)
5576 			stripe_index += mirror_num - 1;
5577 		else {
5578 			int old_stripe_index = stripe_index;
5579 			stripe_index = find_live_mirror(fs_info, map,
5580 					      stripe_index,
5581 					      map->sub_stripes, stripe_index +
5582 					      current->pid % map->sub_stripes,
5583 					      dev_replace_is_ongoing);
5584 			mirror_num = stripe_index - old_stripe_index + 1;
5585 		}
5586 
5587 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5588 		if (need_raid_map &&
5589 		    (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5590 		     mirror_num > 1)) {
5591 			/* push stripe_nr back to the start of the full stripe */
5592 			stripe_nr = div_u64(raid56_full_stripe_start,
5593 					stripe_len * nr_data_stripes(map));
5594 
5595 			/* RAID[56] write or recovery. Return all stripes */
5596 			num_stripes = map->num_stripes;
5597 			max_errors = nr_parity_stripes(map);
5598 
5599 			*length = map->stripe_len;
5600 			stripe_index = 0;
5601 			stripe_offset = 0;
5602 		} else {
5603 			/*
5604 			 * Mirror #0 or #1 means the original data block.
5605 			 * Mirror #2 is RAID5 parity block.
5606 			 * Mirror #3 is RAID6 Q block.
5607 			 */
5608 			stripe_nr = div_u64_rem(stripe_nr,
5609 					nr_data_stripes(map), &stripe_index);
5610 			if (mirror_num > 1)
5611 				stripe_index = nr_data_stripes(map) +
5612 						mirror_num - 2;
5613 
5614 			/* We distribute the parity blocks across stripes */
5615 			div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5616 					&stripe_index);
5617 			if ((op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD &&
5618 			    op != BTRFS_MAP_GET_READ_MIRRORS) && mirror_num <= 1)
5619 				mirror_num = 1;
5620 		}
5621 	} else {
5622 		/*
5623 		 * after this, stripe_nr is the number of stripes on this
5624 		 * device we have to walk to find the data, and stripe_index is
5625 		 * the number of our device in the stripe array
5626 		 */
5627 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5628 				&stripe_index);
5629 		mirror_num = stripe_index + 1;
5630 	}
5631 	if (stripe_index >= map->num_stripes) {
5632 		btrfs_crit(fs_info,
5633 			   "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5634 			   stripe_index, map->num_stripes);
5635 		ret = -EINVAL;
5636 		goto out;
5637 	}
5638 
5639 	num_alloc_stripes = num_stripes;
5640 	if (dev_replace_is_ongoing) {
5641 		if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD)
5642 			num_alloc_stripes <<= 1;
5643 		if (op == BTRFS_MAP_GET_READ_MIRRORS)
5644 			num_alloc_stripes++;
5645 		tgtdev_indexes = num_stripes;
5646 	}
5647 
5648 	bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5649 	if (!bbio) {
5650 		ret = -ENOMEM;
5651 		goto out;
5652 	}
5653 	if (dev_replace_is_ongoing)
5654 		bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5655 
5656 	/* build raid_map */
5657 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5658 	    need_raid_map &&
5659 	    ((op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) ||
5660 	    mirror_num > 1)) {
5661 		u64 tmp;
5662 		unsigned rot;
5663 
5664 		bbio->raid_map = (u64 *)((void *)bbio->stripes +
5665 				 sizeof(struct btrfs_bio_stripe) *
5666 				 num_alloc_stripes +
5667 				 sizeof(int) * tgtdev_indexes);
5668 
5669 		/* Work out the disk rotation on this stripe-set */
5670 		div_u64_rem(stripe_nr, num_stripes, &rot);
5671 
5672 		/* Fill in the logical address of each stripe */
5673 		tmp = stripe_nr * nr_data_stripes(map);
5674 		for (i = 0; i < nr_data_stripes(map); i++)
5675 			bbio->raid_map[(i+rot) % num_stripes] =
5676 				em->start + (tmp + i) * map->stripe_len;
5677 
5678 		bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5679 		if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5680 			bbio->raid_map[(i+rot+1) % num_stripes] =
5681 				RAID6_Q_STRIPE;
5682 	}
5683 
5684 	if (op == BTRFS_MAP_DISCARD) {
5685 		u32 factor = 0;
5686 		u32 sub_stripes = 0;
5687 		u64 stripes_per_dev = 0;
5688 		u32 remaining_stripes = 0;
5689 		u32 last_stripe = 0;
5690 
5691 		if (map->type &
5692 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5693 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5694 				sub_stripes = 1;
5695 			else
5696 				sub_stripes = map->sub_stripes;
5697 
5698 			factor = map->num_stripes / sub_stripes;
5699 			stripes_per_dev = div_u64_rem(stripe_nr_end -
5700 						      stripe_nr_orig,
5701 						      factor,
5702 						      &remaining_stripes);
5703 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5704 			last_stripe *= sub_stripes;
5705 		}
5706 
5707 		for (i = 0; i < num_stripes; i++) {
5708 			bbio->stripes[i].physical =
5709 				map->stripes[stripe_index].physical +
5710 				stripe_offset + stripe_nr * map->stripe_len;
5711 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5712 
5713 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5714 					 BTRFS_BLOCK_GROUP_RAID10)) {
5715 				bbio->stripes[i].length = stripes_per_dev *
5716 							  map->stripe_len;
5717 
5718 				if (i / sub_stripes < remaining_stripes)
5719 					bbio->stripes[i].length +=
5720 						map->stripe_len;
5721 
5722 				/*
5723 				 * Special for the first stripe and
5724 				 * the last stripe:
5725 				 *
5726 				 * |-------|...|-------|
5727 				 *     |----------|
5728 				 *    off     end_off
5729 				 */
5730 				if (i < sub_stripes)
5731 					bbio->stripes[i].length -=
5732 						stripe_offset;
5733 
5734 				if (stripe_index >= last_stripe &&
5735 				    stripe_index <= (last_stripe +
5736 						     sub_stripes - 1))
5737 					bbio->stripes[i].length -=
5738 						stripe_end_offset;
5739 
5740 				if (i == sub_stripes - 1)
5741 					stripe_offset = 0;
5742 			} else
5743 				bbio->stripes[i].length = *length;
5744 
5745 			stripe_index++;
5746 			if (stripe_index == map->num_stripes) {
5747 				/* This could only happen for RAID0/10 */
5748 				stripe_index = 0;
5749 				stripe_nr++;
5750 			}
5751 		}
5752 	} else {
5753 		for (i = 0; i < num_stripes; i++) {
5754 			bbio->stripes[i].physical =
5755 				map->stripes[stripe_index].physical +
5756 				stripe_offset +
5757 				stripe_nr * map->stripe_len;
5758 			bbio->stripes[i].dev =
5759 				map->stripes[stripe_index].dev;
5760 			stripe_index++;
5761 		}
5762 	}
5763 
5764 	if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5765 		max_errors = btrfs_chunk_max_errors(map);
5766 
5767 	if (bbio->raid_map)
5768 		sort_parity_stripes(bbio, num_stripes);
5769 
5770 	tgtdev_indexes = 0;
5771 	if (dev_replace_is_ongoing &&
5772 	   (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD) &&
5773 	    dev_replace->tgtdev != NULL) {
5774 		int index_where_to_add;
5775 		u64 srcdev_devid = dev_replace->srcdev->devid;
5776 
5777 		/*
5778 		 * duplicate the write operations while the dev replace
5779 		 * procedure is running. Since the copying of the old disk
5780 		 * to the new disk takes place at run time while the
5781 		 * filesystem is mounted writable, the regular write
5782 		 * operations to the old disk have to be duplicated to go
5783 		 * to the new disk as well.
5784 		 * Note that device->missing is handled by the caller, and
5785 		 * that the write to the old disk is already set up in the
5786 		 * stripes array.
5787 		 */
5788 		index_where_to_add = num_stripes;
5789 		for (i = 0; i < num_stripes; i++) {
5790 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5791 				/* write to new disk, too */
5792 				struct btrfs_bio_stripe *new =
5793 					bbio->stripes + index_where_to_add;
5794 				struct btrfs_bio_stripe *old =
5795 					bbio->stripes + i;
5796 
5797 				new->physical = old->physical;
5798 				new->length = old->length;
5799 				new->dev = dev_replace->tgtdev;
5800 				bbio->tgtdev_map[i] = index_where_to_add;
5801 				index_where_to_add++;
5802 				max_errors++;
5803 				tgtdev_indexes++;
5804 			}
5805 		}
5806 		num_stripes = index_where_to_add;
5807 	} else if (dev_replace_is_ongoing &&
5808 		   op == BTRFS_MAP_GET_READ_MIRRORS &&
5809 		   dev_replace->tgtdev != NULL) {
5810 		u64 srcdev_devid = dev_replace->srcdev->devid;
5811 		int index_srcdev = 0;
5812 		int found = 0;
5813 		u64 physical_of_found = 0;
5814 
5815 		/*
5816 		 * During the dev-replace procedure, the target drive can
5817 		 * also be used to read data in case it is needed to repair
5818 		 * a corrupt block elsewhere. This is possible if the
5819 		 * requested area is left of the left cursor. In this area,
5820 		 * the target drive is a full copy of the source drive.
5821 		 */
5822 		for (i = 0; i < num_stripes; i++) {
5823 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5824 				/*
5825 				 * In case of DUP, in order to keep it
5826 				 * simple, only add the mirror with the
5827 				 * lowest physical address
5828 				 */
5829 				if (found &&
5830 				    physical_of_found <=
5831 				     bbio->stripes[i].physical)
5832 					continue;
5833 				index_srcdev = i;
5834 				found = 1;
5835 				physical_of_found = bbio->stripes[i].physical;
5836 			}
5837 		}
5838 		if (found) {
5839 			struct btrfs_bio_stripe *tgtdev_stripe =
5840 				bbio->stripes + num_stripes;
5841 
5842 			tgtdev_stripe->physical = physical_of_found;
5843 			tgtdev_stripe->length =
5844 				bbio->stripes[index_srcdev].length;
5845 			tgtdev_stripe->dev = dev_replace->tgtdev;
5846 			bbio->tgtdev_map[index_srcdev] = num_stripes;
5847 
5848 			tgtdev_indexes++;
5849 			num_stripes++;
5850 		}
5851 	}
5852 
5853 	*bbio_ret = bbio;
5854 	bbio->map_type = map->type;
5855 	bbio->num_stripes = num_stripes;
5856 	bbio->max_errors = max_errors;
5857 	bbio->mirror_num = mirror_num;
5858 	bbio->num_tgtdevs = tgtdev_indexes;
5859 
5860 	/*
5861 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5862 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
5863 	 * available as a mirror
5864 	 */
5865 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5866 		WARN_ON(num_stripes > 1);
5867 		bbio->stripes[0].dev = dev_replace->tgtdev;
5868 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5869 		bbio->mirror_num = map->num_stripes + 1;
5870 	}
5871 out:
5872 	if (dev_replace_is_ongoing) {
5873 		btrfs_dev_replace_clear_lock_blocking(dev_replace);
5874 		btrfs_dev_replace_unlock(dev_replace, 0);
5875 	}
5876 	free_extent_map(em);
5877 	return ret;
5878 }
5879 
5880 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5881 		      u64 logical, u64 *length,
5882 		      struct btrfs_bio **bbio_ret, int mirror_num)
5883 {
5884 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5885 				 mirror_num, 0);
5886 }
5887 
5888 /* For Scrub/replace */
5889 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5890 		     u64 logical, u64 *length,
5891 		     struct btrfs_bio **bbio_ret, int mirror_num,
5892 		     int need_raid_map)
5893 {
5894 	return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5895 				 mirror_num, need_raid_map);
5896 }
5897 
5898 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5899 		     u64 chunk_start, u64 physical, u64 devid,
5900 		     u64 **logical, int *naddrs, int *stripe_len)
5901 {
5902 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5903 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5904 	struct extent_map *em;
5905 	struct map_lookup *map;
5906 	u64 *buf;
5907 	u64 bytenr;
5908 	u64 length;
5909 	u64 stripe_nr;
5910 	u64 rmap_len;
5911 	int i, j, nr = 0;
5912 
5913 	read_lock(&em_tree->lock);
5914 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
5915 	read_unlock(&em_tree->lock);
5916 
5917 	if (!em) {
5918 		btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5919 			chunk_start);
5920 		return -EIO;
5921 	}
5922 
5923 	if (em->start != chunk_start) {
5924 		btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
5925 		       em->start, chunk_start);
5926 		free_extent_map(em);
5927 		return -EIO;
5928 	}
5929 	map = em->map_lookup;
5930 
5931 	length = em->len;
5932 	rmap_len = map->stripe_len;
5933 
5934 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5935 		length = div_u64(length, map->num_stripes / map->sub_stripes);
5936 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5937 		length = div_u64(length, map->num_stripes);
5938 	else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5939 		length = div_u64(length, nr_data_stripes(map));
5940 		rmap_len = map->stripe_len * nr_data_stripes(map);
5941 	}
5942 
5943 	buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5944 	BUG_ON(!buf); /* -ENOMEM */
5945 
5946 	for (i = 0; i < map->num_stripes; i++) {
5947 		if (devid && map->stripes[i].dev->devid != devid)
5948 			continue;
5949 		if (map->stripes[i].physical > physical ||
5950 		    map->stripes[i].physical + length <= physical)
5951 			continue;
5952 
5953 		stripe_nr = physical - map->stripes[i].physical;
5954 		stripe_nr = div_u64(stripe_nr, map->stripe_len);
5955 
5956 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5957 			stripe_nr = stripe_nr * map->num_stripes + i;
5958 			stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5959 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5960 			stripe_nr = stripe_nr * map->num_stripes + i;
5961 		} /* else if RAID[56], multiply by nr_data_stripes().
5962 		   * Alternatively, just use rmap_len below instead of
5963 		   * map->stripe_len */
5964 
5965 		bytenr = chunk_start + stripe_nr * rmap_len;
5966 		WARN_ON(nr >= map->num_stripes);
5967 		for (j = 0; j < nr; j++) {
5968 			if (buf[j] == bytenr)
5969 				break;
5970 		}
5971 		if (j == nr) {
5972 			WARN_ON(nr >= map->num_stripes);
5973 			buf[nr++] = bytenr;
5974 		}
5975 	}
5976 
5977 	*logical = buf;
5978 	*naddrs = nr;
5979 	*stripe_len = rmap_len;
5980 
5981 	free_extent_map(em);
5982 	return 0;
5983 }
5984 
5985 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5986 {
5987 	bio->bi_private = bbio->private;
5988 	bio->bi_end_io = bbio->end_io;
5989 	bio_endio(bio);
5990 
5991 	btrfs_put_bbio(bbio);
5992 }
5993 
5994 static void btrfs_end_bio(struct bio *bio)
5995 {
5996 	struct btrfs_bio *bbio = bio->bi_private;
5997 	int is_orig_bio = 0;
5998 
5999 	if (bio->bi_error) {
6000 		atomic_inc(&bbio->error);
6001 		if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6002 			unsigned int stripe_index =
6003 				btrfs_io_bio(bio)->stripe_index;
6004 			struct btrfs_device *dev;
6005 
6006 			BUG_ON(stripe_index >= bbio->num_stripes);
6007 			dev = bbio->stripes[stripe_index].dev;
6008 			if (dev->bdev) {
6009 				if (bio_op(bio) == REQ_OP_WRITE)
6010 					btrfs_dev_stat_inc(dev,
6011 						BTRFS_DEV_STAT_WRITE_ERRS);
6012 				else
6013 					btrfs_dev_stat_inc(dev,
6014 						BTRFS_DEV_STAT_READ_ERRS);
6015 				if (bio->bi_opf & REQ_PREFLUSH)
6016 					btrfs_dev_stat_inc(dev,
6017 						BTRFS_DEV_STAT_FLUSH_ERRS);
6018 				btrfs_dev_stat_print_on_error(dev);
6019 			}
6020 		}
6021 	}
6022 
6023 	if (bio == bbio->orig_bio)
6024 		is_orig_bio = 1;
6025 
6026 	btrfs_bio_counter_dec(bbio->fs_info);
6027 
6028 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6029 		if (!is_orig_bio) {
6030 			bio_put(bio);
6031 			bio = bbio->orig_bio;
6032 		}
6033 
6034 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6035 		/* only send an error to the higher layers if it is
6036 		 * beyond the tolerance of the btrfs bio
6037 		 */
6038 		if (atomic_read(&bbio->error) > bbio->max_errors) {
6039 			bio->bi_error = -EIO;
6040 		} else {
6041 			/*
6042 			 * this bio is actually up to date, we didn't
6043 			 * go over the max number of errors
6044 			 */
6045 			bio->bi_error = 0;
6046 		}
6047 
6048 		btrfs_end_bbio(bbio, bio);
6049 	} else if (!is_orig_bio) {
6050 		bio_put(bio);
6051 	}
6052 }
6053 
6054 /*
6055  * see run_scheduled_bios for a description of why bios are collected for
6056  * async submit.
6057  *
6058  * This will add one bio to the pending list for a device and make sure
6059  * the work struct is scheduled.
6060  */
6061 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6062 					struct bio *bio)
6063 {
6064 	struct btrfs_fs_info *fs_info = device->fs_info;
6065 	int should_queue = 1;
6066 	struct btrfs_pending_bios *pending_bios;
6067 
6068 	if (device->missing || !device->bdev) {
6069 		bio_io_error(bio);
6070 		return;
6071 	}
6072 
6073 	/* don't bother with additional async steps for reads, right now */
6074 	if (bio_op(bio) == REQ_OP_READ) {
6075 		bio_get(bio);
6076 		btrfsic_submit_bio(bio);
6077 		bio_put(bio);
6078 		return;
6079 	}
6080 
6081 	/*
6082 	 * nr_async_bios allows us to reliably return congestion to the
6083 	 * higher layers.  Otherwise, the async bio makes it appear we have
6084 	 * made progress against dirty pages when we've really just put it
6085 	 * on a queue for later
6086 	 */
6087 	atomic_inc(&fs_info->nr_async_bios);
6088 	WARN_ON(bio->bi_next);
6089 	bio->bi_next = NULL;
6090 
6091 	spin_lock(&device->io_lock);
6092 	if (op_is_sync(bio->bi_opf))
6093 		pending_bios = &device->pending_sync_bios;
6094 	else
6095 		pending_bios = &device->pending_bios;
6096 
6097 	if (pending_bios->tail)
6098 		pending_bios->tail->bi_next = bio;
6099 
6100 	pending_bios->tail = bio;
6101 	if (!pending_bios->head)
6102 		pending_bios->head = bio;
6103 	if (device->running_pending)
6104 		should_queue = 0;
6105 
6106 	spin_unlock(&device->io_lock);
6107 
6108 	if (should_queue)
6109 		btrfs_queue_work(fs_info->submit_workers, &device->work);
6110 }
6111 
6112 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6113 			      u64 physical, int dev_nr, int async)
6114 {
6115 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6116 	struct btrfs_fs_info *fs_info = bbio->fs_info;
6117 
6118 	bio->bi_private = bbio;
6119 	btrfs_io_bio(bio)->stripe_index = dev_nr;
6120 	bio->bi_end_io = btrfs_end_bio;
6121 	bio->bi_iter.bi_sector = physical >> 9;
6122 #ifdef DEBUG
6123 	{
6124 		struct rcu_string *name;
6125 
6126 		rcu_read_lock();
6127 		name = rcu_dereference(dev->name);
6128 		btrfs_debug(fs_info,
6129 			"btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6130 			bio_op(bio), bio->bi_opf,
6131 			(u64)bio->bi_iter.bi_sector,
6132 			(u_long)dev->bdev->bd_dev, name->str, dev->devid,
6133 			bio->bi_iter.bi_size);
6134 		rcu_read_unlock();
6135 	}
6136 #endif
6137 	bio->bi_bdev = dev->bdev;
6138 
6139 	btrfs_bio_counter_inc_noblocked(fs_info);
6140 
6141 	if (async)
6142 		btrfs_schedule_bio(dev, bio);
6143 	else
6144 		btrfsic_submit_bio(bio);
6145 }
6146 
6147 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6148 {
6149 	atomic_inc(&bbio->error);
6150 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
6151 		/* Should be the original bio. */
6152 		WARN_ON(bio != bbio->orig_bio);
6153 
6154 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6155 		bio->bi_iter.bi_sector = logical >> 9;
6156 		bio->bi_error = -EIO;
6157 		btrfs_end_bbio(bbio, bio);
6158 	}
6159 }
6160 
6161 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6162 		  int mirror_num, int async_submit)
6163 {
6164 	struct btrfs_device *dev;
6165 	struct bio *first_bio = bio;
6166 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6167 	u64 length = 0;
6168 	u64 map_length;
6169 	int ret;
6170 	int dev_nr;
6171 	int total_devs;
6172 	struct btrfs_bio *bbio = NULL;
6173 
6174 	length = bio->bi_iter.bi_size;
6175 	map_length = length;
6176 
6177 	btrfs_bio_counter_inc_blocked(fs_info);
6178 	ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6179 				&map_length, &bbio, mirror_num, 1);
6180 	if (ret) {
6181 		btrfs_bio_counter_dec(fs_info);
6182 		return ret;
6183 	}
6184 
6185 	total_devs = bbio->num_stripes;
6186 	bbio->orig_bio = first_bio;
6187 	bbio->private = first_bio->bi_private;
6188 	bbio->end_io = first_bio->bi_end_io;
6189 	bbio->fs_info = fs_info;
6190 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6191 
6192 	if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6193 	    ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6194 		/* In this case, map_length has been set to the length of
6195 		   a single stripe; not the whole write */
6196 		if (bio_op(bio) == REQ_OP_WRITE) {
6197 			ret = raid56_parity_write(fs_info, bio, bbio,
6198 						  map_length);
6199 		} else {
6200 			ret = raid56_parity_recover(fs_info, bio, bbio,
6201 						    map_length, mirror_num, 1);
6202 		}
6203 
6204 		btrfs_bio_counter_dec(fs_info);
6205 		return ret;
6206 	}
6207 
6208 	if (map_length < length) {
6209 		btrfs_crit(fs_info,
6210 			   "mapping failed logical %llu bio len %llu len %llu",
6211 			   logical, length, map_length);
6212 		BUG();
6213 	}
6214 
6215 	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6216 		dev = bbio->stripes[dev_nr].dev;
6217 		if (!dev || !dev->bdev ||
6218 		    (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6219 			bbio_error(bbio, first_bio, logical);
6220 			continue;
6221 		}
6222 
6223 		if (dev_nr < total_devs - 1) {
6224 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6225 			BUG_ON(!bio); /* -ENOMEM */
6226 		} else
6227 			bio = first_bio;
6228 
6229 		submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6230 				  dev_nr, async_submit);
6231 	}
6232 	btrfs_bio_counter_dec(fs_info);
6233 	return 0;
6234 }
6235 
6236 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6237 				       u8 *uuid, u8 *fsid)
6238 {
6239 	struct btrfs_device *device;
6240 	struct btrfs_fs_devices *cur_devices;
6241 
6242 	cur_devices = fs_info->fs_devices;
6243 	while (cur_devices) {
6244 		if (!fsid ||
6245 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6246 			device = __find_device(&cur_devices->devices,
6247 					       devid, uuid);
6248 			if (device)
6249 				return device;
6250 		}
6251 		cur_devices = cur_devices->seed;
6252 	}
6253 	return NULL;
6254 }
6255 
6256 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6257 					    u64 devid, u8 *dev_uuid)
6258 {
6259 	struct btrfs_device *device;
6260 
6261 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6262 	if (IS_ERR(device))
6263 		return NULL;
6264 
6265 	list_add(&device->dev_list, &fs_devices->devices);
6266 	device->fs_devices = fs_devices;
6267 	fs_devices->num_devices++;
6268 
6269 	device->missing = 1;
6270 	fs_devices->missing_devices++;
6271 
6272 	return device;
6273 }
6274 
6275 /**
6276  * btrfs_alloc_device - allocate struct btrfs_device
6277  * @fs_info:	used only for generating a new devid, can be NULL if
6278  *		devid is provided (i.e. @devid != NULL).
6279  * @devid:	a pointer to devid for this device.  If NULL a new devid
6280  *		is generated.
6281  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
6282  *		is generated.
6283  *
6284  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6285  * on error.  Returned struct is not linked onto any lists and can be
6286  * destroyed with kfree() right away.
6287  */
6288 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6289 					const u64 *devid,
6290 					const u8 *uuid)
6291 {
6292 	struct btrfs_device *dev;
6293 	u64 tmp;
6294 
6295 	if (WARN_ON(!devid && !fs_info))
6296 		return ERR_PTR(-EINVAL);
6297 
6298 	dev = __alloc_device();
6299 	if (IS_ERR(dev))
6300 		return dev;
6301 
6302 	if (devid)
6303 		tmp = *devid;
6304 	else {
6305 		int ret;
6306 
6307 		ret = find_next_devid(fs_info, &tmp);
6308 		if (ret) {
6309 			kfree(dev);
6310 			return ERR_PTR(ret);
6311 		}
6312 	}
6313 	dev->devid = tmp;
6314 
6315 	if (uuid)
6316 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6317 	else
6318 		generate_random_uuid(dev->uuid);
6319 
6320 	btrfs_init_work(&dev->work, btrfs_submit_helper,
6321 			pending_bios_fn, NULL, NULL);
6322 
6323 	return dev;
6324 }
6325 
6326 /* Return -EIO if any error, otherwise return 0. */
6327 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6328 				   struct extent_buffer *leaf,
6329 				   struct btrfs_chunk *chunk, u64 logical)
6330 {
6331 	u64 length;
6332 	u64 stripe_len;
6333 	u16 num_stripes;
6334 	u16 sub_stripes;
6335 	u64 type;
6336 
6337 	length = btrfs_chunk_length(leaf, chunk);
6338 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6339 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6340 	sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6341 	type = btrfs_chunk_type(leaf, chunk);
6342 
6343 	if (!num_stripes) {
6344 		btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6345 			  num_stripes);
6346 		return -EIO;
6347 	}
6348 	if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6349 		btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6350 		return -EIO;
6351 	}
6352 	if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6353 		btrfs_err(fs_info, "invalid chunk sectorsize %u",
6354 			  btrfs_chunk_sector_size(leaf, chunk));
6355 		return -EIO;
6356 	}
6357 	if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6358 		btrfs_err(fs_info, "invalid chunk length %llu", length);
6359 		return -EIO;
6360 	}
6361 	if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6362 		btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6363 			  stripe_len);
6364 		return -EIO;
6365 	}
6366 	if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6367 	    type) {
6368 		btrfs_err(fs_info, "unrecognized chunk type: %llu",
6369 			  ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6370 			    BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6371 			  btrfs_chunk_type(leaf, chunk));
6372 		return -EIO;
6373 	}
6374 	if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6375 	    (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6376 	    (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6377 	    (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6378 	    (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6379 	    ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6380 	     num_stripes != 1)) {
6381 		btrfs_err(fs_info,
6382 			"invalid num_stripes:sub_stripes %u:%u for profile %llu",
6383 			num_stripes, sub_stripes,
6384 			type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6385 		return -EIO;
6386 	}
6387 
6388 	return 0;
6389 }
6390 
6391 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6392 			  struct extent_buffer *leaf,
6393 			  struct btrfs_chunk *chunk)
6394 {
6395 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6396 	struct map_lookup *map;
6397 	struct extent_map *em;
6398 	u64 logical;
6399 	u64 length;
6400 	u64 stripe_len;
6401 	u64 devid;
6402 	u8 uuid[BTRFS_UUID_SIZE];
6403 	int num_stripes;
6404 	int ret;
6405 	int i;
6406 
6407 	logical = key->offset;
6408 	length = btrfs_chunk_length(leaf, chunk);
6409 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6410 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6411 
6412 	ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6413 	if (ret)
6414 		return ret;
6415 
6416 	read_lock(&map_tree->map_tree.lock);
6417 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6418 	read_unlock(&map_tree->map_tree.lock);
6419 
6420 	/* already mapped? */
6421 	if (em && em->start <= logical && em->start + em->len > logical) {
6422 		free_extent_map(em);
6423 		return 0;
6424 	} else if (em) {
6425 		free_extent_map(em);
6426 	}
6427 
6428 	em = alloc_extent_map();
6429 	if (!em)
6430 		return -ENOMEM;
6431 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6432 	if (!map) {
6433 		free_extent_map(em);
6434 		return -ENOMEM;
6435 	}
6436 
6437 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6438 	em->map_lookup = map;
6439 	em->start = logical;
6440 	em->len = length;
6441 	em->orig_start = 0;
6442 	em->block_start = 0;
6443 	em->block_len = em->len;
6444 
6445 	map->num_stripes = num_stripes;
6446 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
6447 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
6448 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6449 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6450 	map->type = btrfs_chunk_type(leaf, chunk);
6451 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6452 	for (i = 0; i < num_stripes; i++) {
6453 		map->stripes[i].physical =
6454 			btrfs_stripe_offset_nr(leaf, chunk, i);
6455 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6456 		read_extent_buffer(leaf, uuid, (unsigned long)
6457 				   btrfs_stripe_dev_uuid_nr(chunk, i),
6458 				   BTRFS_UUID_SIZE);
6459 		map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6460 							uuid, NULL);
6461 		if (!map->stripes[i].dev &&
6462 		    !btrfs_test_opt(fs_info, DEGRADED)) {
6463 			free_extent_map(em);
6464 			return -EIO;
6465 		}
6466 		if (!map->stripes[i].dev) {
6467 			map->stripes[i].dev =
6468 				add_missing_dev(fs_info->fs_devices, devid,
6469 						uuid);
6470 			if (!map->stripes[i].dev) {
6471 				free_extent_map(em);
6472 				return -EIO;
6473 			}
6474 			btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6475 				   devid, uuid);
6476 		}
6477 		map->stripes[i].dev->in_fs_metadata = 1;
6478 	}
6479 
6480 	write_lock(&map_tree->map_tree.lock);
6481 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6482 	write_unlock(&map_tree->map_tree.lock);
6483 	BUG_ON(ret); /* Tree corruption */
6484 	free_extent_map(em);
6485 
6486 	return 0;
6487 }
6488 
6489 static void fill_device_from_item(struct extent_buffer *leaf,
6490 				 struct btrfs_dev_item *dev_item,
6491 				 struct btrfs_device *device)
6492 {
6493 	unsigned long ptr;
6494 
6495 	device->devid = btrfs_device_id(leaf, dev_item);
6496 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6497 	device->total_bytes = device->disk_total_bytes;
6498 	device->commit_total_bytes = device->disk_total_bytes;
6499 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6500 	device->commit_bytes_used = device->bytes_used;
6501 	device->type = btrfs_device_type(leaf, dev_item);
6502 	device->io_align = btrfs_device_io_align(leaf, dev_item);
6503 	device->io_width = btrfs_device_io_width(leaf, dev_item);
6504 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6505 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6506 	device->is_tgtdev_for_dev_replace = 0;
6507 
6508 	ptr = btrfs_device_uuid(dev_item);
6509 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6510 }
6511 
6512 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6513 						  u8 *fsid)
6514 {
6515 	struct btrfs_fs_devices *fs_devices;
6516 	int ret;
6517 
6518 	BUG_ON(!mutex_is_locked(&uuid_mutex));
6519 
6520 	fs_devices = fs_info->fs_devices->seed;
6521 	while (fs_devices) {
6522 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6523 			return fs_devices;
6524 
6525 		fs_devices = fs_devices->seed;
6526 	}
6527 
6528 	fs_devices = find_fsid(fsid);
6529 	if (!fs_devices) {
6530 		if (!btrfs_test_opt(fs_info, DEGRADED))
6531 			return ERR_PTR(-ENOENT);
6532 
6533 		fs_devices = alloc_fs_devices(fsid);
6534 		if (IS_ERR(fs_devices))
6535 			return fs_devices;
6536 
6537 		fs_devices->seeding = 1;
6538 		fs_devices->opened = 1;
6539 		return fs_devices;
6540 	}
6541 
6542 	fs_devices = clone_fs_devices(fs_devices);
6543 	if (IS_ERR(fs_devices))
6544 		return fs_devices;
6545 
6546 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6547 				   fs_info->bdev_holder);
6548 	if (ret) {
6549 		free_fs_devices(fs_devices);
6550 		fs_devices = ERR_PTR(ret);
6551 		goto out;
6552 	}
6553 
6554 	if (!fs_devices->seeding) {
6555 		__btrfs_close_devices(fs_devices);
6556 		free_fs_devices(fs_devices);
6557 		fs_devices = ERR_PTR(-EINVAL);
6558 		goto out;
6559 	}
6560 
6561 	fs_devices->seed = fs_info->fs_devices->seed;
6562 	fs_info->fs_devices->seed = fs_devices;
6563 out:
6564 	return fs_devices;
6565 }
6566 
6567 static int read_one_dev(struct btrfs_fs_info *fs_info,
6568 			struct extent_buffer *leaf,
6569 			struct btrfs_dev_item *dev_item)
6570 {
6571 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6572 	struct btrfs_device *device;
6573 	u64 devid;
6574 	int ret;
6575 	u8 fs_uuid[BTRFS_UUID_SIZE];
6576 	u8 dev_uuid[BTRFS_UUID_SIZE];
6577 
6578 	devid = btrfs_device_id(leaf, dev_item);
6579 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6580 			   BTRFS_UUID_SIZE);
6581 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6582 			   BTRFS_UUID_SIZE);
6583 
6584 	if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6585 		fs_devices = open_seed_devices(fs_info, fs_uuid);
6586 		if (IS_ERR(fs_devices))
6587 			return PTR_ERR(fs_devices);
6588 	}
6589 
6590 	device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6591 	if (!device) {
6592 		if (!btrfs_test_opt(fs_info, DEGRADED))
6593 			return -EIO;
6594 
6595 		device = add_missing_dev(fs_devices, devid, dev_uuid);
6596 		if (!device)
6597 			return -ENOMEM;
6598 		btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6599 				devid, dev_uuid);
6600 	} else {
6601 		if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6602 			return -EIO;
6603 
6604 		if(!device->bdev && !device->missing) {
6605 			/*
6606 			 * this happens when a device that was properly setup
6607 			 * in the device info lists suddenly goes bad.
6608 			 * device->bdev is NULL, and so we have to set
6609 			 * device->missing to one here
6610 			 */
6611 			device->fs_devices->missing_devices++;
6612 			device->missing = 1;
6613 		}
6614 
6615 		/* Move the device to its own fs_devices */
6616 		if (device->fs_devices != fs_devices) {
6617 			ASSERT(device->missing);
6618 
6619 			list_move(&device->dev_list, &fs_devices->devices);
6620 			device->fs_devices->num_devices--;
6621 			fs_devices->num_devices++;
6622 
6623 			device->fs_devices->missing_devices--;
6624 			fs_devices->missing_devices++;
6625 
6626 			device->fs_devices = fs_devices;
6627 		}
6628 	}
6629 
6630 	if (device->fs_devices != fs_info->fs_devices) {
6631 		BUG_ON(device->writeable);
6632 		if (device->generation !=
6633 		    btrfs_device_generation(leaf, dev_item))
6634 			return -EINVAL;
6635 	}
6636 
6637 	fill_device_from_item(leaf, dev_item, device);
6638 	device->in_fs_metadata = 1;
6639 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6640 		device->fs_devices->total_rw_bytes += device->total_bytes;
6641 		spin_lock(&fs_info->free_chunk_lock);
6642 		fs_info->free_chunk_space += device->total_bytes -
6643 			device->bytes_used;
6644 		spin_unlock(&fs_info->free_chunk_lock);
6645 	}
6646 	ret = 0;
6647 	return ret;
6648 }
6649 
6650 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6651 {
6652 	struct btrfs_root *root = fs_info->tree_root;
6653 	struct btrfs_super_block *super_copy = fs_info->super_copy;
6654 	struct extent_buffer *sb;
6655 	struct btrfs_disk_key *disk_key;
6656 	struct btrfs_chunk *chunk;
6657 	u8 *array_ptr;
6658 	unsigned long sb_array_offset;
6659 	int ret = 0;
6660 	u32 num_stripes;
6661 	u32 array_size;
6662 	u32 len = 0;
6663 	u32 cur_offset;
6664 	u64 type;
6665 	struct btrfs_key key;
6666 
6667 	ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6668 	/*
6669 	 * This will create extent buffer of nodesize, superblock size is
6670 	 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6671 	 * overallocate but we can keep it as-is, only the first page is used.
6672 	 */
6673 	sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6674 	if (IS_ERR(sb))
6675 		return PTR_ERR(sb);
6676 	set_extent_buffer_uptodate(sb);
6677 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6678 	/*
6679 	 * The sb extent buffer is artificial and just used to read the system array.
6680 	 * set_extent_buffer_uptodate() call does not properly mark all it's
6681 	 * pages up-to-date when the page is larger: extent does not cover the
6682 	 * whole page and consequently check_page_uptodate does not find all
6683 	 * the page's extents up-to-date (the hole beyond sb),
6684 	 * write_extent_buffer then triggers a WARN_ON.
6685 	 *
6686 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6687 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6688 	 * to silence the warning eg. on PowerPC 64.
6689 	 */
6690 	if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6691 		SetPageUptodate(sb->pages[0]);
6692 
6693 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6694 	array_size = btrfs_super_sys_array_size(super_copy);
6695 
6696 	array_ptr = super_copy->sys_chunk_array;
6697 	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6698 	cur_offset = 0;
6699 
6700 	while (cur_offset < array_size) {
6701 		disk_key = (struct btrfs_disk_key *)array_ptr;
6702 		len = sizeof(*disk_key);
6703 		if (cur_offset + len > array_size)
6704 			goto out_short_read;
6705 
6706 		btrfs_disk_key_to_cpu(&key, disk_key);
6707 
6708 		array_ptr += len;
6709 		sb_array_offset += len;
6710 		cur_offset += len;
6711 
6712 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6713 			chunk = (struct btrfs_chunk *)sb_array_offset;
6714 			/*
6715 			 * At least one btrfs_chunk with one stripe must be
6716 			 * present, exact stripe count check comes afterwards
6717 			 */
6718 			len = btrfs_chunk_item_size(1);
6719 			if (cur_offset + len > array_size)
6720 				goto out_short_read;
6721 
6722 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6723 			if (!num_stripes) {
6724 				btrfs_err(fs_info,
6725 					"invalid number of stripes %u in sys_array at offset %u",
6726 					num_stripes, cur_offset);
6727 				ret = -EIO;
6728 				break;
6729 			}
6730 
6731 			type = btrfs_chunk_type(sb, chunk);
6732 			if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6733 				btrfs_err(fs_info,
6734 			    "invalid chunk type %llu in sys_array at offset %u",
6735 					type, cur_offset);
6736 				ret = -EIO;
6737 				break;
6738 			}
6739 
6740 			len = btrfs_chunk_item_size(num_stripes);
6741 			if (cur_offset + len > array_size)
6742 				goto out_short_read;
6743 
6744 			ret = read_one_chunk(fs_info, &key, sb, chunk);
6745 			if (ret)
6746 				break;
6747 		} else {
6748 			btrfs_err(fs_info,
6749 			    "unexpected item type %u in sys_array at offset %u",
6750 				  (u32)key.type, cur_offset);
6751 			ret = -EIO;
6752 			break;
6753 		}
6754 		array_ptr += len;
6755 		sb_array_offset += len;
6756 		cur_offset += len;
6757 	}
6758 	clear_extent_buffer_uptodate(sb);
6759 	free_extent_buffer_stale(sb);
6760 	return ret;
6761 
6762 out_short_read:
6763 	btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6764 			len, cur_offset);
6765 	clear_extent_buffer_uptodate(sb);
6766 	free_extent_buffer_stale(sb);
6767 	return -EIO;
6768 }
6769 
6770 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6771 {
6772 	struct btrfs_root *root = fs_info->chunk_root;
6773 	struct btrfs_path *path;
6774 	struct extent_buffer *leaf;
6775 	struct btrfs_key key;
6776 	struct btrfs_key found_key;
6777 	int ret;
6778 	int slot;
6779 	u64 total_dev = 0;
6780 
6781 	path = btrfs_alloc_path();
6782 	if (!path)
6783 		return -ENOMEM;
6784 
6785 	mutex_lock(&uuid_mutex);
6786 	mutex_lock(&fs_info->chunk_mutex);
6787 
6788 	/*
6789 	 * Read all device items, and then all the chunk items. All
6790 	 * device items are found before any chunk item (their object id
6791 	 * is smaller than the lowest possible object id for a chunk
6792 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6793 	 */
6794 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6795 	key.offset = 0;
6796 	key.type = 0;
6797 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6798 	if (ret < 0)
6799 		goto error;
6800 	while (1) {
6801 		leaf = path->nodes[0];
6802 		slot = path->slots[0];
6803 		if (slot >= btrfs_header_nritems(leaf)) {
6804 			ret = btrfs_next_leaf(root, path);
6805 			if (ret == 0)
6806 				continue;
6807 			if (ret < 0)
6808 				goto error;
6809 			break;
6810 		}
6811 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
6812 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6813 			struct btrfs_dev_item *dev_item;
6814 			dev_item = btrfs_item_ptr(leaf, slot,
6815 						  struct btrfs_dev_item);
6816 			ret = read_one_dev(fs_info, leaf, dev_item);
6817 			if (ret)
6818 				goto error;
6819 			total_dev++;
6820 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6821 			struct btrfs_chunk *chunk;
6822 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6823 			ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6824 			if (ret)
6825 				goto error;
6826 		}
6827 		path->slots[0]++;
6828 	}
6829 
6830 	/*
6831 	 * After loading chunk tree, we've got all device information,
6832 	 * do another round of validation checks.
6833 	 */
6834 	if (total_dev != fs_info->fs_devices->total_devices) {
6835 		btrfs_err(fs_info,
6836 	   "super_num_devices %llu mismatch with num_devices %llu found here",
6837 			  btrfs_super_num_devices(fs_info->super_copy),
6838 			  total_dev);
6839 		ret = -EINVAL;
6840 		goto error;
6841 	}
6842 	if (btrfs_super_total_bytes(fs_info->super_copy) <
6843 	    fs_info->fs_devices->total_rw_bytes) {
6844 		btrfs_err(fs_info,
6845 	"super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6846 			  btrfs_super_total_bytes(fs_info->super_copy),
6847 			  fs_info->fs_devices->total_rw_bytes);
6848 		ret = -EINVAL;
6849 		goto error;
6850 	}
6851 	ret = 0;
6852 error:
6853 	mutex_unlock(&fs_info->chunk_mutex);
6854 	mutex_unlock(&uuid_mutex);
6855 
6856 	btrfs_free_path(path);
6857 	return ret;
6858 }
6859 
6860 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6861 {
6862 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6863 	struct btrfs_device *device;
6864 
6865 	while (fs_devices) {
6866 		mutex_lock(&fs_devices->device_list_mutex);
6867 		list_for_each_entry(device, &fs_devices->devices, dev_list)
6868 			device->fs_info = fs_info;
6869 		mutex_unlock(&fs_devices->device_list_mutex);
6870 
6871 		fs_devices = fs_devices->seed;
6872 	}
6873 }
6874 
6875 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6876 {
6877 	int i;
6878 
6879 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6880 		btrfs_dev_stat_reset(dev, i);
6881 }
6882 
6883 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6884 {
6885 	struct btrfs_key key;
6886 	struct btrfs_key found_key;
6887 	struct btrfs_root *dev_root = fs_info->dev_root;
6888 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6889 	struct extent_buffer *eb;
6890 	int slot;
6891 	int ret = 0;
6892 	struct btrfs_device *device;
6893 	struct btrfs_path *path = NULL;
6894 	int i;
6895 
6896 	path = btrfs_alloc_path();
6897 	if (!path) {
6898 		ret = -ENOMEM;
6899 		goto out;
6900 	}
6901 
6902 	mutex_lock(&fs_devices->device_list_mutex);
6903 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6904 		int item_size;
6905 		struct btrfs_dev_stats_item *ptr;
6906 
6907 		key.objectid = BTRFS_DEV_STATS_OBJECTID;
6908 		key.type = BTRFS_PERSISTENT_ITEM_KEY;
6909 		key.offset = device->devid;
6910 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6911 		if (ret) {
6912 			__btrfs_reset_dev_stats(device);
6913 			device->dev_stats_valid = 1;
6914 			btrfs_release_path(path);
6915 			continue;
6916 		}
6917 		slot = path->slots[0];
6918 		eb = path->nodes[0];
6919 		btrfs_item_key_to_cpu(eb, &found_key, slot);
6920 		item_size = btrfs_item_size_nr(eb, slot);
6921 
6922 		ptr = btrfs_item_ptr(eb, slot,
6923 				     struct btrfs_dev_stats_item);
6924 
6925 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6926 			if (item_size >= (1 + i) * sizeof(__le64))
6927 				btrfs_dev_stat_set(device, i,
6928 					btrfs_dev_stats_value(eb, ptr, i));
6929 			else
6930 				btrfs_dev_stat_reset(device, i);
6931 		}
6932 
6933 		device->dev_stats_valid = 1;
6934 		btrfs_dev_stat_print_on_load(device);
6935 		btrfs_release_path(path);
6936 	}
6937 	mutex_unlock(&fs_devices->device_list_mutex);
6938 
6939 out:
6940 	btrfs_free_path(path);
6941 	return ret < 0 ? ret : 0;
6942 }
6943 
6944 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6945 				struct btrfs_fs_info *fs_info,
6946 				struct btrfs_device *device)
6947 {
6948 	struct btrfs_root *dev_root = fs_info->dev_root;
6949 	struct btrfs_path *path;
6950 	struct btrfs_key key;
6951 	struct extent_buffer *eb;
6952 	struct btrfs_dev_stats_item *ptr;
6953 	int ret;
6954 	int i;
6955 
6956 	key.objectid = BTRFS_DEV_STATS_OBJECTID;
6957 	key.type = BTRFS_PERSISTENT_ITEM_KEY;
6958 	key.offset = device->devid;
6959 
6960 	path = btrfs_alloc_path();
6961 	BUG_ON(!path);
6962 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6963 	if (ret < 0) {
6964 		btrfs_warn_in_rcu(fs_info,
6965 			"error %d while searching for dev_stats item for device %s",
6966 			      ret, rcu_str_deref(device->name));
6967 		goto out;
6968 	}
6969 
6970 	if (ret == 0 &&
6971 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6972 		/* need to delete old one and insert a new one */
6973 		ret = btrfs_del_item(trans, dev_root, path);
6974 		if (ret != 0) {
6975 			btrfs_warn_in_rcu(fs_info,
6976 				"delete too small dev_stats item for device %s failed %d",
6977 				      rcu_str_deref(device->name), ret);
6978 			goto out;
6979 		}
6980 		ret = 1;
6981 	}
6982 
6983 	if (ret == 1) {
6984 		/* need to insert a new item */
6985 		btrfs_release_path(path);
6986 		ret = btrfs_insert_empty_item(trans, dev_root, path,
6987 					      &key, sizeof(*ptr));
6988 		if (ret < 0) {
6989 			btrfs_warn_in_rcu(fs_info,
6990 				"insert dev_stats item for device %s failed %d",
6991 				rcu_str_deref(device->name), ret);
6992 			goto out;
6993 		}
6994 	}
6995 
6996 	eb = path->nodes[0];
6997 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6998 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6999 		btrfs_set_dev_stats_value(eb, ptr, i,
7000 					  btrfs_dev_stat_read(device, i));
7001 	btrfs_mark_buffer_dirty(eb);
7002 
7003 out:
7004 	btrfs_free_path(path);
7005 	return ret;
7006 }
7007 
7008 /*
7009  * called from commit_transaction. Writes all changed device stats to disk.
7010  */
7011 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7012 			struct btrfs_fs_info *fs_info)
7013 {
7014 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7015 	struct btrfs_device *device;
7016 	int stats_cnt;
7017 	int ret = 0;
7018 
7019 	mutex_lock(&fs_devices->device_list_mutex);
7020 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
7021 		if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7022 			continue;
7023 
7024 		stats_cnt = atomic_read(&device->dev_stats_ccnt);
7025 		ret = update_dev_stat_item(trans, fs_info, device);
7026 		if (!ret)
7027 			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7028 	}
7029 	mutex_unlock(&fs_devices->device_list_mutex);
7030 
7031 	return ret;
7032 }
7033 
7034 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7035 {
7036 	btrfs_dev_stat_inc(dev, index);
7037 	btrfs_dev_stat_print_on_error(dev);
7038 }
7039 
7040 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7041 {
7042 	if (!dev->dev_stats_valid)
7043 		return;
7044 	btrfs_err_rl_in_rcu(dev->fs_info,
7045 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7046 			   rcu_str_deref(dev->name),
7047 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7048 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7049 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7050 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7051 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7052 }
7053 
7054 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7055 {
7056 	int i;
7057 
7058 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7059 		if (btrfs_dev_stat_read(dev, i) != 0)
7060 			break;
7061 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
7062 		return; /* all values == 0, suppress message */
7063 
7064 	btrfs_info_in_rcu(dev->fs_info,
7065 		"bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7066 	       rcu_str_deref(dev->name),
7067 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7068 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7069 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7070 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7071 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7072 }
7073 
7074 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7075 			struct btrfs_ioctl_get_dev_stats *stats)
7076 {
7077 	struct btrfs_device *dev;
7078 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7079 	int i;
7080 
7081 	mutex_lock(&fs_devices->device_list_mutex);
7082 	dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7083 	mutex_unlock(&fs_devices->device_list_mutex);
7084 
7085 	if (!dev) {
7086 		btrfs_warn(fs_info, "get dev_stats failed, device not found");
7087 		return -ENODEV;
7088 	} else if (!dev->dev_stats_valid) {
7089 		btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7090 		return -ENODEV;
7091 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7092 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7093 			if (stats->nr_items > i)
7094 				stats->values[i] =
7095 					btrfs_dev_stat_read_and_reset(dev, i);
7096 			else
7097 				btrfs_dev_stat_reset(dev, i);
7098 		}
7099 	} else {
7100 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7101 			if (stats->nr_items > i)
7102 				stats->values[i] = btrfs_dev_stat_read(dev, i);
7103 	}
7104 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7105 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7106 	return 0;
7107 }
7108 
7109 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7110 {
7111 	struct buffer_head *bh;
7112 	struct btrfs_super_block *disk_super;
7113 	int copy_num;
7114 
7115 	if (!bdev)
7116 		return;
7117 
7118 	for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7119 		copy_num++) {
7120 
7121 		if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7122 			continue;
7123 
7124 		disk_super = (struct btrfs_super_block *)bh->b_data;
7125 
7126 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7127 		set_buffer_dirty(bh);
7128 		sync_dirty_buffer(bh);
7129 		brelse(bh);
7130 	}
7131 
7132 	/* Notify udev that device has changed */
7133 	btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7134 
7135 	/* Update ctime/mtime for device path for libblkid */
7136 	update_dev_time(device_path);
7137 }
7138 
7139 /*
7140  * Update the size of all devices, which is used for writing out the
7141  * super blocks.
7142  */
7143 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7144 {
7145 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7146 	struct btrfs_device *curr, *next;
7147 
7148 	if (list_empty(&fs_devices->resized_devices))
7149 		return;
7150 
7151 	mutex_lock(&fs_devices->device_list_mutex);
7152 	mutex_lock(&fs_info->chunk_mutex);
7153 	list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7154 				 resized_list) {
7155 		list_del_init(&curr->resized_list);
7156 		curr->commit_total_bytes = curr->disk_total_bytes;
7157 	}
7158 	mutex_unlock(&fs_info->chunk_mutex);
7159 	mutex_unlock(&fs_devices->device_list_mutex);
7160 }
7161 
7162 /* Must be invoked during the transaction commit */
7163 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7164 					struct btrfs_transaction *transaction)
7165 {
7166 	struct extent_map *em;
7167 	struct map_lookup *map;
7168 	struct btrfs_device *dev;
7169 	int i;
7170 
7171 	if (list_empty(&transaction->pending_chunks))
7172 		return;
7173 
7174 	/* In order to kick the device replace finish process */
7175 	mutex_lock(&fs_info->chunk_mutex);
7176 	list_for_each_entry(em, &transaction->pending_chunks, list) {
7177 		map = em->map_lookup;
7178 
7179 		for (i = 0; i < map->num_stripes; i++) {
7180 			dev = map->stripes[i].dev;
7181 			dev->commit_bytes_used = dev->bytes_used;
7182 		}
7183 	}
7184 	mutex_unlock(&fs_info->chunk_mutex);
7185 }
7186 
7187 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7188 {
7189 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7190 	while (fs_devices) {
7191 		fs_devices->fs_info = fs_info;
7192 		fs_devices = fs_devices->seed;
7193 	}
7194 }
7195 
7196 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7197 {
7198 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7199 	while (fs_devices) {
7200 		fs_devices->fs_info = NULL;
7201 		fs_devices = fs_devices->seed;
7202 	}
7203 }
7204