xref: /openbmc/linux/fs/btrfs/volumes.c (revision 3d3337de)
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/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.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 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 				struct btrfs_root *root,
47 				struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
52 
53 DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
55 
56 static struct btrfs_fs_devices *__alloc_fs_devices(void)
57 {
58 	struct btrfs_fs_devices *fs_devs;
59 
60 	fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
61 	if (!fs_devs)
62 		return ERR_PTR(-ENOMEM);
63 
64 	mutex_init(&fs_devs->device_list_mutex);
65 
66 	INIT_LIST_HEAD(&fs_devs->devices);
67 	INIT_LIST_HEAD(&fs_devs->resized_devices);
68 	INIT_LIST_HEAD(&fs_devs->alloc_list);
69 	INIT_LIST_HEAD(&fs_devs->list);
70 
71 	return fs_devs;
72 }
73 
74 /**
75  * alloc_fs_devices - allocate struct btrfs_fs_devices
76  * @fsid:	a pointer to UUID for this FS.  If NULL a new UUID is
77  *		generated.
78  *
79  * Return: a pointer to a new &struct btrfs_fs_devices on success;
80  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
81  * can be destroyed with kfree() right away.
82  */
83 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
84 {
85 	struct btrfs_fs_devices *fs_devs;
86 
87 	fs_devs = __alloc_fs_devices();
88 	if (IS_ERR(fs_devs))
89 		return fs_devs;
90 
91 	if (fsid)
92 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
93 	else
94 		generate_random_uuid(fs_devs->fsid);
95 
96 	return fs_devs;
97 }
98 
99 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
100 {
101 	struct btrfs_device *device;
102 	WARN_ON(fs_devices->opened);
103 	while (!list_empty(&fs_devices->devices)) {
104 		device = list_entry(fs_devices->devices.next,
105 				    struct btrfs_device, dev_list);
106 		list_del(&device->dev_list);
107 		rcu_string_free(device->name);
108 		kfree(device);
109 	}
110 	kfree(fs_devices);
111 }
112 
113 static void btrfs_kobject_uevent(struct block_device *bdev,
114 				 enum kobject_action action)
115 {
116 	int ret;
117 
118 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
119 	if (ret)
120 		pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
121 			action,
122 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
123 			&disk_to_dev(bdev->bd_disk)->kobj);
124 }
125 
126 void btrfs_cleanup_fs_uuids(void)
127 {
128 	struct btrfs_fs_devices *fs_devices;
129 
130 	while (!list_empty(&fs_uuids)) {
131 		fs_devices = list_entry(fs_uuids.next,
132 					struct btrfs_fs_devices, list);
133 		list_del(&fs_devices->list);
134 		free_fs_devices(fs_devices);
135 	}
136 }
137 
138 static struct btrfs_device *__alloc_device(void)
139 {
140 	struct btrfs_device *dev;
141 
142 	dev = kzalloc(sizeof(*dev), GFP_NOFS);
143 	if (!dev)
144 		return ERR_PTR(-ENOMEM);
145 
146 	INIT_LIST_HEAD(&dev->dev_list);
147 	INIT_LIST_HEAD(&dev->dev_alloc_list);
148 	INIT_LIST_HEAD(&dev->resized_list);
149 
150 	spin_lock_init(&dev->io_lock);
151 
152 	spin_lock_init(&dev->reada_lock);
153 	atomic_set(&dev->reada_in_flight, 0);
154 	atomic_set(&dev->dev_stats_ccnt, 0);
155 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
156 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
157 
158 	return dev;
159 }
160 
161 static noinline struct btrfs_device *__find_device(struct list_head *head,
162 						   u64 devid, u8 *uuid)
163 {
164 	struct btrfs_device *dev;
165 
166 	list_for_each_entry(dev, head, dev_list) {
167 		if (dev->devid == devid &&
168 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
169 			return dev;
170 		}
171 	}
172 	return NULL;
173 }
174 
175 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
176 {
177 	struct btrfs_fs_devices *fs_devices;
178 
179 	list_for_each_entry(fs_devices, &fs_uuids, list) {
180 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
181 			return fs_devices;
182 	}
183 	return NULL;
184 }
185 
186 static int
187 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
188 		      int flush, struct block_device **bdev,
189 		      struct buffer_head **bh)
190 {
191 	int ret;
192 
193 	*bdev = blkdev_get_by_path(device_path, flags, holder);
194 
195 	if (IS_ERR(*bdev)) {
196 		ret = PTR_ERR(*bdev);
197 		printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
198 		goto error;
199 	}
200 
201 	if (flush)
202 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
203 	ret = set_blocksize(*bdev, 4096);
204 	if (ret) {
205 		blkdev_put(*bdev, flags);
206 		goto error;
207 	}
208 	invalidate_bdev(*bdev);
209 	*bh = btrfs_read_dev_super(*bdev);
210 	if (!*bh) {
211 		ret = -EINVAL;
212 		blkdev_put(*bdev, flags);
213 		goto error;
214 	}
215 
216 	return 0;
217 
218 error:
219 	*bdev = NULL;
220 	*bh = NULL;
221 	return ret;
222 }
223 
224 static void requeue_list(struct btrfs_pending_bios *pending_bios,
225 			struct bio *head, struct bio *tail)
226 {
227 
228 	struct bio *old_head;
229 
230 	old_head = pending_bios->head;
231 	pending_bios->head = head;
232 	if (pending_bios->tail)
233 		tail->bi_next = old_head;
234 	else
235 		pending_bios->tail = tail;
236 }
237 
238 /*
239  * we try to collect pending bios for a device so we don't get a large
240  * number of procs sending bios down to the same device.  This greatly
241  * improves the schedulers ability to collect and merge the bios.
242  *
243  * But, it also turns into a long list of bios to process and that is sure
244  * to eventually make the worker thread block.  The solution here is to
245  * make some progress and then put this work struct back at the end of
246  * the list if the block device is congested.  This way, multiple devices
247  * can make progress from a single worker thread.
248  */
249 static noinline void run_scheduled_bios(struct btrfs_device *device)
250 {
251 	struct bio *pending;
252 	struct backing_dev_info *bdi;
253 	struct btrfs_fs_info *fs_info;
254 	struct btrfs_pending_bios *pending_bios;
255 	struct bio *tail;
256 	struct bio *cur;
257 	int again = 0;
258 	unsigned long num_run;
259 	unsigned long batch_run = 0;
260 	unsigned long limit;
261 	unsigned long last_waited = 0;
262 	int force_reg = 0;
263 	int sync_pending = 0;
264 	struct blk_plug plug;
265 
266 	/*
267 	 * this function runs all the bios we've collected for
268 	 * a particular device.  We don't want to wander off to
269 	 * another device without first sending all of these down.
270 	 * So, setup a plug here and finish it off before we return
271 	 */
272 	blk_start_plug(&plug);
273 
274 	bdi = blk_get_backing_dev_info(device->bdev);
275 	fs_info = device->dev_root->fs_info;
276 	limit = btrfs_async_submit_limit(fs_info);
277 	limit = limit * 2 / 3;
278 
279 loop:
280 	spin_lock(&device->io_lock);
281 
282 loop_lock:
283 	num_run = 0;
284 
285 	/* take all the bios off the list at once and process them
286 	 * later on (without the lock held).  But, remember the
287 	 * tail and other pointers so the bios can be properly reinserted
288 	 * into the list if we hit congestion
289 	 */
290 	if (!force_reg && device->pending_sync_bios.head) {
291 		pending_bios = &device->pending_sync_bios;
292 		force_reg = 1;
293 	} else {
294 		pending_bios = &device->pending_bios;
295 		force_reg = 0;
296 	}
297 
298 	pending = pending_bios->head;
299 	tail = pending_bios->tail;
300 	WARN_ON(pending && !tail);
301 
302 	/*
303 	 * if pending was null this time around, no bios need processing
304 	 * at all and we can stop.  Otherwise it'll loop back up again
305 	 * and do an additional check so no bios are missed.
306 	 *
307 	 * device->running_pending is used to synchronize with the
308 	 * schedule_bio code.
309 	 */
310 	if (device->pending_sync_bios.head == NULL &&
311 	    device->pending_bios.head == NULL) {
312 		again = 0;
313 		device->running_pending = 0;
314 	} else {
315 		again = 1;
316 		device->running_pending = 1;
317 	}
318 
319 	pending_bios->head = NULL;
320 	pending_bios->tail = NULL;
321 
322 	spin_unlock(&device->io_lock);
323 
324 	while (pending) {
325 
326 		rmb();
327 		/* we want to work on both lists, but do more bios on the
328 		 * sync list than the regular list
329 		 */
330 		if ((num_run > 32 &&
331 		    pending_bios != &device->pending_sync_bios &&
332 		    device->pending_sync_bios.head) ||
333 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
334 		    device->pending_bios.head)) {
335 			spin_lock(&device->io_lock);
336 			requeue_list(pending_bios, pending, tail);
337 			goto loop_lock;
338 		}
339 
340 		cur = pending;
341 		pending = pending->bi_next;
342 		cur->bi_next = NULL;
343 
344 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
345 		    waitqueue_active(&fs_info->async_submit_wait))
346 			wake_up(&fs_info->async_submit_wait);
347 
348 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
349 
350 		/*
351 		 * if we're doing the sync list, record that our
352 		 * plug has some sync requests on it
353 		 *
354 		 * If we're doing the regular list and there are
355 		 * sync requests sitting around, unplug before
356 		 * we add more
357 		 */
358 		if (pending_bios == &device->pending_sync_bios) {
359 			sync_pending = 1;
360 		} else if (sync_pending) {
361 			blk_finish_plug(&plug);
362 			blk_start_plug(&plug);
363 			sync_pending = 0;
364 		}
365 
366 		btrfsic_submit_bio(cur->bi_rw, cur);
367 		num_run++;
368 		batch_run++;
369 
370 		cond_resched();
371 
372 		/*
373 		 * we made progress, there is more work to do and the bdi
374 		 * is now congested.  Back off and let other work structs
375 		 * run instead
376 		 */
377 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
378 		    fs_info->fs_devices->open_devices > 1) {
379 			struct io_context *ioc;
380 
381 			ioc = current->io_context;
382 
383 			/*
384 			 * the main goal here is that we don't want to
385 			 * block if we're going to be able to submit
386 			 * more requests without blocking.
387 			 *
388 			 * This code does two great things, it pokes into
389 			 * the elevator code from a filesystem _and_
390 			 * it makes assumptions about how batching works.
391 			 */
392 			if (ioc && ioc->nr_batch_requests > 0 &&
393 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
394 			    (last_waited == 0 ||
395 			     ioc->last_waited == last_waited)) {
396 				/*
397 				 * we want to go through our batch of
398 				 * requests and stop.  So, we copy out
399 				 * the ioc->last_waited time and test
400 				 * against it before looping
401 				 */
402 				last_waited = ioc->last_waited;
403 				cond_resched();
404 				continue;
405 			}
406 			spin_lock(&device->io_lock);
407 			requeue_list(pending_bios, pending, tail);
408 			device->running_pending = 1;
409 
410 			spin_unlock(&device->io_lock);
411 			btrfs_queue_work(fs_info->submit_workers,
412 					 &device->work);
413 			goto done;
414 		}
415 		/* unplug every 64 requests just for good measure */
416 		if (batch_run % 64 == 0) {
417 			blk_finish_plug(&plug);
418 			blk_start_plug(&plug);
419 			sync_pending = 0;
420 		}
421 	}
422 
423 	cond_resched();
424 	if (again)
425 		goto loop;
426 
427 	spin_lock(&device->io_lock);
428 	if (device->pending_bios.head || device->pending_sync_bios.head)
429 		goto loop_lock;
430 	spin_unlock(&device->io_lock);
431 
432 done:
433 	blk_finish_plug(&plug);
434 }
435 
436 static void pending_bios_fn(struct btrfs_work *work)
437 {
438 	struct btrfs_device *device;
439 
440 	device = container_of(work, struct btrfs_device, work);
441 	run_scheduled_bios(device);
442 }
443 
444 /*
445  * Add new device to list of registered devices
446  *
447  * Returns:
448  * 1   - first time device is seen
449  * 0   - device already known
450  * < 0 - error
451  */
452 static noinline int device_list_add(const char *path,
453 			   struct btrfs_super_block *disk_super,
454 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
455 {
456 	struct btrfs_device *device;
457 	struct btrfs_fs_devices *fs_devices;
458 	struct rcu_string *name;
459 	int ret = 0;
460 	u64 found_transid = btrfs_super_generation(disk_super);
461 
462 	fs_devices = find_fsid(disk_super->fsid);
463 	if (!fs_devices) {
464 		fs_devices = alloc_fs_devices(disk_super->fsid);
465 		if (IS_ERR(fs_devices))
466 			return PTR_ERR(fs_devices);
467 
468 		list_add(&fs_devices->list, &fs_uuids);
469 
470 		device = NULL;
471 	} else {
472 		device = __find_device(&fs_devices->devices, devid,
473 				       disk_super->dev_item.uuid);
474 	}
475 
476 	if (!device) {
477 		if (fs_devices->opened)
478 			return -EBUSY;
479 
480 		device = btrfs_alloc_device(NULL, &devid,
481 					    disk_super->dev_item.uuid);
482 		if (IS_ERR(device)) {
483 			/* we can safely leave the fs_devices entry around */
484 			return PTR_ERR(device);
485 		}
486 
487 		name = rcu_string_strdup(path, GFP_NOFS);
488 		if (!name) {
489 			kfree(device);
490 			return -ENOMEM;
491 		}
492 		rcu_assign_pointer(device->name, name);
493 
494 		mutex_lock(&fs_devices->device_list_mutex);
495 		list_add_rcu(&device->dev_list, &fs_devices->devices);
496 		fs_devices->num_devices++;
497 		mutex_unlock(&fs_devices->device_list_mutex);
498 
499 		ret = 1;
500 		device->fs_devices = fs_devices;
501 	} else if (!device->name || strcmp(device->name->str, path)) {
502 		/*
503 		 * When FS is already mounted.
504 		 * 1. If you are here and if the device->name is NULL that
505 		 *    means this device was missing at time of FS mount.
506 		 * 2. If you are here and if the device->name is different
507 		 *    from 'path' that means either
508 		 *      a. The same device disappeared and reappeared with
509 		 *         different name. or
510 		 *      b. The missing-disk-which-was-replaced, has
511 		 *         reappeared now.
512 		 *
513 		 * We must allow 1 and 2a above. But 2b would be a spurious
514 		 * and unintentional.
515 		 *
516 		 * Further in case of 1 and 2a above, the disk at 'path'
517 		 * would have missed some transaction when it was away and
518 		 * in case of 2a the stale bdev has to be updated as well.
519 		 * 2b must not be allowed at all time.
520 		 */
521 
522 		/*
523 		 * For now, we do allow update to btrfs_fs_device through the
524 		 * btrfs dev scan cli after FS has been mounted.  We're still
525 		 * tracking a problem where systems fail mount by subvolume id
526 		 * when we reject replacement on a mounted FS.
527 		 */
528 		if (!fs_devices->opened && found_transid < device->generation) {
529 			/*
530 			 * That is if the FS is _not_ mounted and if you
531 			 * are here, that means there is more than one
532 			 * disk with same uuid and devid.We keep the one
533 			 * with larger generation number or the last-in if
534 			 * generation are equal.
535 			 */
536 			return -EEXIST;
537 		}
538 
539 		name = rcu_string_strdup(path, GFP_NOFS);
540 		if (!name)
541 			return -ENOMEM;
542 		rcu_string_free(device->name);
543 		rcu_assign_pointer(device->name, name);
544 		if (device->missing) {
545 			fs_devices->missing_devices--;
546 			device->missing = 0;
547 		}
548 	}
549 
550 	/*
551 	 * Unmount does not free the btrfs_device struct but would zero
552 	 * generation along with most of the other members. So just update
553 	 * it back. We need it to pick the disk with largest generation
554 	 * (as above).
555 	 */
556 	if (!fs_devices->opened)
557 		device->generation = found_transid;
558 
559 	*fs_devices_ret = fs_devices;
560 
561 	return ret;
562 }
563 
564 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
565 {
566 	struct btrfs_fs_devices *fs_devices;
567 	struct btrfs_device *device;
568 	struct btrfs_device *orig_dev;
569 
570 	fs_devices = alloc_fs_devices(orig->fsid);
571 	if (IS_ERR(fs_devices))
572 		return fs_devices;
573 
574 	mutex_lock(&orig->device_list_mutex);
575 	fs_devices->total_devices = orig->total_devices;
576 
577 	/* We have held the volume lock, it is safe to get the devices. */
578 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
579 		struct rcu_string *name;
580 
581 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
582 					    orig_dev->uuid);
583 		if (IS_ERR(device))
584 			goto error;
585 
586 		/*
587 		 * This is ok to do without rcu read locked because we hold the
588 		 * uuid mutex so nothing we touch in here is going to disappear.
589 		 */
590 		if (orig_dev->name) {
591 			name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
592 			if (!name) {
593 				kfree(device);
594 				goto error;
595 			}
596 			rcu_assign_pointer(device->name, name);
597 		}
598 
599 		list_add(&device->dev_list, &fs_devices->devices);
600 		device->fs_devices = fs_devices;
601 		fs_devices->num_devices++;
602 	}
603 	mutex_unlock(&orig->device_list_mutex);
604 	return fs_devices;
605 error:
606 	mutex_unlock(&orig->device_list_mutex);
607 	free_fs_devices(fs_devices);
608 	return ERR_PTR(-ENOMEM);
609 }
610 
611 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
612 {
613 	struct btrfs_device *device, *next;
614 	struct btrfs_device *latest_dev = NULL;
615 
616 	mutex_lock(&uuid_mutex);
617 again:
618 	/* This is the initialized path, it is safe to release the devices. */
619 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
620 		if (device->in_fs_metadata) {
621 			if (!device->is_tgtdev_for_dev_replace &&
622 			    (!latest_dev ||
623 			     device->generation > latest_dev->generation)) {
624 				latest_dev = device;
625 			}
626 			continue;
627 		}
628 
629 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
630 			/*
631 			 * In the first step, keep the device which has
632 			 * the correct fsid and the devid that is used
633 			 * for the dev_replace procedure.
634 			 * In the second step, the dev_replace state is
635 			 * read from the device tree and it is known
636 			 * whether the procedure is really active or
637 			 * not, which means whether this device is
638 			 * used or whether it should be removed.
639 			 */
640 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
641 				continue;
642 			}
643 		}
644 		if (device->bdev) {
645 			blkdev_put(device->bdev, device->mode);
646 			device->bdev = NULL;
647 			fs_devices->open_devices--;
648 		}
649 		if (device->writeable) {
650 			list_del_init(&device->dev_alloc_list);
651 			device->writeable = 0;
652 			if (!device->is_tgtdev_for_dev_replace)
653 				fs_devices->rw_devices--;
654 		}
655 		list_del_init(&device->dev_list);
656 		fs_devices->num_devices--;
657 		rcu_string_free(device->name);
658 		kfree(device);
659 	}
660 
661 	if (fs_devices->seed) {
662 		fs_devices = fs_devices->seed;
663 		goto again;
664 	}
665 
666 	fs_devices->latest_bdev = latest_dev->bdev;
667 
668 	mutex_unlock(&uuid_mutex);
669 }
670 
671 static void __free_device(struct work_struct *work)
672 {
673 	struct btrfs_device *device;
674 
675 	device = container_of(work, struct btrfs_device, rcu_work);
676 
677 	if (device->bdev)
678 		blkdev_put(device->bdev, device->mode);
679 
680 	rcu_string_free(device->name);
681 	kfree(device);
682 }
683 
684 static void free_device(struct rcu_head *head)
685 {
686 	struct btrfs_device *device;
687 
688 	device = container_of(head, struct btrfs_device, rcu);
689 
690 	INIT_WORK(&device->rcu_work, __free_device);
691 	schedule_work(&device->rcu_work);
692 }
693 
694 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
695 {
696 	struct btrfs_device *device;
697 
698 	if (--fs_devices->opened > 0)
699 		return 0;
700 
701 	mutex_lock(&fs_devices->device_list_mutex);
702 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
703 		struct btrfs_device *new_device;
704 		struct rcu_string *name;
705 
706 		if (device->bdev)
707 			fs_devices->open_devices--;
708 
709 		if (device->writeable &&
710 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
711 			list_del_init(&device->dev_alloc_list);
712 			fs_devices->rw_devices--;
713 		}
714 
715 		if (device->missing)
716 			fs_devices->missing_devices--;
717 
718 		new_device = btrfs_alloc_device(NULL, &device->devid,
719 						device->uuid);
720 		BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
721 
722 		/* Safe because we are under uuid_mutex */
723 		if (device->name) {
724 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
725 			BUG_ON(!name); /* -ENOMEM */
726 			rcu_assign_pointer(new_device->name, name);
727 		}
728 
729 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
730 		new_device->fs_devices = device->fs_devices;
731 
732 		call_rcu(&device->rcu, free_device);
733 	}
734 	mutex_unlock(&fs_devices->device_list_mutex);
735 
736 	WARN_ON(fs_devices->open_devices);
737 	WARN_ON(fs_devices->rw_devices);
738 	fs_devices->opened = 0;
739 	fs_devices->seeding = 0;
740 
741 	return 0;
742 }
743 
744 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
745 {
746 	struct btrfs_fs_devices *seed_devices = NULL;
747 	int ret;
748 
749 	mutex_lock(&uuid_mutex);
750 	ret = __btrfs_close_devices(fs_devices);
751 	if (!fs_devices->opened) {
752 		seed_devices = fs_devices->seed;
753 		fs_devices->seed = NULL;
754 	}
755 	mutex_unlock(&uuid_mutex);
756 
757 	while (seed_devices) {
758 		fs_devices = seed_devices;
759 		seed_devices = fs_devices->seed;
760 		__btrfs_close_devices(fs_devices);
761 		free_fs_devices(fs_devices);
762 	}
763 	/*
764 	 * Wait for rcu kworkers under __btrfs_close_devices
765 	 * to finish all blkdev_puts so device is really
766 	 * free when umount is done.
767 	 */
768 	rcu_barrier();
769 	return ret;
770 }
771 
772 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
773 				fmode_t flags, void *holder)
774 {
775 	struct request_queue *q;
776 	struct block_device *bdev;
777 	struct list_head *head = &fs_devices->devices;
778 	struct btrfs_device *device;
779 	struct btrfs_device *latest_dev = NULL;
780 	struct buffer_head *bh;
781 	struct btrfs_super_block *disk_super;
782 	u64 devid;
783 	int seeding = 1;
784 	int ret = 0;
785 
786 	flags |= FMODE_EXCL;
787 
788 	list_for_each_entry(device, head, dev_list) {
789 		if (device->bdev)
790 			continue;
791 		if (!device->name)
792 			continue;
793 
794 		/* Just open everything we can; ignore failures here */
795 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
796 					    &bdev, &bh))
797 			continue;
798 
799 		disk_super = (struct btrfs_super_block *)bh->b_data;
800 		devid = btrfs_stack_device_id(&disk_super->dev_item);
801 		if (devid != device->devid)
802 			goto error_brelse;
803 
804 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
805 			   BTRFS_UUID_SIZE))
806 			goto error_brelse;
807 
808 		device->generation = btrfs_super_generation(disk_super);
809 		if (!latest_dev ||
810 		    device->generation > latest_dev->generation)
811 			latest_dev = device;
812 
813 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
814 			device->writeable = 0;
815 		} else {
816 			device->writeable = !bdev_read_only(bdev);
817 			seeding = 0;
818 		}
819 
820 		q = bdev_get_queue(bdev);
821 		if (blk_queue_discard(q))
822 			device->can_discard = 1;
823 
824 		device->bdev = bdev;
825 		device->in_fs_metadata = 0;
826 		device->mode = flags;
827 
828 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
829 			fs_devices->rotating = 1;
830 
831 		fs_devices->open_devices++;
832 		if (device->writeable &&
833 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
834 			fs_devices->rw_devices++;
835 			list_add(&device->dev_alloc_list,
836 				 &fs_devices->alloc_list);
837 		}
838 		brelse(bh);
839 		continue;
840 
841 error_brelse:
842 		brelse(bh);
843 		blkdev_put(bdev, flags);
844 		continue;
845 	}
846 	if (fs_devices->open_devices == 0) {
847 		ret = -EINVAL;
848 		goto out;
849 	}
850 	fs_devices->seeding = seeding;
851 	fs_devices->opened = 1;
852 	fs_devices->latest_bdev = latest_dev->bdev;
853 	fs_devices->total_rw_bytes = 0;
854 out:
855 	return ret;
856 }
857 
858 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
859 		       fmode_t flags, void *holder)
860 {
861 	int ret;
862 
863 	mutex_lock(&uuid_mutex);
864 	if (fs_devices->opened) {
865 		fs_devices->opened++;
866 		ret = 0;
867 	} else {
868 		ret = __btrfs_open_devices(fs_devices, flags, holder);
869 	}
870 	mutex_unlock(&uuid_mutex);
871 	return ret;
872 }
873 
874 /*
875  * Look for a btrfs signature on a device. This may be called out of the mount path
876  * and we are not allowed to call set_blocksize during the scan. The superblock
877  * is read via pagecache
878  */
879 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
880 			  struct btrfs_fs_devices **fs_devices_ret)
881 {
882 	struct btrfs_super_block *disk_super;
883 	struct block_device *bdev;
884 	struct page *page;
885 	void *p;
886 	int ret = -EINVAL;
887 	u64 devid;
888 	u64 transid;
889 	u64 total_devices;
890 	u64 bytenr;
891 	pgoff_t index;
892 
893 	/*
894 	 * we would like to check all the supers, but that would make
895 	 * a btrfs mount succeed after a mkfs from a different FS.
896 	 * So, we need to add a special mount option to scan for
897 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
898 	 */
899 	bytenr = btrfs_sb_offset(0);
900 	flags |= FMODE_EXCL;
901 	mutex_lock(&uuid_mutex);
902 
903 	bdev = blkdev_get_by_path(path, flags, holder);
904 
905 	if (IS_ERR(bdev)) {
906 		ret = PTR_ERR(bdev);
907 		goto error;
908 	}
909 
910 	/* make sure our super fits in the device */
911 	if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
912 		goto error_bdev_put;
913 
914 	/* make sure our super fits in the page */
915 	if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
916 		goto error_bdev_put;
917 
918 	/* make sure our super doesn't straddle pages on disk */
919 	index = bytenr >> PAGE_CACHE_SHIFT;
920 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
921 		goto error_bdev_put;
922 
923 	/* pull in the page with our super */
924 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
925 				   index, GFP_NOFS);
926 
927 	if (IS_ERR_OR_NULL(page))
928 		goto error_bdev_put;
929 
930 	p = kmap(page);
931 
932 	/* align our pointer to the offset of the super block */
933 	disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
934 
935 	if (btrfs_super_bytenr(disk_super) != bytenr ||
936 	    btrfs_super_magic(disk_super) != BTRFS_MAGIC)
937 		goto error_unmap;
938 
939 	devid = btrfs_stack_device_id(&disk_super->dev_item);
940 	transid = btrfs_super_generation(disk_super);
941 	total_devices = btrfs_super_num_devices(disk_super);
942 
943 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
944 	if (ret > 0) {
945 		if (disk_super->label[0]) {
946 			if (disk_super->label[BTRFS_LABEL_SIZE - 1])
947 				disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
948 			printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
949 		} else {
950 			printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
951 		}
952 
953 		printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
954 		ret = 0;
955 	}
956 	if (!ret && fs_devices_ret)
957 		(*fs_devices_ret)->total_devices = total_devices;
958 
959 error_unmap:
960 	kunmap(page);
961 	page_cache_release(page);
962 
963 error_bdev_put:
964 	blkdev_put(bdev, flags);
965 error:
966 	mutex_unlock(&uuid_mutex);
967 	return ret;
968 }
969 
970 /* helper to account the used device space in the range */
971 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
972 				   u64 end, u64 *length)
973 {
974 	struct btrfs_key key;
975 	struct btrfs_root *root = device->dev_root;
976 	struct btrfs_dev_extent *dev_extent;
977 	struct btrfs_path *path;
978 	u64 extent_end;
979 	int ret;
980 	int slot;
981 	struct extent_buffer *l;
982 
983 	*length = 0;
984 
985 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
986 		return 0;
987 
988 	path = btrfs_alloc_path();
989 	if (!path)
990 		return -ENOMEM;
991 	path->reada = 2;
992 
993 	key.objectid = device->devid;
994 	key.offset = start;
995 	key.type = BTRFS_DEV_EXTENT_KEY;
996 
997 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
998 	if (ret < 0)
999 		goto out;
1000 	if (ret > 0) {
1001 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1002 		if (ret < 0)
1003 			goto out;
1004 	}
1005 
1006 	while (1) {
1007 		l = path->nodes[0];
1008 		slot = path->slots[0];
1009 		if (slot >= btrfs_header_nritems(l)) {
1010 			ret = btrfs_next_leaf(root, path);
1011 			if (ret == 0)
1012 				continue;
1013 			if (ret < 0)
1014 				goto out;
1015 
1016 			break;
1017 		}
1018 		btrfs_item_key_to_cpu(l, &key, slot);
1019 
1020 		if (key.objectid < device->devid)
1021 			goto next;
1022 
1023 		if (key.objectid > device->devid)
1024 			break;
1025 
1026 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1027 			goto next;
1028 
1029 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1030 		extent_end = key.offset + btrfs_dev_extent_length(l,
1031 								  dev_extent);
1032 		if (key.offset <= start && extent_end > end) {
1033 			*length = end - start + 1;
1034 			break;
1035 		} else if (key.offset <= start && extent_end > start)
1036 			*length += extent_end - start;
1037 		else if (key.offset > start && extent_end <= end)
1038 			*length += extent_end - key.offset;
1039 		else if (key.offset > start && key.offset <= end) {
1040 			*length += end - key.offset + 1;
1041 			break;
1042 		} else if (key.offset > end)
1043 			break;
1044 
1045 next:
1046 		path->slots[0]++;
1047 	}
1048 	ret = 0;
1049 out:
1050 	btrfs_free_path(path);
1051 	return ret;
1052 }
1053 
1054 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1055 				   struct btrfs_device *device,
1056 				   u64 *start, u64 len)
1057 {
1058 	struct extent_map *em;
1059 	struct list_head *search_list = &trans->transaction->pending_chunks;
1060 	int ret = 0;
1061 	u64 physical_start = *start;
1062 
1063 again:
1064 	list_for_each_entry(em, search_list, list) {
1065 		struct map_lookup *map;
1066 		int i;
1067 
1068 		map = (struct map_lookup *)em->bdev;
1069 		for (i = 0; i < map->num_stripes; i++) {
1070 			if (map->stripes[i].dev != device)
1071 				continue;
1072 			if (map->stripes[i].physical >= physical_start + len ||
1073 			    map->stripes[i].physical + em->orig_block_len <=
1074 			    physical_start)
1075 				continue;
1076 			*start = map->stripes[i].physical +
1077 				em->orig_block_len;
1078 			ret = 1;
1079 		}
1080 	}
1081 	if (search_list == &trans->transaction->pending_chunks) {
1082 		search_list = &trans->root->fs_info->pinned_chunks;
1083 		goto again;
1084 	}
1085 
1086 	return ret;
1087 }
1088 
1089 
1090 /*
1091  * find_free_dev_extent - find free space in the specified device
1092  * @device:	the device which we search the free space in
1093  * @num_bytes:	the size of the free space that we need
1094  * @start:	store the start of the free space.
1095  * @len:	the size of the free space. that we find, or the size of the max
1096  * 		free space if we don't find suitable free space
1097  *
1098  * this uses a pretty simple search, the expectation is that it is
1099  * called very infrequently and that a given device has a small number
1100  * of extents
1101  *
1102  * @start is used to store the start of the free space if we find. But if we
1103  * don't find suitable free space, it will be used to store the start position
1104  * of the max free space.
1105  *
1106  * @len is used to store the size of the free space that we find.
1107  * But if we don't find suitable free space, it is used to store the size of
1108  * the max free space.
1109  */
1110 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1111 			 struct btrfs_device *device, u64 num_bytes,
1112 			 u64 *start, u64 *len)
1113 {
1114 	struct btrfs_key key;
1115 	struct btrfs_root *root = device->dev_root;
1116 	struct btrfs_dev_extent *dev_extent;
1117 	struct btrfs_path *path;
1118 	u64 hole_size;
1119 	u64 max_hole_start;
1120 	u64 max_hole_size;
1121 	u64 extent_end;
1122 	u64 search_start;
1123 	u64 search_end = device->total_bytes;
1124 	int ret;
1125 	int slot;
1126 	struct extent_buffer *l;
1127 
1128 	/* FIXME use last free of some kind */
1129 
1130 	/* we don't want to overwrite the superblock on the drive,
1131 	 * so we make sure to start at an offset of at least 1MB
1132 	 */
1133 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1134 
1135 	path = btrfs_alloc_path();
1136 	if (!path)
1137 		return -ENOMEM;
1138 
1139 	max_hole_start = search_start;
1140 	max_hole_size = 0;
1141 
1142 again:
1143 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1144 		ret = -ENOSPC;
1145 		goto out;
1146 	}
1147 
1148 	path->reada = 2;
1149 	path->search_commit_root = 1;
1150 	path->skip_locking = 1;
1151 
1152 	key.objectid = device->devid;
1153 	key.offset = search_start;
1154 	key.type = BTRFS_DEV_EXTENT_KEY;
1155 
1156 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1157 	if (ret < 0)
1158 		goto out;
1159 	if (ret > 0) {
1160 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1161 		if (ret < 0)
1162 			goto out;
1163 	}
1164 
1165 	while (1) {
1166 		l = path->nodes[0];
1167 		slot = path->slots[0];
1168 		if (slot >= btrfs_header_nritems(l)) {
1169 			ret = btrfs_next_leaf(root, path);
1170 			if (ret == 0)
1171 				continue;
1172 			if (ret < 0)
1173 				goto out;
1174 
1175 			break;
1176 		}
1177 		btrfs_item_key_to_cpu(l, &key, slot);
1178 
1179 		if (key.objectid < device->devid)
1180 			goto next;
1181 
1182 		if (key.objectid > device->devid)
1183 			break;
1184 
1185 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1186 			goto next;
1187 
1188 		if (key.offset > search_start) {
1189 			hole_size = key.offset - search_start;
1190 
1191 			/*
1192 			 * Have to check before we set max_hole_start, otherwise
1193 			 * we could end up sending back this offset anyway.
1194 			 */
1195 			if (contains_pending_extent(trans, device,
1196 						    &search_start,
1197 						    hole_size)) {
1198 				if (key.offset >= search_start) {
1199 					hole_size = key.offset - search_start;
1200 				} else {
1201 					WARN_ON_ONCE(1);
1202 					hole_size = 0;
1203 				}
1204 			}
1205 
1206 			if (hole_size > max_hole_size) {
1207 				max_hole_start = search_start;
1208 				max_hole_size = hole_size;
1209 			}
1210 
1211 			/*
1212 			 * If this free space is greater than which we need,
1213 			 * it must be the max free space that we have found
1214 			 * until now, so max_hole_start must point to the start
1215 			 * of this free space and the length of this free space
1216 			 * is stored in max_hole_size. Thus, we return
1217 			 * max_hole_start and max_hole_size and go back to the
1218 			 * caller.
1219 			 */
1220 			if (hole_size >= num_bytes) {
1221 				ret = 0;
1222 				goto out;
1223 			}
1224 		}
1225 
1226 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1227 		extent_end = key.offset + btrfs_dev_extent_length(l,
1228 								  dev_extent);
1229 		if (extent_end > search_start)
1230 			search_start = extent_end;
1231 next:
1232 		path->slots[0]++;
1233 		cond_resched();
1234 	}
1235 
1236 	/*
1237 	 * At this point, search_start should be the end of
1238 	 * allocated dev extents, and when shrinking the device,
1239 	 * search_end may be smaller than search_start.
1240 	 */
1241 	if (search_end > search_start) {
1242 		hole_size = search_end - search_start;
1243 
1244 		if (contains_pending_extent(trans, device, &search_start,
1245 					    hole_size)) {
1246 			btrfs_release_path(path);
1247 			goto again;
1248 		}
1249 
1250 		if (hole_size > max_hole_size) {
1251 			max_hole_start = search_start;
1252 			max_hole_size = hole_size;
1253 		}
1254 	}
1255 
1256 	/* See above. */
1257 	if (max_hole_size < num_bytes)
1258 		ret = -ENOSPC;
1259 	else
1260 		ret = 0;
1261 
1262 out:
1263 	btrfs_free_path(path);
1264 	*start = max_hole_start;
1265 	if (len)
1266 		*len = max_hole_size;
1267 	return ret;
1268 }
1269 
1270 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1271 			  struct btrfs_device *device,
1272 			  u64 start, u64 *dev_extent_len)
1273 {
1274 	int ret;
1275 	struct btrfs_path *path;
1276 	struct btrfs_root *root = device->dev_root;
1277 	struct btrfs_key key;
1278 	struct btrfs_key found_key;
1279 	struct extent_buffer *leaf = NULL;
1280 	struct btrfs_dev_extent *extent = NULL;
1281 
1282 	path = btrfs_alloc_path();
1283 	if (!path)
1284 		return -ENOMEM;
1285 
1286 	key.objectid = device->devid;
1287 	key.offset = start;
1288 	key.type = BTRFS_DEV_EXTENT_KEY;
1289 again:
1290 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1291 	if (ret > 0) {
1292 		ret = btrfs_previous_item(root, path, key.objectid,
1293 					  BTRFS_DEV_EXTENT_KEY);
1294 		if (ret)
1295 			goto out;
1296 		leaf = path->nodes[0];
1297 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1298 		extent = btrfs_item_ptr(leaf, path->slots[0],
1299 					struct btrfs_dev_extent);
1300 		BUG_ON(found_key.offset > start || found_key.offset +
1301 		       btrfs_dev_extent_length(leaf, extent) < start);
1302 		key = found_key;
1303 		btrfs_release_path(path);
1304 		goto again;
1305 	} else if (ret == 0) {
1306 		leaf = path->nodes[0];
1307 		extent = btrfs_item_ptr(leaf, path->slots[0],
1308 					struct btrfs_dev_extent);
1309 	} else {
1310 		btrfs_error(root->fs_info, ret, "Slot search failed");
1311 		goto out;
1312 	}
1313 
1314 	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1315 
1316 	ret = btrfs_del_item(trans, root, path);
1317 	if (ret) {
1318 		btrfs_error(root->fs_info, ret,
1319 			    "Failed to remove dev extent item");
1320 	} else {
1321 		trans->transaction->have_free_bgs = 1;
1322 	}
1323 out:
1324 	btrfs_free_path(path);
1325 	return ret;
1326 }
1327 
1328 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1329 				  struct btrfs_device *device,
1330 				  u64 chunk_tree, u64 chunk_objectid,
1331 				  u64 chunk_offset, u64 start, u64 num_bytes)
1332 {
1333 	int ret;
1334 	struct btrfs_path *path;
1335 	struct btrfs_root *root = device->dev_root;
1336 	struct btrfs_dev_extent *extent;
1337 	struct extent_buffer *leaf;
1338 	struct btrfs_key key;
1339 
1340 	WARN_ON(!device->in_fs_metadata);
1341 	WARN_ON(device->is_tgtdev_for_dev_replace);
1342 	path = btrfs_alloc_path();
1343 	if (!path)
1344 		return -ENOMEM;
1345 
1346 	key.objectid = device->devid;
1347 	key.offset = start;
1348 	key.type = BTRFS_DEV_EXTENT_KEY;
1349 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1350 				      sizeof(*extent));
1351 	if (ret)
1352 		goto out;
1353 
1354 	leaf = path->nodes[0];
1355 	extent = btrfs_item_ptr(leaf, path->slots[0],
1356 				struct btrfs_dev_extent);
1357 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1358 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1359 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1360 
1361 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1362 		    btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1363 
1364 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1365 	btrfs_mark_buffer_dirty(leaf);
1366 out:
1367 	btrfs_free_path(path);
1368 	return ret;
1369 }
1370 
1371 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1372 {
1373 	struct extent_map_tree *em_tree;
1374 	struct extent_map *em;
1375 	struct rb_node *n;
1376 	u64 ret = 0;
1377 
1378 	em_tree = &fs_info->mapping_tree.map_tree;
1379 	read_lock(&em_tree->lock);
1380 	n = rb_last(&em_tree->map);
1381 	if (n) {
1382 		em = rb_entry(n, struct extent_map, rb_node);
1383 		ret = em->start + em->len;
1384 	}
1385 	read_unlock(&em_tree->lock);
1386 
1387 	return ret;
1388 }
1389 
1390 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1391 				    u64 *devid_ret)
1392 {
1393 	int ret;
1394 	struct btrfs_key key;
1395 	struct btrfs_key found_key;
1396 	struct btrfs_path *path;
1397 
1398 	path = btrfs_alloc_path();
1399 	if (!path)
1400 		return -ENOMEM;
1401 
1402 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1403 	key.type = BTRFS_DEV_ITEM_KEY;
1404 	key.offset = (u64)-1;
1405 
1406 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1407 	if (ret < 0)
1408 		goto error;
1409 
1410 	BUG_ON(ret == 0); /* Corruption */
1411 
1412 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1413 				  BTRFS_DEV_ITEMS_OBJECTID,
1414 				  BTRFS_DEV_ITEM_KEY);
1415 	if (ret) {
1416 		*devid_ret = 1;
1417 	} else {
1418 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1419 				      path->slots[0]);
1420 		*devid_ret = found_key.offset + 1;
1421 	}
1422 	ret = 0;
1423 error:
1424 	btrfs_free_path(path);
1425 	return ret;
1426 }
1427 
1428 /*
1429  * the device information is stored in the chunk root
1430  * the btrfs_device struct should be fully filled in
1431  */
1432 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1433 			    struct btrfs_root *root,
1434 			    struct btrfs_device *device)
1435 {
1436 	int ret;
1437 	struct btrfs_path *path;
1438 	struct btrfs_dev_item *dev_item;
1439 	struct extent_buffer *leaf;
1440 	struct btrfs_key key;
1441 	unsigned long ptr;
1442 
1443 	root = root->fs_info->chunk_root;
1444 
1445 	path = btrfs_alloc_path();
1446 	if (!path)
1447 		return -ENOMEM;
1448 
1449 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1450 	key.type = BTRFS_DEV_ITEM_KEY;
1451 	key.offset = device->devid;
1452 
1453 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1454 				      sizeof(*dev_item));
1455 	if (ret)
1456 		goto out;
1457 
1458 	leaf = path->nodes[0];
1459 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1460 
1461 	btrfs_set_device_id(leaf, dev_item, device->devid);
1462 	btrfs_set_device_generation(leaf, dev_item, 0);
1463 	btrfs_set_device_type(leaf, dev_item, device->type);
1464 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1465 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1466 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1467 	btrfs_set_device_total_bytes(leaf, dev_item,
1468 				     btrfs_device_get_disk_total_bytes(device));
1469 	btrfs_set_device_bytes_used(leaf, dev_item,
1470 				    btrfs_device_get_bytes_used(device));
1471 	btrfs_set_device_group(leaf, dev_item, 0);
1472 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1473 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1474 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1475 
1476 	ptr = btrfs_device_uuid(dev_item);
1477 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1478 	ptr = btrfs_device_fsid(dev_item);
1479 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1480 	btrfs_mark_buffer_dirty(leaf);
1481 
1482 	ret = 0;
1483 out:
1484 	btrfs_free_path(path);
1485 	return ret;
1486 }
1487 
1488 /*
1489  * Function to update ctime/mtime for a given device path.
1490  * Mainly used for ctime/mtime based probe like libblkid.
1491  */
1492 static void update_dev_time(char *path_name)
1493 {
1494 	struct file *filp;
1495 
1496 	filp = filp_open(path_name, O_RDWR, 0);
1497 	if (IS_ERR(filp))
1498 		return;
1499 	file_update_time(filp);
1500 	filp_close(filp, NULL);
1501 	return;
1502 }
1503 
1504 static int btrfs_rm_dev_item(struct btrfs_root *root,
1505 			     struct btrfs_device *device)
1506 {
1507 	int ret;
1508 	struct btrfs_path *path;
1509 	struct btrfs_key key;
1510 	struct btrfs_trans_handle *trans;
1511 
1512 	root = root->fs_info->chunk_root;
1513 
1514 	path = btrfs_alloc_path();
1515 	if (!path)
1516 		return -ENOMEM;
1517 
1518 	trans = btrfs_start_transaction(root, 0);
1519 	if (IS_ERR(trans)) {
1520 		btrfs_free_path(path);
1521 		return PTR_ERR(trans);
1522 	}
1523 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1524 	key.type = BTRFS_DEV_ITEM_KEY;
1525 	key.offset = device->devid;
1526 
1527 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1528 	if (ret < 0)
1529 		goto out;
1530 
1531 	if (ret > 0) {
1532 		ret = -ENOENT;
1533 		goto out;
1534 	}
1535 
1536 	ret = btrfs_del_item(trans, root, path);
1537 	if (ret)
1538 		goto out;
1539 out:
1540 	btrfs_free_path(path);
1541 	btrfs_commit_transaction(trans, root);
1542 	return ret;
1543 }
1544 
1545 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1546 {
1547 	struct btrfs_device *device;
1548 	struct btrfs_device *next_device;
1549 	struct block_device *bdev;
1550 	struct buffer_head *bh = NULL;
1551 	struct btrfs_super_block *disk_super;
1552 	struct btrfs_fs_devices *cur_devices;
1553 	u64 all_avail;
1554 	u64 devid;
1555 	u64 num_devices;
1556 	u8 *dev_uuid;
1557 	unsigned seq;
1558 	int ret = 0;
1559 	bool clear_super = false;
1560 
1561 	mutex_lock(&uuid_mutex);
1562 
1563 	do {
1564 		seq = read_seqbegin(&root->fs_info->profiles_lock);
1565 
1566 		all_avail = root->fs_info->avail_data_alloc_bits |
1567 			    root->fs_info->avail_system_alloc_bits |
1568 			    root->fs_info->avail_metadata_alloc_bits;
1569 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
1570 
1571 	num_devices = root->fs_info->fs_devices->num_devices;
1572 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1573 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1574 		WARN_ON(num_devices < 1);
1575 		num_devices--;
1576 	}
1577 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1578 
1579 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1580 		ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1581 		goto out;
1582 	}
1583 
1584 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1585 		ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1586 		goto out;
1587 	}
1588 
1589 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1590 	    root->fs_info->fs_devices->rw_devices <= 2) {
1591 		ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1592 		goto out;
1593 	}
1594 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1595 	    root->fs_info->fs_devices->rw_devices <= 3) {
1596 		ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1597 		goto out;
1598 	}
1599 
1600 	if (strcmp(device_path, "missing") == 0) {
1601 		struct list_head *devices;
1602 		struct btrfs_device *tmp;
1603 
1604 		device = NULL;
1605 		devices = &root->fs_info->fs_devices->devices;
1606 		/*
1607 		 * It is safe to read the devices since the volume_mutex
1608 		 * is held.
1609 		 */
1610 		list_for_each_entry(tmp, devices, dev_list) {
1611 			if (tmp->in_fs_metadata &&
1612 			    !tmp->is_tgtdev_for_dev_replace &&
1613 			    !tmp->bdev) {
1614 				device = tmp;
1615 				break;
1616 			}
1617 		}
1618 		bdev = NULL;
1619 		bh = NULL;
1620 		disk_super = NULL;
1621 		if (!device) {
1622 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1623 			goto out;
1624 		}
1625 	} else {
1626 		ret = btrfs_get_bdev_and_sb(device_path,
1627 					    FMODE_WRITE | FMODE_EXCL,
1628 					    root->fs_info->bdev_holder, 0,
1629 					    &bdev, &bh);
1630 		if (ret)
1631 			goto out;
1632 		disk_super = (struct btrfs_super_block *)bh->b_data;
1633 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1634 		dev_uuid = disk_super->dev_item.uuid;
1635 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1636 					   disk_super->fsid);
1637 		if (!device) {
1638 			ret = -ENOENT;
1639 			goto error_brelse;
1640 		}
1641 	}
1642 
1643 	if (device->is_tgtdev_for_dev_replace) {
1644 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1645 		goto error_brelse;
1646 	}
1647 
1648 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1649 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1650 		goto error_brelse;
1651 	}
1652 
1653 	if (device->writeable) {
1654 		lock_chunks(root);
1655 		list_del_init(&device->dev_alloc_list);
1656 		device->fs_devices->rw_devices--;
1657 		unlock_chunks(root);
1658 		clear_super = true;
1659 	}
1660 
1661 	mutex_unlock(&uuid_mutex);
1662 	ret = btrfs_shrink_device(device, 0);
1663 	mutex_lock(&uuid_mutex);
1664 	if (ret)
1665 		goto error_undo;
1666 
1667 	/*
1668 	 * TODO: the superblock still includes this device in its num_devices
1669 	 * counter although write_all_supers() is not locked out. This
1670 	 * could give a filesystem state which requires a degraded mount.
1671 	 */
1672 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1673 	if (ret)
1674 		goto error_undo;
1675 
1676 	device->in_fs_metadata = 0;
1677 	btrfs_scrub_cancel_dev(root->fs_info, device);
1678 
1679 	/*
1680 	 * the device list mutex makes sure that we don't change
1681 	 * the device list while someone else is writing out all
1682 	 * the device supers. Whoever is writing all supers, should
1683 	 * lock the device list mutex before getting the number of
1684 	 * devices in the super block (super_copy). Conversely,
1685 	 * whoever updates the number of devices in the super block
1686 	 * (super_copy) should hold the device list mutex.
1687 	 */
1688 
1689 	cur_devices = device->fs_devices;
1690 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1691 	list_del_rcu(&device->dev_list);
1692 
1693 	device->fs_devices->num_devices--;
1694 	device->fs_devices->total_devices--;
1695 
1696 	if (device->missing)
1697 		device->fs_devices->missing_devices--;
1698 
1699 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1700 				 struct btrfs_device, dev_list);
1701 	if (device->bdev == root->fs_info->sb->s_bdev)
1702 		root->fs_info->sb->s_bdev = next_device->bdev;
1703 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1704 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1705 
1706 	if (device->bdev) {
1707 		device->fs_devices->open_devices--;
1708 		/* remove sysfs entry */
1709 		btrfs_kobj_rm_device(root->fs_info, device);
1710 	}
1711 
1712 	call_rcu(&device->rcu, free_device);
1713 
1714 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1715 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1716 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1717 
1718 	if (cur_devices->open_devices == 0) {
1719 		struct btrfs_fs_devices *fs_devices;
1720 		fs_devices = root->fs_info->fs_devices;
1721 		while (fs_devices) {
1722 			if (fs_devices->seed == cur_devices) {
1723 				fs_devices->seed = cur_devices->seed;
1724 				break;
1725 			}
1726 			fs_devices = fs_devices->seed;
1727 		}
1728 		cur_devices->seed = NULL;
1729 		__btrfs_close_devices(cur_devices);
1730 		free_fs_devices(cur_devices);
1731 	}
1732 
1733 	root->fs_info->num_tolerated_disk_barrier_failures =
1734 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1735 
1736 	/*
1737 	 * at this point, the device is zero sized.  We want to
1738 	 * remove it from the devices list and zero out the old super
1739 	 */
1740 	if (clear_super && disk_super) {
1741 		u64 bytenr;
1742 		int i;
1743 
1744 		/* make sure this device isn't detected as part of
1745 		 * the FS anymore
1746 		 */
1747 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1748 		set_buffer_dirty(bh);
1749 		sync_dirty_buffer(bh);
1750 
1751 		/* clear the mirror copies of super block on the disk
1752 		 * being removed, 0th copy is been taken care above and
1753 		 * the below would take of the rest
1754 		 */
1755 		for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1756 			bytenr = btrfs_sb_offset(i);
1757 			if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1758 					i_size_read(bdev->bd_inode))
1759 				break;
1760 
1761 			brelse(bh);
1762 			bh = __bread(bdev, bytenr / 4096,
1763 					BTRFS_SUPER_INFO_SIZE);
1764 			if (!bh)
1765 				continue;
1766 
1767 			disk_super = (struct btrfs_super_block *)bh->b_data;
1768 
1769 			if (btrfs_super_bytenr(disk_super) != bytenr ||
1770 				btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1771 				continue;
1772 			}
1773 			memset(&disk_super->magic, 0,
1774 						sizeof(disk_super->magic));
1775 			set_buffer_dirty(bh);
1776 			sync_dirty_buffer(bh);
1777 		}
1778 	}
1779 
1780 	ret = 0;
1781 
1782 	if (bdev) {
1783 		/* Notify udev that device has changed */
1784 		btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1785 
1786 		/* Update ctime/mtime for device path for libblkid */
1787 		update_dev_time(device_path);
1788 	}
1789 
1790 error_brelse:
1791 	brelse(bh);
1792 	if (bdev)
1793 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1794 out:
1795 	mutex_unlock(&uuid_mutex);
1796 	return ret;
1797 error_undo:
1798 	if (device->writeable) {
1799 		lock_chunks(root);
1800 		list_add(&device->dev_alloc_list,
1801 			 &root->fs_info->fs_devices->alloc_list);
1802 		device->fs_devices->rw_devices++;
1803 		unlock_chunks(root);
1804 	}
1805 	goto error_brelse;
1806 }
1807 
1808 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1809 					struct btrfs_device *srcdev)
1810 {
1811 	struct btrfs_fs_devices *fs_devices;
1812 
1813 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1814 
1815 	/*
1816 	 * in case of fs with no seed, srcdev->fs_devices will point
1817 	 * to fs_devices of fs_info. However when the dev being replaced is
1818 	 * a seed dev it will point to the seed's local fs_devices. In short
1819 	 * srcdev will have its correct fs_devices in both the cases.
1820 	 */
1821 	fs_devices = srcdev->fs_devices;
1822 
1823 	list_del_rcu(&srcdev->dev_list);
1824 	list_del_rcu(&srcdev->dev_alloc_list);
1825 	fs_devices->num_devices--;
1826 	if (srcdev->missing)
1827 		fs_devices->missing_devices--;
1828 
1829 	if (srcdev->writeable) {
1830 		fs_devices->rw_devices--;
1831 		/* zero out the old super if it is writable */
1832 		btrfs_scratch_superblock(srcdev);
1833 	}
1834 
1835 	if (srcdev->bdev)
1836 		fs_devices->open_devices--;
1837 }
1838 
1839 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1840 				      struct btrfs_device *srcdev)
1841 {
1842 	struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1843 
1844 	call_rcu(&srcdev->rcu, free_device);
1845 
1846 	/*
1847 	 * unless fs_devices is seed fs, num_devices shouldn't go
1848 	 * zero
1849 	 */
1850 	BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1851 
1852 	/* if this is no devs we rather delete the fs_devices */
1853 	if (!fs_devices->num_devices) {
1854 		struct btrfs_fs_devices *tmp_fs_devices;
1855 
1856 		tmp_fs_devices = fs_info->fs_devices;
1857 		while (tmp_fs_devices) {
1858 			if (tmp_fs_devices->seed == fs_devices) {
1859 				tmp_fs_devices->seed = fs_devices->seed;
1860 				break;
1861 			}
1862 			tmp_fs_devices = tmp_fs_devices->seed;
1863 		}
1864 		fs_devices->seed = NULL;
1865 		__btrfs_close_devices(fs_devices);
1866 		free_fs_devices(fs_devices);
1867 	}
1868 }
1869 
1870 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1871 				      struct btrfs_device *tgtdev)
1872 {
1873 	struct btrfs_device *next_device;
1874 
1875 	mutex_lock(&uuid_mutex);
1876 	WARN_ON(!tgtdev);
1877 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1878 	if (tgtdev->bdev) {
1879 		btrfs_scratch_superblock(tgtdev);
1880 		fs_info->fs_devices->open_devices--;
1881 	}
1882 	fs_info->fs_devices->num_devices--;
1883 
1884 	next_device = list_entry(fs_info->fs_devices->devices.next,
1885 				 struct btrfs_device, dev_list);
1886 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1887 		fs_info->sb->s_bdev = next_device->bdev;
1888 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1889 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1890 	list_del_rcu(&tgtdev->dev_list);
1891 
1892 	call_rcu(&tgtdev->rcu, free_device);
1893 
1894 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1895 	mutex_unlock(&uuid_mutex);
1896 }
1897 
1898 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1899 				     struct btrfs_device **device)
1900 {
1901 	int ret = 0;
1902 	struct btrfs_super_block *disk_super;
1903 	u64 devid;
1904 	u8 *dev_uuid;
1905 	struct block_device *bdev;
1906 	struct buffer_head *bh;
1907 
1908 	*device = NULL;
1909 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1910 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1911 	if (ret)
1912 		return ret;
1913 	disk_super = (struct btrfs_super_block *)bh->b_data;
1914 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1915 	dev_uuid = disk_super->dev_item.uuid;
1916 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1917 				    disk_super->fsid);
1918 	brelse(bh);
1919 	if (!*device)
1920 		ret = -ENOENT;
1921 	blkdev_put(bdev, FMODE_READ);
1922 	return ret;
1923 }
1924 
1925 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1926 					 char *device_path,
1927 					 struct btrfs_device **device)
1928 {
1929 	*device = NULL;
1930 	if (strcmp(device_path, "missing") == 0) {
1931 		struct list_head *devices;
1932 		struct btrfs_device *tmp;
1933 
1934 		devices = &root->fs_info->fs_devices->devices;
1935 		/*
1936 		 * It is safe to read the devices since the volume_mutex
1937 		 * is held by the caller.
1938 		 */
1939 		list_for_each_entry(tmp, devices, dev_list) {
1940 			if (tmp->in_fs_metadata && !tmp->bdev) {
1941 				*device = tmp;
1942 				break;
1943 			}
1944 		}
1945 
1946 		if (!*device) {
1947 			btrfs_err(root->fs_info, "no missing device found");
1948 			return -ENOENT;
1949 		}
1950 
1951 		return 0;
1952 	} else {
1953 		return btrfs_find_device_by_path(root, device_path, device);
1954 	}
1955 }
1956 
1957 /*
1958  * does all the dirty work required for changing file system's UUID.
1959  */
1960 static int btrfs_prepare_sprout(struct btrfs_root *root)
1961 {
1962 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1963 	struct btrfs_fs_devices *old_devices;
1964 	struct btrfs_fs_devices *seed_devices;
1965 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1966 	struct btrfs_device *device;
1967 	u64 super_flags;
1968 
1969 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1970 	if (!fs_devices->seeding)
1971 		return -EINVAL;
1972 
1973 	seed_devices = __alloc_fs_devices();
1974 	if (IS_ERR(seed_devices))
1975 		return PTR_ERR(seed_devices);
1976 
1977 	old_devices = clone_fs_devices(fs_devices);
1978 	if (IS_ERR(old_devices)) {
1979 		kfree(seed_devices);
1980 		return PTR_ERR(old_devices);
1981 	}
1982 
1983 	list_add(&old_devices->list, &fs_uuids);
1984 
1985 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1986 	seed_devices->opened = 1;
1987 	INIT_LIST_HEAD(&seed_devices->devices);
1988 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1989 	mutex_init(&seed_devices->device_list_mutex);
1990 
1991 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1992 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1993 			      synchronize_rcu);
1994 	list_for_each_entry(device, &seed_devices->devices, dev_list)
1995 		device->fs_devices = seed_devices;
1996 
1997 	lock_chunks(root);
1998 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1999 	unlock_chunks(root);
2000 
2001 	fs_devices->seeding = 0;
2002 	fs_devices->num_devices = 0;
2003 	fs_devices->open_devices = 0;
2004 	fs_devices->missing_devices = 0;
2005 	fs_devices->rotating = 0;
2006 	fs_devices->seed = seed_devices;
2007 
2008 	generate_random_uuid(fs_devices->fsid);
2009 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2010 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2011 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2012 
2013 	super_flags = btrfs_super_flags(disk_super) &
2014 		      ~BTRFS_SUPER_FLAG_SEEDING;
2015 	btrfs_set_super_flags(disk_super, super_flags);
2016 
2017 	return 0;
2018 }
2019 
2020 /*
2021  * strore the expected generation for seed devices in device items.
2022  */
2023 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2024 			       struct btrfs_root *root)
2025 {
2026 	struct btrfs_path *path;
2027 	struct extent_buffer *leaf;
2028 	struct btrfs_dev_item *dev_item;
2029 	struct btrfs_device *device;
2030 	struct btrfs_key key;
2031 	u8 fs_uuid[BTRFS_UUID_SIZE];
2032 	u8 dev_uuid[BTRFS_UUID_SIZE];
2033 	u64 devid;
2034 	int ret;
2035 
2036 	path = btrfs_alloc_path();
2037 	if (!path)
2038 		return -ENOMEM;
2039 
2040 	root = root->fs_info->chunk_root;
2041 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2042 	key.offset = 0;
2043 	key.type = BTRFS_DEV_ITEM_KEY;
2044 
2045 	while (1) {
2046 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2047 		if (ret < 0)
2048 			goto error;
2049 
2050 		leaf = path->nodes[0];
2051 next_slot:
2052 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2053 			ret = btrfs_next_leaf(root, path);
2054 			if (ret > 0)
2055 				break;
2056 			if (ret < 0)
2057 				goto error;
2058 			leaf = path->nodes[0];
2059 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2060 			btrfs_release_path(path);
2061 			continue;
2062 		}
2063 
2064 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2065 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2066 		    key.type != BTRFS_DEV_ITEM_KEY)
2067 			break;
2068 
2069 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2070 					  struct btrfs_dev_item);
2071 		devid = btrfs_device_id(leaf, dev_item);
2072 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2073 				   BTRFS_UUID_SIZE);
2074 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2075 				   BTRFS_UUID_SIZE);
2076 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2077 					   fs_uuid);
2078 		BUG_ON(!device); /* Logic error */
2079 
2080 		if (device->fs_devices->seeding) {
2081 			btrfs_set_device_generation(leaf, dev_item,
2082 						    device->generation);
2083 			btrfs_mark_buffer_dirty(leaf);
2084 		}
2085 
2086 		path->slots[0]++;
2087 		goto next_slot;
2088 	}
2089 	ret = 0;
2090 error:
2091 	btrfs_free_path(path);
2092 	return ret;
2093 }
2094 
2095 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2096 {
2097 	struct request_queue *q;
2098 	struct btrfs_trans_handle *trans;
2099 	struct btrfs_device *device;
2100 	struct block_device *bdev;
2101 	struct list_head *devices;
2102 	struct super_block *sb = root->fs_info->sb;
2103 	struct rcu_string *name;
2104 	u64 tmp;
2105 	int seeding_dev = 0;
2106 	int ret = 0;
2107 
2108 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2109 		return -EROFS;
2110 
2111 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2112 				  root->fs_info->bdev_holder);
2113 	if (IS_ERR(bdev))
2114 		return PTR_ERR(bdev);
2115 
2116 	if (root->fs_info->fs_devices->seeding) {
2117 		seeding_dev = 1;
2118 		down_write(&sb->s_umount);
2119 		mutex_lock(&uuid_mutex);
2120 	}
2121 
2122 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2123 
2124 	devices = &root->fs_info->fs_devices->devices;
2125 
2126 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2127 	list_for_each_entry(device, devices, dev_list) {
2128 		if (device->bdev == bdev) {
2129 			ret = -EEXIST;
2130 			mutex_unlock(
2131 				&root->fs_info->fs_devices->device_list_mutex);
2132 			goto error;
2133 		}
2134 	}
2135 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2136 
2137 	device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2138 	if (IS_ERR(device)) {
2139 		/* we can safely leave the fs_devices entry around */
2140 		ret = PTR_ERR(device);
2141 		goto error;
2142 	}
2143 
2144 	name = rcu_string_strdup(device_path, GFP_NOFS);
2145 	if (!name) {
2146 		kfree(device);
2147 		ret = -ENOMEM;
2148 		goto error;
2149 	}
2150 	rcu_assign_pointer(device->name, name);
2151 
2152 	trans = btrfs_start_transaction(root, 0);
2153 	if (IS_ERR(trans)) {
2154 		rcu_string_free(device->name);
2155 		kfree(device);
2156 		ret = PTR_ERR(trans);
2157 		goto error;
2158 	}
2159 
2160 	q = bdev_get_queue(bdev);
2161 	if (blk_queue_discard(q))
2162 		device->can_discard = 1;
2163 	device->writeable = 1;
2164 	device->generation = trans->transid;
2165 	device->io_width = root->sectorsize;
2166 	device->io_align = root->sectorsize;
2167 	device->sector_size = root->sectorsize;
2168 	device->total_bytes = i_size_read(bdev->bd_inode);
2169 	device->disk_total_bytes = device->total_bytes;
2170 	device->commit_total_bytes = device->total_bytes;
2171 	device->dev_root = root->fs_info->dev_root;
2172 	device->bdev = bdev;
2173 	device->in_fs_metadata = 1;
2174 	device->is_tgtdev_for_dev_replace = 0;
2175 	device->mode = FMODE_EXCL;
2176 	device->dev_stats_valid = 1;
2177 	set_blocksize(device->bdev, 4096);
2178 
2179 	if (seeding_dev) {
2180 		sb->s_flags &= ~MS_RDONLY;
2181 		ret = btrfs_prepare_sprout(root);
2182 		BUG_ON(ret); /* -ENOMEM */
2183 	}
2184 
2185 	device->fs_devices = root->fs_info->fs_devices;
2186 
2187 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2188 	lock_chunks(root);
2189 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2190 	list_add(&device->dev_alloc_list,
2191 		 &root->fs_info->fs_devices->alloc_list);
2192 	root->fs_info->fs_devices->num_devices++;
2193 	root->fs_info->fs_devices->open_devices++;
2194 	root->fs_info->fs_devices->rw_devices++;
2195 	root->fs_info->fs_devices->total_devices++;
2196 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2197 
2198 	spin_lock(&root->fs_info->free_chunk_lock);
2199 	root->fs_info->free_chunk_space += device->total_bytes;
2200 	spin_unlock(&root->fs_info->free_chunk_lock);
2201 
2202 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2203 		root->fs_info->fs_devices->rotating = 1;
2204 
2205 	tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2206 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
2207 				    tmp + device->total_bytes);
2208 
2209 	tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2210 	btrfs_set_super_num_devices(root->fs_info->super_copy,
2211 				    tmp + 1);
2212 
2213 	/* add sysfs device entry */
2214 	btrfs_kobj_add_device(root->fs_info, device);
2215 
2216 	/*
2217 	 * we've got more storage, clear any full flags on the space
2218 	 * infos
2219 	 */
2220 	btrfs_clear_space_info_full(root->fs_info);
2221 
2222 	unlock_chunks(root);
2223 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2224 
2225 	if (seeding_dev) {
2226 		lock_chunks(root);
2227 		ret = init_first_rw_device(trans, root, device);
2228 		unlock_chunks(root);
2229 		if (ret) {
2230 			btrfs_abort_transaction(trans, root, ret);
2231 			goto error_trans;
2232 		}
2233 	}
2234 
2235 	ret = btrfs_add_device(trans, root, device);
2236 	if (ret) {
2237 		btrfs_abort_transaction(trans, root, ret);
2238 		goto error_trans;
2239 	}
2240 
2241 	if (seeding_dev) {
2242 		char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2243 
2244 		ret = btrfs_finish_sprout(trans, root);
2245 		if (ret) {
2246 			btrfs_abort_transaction(trans, root, ret);
2247 			goto error_trans;
2248 		}
2249 
2250 		/* Sprouting would change fsid of the mounted root,
2251 		 * so rename the fsid on the sysfs
2252 		 */
2253 		snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2254 						root->fs_info->fsid);
2255 		if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2256 			goto error_trans;
2257 	}
2258 
2259 	root->fs_info->num_tolerated_disk_barrier_failures =
2260 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2261 	ret = btrfs_commit_transaction(trans, root);
2262 
2263 	if (seeding_dev) {
2264 		mutex_unlock(&uuid_mutex);
2265 		up_write(&sb->s_umount);
2266 
2267 		if (ret) /* transaction commit */
2268 			return ret;
2269 
2270 		ret = btrfs_relocate_sys_chunks(root);
2271 		if (ret < 0)
2272 			btrfs_error(root->fs_info, ret,
2273 				    "Failed to relocate sys chunks after "
2274 				    "device initialization. This can be fixed "
2275 				    "using the \"btrfs balance\" command.");
2276 		trans = btrfs_attach_transaction(root);
2277 		if (IS_ERR(trans)) {
2278 			if (PTR_ERR(trans) == -ENOENT)
2279 				return 0;
2280 			return PTR_ERR(trans);
2281 		}
2282 		ret = btrfs_commit_transaction(trans, root);
2283 	}
2284 
2285 	/* Update ctime/mtime for libblkid */
2286 	update_dev_time(device_path);
2287 	return ret;
2288 
2289 error_trans:
2290 	btrfs_end_transaction(trans, root);
2291 	rcu_string_free(device->name);
2292 	btrfs_kobj_rm_device(root->fs_info, device);
2293 	kfree(device);
2294 error:
2295 	blkdev_put(bdev, FMODE_EXCL);
2296 	if (seeding_dev) {
2297 		mutex_unlock(&uuid_mutex);
2298 		up_write(&sb->s_umount);
2299 	}
2300 	return ret;
2301 }
2302 
2303 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2304 				  struct btrfs_device *srcdev,
2305 				  struct btrfs_device **device_out)
2306 {
2307 	struct request_queue *q;
2308 	struct btrfs_device *device;
2309 	struct block_device *bdev;
2310 	struct btrfs_fs_info *fs_info = root->fs_info;
2311 	struct list_head *devices;
2312 	struct rcu_string *name;
2313 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
2314 	int ret = 0;
2315 
2316 	*device_out = NULL;
2317 	if (fs_info->fs_devices->seeding) {
2318 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2319 		return -EINVAL;
2320 	}
2321 
2322 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2323 				  fs_info->bdev_holder);
2324 	if (IS_ERR(bdev)) {
2325 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
2326 		return PTR_ERR(bdev);
2327 	}
2328 
2329 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2330 
2331 	devices = &fs_info->fs_devices->devices;
2332 	list_for_each_entry(device, devices, dev_list) {
2333 		if (device->bdev == bdev) {
2334 			btrfs_err(fs_info, "target device is in the filesystem!");
2335 			ret = -EEXIST;
2336 			goto error;
2337 		}
2338 	}
2339 
2340 
2341 	if (i_size_read(bdev->bd_inode) <
2342 	    btrfs_device_get_total_bytes(srcdev)) {
2343 		btrfs_err(fs_info, "target device is smaller than source device!");
2344 		ret = -EINVAL;
2345 		goto error;
2346 	}
2347 
2348 
2349 	device = btrfs_alloc_device(NULL, &devid, NULL);
2350 	if (IS_ERR(device)) {
2351 		ret = PTR_ERR(device);
2352 		goto error;
2353 	}
2354 
2355 	name = rcu_string_strdup(device_path, GFP_NOFS);
2356 	if (!name) {
2357 		kfree(device);
2358 		ret = -ENOMEM;
2359 		goto error;
2360 	}
2361 	rcu_assign_pointer(device->name, name);
2362 
2363 	q = bdev_get_queue(bdev);
2364 	if (blk_queue_discard(q))
2365 		device->can_discard = 1;
2366 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2367 	device->writeable = 1;
2368 	device->generation = 0;
2369 	device->io_width = root->sectorsize;
2370 	device->io_align = root->sectorsize;
2371 	device->sector_size = root->sectorsize;
2372 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2373 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2374 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2375 	ASSERT(list_empty(&srcdev->resized_list));
2376 	device->commit_total_bytes = srcdev->commit_total_bytes;
2377 	device->commit_bytes_used = device->bytes_used;
2378 	device->dev_root = fs_info->dev_root;
2379 	device->bdev = bdev;
2380 	device->in_fs_metadata = 1;
2381 	device->is_tgtdev_for_dev_replace = 1;
2382 	device->mode = FMODE_EXCL;
2383 	device->dev_stats_valid = 1;
2384 	set_blocksize(device->bdev, 4096);
2385 	device->fs_devices = fs_info->fs_devices;
2386 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2387 	fs_info->fs_devices->num_devices++;
2388 	fs_info->fs_devices->open_devices++;
2389 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2390 
2391 	*device_out = device;
2392 	return ret;
2393 
2394 error:
2395 	blkdev_put(bdev, FMODE_EXCL);
2396 	return ret;
2397 }
2398 
2399 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2400 					      struct btrfs_device *tgtdev)
2401 {
2402 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2403 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2404 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2405 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2406 	tgtdev->dev_root = fs_info->dev_root;
2407 	tgtdev->in_fs_metadata = 1;
2408 }
2409 
2410 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2411 					struct btrfs_device *device)
2412 {
2413 	int ret;
2414 	struct btrfs_path *path;
2415 	struct btrfs_root *root;
2416 	struct btrfs_dev_item *dev_item;
2417 	struct extent_buffer *leaf;
2418 	struct btrfs_key key;
2419 
2420 	root = device->dev_root->fs_info->chunk_root;
2421 
2422 	path = btrfs_alloc_path();
2423 	if (!path)
2424 		return -ENOMEM;
2425 
2426 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2427 	key.type = BTRFS_DEV_ITEM_KEY;
2428 	key.offset = device->devid;
2429 
2430 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2431 	if (ret < 0)
2432 		goto out;
2433 
2434 	if (ret > 0) {
2435 		ret = -ENOENT;
2436 		goto out;
2437 	}
2438 
2439 	leaf = path->nodes[0];
2440 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2441 
2442 	btrfs_set_device_id(leaf, dev_item, device->devid);
2443 	btrfs_set_device_type(leaf, dev_item, device->type);
2444 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2445 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2446 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2447 	btrfs_set_device_total_bytes(leaf, dev_item,
2448 				     btrfs_device_get_disk_total_bytes(device));
2449 	btrfs_set_device_bytes_used(leaf, dev_item,
2450 				    btrfs_device_get_bytes_used(device));
2451 	btrfs_mark_buffer_dirty(leaf);
2452 
2453 out:
2454 	btrfs_free_path(path);
2455 	return ret;
2456 }
2457 
2458 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2459 		      struct btrfs_device *device, u64 new_size)
2460 {
2461 	struct btrfs_super_block *super_copy =
2462 		device->dev_root->fs_info->super_copy;
2463 	struct btrfs_fs_devices *fs_devices;
2464 	u64 old_total;
2465 	u64 diff;
2466 
2467 	if (!device->writeable)
2468 		return -EACCES;
2469 
2470 	lock_chunks(device->dev_root);
2471 	old_total = btrfs_super_total_bytes(super_copy);
2472 	diff = new_size - device->total_bytes;
2473 
2474 	if (new_size <= device->total_bytes ||
2475 	    device->is_tgtdev_for_dev_replace) {
2476 		unlock_chunks(device->dev_root);
2477 		return -EINVAL;
2478 	}
2479 
2480 	fs_devices = device->dev_root->fs_info->fs_devices;
2481 
2482 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2483 	device->fs_devices->total_rw_bytes += diff;
2484 
2485 	btrfs_device_set_total_bytes(device, new_size);
2486 	btrfs_device_set_disk_total_bytes(device, new_size);
2487 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2488 	if (list_empty(&device->resized_list))
2489 		list_add_tail(&device->resized_list,
2490 			      &fs_devices->resized_devices);
2491 	unlock_chunks(device->dev_root);
2492 
2493 	return btrfs_update_device(trans, device);
2494 }
2495 
2496 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2497 			    struct btrfs_root *root, u64 chunk_objectid,
2498 			    u64 chunk_offset)
2499 {
2500 	int ret;
2501 	struct btrfs_path *path;
2502 	struct btrfs_key key;
2503 
2504 	root = root->fs_info->chunk_root;
2505 	path = btrfs_alloc_path();
2506 	if (!path)
2507 		return -ENOMEM;
2508 
2509 	key.objectid = chunk_objectid;
2510 	key.offset = chunk_offset;
2511 	key.type = BTRFS_CHUNK_ITEM_KEY;
2512 
2513 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2514 	if (ret < 0)
2515 		goto out;
2516 	else if (ret > 0) { /* Logic error or corruption */
2517 		btrfs_error(root->fs_info, -ENOENT,
2518 			    "Failed lookup while freeing chunk.");
2519 		ret = -ENOENT;
2520 		goto out;
2521 	}
2522 
2523 	ret = btrfs_del_item(trans, root, path);
2524 	if (ret < 0)
2525 		btrfs_error(root->fs_info, ret,
2526 			    "Failed to delete chunk item.");
2527 out:
2528 	btrfs_free_path(path);
2529 	return ret;
2530 }
2531 
2532 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2533 			chunk_offset)
2534 {
2535 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2536 	struct btrfs_disk_key *disk_key;
2537 	struct btrfs_chunk *chunk;
2538 	u8 *ptr;
2539 	int ret = 0;
2540 	u32 num_stripes;
2541 	u32 array_size;
2542 	u32 len = 0;
2543 	u32 cur;
2544 	struct btrfs_key key;
2545 
2546 	lock_chunks(root);
2547 	array_size = btrfs_super_sys_array_size(super_copy);
2548 
2549 	ptr = super_copy->sys_chunk_array;
2550 	cur = 0;
2551 
2552 	while (cur < array_size) {
2553 		disk_key = (struct btrfs_disk_key *)ptr;
2554 		btrfs_disk_key_to_cpu(&key, disk_key);
2555 
2556 		len = sizeof(*disk_key);
2557 
2558 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2559 			chunk = (struct btrfs_chunk *)(ptr + len);
2560 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2561 			len += btrfs_chunk_item_size(num_stripes);
2562 		} else {
2563 			ret = -EIO;
2564 			break;
2565 		}
2566 		if (key.objectid == chunk_objectid &&
2567 		    key.offset == chunk_offset) {
2568 			memmove(ptr, ptr + len, array_size - (cur + len));
2569 			array_size -= len;
2570 			btrfs_set_super_sys_array_size(super_copy, array_size);
2571 		} else {
2572 			ptr += len;
2573 			cur += len;
2574 		}
2575 	}
2576 	unlock_chunks(root);
2577 	return ret;
2578 }
2579 
2580 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2581 		       struct btrfs_root *root, u64 chunk_offset)
2582 {
2583 	struct extent_map_tree *em_tree;
2584 	struct extent_map *em;
2585 	struct btrfs_root *extent_root = root->fs_info->extent_root;
2586 	struct map_lookup *map;
2587 	u64 dev_extent_len = 0;
2588 	u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2589 	int i, ret = 0;
2590 
2591 	/* Just in case */
2592 	root = root->fs_info->chunk_root;
2593 	em_tree = &root->fs_info->mapping_tree.map_tree;
2594 
2595 	read_lock(&em_tree->lock);
2596 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2597 	read_unlock(&em_tree->lock);
2598 
2599 	if (!em || em->start > chunk_offset ||
2600 	    em->start + em->len < chunk_offset) {
2601 		/*
2602 		 * This is a logic error, but we don't want to just rely on the
2603 		 * user having built with ASSERT enabled, so if ASSERT doens't
2604 		 * do anything we still error out.
2605 		 */
2606 		ASSERT(0);
2607 		if (em)
2608 			free_extent_map(em);
2609 		return -EINVAL;
2610 	}
2611 	map = (struct map_lookup *)em->bdev;
2612 
2613 	for (i = 0; i < map->num_stripes; i++) {
2614 		struct btrfs_device *device = map->stripes[i].dev;
2615 		ret = btrfs_free_dev_extent(trans, device,
2616 					    map->stripes[i].physical,
2617 					    &dev_extent_len);
2618 		if (ret) {
2619 			btrfs_abort_transaction(trans, root, ret);
2620 			goto out;
2621 		}
2622 
2623 		if (device->bytes_used > 0) {
2624 			lock_chunks(root);
2625 			btrfs_device_set_bytes_used(device,
2626 					device->bytes_used - dev_extent_len);
2627 			spin_lock(&root->fs_info->free_chunk_lock);
2628 			root->fs_info->free_chunk_space += dev_extent_len;
2629 			spin_unlock(&root->fs_info->free_chunk_lock);
2630 			btrfs_clear_space_info_full(root->fs_info);
2631 			unlock_chunks(root);
2632 		}
2633 
2634 		if (map->stripes[i].dev) {
2635 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2636 			if (ret) {
2637 				btrfs_abort_transaction(trans, root, ret);
2638 				goto out;
2639 			}
2640 		}
2641 	}
2642 	ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2643 	if (ret) {
2644 		btrfs_abort_transaction(trans, root, ret);
2645 		goto out;
2646 	}
2647 
2648 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2649 
2650 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2651 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2652 		if (ret) {
2653 			btrfs_abort_transaction(trans, root, ret);
2654 			goto out;
2655 		}
2656 	}
2657 
2658 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2659 	if (ret) {
2660 		btrfs_abort_transaction(trans, extent_root, ret);
2661 		goto out;
2662 	}
2663 
2664 out:
2665 	/* once for us */
2666 	free_extent_map(em);
2667 	return ret;
2668 }
2669 
2670 static int btrfs_relocate_chunk(struct btrfs_root *root,
2671 				u64 chunk_objectid,
2672 				u64 chunk_offset)
2673 {
2674 	struct btrfs_root *extent_root;
2675 	struct btrfs_trans_handle *trans;
2676 	int ret;
2677 
2678 	root = root->fs_info->chunk_root;
2679 	extent_root = root->fs_info->extent_root;
2680 
2681 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2682 	if (ret)
2683 		return -ENOSPC;
2684 
2685 	/* step one, relocate all the extents inside this chunk */
2686 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2687 	if (ret)
2688 		return ret;
2689 
2690 	trans = btrfs_start_transaction(root, 0);
2691 	if (IS_ERR(trans)) {
2692 		ret = PTR_ERR(trans);
2693 		btrfs_std_error(root->fs_info, ret);
2694 		return ret;
2695 	}
2696 
2697 	/*
2698 	 * step two, delete the device extents and the
2699 	 * chunk tree entries
2700 	 */
2701 	ret = btrfs_remove_chunk(trans, root, chunk_offset);
2702 	btrfs_end_transaction(trans, root);
2703 	return ret;
2704 }
2705 
2706 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2707 {
2708 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2709 	struct btrfs_path *path;
2710 	struct extent_buffer *leaf;
2711 	struct btrfs_chunk *chunk;
2712 	struct btrfs_key key;
2713 	struct btrfs_key found_key;
2714 	u64 chunk_type;
2715 	bool retried = false;
2716 	int failed = 0;
2717 	int ret;
2718 
2719 	path = btrfs_alloc_path();
2720 	if (!path)
2721 		return -ENOMEM;
2722 
2723 again:
2724 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2725 	key.offset = (u64)-1;
2726 	key.type = BTRFS_CHUNK_ITEM_KEY;
2727 
2728 	while (1) {
2729 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2730 		if (ret < 0)
2731 			goto error;
2732 		BUG_ON(ret == 0); /* Corruption */
2733 
2734 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2735 					  key.type);
2736 		if (ret < 0)
2737 			goto error;
2738 		if (ret > 0)
2739 			break;
2740 
2741 		leaf = path->nodes[0];
2742 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2743 
2744 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2745 				       struct btrfs_chunk);
2746 		chunk_type = btrfs_chunk_type(leaf, chunk);
2747 		btrfs_release_path(path);
2748 
2749 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2750 			ret = btrfs_relocate_chunk(chunk_root,
2751 						   found_key.objectid,
2752 						   found_key.offset);
2753 			if (ret == -ENOSPC)
2754 				failed++;
2755 			else
2756 				BUG_ON(ret);
2757 		}
2758 
2759 		if (found_key.offset == 0)
2760 			break;
2761 		key.offset = found_key.offset - 1;
2762 	}
2763 	ret = 0;
2764 	if (failed && !retried) {
2765 		failed = 0;
2766 		retried = true;
2767 		goto again;
2768 	} else if (WARN_ON(failed && retried)) {
2769 		ret = -ENOSPC;
2770 	}
2771 error:
2772 	btrfs_free_path(path);
2773 	return ret;
2774 }
2775 
2776 static int insert_balance_item(struct btrfs_root *root,
2777 			       struct btrfs_balance_control *bctl)
2778 {
2779 	struct btrfs_trans_handle *trans;
2780 	struct btrfs_balance_item *item;
2781 	struct btrfs_disk_balance_args disk_bargs;
2782 	struct btrfs_path *path;
2783 	struct extent_buffer *leaf;
2784 	struct btrfs_key key;
2785 	int ret, err;
2786 
2787 	path = btrfs_alloc_path();
2788 	if (!path)
2789 		return -ENOMEM;
2790 
2791 	trans = btrfs_start_transaction(root, 0);
2792 	if (IS_ERR(trans)) {
2793 		btrfs_free_path(path);
2794 		return PTR_ERR(trans);
2795 	}
2796 
2797 	key.objectid = BTRFS_BALANCE_OBJECTID;
2798 	key.type = BTRFS_BALANCE_ITEM_KEY;
2799 	key.offset = 0;
2800 
2801 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2802 				      sizeof(*item));
2803 	if (ret)
2804 		goto out;
2805 
2806 	leaf = path->nodes[0];
2807 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2808 
2809 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2810 
2811 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2812 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2813 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2814 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2815 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2816 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2817 
2818 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2819 
2820 	btrfs_mark_buffer_dirty(leaf);
2821 out:
2822 	btrfs_free_path(path);
2823 	err = btrfs_commit_transaction(trans, root);
2824 	if (err && !ret)
2825 		ret = err;
2826 	return ret;
2827 }
2828 
2829 static int del_balance_item(struct btrfs_root *root)
2830 {
2831 	struct btrfs_trans_handle *trans;
2832 	struct btrfs_path *path;
2833 	struct btrfs_key key;
2834 	int ret, err;
2835 
2836 	path = btrfs_alloc_path();
2837 	if (!path)
2838 		return -ENOMEM;
2839 
2840 	trans = btrfs_start_transaction(root, 0);
2841 	if (IS_ERR(trans)) {
2842 		btrfs_free_path(path);
2843 		return PTR_ERR(trans);
2844 	}
2845 
2846 	key.objectid = BTRFS_BALANCE_OBJECTID;
2847 	key.type = BTRFS_BALANCE_ITEM_KEY;
2848 	key.offset = 0;
2849 
2850 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2851 	if (ret < 0)
2852 		goto out;
2853 	if (ret > 0) {
2854 		ret = -ENOENT;
2855 		goto out;
2856 	}
2857 
2858 	ret = btrfs_del_item(trans, root, path);
2859 out:
2860 	btrfs_free_path(path);
2861 	err = btrfs_commit_transaction(trans, root);
2862 	if (err && !ret)
2863 		ret = err;
2864 	return ret;
2865 }
2866 
2867 /*
2868  * This is a heuristic used to reduce the number of chunks balanced on
2869  * resume after balance was interrupted.
2870  */
2871 static void update_balance_args(struct btrfs_balance_control *bctl)
2872 {
2873 	/*
2874 	 * Turn on soft mode for chunk types that were being converted.
2875 	 */
2876 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2877 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2878 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2879 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2880 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2881 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2882 
2883 	/*
2884 	 * Turn on usage filter if is not already used.  The idea is
2885 	 * that chunks that we have already balanced should be
2886 	 * reasonably full.  Don't do it for chunks that are being
2887 	 * converted - that will keep us from relocating unconverted
2888 	 * (albeit full) chunks.
2889 	 */
2890 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2891 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2892 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2893 		bctl->data.usage = 90;
2894 	}
2895 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2896 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2897 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2898 		bctl->sys.usage = 90;
2899 	}
2900 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2901 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2902 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2903 		bctl->meta.usage = 90;
2904 	}
2905 }
2906 
2907 /*
2908  * Should be called with both balance and volume mutexes held to
2909  * serialize other volume operations (add_dev/rm_dev/resize) with
2910  * restriper.  Same goes for unset_balance_control.
2911  */
2912 static void set_balance_control(struct btrfs_balance_control *bctl)
2913 {
2914 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2915 
2916 	BUG_ON(fs_info->balance_ctl);
2917 
2918 	spin_lock(&fs_info->balance_lock);
2919 	fs_info->balance_ctl = bctl;
2920 	spin_unlock(&fs_info->balance_lock);
2921 }
2922 
2923 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2924 {
2925 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2926 
2927 	BUG_ON(!fs_info->balance_ctl);
2928 
2929 	spin_lock(&fs_info->balance_lock);
2930 	fs_info->balance_ctl = NULL;
2931 	spin_unlock(&fs_info->balance_lock);
2932 
2933 	kfree(bctl);
2934 }
2935 
2936 /*
2937  * Balance filters.  Return 1 if chunk should be filtered out
2938  * (should not be balanced).
2939  */
2940 static int chunk_profiles_filter(u64 chunk_type,
2941 				 struct btrfs_balance_args *bargs)
2942 {
2943 	chunk_type = chunk_to_extended(chunk_type) &
2944 				BTRFS_EXTENDED_PROFILE_MASK;
2945 
2946 	if (bargs->profiles & chunk_type)
2947 		return 0;
2948 
2949 	return 1;
2950 }
2951 
2952 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2953 			      struct btrfs_balance_args *bargs)
2954 {
2955 	struct btrfs_block_group_cache *cache;
2956 	u64 chunk_used, user_thresh;
2957 	int ret = 1;
2958 
2959 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2960 	chunk_used = btrfs_block_group_used(&cache->item);
2961 
2962 	if (bargs->usage == 0)
2963 		user_thresh = 1;
2964 	else if (bargs->usage > 100)
2965 		user_thresh = cache->key.offset;
2966 	else
2967 		user_thresh = div_factor_fine(cache->key.offset,
2968 					      bargs->usage);
2969 
2970 	if (chunk_used < user_thresh)
2971 		ret = 0;
2972 
2973 	btrfs_put_block_group(cache);
2974 	return ret;
2975 }
2976 
2977 static int chunk_devid_filter(struct extent_buffer *leaf,
2978 			      struct btrfs_chunk *chunk,
2979 			      struct btrfs_balance_args *bargs)
2980 {
2981 	struct btrfs_stripe *stripe;
2982 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2983 	int i;
2984 
2985 	for (i = 0; i < num_stripes; i++) {
2986 		stripe = btrfs_stripe_nr(chunk, i);
2987 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2988 			return 0;
2989 	}
2990 
2991 	return 1;
2992 }
2993 
2994 /* [pstart, pend) */
2995 static int chunk_drange_filter(struct extent_buffer *leaf,
2996 			       struct btrfs_chunk *chunk,
2997 			       u64 chunk_offset,
2998 			       struct btrfs_balance_args *bargs)
2999 {
3000 	struct btrfs_stripe *stripe;
3001 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3002 	u64 stripe_offset;
3003 	u64 stripe_length;
3004 	int factor;
3005 	int i;
3006 
3007 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3008 		return 0;
3009 
3010 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3011 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3012 		factor = num_stripes / 2;
3013 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3014 		factor = num_stripes - 1;
3015 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3016 		factor = num_stripes - 2;
3017 	} else {
3018 		factor = num_stripes;
3019 	}
3020 
3021 	for (i = 0; i < num_stripes; i++) {
3022 		stripe = btrfs_stripe_nr(chunk, i);
3023 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3024 			continue;
3025 
3026 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3027 		stripe_length = btrfs_chunk_length(leaf, chunk);
3028 		stripe_length = div_u64(stripe_length, factor);
3029 
3030 		if (stripe_offset < bargs->pend &&
3031 		    stripe_offset + stripe_length > bargs->pstart)
3032 			return 0;
3033 	}
3034 
3035 	return 1;
3036 }
3037 
3038 /* [vstart, vend) */
3039 static int chunk_vrange_filter(struct extent_buffer *leaf,
3040 			       struct btrfs_chunk *chunk,
3041 			       u64 chunk_offset,
3042 			       struct btrfs_balance_args *bargs)
3043 {
3044 	if (chunk_offset < bargs->vend &&
3045 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3046 		/* at least part of the chunk is inside this vrange */
3047 		return 0;
3048 
3049 	return 1;
3050 }
3051 
3052 static int chunk_soft_convert_filter(u64 chunk_type,
3053 				     struct btrfs_balance_args *bargs)
3054 {
3055 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3056 		return 0;
3057 
3058 	chunk_type = chunk_to_extended(chunk_type) &
3059 				BTRFS_EXTENDED_PROFILE_MASK;
3060 
3061 	if (bargs->target == chunk_type)
3062 		return 1;
3063 
3064 	return 0;
3065 }
3066 
3067 static int should_balance_chunk(struct btrfs_root *root,
3068 				struct extent_buffer *leaf,
3069 				struct btrfs_chunk *chunk, u64 chunk_offset)
3070 {
3071 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3072 	struct btrfs_balance_args *bargs = NULL;
3073 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3074 
3075 	/* type filter */
3076 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3077 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3078 		return 0;
3079 	}
3080 
3081 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3082 		bargs = &bctl->data;
3083 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3084 		bargs = &bctl->sys;
3085 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3086 		bargs = &bctl->meta;
3087 
3088 	/* profiles filter */
3089 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3090 	    chunk_profiles_filter(chunk_type, bargs)) {
3091 		return 0;
3092 	}
3093 
3094 	/* usage filter */
3095 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3096 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3097 		return 0;
3098 	}
3099 
3100 	/* devid filter */
3101 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3102 	    chunk_devid_filter(leaf, chunk, bargs)) {
3103 		return 0;
3104 	}
3105 
3106 	/* drange filter, makes sense only with devid filter */
3107 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3108 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3109 		return 0;
3110 	}
3111 
3112 	/* vrange filter */
3113 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3114 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3115 		return 0;
3116 	}
3117 
3118 	/* soft profile changing mode */
3119 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3120 	    chunk_soft_convert_filter(chunk_type, bargs)) {
3121 		return 0;
3122 	}
3123 
3124 	/*
3125 	 * limited by count, must be the last filter
3126 	 */
3127 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3128 		if (bargs->limit == 0)
3129 			return 0;
3130 		else
3131 			bargs->limit--;
3132 	}
3133 
3134 	return 1;
3135 }
3136 
3137 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3138 {
3139 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3140 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3141 	struct btrfs_root *dev_root = fs_info->dev_root;
3142 	struct list_head *devices;
3143 	struct btrfs_device *device;
3144 	u64 old_size;
3145 	u64 size_to_free;
3146 	struct btrfs_chunk *chunk;
3147 	struct btrfs_path *path;
3148 	struct btrfs_key key;
3149 	struct btrfs_key found_key;
3150 	struct btrfs_trans_handle *trans;
3151 	struct extent_buffer *leaf;
3152 	int slot;
3153 	int ret;
3154 	int enospc_errors = 0;
3155 	bool counting = true;
3156 	u64 limit_data = bctl->data.limit;
3157 	u64 limit_meta = bctl->meta.limit;
3158 	u64 limit_sys = bctl->sys.limit;
3159 
3160 	/* step one make some room on all the devices */
3161 	devices = &fs_info->fs_devices->devices;
3162 	list_for_each_entry(device, devices, dev_list) {
3163 		old_size = btrfs_device_get_total_bytes(device);
3164 		size_to_free = div_factor(old_size, 1);
3165 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3166 		if (!device->writeable ||
3167 		    btrfs_device_get_total_bytes(device) -
3168 		    btrfs_device_get_bytes_used(device) > size_to_free ||
3169 		    device->is_tgtdev_for_dev_replace)
3170 			continue;
3171 
3172 		ret = btrfs_shrink_device(device, old_size - size_to_free);
3173 		if (ret == -ENOSPC)
3174 			break;
3175 		BUG_ON(ret);
3176 
3177 		trans = btrfs_start_transaction(dev_root, 0);
3178 		BUG_ON(IS_ERR(trans));
3179 
3180 		ret = btrfs_grow_device(trans, device, old_size);
3181 		BUG_ON(ret);
3182 
3183 		btrfs_end_transaction(trans, dev_root);
3184 	}
3185 
3186 	/* step two, relocate all the chunks */
3187 	path = btrfs_alloc_path();
3188 	if (!path) {
3189 		ret = -ENOMEM;
3190 		goto error;
3191 	}
3192 
3193 	/* zero out stat counters */
3194 	spin_lock(&fs_info->balance_lock);
3195 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3196 	spin_unlock(&fs_info->balance_lock);
3197 again:
3198 	if (!counting) {
3199 		bctl->data.limit = limit_data;
3200 		bctl->meta.limit = limit_meta;
3201 		bctl->sys.limit = limit_sys;
3202 	}
3203 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3204 	key.offset = (u64)-1;
3205 	key.type = BTRFS_CHUNK_ITEM_KEY;
3206 
3207 	while (1) {
3208 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3209 		    atomic_read(&fs_info->balance_cancel_req)) {
3210 			ret = -ECANCELED;
3211 			goto error;
3212 		}
3213 
3214 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3215 		if (ret < 0)
3216 			goto error;
3217 
3218 		/*
3219 		 * this shouldn't happen, it means the last relocate
3220 		 * failed
3221 		 */
3222 		if (ret == 0)
3223 			BUG(); /* FIXME break ? */
3224 
3225 		ret = btrfs_previous_item(chunk_root, path, 0,
3226 					  BTRFS_CHUNK_ITEM_KEY);
3227 		if (ret) {
3228 			ret = 0;
3229 			break;
3230 		}
3231 
3232 		leaf = path->nodes[0];
3233 		slot = path->slots[0];
3234 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3235 
3236 		if (found_key.objectid != key.objectid)
3237 			break;
3238 
3239 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3240 
3241 		if (!counting) {
3242 			spin_lock(&fs_info->balance_lock);
3243 			bctl->stat.considered++;
3244 			spin_unlock(&fs_info->balance_lock);
3245 		}
3246 
3247 		ret = should_balance_chunk(chunk_root, leaf, chunk,
3248 					   found_key.offset);
3249 		btrfs_release_path(path);
3250 		if (!ret)
3251 			goto loop;
3252 
3253 		if (counting) {
3254 			spin_lock(&fs_info->balance_lock);
3255 			bctl->stat.expected++;
3256 			spin_unlock(&fs_info->balance_lock);
3257 			goto loop;
3258 		}
3259 
3260 		ret = btrfs_relocate_chunk(chunk_root,
3261 					   found_key.objectid,
3262 					   found_key.offset);
3263 		if (ret && ret != -ENOSPC)
3264 			goto error;
3265 		if (ret == -ENOSPC) {
3266 			enospc_errors++;
3267 		} else {
3268 			spin_lock(&fs_info->balance_lock);
3269 			bctl->stat.completed++;
3270 			spin_unlock(&fs_info->balance_lock);
3271 		}
3272 loop:
3273 		if (found_key.offset == 0)
3274 			break;
3275 		key.offset = found_key.offset - 1;
3276 	}
3277 
3278 	if (counting) {
3279 		btrfs_release_path(path);
3280 		counting = false;
3281 		goto again;
3282 	}
3283 error:
3284 	btrfs_free_path(path);
3285 	if (enospc_errors) {
3286 		btrfs_info(fs_info, "%d enospc errors during balance",
3287 		       enospc_errors);
3288 		if (!ret)
3289 			ret = -ENOSPC;
3290 	}
3291 
3292 	return ret;
3293 }
3294 
3295 /**
3296  * alloc_profile_is_valid - see if a given profile is valid and reduced
3297  * @flags: profile to validate
3298  * @extended: if true @flags is treated as an extended profile
3299  */
3300 static int alloc_profile_is_valid(u64 flags, int extended)
3301 {
3302 	u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3303 			       BTRFS_BLOCK_GROUP_PROFILE_MASK);
3304 
3305 	flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3306 
3307 	/* 1) check that all other bits are zeroed */
3308 	if (flags & ~mask)
3309 		return 0;
3310 
3311 	/* 2) see if profile is reduced */
3312 	if (flags == 0)
3313 		return !extended; /* "0" is valid for usual profiles */
3314 
3315 	/* true if exactly one bit set */
3316 	return (flags & (flags - 1)) == 0;
3317 }
3318 
3319 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3320 {
3321 	/* cancel requested || normal exit path */
3322 	return atomic_read(&fs_info->balance_cancel_req) ||
3323 		(atomic_read(&fs_info->balance_pause_req) == 0 &&
3324 		 atomic_read(&fs_info->balance_cancel_req) == 0);
3325 }
3326 
3327 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3328 {
3329 	int ret;
3330 
3331 	unset_balance_control(fs_info);
3332 	ret = del_balance_item(fs_info->tree_root);
3333 	if (ret)
3334 		btrfs_std_error(fs_info, ret);
3335 
3336 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3337 }
3338 
3339 /*
3340  * Should be called with both balance and volume mutexes held
3341  */
3342 int btrfs_balance(struct btrfs_balance_control *bctl,
3343 		  struct btrfs_ioctl_balance_args *bargs)
3344 {
3345 	struct btrfs_fs_info *fs_info = bctl->fs_info;
3346 	u64 allowed;
3347 	int mixed = 0;
3348 	int ret;
3349 	u64 num_devices;
3350 	unsigned seq;
3351 
3352 	if (btrfs_fs_closing(fs_info) ||
3353 	    atomic_read(&fs_info->balance_pause_req) ||
3354 	    atomic_read(&fs_info->balance_cancel_req)) {
3355 		ret = -EINVAL;
3356 		goto out;
3357 	}
3358 
3359 	allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3360 	if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3361 		mixed = 1;
3362 
3363 	/*
3364 	 * In case of mixed groups both data and meta should be picked,
3365 	 * and identical options should be given for both of them.
3366 	 */
3367 	allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3368 	if (mixed && (bctl->flags & allowed)) {
3369 		if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3370 		    !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3371 		    memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3372 			btrfs_err(fs_info, "with mixed groups data and "
3373 				   "metadata balance options must be the same");
3374 			ret = -EINVAL;
3375 			goto out;
3376 		}
3377 	}
3378 
3379 	num_devices = fs_info->fs_devices->num_devices;
3380 	btrfs_dev_replace_lock(&fs_info->dev_replace);
3381 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3382 		BUG_ON(num_devices < 1);
3383 		num_devices--;
3384 	}
3385 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
3386 	allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3387 	if (num_devices == 1)
3388 		allowed |= BTRFS_BLOCK_GROUP_DUP;
3389 	else if (num_devices > 1)
3390 		allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3391 	if (num_devices > 2)
3392 		allowed |= BTRFS_BLOCK_GROUP_RAID5;
3393 	if (num_devices > 3)
3394 		allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3395 			    BTRFS_BLOCK_GROUP_RAID6);
3396 	if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3397 	    (!alloc_profile_is_valid(bctl->data.target, 1) ||
3398 	     (bctl->data.target & ~allowed))) {
3399 		btrfs_err(fs_info, "unable to start balance with target "
3400 			   "data profile %llu",
3401 		       bctl->data.target);
3402 		ret = -EINVAL;
3403 		goto out;
3404 	}
3405 	if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3406 	    (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3407 	     (bctl->meta.target & ~allowed))) {
3408 		btrfs_err(fs_info,
3409 			   "unable to start balance with target metadata profile %llu",
3410 		       bctl->meta.target);
3411 		ret = -EINVAL;
3412 		goto out;
3413 	}
3414 	if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3415 	    (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3416 	     (bctl->sys.target & ~allowed))) {
3417 		btrfs_err(fs_info,
3418 			   "unable to start balance with target system profile %llu",
3419 		       bctl->sys.target);
3420 		ret = -EINVAL;
3421 		goto out;
3422 	}
3423 
3424 	/* allow dup'ed data chunks only in mixed mode */
3425 	if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3426 	    (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3427 		btrfs_err(fs_info, "dup for data is not allowed");
3428 		ret = -EINVAL;
3429 		goto out;
3430 	}
3431 
3432 	/* allow to reduce meta or sys integrity only if force set */
3433 	allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3434 			BTRFS_BLOCK_GROUP_RAID10 |
3435 			BTRFS_BLOCK_GROUP_RAID5 |
3436 			BTRFS_BLOCK_GROUP_RAID6;
3437 	do {
3438 		seq = read_seqbegin(&fs_info->profiles_lock);
3439 
3440 		if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3441 		     (fs_info->avail_system_alloc_bits & allowed) &&
3442 		     !(bctl->sys.target & allowed)) ||
3443 		    ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3444 		     (fs_info->avail_metadata_alloc_bits & allowed) &&
3445 		     !(bctl->meta.target & allowed))) {
3446 			if (bctl->flags & BTRFS_BALANCE_FORCE) {
3447 				btrfs_info(fs_info, "force reducing metadata integrity");
3448 			} else {
3449 				btrfs_err(fs_info, "balance will reduce metadata "
3450 					   "integrity, use force if you want this");
3451 				ret = -EINVAL;
3452 				goto out;
3453 			}
3454 		}
3455 	} while (read_seqretry(&fs_info->profiles_lock, seq));
3456 
3457 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3458 		int num_tolerated_disk_barrier_failures;
3459 		u64 target = bctl->sys.target;
3460 
3461 		num_tolerated_disk_barrier_failures =
3462 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3463 		if (num_tolerated_disk_barrier_failures > 0 &&
3464 		    (target &
3465 		     (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3466 		      BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3467 			num_tolerated_disk_barrier_failures = 0;
3468 		else if (num_tolerated_disk_barrier_failures > 1 &&
3469 			 (target &
3470 			  (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3471 			num_tolerated_disk_barrier_failures = 1;
3472 
3473 		fs_info->num_tolerated_disk_barrier_failures =
3474 			num_tolerated_disk_barrier_failures;
3475 	}
3476 
3477 	ret = insert_balance_item(fs_info->tree_root, bctl);
3478 	if (ret && ret != -EEXIST)
3479 		goto out;
3480 
3481 	if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3482 		BUG_ON(ret == -EEXIST);
3483 		set_balance_control(bctl);
3484 	} else {
3485 		BUG_ON(ret != -EEXIST);
3486 		spin_lock(&fs_info->balance_lock);
3487 		update_balance_args(bctl);
3488 		spin_unlock(&fs_info->balance_lock);
3489 	}
3490 
3491 	atomic_inc(&fs_info->balance_running);
3492 	mutex_unlock(&fs_info->balance_mutex);
3493 
3494 	ret = __btrfs_balance(fs_info);
3495 
3496 	mutex_lock(&fs_info->balance_mutex);
3497 	atomic_dec(&fs_info->balance_running);
3498 
3499 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3500 		fs_info->num_tolerated_disk_barrier_failures =
3501 			btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3502 	}
3503 
3504 	if (bargs) {
3505 		memset(bargs, 0, sizeof(*bargs));
3506 		update_ioctl_balance_args(fs_info, 0, bargs);
3507 	}
3508 
3509 	if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3510 	    balance_need_close(fs_info)) {
3511 		__cancel_balance(fs_info);
3512 	}
3513 
3514 	wake_up(&fs_info->balance_wait_q);
3515 
3516 	return ret;
3517 out:
3518 	if (bctl->flags & BTRFS_BALANCE_RESUME)
3519 		__cancel_balance(fs_info);
3520 	else {
3521 		kfree(bctl);
3522 		atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3523 	}
3524 	return ret;
3525 }
3526 
3527 static int balance_kthread(void *data)
3528 {
3529 	struct btrfs_fs_info *fs_info = data;
3530 	int ret = 0;
3531 
3532 	mutex_lock(&fs_info->volume_mutex);
3533 	mutex_lock(&fs_info->balance_mutex);
3534 
3535 	if (fs_info->balance_ctl) {
3536 		btrfs_info(fs_info, "continuing balance");
3537 		ret = btrfs_balance(fs_info->balance_ctl, NULL);
3538 	}
3539 
3540 	mutex_unlock(&fs_info->balance_mutex);
3541 	mutex_unlock(&fs_info->volume_mutex);
3542 
3543 	return ret;
3544 }
3545 
3546 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3547 {
3548 	struct task_struct *tsk;
3549 
3550 	spin_lock(&fs_info->balance_lock);
3551 	if (!fs_info->balance_ctl) {
3552 		spin_unlock(&fs_info->balance_lock);
3553 		return 0;
3554 	}
3555 	spin_unlock(&fs_info->balance_lock);
3556 
3557 	if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3558 		btrfs_info(fs_info, "force skipping balance");
3559 		return 0;
3560 	}
3561 
3562 	tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3563 	return PTR_ERR_OR_ZERO(tsk);
3564 }
3565 
3566 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3567 {
3568 	struct btrfs_balance_control *bctl;
3569 	struct btrfs_balance_item *item;
3570 	struct btrfs_disk_balance_args disk_bargs;
3571 	struct btrfs_path *path;
3572 	struct extent_buffer *leaf;
3573 	struct btrfs_key key;
3574 	int ret;
3575 
3576 	path = btrfs_alloc_path();
3577 	if (!path)
3578 		return -ENOMEM;
3579 
3580 	key.objectid = BTRFS_BALANCE_OBJECTID;
3581 	key.type = BTRFS_BALANCE_ITEM_KEY;
3582 	key.offset = 0;
3583 
3584 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3585 	if (ret < 0)
3586 		goto out;
3587 	if (ret > 0) { /* ret = -ENOENT; */
3588 		ret = 0;
3589 		goto out;
3590 	}
3591 
3592 	bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3593 	if (!bctl) {
3594 		ret = -ENOMEM;
3595 		goto out;
3596 	}
3597 
3598 	leaf = path->nodes[0];
3599 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3600 
3601 	bctl->fs_info = fs_info;
3602 	bctl->flags = btrfs_balance_flags(leaf, item);
3603 	bctl->flags |= BTRFS_BALANCE_RESUME;
3604 
3605 	btrfs_balance_data(leaf, item, &disk_bargs);
3606 	btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3607 	btrfs_balance_meta(leaf, item, &disk_bargs);
3608 	btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3609 	btrfs_balance_sys(leaf, item, &disk_bargs);
3610 	btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3611 
3612 	WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3613 
3614 	mutex_lock(&fs_info->volume_mutex);
3615 	mutex_lock(&fs_info->balance_mutex);
3616 
3617 	set_balance_control(bctl);
3618 
3619 	mutex_unlock(&fs_info->balance_mutex);
3620 	mutex_unlock(&fs_info->volume_mutex);
3621 out:
3622 	btrfs_free_path(path);
3623 	return ret;
3624 }
3625 
3626 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3627 {
3628 	int ret = 0;
3629 
3630 	mutex_lock(&fs_info->balance_mutex);
3631 	if (!fs_info->balance_ctl) {
3632 		mutex_unlock(&fs_info->balance_mutex);
3633 		return -ENOTCONN;
3634 	}
3635 
3636 	if (atomic_read(&fs_info->balance_running)) {
3637 		atomic_inc(&fs_info->balance_pause_req);
3638 		mutex_unlock(&fs_info->balance_mutex);
3639 
3640 		wait_event(fs_info->balance_wait_q,
3641 			   atomic_read(&fs_info->balance_running) == 0);
3642 
3643 		mutex_lock(&fs_info->balance_mutex);
3644 		/* we are good with balance_ctl ripped off from under us */
3645 		BUG_ON(atomic_read(&fs_info->balance_running));
3646 		atomic_dec(&fs_info->balance_pause_req);
3647 	} else {
3648 		ret = -ENOTCONN;
3649 	}
3650 
3651 	mutex_unlock(&fs_info->balance_mutex);
3652 	return ret;
3653 }
3654 
3655 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3656 {
3657 	if (fs_info->sb->s_flags & MS_RDONLY)
3658 		return -EROFS;
3659 
3660 	mutex_lock(&fs_info->balance_mutex);
3661 	if (!fs_info->balance_ctl) {
3662 		mutex_unlock(&fs_info->balance_mutex);
3663 		return -ENOTCONN;
3664 	}
3665 
3666 	atomic_inc(&fs_info->balance_cancel_req);
3667 	/*
3668 	 * if we are running just wait and return, balance item is
3669 	 * deleted in btrfs_balance in this case
3670 	 */
3671 	if (atomic_read(&fs_info->balance_running)) {
3672 		mutex_unlock(&fs_info->balance_mutex);
3673 		wait_event(fs_info->balance_wait_q,
3674 			   atomic_read(&fs_info->balance_running) == 0);
3675 		mutex_lock(&fs_info->balance_mutex);
3676 	} else {
3677 		/* __cancel_balance needs volume_mutex */
3678 		mutex_unlock(&fs_info->balance_mutex);
3679 		mutex_lock(&fs_info->volume_mutex);
3680 		mutex_lock(&fs_info->balance_mutex);
3681 
3682 		if (fs_info->balance_ctl)
3683 			__cancel_balance(fs_info);
3684 
3685 		mutex_unlock(&fs_info->volume_mutex);
3686 	}
3687 
3688 	BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3689 	atomic_dec(&fs_info->balance_cancel_req);
3690 	mutex_unlock(&fs_info->balance_mutex);
3691 	return 0;
3692 }
3693 
3694 static int btrfs_uuid_scan_kthread(void *data)
3695 {
3696 	struct btrfs_fs_info *fs_info = data;
3697 	struct btrfs_root *root = fs_info->tree_root;
3698 	struct btrfs_key key;
3699 	struct btrfs_key max_key;
3700 	struct btrfs_path *path = NULL;
3701 	int ret = 0;
3702 	struct extent_buffer *eb;
3703 	int slot;
3704 	struct btrfs_root_item root_item;
3705 	u32 item_size;
3706 	struct btrfs_trans_handle *trans = NULL;
3707 
3708 	path = btrfs_alloc_path();
3709 	if (!path) {
3710 		ret = -ENOMEM;
3711 		goto out;
3712 	}
3713 
3714 	key.objectid = 0;
3715 	key.type = BTRFS_ROOT_ITEM_KEY;
3716 	key.offset = 0;
3717 
3718 	max_key.objectid = (u64)-1;
3719 	max_key.type = BTRFS_ROOT_ITEM_KEY;
3720 	max_key.offset = (u64)-1;
3721 
3722 	while (1) {
3723 		ret = btrfs_search_forward(root, &key, path, 0);
3724 		if (ret) {
3725 			if (ret > 0)
3726 				ret = 0;
3727 			break;
3728 		}
3729 
3730 		if (key.type != BTRFS_ROOT_ITEM_KEY ||
3731 		    (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3732 		     key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3733 		    key.objectid > BTRFS_LAST_FREE_OBJECTID)
3734 			goto skip;
3735 
3736 		eb = path->nodes[0];
3737 		slot = path->slots[0];
3738 		item_size = btrfs_item_size_nr(eb, slot);
3739 		if (item_size < sizeof(root_item))
3740 			goto skip;
3741 
3742 		read_extent_buffer(eb, &root_item,
3743 				   btrfs_item_ptr_offset(eb, slot),
3744 				   (int)sizeof(root_item));
3745 		if (btrfs_root_refs(&root_item) == 0)
3746 			goto skip;
3747 
3748 		if (!btrfs_is_empty_uuid(root_item.uuid) ||
3749 		    !btrfs_is_empty_uuid(root_item.received_uuid)) {
3750 			if (trans)
3751 				goto update_tree;
3752 
3753 			btrfs_release_path(path);
3754 			/*
3755 			 * 1 - subvol uuid item
3756 			 * 1 - received_subvol uuid item
3757 			 */
3758 			trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3759 			if (IS_ERR(trans)) {
3760 				ret = PTR_ERR(trans);
3761 				break;
3762 			}
3763 			continue;
3764 		} else {
3765 			goto skip;
3766 		}
3767 update_tree:
3768 		if (!btrfs_is_empty_uuid(root_item.uuid)) {
3769 			ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3770 						  root_item.uuid,
3771 						  BTRFS_UUID_KEY_SUBVOL,
3772 						  key.objectid);
3773 			if (ret < 0) {
3774 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
3775 					ret);
3776 				break;
3777 			}
3778 		}
3779 
3780 		if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3781 			ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3782 						  root_item.received_uuid,
3783 						 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3784 						  key.objectid);
3785 			if (ret < 0) {
3786 				btrfs_warn(fs_info, "uuid_tree_add failed %d",
3787 					ret);
3788 				break;
3789 			}
3790 		}
3791 
3792 skip:
3793 		if (trans) {
3794 			ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3795 			trans = NULL;
3796 			if (ret)
3797 				break;
3798 		}
3799 
3800 		btrfs_release_path(path);
3801 		if (key.offset < (u64)-1) {
3802 			key.offset++;
3803 		} else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3804 			key.offset = 0;
3805 			key.type = BTRFS_ROOT_ITEM_KEY;
3806 		} else if (key.objectid < (u64)-1) {
3807 			key.offset = 0;
3808 			key.type = BTRFS_ROOT_ITEM_KEY;
3809 			key.objectid++;
3810 		} else {
3811 			break;
3812 		}
3813 		cond_resched();
3814 	}
3815 
3816 out:
3817 	btrfs_free_path(path);
3818 	if (trans && !IS_ERR(trans))
3819 		btrfs_end_transaction(trans, fs_info->uuid_root);
3820 	if (ret)
3821 		btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3822 	else
3823 		fs_info->update_uuid_tree_gen = 1;
3824 	up(&fs_info->uuid_tree_rescan_sem);
3825 	return 0;
3826 }
3827 
3828 /*
3829  * Callback for btrfs_uuid_tree_iterate().
3830  * returns:
3831  * 0	check succeeded, the entry is not outdated.
3832  * < 0	if an error occured.
3833  * > 0	if the check failed, which means the caller shall remove the entry.
3834  */
3835 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3836 				       u8 *uuid, u8 type, u64 subid)
3837 {
3838 	struct btrfs_key key;
3839 	int ret = 0;
3840 	struct btrfs_root *subvol_root;
3841 
3842 	if (type != BTRFS_UUID_KEY_SUBVOL &&
3843 	    type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3844 		goto out;
3845 
3846 	key.objectid = subid;
3847 	key.type = BTRFS_ROOT_ITEM_KEY;
3848 	key.offset = (u64)-1;
3849 	subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3850 	if (IS_ERR(subvol_root)) {
3851 		ret = PTR_ERR(subvol_root);
3852 		if (ret == -ENOENT)
3853 			ret = 1;
3854 		goto out;
3855 	}
3856 
3857 	switch (type) {
3858 	case BTRFS_UUID_KEY_SUBVOL:
3859 		if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3860 			ret = 1;
3861 		break;
3862 	case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3863 		if (memcmp(uuid, subvol_root->root_item.received_uuid,
3864 			   BTRFS_UUID_SIZE))
3865 			ret = 1;
3866 		break;
3867 	}
3868 
3869 out:
3870 	return ret;
3871 }
3872 
3873 static int btrfs_uuid_rescan_kthread(void *data)
3874 {
3875 	struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3876 	int ret;
3877 
3878 	/*
3879 	 * 1st step is to iterate through the existing UUID tree and
3880 	 * to delete all entries that contain outdated data.
3881 	 * 2nd step is to add all missing entries to the UUID tree.
3882 	 */
3883 	ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3884 	if (ret < 0) {
3885 		btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3886 		up(&fs_info->uuid_tree_rescan_sem);
3887 		return ret;
3888 	}
3889 	return btrfs_uuid_scan_kthread(data);
3890 }
3891 
3892 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3893 {
3894 	struct btrfs_trans_handle *trans;
3895 	struct btrfs_root *tree_root = fs_info->tree_root;
3896 	struct btrfs_root *uuid_root;
3897 	struct task_struct *task;
3898 	int ret;
3899 
3900 	/*
3901 	 * 1 - root node
3902 	 * 1 - root item
3903 	 */
3904 	trans = btrfs_start_transaction(tree_root, 2);
3905 	if (IS_ERR(trans))
3906 		return PTR_ERR(trans);
3907 
3908 	uuid_root = btrfs_create_tree(trans, fs_info,
3909 				      BTRFS_UUID_TREE_OBJECTID);
3910 	if (IS_ERR(uuid_root)) {
3911 		btrfs_abort_transaction(trans, tree_root,
3912 					PTR_ERR(uuid_root));
3913 		return PTR_ERR(uuid_root);
3914 	}
3915 
3916 	fs_info->uuid_root = uuid_root;
3917 
3918 	ret = btrfs_commit_transaction(trans, tree_root);
3919 	if (ret)
3920 		return ret;
3921 
3922 	down(&fs_info->uuid_tree_rescan_sem);
3923 	task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3924 	if (IS_ERR(task)) {
3925 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3926 		btrfs_warn(fs_info, "failed to start uuid_scan task");
3927 		up(&fs_info->uuid_tree_rescan_sem);
3928 		return PTR_ERR(task);
3929 	}
3930 
3931 	return 0;
3932 }
3933 
3934 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3935 {
3936 	struct task_struct *task;
3937 
3938 	down(&fs_info->uuid_tree_rescan_sem);
3939 	task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3940 	if (IS_ERR(task)) {
3941 		/* fs_info->update_uuid_tree_gen remains 0 in all error case */
3942 		btrfs_warn(fs_info, "failed to start uuid_rescan task");
3943 		up(&fs_info->uuid_tree_rescan_sem);
3944 		return PTR_ERR(task);
3945 	}
3946 
3947 	return 0;
3948 }
3949 
3950 /*
3951  * shrinking a device means finding all of the device extents past
3952  * the new size, and then following the back refs to the chunks.
3953  * The chunk relocation code actually frees the device extent
3954  */
3955 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3956 {
3957 	struct btrfs_trans_handle *trans;
3958 	struct btrfs_root *root = device->dev_root;
3959 	struct btrfs_dev_extent *dev_extent = NULL;
3960 	struct btrfs_path *path;
3961 	u64 length;
3962 	u64 chunk_objectid;
3963 	u64 chunk_offset;
3964 	int ret;
3965 	int slot;
3966 	int failed = 0;
3967 	bool retried = false;
3968 	struct extent_buffer *l;
3969 	struct btrfs_key key;
3970 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3971 	u64 old_total = btrfs_super_total_bytes(super_copy);
3972 	u64 old_size = btrfs_device_get_total_bytes(device);
3973 	u64 diff = old_size - new_size;
3974 
3975 	if (device->is_tgtdev_for_dev_replace)
3976 		return -EINVAL;
3977 
3978 	path = btrfs_alloc_path();
3979 	if (!path)
3980 		return -ENOMEM;
3981 
3982 	path->reada = 2;
3983 
3984 	lock_chunks(root);
3985 
3986 	btrfs_device_set_total_bytes(device, new_size);
3987 	if (device->writeable) {
3988 		device->fs_devices->total_rw_bytes -= diff;
3989 		spin_lock(&root->fs_info->free_chunk_lock);
3990 		root->fs_info->free_chunk_space -= diff;
3991 		spin_unlock(&root->fs_info->free_chunk_lock);
3992 	}
3993 	unlock_chunks(root);
3994 
3995 again:
3996 	key.objectid = device->devid;
3997 	key.offset = (u64)-1;
3998 	key.type = BTRFS_DEV_EXTENT_KEY;
3999 
4000 	do {
4001 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4002 		if (ret < 0)
4003 			goto done;
4004 
4005 		ret = btrfs_previous_item(root, path, 0, key.type);
4006 		if (ret < 0)
4007 			goto done;
4008 		if (ret) {
4009 			ret = 0;
4010 			btrfs_release_path(path);
4011 			break;
4012 		}
4013 
4014 		l = path->nodes[0];
4015 		slot = path->slots[0];
4016 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4017 
4018 		if (key.objectid != device->devid) {
4019 			btrfs_release_path(path);
4020 			break;
4021 		}
4022 
4023 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4024 		length = btrfs_dev_extent_length(l, dev_extent);
4025 
4026 		if (key.offset + length <= new_size) {
4027 			btrfs_release_path(path);
4028 			break;
4029 		}
4030 
4031 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4032 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4033 		btrfs_release_path(path);
4034 
4035 		ret = btrfs_relocate_chunk(root, chunk_objectid, chunk_offset);
4036 		if (ret && ret != -ENOSPC)
4037 			goto done;
4038 		if (ret == -ENOSPC)
4039 			failed++;
4040 	} while (key.offset-- > 0);
4041 
4042 	if (failed && !retried) {
4043 		failed = 0;
4044 		retried = true;
4045 		goto again;
4046 	} else if (failed && retried) {
4047 		ret = -ENOSPC;
4048 		lock_chunks(root);
4049 
4050 		btrfs_device_set_total_bytes(device, old_size);
4051 		if (device->writeable)
4052 			device->fs_devices->total_rw_bytes += diff;
4053 		spin_lock(&root->fs_info->free_chunk_lock);
4054 		root->fs_info->free_chunk_space += diff;
4055 		spin_unlock(&root->fs_info->free_chunk_lock);
4056 		unlock_chunks(root);
4057 		goto done;
4058 	}
4059 
4060 	/* Shrinking succeeded, else we would be at "done". */
4061 	trans = btrfs_start_transaction(root, 0);
4062 	if (IS_ERR(trans)) {
4063 		ret = PTR_ERR(trans);
4064 		goto done;
4065 	}
4066 
4067 	lock_chunks(root);
4068 	btrfs_device_set_disk_total_bytes(device, new_size);
4069 	if (list_empty(&device->resized_list))
4070 		list_add_tail(&device->resized_list,
4071 			      &root->fs_info->fs_devices->resized_devices);
4072 
4073 	WARN_ON(diff > old_total);
4074 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
4075 	unlock_chunks(root);
4076 
4077 	/* Now btrfs_update_device() will change the on-disk size. */
4078 	ret = btrfs_update_device(trans, device);
4079 	btrfs_end_transaction(trans, root);
4080 done:
4081 	btrfs_free_path(path);
4082 	return ret;
4083 }
4084 
4085 static int btrfs_add_system_chunk(struct btrfs_root *root,
4086 			   struct btrfs_key *key,
4087 			   struct btrfs_chunk *chunk, int item_size)
4088 {
4089 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4090 	struct btrfs_disk_key disk_key;
4091 	u32 array_size;
4092 	u8 *ptr;
4093 
4094 	lock_chunks(root);
4095 	array_size = btrfs_super_sys_array_size(super_copy);
4096 	if (array_size + item_size + sizeof(disk_key)
4097 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4098 		unlock_chunks(root);
4099 		return -EFBIG;
4100 	}
4101 
4102 	ptr = super_copy->sys_chunk_array + array_size;
4103 	btrfs_cpu_key_to_disk(&disk_key, key);
4104 	memcpy(ptr, &disk_key, sizeof(disk_key));
4105 	ptr += sizeof(disk_key);
4106 	memcpy(ptr, chunk, item_size);
4107 	item_size += sizeof(disk_key);
4108 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4109 	unlock_chunks(root);
4110 
4111 	return 0;
4112 }
4113 
4114 /*
4115  * sort the devices in descending order by max_avail, total_avail
4116  */
4117 static int btrfs_cmp_device_info(const void *a, const void *b)
4118 {
4119 	const struct btrfs_device_info *di_a = a;
4120 	const struct btrfs_device_info *di_b = b;
4121 
4122 	if (di_a->max_avail > di_b->max_avail)
4123 		return -1;
4124 	if (di_a->max_avail < di_b->max_avail)
4125 		return 1;
4126 	if (di_a->total_avail > di_b->total_avail)
4127 		return -1;
4128 	if (di_a->total_avail < di_b->total_avail)
4129 		return 1;
4130 	return 0;
4131 }
4132 
4133 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4134 	[BTRFS_RAID_RAID10] = {
4135 		.sub_stripes	= 2,
4136 		.dev_stripes	= 1,
4137 		.devs_max	= 0,	/* 0 == as many as possible */
4138 		.devs_min	= 4,
4139 		.devs_increment	= 2,
4140 		.ncopies	= 2,
4141 	},
4142 	[BTRFS_RAID_RAID1] = {
4143 		.sub_stripes	= 1,
4144 		.dev_stripes	= 1,
4145 		.devs_max	= 2,
4146 		.devs_min	= 2,
4147 		.devs_increment	= 2,
4148 		.ncopies	= 2,
4149 	},
4150 	[BTRFS_RAID_DUP] = {
4151 		.sub_stripes	= 1,
4152 		.dev_stripes	= 2,
4153 		.devs_max	= 1,
4154 		.devs_min	= 1,
4155 		.devs_increment	= 1,
4156 		.ncopies	= 2,
4157 	},
4158 	[BTRFS_RAID_RAID0] = {
4159 		.sub_stripes	= 1,
4160 		.dev_stripes	= 1,
4161 		.devs_max	= 0,
4162 		.devs_min	= 2,
4163 		.devs_increment	= 1,
4164 		.ncopies	= 1,
4165 	},
4166 	[BTRFS_RAID_SINGLE] = {
4167 		.sub_stripes	= 1,
4168 		.dev_stripes	= 1,
4169 		.devs_max	= 1,
4170 		.devs_min	= 1,
4171 		.devs_increment	= 1,
4172 		.ncopies	= 1,
4173 	},
4174 	[BTRFS_RAID_RAID5] = {
4175 		.sub_stripes	= 1,
4176 		.dev_stripes	= 1,
4177 		.devs_max	= 0,
4178 		.devs_min	= 2,
4179 		.devs_increment	= 1,
4180 		.ncopies	= 2,
4181 	},
4182 	[BTRFS_RAID_RAID6] = {
4183 		.sub_stripes	= 1,
4184 		.dev_stripes	= 1,
4185 		.devs_max	= 0,
4186 		.devs_min	= 3,
4187 		.devs_increment	= 1,
4188 		.ncopies	= 3,
4189 	},
4190 };
4191 
4192 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4193 {
4194 	/* TODO allow them to set a preferred stripe size */
4195 	return 64 * 1024;
4196 }
4197 
4198 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4199 {
4200 	if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4201 		return;
4202 
4203 	btrfs_set_fs_incompat(info, RAID56);
4204 }
4205 
4206 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)		\
4207 			- sizeof(struct btrfs_item)		\
4208 			- sizeof(struct btrfs_chunk))		\
4209 			/ sizeof(struct btrfs_stripe) + 1)
4210 
4211 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
4212 				- 2 * sizeof(struct btrfs_disk_key)	\
4213 				- 2 * sizeof(struct btrfs_chunk))	\
4214 				/ sizeof(struct btrfs_stripe) + 1)
4215 
4216 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4217 			       struct btrfs_root *extent_root, u64 start,
4218 			       u64 type)
4219 {
4220 	struct btrfs_fs_info *info = extent_root->fs_info;
4221 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
4222 	struct list_head *cur;
4223 	struct map_lookup *map = NULL;
4224 	struct extent_map_tree *em_tree;
4225 	struct extent_map *em;
4226 	struct btrfs_device_info *devices_info = NULL;
4227 	u64 total_avail;
4228 	int num_stripes;	/* total number of stripes to allocate */
4229 	int data_stripes;	/* number of stripes that count for
4230 				   block group size */
4231 	int sub_stripes;	/* sub_stripes info for map */
4232 	int dev_stripes;	/* stripes per dev */
4233 	int devs_max;		/* max devs to use */
4234 	int devs_min;		/* min devs needed */
4235 	int devs_increment;	/* ndevs has to be a multiple of this */
4236 	int ncopies;		/* how many copies to data has */
4237 	int ret;
4238 	u64 max_stripe_size;
4239 	u64 max_chunk_size;
4240 	u64 stripe_size;
4241 	u64 num_bytes;
4242 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4243 	int ndevs;
4244 	int i;
4245 	int j;
4246 	int index;
4247 
4248 	BUG_ON(!alloc_profile_is_valid(type, 0));
4249 
4250 	if (list_empty(&fs_devices->alloc_list))
4251 		return -ENOSPC;
4252 
4253 	index = __get_raid_index(type);
4254 
4255 	sub_stripes = btrfs_raid_array[index].sub_stripes;
4256 	dev_stripes = btrfs_raid_array[index].dev_stripes;
4257 	devs_max = btrfs_raid_array[index].devs_max;
4258 	devs_min = btrfs_raid_array[index].devs_min;
4259 	devs_increment = btrfs_raid_array[index].devs_increment;
4260 	ncopies = btrfs_raid_array[index].ncopies;
4261 
4262 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4263 		max_stripe_size = 1024 * 1024 * 1024;
4264 		max_chunk_size = 10 * max_stripe_size;
4265 		if (!devs_max)
4266 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4267 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4268 		/* for larger filesystems, use larger metadata chunks */
4269 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4270 			max_stripe_size = 1024 * 1024 * 1024;
4271 		else
4272 			max_stripe_size = 256 * 1024 * 1024;
4273 		max_chunk_size = max_stripe_size;
4274 		if (!devs_max)
4275 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4276 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4277 		max_stripe_size = 32 * 1024 * 1024;
4278 		max_chunk_size = 2 * max_stripe_size;
4279 		if (!devs_max)
4280 			devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4281 	} else {
4282 		btrfs_err(info, "invalid chunk type 0x%llx requested",
4283 		       type);
4284 		BUG_ON(1);
4285 	}
4286 
4287 	/* we don't want a chunk larger than 10% of writeable space */
4288 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4289 			     max_chunk_size);
4290 
4291 	devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4292 			       GFP_NOFS);
4293 	if (!devices_info)
4294 		return -ENOMEM;
4295 
4296 	cur = fs_devices->alloc_list.next;
4297 
4298 	/*
4299 	 * in the first pass through the devices list, we gather information
4300 	 * about the available holes on each device.
4301 	 */
4302 	ndevs = 0;
4303 	while (cur != &fs_devices->alloc_list) {
4304 		struct btrfs_device *device;
4305 		u64 max_avail;
4306 		u64 dev_offset;
4307 
4308 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4309 
4310 		cur = cur->next;
4311 
4312 		if (!device->writeable) {
4313 			WARN(1, KERN_ERR
4314 			       "BTRFS: read-only device in alloc_list\n");
4315 			continue;
4316 		}
4317 
4318 		if (!device->in_fs_metadata ||
4319 		    device->is_tgtdev_for_dev_replace)
4320 			continue;
4321 
4322 		if (device->total_bytes > device->bytes_used)
4323 			total_avail = device->total_bytes - device->bytes_used;
4324 		else
4325 			total_avail = 0;
4326 
4327 		/* If there is no space on this device, skip it. */
4328 		if (total_avail == 0)
4329 			continue;
4330 
4331 		ret = find_free_dev_extent(trans, device,
4332 					   max_stripe_size * dev_stripes,
4333 					   &dev_offset, &max_avail);
4334 		if (ret && ret != -ENOSPC)
4335 			goto error;
4336 
4337 		if (ret == 0)
4338 			max_avail = max_stripe_size * dev_stripes;
4339 
4340 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4341 			continue;
4342 
4343 		if (ndevs == fs_devices->rw_devices) {
4344 			WARN(1, "%s: found more than %llu devices\n",
4345 			     __func__, fs_devices->rw_devices);
4346 			break;
4347 		}
4348 		devices_info[ndevs].dev_offset = dev_offset;
4349 		devices_info[ndevs].max_avail = max_avail;
4350 		devices_info[ndevs].total_avail = total_avail;
4351 		devices_info[ndevs].dev = device;
4352 		++ndevs;
4353 	}
4354 
4355 	/*
4356 	 * now sort the devices by hole size / available space
4357 	 */
4358 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4359 	     btrfs_cmp_device_info, NULL);
4360 
4361 	/* round down to number of usable stripes */
4362 	ndevs -= ndevs % devs_increment;
4363 
4364 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4365 		ret = -ENOSPC;
4366 		goto error;
4367 	}
4368 
4369 	if (devs_max && ndevs > devs_max)
4370 		ndevs = devs_max;
4371 	/*
4372 	 * the primary goal is to maximize the number of stripes, so use as many
4373 	 * devices as possible, even if the stripes are not maximum sized.
4374 	 */
4375 	stripe_size = devices_info[ndevs-1].max_avail;
4376 	num_stripes = ndevs * dev_stripes;
4377 
4378 	/*
4379 	 * this will have to be fixed for RAID1 and RAID10 over
4380 	 * more drives
4381 	 */
4382 	data_stripes = num_stripes / ncopies;
4383 
4384 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
4385 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4386 				 btrfs_super_stripesize(info->super_copy));
4387 		data_stripes = num_stripes - 1;
4388 	}
4389 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
4390 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4391 				 btrfs_super_stripesize(info->super_copy));
4392 		data_stripes = num_stripes - 2;
4393 	}
4394 
4395 	/*
4396 	 * Use the number of data stripes to figure out how big this chunk
4397 	 * is really going to be in terms of logical address space,
4398 	 * and compare that answer with the max chunk size
4399 	 */
4400 	if (stripe_size * data_stripes > max_chunk_size) {
4401 		u64 mask = (1ULL << 24) - 1;
4402 
4403 		stripe_size = div_u64(max_chunk_size, data_stripes);
4404 
4405 		/* bump the answer up to a 16MB boundary */
4406 		stripe_size = (stripe_size + mask) & ~mask;
4407 
4408 		/* but don't go higher than the limits we found
4409 		 * while searching for free extents
4410 		 */
4411 		if (stripe_size > devices_info[ndevs-1].max_avail)
4412 			stripe_size = devices_info[ndevs-1].max_avail;
4413 	}
4414 
4415 	stripe_size = div_u64(stripe_size, dev_stripes);
4416 
4417 	/* align to BTRFS_STRIPE_LEN */
4418 	stripe_size = div_u64(stripe_size, raid_stripe_len);
4419 	stripe_size *= raid_stripe_len;
4420 
4421 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4422 	if (!map) {
4423 		ret = -ENOMEM;
4424 		goto error;
4425 	}
4426 	map->num_stripes = num_stripes;
4427 
4428 	for (i = 0; i < ndevs; ++i) {
4429 		for (j = 0; j < dev_stripes; ++j) {
4430 			int s = i * dev_stripes + j;
4431 			map->stripes[s].dev = devices_info[i].dev;
4432 			map->stripes[s].physical = devices_info[i].dev_offset +
4433 						   j * stripe_size;
4434 		}
4435 	}
4436 	map->sector_size = extent_root->sectorsize;
4437 	map->stripe_len = raid_stripe_len;
4438 	map->io_align = raid_stripe_len;
4439 	map->io_width = raid_stripe_len;
4440 	map->type = type;
4441 	map->sub_stripes = sub_stripes;
4442 
4443 	num_bytes = stripe_size * data_stripes;
4444 
4445 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4446 
4447 	em = alloc_extent_map();
4448 	if (!em) {
4449 		kfree(map);
4450 		ret = -ENOMEM;
4451 		goto error;
4452 	}
4453 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4454 	em->bdev = (struct block_device *)map;
4455 	em->start = start;
4456 	em->len = num_bytes;
4457 	em->block_start = 0;
4458 	em->block_len = em->len;
4459 	em->orig_block_len = stripe_size;
4460 
4461 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4462 	write_lock(&em_tree->lock);
4463 	ret = add_extent_mapping(em_tree, em, 0);
4464 	if (!ret) {
4465 		list_add_tail(&em->list, &trans->transaction->pending_chunks);
4466 		atomic_inc(&em->refs);
4467 	}
4468 	write_unlock(&em_tree->lock);
4469 	if (ret) {
4470 		free_extent_map(em);
4471 		goto error;
4472 	}
4473 
4474 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
4475 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4476 				     start, num_bytes);
4477 	if (ret)
4478 		goto error_del_extent;
4479 
4480 	for (i = 0; i < map->num_stripes; i++) {
4481 		num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4482 		btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4483 	}
4484 
4485 	spin_lock(&extent_root->fs_info->free_chunk_lock);
4486 	extent_root->fs_info->free_chunk_space -= (stripe_size *
4487 						   map->num_stripes);
4488 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
4489 
4490 	free_extent_map(em);
4491 	check_raid56_incompat_flag(extent_root->fs_info, type);
4492 
4493 	kfree(devices_info);
4494 	return 0;
4495 
4496 error_del_extent:
4497 	write_lock(&em_tree->lock);
4498 	remove_extent_mapping(em_tree, em);
4499 	write_unlock(&em_tree->lock);
4500 
4501 	/* One for our allocation */
4502 	free_extent_map(em);
4503 	/* One for the tree reference */
4504 	free_extent_map(em);
4505 	/* One for the pending_chunks list reference */
4506 	free_extent_map(em);
4507 error:
4508 	kfree(devices_info);
4509 	return ret;
4510 }
4511 
4512 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4513 				struct btrfs_root *extent_root,
4514 				u64 chunk_offset, u64 chunk_size)
4515 {
4516 	struct btrfs_key key;
4517 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4518 	struct btrfs_device *device;
4519 	struct btrfs_chunk *chunk;
4520 	struct btrfs_stripe *stripe;
4521 	struct extent_map_tree *em_tree;
4522 	struct extent_map *em;
4523 	struct map_lookup *map;
4524 	size_t item_size;
4525 	u64 dev_offset;
4526 	u64 stripe_size;
4527 	int i = 0;
4528 	int ret;
4529 
4530 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4531 	read_lock(&em_tree->lock);
4532 	em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4533 	read_unlock(&em_tree->lock);
4534 
4535 	if (!em) {
4536 		btrfs_crit(extent_root->fs_info, "unable to find logical "
4537 			   "%Lu len %Lu", chunk_offset, chunk_size);
4538 		return -EINVAL;
4539 	}
4540 
4541 	if (em->start != chunk_offset || em->len != chunk_size) {
4542 		btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4543 			  " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4544 			  chunk_size, em->start, em->len);
4545 		free_extent_map(em);
4546 		return -EINVAL;
4547 	}
4548 
4549 	map = (struct map_lookup *)em->bdev;
4550 	item_size = btrfs_chunk_item_size(map->num_stripes);
4551 	stripe_size = em->orig_block_len;
4552 
4553 	chunk = kzalloc(item_size, GFP_NOFS);
4554 	if (!chunk) {
4555 		ret = -ENOMEM;
4556 		goto out;
4557 	}
4558 
4559 	for (i = 0; i < map->num_stripes; i++) {
4560 		device = map->stripes[i].dev;
4561 		dev_offset = map->stripes[i].physical;
4562 
4563 		ret = btrfs_update_device(trans, device);
4564 		if (ret)
4565 			goto out;
4566 		ret = btrfs_alloc_dev_extent(trans, device,
4567 					     chunk_root->root_key.objectid,
4568 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4569 					     chunk_offset, dev_offset,
4570 					     stripe_size);
4571 		if (ret)
4572 			goto out;
4573 	}
4574 
4575 	stripe = &chunk->stripe;
4576 	for (i = 0; i < map->num_stripes; i++) {
4577 		device = map->stripes[i].dev;
4578 		dev_offset = map->stripes[i].physical;
4579 
4580 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4581 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4582 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4583 		stripe++;
4584 	}
4585 
4586 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4587 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4588 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4589 	btrfs_set_stack_chunk_type(chunk, map->type);
4590 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4591 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4592 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4593 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4594 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4595 
4596 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4597 	key.type = BTRFS_CHUNK_ITEM_KEY;
4598 	key.offset = chunk_offset;
4599 
4600 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4601 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4602 		/*
4603 		 * TODO: Cleanup of inserted chunk root in case of
4604 		 * failure.
4605 		 */
4606 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4607 					     item_size);
4608 	}
4609 
4610 out:
4611 	kfree(chunk);
4612 	free_extent_map(em);
4613 	return ret;
4614 }
4615 
4616 /*
4617  * Chunk allocation falls into two parts. The first part does works
4618  * that make the new allocated chunk useable, but not do any operation
4619  * that modifies the chunk tree. The second part does the works that
4620  * require modifying the chunk tree. This division is important for the
4621  * bootstrap process of adding storage to a seed btrfs.
4622  */
4623 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4624 		      struct btrfs_root *extent_root, u64 type)
4625 {
4626 	u64 chunk_offset;
4627 
4628 	chunk_offset = find_next_chunk(extent_root->fs_info);
4629 	return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4630 }
4631 
4632 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4633 					 struct btrfs_root *root,
4634 					 struct btrfs_device *device)
4635 {
4636 	u64 chunk_offset;
4637 	u64 sys_chunk_offset;
4638 	u64 alloc_profile;
4639 	struct btrfs_fs_info *fs_info = root->fs_info;
4640 	struct btrfs_root *extent_root = fs_info->extent_root;
4641 	int ret;
4642 
4643 	chunk_offset = find_next_chunk(fs_info);
4644 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4645 	ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4646 				  alloc_profile);
4647 	if (ret)
4648 		return ret;
4649 
4650 	sys_chunk_offset = find_next_chunk(root->fs_info);
4651 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4652 	ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4653 				  alloc_profile);
4654 	return ret;
4655 }
4656 
4657 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4658 {
4659 	int max_errors;
4660 
4661 	if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4662 			 BTRFS_BLOCK_GROUP_RAID10 |
4663 			 BTRFS_BLOCK_GROUP_RAID5 |
4664 			 BTRFS_BLOCK_GROUP_DUP)) {
4665 		max_errors = 1;
4666 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4667 		max_errors = 2;
4668 	} else {
4669 		max_errors = 0;
4670 	}
4671 
4672 	return max_errors;
4673 }
4674 
4675 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4676 {
4677 	struct extent_map *em;
4678 	struct map_lookup *map;
4679 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4680 	int readonly = 0;
4681 	int miss_ndevs = 0;
4682 	int i;
4683 
4684 	read_lock(&map_tree->map_tree.lock);
4685 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4686 	read_unlock(&map_tree->map_tree.lock);
4687 	if (!em)
4688 		return 1;
4689 
4690 	map = (struct map_lookup *)em->bdev;
4691 	for (i = 0; i < map->num_stripes; i++) {
4692 		if (map->stripes[i].dev->missing) {
4693 			miss_ndevs++;
4694 			continue;
4695 		}
4696 
4697 		if (!map->stripes[i].dev->writeable) {
4698 			readonly = 1;
4699 			goto end;
4700 		}
4701 	}
4702 
4703 	/*
4704 	 * If the number of missing devices is larger than max errors,
4705 	 * we can not write the data into that chunk successfully, so
4706 	 * set it readonly.
4707 	 */
4708 	if (miss_ndevs > btrfs_chunk_max_errors(map))
4709 		readonly = 1;
4710 end:
4711 	free_extent_map(em);
4712 	return readonly;
4713 }
4714 
4715 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4716 {
4717 	extent_map_tree_init(&tree->map_tree);
4718 }
4719 
4720 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4721 {
4722 	struct extent_map *em;
4723 
4724 	while (1) {
4725 		write_lock(&tree->map_tree.lock);
4726 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4727 		if (em)
4728 			remove_extent_mapping(&tree->map_tree, em);
4729 		write_unlock(&tree->map_tree.lock);
4730 		if (!em)
4731 			break;
4732 		/* once for us */
4733 		free_extent_map(em);
4734 		/* once for the tree */
4735 		free_extent_map(em);
4736 	}
4737 }
4738 
4739 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4740 {
4741 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4742 	struct extent_map *em;
4743 	struct map_lookup *map;
4744 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4745 	int ret;
4746 
4747 	read_lock(&em_tree->lock);
4748 	em = lookup_extent_mapping(em_tree, logical, len);
4749 	read_unlock(&em_tree->lock);
4750 
4751 	/*
4752 	 * We could return errors for these cases, but that could get ugly and
4753 	 * we'd probably do the same thing which is just not do anything else
4754 	 * and exit, so return 1 so the callers don't try to use other copies.
4755 	 */
4756 	if (!em) {
4757 		btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4758 			    logical+len);
4759 		return 1;
4760 	}
4761 
4762 	if (em->start > logical || em->start + em->len < logical) {
4763 		btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4764 			    "%Lu-%Lu", logical, logical+len, em->start,
4765 			    em->start + em->len);
4766 		free_extent_map(em);
4767 		return 1;
4768 	}
4769 
4770 	map = (struct map_lookup *)em->bdev;
4771 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4772 		ret = map->num_stripes;
4773 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4774 		ret = map->sub_stripes;
4775 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4776 		ret = 2;
4777 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4778 		ret = 3;
4779 	else
4780 		ret = 1;
4781 	free_extent_map(em);
4782 
4783 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4784 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4785 		ret++;
4786 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4787 
4788 	return ret;
4789 }
4790 
4791 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4792 				    struct btrfs_mapping_tree *map_tree,
4793 				    u64 logical)
4794 {
4795 	struct extent_map *em;
4796 	struct map_lookup *map;
4797 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4798 	unsigned long len = root->sectorsize;
4799 
4800 	read_lock(&em_tree->lock);
4801 	em = lookup_extent_mapping(em_tree, logical, len);
4802 	read_unlock(&em_tree->lock);
4803 	BUG_ON(!em);
4804 
4805 	BUG_ON(em->start > logical || em->start + em->len < logical);
4806 	map = (struct map_lookup *)em->bdev;
4807 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4808 		len = map->stripe_len * nr_data_stripes(map);
4809 	free_extent_map(em);
4810 	return len;
4811 }
4812 
4813 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4814 			   u64 logical, u64 len, int mirror_num)
4815 {
4816 	struct extent_map *em;
4817 	struct map_lookup *map;
4818 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4819 	int ret = 0;
4820 
4821 	read_lock(&em_tree->lock);
4822 	em = lookup_extent_mapping(em_tree, logical, len);
4823 	read_unlock(&em_tree->lock);
4824 	BUG_ON(!em);
4825 
4826 	BUG_ON(em->start > logical || em->start + em->len < logical);
4827 	map = (struct map_lookup *)em->bdev;
4828 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4829 		ret = 1;
4830 	free_extent_map(em);
4831 	return ret;
4832 }
4833 
4834 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4835 			    struct map_lookup *map, int first, int num,
4836 			    int optimal, int dev_replace_is_ongoing)
4837 {
4838 	int i;
4839 	int tolerance;
4840 	struct btrfs_device *srcdev;
4841 
4842 	if (dev_replace_is_ongoing &&
4843 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4844 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4845 		srcdev = fs_info->dev_replace.srcdev;
4846 	else
4847 		srcdev = NULL;
4848 
4849 	/*
4850 	 * try to avoid the drive that is the source drive for a
4851 	 * dev-replace procedure, only choose it if no other non-missing
4852 	 * mirror is available
4853 	 */
4854 	for (tolerance = 0; tolerance < 2; tolerance++) {
4855 		if (map->stripes[optimal].dev->bdev &&
4856 		    (tolerance || map->stripes[optimal].dev != srcdev))
4857 			return optimal;
4858 		for (i = first; i < first + num; i++) {
4859 			if (map->stripes[i].dev->bdev &&
4860 			    (tolerance || map->stripes[i].dev != srcdev))
4861 				return i;
4862 		}
4863 	}
4864 
4865 	/* we couldn't find one that doesn't fail.  Just return something
4866 	 * and the io error handling code will clean up eventually
4867 	 */
4868 	return optimal;
4869 }
4870 
4871 static inline int parity_smaller(u64 a, u64 b)
4872 {
4873 	return a > b;
4874 }
4875 
4876 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4877 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
4878 {
4879 	struct btrfs_bio_stripe s;
4880 	int i;
4881 	u64 l;
4882 	int again = 1;
4883 
4884 	while (again) {
4885 		again = 0;
4886 		for (i = 0; i < num_stripes - 1; i++) {
4887 			if (parity_smaller(bbio->raid_map[i],
4888 					   bbio->raid_map[i+1])) {
4889 				s = bbio->stripes[i];
4890 				l = bbio->raid_map[i];
4891 				bbio->stripes[i] = bbio->stripes[i+1];
4892 				bbio->raid_map[i] = bbio->raid_map[i+1];
4893 				bbio->stripes[i+1] = s;
4894 				bbio->raid_map[i+1] = l;
4895 
4896 				again = 1;
4897 			}
4898 		}
4899 	}
4900 }
4901 
4902 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
4903 {
4904 	struct btrfs_bio *bbio = kzalloc(
4905 		 /* the size of the btrfs_bio */
4906 		sizeof(struct btrfs_bio) +
4907 		/* plus the variable array for the stripes */
4908 		sizeof(struct btrfs_bio_stripe) * (total_stripes) +
4909 		/* plus the variable array for the tgt dev */
4910 		sizeof(int) * (real_stripes) +
4911 		/*
4912 		 * plus the raid_map, which includes both the tgt dev
4913 		 * and the stripes
4914 		 */
4915 		sizeof(u64) * (total_stripes),
4916 		GFP_NOFS);
4917 	if (!bbio)
4918 		return NULL;
4919 
4920 	atomic_set(&bbio->error, 0);
4921 	atomic_set(&bbio->refs, 1);
4922 
4923 	return bbio;
4924 }
4925 
4926 void btrfs_get_bbio(struct btrfs_bio *bbio)
4927 {
4928 	WARN_ON(!atomic_read(&bbio->refs));
4929 	atomic_inc(&bbio->refs);
4930 }
4931 
4932 void btrfs_put_bbio(struct btrfs_bio *bbio)
4933 {
4934 	if (!bbio)
4935 		return;
4936 	if (atomic_dec_and_test(&bbio->refs))
4937 		kfree(bbio);
4938 }
4939 
4940 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4941 			     u64 logical, u64 *length,
4942 			     struct btrfs_bio **bbio_ret,
4943 			     int mirror_num, int need_raid_map)
4944 {
4945 	struct extent_map *em;
4946 	struct map_lookup *map;
4947 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4948 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4949 	u64 offset;
4950 	u64 stripe_offset;
4951 	u64 stripe_end_offset;
4952 	u64 stripe_nr;
4953 	u64 stripe_nr_orig;
4954 	u64 stripe_nr_end;
4955 	u64 stripe_len;
4956 	u32 stripe_index;
4957 	int i;
4958 	int ret = 0;
4959 	int num_stripes;
4960 	int max_errors = 0;
4961 	int tgtdev_indexes = 0;
4962 	struct btrfs_bio *bbio = NULL;
4963 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4964 	int dev_replace_is_ongoing = 0;
4965 	int num_alloc_stripes;
4966 	int patch_the_first_stripe_for_dev_replace = 0;
4967 	u64 physical_to_patch_in_first_stripe = 0;
4968 	u64 raid56_full_stripe_start = (u64)-1;
4969 
4970 	read_lock(&em_tree->lock);
4971 	em = lookup_extent_mapping(em_tree, logical, *length);
4972 	read_unlock(&em_tree->lock);
4973 
4974 	if (!em) {
4975 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4976 			logical, *length);
4977 		return -EINVAL;
4978 	}
4979 
4980 	if (em->start > logical || em->start + em->len < logical) {
4981 		btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4982 			   "found %Lu-%Lu", logical, em->start,
4983 			   em->start + em->len);
4984 		free_extent_map(em);
4985 		return -EINVAL;
4986 	}
4987 
4988 	map = (struct map_lookup *)em->bdev;
4989 	offset = logical - em->start;
4990 
4991 	stripe_len = map->stripe_len;
4992 	stripe_nr = offset;
4993 	/*
4994 	 * stripe_nr counts the total number of stripes we have to stride
4995 	 * to get to this block
4996 	 */
4997 	stripe_nr = div64_u64(stripe_nr, stripe_len);
4998 
4999 	stripe_offset = stripe_nr * stripe_len;
5000 	BUG_ON(offset < stripe_offset);
5001 
5002 	/* stripe_offset is the offset of this block in its stripe*/
5003 	stripe_offset = offset - stripe_offset;
5004 
5005 	/* if we're here for raid56, we need to know the stripe aligned start */
5006 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5007 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5008 		raid56_full_stripe_start = offset;
5009 
5010 		/* allow a write of a full stripe, but make sure we don't
5011 		 * allow straddling of stripes
5012 		 */
5013 		raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5014 				full_stripe_len);
5015 		raid56_full_stripe_start *= full_stripe_len;
5016 	}
5017 
5018 	if (rw & REQ_DISCARD) {
5019 		/* we don't discard raid56 yet */
5020 		if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5021 			ret = -EOPNOTSUPP;
5022 			goto out;
5023 		}
5024 		*length = min_t(u64, em->len - offset, *length);
5025 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5026 		u64 max_len;
5027 		/* For writes to RAID[56], allow a full stripeset across all disks.
5028 		   For other RAID types and for RAID[56] reads, just allow a single
5029 		   stripe (on a single disk). */
5030 		if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5031 		    (rw & REQ_WRITE)) {
5032 			max_len = stripe_len * nr_data_stripes(map) -
5033 				(offset - raid56_full_stripe_start);
5034 		} else {
5035 			/* we limit the length of each bio to what fits in a stripe */
5036 			max_len = stripe_len - stripe_offset;
5037 		}
5038 		*length = min_t(u64, em->len - offset, max_len);
5039 	} else {
5040 		*length = em->len - offset;
5041 	}
5042 
5043 	/* This is for when we're called from btrfs_merge_bio_hook() and all
5044 	   it cares about is the length */
5045 	if (!bbio_ret)
5046 		goto out;
5047 
5048 	btrfs_dev_replace_lock(dev_replace);
5049 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5050 	if (!dev_replace_is_ongoing)
5051 		btrfs_dev_replace_unlock(dev_replace);
5052 
5053 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5054 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5055 	    dev_replace->tgtdev != NULL) {
5056 		/*
5057 		 * in dev-replace case, for repair case (that's the only
5058 		 * case where the mirror is selected explicitly when
5059 		 * calling btrfs_map_block), blocks left of the left cursor
5060 		 * can also be read from the target drive.
5061 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
5062 		 * the last one to the array of stripes. For READ, it also
5063 		 * needs to be supported using the same mirror number.
5064 		 * If the requested block is not left of the left cursor,
5065 		 * EIO is returned. This can happen because btrfs_num_copies()
5066 		 * returns one more in the dev-replace case.
5067 		 */
5068 		u64 tmp_length = *length;
5069 		struct btrfs_bio *tmp_bbio = NULL;
5070 		int tmp_num_stripes;
5071 		u64 srcdev_devid = dev_replace->srcdev->devid;
5072 		int index_srcdev = 0;
5073 		int found = 0;
5074 		u64 physical_of_found = 0;
5075 
5076 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5077 			     logical, &tmp_length, &tmp_bbio, 0, 0);
5078 		if (ret) {
5079 			WARN_ON(tmp_bbio != NULL);
5080 			goto out;
5081 		}
5082 
5083 		tmp_num_stripes = tmp_bbio->num_stripes;
5084 		if (mirror_num > tmp_num_stripes) {
5085 			/*
5086 			 * REQ_GET_READ_MIRRORS does not contain this
5087 			 * mirror, that means that the requested area
5088 			 * is not left of the left cursor
5089 			 */
5090 			ret = -EIO;
5091 			btrfs_put_bbio(tmp_bbio);
5092 			goto out;
5093 		}
5094 
5095 		/*
5096 		 * process the rest of the function using the mirror_num
5097 		 * of the source drive. Therefore look it up first.
5098 		 * At the end, patch the device pointer to the one of the
5099 		 * target drive.
5100 		 */
5101 		for (i = 0; i < tmp_num_stripes; i++) {
5102 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5103 				/*
5104 				 * In case of DUP, in order to keep it
5105 				 * simple, only add the mirror with the
5106 				 * lowest physical address
5107 				 */
5108 				if (found &&
5109 				    physical_of_found <=
5110 				     tmp_bbio->stripes[i].physical)
5111 					continue;
5112 				index_srcdev = i;
5113 				found = 1;
5114 				physical_of_found =
5115 					tmp_bbio->stripes[i].physical;
5116 			}
5117 		}
5118 
5119 		if (found) {
5120 			mirror_num = index_srcdev + 1;
5121 			patch_the_first_stripe_for_dev_replace = 1;
5122 			physical_to_patch_in_first_stripe = physical_of_found;
5123 		} else {
5124 			WARN_ON(1);
5125 			ret = -EIO;
5126 			btrfs_put_bbio(tmp_bbio);
5127 			goto out;
5128 		}
5129 
5130 		btrfs_put_bbio(tmp_bbio);
5131 	} else if (mirror_num > map->num_stripes) {
5132 		mirror_num = 0;
5133 	}
5134 
5135 	num_stripes = 1;
5136 	stripe_index = 0;
5137 	stripe_nr_orig = stripe_nr;
5138 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5139 	stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5140 	stripe_end_offset = stripe_nr_end * map->stripe_len -
5141 			    (offset + *length);
5142 
5143 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5144 		if (rw & REQ_DISCARD)
5145 			num_stripes = min_t(u64, map->num_stripes,
5146 					    stripe_nr_end - stripe_nr_orig);
5147 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5148 				&stripe_index);
5149 		if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5150 			mirror_num = 1;
5151 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5152 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5153 			num_stripes = map->num_stripes;
5154 		else if (mirror_num)
5155 			stripe_index = mirror_num - 1;
5156 		else {
5157 			stripe_index = find_live_mirror(fs_info, map, 0,
5158 					    map->num_stripes,
5159 					    current->pid % map->num_stripes,
5160 					    dev_replace_is_ongoing);
5161 			mirror_num = stripe_index + 1;
5162 		}
5163 
5164 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5165 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5166 			num_stripes = map->num_stripes;
5167 		} else if (mirror_num) {
5168 			stripe_index = mirror_num - 1;
5169 		} else {
5170 			mirror_num = 1;
5171 		}
5172 
5173 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5174 		u32 factor = map->num_stripes / map->sub_stripes;
5175 
5176 		stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5177 		stripe_index *= map->sub_stripes;
5178 
5179 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5180 			num_stripes = map->sub_stripes;
5181 		else if (rw & REQ_DISCARD)
5182 			num_stripes = min_t(u64, map->sub_stripes *
5183 					    (stripe_nr_end - stripe_nr_orig),
5184 					    map->num_stripes);
5185 		else if (mirror_num)
5186 			stripe_index += mirror_num - 1;
5187 		else {
5188 			int old_stripe_index = stripe_index;
5189 			stripe_index = find_live_mirror(fs_info, map,
5190 					      stripe_index,
5191 					      map->sub_stripes, stripe_index +
5192 					      current->pid % map->sub_stripes,
5193 					      dev_replace_is_ongoing);
5194 			mirror_num = stripe_index - old_stripe_index + 1;
5195 		}
5196 
5197 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5198 		if (need_raid_map &&
5199 		    ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5200 		     mirror_num > 1)) {
5201 			/* push stripe_nr back to the start of the full stripe */
5202 			stripe_nr = div_u64(raid56_full_stripe_start,
5203 					stripe_len * nr_data_stripes(map));
5204 
5205 			/* RAID[56] write or recovery. Return all stripes */
5206 			num_stripes = map->num_stripes;
5207 			max_errors = nr_parity_stripes(map);
5208 
5209 			*length = map->stripe_len;
5210 			stripe_index = 0;
5211 			stripe_offset = 0;
5212 		} else {
5213 			/*
5214 			 * Mirror #0 or #1 means the original data block.
5215 			 * Mirror #2 is RAID5 parity block.
5216 			 * Mirror #3 is RAID6 Q block.
5217 			 */
5218 			stripe_nr = div_u64_rem(stripe_nr,
5219 					nr_data_stripes(map), &stripe_index);
5220 			if (mirror_num > 1)
5221 				stripe_index = nr_data_stripes(map) +
5222 						mirror_num - 2;
5223 
5224 			/* We distribute the parity blocks across stripes */
5225 			div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5226 					&stripe_index);
5227 			if (!(rw & (REQ_WRITE | REQ_DISCARD |
5228 				    REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5229 				mirror_num = 1;
5230 		}
5231 	} else {
5232 		/*
5233 		 * after this, stripe_nr is the number of stripes on this
5234 		 * device we have to walk to find the data, and stripe_index is
5235 		 * the number of our device in the stripe array
5236 		 */
5237 		stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5238 				&stripe_index);
5239 		mirror_num = stripe_index + 1;
5240 	}
5241 	BUG_ON(stripe_index >= map->num_stripes);
5242 
5243 	num_alloc_stripes = num_stripes;
5244 	if (dev_replace_is_ongoing) {
5245 		if (rw & (REQ_WRITE | REQ_DISCARD))
5246 			num_alloc_stripes <<= 1;
5247 		if (rw & REQ_GET_READ_MIRRORS)
5248 			num_alloc_stripes++;
5249 		tgtdev_indexes = num_stripes;
5250 	}
5251 
5252 	bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5253 	if (!bbio) {
5254 		ret = -ENOMEM;
5255 		goto out;
5256 	}
5257 	if (dev_replace_is_ongoing)
5258 		bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5259 
5260 	/* build raid_map */
5261 	if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5262 	    need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5263 	    mirror_num > 1)) {
5264 		u64 tmp;
5265 		unsigned rot;
5266 
5267 		bbio->raid_map = (u64 *)((void *)bbio->stripes +
5268 				 sizeof(struct btrfs_bio_stripe) *
5269 				 num_alloc_stripes +
5270 				 sizeof(int) * tgtdev_indexes);
5271 
5272 		/* Work out the disk rotation on this stripe-set */
5273 		div_u64_rem(stripe_nr, num_stripes, &rot);
5274 
5275 		/* Fill in the logical address of each stripe */
5276 		tmp = stripe_nr * nr_data_stripes(map);
5277 		for (i = 0; i < nr_data_stripes(map); i++)
5278 			bbio->raid_map[(i+rot) % num_stripes] =
5279 				em->start + (tmp + i) * map->stripe_len;
5280 
5281 		bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5282 		if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5283 			bbio->raid_map[(i+rot+1) % num_stripes] =
5284 				RAID6_Q_STRIPE;
5285 	}
5286 
5287 	if (rw & REQ_DISCARD) {
5288 		u32 factor = 0;
5289 		u32 sub_stripes = 0;
5290 		u64 stripes_per_dev = 0;
5291 		u32 remaining_stripes = 0;
5292 		u32 last_stripe = 0;
5293 
5294 		if (map->type &
5295 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5296 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5297 				sub_stripes = 1;
5298 			else
5299 				sub_stripes = map->sub_stripes;
5300 
5301 			factor = map->num_stripes / sub_stripes;
5302 			stripes_per_dev = div_u64_rem(stripe_nr_end -
5303 						      stripe_nr_orig,
5304 						      factor,
5305 						      &remaining_stripes);
5306 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5307 			last_stripe *= sub_stripes;
5308 		}
5309 
5310 		for (i = 0; i < num_stripes; i++) {
5311 			bbio->stripes[i].physical =
5312 				map->stripes[stripe_index].physical +
5313 				stripe_offset + stripe_nr * map->stripe_len;
5314 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5315 
5316 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5317 					 BTRFS_BLOCK_GROUP_RAID10)) {
5318 				bbio->stripes[i].length = stripes_per_dev *
5319 							  map->stripe_len;
5320 
5321 				if (i / sub_stripes < remaining_stripes)
5322 					bbio->stripes[i].length +=
5323 						map->stripe_len;
5324 
5325 				/*
5326 				 * Special for the first stripe and
5327 				 * the last stripe:
5328 				 *
5329 				 * |-------|...|-------|
5330 				 *     |----------|
5331 				 *    off     end_off
5332 				 */
5333 				if (i < sub_stripes)
5334 					bbio->stripes[i].length -=
5335 						stripe_offset;
5336 
5337 				if (stripe_index >= last_stripe &&
5338 				    stripe_index <= (last_stripe +
5339 						     sub_stripes - 1))
5340 					bbio->stripes[i].length -=
5341 						stripe_end_offset;
5342 
5343 				if (i == sub_stripes - 1)
5344 					stripe_offset = 0;
5345 			} else
5346 				bbio->stripes[i].length = *length;
5347 
5348 			stripe_index++;
5349 			if (stripe_index == map->num_stripes) {
5350 				/* This could only happen for RAID0/10 */
5351 				stripe_index = 0;
5352 				stripe_nr++;
5353 			}
5354 		}
5355 	} else {
5356 		for (i = 0; i < num_stripes; i++) {
5357 			bbio->stripes[i].physical =
5358 				map->stripes[stripe_index].physical +
5359 				stripe_offset +
5360 				stripe_nr * map->stripe_len;
5361 			bbio->stripes[i].dev =
5362 				map->stripes[stripe_index].dev;
5363 			stripe_index++;
5364 		}
5365 	}
5366 
5367 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5368 		max_errors = btrfs_chunk_max_errors(map);
5369 
5370 	if (bbio->raid_map)
5371 		sort_parity_stripes(bbio, num_stripes);
5372 
5373 	tgtdev_indexes = 0;
5374 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5375 	    dev_replace->tgtdev != NULL) {
5376 		int index_where_to_add;
5377 		u64 srcdev_devid = dev_replace->srcdev->devid;
5378 
5379 		/*
5380 		 * duplicate the write operations while the dev replace
5381 		 * procedure is running. Since the copying of the old disk
5382 		 * to the new disk takes place at run time while the
5383 		 * filesystem is mounted writable, the regular write
5384 		 * operations to the old disk have to be duplicated to go
5385 		 * to the new disk as well.
5386 		 * Note that device->missing is handled by the caller, and
5387 		 * that the write to the old disk is already set up in the
5388 		 * stripes array.
5389 		 */
5390 		index_where_to_add = num_stripes;
5391 		for (i = 0; i < num_stripes; i++) {
5392 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5393 				/* write to new disk, too */
5394 				struct btrfs_bio_stripe *new =
5395 					bbio->stripes + index_where_to_add;
5396 				struct btrfs_bio_stripe *old =
5397 					bbio->stripes + i;
5398 
5399 				new->physical = old->physical;
5400 				new->length = old->length;
5401 				new->dev = dev_replace->tgtdev;
5402 				bbio->tgtdev_map[i] = index_where_to_add;
5403 				index_where_to_add++;
5404 				max_errors++;
5405 				tgtdev_indexes++;
5406 			}
5407 		}
5408 		num_stripes = index_where_to_add;
5409 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5410 		   dev_replace->tgtdev != NULL) {
5411 		u64 srcdev_devid = dev_replace->srcdev->devid;
5412 		int index_srcdev = 0;
5413 		int found = 0;
5414 		u64 physical_of_found = 0;
5415 
5416 		/*
5417 		 * During the dev-replace procedure, the target drive can
5418 		 * also be used to read data in case it is needed to repair
5419 		 * a corrupt block elsewhere. This is possible if the
5420 		 * requested area is left of the left cursor. In this area,
5421 		 * the target drive is a full copy of the source drive.
5422 		 */
5423 		for (i = 0; i < num_stripes; i++) {
5424 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5425 				/*
5426 				 * In case of DUP, in order to keep it
5427 				 * simple, only add the mirror with the
5428 				 * lowest physical address
5429 				 */
5430 				if (found &&
5431 				    physical_of_found <=
5432 				     bbio->stripes[i].physical)
5433 					continue;
5434 				index_srcdev = i;
5435 				found = 1;
5436 				physical_of_found = bbio->stripes[i].physical;
5437 			}
5438 		}
5439 		if (found) {
5440 			if (physical_of_found + map->stripe_len <=
5441 			    dev_replace->cursor_left) {
5442 				struct btrfs_bio_stripe *tgtdev_stripe =
5443 					bbio->stripes + num_stripes;
5444 
5445 				tgtdev_stripe->physical = physical_of_found;
5446 				tgtdev_stripe->length =
5447 					bbio->stripes[index_srcdev].length;
5448 				tgtdev_stripe->dev = dev_replace->tgtdev;
5449 				bbio->tgtdev_map[index_srcdev] = num_stripes;
5450 
5451 				tgtdev_indexes++;
5452 				num_stripes++;
5453 			}
5454 		}
5455 	}
5456 
5457 	*bbio_ret = bbio;
5458 	bbio->map_type = map->type;
5459 	bbio->num_stripes = num_stripes;
5460 	bbio->max_errors = max_errors;
5461 	bbio->mirror_num = mirror_num;
5462 	bbio->num_tgtdevs = tgtdev_indexes;
5463 
5464 	/*
5465 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5466 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
5467 	 * available as a mirror
5468 	 */
5469 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5470 		WARN_ON(num_stripes > 1);
5471 		bbio->stripes[0].dev = dev_replace->tgtdev;
5472 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5473 		bbio->mirror_num = map->num_stripes + 1;
5474 	}
5475 out:
5476 	if (dev_replace_is_ongoing)
5477 		btrfs_dev_replace_unlock(dev_replace);
5478 	free_extent_map(em);
5479 	return ret;
5480 }
5481 
5482 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5483 		      u64 logical, u64 *length,
5484 		      struct btrfs_bio **bbio_ret, int mirror_num)
5485 {
5486 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5487 				 mirror_num, 0);
5488 }
5489 
5490 /* For Scrub/replace */
5491 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5492 		     u64 logical, u64 *length,
5493 		     struct btrfs_bio **bbio_ret, int mirror_num,
5494 		     int need_raid_map)
5495 {
5496 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5497 				 mirror_num, need_raid_map);
5498 }
5499 
5500 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5501 		     u64 chunk_start, u64 physical, u64 devid,
5502 		     u64 **logical, int *naddrs, int *stripe_len)
5503 {
5504 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5505 	struct extent_map *em;
5506 	struct map_lookup *map;
5507 	u64 *buf;
5508 	u64 bytenr;
5509 	u64 length;
5510 	u64 stripe_nr;
5511 	u64 rmap_len;
5512 	int i, j, nr = 0;
5513 
5514 	read_lock(&em_tree->lock);
5515 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
5516 	read_unlock(&em_tree->lock);
5517 
5518 	if (!em) {
5519 		printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5520 		       chunk_start);
5521 		return -EIO;
5522 	}
5523 
5524 	if (em->start != chunk_start) {
5525 		printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5526 		       em->start, chunk_start);
5527 		free_extent_map(em);
5528 		return -EIO;
5529 	}
5530 	map = (struct map_lookup *)em->bdev;
5531 
5532 	length = em->len;
5533 	rmap_len = map->stripe_len;
5534 
5535 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5536 		length = div_u64(length, map->num_stripes / map->sub_stripes);
5537 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5538 		length = div_u64(length, map->num_stripes);
5539 	else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5540 		length = div_u64(length, nr_data_stripes(map));
5541 		rmap_len = map->stripe_len * nr_data_stripes(map);
5542 	}
5543 
5544 	buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5545 	BUG_ON(!buf); /* -ENOMEM */
5546 
5547 	for (i = 0; i < map->num_stripes; i++) {
5548 		if (devid && map->stripes[i].dev->devid != devid)
5549 			continue;
5550 		if (map->stripes[i].physical > physical ||
5551 		    map->stripes[i].physical + length <= physical)
5552 			continue;
5553 
5554 		stripe_nr = physical - map->stripes[i].physical;
5555 		stripe_nr = div_u64(stripe_nr, map->stripe_len);
5556 
5557 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5558 			stripe_nr = stripe_nr * map->num_stripes + i;
5559 			stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5560 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5561 			stripe_nr = stripe_nr * map->num_stripes + i;
5562 		} /* else if RAID[56], multiply by nr_data_stripes().
5563 		   * Alternatively, just use rmap_len below instead of
5564 		   * map->stripe_len */
5565 
5566 		bytenr = chunk_start + stripe_nr * rmap_len;
5567 		WARN_ON(nr >= map->num_stripes);
5568 		for (j = 0; j < nr; j++) {
5569 			if (buf[j] == bytenr)
5570 				break;
5571 		}
5572 		if (j == nr) {
5573 			WARN_ON(nr >= map->num_stripes);
5574 			buf[nr++] = bytenr;
5575 		}
5576 	}
5577 
5578 	*logical = buf;
5579 	*naddrs = nr;
5580 	*stripe_len = rmap_len;
5581 
5582 	free_extent_map(em);
5583 	return 0;
5584 }
5585 
5586 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5587 {
5588 	if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5589 		bio_endio_nodec(bio, err);
5590 	else
5591 		bio_endio(bio, err);
5592 	btrfs_put_bbio(bbio);
5593 }
5594 
5595 static void btrfs_end_bio(struct bio *bio, int err)
5596 {
5597 	struct btrfs_bio *bbio = bio->bi_private;
5598 	struct btrfs_device *dev = bbio->stripes[0].dev;
5599 	int is_orig_bio = 0;
5600 
5601 	if (err) {
5602 		atomic_inc(&bbio->error);
5603 		if (err == -EIO || err == -EREMOTEIO) {
5604 			unsigned int stripe_index =
5605 				btrfs_io_bio(bio)->stripe_index;
5606 
5607 			BUG_ON(stripe_index >= bbio->num_stripes);
5608 			dev = bbio->stripes[stripe_index].dev;
5609 			if (dev->bdev) {
5610 				if (bio->bi_rw & WRITE)
5611 					btrfs_dev_stat_inc(dev,
5612 						BTRFS_DEV_STAT_WRITE_ERRS);
5613 				else
5614 					btrfs_dev_stat_inc(dev,
5615 						BTRFS_DEV_STAT_READ_ERRS);
5616 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5617 					btrfs_dev_stat_inc(dev,
5618 						BTRFS_DEV_STAT_FLUSH_ERRS);
5619 				btrfs_dev_stat_print_on_error(dev);
5620 			}
5621 		}
5622 	}
5623 
5624 	if (bio == bbio->orig_bio)
5625 		is_orig_bio = 1;
5626 
5627 	btrfs_bio_counter_dec(bbio->fs_info);
5628 
5629 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5630 		if (!is_orig_bio) {
5631 			bio_put(bio);
5632 			bio = bbio->orig_bio;
5633 		}
5634 
5635 		bio->bi_private = bbio->private;
5636 		bio->bi_end_io = bbio->end_io;
5637 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5638 		/* only send an error to the higher layers if it is
5639 		 * beyond the tolerance of the btrfs bio
5640 		 */
5641 		if (atomic_read(&bbio->error) > bbio->max_errors) {
5642 			err = -EIO;
5643 		} else {
5644 			/*
5645 			 * this bio is actually up to date, we didn't
5646 			 * go over the max number of errors
5647 			 */
5648 			set_bit(BIO_UPTODATE, &bio->bi_flags);
5649 			err = 0;
5650 		}
5651 
5652 		btrfs_end_bbio(bbio, bio, err);
5653 	} else if (!is_orig_bio) {
5654 		bio_put(bio);
5655 	}
5656 }
5657 
5658 /*
5659  * see run_scheduled_bios for a description of why bios are collected for
5660  * async submit.
5661  *
5662  * This will add one bio to the pending list for a device and make sure
5663  * the work struct is scheduled.
5664  */
5665 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5666 					struct btrfs_device *device,
5667 					int rw, struct bio *bio)
5668 {
5669 	int should_queue = 1;
5670 	struct btrfs_pending_bios *pending_bios;
5671 
5672 	if (device->missing || !device->bdev) {
5673 		bio_endio(bio, -EIO);
5674 		return;
5675 	}
5676 
5677 	/* don't bother with additional async steps for reads, right now */
5678 	if (!(rw & REQ_WRITE)) {
5679 		bio_get(bio);
5680 		btrfsic_submit_bio(rw, bio);
5681 		bio_put(bio);
5682 		return;
5683 	}
5684 
5685 	/*
5686 	 * nr_async_bios allows us to reliably return congestion to the
5687 	 * higher layers.  Otherwise, the async bio makes it appear we have
5688 	 * made progress against dirty pages when we've really just put it
5689 	 * on a queue for later
5690 	 */
5691 	atomic_inc(&root->fs_info->nr_async_bios);
5692 	WARN_ON(bio->bi_next);
5693 	bio->bi_next = NULL;
5694 	bio->bi_rw |= rw;
5695 
5696 	spin_lock(&device->io_lock);
5697 	if (bio->bi_rw & REQ_SYNC)
5698 		pending_bios = &device->pending_sync_bios;
5699 	else
5700 		pending_bios = &device->pending_bios;
5701 
5702 	if (pending_bios->tail)
5703 		pending_bios->tail->bi_next = bio;
5704 
5705 	pending_bios->tail = bio;
5706 	if (!pending_bios->head)
5707 		pending_bios->head = bio;
5708 	if (device->running_pending)
5709 		should_queue = 0;
5710 
5711 	spin_unlock(&device->io_lock);
5712 
5713 	if (should_queue)
5714 		btrfs_queue_work(root->fs_info->submit_workers,
5715 				 &device->work);
5716 }
5717 
5718 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5719 		       sector_t sector)
5720 {
5721 	struct bio_vec *prev;
5722 	struct request_queue *q = bdev_get_queue(bdev);
5723 	unsigned int max_sectors = queue_max_sectors(q);
5724 	struct bvec_merge_data bvm = {
5725 		.bi_bdev = bdev,
5726 		.bi_sector = sector,
5727 		.bi_rw = bio->bi_rw,
5728 	};
5729 
5730 	if (WARN_ON(bio->bi_vcnt == 0))
5731 		return 1;
5732 
5733 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5734 	if (bio_sectors(bio) > max_sectors)
5735 		return 0;
5736 
5737 	if (!q->merge_bvec_fn)
5738 		return 1;
5739 
5740 	bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5741 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5742 		return 0;
5743 	return 1;
5744 }
5745 
5746 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5747 			      struct bio *bio, u64 physical, int dev_nr,
5748 			      int rw, int async)
5749 {
5750 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5751 
5752 	bio->bi_private = bbio;
5753 	btrfs_io_bio(bio)->stripe_index = dev_nr;
5754 	bio->bi_end_io = btrfs_end_bio;
5755 	bio->bi_iter.bi_sector = physical >> 9;
5756 #ifdef DEBUG
5757 	{
5758 		struct rcu_string *name;
5759 
5760 		rcu_read_lock();
5761 		name = rcu_dereference(dev->name);
5762 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5763 			 "(%s id %llu), size=%u\n", rw,
5764 			 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5765 			 name->str, dev->devid, bio->bi_iter.bi_size);
5766 		rcu_read_unlock();
5767 	}
5768 #endif
5769 	bio->bi_bdev = dev->bdev;
5770 
5771 	btrfs_bio_counter_inc_noblocked(root->fs_info);
5772 
5773 	if (async)
5774 		btrfs_schedule_bio(root, dev, rw, bio);
5775 	else
5776 		btrfsic_submit_bio(rw, bio);
5777 }
5778 
5779 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5780 			      struct bio *first_bio, struct btrfs_device *dev,
5781 			      int dev_nr, int rw, int async)
5782 {
5783 	struct bio_vec *bvec = first_bio->bi_io_vec;
5784 	struct bio *bio;
5785 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
5786 	u64 physical = bbio->stripes[dev_nr].physical;
5787 
5788 again:
5789 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5790 	if (!bio)
5791 		return -ENOMEM;
5792 
5793 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5794 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5795 				 bvec->bv_offset) < bvec->bv_len) {
5796 			u64 len = bio->bi_iter.bi_size;
5797 
5798 			atomic_inc(&bbio->stripes_pending);
5799 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5800 					  rw, async);
5801 			physical += len;
5802 			goto again;
5803 		}
5804 		bvec++;
5805 	}
5806 
5807 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5808 	return 0;
5809 }
5810 
5811 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5812 {
5813 	atomic_inc(&bbio->error);
5814 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5815 		/* Shoud be the original bio. */
5816 		WARN_ON(bio != bbio->orig_bio);
5817 
5818 		bio->bi_private = bbio->private;
5819 		bio->bi_end_io = bbio->end_io;
5820 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5821 		bio->bi_iter.bi_sector = logical >> 9;
5822 
5823 		btrfs_end_bbio(bbio, bio, -EIO);
5824 	}
5825 }
5826 
5827 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5828 		  int mirror_num, int async_submit)
5829 {
5830 	struct btrfs_device *dev;
5831 	struct bio *first_bio = bio;
5832 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5833 	u64 length = 0;
5834 	u64 map_length;
5835 	int ret;
5836 	int dev_nr;
5837 	int total_devs;
5838 	struct btrfs_bio *bbio = NULL;
5839 
5840 	length = bio->bi_iter.bi_size;
5841 	map_length = length;
5842 
5843 	btrfs_bio_counter_inc_blocked(root->fs_info);
5844 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5845 			      mirror_num, 1);
5846 	if (ret) {
5847 		btrfs_bio_counter_dec(root->fs_info);
5848 		return ret;
5849 	}
5850 
5851 	total_devs = bbio->num_stripes;
5852 	bbio->orig_bio = first_bio;
5853 	bbio->private = first_bio->bi_private;
5854 	bbio->end_io = first_bio->bi_end_io;
5855 	bbio->fs_info = root->fs_info;
5856 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5857 
5858 	if (bbio->raid_map) {
5859 		/* In this case, map_length has been set to the length of
5860 		   a single stripe; not the whole write */
5861 		if (rw & WRITE) {
5862 			ret = raid56_parity_write(root, bio, bbio, map_length);
5863 		} else {
5864 			ret = raid56_parity_recover(root, bio, bbio, map_length,
5865 						    mirror_num, 1);
5866 		}
5867 
5868 		btrfs_bio_counter_dec(root->fs_info);
5869 		return ret;
5870 	}
5871 
5872 	if (map_length < length) {
5873 		btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5874 			logical, length, map_length);
5875 		BUG();
5876 	}
5877 
5878 	for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
5879 		dev = bbio->stripes[dev_nr].dev;
5880 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5881 			bbio_error(bbio, first_bio, logical);
5882 			continue;
5883 		}
5884 
5885 		/*
5886 		 * Check and see if we're ok with this bio based on it's size
5887 		 * and offset with the given device.
5888 		 */
5889 		if (!bio_size_ok(dev->bdev, first_bio,
5890 				 bbio->stripes[dev_nr].physical >> 9)) {
5891 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5892 						 dev_nr, rw, async_submit);
5893 			BUG_ON(ret);
5894 			continue;
5895 		}
5896 
5897 		if (dev_nr < total_devs - 1) {
5898 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5899 			BUG_ON(!bio); /* -ENOMEM */
5900 		} else {
5901 			bio = first_bio;
5902 			bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5903 		}
5904 
5905 		submit_stripe_bio(root, bbio, bio,
5906 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
5907 				  async_submit);
5908 	}
5909 	btrfs_bio_counter_dec(root->fs_info);
5910 	return 0;
5911 }
5912 
5913 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5914 				       u8 *uuid, u8 *fsid)
5915 {
5916 	struct btrfs_device *device;
5917 	struct btrfs_fs_devices *cur_devices;
5918 
5919 	cur_devices = fs_info->fs_devices;
5920 	while (cur_devices) {
5921 		if (!fsid ||
5922 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5923 			device = __find_device(&cur_devices->devices,
5924 					       devid, uuid);
5925 			if (device)
5926 				return device;
5927 		}
5928 		cur_devices = cur_devices->seed;
5929 	}
5930 	return NULL;
5931 }
5932 
5933 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5934 					    struct btrfs_fs_devices *fs_devices,
5935 					    u64 devid, u8 *dev_uuid)
5936 {
5937 	struct btrfs_device *device;
5938 
5939 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5940 	if (IS_ERR(device))
5941 		return NULL;
5942 
5943 	list_add(&device->dev_list, &fs_devices->devices);
5944 	device->fs_devices = fs_devices;
5945 	fs_devices->num_devices++;
5946 
5947 	device->missing = 1;
5948 	fs_devices->missing_devices++;
5949 
5950 	return device;
5951 }
5952 
5953 /**
5954  * btrfs_alloc_device - allocate struct btrfs_device
5955  * @fs_info:	used only for generating a new devid, can be NULL if
5956  *		devid is provided (i.e. @devid != NULL).
5957  * @devid:	a pointer to devid for this device.  If NULL a new devid
5958  *		is generated.
5959  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
5960  *		is generated.
5961  *
5962  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5963  * on error.  Returned struct is not linked onto any lists and can be
5964  * destroyed with kfree() right away.
5965  */
5966 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5967 					const u64 *devid,
5968 					const u8 *uuid)
5969 {
5970 	struct btrfs_device *dev;
5971 	u64 tmp;
5972 
5973 	if (WARN_ON(!devid && !fs_info))
5974 		return ERR_PTR(-EINVAL);
5975 
5976 	dev = __alloc_device();
5977 	if (IS_ERR(dev))
5978 		return dev;
5979 
5980 	if (devid)
5981 		tmp = *devid;
5982 	else {
5983 		int ret;
5984 
5985 		ret = find_next_devid(fs_info, &tmp);
5986 		if (ret) {
5987 			kfree(dev);
5988 			return ERR_PTR(ret);
5989 		}
5990 	}
5991 	dev->devid = tmp;
5992 
5993 	if (uuid)
5994 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5995 	else
5996 		generate_random_uuid(dev->uuid);
5997 
5998 	btrfs_init_work(&dev->work, btrfs_submit_helper,
5999 			pending_bios_fn, NULL, NULL);
6000 
6001 	return dev;
6002 }
6003 
6004 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6005 			  struct extent_buffer *leaf,
6006 			  struct btrfs_chunk *chunk)
6007 {
6008 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6009 	struct map_lookup *map;
6010 	struct extent_map *em;
6011 	u64 logical;
6012 	u64 length;
6013 	u64 devid;
6014 	u8 uuid[BTRFS_UUID_SIZE];
6015 	int num_stripes;
6016 	int ret;
6017 	int i;
6018 
6019 	logical = key->offset;
6020 	length = btrfs_chunk_length(leaf, chunk);
6021 
6022 	read_lock(&map_tree->map_tree.lock);
6023 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6024 	read_unlock(&map_tree->map_tree.lock);
6025 
6026 	/* already mapped? */
6027 	if (em && em->start <= logical && em->start + em->len > logical) {
6028 		free_extent_map(em);
6029 		return 0;
6030 	} else if (em) {
6031 		free_extent_map(em);
6032 	}
6033 
6034 	em = alloc_extent_map();
6035 	if (!em)
6036 		return -ENOMEM;
6037 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6038 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6039 	if (!map) {
6040 		free_extent_map(em);
6041 		return -ENOMEM;
6042 	}
6043 
6044 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6045 	em->bdev = (struct block_device *)map;
6046 	em->start = logical;
6047 	em->len = length;
6048 	em->orig_start = 0;
6049 	em->block_start = 0;
6050 	em->block_len = em->len;
6051 
6052 	map->num_stripes = num_stripes;
6053 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
6054 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
6055 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6056 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6057 	map->type = btrfs_chunk_type(leaf, chunk);
6058 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6059 	for (i = 0; i < num_stripes; i++) {
6060 		map->stripes[i].physical =
6061 			btrfs_stripe_offset_nr(leaf, chunk, i);
6062 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6063 		read_extent_buffer(leaf, uuid, (unsigned long)
6064 				   btrfs_stripe_dev_uuid_nr(chunk, i),
6065 				   BTRFS_UUID_SIZE);
6066 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6067 							uuid, NULL);
6068 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6069 			free_extent_map(em);
6070 			return -EIO;
6071 		}
6072 		if (!map->stripes[i].dev) {
6073 			map->stripes[i].dev =
6074 				add_missing_dev(root, root->fs_info->fs_devices,
6075 						devid, uuid);
6076 			if (!map->stripes[i].dev) {
6077 				free_extent_map(em);
6078 				return -EIO;
6079 			}
6080 		}
6081 		map->stripes[i].dev->in_fs_metadata = 1;
6082 	}
6083 
6084 	write_lock(&map_tree->map_tree.lock);
6085 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6086 	write_unlock(&map_tree->map_tree.lock);
6087 	BUG_ON(ret); /* Tree corruption */
6088 	free_extent_map(em);
6089 
6090 	return 0;
6091 }
6092 
6093 static void fill_device_from_item(struct extent_buffer *leaf,
6094 				 struct btrfs_dev_item *dev_item,
6095 				 struct btrfs_device *device)
6096 {
6097 	unsigned long ptr;
6098 
6099 	device->devid = btrfs_device_id(leaf, dev_item);
6100 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6101 	device->total_bytes = device->disk_total_bytes;
6102 	device->commit_total_bytes = device->disk_total_bytes;
6103 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6104 	device->commit_bytes_used = device->bytes_used;
6105 	device->type = btrfs_device_type(leaf, dev_item);
6106 	device->io_align = btrfs_device_io_align(leaf, dev_item);
6107 	device->io_width = btrfs_device_io_width(leaf, dev_item);
6108 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6109 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6110 	device->is_tgtdev_for_dev_replace = 0;
6111 
6112 	ptr = btrfs_device_uuid(dev_item);
6113 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6114 }
6115 
6116 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6117 						  u8 *fsid)
6118 {
6119 	struct btrfs_fs_devices *fs_devices;
6120 	int ret;
6121 
6122 	BUG_ON(!mutex_is_locked(&uuid_mutex));
6123 
6124 	fs_devices = root->fs_info->fs_devices->seed;
6125 	while (fs_devices) {
6126 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6127 			return fs_devices;
6128 
6129 		fs_devices = fs_devices->seed;
6130 	}
6131 
6132 	fs_devices = find_fsid(fsid);
6133 	if (!fs_devices) {
6134 		if (!btrfs_test_opt(root, DEGRADED))
6135 			return ERR_PTR(-ENOENT);
6136 
6137 		fs_devices = alloc_fs_devices(fsid);
6138 		if (IS_ERR(fs_devices))
6139 			return fs_devices;
6140 
6141 		fs_devices->seeding = 1;
6142 		fs_devices->opened = 1;
6143 		return fs_devices;
6144 	}
6145 
6146 	fs_devices = clone_fs_devices(fs_devices);
6147 	if (IS_ERR(fs_devices))
6148 		return fs_devices;
6149 
6150 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6151 				   root->fs_info->bdev_holder);
6152 	if (ret) {
6153 		free_fs_devices(fs_devices);
6154 		fs_devices = ERR_PTR(ret);
6155 		goto out;
6156 	}
6157 
6158 	if (!fs_devices->seeding) {
6159 		__btrfs_close_devices(fs_devices);
6160 		free_fs_devices(fs_devices);
6161 		fs_devices = ERR_PTR(-EINVAL);
6162 		goto out;
6163 	}
6164 
6165 	fs_devices->seed = root->fs_info->fs_devices->seed;
6166 	root->fs_info->fs_devices->seed = fs_devices;
6167 out:
6168 	return fs_devices;
6169 }
6170 
6171 static int read_one_dev(struct btrfs_root *root,
6172 			struct extent_buffer *leaf,
6173 			struct btrfs_dev_item *dev_item)
6174 {
6175 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6176 	struct btrfs_device *device;
6177 	u64 devid;
6178 	int ret;
6179 	u8 fs_uuid[BTRFS_UUID_SIZE];
6180 	u8 dev_uuid[BTRFS_UUID_SIZE];
6181 
6182 	devid = btrfs_device_id(leaf, dev_item);
6183 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6184 			   BTRFS_UUID_SIZE);
6185 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6186 			   BTRFS_UUID_SIZE);
6187 
6188 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6189 		fs_devices = open_seed_devices(root, fs_uuid);
6190 		if (IS_ERR(fs_devices))
6191 			return PTR_ERR(fs_devices);
6192 	}
6193 
6194 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6195 	if (!device) {
6196 		if (!btrfs_test_opt(root, DEGRADED))
6197 			return -EIO;
6198 
6199 		btrfs_warn(root->fs_info, "devid %llu missing", devid);
6200 		device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6201 		if (!device)
6202 			return -ENOMEM;
6203 	} else {
6204 		if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6205 			return -EIO;
6206 
6207 		if(!device->bdev && !device->missing) {
6208 			/*
6209 			 * this happens when a device that was properly setup
6210 			 * in the device info lists suddenly goes bad.
6211 			 * device->bdev is NULL, and so we have to set
6212 			 * device->missing to one here
6213 			 */
6214 			device->fs_devices->missing_devices++;
6215 			device->missing = 1;
6216 		}
6217 
6218 		/* Move the device to its own fs_devices */
6219 		if (device->fs_devices != fs_devices) {
6220 			ASSERT(device->missing);
6221 
6222 			list_move(&device->dev_list, &fs_devices->devices);
6223 			device->fs_devices->num_devices--;
6224 			fs_devices->num_devices++;
6225 
6226 			device->fs_devices->missing_devices--;
6227 			fs_devices->missing_devices++;
6228 
6229 			device->fs_devices = fs_devices;
6230 		}
6231 	}
6232 
6233 	if (device->fs_devices != root->fs_info->fs_devices) {
6234 		BUG_ON(device->writeable);
6235 		if (device->generation !=
6236 		    btrfs_device_generation(leaf, dev_item))
6237 			return -EINVAL;
6238 	}
6239 
6240 	fill_device_from_item(leaf, dev_item, device);
6241 	device->in_fs_metadata = 1;
6242 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6243 		device->fs_devices->total_rw_bytes += device->total_bytes;
6244 		spin_lock(&root->fs_info->free_chunk_lock);
6245 		root->fs_info->free_chunk_space += device->total_bytes -
6246 			device->bytes_used;
6247 		spin_unlock(&root->fs_info->free_chunk_lock);
6248 	}
6249 	ret = 0;
6250 	return ret;
6251 }
6252 
6253 int btrfs_read_sys_array(struct btrfs_root *root)
6254 {
6255 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6256 	struct extent_buffer *sb;
6257 	struct btrfs_disk_key *disk_key;
6258 	struct btrfs_chunk *chunk;
6259 	u8 *array_ptr;
6260 	unsigned long sb_array_offset;
6261 	int ret = 0;
6262 	u32 num_stripes;
6263 	u32 array_size;
6264 	u32 len = 0;
6265 	u32 cur_offset;
6266 	struct btrfs_key key;
6267 
6268 	ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6269 	/*
6270 	 * This will create extent buffer of nodesize, superblock size is
6271 	 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6272 	 * overallocate but we can keep it as-is, only the first page is used.
6273 	 */
6274 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6275 	if (!sb)
6276 		return -ENOMEM;
6277 	btrfs_set_buffer_uptodate(sb);
6278 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6279 	/*
6280 	 * The sb extent buffer is artifical and just used to read the system array.
6281 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6282 	 * pages up-to-date when the page is larger: extent does not cover the
6283 	 * whole page and consequently check_page_uptodate does not find all
6284 	 * the page's extents up-to-date (the hole beyond sb),
6285 	 * write_extent_buffer then triggers a WARN_ON.
6286 	 *
6287 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6288 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6289 	 * to silence the warning eg. on PowerPC 64.
6290 	 */
6291 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6292 		SetPageUptodate(sb->pages[0]);
6293 
6294 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6295 	array_size = btrfs_super_sys_array_size(super_copy);
6296 
6297 	array_ptr = super_copy->sys_chunk_array;
6298 	sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6299 	cur_offset = 0;
6300 
6301 	while (cur_offset < array_size) {
6302 		disk_key = (struct btrfs_disk_key *)array_ptr;
6303 		len = sizeof(*disk_key);
6304 		if (cur_offset + len > array_size)
6305 			goto out_short_read;
6306 
6307 		btrfs_disk_key_to_cpu(&key, disk_key);
6308 
6309 		array_ptr += len;
6310 		sb_array_offset += len;
6311 		cur_offset += len;
6312 
6313 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6314 			chunk = (struct btrfs_chunk *)sb_array_offset;
6315 			/*
6316 			 * At least one btrfs_chunk with one stripe must be
6317 			 * present, exact stripe count check comes afterwards
6318 			 */
6319 			len = btrfs_chunk_item_size(1);
6320 			if (cur_offset + len > array_size)
6321 				goto out_short_read;
6322 
6323 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6324 			len = btrfs_chunk_item_size(num_stripes);
6325 			if (cur_offset + len > array_size)
6326 				goto out_short_read;
6327 
6328 			ret = read_one_chunk(root, &key, sb, chunk);
6329 			if (ret)
6330 				break;
6331 		} else {
6332 			ret = -EIO;
6333 			break;
6334 		}
6335 		array_ptr += len;
6336 		sb_array_offset += len;
6337 		cur_offset += len;
6338 	}
6339 	free_extent_buffer(sb);
6340 	return ret;
6341 
6342 out_short_read:
6343 	printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6344 			len, cur_offset);
6345 	free_extent_buffer(sb);
6346 	return -EIO;
6347 }
6348 
6349 int btrfs_read_chunk_tree(struct btrfs_root *root)
6350 {
6351 	struct btrfs_path *path;
6352 	struct extent_buffer *leaf;
6353 	struct btrfs_key key;
6354 	struct btrfs_key found_key;
6355 	int ret;
6356 	int slot;
6357 
6358 	root = root->fs_info->chunk_root;
6359 
6360 	path = btrfs_alloc_path();
6361 	if (!path)
6362 		return -ENOMEM;
6363 
6364 	mutex_lock(&uuid_mutex);
6365 	lock_chunks(root);
6366 
6367 	/*
6368 	 * Read all device items, and then all the chunk items. All
6369 	 * device items are found before any chunk item (their object id
6370 	 * is smaller than the lowest possible object id for a chunk
6371 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6372 	 */
6373 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6374 	key.offset = 0;
6375 	key.type = 0;
6376 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6377 	if (ret < 0)
6378 		goto error;
6379 	while (1) {
6380 		leaf = path->nodes[0];
6381 		slot = path->slots[0];
6382 		if (slot >= btrfs_header_nritems(leaf)) {
6383 			ret = btrfs_next_leaf(root, path);
6384 			if (ret == 0)
6385 				continue;
6386 			if (ret < 0)
6387 				goto error;
6388 			break;
6389 		}
6390 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
6391 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6392 			struct btrfs_dev_item *dev_item;
6393 			dev_item = btrfs_item_ptr(leaf, slot,
6394 						  struct btrfs_dev_item);
6395 			ret = read_one_dev(root, leaf, dev_item);
6396 			if (ret)
6397 				goto error;
6398 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6399 			struct btrfs_chunk *chunk;
6400 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6401 			ret = read_one_chunk(root, &found_key, leaf, chunk);
6402 			if (ret)
6403 				goto error;
6404 		}
6405 		path->slots[0]++;
6406 	}
6407 	ret = 0;
6408 error:
6409 	unlock_chunks(root);
6410 	mutex_unlock(&uuid_mutex);
6411 
6412 	btrfs_free_path(path);
6413 	return ret;
6414 }
6415 
6416 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6417 {
6418 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6419 	struct btrfs_device *device;
6420 
6421 	while (fs_devices) {
6422 		mutex_lock(&fs_devices->device_list_mutex);
6423 		list_for_each_entry(device, &fs_devices->devices, dev_list)
6424 			device->dev_root = fs_info->dev_root;
6425 		mutex_unlock(&fs_devices->device_list_mutex);
6426 
6427 		fs_devices = fs_devices->seed;
6428 	}
6429 }
6430 
6431 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6432 {
6433 	int i;
6434 
6435 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6436 		btrfs_dev_stat_reset(dev, i);
6437 }
6438 
6439 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6440 {
6441 	struct btrfs_key key;
6442 	struct btrfs_key found_key;
6443 	struct btrfs_root *dev_root = fs_info->dev_root;
6444 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6445 	struct extent_buffer *eb;
6446 	int slot;
6447 	int ret = 0;
6448 	struct btrfs_device *device;
6449 	struct btrfs_path *path = NULL;
6450 	int i;
6451 
6452 	path = btrfs_alloc_path();
6453 	if (!path) {
6454 		ret = -ENOMEM;
6455 		goto out;
6456 	}
6457 
6458 	mutex_lock(&fs_devices->device_list_mutex);
6459 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6460 		int item_size;
6461 		struct btrfs_dev_stats_item *ptr;
6462 
6463 		key.objectid = 0;
6464 		key.type = BTRFS_DEV_STATS_KEY;
6465 		key.offset = device->devid;
6466 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6467 		if (ret) {
6468 			__btrfs_reset_dev_stats(device);
6469 			device->dev_stats_valid = 1;
6470 			btrfs_release_path(path);
6471 			continue;
6472 		}
6473 		slot = path->slots[0];
6474 		eb = path->nodes[0];
6475 		btrfs_item_key_to_cpu(eb, &found_key, slot);
6476 		item_size = btrfs_item_size_nr(eb, slot);
6477 
6478 		ptr = btrfs_item_ptr(eb, slot,
6479 				     struct btrfs_dev_stats_item);
6480 
6481 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6482 			if (item_size >= (1 + i) * sizeof(__le64))
6483 				btrfs_dev_stat_set(device, i,
6484 					btrfs_dev_stats_value(eb, ptr, i));
6485 			else
6486 				btrfs_dev_stat_reset(device, i);
6487 		}
6488 
6489 		device->dev_stats_valid = 1;
6490 		btrfs_dev_stat_print_on_load(device);
6491 		btrfs_release_path(path);
6492 	}
6493 	mutex_unlock(&fs_devices->device_list_mutex);
6494 
6495 out:
6496 	btrfs_free_path(path);
6497 	return ret < 0 ? ret : 0;
6498 }
6499 
6500 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6501 				struct btrfs_root *dev_root,
6502 				struct btrfs_device *device)
6503 {
6504 	struct btrfs_path *path;
6505 	struct btrfs_key key;
6506 	struct extent_buffer *eb;
6507 	struct btrfs_dev_stats_item *ptr;
6508 	int ret;
6509 	int i;
6510 
6511 	key.objectid = 0;
6512 	key.type = BTRFS_DEV_STATS_KEY;
6513 	key.offset = device->devid;
6514 
6515 	path = btrfs_alloc_path();
6516 	BUG_ON(!path);
6517 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6518 	if (ret < 0) {
6519 		printk_in_rcu(KERN_WARNING "BTRFS: "
6520 			"error %d while searching for dev_stats item for device %s!\n",
6521 			      ret, rcu_str_deref(device->name));
6522 		goto out;
6523 	}
6524 
6525 	if (ret == 0 &&
6526 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6527 		/* need to delete old one and insert a new one */
6528 		ret = btrfs_del_item(trans, dev_root, path);
6529 		if (ret != 0) {
6530 			printk_in_rcu(KERN_WARNING "BTRFS: "
6531 				"delete too small dev_stats item for device %s failed %d!\n",
6532 				      rcu_str_deref(device->name), ret);
6533 			goto out;
6534 		}
6535 		ret = 1;
6536 	}
6537 
6538 	if (ret == 1) {
6539 		/* need to insert a new item */
6540 		btrfs_release_path(path);
6541 		ret = btrfs_insert_empty_item(trans, dev_root, path,
6542 					      &key, sizeof(*ptr));
6543 		if (ret < 0) {
6544 			printk_in_rcu(KERN_WARNING "BTRFS: "
6545 					  "insert dev_stats item for device %s failed %d!\n",
6546 				      rcu_str_deref(device->name), ret);
6547 			goto out;
6548 		}
6549 	}
6550 
6551 	eb = path->nodes[0];
6552 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6553 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6554 		btrfs_set_dev_stats_value(eb, ptr, i,
6555 					  btrfs_dev_stat_read(device, i));
6556 	btrfs_mark_buffer_dirty(eb);
6557 
6558 out:
6559 	btrfs_free_path(path);
6560 	return ret;
6561 }
6562 
6563 /*
6564  * called from commit_transaction. Writes all changed device stats to disk.
6565  */
6566 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6567 			struct btrfs_fs_info *fs_info)
6568 {
6569 	struct btrfs_root *dev_root = fs_info->dev_root;
6570 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6571 	struct btrfs_device *device;
6572 	int stats_cnt;
6573 	int ret = 0;
6574 
6575 	mutex_lock(&fs_devices->device_list_mutex);
6576 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6577 		if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6578 			continue;
6579 
6580 		stats_cnt = atomic_read(&device->dev_stats_ccnt);
6581 		ret = update_dev_stat_item(trans, dev_root, device);
6582 		if (!ret)
6583 			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6584 	}
6585 	mutex_unlock(&fs_devices->device_list_mutex);
6586 
6587 	return ret;
6588 }
6589 
6590 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6591 {
6592 	btrfs_dev_stat_inc(dev, index);
6593 	btrfs_dev_stat_print_on_error(dev);
6594 }
6595 
6596 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6597 {
6598 	if (!dev->dev_stats_valid)
6599 		return;
6600 	printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6601 			   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6602 			   rcu_str_deref(dev->name),
6603 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6604 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6605 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6606 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6607 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6608 }
6609 
6610 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6611 {
6612 	int i;
6613 
6614 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6615 		if (btrfs_dev_stat_read(dev, i) != 0)
6616 			break;
6617 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
6618 		return; /* all values == 0, suppress message */
6619 
6620 	printk_in_rcu(KERN_INFO "BTRFS: "
6621 		   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6622 	       rcu_str_deref(dev->name),
6623 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6624 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6625 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6626 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6627 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6628 }
6629 
6630 int btrfs_get_dev_stats(struct btrfs_root *root,
6631 			struct btrfs_ioctl_get_dev_stats *stats)
6632 {
6633 	struct btrfs_device *dev;
6634 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6635 	int i;
6636 
6637 	mutex_lock(&fs_devices->device_list_mutex);
6638 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6639 	mutex_unlock(&fs_devices->device_list_mutex);
6640 
6641 	if (!dev) {
6642 		btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6643 		return -ENODEV;
6644 	} else if (!dev->dev_stats_valid) {
6645 		btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6646 		return -ENODEV;
6647 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6648 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6649 			if (stats->nr_items > i)
6650 				stats->values[i] =
6651 					btrfs_dev_stat_read_and_reset(dev, i);
6652 			else
6653 				btrfs_dev_stat_reset(dev, i);
6654 		}
6655 	} else {
6656 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6657 			if (stats->nr_items > i)
6658 				stats->values[i] = btrfs_dev_stat_read(dev, i);
6659 	}
6660 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6661 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6662 	return 0;
6663 }
6664 
6665 int btrfs_scratch_superblock(struct btrfs_device *device)
6666 {
6667 	struct buffer_head *bh;
6668 	struct btrfs_super_block *disk_super;
6669 
6670 	bh = btrfs_read_dev_super(device->bdev);
6671 	if (!bh)
6672 		return -EINVAL;
6673 	disk_super = (struct btrfs_super_block *)bh->b_data;
6674 
6675 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6676 	set_buffer_dirty(bh);
6677 	sync_dirty_buffer(bh);
6678 	brelse(bh);
6679 
6680 	return 0;
6681 }
6682 
6683 /*
6684  * Update the size of all devices, which is used for writing out the
6685  * super blocks.
6686  */
6687 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6688 {
6689 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6690 	struct btrfs_device *curr, *next;
6691 
6692 	if (list_empty(&fs_devices->resized_devices))
6693 		return;
6694 
6695 	mutex_lock(&fs_devices->device_list_mutex);
6696 	lock_chunks(fs_info->dev_root);
6697 	list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6698 				 resized_list) {
6699 		list_del_init(&curr->resized_list);
6700 		curr->commit_total_bytes = curr->disk_total_bytes;
6701 	}
6702 	unlock_chunks(fs_info->dev_root);
6703 	mutex_unlock(&fs_devices->device_list_mutex);
6704 }
6705 
6706 /* Must be invoked during the transaction commit */
6707 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6708 					struct btrfs_transaction *transaction)
6709 {
6710 	struct extent_map *em;
6711 	struct map_lookup *map;
6712 	struct btrfs_device *dev;
6713 	int i;
6714 
6715 	if (list_empty(&transaction->pending_chunks))
6716 		return;
6717 
6718 	/* In order to kick the device replace finish process */
6719 	lock_chunks(root);
6720 	list_for_each_entry(em, &transaction->pending_chunks, list) {
6721 		map = (struct map_lookup *)em->bdev;
6722 
6723 		for (i = 0; i < map->num_stripes; i++) {
6724 			dev = map->stripes[i].dev;
6725 			dev->commit_bytes_used = dev->bytes_used;
6726 		}
6727 	}
6728 	unlock_chunks(root);
6729 }
6730