xref: /linux/fs/btrfs/volumes.c (revision 04eeb606a8383b306f4bc6991da8231b5f3924b0)
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 void lock_chunks(struct btrfs_root *root)
57 {
58 	mutex_lock(&root->fs_info->chunk_mutex);
59 }
60 
61 static void unlock_chunks(struct btrfs_root *root)
62 {
63 	mutex_unlock(&root->fs_info->chunk_mutex);
64 }
65 
66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
67 {
68 	struct btrfs_fs_devices *fs_devs;
69 
70 	fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
71 	if (!fs_devs)
72 		return ERR_PTR(-ENOMEM);
73 
74 	mutex_init(&fs_devs->device_list_mutex);
75 
76 	INIT_LIST_HEAD(&fs_devs->devices);
77 	INIT_LIST_HEAD(&fs_devs->resized_devices);
78 	INIT_LIST_HEAD(&fs_devs->alloc_list);
79 	INIT_LIST_HEAD(&fs_devs->list);
80 
81 	return fs_devs;
82 }
83 
84 /**
85  * alloc_fs_devices - allocate struct btrfs_fs_devices
86  * @fsid:	a pointer to UUID for this FS.  If NULL a new UUID is
87  *		generated.
88  *
89  * Return: a pointer to a new &struct btrfs_fs_devices on success;
90  * ERR_PTR() on error.  Returned struct is not linked onto any lists and
91  * can be destroyed with kfree() right away.
92  */
93 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
94 {
95 	struct btrfs_fs_devices *fs_devs;
96 
97 	fs_devs = __alloc_fs_devices();
98 	if (IS_ERR(fs_devs))
99 		return fs_devs;
100 
101 	if (fsid)
102 		memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
103 	else
104 		generate_random_uuid(fs_devs->fsid);
105 
106 	return fs_devs;
107 }
108 
109 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
110 {
111 	struct btrfs_device *device;
112 	WARN_ON(fs_devices->opened);
113 	while (!list_empty(&fs_devices->devices)) {
114 		device = list_entry(fs_devices->devices.next,
115 				    struct btrfs_device, dev_list);
116 		list_del(&device->dev_list);
117 		rcu_string_free(device->name);
118 		kfree(device);
119 	}
120 	kfree(fs_devices);
121 }
122 
123 static void btrfs_kobject_uevent(struct block_device *bdev,
124 				 enum kobject_action action)
125 {
126 	int ret;
127 
128 	ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
129 	if (ret)
130 		pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
131 			action,
132 			kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
133 			&disk_to_dev(bdev->bd_disk)->kobj);
134 }
135 
136 void btrfs_cleanup_fs_uuids(void)
137 {
138 	struct btrfs_fs_devices *fs_devices;
139 
140 	while (!list_empty(&fs_uuids)) {
141 		fs_devices = list_entry(fs_uuids.next,
142 					struct btrfs_fs_devices, list);
143 		list_del(&fs_devices->list);
144 		free_fs_devices(fs_devices);
145 	}
146 }
147 
148 static struct btrfs_device *__alloc_device(void)
149 {
150 	struct btrfs_device *dev;
151 
152 	dev = kzalloc(sizeof(*dev), GFP_NOFS);
153 	if (!dev)
154 		return ERR_PTR(-ENOMEM);
155 
156 	INIT_LIST_HEAD(&dev->dev_list);
157 	INIT_LIST_HEAD(&dev->dev_alloc_list);
158 	INIT_LIST_HEAD(&dev->resized_list);
159 
160 	spin_lock_init(&dev->io_lock);
161 
162 	spin_lock_init(&dev->reada_lock);
163 	atomic_set(&dev->reada_in_flight, 0);
164 	atomic_set(&dev->dev_stats_ccnt, 0);
165 	INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
166 	INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
167 
168 	return dev;
169 }
170 
171 static noinline struct btrfs_device *__find_device(struct list_head *head,
172 						   u64 devid, u8 *uuid)
173 {
174 	struct btrfs_device *dev;
175 
176 	list_for_each_entry(dev, head, dev_list) {
177 		if (dev->devid == devid &&
178 		    (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
179 			return dev;
180 		}
181 	}
182 	return NULL;
183 }
184 
185 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
186 {
187 	struct btrfs_fs_devices *fs_devices;
188 
189 	list_for_each_entry(fs_devices, &fs_uuids, list) {
190 		if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
191 			return fs_devices;
192 	}
193 	return NULL;
194 }
195 
196 static int
197 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
198 		      int flush, struct block_device **bdev,
199 		      struct buffer_head **bh)
200 {
201 	int ret;
202 
203 	*bdev = blkdev_get_by_path(device_path, flags, holder);
204 
205 	if (IS_ERR(*bdev)) {
206 		ret = PTR_ERR(*bdev);
207 		printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
208 		goto error;
209 	}
210 
211 	if (flush)
212 		filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
213 	ret = set_blocksize(*bdev, 4096);
214 	if (ret) {
215 		blkdev_put(*bdev, flags);
216 		goto error;
217 	}
218 	invalidate_bdev(*bdev);
219 	*bh = btrfs_read_dev_super(*bdev);
220 	if (!*bh) {
221 		ret = -EINVAL;
222 		blkdev_put(*bdev, flags);
223 		goto error;
224 	}
225 
226 	return 0;
227 
228 error:
229 	*bdev = NULL;
230 	*bh = NULL;
231 	return ret;
232 }
233 
234 static void requeue_list(struct btrfs_pending_bios *pending_bios,
235 			struct bio *head, struct bio *tail)
236 {
237 
238 	struct bio *old_head;
239 
240 	old_head = pending_bios->head;
241 	pending_bios->head = head;
242 	if (pending_bios->tail)
243 		tail->bi_next = old_head;
244 	else
245 		pending_bios->tail = tail;
246 }
247 
248 /*
249  * we try to collect pending bios for a device so we don't get a large
250  * number of procs sending bios down to the same device.  This greatly
251  * improves the schedulers ability to collect and merge the bios.
252  *
253  * But, it also turns into a long list of bios to process and that is sure
254  * to eventually make the worker thread block.  The solution here is to
255  * make some progress and then put this work struct back at the end of
256  * the list if the block device is congested.  This way, multiple devices
257  * can make progress from a single worker thread.
258  */
259 static noinline void run_scheduled_bios(struct btrfs_device *device)
260 {
261 	struct bio *pending;
262 	struct backing_dev_info *bdi;
263 	struct btrfs_fs_info *fs_info;
264 	struct btrfs_pending_bios *pending_bios;
265 	struct bio *tail;
266 	struct bio *cur;
267 	int again = 0;
268 	unsigned long num_run;
269 	unsigned long batch_run = 0;
270 	unsigned long limit;
271 	unsigned long last_waited = 0;
272 	int force_reg = 0;
273 	int sync_pending = 0;
274 	struct blk_plug plug;
275 
276 	/*
277 	 * this function runs all the bios we've collected for
278 	 * a particular device.  We don't want to wander off to
279 	 * another device without first sending all of these down.
280 	 * So, setup a plug here and finish it off before we return
281 	 */
282 	blk_start_plug(&plug);
283 
284 	bdi = blk_get_backing_dev_info(device->bdev);
285 	fs_info = device->dev_root->fs_info;
286 	limit = btrfs_async_submit_limit(fs_info);
287 	limit = limit * 2 / 3;
288 
289 loop:
290 	spin_lock(&device->io_lock);
291 
292 loop_lock:
293 	num_run = 0;
294 
295 	/* take all the bios off the list at once and process them
296 	 * later on (without the lock held).  But, remember the
297 	 * tail and other pointers so the bios can be properly reinserted
298 	 * into the list if we hit congestion
299 	 */
300 	if (!force_reg && device->pending_sync_bios.head) {
301 		pending_bios = &device->pending_sync_bios;
302 		force_reg = 1;
303 	} else {
304 		pending_bios = &device->pending_bios;
305 		force_reg = 0;
306 	}
307 
308 	pending = pending_bios->head;
309 	tail = pending_bios->tail;
310 	WARN_ON(pending && !tail);
311 
312 	/*
313 	 * if pending was null this time around, no bios need processing
314 	 * at all and we can stop.  Otherwise it'll loop back up again
315 	 * and do an additional check so no bios are missed.
316 	 *
317 	 * device->running_pending is used to synchronize with the
318 	 * schedule_bio code.
319 	 */
320 	if (device->pending_sync_bios.head == NULL &&
321 	    device->pending_bios.head == NULL) {
322 		again = 0;
323 		device->running_pending = 0;
324 	} else {
325 		again = 1;
326 		device->running_pending = 1;
327 	}
328 
329 	pending_bios->head = NULL;
330 	pending_bios->tail = NULL;
331 
332 	spin_unlock(&device->io_lock);
333 
334 	while (pending) {
335 
336 		rmb();
337 		/* we want to work on both lists, but do more bios on the
338 		 * sync list than the regular list
339 		 */
340 		if ((num_run > 32 &&
341 		    pending_bios != &device->pending_sync_bios &&
342 		    device->pending_sync_bios.head) ||
343 		   (num_run > 64 && pending_bios == &device->pending_sync_bios &&
344 		    device->pending_bios.head)) {
345 			spin_lock(&device->io_lock);
346 			requeue_list(pending_bios, pending, tail);
347 			goto loop_lock;
348 		}
349 
350 		cur = pending;
351 		pending = pending->bi_next;
352 		cur->bi_next = NULL;
353 
354 		if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
355 		    waitqueue_active(&fs_info->async_submit_wait))
356 			wake_up(&fs_info->async_submit_wait);
357 
358 		BUG_ON(atomic_read(&cur->bi_cnt) == 0);
359 
360 		/*
361 		 * if we're doing the sync list, record that our
362 		 * plug has some sync requests on it
363 		 *
364 		 * If we're doing the regular list and there are
365 		 * sync requests sitting around, unplug before
366 		 * we add more
367 		 */
368 		if (pending_bios == &device->pending_sync_bios) {
369 			sync_pending = 1;
370 		} else if (sync_pending) {
371 			blk_finish_plug(&plug);
372 			blk_start_plug(&plug);
373 			sync_pending = 0;
374 		}
375 
376 		btrfsic_submit_bio(cur->bi_rw, cur);
377 		num_run++;
378 		batch_run++;
379 		if (need_resched())
380 			cond_resched();
381 
382 		/*
383 		 * we made progress, there is more work to do and the bdi
384 		 * is now congested.  Back off and let other work structs
385 		 * run instead
386 		 */
387 		if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
388 		    fs_info->fs_devices->open_devices > 1) {
389 			struct io_context *ioc;
390 
391 			ioc = current->io_context;
392 
393 			/*
394 			 * the main goal here is that we don't want to
395 			 * block if we're going to be able to submit
396 			 * more requests without blocking.
397 			 *
398 			 * This code does two great things, it pokes into
399 			 * the elevator code from a filesystem _and_
400 			 * it makes assumptions about how batching works.
401 			 */
402 			if (ioc && ioc->nr_batch_requests > 0 &&
403 			    time_before(jiffies, ioc->last_waited + HZ/50UL) &&
404 			    (last_waited == 0 ||
405 			     ioc->last_waited == last_waited)) {
406 				/*
407 				 * we want to go through our batch of
408 				 * requests and stop.  So, we copy out
409 				 * the ioc->last_waited time and test
410 				 * against it before looping
411 				 */
412 				last_waited = ioc->last_waited;
413 				if (need_resched())
414 					cond_resched();
415 				continue;
416 			}
417 			spin_lock(&device->io_lock);
418 			requeue_list(pending_bios, pending, tail);
419 			device->running_pending = 1;
420 
421 			spin_unlock(&device->io_lock);
422 			btrfs_queue_work(fs_info->submit_workers,
423 					 &device->work);
424 			goto done;
425 		}
426 		/* unplug every 64 requests just for good measure */
427 		if (batch_run % 64 == 0) {
428 			blk_finish_plug(&plug);
429 			blk_start_plug(&plug);
430 			sync_pending = 0;
431 		}
432 	}
433 
434 	cond_resched();
435 	if (again)
436 		goto loop;
437 
438 	spin_lock(&device->io_lock);
439 	if (device->pending_bios.head || device->pending_sync_bios.head)
440 		goto loop_lock;
441 	spin_unlock(&device->io_lock);
442 
443 done:
444 	blk_finish_plug(&plug);
445 }
446 
447 static void pending_bios_fn(struct btrfs_work *work)
448 {
449 	struct btrfs_device *device;
450 
451 	device = container_of(work, struct btrfs_device, work);
452 	run_scheduled_bios(device);
453 }
454 
455 /*
456  * Add new device to list of registered devices
457  *
458  * Returns:
459  * 1   - first time device is seen
460  * 0   - device already known
461  * < 0 - error
462  */
463 static noinline int device_list_add(const char *path,
464 			   struct btrfs_super_block *disk_super,
465 			   u64 devid, struct btrfs_fs_devices **fs_devices_ret)
466 {
467 	struct btrfs_device *device;
468 	struct btrfs_fs_devices *fs_devices;
469 	struct rcu_string *name;
470 	int ret = 0;
471 	u64 found_transid = btrfs_super_generation(disk_super);
472 
473 	fs_devices = find_fsid(disk_super->fsid);
474 	if (!fs_devices) {
475 		fs_devices = alloc_fs_devices(disk_super->fsid);
476 		if (IS_ERR(fs_devices))
477 			return PTR_ERR(fs_devices);
478 
479 		list_add(&fs_devices->list, &fs_uuids);
480 
481 		device = NULL;
482 	} else {
483 		device = __find_device(&fs_devices->devices, devid,
484 				       disk_super->dev_item.uuid);
485 	}
486 
487 	if (!device) {
488 		if (fs_devices->opened)
489 			return -EBUSY;
490 
491 		device = btrfs_alloc_device(NULL, &devid,
492 					    disk_super->dev_item.uuid);
493 		if (IS_ERR(device)) {
494 			/* we can safely leave the fs_devices entry around */
495 			return PTR_ERR(device);
496 		}
497 
498 		name = rcu_string_strdup(path, GFP_NOFS);
499 		if (!name) {
500 			kfree(device);
501 			return -ENOMEM;
502 		}
503 		rcu_assign_pointer(device->name, name);
504 
505 		mutex_lock(&fs_devices->device_list_mutex);
506 		list_add_rcu(&device->dev_list, &fs_devices->devices);
507 		fs_devices->num_devices++;
508 		mutex_unlock(&fs_devices->device_list_mutex);
509 
510 		ret = 1;
511 		device->fs_devices = fs_devices;
512 	} else if (!device->name || strcmp(device->name->str, path)) {
513 		/*
514 		 * When FS is already mounted.
515 		 * 1. If you are here and if the device->name is NULL that
516 		 *    means this device was missing at time of FS mount.
517 		 * 2. If you are here and if the device->name is different
518 		 *    from 'path' that means either
519 		 *      a. The same device disappeared and reappeared with
520 		 *         different name. or
521 		 *      b. The missing-disk-which-was-replaced, has
522 		 *         reappeared now.
523 		 *
524 		 * We must allow 1 and 2a above. But 2b would be a spurious
525 		 * and unintentional.
526 		 *
527 		 * Further in case of 1 and 2a above, the disk at 'path'
528 		 * would have missed some transaction when it was away and
529 		 * in case of 2a the stale bdev has to be updated as well.
530 		 * 2b must not be allowed at all time.
531 		 */
532 
533 		/*
534 		 * For now, we do allow update to btrfs_fs_device through the
535 		 * btrfs dev scan cli after FS has been mounted.  We're still
536 		 * tracking a problem where systems fail mount by subvolume id
537 		 * when we reject replacement on a mounted FS.
538 		 */
539 		if (!fs_devices->opened && found_transid < device->generation) {
540 			/*
541 			 * That is if the FS is _not_ mounted and if you
542 			 * are here, that means there is more than one
543 			 * disk with same uuid and devid.We keep the one
544 			 * with larger generation number or the last-in if
545 			 * generation are equal.
546 			 */
547 			return -EEXIST;
548 		}
549 
550 		name = rcu_string_strdup(path, GFP_NOFS);
551 		if (!name)
552 			return -ENOMEM;
553 		rcu_string_free(device->name);
554 		rcu_assign_pointer(device->name, name);
555 		if (device->missing) {
556 			fs_devices->missing_devices--;
557 			device->missing = 0;
558 		}
559 	}
560 
561 	/*
562 	 * Unmount does not free the btrfs_device struct but would zero
563 	 * generation along with most of the other members. So just update
564 	 * it back. We need it to pick the disk with largest generation
565 	 * (as above).
566 	 */
567 	if (!fs_devices->opened)
568 		device->generation = found_transid;
569 
570 	*fs_devices_ret = fs_devices;
571 
572 	return ret;
573 }
574 
575 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
576 {
577 	struct btrfs_fs_devices *fs_devices;
578 	struct btrfs_device *device;
579 	struct btrfs_device *orig_dev;
580 
581 	fs_devices = alloc_fs_devices(orig->fsid);
582 	if (IS_ERR(fs_devices))
583 		return fs_devices;
584 
585 	mutex_lock(&orig->device_list_mutex);
586 	fs_devices->total_devices = orig->total_devices;
587 
588 	/* We have held the volume lock, it is safe to get the devices. */
589 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
590 		struct rcu_string *name;
591 
592 		device = btrfs_alloc_device(NULL, &orig_dev->devid,
593 					    orig_dev->uuid);
594 		if (IS_ERR(device))
595 			goto error;
596 
597 		/*
598 		 * This is ok to do without rcu read locked because we hold the
599 		 * uuid mutex so nothing we touch in here is going to disappear.
600 		 */
601 		if (orig_dev->name) {
602 			name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
603 			if (!name) {
604 				kfree(device);
605 				goto error;
606 			}
607 			rcu_assign_pointer(device->name, name);
608 		}
609 
610 		list_add(&device->dev_list, &fs_devices->devices);
611 		device->fs_devices = fs_devices;
612 		fs_devices->num_devices++;
613 	}
614 	mutex_unlock(&orig->device_list_mutex);
615 	return fs_devices;
616 error:
617 	mutex_unlock(&orig->device_list_mutex);
618 	free_fs_devices(fs_devices);
619 	return ERR_PTR(-ENOMEM);
620 }
621 
622 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
623 			       struct btrfs_fs_devices *fs_devices, int step)
624 {
625 	struct btrfs_device *device, *next;
626 	struct btrfs_device *latest_dev = NULL;
627 
628 	mutex_lock(&uuid_mutex);
629 again:
630 	/* This is the initialized path, it is safe to release the devices. */
631 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
632 		if (device->in_fs_metadata) {
633 			if (!device->is_tgtdev_for_dev_replace &&
634 			    (!latest_dev ||
635 			     device->generation > latest_dev->generation)) {
636 				latest_dev = device;
637 			}
638 			continue;
639 		}
640 
641 		if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
642 			/*
643 			 * In the first step, keep the device which has
644 			 * the correct fsid and the devid that is used
645 			 * for the dev_replace procedure.
646 			 * In the second step, the dev_replace state is
647 			 * read from the device tree and it is known
648 			 * whether the procedure is really active or
649 			 * not, which means whether this device is
650 			 * used or whether it should be removed.
651 			 */
652 			if (step == 0 || device->is_tgtdev_for_dev_replace) {
653 				continue;
654 			}
655 		}
656 		if (device->bdev) {
657 			blkdev_put(device->bdev, device->mode);
658 			device->bdev = NULL;
659 			fs_devices->open_devices--;
660 		}
661 		if (device->writeable) {
662 			list_del_init(&device->dev_alloc_list);
663 			device->writeable = 0;
664 			if (!device->is_tgtdev_for_dev_replace)
665 				fs_devices->rw_devices--;
666 		}
667 		list_del_init(&device->dev_list);
668 		fs_devices->num_devices--;
669 		rcu_string_free(device->name);
670 		kfree(device);
671 	}
672 
673 	if (fs_devices->seed) {
674 		fs_devices = fs_devices->seed;
675 		goto again;
676 	}
677 
678 	fs_devices->latest_bdev = latest_dev->bdev;
679 
680 	mutex_unlock(&uuid_mutex);
681 }
682 
683 static void __free_device(struct work_struct *work)
684 {
685 	struct btrfs_device *device;
686 
687 	device = container_of(work, struct btrfs_device, rcu_work);
688 
689 	if (device->bdev)
690 		blkdev_put(device->bdev, device->mode);
691 
692 	rcu_string_free(device->name);
693 	kfree(device);
694 }
695 
696 static void free_device(struct rcu_head *head)
697 {
698 	struct btrfs_device *device;
699 
700 	device = container_of(head, struct btrfs_device, rcu);
701 
702 	INIT_WORK(&device->rcu_work, __free_device);
703 	schedule_work(&device->rcu_work);
704 }
705 
706 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
707 {
708 	struct btrfs_device *device;
709 
710 	if (--fs_devices->opened > 0)
711 		return 0;
712 
713 	mutex_lock(&fs_devices->device_list_mutex);
714 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
715 		struct btrfs_device *new_device;
716 		struct rcu_string *name;
717 
718 		if (device->bdev)
719 			fs_devices->open_devices--;
720 
721 		if (device->writeable &&
722 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
723 			list_del_init(&device->dev_alloc_list);
724 			fs_devices->rw_devices--;
725 		}
726 
727 		if (device->missing)
728 			fs_devices->missing_devices--;
729 
730 		new_device = btrfs_alloc_device(NULL, &device->devid,
731 						device->uuid);
732 		BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
733 
734 		/* Safe because we are under uuid_mutex */
735 		if (device->name) {
736 			name = rcu_string_strdup(device->name->str, GFP_NOFS);
737 			BUG_ON(!name); /* -ENOMEM */
738 			rcu_assign_pointer(new_device->name, name);
739 		}
740 
741 		list_replace_rcu(&device->dev_list, &new_device->dev_list);
742 		new_device->fs_devices = device->fs_devices;
743 
744 		call_rcu(&device->rcu, free_device);
745 	}
746 	mutex_unlock(&fs_devices->device_list_mutex);
747 
748 	WARN_ON(fs_devices->open_devices);
749 	WARN_ON(fs_devices->rw_devices);
750 	fs_devices->opened = 0;
751 	fs_devices->seeding = 0;
752 
753 	return 0;
754 }
755 
756 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
757 {
758 	struct btrfs_fs_devices *seed_devices = NULL;
759 	int ret;
760 
761 	mutex_lock(&uuid_mutex);
762 	ret = __btrfs_close_devices(fs_devices);
763 	if (!fs_devices->opened) {
764 		seed_devices = fs_devices->seed;
765 		fs_devices->seed = NULL;
766 	}
767 	mutex_unlock(&uuid_mutex);
768 
769 	while (seed_devices) {
770 		fs_devices = seed_devices;
771 		seed_devices = fs_devices->seed;
772 		__btrfs_close_devices(fs_devices);
773 		free_fs_devices(fs_devices);
774 	}
775 	/*
776 	 * Wait for rcu kworkers under __btrfs_close_devices
777 	 * to finish all blkdev_puts so device is really
778 	 * free when umount is done.
779 	 */
780 	rcu_barrier();
781 	return ret;
782 }
783 
784 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
785 				fmode_t flags, void *holder)
786 {
787 	struct request_queue *q;
788 	struct block_device *bdev;
789 	struct list_head *head = &fs_devices->devices;
790 	struct btrfs_device *device;
791 	struct btrfs_device *latest_dev = NULL;
792 	struct buffer_head *bh;
793 	struct btrfs_super_block *disk_super;
794 	u64 devid;
795 	int seeding = 1;
796 	int ret = 0;
797 
798 	flags |= FMODE_EXCL;
799 
800 	list_for_each_entry(device, head, dev_list) {
801 		if (device->bdev)
802 			continue;
803 		if (!device->name)
804 			continue;
805 
806 		/* Just open everything we can; ignore failures here */
807 		if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
808 					    &bdev, &bh))
809 			continue;
810 
811 		disk_super = (struct btrfs_super_block *)bh->b_data;
812 		devid = btrfs_stack_device_id(&disk_super->dev_item);
813 		if (devid != device->devid)
814 			goto error_brelse;
815 
816 		if (memcmp(device->uuid, disk_super->dev_item.uuid,
817 			   BTRFS_UUID_SIZE))
818 			goto error_brelse;
819 
820 		device->generation = btrfs_super_generation(disk_super);
821 		if (!latest_dev ||
822 		    device->generation > latest_dev->generation)
823 			latest_dev = device;
824 
825 		if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
826 			device->writeable = 0;
827 		} else {
828 			device->writeable = !bdev_read_only(bdev);
829 			seeding = 0;
830 		}
831 
832 		q = bdev_get_queue(bdev);
833 		if (blk_queue_discard(q))
834 			device->can_discard = 1;
835 
836 		device->bdev = bdev;
837 		device->in_fs_metadata = 0;
838 		device->mode = flags;
839 
840 		if (!blk_queue_nonrot(bdev_get_queue(bdev)))
841 			fs_devices->rotating = 1;
842 
843 		fs_devices->open_devices++;
844 		if (device->writeable &&
845 		    device->devid != BTRFS_DEV_REPLACE_DEVID) {
846 			fs_devices->rw_devices++;
847 			list_add(&device->dev_alloc_list,
848 				 &fs_devices->alloc_list);
849 		}
850 		brelse(bh);
851 		continue;
852 
853 error_brelse:
854 		brelse(bh);
855 		blkdev_put(bdev, flags);
856 		continue;
857 	}
858 	if (fs_devices->open_devices == 0) {
859 		ret = -EINVAL;
860 		goto out;
861 	}
862 	fs_devices->seeding = seeding;
863 	fs_devices->opened = 1;
864 	fs_devices->latest_bdev = latest_dev->bdev;
865 	fs_devices->total_rw_bytes = 0;
866 out:
867 	return ret;
868 }
869 
870 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
871 		       fmode_t flags, void *holder)
872 {
873 	int ret;
874 
875 	mutex_lock(&uuid_mutex);
876 	if (fs_devices->opened) {
877 		fs_devices->opened++;
878 		ret = 0;
879 	} else {
880 		ret = __btrfs_open_devices(fs_devices, flags, holder);
881 	}
882 	mutex_unlock(&uuid_mutex);
883 	return ret;
884 }
885 
886 /*
887  * Look for a btrfs signature on a device. This may be called out of the mount path
888  * and we are not allowed to call set_blocksize during the scan. The superblock
889  * is read via pagecache
890  */
891 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
892 			  struct btrfs_fs_devices **fs_devices_ret)
893 {
894 	struct btrfs_super_block *disk_super;
895 	struct block_device *bdev;
896 	struct page *page;
897 	void *p;
898 	int ret = -EINVAL;
899 	u64 devid;
900 	u64 transid;
901 	u64 total_devices;
902 	u64 bytenr;
903 	pgoff_t index;
904 
905 	/*
906 	 * we would like to check all the supers, but that would make
907 	 * a btrfs mount succeed after a mkfs from a different FS.
908 	 * So, we need to add a special mount option to scan for
909 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
910 	 */
911 	bytenr = btrfs_sb_offset(0);
912 	flags |= FMODE_EXCL;
913 	mutex_lock(&uuid_mutex);
914 
915 	bdev = blkdev_get_by_path(path, flags, holder);
916 
917 	if (IS_ERR(bdev)) {
918 		ret = PTR_ERR(bdev);
919 		goto error;
920 	}
921 
922 	/* make sure our super fits in the device */
923 	if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
924 		goto error_bdev_put;
925 
926 	/* make sure our super fits in the page */
927 	if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
928 		goto error_bdev_put;
929 
930 	/* make sure our super doesn't straddle pages on disk */
931 	index = bytenr >> PAGE_CACHE_SHIFT;
932 	if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
933 		goto error_bdev_put;
934 
935 	/* pull in the page with our super */
936 	page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
937 				   index, GFP_NOFS);
938 
939 	if (IS_ERR_OR_NULL(page))
940 		goto error_bdev_put;
941 
942 	p = kmap(page);
943 
944 	/* align our pointer to the offset of the super block */
945 	disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
946 
947 	if (btrfs_super_bytenr(disk_super) != bytenr ||
948 	    btrfs_super_magic(disk_super) != BTRFS_MAGIC)
949 		goto error_unmap;
950 
951 	devid = btrfs_stack_device_id(&disk_super->dev_item);
952 	transid = btrfs_super_generation(disk_super);
953 	total_devices = btrfs_super_num_devices(disk_super);
954 
955 	ret = device_list_add(path, disk_super, devid, fs_devices_ret);
956 	if (ret > 0) {
957 		if (disk_super->label[0]) {
958 			if (disk_super->label[BTRFS_LABEL_SIZE - 1])
959 				disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
960 			printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
961 		} else {
962 			printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
963 		}
964 
965 		printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
966 		ret = 0;
967 	}
968 	if (!ret && fs_devices_ret)
969 		(*fs_devices_ret)->total_devices = total_devices;
970 
971 error_unmap:
972 	kunmap(page);
973 	page_cache_release(page);
974 
975 error_bdev_put:
976 	blkdev_put(bdev, flags);
977 error:
978 	mutex_unlock(&uuid_mutex);
979 	return ret;
980 }
981 
982 /* helper to account the used device space in the range */
983 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
984 				   u64 end, u64 *length)
985 {
986 	struct btrfs_key key;
987 	struct btrfs_root *root = device->dev_root;
988 	struct btrfs_dev_extent *dev_extent;
989 	struct btrfs_path *path;
990 	u64 extent_end;
991 	int ret;
992 	int slot;
993 	struct extent_buffer *l;
994 
995 	*length = 0;
996 
997 	if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
998 		return 0;
999 
1000 	path = btrfs_alloc_path();
1001 	if (!path)
1002 		return -ENOMEM;
1003 	path->reada = 2;
1004 
1005 	key.objectid = device->devid;
1006 	key.offset = start;
1007 	key.type = BTRFS_DEV_EXTENT_KEY;
1008 
1009 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1010 	if (ret < 0)
1011 		goto out;
1012 	if (ret > 0) {
1013 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1014 		if (ret < 0)
1015 			goto out;
1016 	}
1017 
1018 	while (1) {
1019 		l = path->nodes[0];
1020 		slot = path->slots[0];
1021 		if (slot >= btrfs_header_nritems(l)) {
1022 			ret = btrfs_next_leaf(root, path);
1023 			if (ret == 0)
1024 				continue;
1025 			if (ret < 0)
1026 				goto out;
1027 
1028 			break;
1029 		}
1030 		btrfs_item_key_to_cpu(l, &key, slot);
1031 
1032 		if (key.objectid < device->devid)
1033 			goto next;
1034 
1035 		if (key.objectid > device->devid)
1036 			break;
1037 
1038 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1039 			goto next;
1040 
1041 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1042 		extent_end = key.offset + btrfs_dev_extent_length(l,
1043 								  dev_extent);
1044 		if (key.offset <= start && extent_end > end) {
1045 			*length = end - start + 1;
1046 			break;
1047 		} else if (key.offset <= start && extent_end > start)
1048 			*length += extent_end - start;
1049 		else if (key.offset > start && extent_end <= end)
1050 			*length += extent_end - key.offset;
1051 		else if (key.offset > start && key.offset <= end) {
1052 			*length += end - key.offset + 1;
1053 			break;
1054 		} else if (key.offset > end)
1055 			break;
1056 
1057 next:
1058 		path->slots[0]++;
1059 	}
1060 	ret = 0;
1061 out:
1062 	btrfs_free_path(path);
1063 	return ret;
1064 }
1065 
1066 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1067 				   struct btrfs_device *device,
1068 				   u64 *start, u64 len)
1069 {
1070 	struct extent_map *em;
1071 	int ret = 0;
1072 
1073 	list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1074 		struct map_lookup *map;
1075 		int i;
1076 
1077 		map = (struct map_lookup *)em->bdev;
1078 		for (i = 0; i < map->num_stripes; i++) {
1079 			if (map->stripes[i].dev != device)
1080 				continue;
1081 			if (map->stripes[i].physical >= *start + len ||
1082 			    map->stripes[i].physical + em->orig_block_len <=
1083 			    *start)
1084 				continue;
1085 			*start = map->stripes[i].physical +
1086 				em->orig_block_len;
1087 			ret = 1;
1088 		}
1089 	}
1090 
1091 	return ret;
1092 }
1093 
1094 
1095 /*
1096  * find_free_dev_extent - find free space in the specified device
1097  * @device:	the device which we search the free space in
1098  * @num_bytes:	the size of the free space that we need
1099  * @start:	store the start of the free space.
1100  * @len:	the size of the free space. that we find, or the size of the max
1101  * 		free space if we don't find suitable free space
1102  *
1103  * this uses a pretty simple search, the expectation is that it is
1104  * called very infrequently and that a given device has a small number
1105  * of extents
1106  *
1107  * @start is used to store the start of the free space if we find. But if we
1108  * don't find suitable free space, it will be used to store the start position
1109  * of the max free space.
1110  *
1111  * @len is used to store the size of the free space that we find.
1112  * But if we don't find suitable free space, it is used to store the size of
1113  * the max free space.
1114  */
1115 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1116 			 struct btrfs_device *device, u64 num_bytes,
1117 			 u64 *start, u64 *len)
1118 {
1119 	struct btrfs_key key;
1120 	struct btrfs_root *root = device->dev_root;
1121 	struct btrfs_dev_extent *dev_extent;
1122 	struct btrfs_path *path;
1123 	u64 hole_size;
1124 	u64 max_hole_start;
1125 	u64 max_hole_size;
1126 	u64 extent_end;
1127 	u64 search_start;
1128 	u64 search_end = device->total_bytes;
1129 	int ret;
1130 	int slot;
1131 	struct extent_buffer *l;
1132 
1133 	/* FIXME use last free of some kind */
1134 
1135 	/* we don't want to overwrite the superblock on the drive,
1136 	 * so we make sure to start at an offset of at least 1MB
1137 	 */
1138 	search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1139 
1140 	path = btrfs_alloc_path();
1141 	if (!path)
1142 		return -ENOMEM;
1143 again:
1144 	max_hole_start = search_start;
1145 	max_hole_size = 0;
1146 	hole_size = 0;
1147 
1148 	if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1149 		ret = -ENOSPC;
1150 		goto out;
1151 	}
1152 
1153 	path->reada = 2;
1154 	path->search_commit_root = 1;
1155 	path->skip_locking = 1;
1156 
1157 	key.objectid = device->devid;
1158 	key.offset = search_start;
1159 	key.type = BTRFS_DEV_EXTENT_KEY;
1160 
1161 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1162 	if (ret < 0)
1163 		goto out;
1164 	if (ret > 0) {
1165 		ret = btrfs_previous_item(root, path, key.objectid, key.type);
1166 		if (ret < 0)
1167 			goto out;
1168 	}
1169 
1170 	while (1) {
1171 		l = path->nodes[0];
1172 		slot = path->slots[0];
1173 		if (slot >= btrfs_header_nritems(l)) {
1174 			ret = btrfs_next_leaf(root, path);
1175 			if (ret == 0)
1176 				continue;
1177 			if (ret < 0)
1178 				goto out;
1179 
1180 			break;
1181 		}
1182 		btrfs_item_key_to_cpu(l, &key, slot);
1183 
1184 		if (key.objectid < device->devid)
1185 			goto next;
1186 
1187 		if (key.objectid > device->devid)
1188 			break;
1189 
1190 		if (key.type != BTRFS_DEV_EXTENT_KEY)
1191 			goto next;
1192 
1193 		if (key.offset > search_start) {
1194 			hole_size = key.offset - search_start;
1195 
1196 			/*
1197 			 * Have to check before we set max_hole_start, otherwise
1198 			 * we could end up sending back this offset anyway.
1199 			 */
1200 			if (contains_pending_extent(trans, device,
1201 						    &search_start,
1202 						    hole_size))
1203 				hole_size = 0;
1204 
1205 			if (hole_size > max_hole_size) {
1206 				max_hole_start = search_start;
1207 				max_hole_size = hole_size;
1208 			}
1209 
1210 			/*
1211 			 * If this free space is greater than which we need,
1212 			 * it must be the max free space that we have found
1213 			 * until now, so max_hole_start must point to the start
1214 			 * of this free space and the length of this free space
1215 			 * is stored in max_hole_size. Thus, we return
1216 			 * max_hole_start and max_hole_size and go back to the
1217 			 * caller.
1218 			 */
1219 			if (hole_size >= num_bytes) {
1220 				ret = 0;
1221 				goto out;
1222 			}
1223 		}
1224 
1225 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1226 		extent_end = key.offset + btrfs_dev_extent_length(l,
1227 								  dev_extent);
1228 		if (extent_end > search_start)
1229 			search_start = extent_end;
1230 next:
1231 		path->slots[0]++;
1232 		cond_resched();
1233 	}
1234 
1235 	/*
1236 	 * At this point, search_start should be the end of
1237 	 * allocated dev extents, and when shrinking the device,
1238 	 * search_end may be smaller than search_start.
1239 	 */
1240 	if (search_end > search_start)
1241 		hole_size = search_end - search_start;
1242 
1243 	if (hole_size > max_hole_size) {
1244 		max_hole_start = search_start;
1245 		max_hole_size = hole_size;
1246 	}
1247 
1248 	if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1249 		btrfs_release_path(path);
1250 		goto again;
1251 	}
1252 
1253 	/* See above. */
1254 	if (hole_size < num_bytes)
1255 		ret = -ENOSPC;
1256 	else
1257 		ret = 0;
1258 
1259 out:
1260 	btrfs_free_path(path);
1261 	*start = max_hole_start;
1262 	if (len)
1263 		*len = max_hole_size;
1264 	return ret;
1265 }
1266 
1267 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1268 			  struct btrfs_device *device,
1269 			  u64 start, u64 *dev_extent_len)
1270 {
1271 	int ret;
1272 	struct btrfs_path *path;
1273 	struct btrfs_root *root = device->dev_root;
1274 	struct btrfs_key key;
1275 	struct btrfs_key found_key;
1276 	struct extent_buffer *leaf = NULL;
1277 	struct btrfs_dev_extent *extent = NULL;
1278 
1279 	path = btrfs_alloc_path();
1280 	if (!path)
1281 		return -ENOMEM;
1282 
1283 	key.objectid = device->devid;
1284 	key.offset = start;
1285 	key.type = BTRFS_DEV_EXTENT_KEY;
1286 again:
1287 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1288 	if (ret > 0) {
1289 		ret = btrfs_previous_item(root, path, key.objectid,
1290 					  BTRFS_DEV_EXTENT_KEY);
1291 		if (ret)
1292 			goto out;
1293 		leaf = path->nodes[0];
1294 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1295 		extent = btrfs_item_ptr(leaf, path->slots[0],
1296 					struct btrfs_dev_extent);
1297 		BUG_ON(found_key.offset > start || found_key.offset +
1298 		       btrfs_dev_extent_length(leaf, extent) < start);
1299 		key = found_key;
1300 		btrfs_release_path(path);
1301 		goto again;
1302 	} else if (ret == 0) {
1303 		leaf = path->nodes[0];
1304 		extent = btrfs_item_ptr(leaf, path->slots[0],
1305 					struct btrfs_dev_extent);
1306 	} else {
1307 		btrfs_error(root->fs_info, ret, "Slot search failed");
1308 		goto out;
1309 	}
1310 
1311 	*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1312 
1313 	ret = btrfs_del_item(trans, root, path);
1314 	if (ret) {
1315 		btrfs_error(root->fs_info, ret,
1316 			    "Failed to remove dev extent item");
1317 	}
1318 out:
1319 	btrfs_free_path(path);
1320 	return ret;
1321 }
1322 
1323 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1324 				  struct btrfs_device *device,
1325 				  u64 chunk_tree, u64 chunk_objectid,
1326 				  u64 chunk_offset, u64 start, u64 num_bytes)
1327 {
1328 	int ret;
1329 	struct btrfs_path *path;
1330 	struct btrfs_root *root = device->dev_root;
1331 	struct btrfs_dev_extent *extent;
1332 	struct extent_buffer *leaf;
1333 	struct btrfs_key key;
1334 
1335 	WARN_ON(!device->in_fs_metadata);
1336 	WARN_ON(device->is_tgtdev_for_dev_replace);
1337 	path = btrfs_alloc_path();
1338 	if (!path)
1339 		return -ENOMEM;
1340 
1341 	key.objectid = device->devid;
1342 	key.offset = start;
1343 	key.type = BTRFS_DEV_EXTENT_KEY;
1344 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1345 				      sizeof(*extent));
1346 	if (ret)
1347 		goto out;
1348 
1349 	leaf = path->nodes[0];
1350 	extent = btrfs_item_ptr(leaf, path->slots[0],
1351 				struct btrfs_dev_extent);
1352 	btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1353 	btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1354 	btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1355 
1356 	write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1357 		    btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1358 
1359 	btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1360 	btrfs_mark_buffer_dirty(leaf);
1361 out:
1362 	btrfs_free_path(path);
1363 	return ret;
1364 }
1365 
1366 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1367 {
1368 	struct extent_map_tree *em_tree;
1369 	struct extent_map *em;
1370 	struct rb_node *n;
1371 	u64 ret = 0;
1372 
1373 	em_tree = &fs_info->mapping_tree.map_tree;
1374 	read_lock(&em_tree->lock);
1375 	n = rb_last(&em_tree->map);
1376 	if (n) {
1377 		em = rb_entry(n, struct extent_map, rb_node);
1378 		ret = em->start + em->len;
1379 	}
1380 	read_unlock(&em_tree->lock);
1381 
1382 	return ret;
1383 }
1384 
1385 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1386 				    u64 *devid_ret)
1387 {
1388 	int ret;
1389 	struct btrfs_key key;
1390 	struct btrfs_key found_key;
1391 	struct btrfs_path *path;
1392 
1393 	path = btrfs_alloc_path();
1394 	if (!path)
1395 		return -ENOMEM;
1396 
1397 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1398 	key.type = BTRFS_DEV_ITEM_KEY;
1399 	key.offset = (u64)-1;
1400 
1401 	ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1402 	if (ret < 0)
1403 		goto error;
1404 
1405 	BUG_ON(ret == 0); /* Corruption */
1406 
1407 	ret = btrfs_previous_item(fs_info->chunk_root, path,
1408 				  BTRFS_DEV_ITEMS_OBJECTID,
1409 				  BTRFS_DEV_ITEM_KEY);
1410 	if (ret) {
1411 		*devid_ret = 1;
1412 	} else {
1413 		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1414 				      path->slots[0]);
1415 		*devid_ret = found_key.offset + 1;
1416 	}
1417 	ret = 0;
1418 error:
1419 	btrfs_free_path(path);
1420 	return ret;
1421 }
1422 
1423 /*
1424  * the device information is stored in the chunk root
1425  * the btrfs_device struct should be fully filled in
1426  */
1427 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1428 			    struct btrfs_root *root,
1429 			    struct btrfs_device *device)
1430 {
1431 	int ret;
1432 	struct btrfs_path *path;
1433 	struct btrfs_dev_item *dev_item;
1434 	struct extent_buffer *leaf;
1435 	struct btrfs_key key;
1436 	unsigned long ptr;
1437 
1438 	root = root->fs_info->chunk_root;
1439 
1440 	path = btrfs_alloc_path();
1441 	if (!path)
1442 		return -ENOMEM;
1443 
1444 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1445 	key.type = BTRFS_DEV_ITEM_KEY;
1446 	key.offset = device->devid;
1447 
1448 	ret = btrfs_insert_empty_item(trans, root, path, &key,
1449 				      sizeof(*dev_item));
1450 	if (ret)
1451 		goto out;
1452 
1453 	leaf = path->nodes[0];
1454 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1455 
1456 	btrfs_set_device_id(leaf, dev_item, device->devid);
1457 	btrfs_set_device_generation(leaf, dev_item, 0);
1458 	btrfs_set_device_type(leaf, dev_item, device->type);
1459 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1460 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1461 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1462 	btrfs_set_device_total_bytes(leaf, dev_item,
1463 				     btrfs_device_get_disk_total_bytes(device));
1464 	btrfs_set_device_bytes_used(leaf, dev_item,
1465 				    btrfs_device_get_bytes_used(device));
1466 	btrfs_set_device_group(leaf, dev_item, 0);
1467 	btrfs_set_device_seek_speed(leaf, dev_item, 0);
1468 	btrfs_set_device_bandwidth(leaf, dev_item, 0);
1469 	btrfs_set_device_start_offset(leaf, dev_item, 0);
1470 
1471 	ptr = btrfs_device_uuid(dev_item);
1472 	write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1473 	ptr = btrfs_device_fsid(dev_item);
1474 	write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1475 	btrfs_mark_buffer_dirty(leaf);
1476 
1477 	ret = 0;
1478 out:
1479 	btrfs_free_path(path);
1480 	return ret;
1481 }
1482 
1483 /*
1484  * Function to update ctime/mtime for a given device path.
1485  * Mainly used for ctime/mtime based probe like libblkid.
1486  */
1487 static void update_dev_time(char *path_name)
1488 {
1489 	struct file *filp;
1490 
1491 	filp = filp_open(path_name, O_RDWR, 0);
1492 	if (!filp)
1493 		return;
1494 	file_update_time(filp);
1495 	filp_close(filp, NULL);
1496 	return;
1497 }
1498 
1499 static int btrfs_rm_dev_item(struct btrfs_root *root,
1500 			     struct btrfs_device *device)
1501 {
1502 	int ret;
1503 	struct btrfs_path *path;
1504 	struct btrfs_key key;
1505 	struct btrfs_trans_handle *trans;
1506 
1507 	root = root->fs_info->chunk_root;
1508 
1509 	path = btrfs_alloc_path();
1510 	if (!path)
1511 		return -ENOMEM;
1512 
1513 	trans = btrfs_start_transaction(root, 0);
1514 	if (IS_ERR(trans)) {
1515 		btrfs_free_path(path);
1516 		return PTR_ERR(trans);
1517 	}
1518 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1519 	key.type = BTRFS_DEV_ITEM_KEY;
1520 	key.offset = device->devid;
1521 
1522 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1523 	if (ret < 0)
1524 		goto out;
1525 
1526 	if (ret > 0) {
1527 		ret = -ENOENT;
1528 		goto out;
1529 	}
1530 
1531 	ret = btrfs_del_item(trans, root, path);
1532 	if (ret)
1533 		goto out;
1534 out:
1535 	btrfs_free_path(path);
1536 	btrfs_commit_transaction(trans, root);
1537 	return ret;
1538 }
1539 
1540 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1541 {
1542 	struct btrfs_device *device;
1543 	struct btrfs_device *next_device;
1544 	struct block_device *bdev;
1545 	struct buffer_head *bh = NULL;
1546 	struct btrfs_super_block *disk_super;
1547 	struct btrfs_fs_devices *cur_devices;
1548 	u64 all_avail;
1549 	u64 devid;
1550 	u64 num_devices;
1551 	u8 *dev_uuid;
1552 	unsigned seq;
1553 	int ret = 0;
1554 	bool clear_super = false;
1555 
1556 	mutex_lock(&uuid_mutex);
1557 
1558 	do {
1559 		seq = read_seqbegin(&root->fs_info->profiles_lock);
1560 
1561 		all_avail = root->fs_info->avail_data_alloc_bits |
1562 			    root->fs_info->avail_system_alloc_bits |
1563 			    root->fs_info->avail_metadata_alloc_bits;
1564 	} while (read_seqretry(&root->fs_info->profiles_lock, seq));
1565 
1566 	num_devices = root->fs_info->fs_devices->num_devices;
1567 	btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1568 	if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1569 		WARN_ON(num_devices < 1);
1570 		num_devices--;
1571 	}
1572 	btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1573 
1574 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1575 		ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1576 		goto out;
1577 	}
1578 
1579 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1580 		ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1581 		goto out;
1582 	}
1583 
1584 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1585 	    root->fs_info->fs_devices->rw_devices <= 2) {
1586 		ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1587 		goto out;
1588 	}
1589 	if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1590 	    root->fs_info->fs_devices->rw_devices <= 3) {
1591 		ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1592 		goto out;
1593 	}
1594 
1595 	if (strcmp(device_path, "missing") == 0) {
1596 		struct list_head *devices;
1597 		struct btrfs_device *tmp;
1598 
1599 		device = NULL;
1600 		devices = &root->fs_info->fs_devices->devices;
1601 		/*
1602 		 * It is safe to read the devices since the volume_mutex
1603 		 * is held.
1604 		 */
1605 		list_for_each_entry(tmp, devices, dev_list) {
1606 			if (tmp->in_fs_metadata &&
1607 			    !tmp->is_tgtdev_for_dev_replace &&
1608 			    !tmp->bdev) {
1609 				device = tmp;
1610 				break;
1611 			}
1612 		}
1613 		bdev = NULL;
1614 		bh = NULL;
1615 		disk_super = NULL;
1616 		if (!device) {
1617 			ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1618 			goto out;
1619 		}
1620 	} else {
1621 		ret = btrfs_get_bdev_and_sb(device_path,
1622 					    FMODE_WRITE | FMODE_EXCL,
1623 					    root->fs_info->bdev_holder, 0,
1624 					    &bdev, &bh);
1625 		if (ret)
1626 			goto out;
1627 		disk_super = (struct btrfs_super_block *)bh->b_data;
1628 		devid = btrfs_stack_device_id(&disk_super->dev_item);
1629 		dev_uuid = disk_super->dev_item.uuid;
1630 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1631 					   disk_super->fsid);
1632 		if (!device) {
1633 			ret = -ENOENT;
1634 			goto error_brelse;
1635 		}
1636 	}
1637 
1638 	if (device->is_tgtdev_for_dev_replace) {
1639 		ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1640 		goto error_brelse;
1641 	}
1642 
1643 	if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1644 		ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1645 		goto error_brelse;
1646 	}
1647 
1648 	if (device->writeable) {
1649 		lock_chunks(root);
1650 		list_del_init(&device->dev_alloc_list);
1651 		device->fs_devices->rw_devices--;
1652 		unlock_chunks(root);
1653 		clear_super = true;
1654 	}
1655 
1656 	mutex_unlock(&uuid_mutex);
1657 	ret = btrfs_shrink_device(device, 0);
1658 	mutex_lock(&uuid_mutex);
1659 	if (ret)
1660 		goto error_undo;
1661 
1662 	/*
1663 	 * TODO: the superblock still includes this device in its num_devices
1664 	 * counter although write_all_supers() is not locked out. This
1665 	 * could give a filesystem state which requires a degraded mount.
1666 	 */
1667 	ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1668 	if (ret)
1669 		goto error_undo;
1670 
1671 	device->in_fs_metadata = 0;
1672 	btrfs_scrub_cancel_dev(root->fs_info, device);
1673 
1674 	/*
1675 	 * the device list mutex makes sure that we don't change
1676 	 * the device list while someone else is writing out all
1677 	 * the device supers. Whoever is writing all supers, should
1678 	 * lock the device list mutex before getting the number of
1679 	 * devices in the super block (super_copy). Conversely,
1680 	 * whoever updates the number of devices in the super block
1681 	 * (super_copy) should hold the device list mutex.
1682 	 */
1683 
1684 	cur_devices = device->fs_devices;
1685 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1686 	list_del_rcu(&device->dev_list);
1687 
1688 	device->fs_devices->num_devices--;
1689 	device->fs_devices->total_devices--;
1690 
1691 	if (device->missing)
1692 		device->fs_devices->missing_devices--;
1693 
1694 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
1695 				 struct btrfs_device, dev_list);
1696 	if (device->bdev == root->fs_info->sb->s_bdev)
1697 		root->fs_info->sb->s_bdev = next_device->bdev;
1698 	if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1699 		root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1700 
1701 	if (device->bdev) {
1702 		device->fs_devices->open_devices--;
1703 		/* remove sysfs entry */
1704 		btrfs_kobj_rm_device(root->fs_info, device);
1705 	}
1706 
1707 	call_rcu(&device->rcu, free_device);
1708 
1709 	num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1710 	btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1711 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1712 
1713 	if (cur_devices->open_devices == 0) {
1714 		struct btrfs_fs_devices *fs_devices;
1715 		fs_devices = root->fs_info->fs_devices;
1716 		while (fs_devices) {
1717 			if (fs_devices->seed == cur_devices) {
1718 				fs_devices->seed = cur_devices->seed;
1719 				break;
1720 			}
1721 			fs_devices = fs_devices->seed;
1722 		}
1723 		cur_devices->seed = NULL;
1724 		__btrfs_close_devices(cur_devices);
1725 		free_fs_devices(cur_devices);
1726 	}
1727 
1728 	root->fs_info->num_tolerated_disk_barrier_failures =
1729 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1730 
1731 	/*
1732 	 * at this point, the device is zero sized.  We want to
1733 	 * remove it from the devices list and zero out the old super
1734 	 */
1735 	if (clear_super && disk_super) {
1736 		u64 bytenr;
1737 		int i;
1738 
1739 		/* make sure this device isn't detected as part of
1740 		 * the FS anymore
1741 		 */
1742 		memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1743 		set_buffer_dirty(bh);
1744 		sync_dirty_buffer(bh);
1745 
1746 		/* clear the mirror copies of super block on the disk
1747 		 * being removed, 0th copy is been taken care above and
1748 		 * the below would take of the rest
1749 		 */
1750 		for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1751 			bytenr = btrfs_sb_offset(i);
1752 			if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1753 					i_size_read(bdev->bd_inode))
1754 				break;
1755 
1756 			brelse(bh);
1757 			bh = __bread(bdev, bytenr / 4096,
1758 					BTRFS_SUPER_INFO_SIZE);
1759 			if (!bh)
1760 				continue;
1761 
1762 			disk_super = (struct btrfs_super_block *)bh->b_data;
1763 
1764 			if (btrfs_super_bytenr(disk_super) != bytenr ||
1765 				btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1766 				continue;
1767 			}
1768 			memset(&disk_super->magic, 0,
1769 						sizeof(disk_super->magic));
1770 			set_buffer_dirty(bh);
1771 			sync_dirty_buffer(bh);
1772 		}
1773 	}
1774 
1775 	ret = 0;
1776 
1777 	if (bdev) {
1778 		/* Notify udev that device has changed */
1779 		btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1780 
1781 		/* Update ctime/mtime for device path for libblkid */
1782 		update_dev_time(device_path);
1783 	}
1784 
1785 error_brelse:
1786 	brelse(bh);
1787 	if (bdev)
1788 		blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1789 out:
1790 	mutex_unlock(&uuid_mutex);
1791 	return ret;
1792 error_undo:
1793 	if (device->writeable) {
1794 		lock_chunks(root);
1795 		list_add(&device->dev_alloc_list,
1796 			 &root->fs_info->fs_devices->alloc_list);
1797 		device->fs_devices->rw_devices++;
1798 		unlock_chunks(root);
1799 	}
1800 	goto error_brelse;
1801 }
1802 
1803 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1804 				 struct btrfs_device *srcdev)
1805 {
1806 	struct btrfs_fs_devices *fs_devices;
1807 
1808 	WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1809 
1810 	/*
1811 	 * in case of fs with no seed, srcdev->fs_devices will point
1812 	 * to fs_devices of fs_info. However when the dev being replaced is
1813 	 * a seed dev it will point to the seed's local fs_devices. In short
1814 	 * srcdev will have its correct fs_devices in both the cases.
1815 	 */
1816 	fs_devices = srcdev->fs_devices;
1817 
1818 	list_del_rcu(&srcdev->dev_list);
1819 	list_del_rcu(&srcdev->dev_alloc_list);
1820 	fs_devices->num_devices--;
1821 	if (srcdev->missing)
1822 		fs_devices->missing_devices--;
1823 
1824 	if (srcdev->writeable) {
1825 		fs_devices->rw_devices--;
1826 		/* zero out the old super if it is writable */
1827 		btrfs_scratch_superblock(srcdev);
1828 	}
1829 
1830 	if (srcdev->bdev)
1831 		fs_devices->open_devices--;
1832 
1833 	call_rcu(&srcdev->rcu, free_device);
1834 
1835 	/*
1836 	 * unless fs_devices is seed fs, num_devices shouldn't go
1837 	 * zero
1838 	 */
1839 	BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1840 
1841 	/* if this is no devs we rather delete the fs_devices */
1842 	if (!fs_devices->num_devices) {
1843 		struct btrfs_fs_devices *tmp_fs_devices;
1844 
1845 		tmp_fs_devices = fs_info->fs_devices;
1846 		while (tmp_fs_devices) {
1847 			if (tmp_fs_devices->seed == fs_devices) {
1848 				tmp_fs_devices->seed = fs_devices->seed;
1849 				break;
1850 			}
1851 			tmp_fs_devices = tmp_fs_devices->seed;
1852 		}
1853 		fs_devices->seed = NULL;
1854 		__btrfs_close_devices(fs_devices);
1855 		free_fs_devices(fs_devices);
1856 	}
1857 }
1858 
1859 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1860 				      struct btrfs_device *tgtdev)
1861 {
1862 	struct btrfs_device *next_device;
1863 
1864 	mutex_lock(&uuid_mutex);
1865 	WARN_ON(!tgtdev);
1866 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
1867 	if (tgtdev->bdev) {
1868 		btrfs_scratch_superblock(tgtdev);
1869 		fs_info->fs_devices->open_devices--;
1870 	}
1871 	fs_info->fs_devices->num_devices--;
1872 
1873 	next_device = list_entry(fs_info->fs_devices->devices.next,
1874 				 struct btrfs_device, dev_list);
1875 	if (tgtdev->bdev == fs_info->sb->s_bdev)
1876 		fs_info->sb->s_bdev = next_device->bdev;
1877 	if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1878 		fs_info->fs_devices->latest_bdev = next_device->bdev;
1879 	list_del_rcu(&tgtdev->dev_list);
1880 
1881 	call_rcu(&tgtdev->rcu, free_device);
1882 
1883 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1884 	mutex_unlock(&uuid_mutex);
1885 }
1886 
1887 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1888 				     struct btrfs_device **device)
1889 {
1890 	int ret = 0;
1891 	struct btrfs_super_block *disk_super;
1892 	u64 devid;
1893 	u8 *dev_uuid;
1894 	struct block_device *bdev;
1895 	struct buffer_head *bh;
1896 
1897 	*device = NULL;
1898 	ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1899 				    root->fs_info->bdev_holder, 0, &bdev, &bh);
1900 	if (ret)
1901 		return ret;
1902 	disk_super = (struct btrfs_super_block *)bh->b_data;
1903 	devid = btrfs_stack_device_id(&disk_super->dev_item);
1904 	dev_uuid = disk_super->dev_item.uuid;
1905 	*device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1906 				    disk_super->fsid);
1907 	brelse(bh);
1908 	if (!*device)
1909 		ret = -ENOENT;
1910 	blkdev_put(bdev, FMODE_READ);
1911 	return ret;
1912 }
1913 
1914 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1915 					 char *device_path,
1916 					 struct btrfs_device **device)
1917 {
1918 	*device = NULL;
1919 	if (strcmp(device_path, "missing") == 0) {
1920 		struct list_head *devices;
1921 		struct btrfs_device *tmp;
1922 
1923 		devices = &root->fs_info->fs_devices->devices;
1924 		/*
1925 		 * It is safe to read the devices since the volume_mutex
1926 		 * is held by the caller.
1927 		 */
1928 		list_for_each_entry(tmp, devices, dev_list) {
1929 			if (tmp->in_fs_metadata && !tmp->bdev) {
1930 				*device = tmp;
1931 				break;
1932 			}
1933 		}
1934 
1935 		if (!*device) {
1936 			btrfs_err(root->fs_info, "no missing device found");
1937 			return -ENOENT;
1938 		}
1939 
1940 		return 0;
1941 	} else {
1942 		return btrfs_find_device_by_path(root, device_path, device);
1943 	}
1944 }
1945 
1946 /*
1947  * does all the dirty work required for changing file system's UUID.
1948  */
1949 static int btrfs_prepare_sprout(struct btrfs_root *root)
1950 {
1951 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1952 	struct btrfs_fs_devices *old_devices;
1953 	struct btrfs_fs_devices *seed_devices;
1954 	struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1955 	struct btrfs_device *device;
1956 	u64 super_flags;
1957 
1958 	BUG_ON(!mutex_is_locked(&uuid_mutex));
1959 	if (!fs_devices->seeding)
1960 		return -EINVAL;
1961 
1962 	seed_devices = __alloc_fs_devices();
1963 	if (IS_ERR(seed_devices))
1964 		return PTR_ERR(seed_devices);
1965 
1966 	old_devices = clone_fs_devices(fs_devices);
1967 	if (IS_ERR(old_devices)) {
1968 		kfree(seed_devices);
1969 		return PTR_ERR(old_devices);
1970 	}
1971 
1972 	list_add(&old_devices->list, &fs_uuids);
1973 
1974 	memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1975 	seed_devices->opened = 1;
1976 	INIT_LIST_HEAD(&seed_devices->devices);
1977 	INIT_LIST_HEAD(&seed_devices->alloc_list);
1978 	mutex_init(&seed_devices->device_list_mutex);
1979 
1980 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1981 	list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1982 			      synchronize_rcu);
1983 	list_for_each_entry(device, &seed_devices->devices, dev_list)
1984 		device->fs_devices = seed_devices;
1985 
1986 	lock_chunks(root);
1987 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1988 	unlock_chunks(root);
1989 
1990 	fs_devices->seeding = 0;
1991 	fs_devices->num_devices = 0;
1992 	fs_devices->open_devices = 0;
1993 	fs_devices->missing_devices = 0;
1994 	fs_devices->rotating = 0;
1995 	fs_devices->seed = seed_devices;
1996 
1997 	generate_random_uuid(fs_devices->fsid);
1998 	memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1999 	memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2000 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2001 
2002 	super_flags = btrfs_super_flags(disk_super) &
2003 		      ~BTRFS_SUPER_FLAG_SEEDING;
2004 	btrfs_set_super_flags(disk_super, super_flags);
2005 
2006 	return 0;
2007 }
2008 
2009 /*
2010  * strore the expected generation for seed devices in device items.
2011  */
2012 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2013 			       struct btrfs_root *root)
2014 {
2015 	struct btrfs_path *path;
2016 	struct extent_buffer *leaf;
2017 	struct btrfs_dev_item *dev_item;
2018 	struct btrfs_device *device;
2019 	struct btrfs_key key;
2020 	u8 fs_uuid[BTRFS_UUID_SIZE];
2021 	u8 dev_uuid[BTRFS_UUID_SIZE];
2022 	u64 devid;
2023 	int ret;
2024 
2025 	path = btrfs_alloc_path();
2026 	if (!path)
2027 		return -ENOMEM;
2028 
2029 	root = root->fs_info->chunk_root;
2030 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2031 	key.offset = 0;
2032 	key.type = BTRFS_DEV_ITEM_KEY;
2033 
2034 	while (1) {
2035 		ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2036 		if (ret < 0)
2037 			goto error;
2038 
2039 		leaf = path->nodes[0];
2040 next_slot:
2041 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2042 			ret = btrfs_next_leaf(root, path);
2043 			if (ret > 0)
2044 				break;
2045 			if (ret < 0)
2046 				goto error;
2047 			leaf = path->nodes[0];
2048 			btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2049 			btrfs_release_path(path);
2050 			continue;
2051 		}
2052 
2053 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2054 		if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2055 		    key.type != BTRFS_DEV_ITEM_KEY)
2056 			break;
2057 
2058 		dev_item = btrfs_item_ptr(leaf, path->slots[0],
2059 					  struct btrfs_dev_item);
2060 		devid = btrfs_device_id(leaf, dev_item);
2061 		read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2062 				   BTRFS_UUID_SIZE);
2063 		read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2064 				   BTRFS_UUID_SIZE);
2065 		device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2066 					   fs_uuid);
2067 		BUG_ON(!device); /* Logic error */
2068 
2069 		if (device->fs_devices->seeding) {
2070 			btrfs_set_device_generation(leaf, dev_item,
2071 						    device->generation);
2072 			btrfs_mark_buffer_dirty(leaf);
2073 		}
2074 
2075 		path->slots[0]++;
2076 		goto next_slot;
2077 	}
2078 	ret = 0;
2079 error:
2080 	btrfs_free_path(path);
2081 	return ret;
2082 }
2083 
2084 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2085 {
2086 	struct request_queue *q;
2087 	struct btrfs_trans_handle *trans;
2088 	struct btrfs_device *device;
2089 	struct block_device *bdev;
2090 	struct list_head *devices;
2091 	struct super_block *sb = root->fs_info->sb;
2092 	struct rcu_string *name;
2093 	u64 tmp;
2094 	int seeding_dev = 0;
2095 	int ret = 0;
2096 
2097 	if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2098 		return -EROFS;
2099 
2100 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2101 				  root->fs_info->bdev_holder);
2102 	if (IS_ERR(bdev))
2103 		return PTR_ERR(bdev);
2104 
2105 	if (root->fs_info->fs_devices->seeding) {
2106 		seeding_dev = 1;
2107 		down_write(&sb->s_umount);
2108 		mutex_lock(&uuid_mutex);
2109 	}
2110 
2111 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2112 
2113 	devices = &root->fs_info->fs_devices->devices;
2114 
2115 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2116 	list_for_each_entry(device, devices, dev_list) {
2117 		if (device->bdev == bdev) {
2118 			ret = -EEXIST;
2119 			mutex_unlock(
2120 				&root->fs_info->fs_devices->device_list_mutex);
2121 			goto error;
2122 		}
2123 	}
2124 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2125 
2126 	device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2127 	if (IS_ERR(device)) {
2128 		/* we can safely leave the fs_devices entry around */
2129 		ret = PTR_ERR(device);
2130 		goto error;
2131 	}
2132 
2133 	name = rcu_string_strdup(device_path, GFP_NOFS);
2134 	if (!name) {
2135 		kfree(device);
2136 		ret = -ENOMEM;
2137 		goto error;
2138 	}
2139 	rcu_assign_pointer(device->name, name);
2140 
2141 	trans = btrfs_start_transaction(root, 0);
2142 	if (IS_ERR(trans)) {
2143 		rcu_string_free(device->name);
2144 		kfree(device);
2145 		ret = PTR_ERR(trans);
2146 		goto error;
2147 	}
2148 
2149 	q = bdev_get_queue(bdev);
2150 	if (blk_queue_discard(q))
2151 		device->can_discard = 1;
2152 	device->writeable = 1;
2153 	device->generation = trans->transid;
2154 	device->io_width = root->sectorsize;
2155 	device->io_align = root->sectorsize;
2156 	device->sector_size = root->sectorsize;
2157 	device->total_bytes = i_size_read(bdev->bd_inode);
2158 	device->disk_total_bytes = device->total_bytes;
2159 	device->commit_total_bytes = device->total_bytes;
2160 	device->dev_root = root->fs_info->dev_root;
2161 	device->bdev = bdev;
2162 	device->in_fs_metadata = 1;
2163 	device->is_tgtdev_for_dev_replace = 0;
2164 	device->mode = FMODE_EXCL;
2165 	device->dev_stats_valid = 1;
2166 	set_blocksize(device->bdev, 4096);
2167 
2168 	if (seeding_dev) {
2169 		sb->s_flags &= ~MS_RDONLY;
2170 		ret = btrfs_prepare_sprout(root);
2171 		BUG_ON(ret); /* -ENOMEM */
2172 	}
2173 
2174 	device->fs_devices = root->fs_info->fs_devices;
2175 
2176 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2177 	lock_chunks(root);
2178 	list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2179 	list_add(&device->dev_alloc_list,
2180 		 &root->fs_info->fs_devices->alloc_list);
2181 	root->fs_info->fs_devices->num_devices++;
2182 	root->fs_info->fs_devices->open_devices++;
2183 	root->fs_info->fs_devices->rw_devices++;
2184 	root->fs_info->fs_devices->total_devices++;
2185 	root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2186 
2187 	spin_lock(&root->fs_info->free_chunk_lock);
2188 	root->fs_info->free_chunk_space += device->total_bytes;
2189 	spin_unlock(&root->fs_info->free_chunk_lock);
2190 
2191 	if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2192 		root->fs_info->fs_devices->rotating = 1;
2193 
2194 	tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2195 	btrfs_set_super_total_bytes(root->fs_info->super_copy,
2196 				    tmp + device->total_bytes);
2197 
2198 	tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2199 	btrfs_set_super_num_devices(root->fs_info->super_copy,
2200 				    tmp + 1);
2201 
2202 	/* add sysfs device entry */
2203 	btrfs_kobj_add_device(root->fs_info, device);
2204 
2205 	/*
2206 	 * we've got more storage, clear any full flags on the space
2207 	 * infos
2208 	 */
2209 	btrfs_clear_space_info_full(root->fs_info);
2210 
2211 	unlock_chunks(root);
2212 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2213 
2214 	if (seeding_dev) {
2215 		lock_chunks(root);
2216 		ret = init_first_rw_device(trans, root, device);
2217 		unlock_chunks(root);
2218 		if (ret) {
2219 			btrfs_abort_transaction(trans, root, ret);
2220 			goto error_trans;
2221 		}
2222 	}
2223 
2224 	ret = btrfs_add_device(trans, root, device);
2225 	if (ret) {
2226 		btrfs_abort_transaction(trans, root, ret);
2227 		goto error_trans;
2228 	}
2229 
2230 	if (seeding_dev) {
2231 		char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2232 
2233 		ret = btrfs_finish_sprout(trans, root);
2234 		if (ret) {
2235 			btrfs_abort_transaction(trans, root, ret);
2236 			goto error_trans;
2237 		}
2238 
2239 		/* Sprouting would change fsid of the mounted root,
2240 		 * so rename the fsid on the sysfs
2241 		 */
2242 		snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2243 						root->fs_info->fsid);
2244 		if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2245 			goto error_trans;
2246 	}
2247 
2248 	root->fs_info->num_tolerated_disk_barrier_failures =
2249 		btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2250 	ret = btrfs_commit_transaction(trans, root);
2251 
2252 	if (seeding_dev) {
2253 		mutex_unlock(&uuid_mutex);
2254 		up_write(&sb->s_umount);
2255 
2256 		if (ret) /* transaction commit */
2257 			return ret;
2258 
2259 		ret = btrfs_relocate_sys_chunks(root);
2260 		if (ret < 0)
2261 			btrfs_error(root->fs_info, ret,
2262 				    "Failed to relocate sys chunks after "
2263 				    "device initialization. This can be fixed "
2264 				    "using the \"btrfs balance\" command.");
2265 		trans = btrfs_attach_transaction(root);
2266 		if (IS_ERR(trans)) {
2267 			if (PTR_ERR(trans) == -ENOENT)
2268 				return 0;
2269 			return PTR_ERR(trans);
2270 		}
2271 		ret = btrfs_commit_transaction(trans, root);
2272 	}
2273 
2274 	/* Update ctime/mtime for libblkid */
2275 	update_dev_time(device_path);
2276 	return ret;
2277 
2278 error_trans:
2279 	btrfs_end_transaction(trans, root);
2280 	rcu_string_free(device->name);
2281 	btrfs_kobj_rm_device(root->fs_info, device);
2282 	kfree(device);
2283 error:
2284 	blkdev_put(bdev, FMODE_EXCL);
2285 	if (seeding_dev) {
2286 		mutex_unlock(&uuid_mutex);
2287 		up_write(&sb->s_umount);
2288 	}
2289 	return ret;
2290 }
2291 
2292 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2293 				  struct btrfs_device *srcdev,
2294 				  struct btrfs_device **device_out)
2295 {
2296 	struct request_queue *q;
2297 	struct btrfs_device *device;
2298 	struct block_device *bdev;
2299 	struct btrfs_fs_info *fs_info = root->fs_info;
2300 	struct list_head *devices;
2301 	struct rcu_string *name;
2302 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
2303 	int ret = 0;
2304 
2305 	*device_out = NULL;
2306 	if (fs_info->fs_devices->seeding) {
2307 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2308 		return -EINVAL;
2309 	}
2310 
2311 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2312 				  fs_info->bdev_holder);
2313 	if (IS_ERR(bdev)) {
2314 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
2315 		return PTR_ERR(bdev);
2316 	}
2317 
2318 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
2319 
2320 	devices = &fs_info->fs_devices->devices;
2321 	list_for_each_entry(device, devices, dev_list) {
2322 		if (device->bdev == bdev) {
2323 			btrfs_err(fs_info, "target device is in the filesystem!");
2324 			ret = -EEXIST;
2325 			goto error;
2326 		}
2327 	}
2328 
2329 
2330 	if (i_size_read(bdev->bd_inode) <
2331 	    btrfs_device_get_total_bytes(srcdev)) {
2332 		btrfs_err(fs_info, "target device is smaller than source device!");
2333 		ret = -EINVAL;
2334 		goto error;
2335 	}
2336 
2337 
2338 	device = btrfs_alloc_device(NULL, &devid, NULL);
2339 	if (IS_ERR(device)) {
2340 		ret = PTR_ERR(device);
2341 		goto error;
2342 	}
2343 
2344 	name = rcu_string_strdup(device_path, GFP_NOFS);
2345 	if (!name) {
2346 		kfree(device);
2347 		ret = -ENOMEM;
2348 		goto error;
2349 	}
2350 	rcu_assign_pointer(device->name, name);
2351 
2352 	q = bdev_get_queue(bdev);
2353 	if (blk_queue_discard(q))
2354 		device->can_discard = 1;
2355 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2356 	device->writeable = 1;
2357 	device->generation = 0;
2358 	device->io_width = root->sectorsize;
2359 	device->io_align = root->sectorsize;
2360 	device->sector_size = root->sectorsize;
2361 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2362 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2363 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2364 	ASSERT(list_empty(&srcdev->resized_list));
2365 	device->commit_total_bytes = srcdev->commit_total_bytes;
2366 	device->commit_bytes_used = device->bytes_used;
2367 	device->dev_root = fs_info->dev_root;
2368 	device->bdev = bdev;
2369 	device->in_fs_metadata = 1;
2370 	device->is_tgtdev_for_dev_replace = 1;
2371 	device->mode = FMODE_EXCL;
2372 	device->dev_stats_valid = 1;
2373 	set_blocksize(device->bdev, 4096);
2374 	device->fs_devices = fs_info->fs_devices;
2375 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
2376 	fs_info->fs_devices->num_devices++;
2377 	fs_info->fs_devices->open_devices++;
2378 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2379 
2380 	*device_out = device;
2381 	return ret;
2382 
2383 error:
2384 	blkdev_put(bdev, FMODE_EXCL);
2385 	return ret;
2386 }
2387 
2388 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2389 					      struct btrfs_device *tgtdev)
2390 {
2391 	WARN_ON(fs_info->fs_devices->rw_devices == 0);
2392 	tgtdev->io_width = fs_info->dev_root->sectorsize;
2393 	tgtdev->io_align = fs_info->dev_root->sectorsize;
2394 	tgtdev->sector_size = fs_info->dev_root->sectorsize;
2395 	tgtdev->dev_root = fs_info->dev_root;
2396 	tgtdev->in_fs_metadata = 1;
2397 }
2398 
2399 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2400 					struct btrfs_device *device)
2401 {
2402 	int ret;
2403 	struct btrfs_path *path;
2404 	struct btrfs_root *root;
2405 	struct btrfs_dev_item *dev_item;
2406 	struct extent_buffer *leaf;
2407 	struct btrfs_key key;
2408 
2409 	root = device->dev_root->fs_info->chunk_root;
2410 
2411 	path = btrfs_alloc_path();
2412 	if (!path)
2413 		return -ENOMEM;
2414 
2415 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2416 	key.type = BTRFS_DEV_ITEM_KEY;
2417 	key.offset = device->devid;
2418 
2419 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2420 	if (ret < 0)
2421 		goto out;
2422 
2423 	if (ret > 0) {
2424 		ret = -ENOENT;
2425 		goto out;
2426 	}
2427 
2428 	leaf = path->nodes[0];
2429 	dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2430 
2431 	btrfs_set_device_id(leaf, dev_item, device->devid);
2432 	btrfs_set_device_type(leaf, dev_item, device->type);
2433 	btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2434 	btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2435 	btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2436 	btrfs_set_device_total_bytes(leaf, dev_item,
2437 				     btrfs_device_get_disk_total_bytes(device));
2438 	btrfs_set_device_bytes_used(leaf, dev_item,
2439 				    btrfs_device_get_bytes_used(device));
2440 	btrfs_mark_buffer_dirty(leaf);
2441 
2442 out:
2443 	btrfs_free_path(path);
2444 	return ret;
2445 }
2446 
2447 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2448 		      struct btrfs_device *device, u64 new_size)
2449 {
2450 	struct btrfs_super_block *super_copy =
2451 		device->dev_root->fs_info->super_copy;
2452 	struct btrfs_fs_devices *fs_devices;
2453 	u64 old_total;
2454 	u64 diff;
2455 
2456 	if (!device->writeable)
2457 		return -EACCES;
2458 
2459 	lock_chunks(device->dev_root);
2460 	old_total = btrfs_super_total_bytes(super_copy);
2461 	diff = new_size - device->total_bytes;
2462 
2463 	if (new_size <= device->total_bytes ||
2464 	    device->is_tgtdev_for_dev_replace) {
2465 		unlock_chunks(device->dev_root);
2466 		return -EINVAL;
2467 	}
2468 
2469 	fs_devices = device->dev_root->fs_info->fs_devices;
2470 
2471 	btrfs_set_super_total_bytes(super_copy, old_total + diff);
2472 	device->fs_devices->total_rw_bytes += diff;
2473 
2474 	btrfs_device_set_total_bytes(device, new_size);
2475 	btrfs_device_set_disk_total_bytes(device, new_size);
2476 	btrfs_clear_space_info_full(device->dev_root->fs_info);
2477 	if (list_empty(&device->resized_list))
2478 		list_add_tail(&device->resized_list,
2479 			      &fs_devices->resized_devices);
2480 	unlock_chunks(device->dev_root);
2481 
2482 	return btrfs_update_device(trans, device);
2483 }
2484 
2485 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2486 			    struct btrfs_root *root,
2487 			    u64 chunk_tree, u64 chunk_objectid,
2488 			    u64 chunk_offset)
2489 {
2490 	int ret;
2491 	struct btrfs_path *path;
2492 	struct btrfs_key key;
2493 
2494 	root = root->fs_info->chunk_root;
2495 	path = btrfs_alloc_path();
2496 	if (!path)
2497 		return -ENOMEM;
2498 
2499 	key.objectid = chunk_objectid;
2500 	key.offset = chunk_offset;
2501 	key.type = BTRFS_CHUNK_ITEM_KEY;
2502 
2503 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2504 	if (ret < 0)
2505 		goto out;
2506 	else if (ret > 0) { /* Logic error or corruption */
2507 		btrfs_error(root->fs_info, -ENOENT,
2508 			    "Failed lookup while freeing chunk.");
2509 		ret = -ENOENT;
2510 		goto out;
2511 	}
2512 
2513 	ret = btrfs_del_item(trans, root, path);
2514 	if (ret < 0)
2515 		btrfs_error(root->fs_info, ret,
2516 			    "Failed to delete chunk item.");
2517 out:
2518 	btrfs_free_path(path);
2519 	return ret;
2520 }
2521 
2522 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2523 			chunk_offset)
2524 {
2525 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2526 	struct btrfs_disk_key *disk_key;
2527 	struct btrfs_chunk *chunk;
2528 	u8 *ptr;
2529 	int ret = 0;
2530 	u32 num_stripes;
2531 	u32 array_size;
2532 	u32 len = 0;
2533 	u32 cur;
2534 	struct btrfs_key key;
2535 
2536 	lock_chunks(root);
2537 	array_size = btrfs_super_sys_array_size(super_copy);
2538 
2539 	ptr = super_copy->sys_chunk_array;
2540 	cur = 0;
2541 
2542 	while (cur < array_size) {
2543 		disk_key = (struct btrfs_disk_key *)ptr;
2544 		btrfs_disk_key_to_cpu(&key, disk_key);
2545 
2546 		len = sizeof(*disk_key);
2547 
2548 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2549 			chunk = (struct btrfs_chunk *)(ptr + len);
2550 			num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2551 			len += btrfs_chunk_item_size(num_stripes);
2552 		} else {
2553 			ret = -EIO;
2554 			break;
2555 		}
2556 		if (key.objectid == chunk_objectid &&
2557 		    key.offset == chunk_offset) {
2558 			memmove(ptr, ptr + len, array_size - (cur + len));
2559 			array_size -= len;
2560 			btrfs_set_super_sys_array_size(super_copy, array_size);
2561 		} else {
2562 			ptr += len;
2563 			cur += len;
2564 		}
2565 	}
2566 	unlock_chunks(root);
2567 	return ret;
2568 }
2569 
2570 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2571 		       struct btrfs_root *root, u64 chunk_offset)
2572 {
2573 	struct extent_map_tree *em_tree;
2574 	struct extent_map *em;
2575 	struct btrfs_root *extent_root = root->fs_info->extent_root;
2576 	struct map_lookup *map;
2577 	u64 dev_extent_len = 0;
2578 	u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2579 	u64 chunk_tree = root->fs_info->chunk_root->objectid;
2580 	int i, ret = 0;
2581 
2582 	/* Just in case */
2583 	root = root->fs_info->chunk_root;
2584 	em_tree = &root->fs_info->mapping_tree.map_tree;
2585 
2586 	read_lock(&em_tree->lock);
2587 	em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2588 	read_unlock(&em_tree->lock);
2589 
2590 	if (!em || em->start > chunk_offset ||
2591 	    em->start + em->len < chunk_offset) {
2592 		/*
2593 		 * This is a logic error, but we don't want to just rely on the
2594 		 * user having built with ASSERT enabled, so if ASSERT doens't
2595 		 * do anything we still error out.
2596 		 */
2597 		ASSERT(0);
2598 		if (em)
2599 			free_extent_map(em);
2600 		return -EINVAL;
2601 	}
2602 	map = (struct map_lookup *)em->bdev;
2603 
2604 	for (i = 0; i < map->num_stripes; i++) {
2605 		struct btrfs_device *device = map->stripes[i].dev;
2606 		ret = btrfs_free_dev_extent(trans, device,
2607 					    map->stripes[i].physical,
2608 					    &dev_extent_len);
2609 		if (ret) {
2610 			btrfs_abort_transaction(trans, root, ret);
2611 			goto out;
2612 		}
2613 
2614 		if (device->bytes_used > 0) {
2615 			lock_chunks(root);
2616 			btrfs_device_set_bytes_used(device,
2617 					device->bytes_used - dev_extent_len);
2618 			spin_lock(&root->fs_info->free_chunk_lock);
2619 			root->fs_info->free_chunk_space += dev_extent_len;
2620 			spin_unlock(&root->fs_info->free_chunk_lock);
2621 			btrfs_clear_space_info_full(root->fs_info);
2622 			unlock_chunks(root);
2623 		}
2624 
2625 		if (map->stripes[i].dev) {
2626 			ret = btrfs_update_device(trans, map->stripes[i].dev);
2627 			if (ret) {
2628 				btrfs_abort_transaction(trans, root, ret);
2629 				goto out;
2630 			}
2631 		}
2632 	}
2633 	ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2634 			       chunk_offset);
2635 	if (ret) {
2636 		btrfs_abort_transaction(trans, root, ret);
2637 		goto out;
2638 	}
2639 
2640 	trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2641 
2642 	if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2643 		ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2644 		if (ret) {
2645 			btrfs_abort_transaction(trans, root, ret);
2646 			goto out;
2647 		}
2648 	}
2649 
2650 	ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2651 	if (ret) {
2652 		btrfs_abort_transaction(trans, extent_root, ret);
2653 		goto out;
2654 	}
2655 
2656 	write_lock(&em_tree->lock);
2657 	remove_extent_mapping(em_tree, em);
2658 	write_unlock(&em_tree->lock);
2659 
2660 	/* once for the tree */
2661 	free_extent_map(em);
2662 out:
2663 	/* once for us */
2664 	free_extent_map(em);
2665 	return ret;
2666 }
2667 
2668 static int btrfs_relocate_chunk(struct btrfs_root *root,
2669 			 u64 chunk_tree, u64 chunk_objectid,
2670 			 u64 chunk_offset)
2671 {
2672 	struct btrfs_root *extent_root;
2673 	struct btrfs_trans_handle *trans;
2674 	int ret;
2675 
2676 	root = root->fs_info->chunk_root;
2677 	extent_root = root->fs_info->extent_root;
2678 
2679 	ret = btrfs_can_relocate(extent_root, chunk_offset);
2680 	if (ret)
2681 		return -ENOSPC;
2682 
2683 	/* step one, relocate all the extents inside this chunk */
2684 	ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2685 	if (ret)
2686 		return ret;
2687 
2688 	trans = btrfs_start_transaction(root, 0);
2689 	if (IS_ERR(trans)) {
2690 		ret = PTR_ERR(trans);
2691 		btrfs_std_error(root->fs_info, ret);
2692 		return ret;
2693 	}
2694 
2695 	/*
2696 	 * step two, delete the device extents and the
2697 	 * chunk tree entries
2698 	 */
2699 	ret = btrfs_remove_chunk(trans, root, chunk_offset);
2700 	btrfs_end_transaction(trans, root);
2701 	return ret;
2702 }
2703 
2704 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2705 {
2706 	struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2707 	struct btrfs_path *path;
2708 	struct extent_buffer *leaf;
2709 	struct btrfs_chunk *chunk;
2710 	struct btrfs_key key;
2711 	struct btrfs_key found_key;
2712 	u64 chunk_tree = chunk_root->root_key.objectid;
2713 	u64 chunk_type;
2714 	bool retried = false;
2715 	int failed = 0;
2716 	int ret;
2717 
2718 	path = btrfs_alloc_path();
2719 	if (!path)
2720 		return -ENOMEM;
2721 
2722 again:
2723 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2724 	key.offset = (u64)-1;
2725 	key.type = BTRFS_CHUNK_ITEM_KEY;
2726 
2727 	while (1) {
2728 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2729 		if (ret < 0)
2730 			goto error;
2731 		BUG_ON(ret == 0); /* Corruption */
2732 
2733 		ret = btrfs_previous_item(chunk_root, path, key.objectid,
2734 					  key.type);
2735 		if (ret < 0)
2736 			goto error;
2737 		if (ret > 0)
2738 			break;
2739 
2740 		leaf = path->nodes[0];
2741 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2742 
2743 		chunk = btrfs_item_ptr(leaf, path->slots[0],
2744 				       struct btrfs_chunk);
2745 		chunk_type = btrfs_chunk_type(leaf, chunk);
2746 		btrfs_release_path(path);
2747 
2748 		if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2749 			ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2750 						   found_key.objectid,
2751 						   found_key.offset);
2752 			if (ret == -ENOSPC)
2753 				failed++;
2754 			else
2755 				BUG_ON(ret);
2756 		}
2757 
2758 		if (found_key.offset == 0)
2759 			break;
2760 		key.offset = found_key.offset - 1;
2761 	}
2762 	ret = 0;
2763 	if (failed && !retried) {
2764 		failed = 0;
2765 		retried = true;
2766 		goto again;
2767 	} else if (WARN_ON(failed && retried)) {
2768 		ret = -ENOSPC;
2769 	}
2770 error:
2771 	btrfs_free_path(path);
2772 	return ret;
2773 }
2774 
2775 static int insert_balance_item(struct btrfs_root *root,
2776 			       struct btrfs_balance_control *bctl)
2777 {
2778 	struct btrfs_trans_handle *trans;
2779 	struct btrfs_balance_item *item;
2780 	struct btrfs_disk_balance_args disk_bargs;
2781 	struct btrfs_path *path;
2782 	struct extent_buffer *leaf;
2783 	struct btrfs_key key;
2784 	int ret, err;
2785 
2786 	path = btrfs_alloc_path();
2787 	if (!path)
2788 		return -ENOMEM;
2789 
2790 	trans = btrfs_start_transaction(root, 0);
2791 	if (IS_ERR(trans)) {
2792 		btrfs_free_path(path);
2793 		return PTR_ERR(trans);
2794 	}
2795 
2796 	key.objectid = BTRFS_BALANCE_OBJECTID;
2797 	key.type = BTRFS_BALANCE_ITEM_KEY;
2798 	key.offset = 0;
2799 
2800 	ret = btrfs_insert_empty_item(trans, root, path, &key,
2801 				      sizeof(*item));
2802 	if (ret)
2803 		goto out;
2804 
2805 	leaf = path->nodes[0];
2806 	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2807 
2808 	memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2809 
2810 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2811 	btrfs_set_balance_data(leaf, item, &disk_bargs);
2812 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2813 	btrfs_set_balance_meta(leaf, item, &disk_bargs);
2814 	btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2815 	btrfs_set_balance_sys(leaf, item, &disk_bargs);
2816 
2817 	btrfs_set_balance_flags(leaf, item, bctl->flags);
2818 
2819 	btrfs_mark_buffer_dirty(leaf);
2820 out:
2821 	btrfs_free_path(path);
2822 	err = btrfs_commit_transaction(trans, root);
2823 	if (err && !ret)
2824 		ret = err;
2825 	return ret;
2826 }
2827 
2828 static int del_balance_item(struct btrfs_root *root)
2829 {
2830 	struct btrfs_trans_handle *trans;
2831 	struct btrfs_path *path;
2832 	struct btrfs_key key;
2833 	int ret, err;
2834 
2835 	path = btrfs_alloc_path();
2836 	if (!path)
2837 		return -ENOMEM;
2838 
2839 	trans = btrfs_start_transaction(root, 0);
2840 	if (IS_ERR(trans)) {
2841 		btrfs_free_path(path);
2842 		return PTR_ERR(trans);
2843 	}
2844 
2845 	key.objectid = BTRFS_BALANCE_OBJECTID;
2846 	key.type = BTRFS_BALANCE_ITEM_KEY;
2847 	key.offset = 0;
2848 
2849 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2850 	if (ret < 0)
2851 		goto out;
2852 	if (ret > 0) {
2853 		ret = -ENOENT;
2854 		goto out;
2855 	}
2856 
2857 	ret = btrfs_del_item(trans, root, path);
2858 out:
2859 	btrfs_free_path(path);
2860 	err = btrfs_commit_transaction(trans, root);
2861 	if (err && !ret)
2862 		ret = err;
2863 	return ret;
2864 }
2865 
2866 /*
2867  * This is a heuristic used to reduce the number of chunks balanced on
2868  * resume after balance was interrupted.
2869  */
2870 static void update_balance_args(struct btrfs_balance_control *bctl)
2871 {
2872 	/*
2873 	 * Turn on soft mode for chunk types that were being converted.
2874 	 */
2875 	if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2876 		bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2877 	if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2878 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2879 	if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2880 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2881 
2882 	/*
2883 	 * Turn on usage filter if is not already used.  The idea is
2884 	 * that chunks that we have already balanced should be
2885 	 * reasonably full.  Don't do it for chunks that are being
2886 	 * converted - that will keep us from relocating unconverted
2887 	 * (albeit full) chunks.
2888 	 */
2889 	if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2890 	    !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2891 		bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2892 		bctl->data.usage = 90;
2893 	}
2894 	if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2895 	    !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2896 		bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2897 		bctl->sys.usage = 90;
2898 	}
2899 	if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2900 	    !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2901 		bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2902 		bctl->meta.usage = 90;
2903 	}
2904 }
2905 
2906 /*
2907  * Should be called with both balance and volume mutexes held to
2908  * serialize other volume operations (add_dev/rm_dev/resize) with
2909  * restriper.  Same goes for unset_balance_control.
2910  */
2911 static void set_balance_control(struct btrfs_balance_control *bctl)
2912 {
2913 	struct btrfs_fs_info *fs_info = bctl->fs_info;
2914 
2915 	BUG_ON(fs_info->balance_ctl);
2916 
2917 	spin_lock(&fs_info->balance_lock);
2918 	fs_info->balance_ctl = bctl;
2919 	spin_unlock(&fs_info->balance_lock);
2920 }
2921 
2922 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2923 {
2924 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2925 
2926 	BUG_ON(!fs_info->balance_ctl);
2927 
2928 	spin_lock(&fs_info->balance_lock);
2929 	fs_info->balance_ctl = NULL;
2930 	spin_unlock(&fs_info->balance_lock);
2931 
2932 	kfree(bctl);
2933 }
2934 
2935 /*
2936  * Balance filters.  Return 1 if chunk should be filtered out
2937  * (should not be balanced).
2938  */
2939 static int chunk_profiles_filter(u64 chunk_type,
2940 				 struct btrfs_balance_args *bargs)
2941 {
2942 	chunk_type = chunk_to_extended(chunk_type) &
2943 				BTRFS_EXTENDED_PROFILE_MASK;
2944 
2945 	if (bargs->profiles & chunk_type)
2946 		return 0;
2947 
2948 	return 1;
2949 }
2950 
2951 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2952 			      struct btrfs_balance_args *bargs)
2953 {
2954 	struct btrfs_block_group_cache *cache;
2955 	u64 chunk_used, user_thresh;
2956 	int ret = 1;
2957 
2958 	cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2959 	chunk_used = btrfs_block_group_used(&cache->item);
2960 
2961 	if (bargs->usage == 0)
2962 		user_thresh = 1;
2963 	else if (bargs->usage > 100)
2964 		user_thresh = cache->key.offset;
2965 	else
2966 		user_thresh = div_factor_fine(cache->key.offset,
2967 					      bargs->usage);
2968 
2969 	if (chunk_used < user_thresh)
2970 		ret = 0;
2971 
2972 	btrfs_put_block_group(cache);
2973 	return ret;
2974 }
2975 
2976 static int chunk_devid_filter(struct extent_buffer *leaf,
2977 			      struct btrfs_chunk *chunk,
2978 			      struct btrfs_balance_args *bargs)
2979 {
2980 	struct btrfs_stripe *stripe;
2981 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2982 	int i;
2983 
2984 	for (i = 0; i < num_stripes; i++) {
2985 		stripe = btrfs_stripe_nr(chunk, i);
2986 		if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2987 			return 0;
2988 	}
2989 
2990 	return 1;
2991 }
2992 
2993 /* [pstart, pend) */
2994 static int chunk_drange_filter(struct extent_buffer *leaf,
2995 			       struct btrfs_chunk *chunk,
2996 			       u64 chunk_offset,
2997 			       struct btrfs_balance_args *bargs)
2998 {
2999 	struct btrfs_stripe *stripe;
3000 	int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3001 	u64 stripe_offset;
3002 	u64 stripe_length;
3003 	int factor;
3004 	int i;
3005 
3006 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3007 		return 0;
3008 
3009 	if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3010 	     BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3011 		factor = num_stripes / 2;
3012 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3013 		factor = num_stripes - 1;
3014 	} else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3015 		factor = num_stripes - 2;
3016 	} else {
3017 		factor = num_stripes;
3018 	}
3019 
3020 	for (i = 0; i < num_stripes; i++) {
3021 		stripe = btrfs_stripe_nr(chunk, i);
3022 		if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3023 			continue;
3024 
3025 		stripe_offset = btrfs_stripe_offset(leaf, stripe);
3026 		stripe_length = btrfs_chunk_length(leaf, chunk);
3027 		do_div(stripe_length, factor);
3028 
3029 		if (stripe_offset < bargs->pend &&
3030 		    stripe_offset + stripe_length > bargs->pstart)
3031 			return 0;
3032 	}
3033 
3034 	return 1;
3035 }
3036 
3037 /* [vstart, vend) */
3038 static int chunk_vrange_filter(struct extent_buffer *leaf,
3039 			       struct btrfs_chunk *chunk,
3040 			       u64 chunk_offset,
3041 			       struct btrfs_balance_args *bargs)
3042 {
3043 	if (chunk_offset < bargs->vend &&
3044 	    chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3045 		/* at least part of the chunk is inside this vrange */
3046 		return 0;
3047 
3048 	return 1;
3049 }
3050 
3051 static int chunk_soft_convert_filter(u64 chunk_type,
3052 				     struct btrfs_balance_args *bargs)
3053 {
3054 	if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3055 		return 0;
3056 
3057 	chunk_type = chunk_to_extended(chunk_type) &
3058 				BTRFS_EXTENDED_PROFILE_MASK;
3059 
3060 	if (bargs->target == chunk_type)
3061 		return 1;
3062 
3063 	return 0;
3064 }
3065 
3066 static int should_balance_chunk(struct btrfs_root *root,
3067 				struct extent_buffer *leaf,
3068 				struct btrfs_chunk *chunk, u64 chunk_offset)
3069 {
3070 	struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3071 	struct btrfs_balance_args *bargs = NULL;
3072 	u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3073 
3074 	/* type filter */
3075 	if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3076 	      (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3077 		return 0;
3078 	}
3079 
3080 	if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3081 		bargs = &bctl->data;
3082 	else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3083 		bargs = &bctl->sys;
3084 	else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3085 		bargs = &bctl->meta;
3086 
3087 	/* profiles filter */
3088 	if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3089 	    chunk_profiles_filter(chunk_type, bargs)) {
3090 		return 0;
3091 	}
3092 
3093 	/* usage filter */
3094 	if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3095 	    chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3096 		return 0;
3097 	}
3098 
3099 	/* devid filter */
3100 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3101 	    chunk_devid_filter(leaf, chunk, bargs)) {
3102 		return 0;
3103 	}
3104 
3105 	/* drange filter, makes sense only with devid filter */
3106 	if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3107 	    chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3108 		return 0;
3109 	}
3110 
3111 	/* vrange filter */
3112 	if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3113 	    chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3114 		return 0;
3115 	}
3116 
3117 	/* soft profile changing mode */
3118 	if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3119 	    chunk_soft_convert_filter(chunk_type, bargs)) {
3120 		return 0;
3121 	}
3122 
3123 	/*
3124 	 * limited by count, must be the last filter
3125 	 */
3126 	if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3127 		if (bargs->limit == 0)
3128 			return 0;
3129 		else
3130 			bargs->limit--;
3131 	}
3132 
3133 	return 1;
3134 }
3135 
3136 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3137 {
3138 	struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3139 	struct btrfs_root *chunk_root = fs_info->chunk_root;
3140 	struct btrfs_root *dev_root = fs_info->dev_root;
3141 	struct list_head *devices;
3142 	struct btrfs_device *device;
3143 	u64 old_size;
3144 	u64 size_to_free;
3145 	struct btrfs_chunk *chunk;
3146 	struct btrfs_path *path;
3147 	struct btrfs_key key;
3148 	struct btrfs_key found_key;
3149 	struct btrfs_trans_handle *trans;
3150 	struct extent_buffer *leaf;
3151 	int slot;
3152 	int ret;
3153 	int enospc_errors = 0;
3154 	bool counting = true;
3155 	u64 limit_data = bctl->data.limit;
3156 	u64 limit_meta = bctl->meta.limit;
3157 	u64 limit_sys = bctl->sys.limit;
3158 
3159 	/* step one make some room on all the devices */
3160 	devices = &fs_info->fs_devices->devices;
3161 	list_for_each_entry(device, devices, dev_list) {
3162 		old_size = btrfs_device_get_total_bytes(device);
3163 		size_to_free = div_factor(old_size, 1);
3164 		size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3165 		if (!device->writeable ||
3166 		    btrfs_device_get_total_bytes(device) -
3167 		    btrfs_device_get_bytes_used(device) > size_to_free ||
3168 		    device->is_tgtdev_for_dev_replace)
3169 			continue;
3170 
3171 		ret = btrfs_shrink_device(device, old_size - size_to_free);
3172 		if (ret == -ENOSPC)
3173 			break;
3174 		BUG_ON(ret);
3175 
3176 		trans = btrfs_start_transaction(dev_root, 0);
3177 		BUG_ON(IS_ERR(trans));
3178 
3179 		ret = btrfs_grow_device(trans, device, old_size);
3180 		BUG_ON(ret);
3181 
3182 		btrfs_end_transaction(trans, dev_root);
3183 	}
3184 
3185 	/* step two, relocate all the chunks */
3186 	path = btrfs_alloc_path();
3187 	if (!path) {
3188 		ret = -ENOMEM;
3189 		goto error;
3190 	}
3191 
3192 	/* zero out stat counters */
3193 	spin_lock(&fs_info->balance_lock);
3194 	memset(&bctl->stat, 0, sizeof(bctl->stat));
3195 	spin_unlock(&fs_info->balance_lock);
3196 again:
3197 	if (!counting) {
3198 		bctl->data.limit = limit_data;
3199 		bctl->meta.limit = limit_meta;
3200 		bctl->sys.limit = limit_sys;
3201 	}
3202 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3203 	key.offset = (u64)-1;
3204 	key.type = BTRFS_CHUNK_ITEM_KEY;
3205 
3206 	while (1) {
3207 		if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3208 		    atomic_read(&fs_info->balance_cancel_req)) {
3209 			ret = -ECANCELED;
3210 			goto error;
3211 		}
3212 
3213 		ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3214 		if (ret < 0)
3215 			goto error;
3216 
3217 		/*
3218 		 * this shouldn't happen, it means the last relocate
3219 		 * failed
3220 		 */
3221 		if (ret == 0)
3222 			BUG(); /* FIXME break ? */
3223 
3224 		ret = btrfs_previous_item(chunk_root, path, 0,
3225 					  BTRFS_CHUNK_ITEM_KEY);
3226 		if (ret) {
3227 			ret = 0;
3228 			break;
3229 		}
3230 
3231 		leaf = path->nodes[0];
3232 		slot = path->slots[0];
3233 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3234 
3235 		if (found_key.objectid != key.objectid)
3236 			break;
3237 
3238 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3239 
3240 		if (!counting) {
3241 			spin_lock(&fs_info->balance_lock);
3242 			bctl->stat.considered++;
3243 			spin_unlock(&fs_info->balance_lock);
3244 		}
3245 
3246 		ret = should_balance_chunk(chunk_root, leaf, chunk,
3247 					   found_key.offset);
3248 		btrfs_release_path(path);
3249 		if (!ret)
3250 			goto loop;
3251 
3252 		if (counting) {
3253 			spin_lock(&fs_info->balance_lock);
3254 			bctl->stat.expected++;
3255 			spin_unlock(&fs_info->balance_lock);
3256 			goto loop;
3257 		}
3258 
3259 		ret = btrfs_relocate_chunk(chunk_root,
3260 					   chunk_root->root_key.objectid,
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_tree;
3963 	u64 chunk_objectid;
3964 	u64 chunk_offset;
3965 	int ret;
3966 	int slot;
3967 	int failed = 0;
3968 	bool retried = false;
3969 	struct extent_buffer *l;
3970 	struct btrfs_key key;
3971 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3972 	u64 old_total = btrfs_super_total_bytes(super_copy);
3973 	u64 old_size = btrfs_device_get_total_bytes(device);
3974 	u64 diff = old_size - new_size;
3975 
3976 	if (device->is_tgtdev_for_dev_replace)
3977 		return -EINVAL;
3978 
3979 	path = btrfs_alloc_path();
3980 	if (!path)
3981 		return -ENOMEM;
3982 
3983 	path->reada = 2;
3984 
3985 	lock_chunks(root);
3986 
3987 	btrfs_device_set_total_bytes(device, new_size);
3988 	if (device->writeable) {
3989 		device->fs_devices->total_rw_bytes -= diff;
3990 		spin_lock(&root->fs_info->free_chunk_lock);
3991 		root->fs_info->free_chunk_space -= diff;
3992 		spin_unlock(&root->fs_info->free_chunk_lock);
3993 	}
3994 	unlock_chunks(root);
3995 
3996 again:
3997 	key.objectid = device->devid;
3998 	key.offset = (u64)-1;
3999 	key.type = BTRFS_DEV_EXTENT_KEY;
4000 
4001 	do {
4002 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4003 		if (ret < 0)
4004 			goto done;
4005 
4006 		ret = btrfs_previous_item(root, path, 0, key.type);
4007 		if (ret < 0)
4008 			goto done;
4009 		if (ret) {
4010 			ret = 0;
4011 			btrfs_release_path(path);
4012 			break;
4013 		}
4014 
4015 		l = path->nodes[0];
4016 		slot = path->slots[0];
4017 		btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4018 
4019 		if (key.objectid != device->devid) {
4020 			btrfs_release_path(path);
4021 			break;
4022 		}
4023 
4024 		dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4025 		length = btrfs_dev_extent_length(l, dev_extent);
4026 
4027 		if (key.offset + length <= new_size) {
4028 			btrfs_release_path(path);
4029 			break;
4030 		}
4031 
4032 		chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
4033 		chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4034 		chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4035 		btrfs_release_path(path);
4036 
4037 		ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
4038 					   chunk_offset);
4039 		if (ret && ret != -ENOSPC)
4040 			goto done;
4041 		if (ret == -ENOSPC)
4042 			failed++;
4043 	} while (key.offset-- > 0);
4044 
4045 	if (failed && !retried) {
4046 		failed = 0;
4047 		retried = true;
4048 		goto again;
4049 	} else if (failed && retried) {
4050 		ret = -ENOSPC;
4051 		lock_chunks(root);
4052 
4053 		btrfs_device_set_total_bytes(device, old_size);
4054 		if (device->writeable)
4055 			device->fs_devices->total_rw_bytes += diff;
4056 		spin_lock(&root->fs_info->free_chunk_lock);
4057 		root->fs_info->free_chunk_space += diff;
4058 		spin_unlock(&root->fs_info->free_chunk_lock);
4059 		unlock_chunks(root);
4060 		goto done;
4061 	}
4062 
4063 	/* Shrinking succeeded, else we would be at "done". */
4064 	trans = btrfs_start_transaction(root, 0);
4065 	if (IS_ERR(trans)) {
4066 		ret = PTR_ERR(trans);
4067 		goto done;
4068 	}
4069 
4070 	lock_chunks(root);
4071 	btrfs_device_set_disk_total_bytes(device, new_size);
4072 	if (list_empty(&device->resized_list))
4073 		list_add_tail(&device->resized_list,
4074 			      &root->fs_info->fs_devices->resized_devices);
4075 
4076 	WARN_ON(diff > old_total);
4077 	btrfs_set_super_total_bytes(super_copy, old_total - diff);
4078 	unlock_chunks(root);
4079 
4080 	/* Now btrfs_update_device() will change the on-disk size. */
4081 	ret = btrfs_update_device(trans, device);
4082 	btrfs_end_transaction(trans, root);
4083 done:
4084 	btrfs_free_path(path);
4085 	return ret;
4086 }
4087 
4088 static int btrfs_add_system_chunk(struct btrfs_root *root,
4089 			   struct btrfs_key *key,
4090 			   struct btrfs_chunk *chunk, int item_size)
4091 {
4092 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4093 	struct btrfs_disk_key disk_key;
4094 	u32 array_size;
4095 	u8 *ptr;
4096 
4097 	lock_chunks(root);
4098 	array_size = btrfs_super_sys_array_size(super_copy);
4099 	if (array_size + item_size + sizeof(disk_key)
4100 			> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4101 		unlock_chunks(root);
4102 		return -EFBIG;
4103 	}
4104 
4105 	ptr = super_copy->sys_chunk_array + array_size;
4106 	btrfs_cpu_key_to_disk(&disk_key, key);
4107 	memcpy(ptr, &disk_key, sizeof(disk_key));
4108 	ptr += sizeof(disk_key);
4109 	memcpy(ptr, chunk, item_size);
4110 	item_size += sizeof(disk_key);
4111 	btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4112 	unlock_chunks(root);
4113 
4114 	return 0;
4115 }
4116 
4117 /*
4118  * sort the devices in descending order by max_avail, total_avail
4119  */
4120 static int btrfs_cmp_device_info(const void *a, const void *b)
4121 {
4122 	const struct btrfs_device_info *di_a = a;
4123 	const struct btrfs_device_info *di_b = b;
4124 
4125 	if (di_a->max_avail > di_b->max_avail)
4126 		return -1;
4127 	if (di_a->max_avail < di_b->max_avail)
4128 		return 1;
4129 	if (di_a->total_avail > di_b->total_avail)
4130 		return -1;
4131 	if (di_a->total_avail < di_b->total_avail)
4132 		return 1;
4133 	return 0;
4134 }
4135 
4136 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4137 	[BTRFS_RAID_RAID10] = {
4138 		.sub_stripes	= 2,
4139 		.dev_stripes	= 1,
4140 		.devs_max	= 0,	/* 0 == as many as possible */
4141 		.devs_min	= 4,
4142 		.devs_increment	= 2,
4143 		.ncopies	= 2,
4144 	},
4145 	[BTRFS_RAID_RAID1] = {
4146 		.sub_stripes	= 1,
4147 		.dev_stripes	= 1,
4148 		.devs_max	= 2,
4149 		.devs_min	= 2,
4150 		.devs_increment	= 2,
4151 		.ncopies	= 2,
4152 	},
4153 	[BTRFS_RAID_DUP] = {
4154 		.sub_stripes	= 1,
4155 		.dev_stripes	= 2,
4156 		.devs_max	= 1,
4157 		.devs_min	= 1,
4158 		.devs_increment	= 1,
4159 		.ncopies	= 2,
4160 	},
4161 	[BTRFS_RAID_RAID0] = {
4162 		.sub_stripes	= 1,
4163 		.dev_stripes	= 1,
4164 		.devs_max	= 0,
4165 		.devs_min	= 2,
4166 		.devs_increment	= 1,
4167 		.ncopies	= 1,
4168 	},
4169 	[BTRFS_RAID_SINGLE] = {
4170 		.sub_stripes	= 1,
4171 		.dev_stripes	= 1,
4172 		.devs_max	= 1,
4173 		.devs_min	= 1,
4174 		.devs_increment	= 1,
4175 		.ncopies	= 1,
4176 	},
4177 	[BTRFS_RAID_RAID5] = {
4178 		.sub_stripes	= 1,
4179 		.dev_stripes	= 1,
4180 		.devs_max	= 0,
4181 		.devs_min	= 2,
4182 		.devs_increment	= 1,
4183 		.ncopies	= 2,
4184 	},
4185 	[BTRFS_RAID_RAID6] = {
4186 		.sub_stripes	= 1,
4187 		.dev_stripes	= 1,
4188 		.devs_max	= 0,
4189 		.devs_min	= 3,
4190 		.devs_increment	= 1,
4191 		.ncopies	= 3,
4192 	},
4193 };
4194 
4195 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4196 {
4197 	/* TODO allow them to set a preferred stripe size */
4198 	return 64 * 1024;
4199 }
4200 
4201 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4202 {
4203 	if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4204 		return;
4205 
4206 	btrfs_set_fs_incompat(info, RAID56);
4207 }
4208 
4209 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r)		\
4210 			- sizeof(struct btrfs_item)		\
4211 			- sizeof(struct btrfs_chunk))		\
4212 			/ sizeof(struct btrfs_stripe) + 1)
4213 
4214 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE	\
4215 				- 2 * sizeof(struct btrfs_disk_key)	\
4216 				- 2 * sizeof(struct btrfs_chunk))	\
4217 				/ sizeof(struct btrfs_stripe) + 1)
4218 
4219 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4220 			       struct btrfs_root *extent_root, u64 start,
4221 			       u64 type)
4222 {
4223 	struct btrfs_fs_info *info = extent_root->fs_info;
4224 	struct btrfs_fs_devices *fs_devices = info->fs_devices;
4225 	struct list_head *cur;
4226 	struct map_lookup *map = NULL;
4227 	struct extent_map_tree *em_tree;
4228 	struct extent_map *em;
4229 	struct btrfs_device_info *devices_info = NULL;
4230 	u64 total_avail;
4231 	int num_stripes;	/* total number of stripes to allocate */
4232 	int data_stripes;	/* number of stripes that count for
4233 				   block group size */
4234 	int sub_stripes;	/* sub_stripes info for map */
4235 	int dev_stripes;	/* stripes per dev */
4236 	int devs_max;		/* max devs to use */
4237 	int devs_min;		/* min devs needed */
4238 	int devs_increment;	/* ndevs has to be a multiple of this */
4239 	int ncopies;		/* how many copies to data has */
4240 	int ret;
4241 	u64 max_stripe_size;
4242 	u64 max_chunk_size;
4243 	u64 stripe_size;
4244 	u64 num_bytes;
4245 	u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4246 	int ndevs;
4247 	int i;
4248 	int j;
4249 	int index;
4250 
4251 	BUG_ON(!alloc_profile_is_valid(type, 0));
4252 
4253 	if (list_empty(&fs_devices->alloc_list))
4254 		return -ENOSPC;
4255 
4256 	index = __get_raid_index(type);
4257 
4258 	sub_stripes = btrfs_raid_array[index].sub_stripes;
4259 	dev_stripes = btrfs_raid_array[index].dev_stripes;
4260 	devs_max = btrfs_raid_array[index].devs_max;
4261 	devs_min = btrfs_raid_array[index].devs_min;
4262 	devs_increment = btrfs_raid_array[index].devs_increment;
4263 	ncopies = btrfs_raid_array[index].ncopies;
4264 
4265 	if (type & BTRFS_BLOCK_GROUP_DATA) {
4266 		max_stripe_size = 1024 * 1024 * 1024;
4267 		max_chunk_size = 10 * max_stripe_size;
4268 		if (!devs_max)
4269 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4270 	} else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4271 		/* for larger filesystems, use larger metadata chunks */
4272 		if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4273 			max_stripe_size = 1024 * 1024 * 1024;
4274 		else
4275 			max_stripe_size = 256 * 1024 * 1024;
4276 		max_chunk_size = max_stripe_size;
4277 		if (!devs_max)
4278 			devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4279 	} else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4280 		max_stripe_size = 32 * 1024 * 1024;
4281 		max_chunk_size = 2 * max_stripe_size;
4282 		if (!devs_max)
4283 			devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4284 	} else {
4285 		btrfs_err(info, "invalid chunk type 0x%llx requested",
4286 		       type);
4287 		BUG_ON(1);
4288 	}
4289 
4290 	/* we don't want a chunk larger than 10% of writeable space */
4291 	max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4292 			     max_chunk_size);
4293 
4294 	devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4295 			       GFP_NOFS);
4296 	if (!devices_info)
4297 		return -ENOMEM;
4298 
4299 	cur = fs_devices->alloc_list.next;
4300 
4301 	/*
4302 	 * in the first pass through the devices list, we gather information
4303 	 * about the available holes on each device.
4304 	 */
4305 	ndevs = 0;
4306 	while (cur != &fs_devices->alloc_list) {
4307 		struct btrfs_device *device;
4308 		u64 max_avail;
4309 		u64 dev_offset;
4310 
4311 		device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4312 
4313 		cur = cur->next;
4314 
4315 		if (!device->writeable) {
4316 			WARN(1, KERN_ERR
4317 			       "BTRFS: read-only device in alloc_list\n");
4318 			continue;
4319 		}
4320 
4321 		if (!device->in_fs_metadata ||
4322 		    device->is_tgtdev_for_dev_replace)
4323 			continue;
4324 
4325 		if (device->total_bytes > device->bytes_used)
4326 			total_avail = device->total_bytes - device->bytes_used;
4327 		else
4328 			total_avail = 0;
4329 
4330 		/* If there is no space on this device, skip it. */
4331 		if (total_avail == 0)
4332 			continue;
4333 
4334 		ret = find_free_dev_extent(trans, device,
4335 					   max_stripe_size * dev_stripes,
4336 					   &dev_offset, &max_avail);
4337 		if (ret && ret != -ENOSPC)
4338 			goto error;
4339 
4340 		if (ret == 0)
4341 			max_avail = max_stripe_size * dev_stripes;
4342 
4343 		if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4344 			continue;
4345 
4346 		if (ndevs == fs_devices->rw_devices) {
4347 			WARN(1, "%s: found more than %llu devices\n",
4348 			     __func__, fs_devices->rw_devices);
4349 			break;
4350 		}
4351 		devices_info[ndevs].dev_offset = dev_offset;
4352 		devices_info[ndevs].max_avail = max_avail;
4353 		devices_info[ndevs].total_avail = total_avail;
4354 		devices_info[ndevs].dev = device;
4355 		++ndevs;
4356 	}
4357 
4358 	/*
4359 	 * now sort the devices by hole size / available space
4360 	 */
4361 	sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4362 	     btrfs_cmp_device_info, NULL);
4363 
4364 	/* round down to number of usable stripes */
4365 	ndevs -= ndevs % devs_increment;
4366 
4367 	if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4368 		ret = -ENOSPC;
4369 		goto error;
4370 	}
4371 
4372 	if (devs_max && ndevs > devs_max)
4373 		ndevs = devs_max;
4374 	/*
4375 	 * the primary goal is to maximize the number of stripes, so use as many
4376 	 * devices as possible, even if the stripes are not maximum sized.
4377 	 */
4378 	stripe_size = devices_info[ndevs-1].max_avail;
4379 	num_stripes = ndevs * dev_stripes;
4380 
4381 	/*
4382 	 * this will have to be fixed for RAID1 and RAID10 over
4383 	 * more drives
4384 	 */
4385 	data_stripes = num_stripes / ncopies;
4386 
4387 	if (type & BTRFS_BLOCK_GROUP_RAID5) {
4388 		raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4389 				 btrfs_super_stripesize(info->super_copy));
4390 		data_stripes = num_stripes - 1;
4391 	}
4392 	if (type & BTRFS_BLOCK_GROUP_RAID6) {
4393 		raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4394 				 btrfs_super_stripesize(info->super_copy));
4395 		data_stripes = num_stripes - 2;
4396 	}
4397 
4398 	/*
4399 	 * Use the number of data stripes to figure out how big this chunk
4400 	 * is really going to be in terms of logical address space,
4401 	 * and compare that answer with the max chunk size
4402 	 */
4403 	if (stripe_size * data_stripes > max_chunk_size) {
4404 		u64 mask = (1ULL << 24) - 1;
4405 		stripe_size = max_chunk_size;
4406 		do_div(stripe_size, data_stripes);
4407 
4408 		/* bump the answer up to a 16MB boundary */
4409 		stripe_size = (stripe_size + mask) & ~mask;
4410 
4411 		/* but don't go higher than the limits we found
4412 		 * while searching for free extents
4413 		 */
4414 		if (stripe_size > devices_info[ndevs-1].max_avail)
4415 			stripe_size = devices_info[ndevs-1].max_avail;
4416 	}
4417 
4418 	do_div(stripe_size, dev_stripes);
4419 
4420 	/* align to BTRFS_STRIPE_LEN */
4421 	do_div(stripe_size, raid_stripe_len);
4422 	stripe_size *= raid_stripe_len;
4423 
4424 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4425 	if (!map) {
4426 		ret = -ENOMEM;
4427 		goto error;
4428 	}
4429 	map->num_stripes = num_stripes;
4430 
4431 	for (i = 0; i < ndevs; ++i) {
4432 		for (j = 0; j < dev_stripes; ++j) {
4433 			int s = i * dev_stripes + j;
4434 			map->stripes[s].dev = devices_info[i].dev;
4435 			map->stripes[s].physical = devices_info[i].dev_offset +
4436 						   j * stripe_size;
4437 		}
4438 	}
4439 	map->sector_size = extent_root->sectorsize;
4440 	map->stripe_len = raid_stripe_len;
4441 	map->io_align = raid_stripe_len;
4442 	map->io_width = raid_stripe_len;
4443 	map->type = type;
4444 	map->sub_stripes = sub_stripes;
4445 
4446 	num_bytes = stripe_size * data_stripes;
4447 
4448 	trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4449 
4450 	em = alloc_extent_map();
4451 	if (!em) {
4452 		kfree(map);
4453 		ret = -ENOMEM;
4454 		goto error;
4455 	}
4456 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4457 	em->bdev = (struct block_device *)map;
4458 	em->start = start;
4459 	em->len = num_bytes;
4460 	em->block_start = 0;
4461 	em->block_len = em->len;
4462 	em->orig_block_len = stripe_size;
4463 
4464 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4465 	write_lock(&em_tree->lock);
4466 	ret = add_extent_mapping(em_tree, em, 0);
4467 	if (!ret) {
4468 		list_add_tail(&em->list, &trans->transaction->pending_chunks);
4469 		atomic_inc(&em->refs);
4470 	}
4471 	write_unlock(&em_tree->lock);
4472 	if (ret) {
4473 		free_extent_map(em);
4474 		goto error;
4475 	}
4476 
4477 	ret = btrfs_make_block_group(trans, extent_root, 0, type,
4478 				     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4479 				     start, num_bytes);
4480 	if (ret)
4481 		goto error_del_extent;
4482 
4483 	for (i = 0; i < map->num_stripes; i++) {
4484 		num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4485 		btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4486 	}
4487 
4488 	spin_lock(&extent_root->fs_info->free_chunk_lock);
4489 	extent_root->fs_info->free_chunk_space -= (stripe_size *
4490 						   map->num_stripes);
4491 	spin_unlock(&extent_root->fs_info->free_chunk_lock);
4492 
4493 	free_extent_map(em);
4494 	check_raid56_incompat_flag(extent_root->fs_info, type);
4495 
4496 	kfree(devices_info);
4497 	return 0;
4498 
4499 error_del_extent:
4500 	write_lock(&em_tree->lock);
4501 	remove_extent_mapping(em_tree, em);
4502 	write_unlock(&em_tree->lock);
4503 
4504 	/* One for our allocation */
4505 	free_extent_map(em);
4506 	/* One for the tree reference */
4507 	free_extent_map(em);
4508 error:
4509 	kfree(devices_info);
4510 	return ret;
4511 }
4512 
4513 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4514 				struct btrfs_root *extent_root,
4515 				u64 chunk_offset, u64 chunk_size)
4516 {
4517 	struct btrfs_key key;
4518 	struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4519 	struct btrfs_device *device;
4520 	struct btrfs_chunk *chunk;
4521 	struct btrfs_stripe *stripe;
4522 	struct extent_map_tree *em_tree;
4523 	struct extent_map *em;
4524 	struct map_lookup *map;
4525 	size_t item_size;
4526 	u64 dev_offset;
4527 	u64 stripe_size;
4528 	int i = 0;
4529 	int ret;
4530 
4531 	em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4532 	read_lock(&em_tree->lock);
4533 	em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4534 	read_unlock(&em_tree->lock);
4535 
4536 	if (!em) {
4537 		btrfs_crit(extent_root->fs_info, "unable to find logical "
4538 			   "%Lu len %Lu", chunk_offset, chunk_size);
4539 		return -EINVAL;
4540 	}
4541 
4542 	if (em->start != chunk_offset || em->len != chunk_size) {
4543 		btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4544 			  " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4545 			  chunk_size, em->start, em->len);
4546 		free_extent_map(em);
4547 		return -EINVAL;
4548 	}
4549 
4550 	map = (struct map_lookup *)em->bdev;
4551 	item_size = btrfs_chunk_item_size(map->num_stripes);
4552 	stripe_size = em->orig_block_len;
4553 
4554 	chunk = kzalloc(item_size, GFP_NOFS);
4555 	if (!chunk) {
4556 		ret = -ENOMEM;
4557 		goto out;
4558 	}
4559 
4560 	for (i = 0; i < map->num_stripes; i++) {
4561 		device = map->stripes[i].dev;
4562 		dev_offset = map->stripes[i].physical;
4563 
4564 		ret = btrfs_update_device(trans, device);
4565 		if (ret)
4566 			goto out;
4567 		ret = btrfs_alloc_dev_extent(trans, device,
4568 					     chunk_root->root_key.objectid,
4569 					     BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4570 					     chunk_offset, dev_offset,
4571 					     stripe_size);
4572 		if (ret)
4573 			goto out;
4574 	}
4575 
4576 	stripe = &chunk->stripe;
4577 	for (i = 0; i < map->num_stripes; i++) {
4578 		device = map->stripes[i].dev;
4579 		dev_offset = map->stripes[i].physical;
4580 
4581 		btrfs_set_stack_stripe_devid(stripe, device->devid);
4582 		btrfs_set_stack_stripe_offset(stripe, dev_offset);
4583 		memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4584 		stripe++;
4585 	}
4586 
4587 	btrfs_set_stack_chunk_length(chunk, chunk_size);
4588 	btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4589 	btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4590 	btrfs_set_stack_chunk_type(chunk, map->type);
4591 	btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4592 	btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4593 	btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4594 	btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4595 	btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4596 
4597 	key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4598 	key.type = BTRFS_CHUNK_ITEM_KEY;
4599 	key.offset = chunk_offset;
4600 
4601 	ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4602 	if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4603 		/*
4604 		 * TODO: Cleanup of inserted chunk root in case of
4605 		 * failure.
4606 		 */
4607 		ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4608 					     item_size);
4609 	}
4610 
4611 out:
4612 	kfree(chunk);
4613 	free_extent_map(em);
4614 	return ret;
4615 }
4616 
4617 /*
4618  * Chunk allocation falls into two parts. The first part does works
4619  * that make the new allocated chunk useable, but not do any operation
4620  * that modifies the chunk tree. The second part does the works that
4621  * require modifying the chunk tree. This division is important for the
4622  * bootstrap process of adding storage to a seed btrfs.
4623  */
4624 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4625 		      struct btrfs_root *extent_root, u64 type)
4626 {
4627 	u64 chunk_offset;
4628 
4629 	chunk_offset = find_next_chunk(extent_root->fs_info);
4630 	return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4631 }
4632 
4633 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4634 					 struct btrfs_root *root,
4635 					 struct btrfs_device *device)
4636 {
4637 	u64 chunk_offset;
4638 	u64 sys_chunk_offset;
4639 	u64 alloc_profile;
4640 	struct btrfs_fs_info *fs_info = root->fs_info;
4641 	struct btrfs_root *extent_root = fs_info->extent_root;
4642 	int ret;
4643 
4644 	chunk_offset = find_next_chunk(fs_info);
4645 	alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4646 	ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4647 				  alloc_profile);
4648 	if (ret)
4649 		return ret;
4650 
4651 	sys_chunk_offset = find_next_chunk(root->fs_info);
4652 	alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4653 	ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4654 				  alloc_profile);
4655 	return ret;
4656 }
4657 
4658 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4659 {
4660 	int max_errors;
4661 
4662 	if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4663 			 BTRFS_BLOCK_GROUP_RAID10 |
4664 			 BTRFS_BLOCK_GROUP_RAID5 |
4665 			 BTRFS_BLOCK_GROUP_DUP)) {
4666 		max_errors = 1;
4667 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4668 		max_errors = 2;
4669 	} else {
4670 		max_errors = 0;
4671 	}
4672 
4673 	return max_errors;
4674 }
4675 
4676 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4677 {
4678 	struct extent_map *em;
4679 	struct map_lookup *map;
4680 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4681 	int readonly = 0;
4682 	int miss_ndevs = 0;
4683 	int i;
4684 
4685 	read_lock(&map_tree->map_tree.lock);
4686 	em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4687 	read_unlock(&map_tree->map_tree.lock);
4688 	if (!em)
4689 		return 1;
4690 
4691 	map = (struct map_lookup *)em->bdev;
4692 	for (i = 0; i < map->num_stripes; i++) {
4693 		if (map->stripes[i].dev->missing) {
4694 			miss_ndevs++;
4695 			continue;
4696 		}
4697 
4698 		if (!map->stripes[i].dev->writeable) {
4699 			readonly = 1;
4700 			goto end;
4701 		}
4702 	}
4703 
4704 	/*
4705 	 * If the number of missing devices is larger than max errors,
4706 	 * we can not write the data into that chunk successfully, so
4707 	 * set it readonly.
4708 	 */
4709 	if (miss_ndevs > btrfs_chunk_max_errors(map))
4710 		readonly = 1;
4711 end:
4712 	free_extent_map(em);
4713 	return readonly;
4714 }
4715 
4716 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4717 {
4718 	extent_map_tree_init(&tree->map_tree);
4719 }
4720 
4721 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4722 {
4723 	struct extent_map *em;
4724 
4725 	while (1) {
4726 		write_lock(&tree->map_tree.lock);
4727 		em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4728 		if (em)
4729 			remove_extent_mapping(&tree->map_tree, em);
4730 		write_unlock(&tree->map_tree.lock);
4731 		if (!em)
4732 			break;
4733 		/* once for us */
4734 		free_extent_map(em);
4735 		/* once for the tree */
4736 		free_extent_map(em);
4737 	}
4738 }
4739 
4740 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4741 {
4742 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4743 	struct extent_map *em;
4744 	struct map_lookup *map;
4745 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4746 	int ret;
4747 
4748 	read_lock(&em_tree->lock);
4749 	em = lookup_extent_mapping(em_tree, logical, len);
4750 	read_unlock(&em_tree->lock);
4751 
4752 	/*
4753 	 * We could return errors for these cases, but that could get ugly and
4754 	 * we'd probably do the same thing which is just not do anything else
4755 	 * and exit, so return 1 so the callers don't try to use other copies.
4756 	 */
4757 	if (!em) {
4758 		btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4759 			    logical+len);
4760 		return 1;
4761 	}
4762 
4763 	if (em->start > logical || em->start + em->len < logical) {
4764 		btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4765 			    "%Lu-%Lu", logical, logical+len, em->start,
4766 			    em->start + em->len);
4767 		free_extent_map(em);
4768 		return 1;
4769 	}
4770 
4771 	map = (struct map_lookup *)em->bdev;
4772 	if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4773 		ret = map->num_stripes;
4774 	else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4775 		ret = map->sub_stripes;
4776 	else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4777 		ret = 2;
4778 	else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4779 		ret = 3;
4780 	else
4781 		ret = 1;
4782 	free_extent_map(em);
4783 
4784 	btrfs_dev_replace_lock(&fs_info->dev_replace);
4785 	if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4786 		ret++;
4787 	btrfs_dev_replace_unlock(&fs_info->dev_replace);
4788 
4789 	return ret;
4790 }
4791 
4792 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4793 				    struct btrfs_mapping_tree *map_tree,
4794 				    u64 logical)
4795 {
4796 	struct extent_map *em;
4797 	struct map_lookup *map;
4798 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4799 	unsigned long len = root->sectorsize;
4800 
4801 	read_lock(&em_tree->lock);
4802 	em = lookup_extent_mapping(em_tree, logical, len);
4803 	read_unlock(&em_tree->lock);
4804 	BUG_ON(!em);
4805 
4806 	BUG_ON(em->start > logical || em->start + em->len < logical);
4807 	map = (struct map_lookup *)em->bdev;
4808 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4809 			 BTRFS_BLOCK_GROUP_RAID6)) {
4810 		len = map->stripe_len * nr_data_stripes(map);
4811 	}
4812 	free_extent_map(em);
4813 	return len;
4814 }
4815 
4816 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4817 			   u64 logical, u64 len, int mirror_num)
4818 {
4819 	struct extent_map *em;
4820 	struct map_lookup *map;
4821 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4822 	int ret = 0;
4823 
4824 	read_lock(&em_tree->lock);
4825 	em = lookup_extent_mapping(em_tree, logical, len);
4826 	read_unlock(&em_tree->lock);
4827 	BUG_ON(!em);
4828 
4829 	BUG_ON(em->start > logical || em->start + em->len < logical);
4830 	map = (struct map_lookup *)em->bdev;
4831 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4832 			 BTRFS_BLOCK_GROUP_RAID6))
4833 		ret = 1;
4834 	free_extent_map(em);
4835 	return ret;
4836 }
4837 
4838 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4839 			    struct map_lookup *map, int first, int num,
4840 			    int optimal, int dev_replace_is_ongoing)
4841 {
4842 	int i;
4843 	int tolerance;
4844 	struct btrfs_device *srcdev;
4845 
4846 	if (dev_replace_is_ongoing &&
4847 	    fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4848 	     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4849 		srcdev = fs_info->dev_replace.srcdev;
4850 	else
4851 		srcdev = NULL;
4852 
4853 	/*
4854 	 * try to avoid the drive that is the source drive for a
4855 	 * dev-replace procedure, only choose it if no other non-missing
4856 	 * mirror is available
4857 	 */
4858 	for (tolerance = 0; tolerance < 2; tolerance++) {
4859 		if (map->stripes[optimal].dev->bdev &&
4860 		    (tolerance || map->stripes[optimal].dev != srcdev))
4861 			return optimal;
4862 		for (i = first; i < first + num; i++) {
4863 			if (map->stripes[i].dev->bdev &&
4864 			    (tolerance || map->stripes[i].dev != srcdev))
4865 				return i;
4866 		}
4867 	}
4868 
4869 	/* we couldn't find one that doesn't fail.  Just return something
4870 	 * and the io error handling code will clean up eventually
4871 	 */
4872 	return optimal;
4873 }
4874 
4875 static inline int parity_smaller(u64 a, u64 b)
4876 {
4877 	return a > b;
4878 }
4879 
4880 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4881 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4882 {
4883 	struct btrfs_bio_stripe s;
4884 	int i;
4885 	u64 l;
4886 	int again = 1;
4887 
4888 	while (again) {
4889 		again = 0;
4890 		for (i = 0; i < bbio->num_stripes - 1; i++) {
4891 			if (parity_smaller(raid_map[i], raid_map[i+1])) {
4892 				s = bbio->stripes[i];
4893 				l = raid_map[i];
4894 				bbio->stripes[i] = bbio->stripes[i+1];
4895 				raid_map[i] = raid_map[i+1];
4896 				bbio->stripes[i+1] = s;
4897 				raid_map[i+1] = l;
4898 				again = 1;
4899 			}
4900 		}
4901 	}
4902 }
4903 
4904 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4905 			     u64 logical, u64 *length,
4906 			     struct btrfs_bio **bbio_ret,
4907 			     int mirror_num, u64 **raid_map_ret)
4908 {
4909 	struct extent_map *em;
4910 	struct map_lookup *map;
4911 	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4912 	struct extent_map_tree *em_tree = &map_tree->map_tree;
4913 	u64 offset;
4914 	u64 stripe_offset;
4915 	u64 stripe_end_offset;
4916 	u64 stripe_nr;
4917 	u64 stripe_nr_orig;
4918 	u64 stripe_nr_end;
4919 	u64 stripe_len;
4920 	u64 *raid_map = NULL;
4921 	int stripe_index;
4922 	int i;
4923 	int ret = 0;
4924 	int num_stripes;
4925 	int max_errors = 0;
4926 	struct btrfs_bio *bbio = NULL;
4927 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4928 	int dev_replace_is_ongoing = 0;
4929 	int num_alloc_stripes;
4930 	int patch_the_first_stripe_for_dev_replace = 0;
4931 	u64 physical_to_patch_in_first_stripe = 0;
4932 	u64 raid56_full_stripe_start = (u64)-1;
4933 
4934 	read_lock(&em_tree->lock);
4935 	em = lookup_extent_mapping(em_tree, logical, *length);
4936 	read_unlock(&em_tree->lock);
4937 
4938 	if (!em) {
4939 		btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4940 			logical, *length);
4941 		return -EINVAL;
4942 	}
4943 
4944 	if (em->start > logical || em->start + em->len < logical) {
4945 		btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4946 			   "found %Lu-%Lu", logical, em->start,
4947 			   em->start + em->len);
4948 		free_extent_map(em);
4949 		return -EINVAL;
4950 	}
4951 
4952 	map = (struct map_lookup *)em->bdev;
4953 	offset = logical - em->start;
4954 
4955 	stripe_len = map->stripe_len;
4956 	stripe_nr = offset;
4957 	/*
4958 	 * stripe_nr counts the total number of stripes we have to stride
4959 	 * to get to this block
4960 	 */
4961 	do_div(stripe_nr, stripe_len);
4962 
4963 	stripe_offset = stripe_nr * stripe_len;
4964 	BUG_ON(offset < stripe_offset);
4965 
4966 	/* stripe_offset is the offset of this block in its stripe*/
4967 	stripe_offset = offset - stripe_offset;
4968 
4969 	/* if we're here for raid56, we need to know the stripe aligned start */
4970 	if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4971 		unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4972 		raid56_full_stripe_start = offset;
4973 
4974 		/* allow a write of a full stripe, but make sure we don't
4975 		 * allow straddling of stripes
4976 		 */
4977 		do_div(raid56_full_stripe_start, full_stripe_len);
4978 		raid56_full_stripe_start *= full_stripe_len;
4979 	}
4980 
4981 	if (rw & REQ_DISCARD) {
4982 		/* we don't discard raid56 yet */
4983 		if (map->type &
4984 		    (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4985 			ret = -EOPNOTSUPP;
4986 			goto out;
4987 		}
4988 		*length = min_t(u64, em->len - offset, *length);
4989 	} else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4990 		u64 max_len;
4991 		/* For writes to RAID[56], allow a full stripeset across all disks.
4992 		   For other RAID types and for RAID[56] reads, just allow a single
4993 		   stripe (on a single disk). */
4994 		if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4995 		    (rw & REQ_WRITE)) {
4996 			max_len = stripe_len * nr_data_stripes(map) -
4997 				(offset - raid56_full_stripe_start);
4998 		} else {
4999 			/* we limit the length of each bio to what fits in a stripe */
5000 			max_len = stripe_len - stripe_offset;
5001 		}
5002 		*length = min_t(u64, em->len - offset, max_len);
5003 	} else {
5004 		*length = em->len - offset;
5005 	}
5006 
5007 	/* This is for when we're called from btrfs_merge_bio_hook() and all
5008 	   it cares about is the length */
5009 	if (!bbio_ret)
5010 		goto out;
5011 
5012 	btrfs_dev_replace_lock(dev_replace);
5013 	dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5014 	if (!dev_replace_is_ongoing)
5015 		btrfs_dev_replace_unlock(dev_replace);
5016 
5017 	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5018 	    !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5019 	    dev_replace->tgtdev != NULL) {
5020 		/*
5021 		 * in dev-replace case, for repair case (that's the only
5022 		 * case where the mirror is selected explicitly when
5023 		 * calling btrfs_map_block), blocks left of the left cursor
5024 		 * can also be read from the target drive.
5025 		 * For REQ_GET_READ_MIRRORS, the target drive is added as
5026 		 * the last one to the array of stripes. For READ, it also
5027 		 * needs to be supported using the same mirror number.
5028 		 * If the requested block is not left of the left cursor,
5029 		 * EIO is returned. This can happen because btrfs_num_copies()
5030 		 * returns one more in the dev-replace case.
5031 		 */
5032 		u64 tmp_length = *length;
5033 		struct btrfs_bio *tmp_bbio = NULL;
5034 		int tmp_num_stripes;
5035 		u64 srcdev_devid = dev_replace->srcdev->devid;
5036 		int index_srcdev = 0;
5037 		int found = 0;
5038 		u64 physical_of_found = 0;
5039 
5040 		ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5041 			     logical, &tmp_length, &tmp_bbio, 0, NULL);
5042 		if (ret) {
5043 			WARN_ON(tmp_bbio != NULL);
5044 			goto out;
5045 		}
5046 
5047 		tmp_num_stripes = tmp_bbio->num_stripes;
5048 		if (mirror_num > tmp_num_stripes) {
5049 			/*
5050 			 * REQ_GET_READ_MIRRORS does not contain this
5051 			 * mirror, that means that the requested area
5052 			 * is not left of the left cursor
5053 			 */
5054 			ret = -EIO;
5055 			kfree(tmp_bbio);
5056 			goto out;
5057 		}
5058 
5059 		/*
5060 		 * process the rest of the function using the mirror_num
5061 		 * of the source drive. Therefore look it up first.
5062 		 * At the end, patch the device pointer to the one of the
5063 		 * target drive.
5064 		 */
5065 		for (i = 0; i < tmp_num_stripes; i++) {
5066 			if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5067 				/*
5068 				 * In case of DUP, in order to keep it
5069 				 * simple, only add the mirror with the
5070 				 * lowest physical address
5071 				 */
5072 				if (found &&
5073 				    physical_of_found <=
5074 				     tmp_bbio->stripes[i].physical)
5075 					continue;
5076 				index_srcdev = i;
5077 				found = 1;
5078 				physical_of_found =
5079 					tmp_bbio->stripes[i].physical;
5080 			}
5081 		}
5082 
5083 		if (found) {
5084 			mirror_num = index_srcdev + 1;
5085 			patch_the_first_stripe_for_dev_replace = 1;
5086 			physical_to_patch_in_first_stripe = physical_of_found;
5087 		} else {
5088 			WARN_ON(1);
5089 			ret = -EIO;
5090 			kfree(tmp_bbio);
5091 			goto out;
5092 		}
5093 
5094 		kfree(tmp_bbio);
5095 	} else if (mirror_num > map->num_stripes) {
5096 		mirror_num = 0;
5097 	}
5098 
5099 	num_stripes = 1;
5100 	stripe_index = 0;
5101 	stripe_nr_orig = stripe_nr;
5102 	stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5103 	do_div(stripe_nr_end, map->stripe_len);
5104 	stripe_end_offset = stripe_nr_end * map->stripe_len -
5105 			    (offset + *length);
5106 
5107 	if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5108 		if (rw & REQ_DISCARD)
5109 			num_stripes = min_t(u64, map->num_stripes,
5110 					    stripe_nr_end - stripe_nr_orig);
5111 		stripe_index = do_div(stripe_nr, map->num_stripes);
5112 		if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5113 			mirror_num = 1;
5114 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5115 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5116 			num_stripes = map->num_stripes;
5117 		else if (mirror_num)
5118 			stripe_index = mirror_num - 1;
5119 		else {
5120 			stripe_index = find_live_mirror(fs_info, map, 0,
5121 					    map->num_stripes,
5122 					    current->pid % map->num_stripes,
5123 					    dev_replace_is_ongoing);
5124 			mirror_num = stripe_index + 1;
5125 		}
5126 
5127 	} else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5128 		if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5129 			num_stripes = map->num_stripes;
5130 		} else if (mirror_num) {
5131 			stripe_index = mirror_num - 1;
5132 		} else {
5133 			mirror_num = 1;
5134 		}
5135 
5136 	} else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5137 		int factor = map->num_stripes / map->sub_stripes;
5138 
5139 		stripe_index = do_div(stripe_nr, factor);
5140 		stripe_index *= map->sub_stripes;
5141 
5142 		if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5143 			num_stripes = map->sub_stripes;
5144 		else if (rw & REQ_DISCARD)
5145 			num_stripes = min_t(u64, map->sub_stripes *
5146 					    (stripe_nr_end - stripe_nr_orig),
5147 					    map->num_stripes);
5148 		else if (mirror_num)
5149 			stripe_index += mirror_num - 1;
5150 		else {
5151 			int old_stripe_index = stripe_index;
5152 			stripe_index = find_live_mirror(fs_info, map,
5153 					      stripe_index,
5154 					      map->sub_stripes, stripe_index +
5155 					      current->pid % map->sub_stripes,
5156 					      dev_replace_is_ongoing);
5157 			mirror_num = stripe_index - old_stripe_index + 1;
5158 		}
5159 
5160 	} else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5161 				BTRFS_BLOCK_GROUP_RAID6)) {
5162 		u64 tmp;
5163 
5164 		if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5165 		    && raid_map_ret) {
5166 			int i, rot;
5167 
5168 			/* push stripe_nr back to the start of the full stripe */
5169 			stripe_nr = raid56_full_stripe_start;
5170 			do_div(stripe_nr, stripe_len);
5171 
5172 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5173 
5174 			/* RAID[56] write or recovery. Return all stripes */
5175 			num_stripes = map->num_stripes;
5176 			max_errors = nr_parity_stripes(map);
5177 
5178 			raid_map = kmalloc_array(num_stripes, sizeof(u64),
5179 					   GFP_NOFS);
5180 			if (!raid_map) {
5181 				ret = -ENOMEM;
5182 				goto out;
5183 			}
5184 
5185 			/* Work out the disk rotation on this stripe-set */
5186 			tmp = stripe_nr;
5187 			rot = do_div(tmp, num_stripes);
5188 
5189 			/* Fill in the logical address of each stripe */
5190 			tmp = stripe_nr * nr_data_stripes(map);
5191 			for (i = 0; i < nr_data_stripes(map); i++)
5192 				raid_map[(i+rot) % num_stripes] =
5193 					em->start + (tmp + i) * map->stripe_len;
5194 
5195 			raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5196 			if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5197 				raid_map[(i+rot+1) % num_stripes] =
5198 					RAID6_Q_STRIPE;
5199 
5200 			*length = map->stripe_len;
5201 			stripe_index = 0;
5202 			stripe_offset = 0;
5203 		} else {
5204 			/*
5205 			 * Mirror #0 or #1 means the original data block.
5206 			 * Mirror #2 is RAID5 parity block.
5207 			 * Mirror #3 is RAID6 Q block.
5208 			 */
5209 			stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5210 			if (mirror_num > 1)
5211 				stripe_index = nr_data_stripes(map) +
5212 						mirror_num - 2;
5213 
5214 			/* We distribute the parity blocks across stripes */
5215 			tmp = stripe_nr + stripe_index;
5216 			stripe_index = do_div(tmp, map->num_stripes);
5217 			if (!(rw & (REQ_WRITE | REQ_DISCARD |
5218 				    REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5219 				mirror_num = 1;
5220 		}
5221 	} else {
5222 		/*
5223 		 * after this do_div call, stripe_nr is the number of stripes
5224 		 * on this device we have to walk to find the data, and
5225 		 * stripe_index is the number of our device in the stripe array
5226 		 */
5227 		stripe_index = do_div(stripe_nr, map->num_stripes);
5228 		mirror_num = stripe_index + 1;
5229 	}
5230 	BUG_ON(stripe_index >= map->num_stripes);
5231 
5232 	num_alloc_stripes = num_stripes;
5233 	if (dev_replace_is_ongoing) {
5234 		if (rw & (REQ_WRITE | REQ_DISCARD))
5235 			num_alloc_stripes <<= 1;
5236 		if (rw & REQ_GET_READ_MIRRORS)
5237 			num_alloc_stripes++;
5238 	}
5239 	bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5240 	if (!bbio) {
5241 		kfree(raid_map);
5242 		ret = -ENOMEM;
5243 		goto out;
5244 	}
5245 	atomic_set(&bbio->error, 0);
5246 
5247 	if (rw & REQ_DISCARD) {
5248 		int factor = 0;
5249 		int sub_stripes = 0;
5250 		u64 stripes_per_dev = 0;
5251 		u32 remaining_stripes = 0;
5252 		u32 last_stripe = 0;
5253 
5254 		if (map->type &
5255 		    (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5256 			if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5257 				sub_stripes = 1;
5258 			else
5259 				sub_stripes = map->sub_stripes;
5260 
5261 			factor = map->num_stripes / sub_stripes;
5262 			stripes_per_dev = div_u64_rem(stripe_nr_end -
5263 						      stripe_nr_orig,
5264 						      factor,
5265 						      &remaining_stripes);
5266 			div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5267 			last_stripe *= sub_stripes;
5268 		}
5269 
5270 		for (i = 0; i < num_stripes; i++) {
5271 			bbio->stripes[i].physical =
5272 				map->stripes[stripe_index].physical +
5273 				stripe_offset + stripe_nr * map->stripe_len;
5274 			bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5275 
5276 			if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5277 					 BTRFS_BLOCK_GROUP_RAID10)) {
5278 				bbio->stripes[i].length = stripes_per_dev *
5279 							  map->stripe_len;
5280 
5281 				if (i / sub_stripes < remaining_stripes)
5282 					bbio->stripes[i].length +=
5283 						map->stripe_len;
5284 
5285 				/*
5286 				 * Special for the first stripe and
5287 				 * the last stripe:
5288 				 *
5289 				 * |-------|...|-------|
5290 				 *     |----------|
5291 				 *    off     end_off
5292 				 */
5293 				if (i < sub_stripes)
5294 					bbio->stripes[i].length -=
5295 						stripe_offset;
5296 
5297 				if (stripe_index >= last_stripe &&
5298 				    stripe_index <= (last_stripe +
5299 						     sub_stripes - 1))
5300 					bbio->stripes[i].length -=
5301 						stripe_end_offset;
5302 
5303 				if (i == sub_stripes - 1)
5304 					stripe_offset = 0;
5305 			} else
5306 				bbio->stripes[i].length = *length;
5307 
5308 			stripe_index++;
5309 			if (stripe_index == map->num_stripes) {
5310 				/* This could only happen for RAID0/10 */
5311 				stripe_index = 0;
5312 				stripe_nr++;
5313 			}
5314 		}
5315 	} else {
5316 		for (i = 0; i < num_stripes; i++) {
5317 			bbio->stripes[i].physical =
5318 				map->stripes[stripe_index].physical +
5319 				stripe_offset +
5320 				stripe_nr * map->stripe_len;
5321 			bbio->stripes[i].dev =
5322 				map->stripes[stripe_index].dev;
5323 			stripe_index++;
5324 		}
5325 	}
5326 
5327 	if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5328 		max_errors = btrfs_chunk_max_errors(map);
5329 
5330 	if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5331 	    dev_replace->tgtdev != NULL) {
5332 		int index_where_to_add;
5333 		u64 srcdev_devid = dev_replace->srcdev->devid;
5334 
5335 		/*
5336 		 * duplicate the write operations while the dev replace
5337 		 * procedure is running. Since the copying of the old disk
5338 		 * to the new disk takes place at run time while the
5339 		 * filesystem is mounted writable, the regular write
5340 		 * operations to the old disk have to be duplicated to go
5341 		 * to the new disk as well.
5342 		 * Note that device->missing is handled by the caller, and
5343 		 * that the write to the old disk is already set up in the
5344 		 * stripes array.
5345 		 */
5346 		index_where_to_add = num_stripes;
5347 		for (i = 0; i < num_stripes; i++) {
5348 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5349 				/* write to new disk, too */
5350 				struct btrfs_bio_stripe *new =
5351 					bbio->stripes + index_where_to_add;
5352 				struct btrfs_bio_stripe *old =
5353 					bbio->stripes + i;
5354 
5355 				new->physical = old->physical;
5356 				new->length = old->length;
5357 				new->dev = dev_replace->tgtdev;
5358 				index_where_to_add++;
5359 				max_errors++;
5360 			}
5361 		}
5362 		num_stripes = index_where_to_add;
5363 	} else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5364 		   dev_replace->tgtdev != NULL) {
5365 		u64 srcdev_devid = dev_replace->srcdev->devid;
5366 		int index_srcdev = 0;
5367 		int found = 0;
5368 		u64 physical_of_found = 0;
5369 
5370 		/*
5371 		 * During the dev-replace procedure, the target drive can
5372 		 * also be used to read data in case it is needed to repair
5373 		 * a corrupt block elsewhere. This is possible if the
5374 		 * requested area is left of the left cursor. In this area,
5375 		 * the target drive is a full copy of the source drive.
5376 		 */
5377 		for (i = 0; i < num_stripes; i++) {
5378 			if (bbio->stripes[i].dev->devid == srcdev_devid) {
5379 				/*
5380 				 * In case of DUP, in order to keep it
5381 				 * simple, only add the mirror with the
5382 				 * lowest physical address
5383 				 */
5384 				if (found &&
5385 				    physical_of_found <=
5386 				     bbio->stripes[i].physical)
5387 					continue;
5388 				index_srcdev = i;
5389 				found = 1;
5390 				physical_of_found = bbio->stripes[i].physical;
5391 			}
5392 		}
5393 		if (found) {
5394 			u64 length = map->stripe_len;
5395 
5396 			if (physical_of_found + length <=
5397 			    dev_replace->cursor_left) {
5398 				struct btrfs_bio_stripe *tgtdev_stripe =
5399 					bbio->stripes + num_stripes;
5400 
5401 				tgtdev_stripe->physical = physical_of_found;
5402 				tgtdev_stripe->length =
5403 					bbio->stripes[index_srcdev].length;
5404 				tgtdev_stripe->dev = dev_replace->tgtdev;
5405 
5406 				num_stripes++;
5407 			}
5408 		}
5409 	}
5410 
5411 	*bbio_ret = bbio;
5412 	bbio->num_stripes = num_stripes;
5413 	bbio->max_errors = max_errors;
5414 	bbio->mirror_num = mirror_num;
5415 
5416 	/*
5417 	 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5418 	 * mirror_num == num_stripes + 1 && dev_replace target drive is
5419 	 * available as a mirror
5420 	 */
5421 	if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5422 		WARN_ON(num_stripes > 1);
5423 		bbio->stripes[0].dev = dev_replace->tgtdev;
5424 		bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5425 		bbio->mirror_num = map->num_stripes + 1;
5426 	}
5427 	if (raid_map) {
5428 		sort_parity_stripes(bbio, raid_map);
5429 		*raid_map_ret = raid_map;
5430 	}
5431 out:
5432 	if (dev_replace_is_ongoing)
5433 		btrfs_dev_replace_unlock(dev_replace);
5434 	free_extent_map(em);
5435 	return ret;
5436 }
5437 
5438 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5439 		      u64 logical, u64 *length,
5440 		      struct btrfs_bio **bbio_ret, int mirror_num)
5441 {
5442 	return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5443 				 mirror_num, NULL);
5444 }
5445 
5446 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5447 		     u64 chunk_start, u64 physical, u64 devid,
5448 		     u64 **logical, int *naddrs, int *stripe_len)
5449 {
5450 	struct extent_map_tree *em_tree = &map_tree->map_tree;
5451 	struct extent_map *em;
5452 	struct map_lookup *map;
5453 	u64 *buf;
5454 	u64 bytenr;
5455 	u64 length;
5456 	u64 stripe_nr;
5457 	u64 rmap_len;
5458 	int i, j, nr = 0;
5459 
5460 	read_lock(&em_tree->lock);
5461 	em = lookup_extent_mapping(em_tree, chunk_start, 1);
5462 	read_unlock(&em_tree->lock);
5463 
5464 	if (!em) {
5465 		printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5466 		       chunk_start);
5467 		return -EIO;
5468 	}
5469 
5470 	if (em->start != chunk_start) {
5471 		printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5472 		       em->start, chunk_start);
5473 		free_extent_map(em);
5474 		return -EIO;
5475 	}
5476 	map = (struct map_lookup *)em->bdev;
5477 
5478 	length = em->len;
5479 	rmap_len = map->stripe_len;
5480 
5481 	if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5482 		do_div(length, map->num_stripes / map->sub_stripes);
5483 	else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5484 		do_div(length, map->num_stripes);
5485 	else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5486 			      BTRFS_BLOCK_GROUP_RAID6)) {
5487 		do_div(length, nr_data_stripes(map));
5488 		rmap_len = map->stripe_len * nr_data_stripes(map);
5489 	}
5490 
5491 	buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5492 	BUG_ON(!buf); /* -ENOMEM */
5493 
5494 	for (i = 0; i < map->num_stripes; i++) {
5495 		if (devid && map->stripes[i].dev->devid != devid)
5496 			continue;
5497 		if (map->stripes[i].physical > physical ||
5498 		    map->stripes[i].physical + length <= physical)
5499 			continue;
5500 
5501 		stripe_nr = physical - map->stripes[i].physical;
5502 		do_div(stripe_nr, map->stripe_len);
5503 
5504 		if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5505 			stripe_nr = stripe_nr * map->num_stripes + i;
5506 			do_div(stripe_nr, map->sub_stripes);
5507 		} else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5508 			stripe_nr = stripe_nr * map->num_stripes + i;
5509 		} /* else if RAID[56], multiply by nr_data_stripes().
5510 		   * Alternatively, just use rmap_len below instead of
5511 		   * map->stripe_len */
5512 
5513 		bytenr = chunk_start + stripe_nr * rmap_len;
5514 		WARN_ON(nr >= map->num_stripes);
5515 		for (j = 0; j < nr; j++) {
5516 			if (buf[j] == bytenr)
5517 				break;
5518 		}
5519 		if (j == nr) {
5520 			WARN_ON(nr >= map->num_stripes);
5521 			buf[nr++] = bytenr;
5522 		}
5523 	}
5524 
5525 	*logical = buf;
5526 	*naddrs = nr;
5527 	*stripe_len = rmap_len;
5528 
5529 	free_extent_map(em);
5530 	return 0;
5531 }
5532 
5533 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5534 {
5535 	if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5536 		bio_endio_nodec(bio, err);
5537 	else
5538 		bio_endio(bio, err);
5539 	kfree(bbio);
5540 }
5541 
5542 static void btrfs_end_bio(struct bio *bio, int err)
5543 {
5544 	struct btrfs_bio *bbio = bio->bi_private;
5545 	struct btrfs_device *dev = bbio->stripes[0].dev;
5546 	int is_orig_bio = 0;
5547 
5548 	if (err) {
5549 		atomic_inc(&bbio->error);
5550 		if (err == -EIO || err == -EREMOTEIO) {
5551 			unsigned int stripe_index =
5552 				btrfs_io_bio(bio)->stripe_index;
5553 
5554 			BUG_ON(stripe_index >= bbio->num_stripes);
5555 			dev = bbio->stripes[stripe_index].dev;
5556 			if (dev->bdev) {
5557 				if (bio->bi_rw & WRITE)
5558 					btrfs_dev_stat_inc(dev,
5559 						BTRFS_DEV_STAT_WRITE_ERRS);
5560 				else
5561 					btrfs_dev_stat_inc(dev,
5562 						BTRFS_DEV_STAT_READ_ERRS);
5563 				if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5564 					btrfs_dev_stat_inc(dev,
5565 						BTRFS_DEV_STAT_FLUSH_ERRS);
5566 				btrfs_dev_stat_print_on_error(dev);
5567 			}
5568 		}
5569 	}
5570 
5571 	if (bio == bbio->orig_bio)
5572 		is_orig_bio = 1;
5573 
5574 	btrfs_bio_counter_dec(bbio->fs_info);
5575 
5576 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5577 		if (!is_orig_bio) {
5578 			bio_put(bio);
5579 			bio = bbio->orig_bio;
5580 		}
5581 
5582 		bio->bi_private = bbio->private;
5583 		bio->bi_end_io = bbio->end_io;
5584 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5585 		/* only send an error to the higher layers if it is
5586 		 * beyond the tolerance of the btrfs bio
5587 		 */
5588 		if (atomic_read(&bbio->error) > bbio->max_errors) {
5589 			err = -EIO;
5590 		} else {
5591 			/*
5592 			 * this bio is actually up to date, we didn't
5593 			 * go over the max number of errors
5594 			 */
5595 			set_bit(BIO_UPTODATE, &bio->bi_flags);
5596 			err = 0;
5597 		}
5598 
5599 		btrfs_end_bbio(bbio, bio, err);
5600 	} else if (!is_orig_bio) {
5601 		bio_put(bio);
5602 	}
5603 }
5604 
5605 /*
5606  * see run_scheduled_bios for a description of why bios are collected for
5607  * async submit.
5608  *
5609  * This will add one bio to the pending list for a device and make sure
5610  * the work struct is scheduled.
5611  */
5612 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5613 					struct btrfs_device *device,
5614 					int rw, struct bio *bio)
5615 {
5616 	int should_queue = 1;
5617 	struct btrfs_pending_bios *pending_bios;
5618 
5619 	if (device->missing || !device->bdev) {
5620 		bio_endio(bio, -EIO);
5621 		return;
5622 	}
5623 
5624 	/* don't bother with additional async steps for reads, right now */
5625 	if (!(rw & REQ_WRITE)) {
5626 		bio_get(bio);
5627 		btrfsic_submit_bio(rw, bio);
5628 		bio_put(bio);
5629 		return;
5630 	}
5631 
5632 	/*
5633 	 * nr_async_bios allows us to reliably return congestion to the
5634 	 * higher layers.  Otherwise, the async bio makes it appear we have
5635 	 * made progress against dirty pages when we've really just put it
5636 	 * on a queue for later
5637 	 */
5638 	atomic_inc(&root->fs_info->nr_async_bios);
5639 	WARN_ON(bio->bi_next);
5640 	bio->bi_next = NULL;
5641 	bio->bi_rw |= rw;
5642 
5643 	spin_lock(&device->io_lock);
5644 	if (bio->bi_rw & REQ_SYNC)
5645 		pending_bios = &device->pending_sync_bios;
5646 	else
5647 		pending_bios = &device->pending_bios;
5648 
5649 	if (pending_bios->tail)
5650 		pending_bios->tail->bi_next = bio;
5651 
5652 	pending_bios->tail = bio;
5653 	if (!pending_bios->head)
5654 		pending_bios->head = bio;
5655 	if (device->running_pending)
5656 		should_queue = 0;
5657 
5658 	spin_unlock(&device->io_lock);
5659 
5660 	if (should_queue)
5661 		btrfs_queue_work(root->fs_info->submit_workers,
5662 				 &device->work);
5663 }
5664 
5665 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5666 		       sector_t sector)
5667 {
5668 	struct bio_vec *prev;
5669 	struct request_queue *q = bdev_get_queue(bdev);
5670 	unsigned int max_sectors = queue_max_sectors(q);
5671 	struct bvec_merge_data bvm = {
5672 		.bi_bdev = bdev,
5673 		.bi_sector = sector,
5674 		.bi_rw = bio->bi_rw,
5675 	};
5676 
5677 	if (WARN_ON(bio->bi_vcnt == 0))
5678 		return 1;
5679 
5680 	prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5681 	if (bio_sectors(bio) > max_sectors)
5682 		return 0;
5683 
5684 	if (!q->merge_bvec_fn)
5685 		return 1;
5686 
5687 	bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5688 	if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5689 		return 0;
5690 	return 1;
5691 }
5692 
5693 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5694 			      struct bio *bio, u64 physical, int dev_nr,
5695 			      int rw, int async)
5696 {
5697 	struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5698 
5699 	bio->bi_private = bbio;
5700 	btrfs_io_bio(bio)->stripe_index = dev_nr;
5701 	bio->bi_end_io = btrfs_end_bio;
5702 	bio->bi_iter.bi_sector = physical >> 9;
5703 #ifdef DEBUG
5704 	{
5705 		struct rcu_string *name;
5706 
5707 		rcu_read_lock();
5708 		name = rcu_dereference(dev->name);
5709 		pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5710 			 "(%s id %llu), size=%u\n", rw,
5711 			 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5712 			 name->str, dev->devid, bio->bi_iter.bi_size);
5713 		rcu_read_unlock();
5714 	}
5715 #endif
5716 	bio->bi_bdev = dev->bdev;
5717 
5718 	btrfs_bio_counter_inc_noblocked(root->fs_info);
5719 
5720 	if (async)
5721 		btrfs_schedule_bio(root, dev, rw, bio);
5722 	else
5723 		btrfsic_submit_bio(rw, bio);
5724 }
5725 
5726 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5727 			      struct bio *first_bio, struct btrfs_device *dev,
5728 			      int dev_nr, int rw, int async)
5729 {
5730 	struct bio_vec *bvec = first_bio->bi_io_vec;
5731 	struct bio *bio;
5732 	int nr_vecs = bio_get_nr_vecs(dev->bdev);
5733 	u64 physical = bbio->stripes[dev_nr].physical;
5734 
5735 again:
5736 	bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5737 	if (!bio)
5738 		return -ENOMEM;
5739 
5740 	while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5741 		if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5742 				 bvec->bv_offset) < bvec->bv_len) {
5743 			u64 len = bio->bi_iter.bi_size;
5744 
5745 			atomic_inc(&bbio->stripes_pending);
5746 			submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5747 					  rw, async);
5748 			physical += len;
5749 			goto again;
5750 		}
5751 		bvec++;
5752 	}
5753 
5754 	submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5755 	return 0;
5756 }
5757 
5758 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5759 {
5760 	atomic_inc(&bbio->error);
5761 	if (atomic_dec_and_test(&bbio->stripes_pending)) {
5762 		/* Shoud be the original bio. */
5763 		WARN_ON(bio != bbio->orig_bio);
5764 
5765 		bio->bi_private = bbio->private;
5766 		bio->bi_end_io = bbio->end_io;
5767 		btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5768 		bio->bi_iter.bi_sector = logical >> 9;
5769 
5770 		btrfs_end_bbio(bbio, bio, -EIO);
5771 	}
5772 }
5773 
5774 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5775 		  int mirror_num, int async_submit)
5776 {
5777 	struct btrfs_device *dev;
5778 	struct bio *first_bio = bio;
5779 	u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5780 	u64 length = 0;
5781 	u64 map_length;
5782 	u64 *raid_map = NULL;
5783 	int ret;
5784 	int dev_nr = 0;
5785 	int total_devs = 1;
5786 	struct btrfs_bio *bbio = NULL;
5787 
5788 	length = bio->bi_iter.bi_size;
5789 	map_length = length;
5790 
5791 	btrfs_bio_counter_inc_blocked(root->fs_info);
5792 	ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5793 			      mirror_num, &raid_map);
5794 	if (ret) {
5795 		btrfs_bio_counter_dec(root->fs_info);
5796 		return ret;
5797 	}
5798 
5799 	total_devs = bbio->num_stripes;
5800 	bbio->orig_bio = first_bio;
5801 	bbio->private = first_bio->bi_private;
5802 	bbio->end_io = first_bio->bi_end_io;
5803 	bbio->fs_info = root->fs_info;
5804 	atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5805 
5806 	if (raid_map) {
5807 		/* In this case, map_length has been set to the length of
5808 		   a single stripe; not the whole write */
5809 		if (rw & WRITE) {
5810 			ret = raid56_parity_write(root, bio, bbio,
5811 						  raid_map, map_length);
5812 		} else {
5813 			ret = raid56_parity_recover(root, bio, bbio,
5814 						    raid_map, map_length,
5815 						    mirror_num);
5816 		}
5817 		/*
5818 		 * FIXME, replace dosen't support raid56 yet, please fix
5819 		 * it in the future.
5820 		 */
5821 		btrfs_bio_counter_dec(root->fs_info);
5822 		return ret;
5823 	}
5824 
5825 	if (map_length < length) {
5826 		btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5827 			logical, length, map_length);
5828 		BUG();
5829 	}
5830 
5831 	while (dev_nr < total_devs) {
5832 		dev = bbio->stripes[dev_nr].dev;
5833 		if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5834 			bbio_error(bbio, first_bio, logical);
5835 			dev_nr++;
5836 			continue;
5837 		}
5838 
5839 		/*
5840 		 * Check and see if we're ok with this bio based on it's size
5841 		 * and offset with the given device.
5842 		 */
5843 		if (!bio_size_ok(dev->bdev, first_bio,
5844 				 bbio->stripes[dev_nr].physical >> 9)) {
5845 			ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5846 						 dev_nr, rw, async_submit);
5847 			BUG_ON(ret);
5848 			dev_nr++;
5849 			continue;
5850 		}
5851 
5852 		if (dev_nr < total_devs - 1) {
5853 			bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5854 			BUG_ON(!bio); /* -ENOMEM */
5855 		} else {
5856 			bio = first_bio;
5857 			bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5858 		}
5859 
5860 		submit_stripe_bio(root, bbio, bio,
5861 				  bbio->stripes[dev_nr].physical, dev_nr, rw,
5862 				  async_submit);
5863 		dev_nr++;
5864 	}
5865 	btrfs_bio_counter_dec(root->fs_info);
5866 	return 0;
5867 }
5868 
5869 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5870 				       u8 *uuid, u8 *fsid)
5871 {
5872 	struct btrfs_device *device;
5873 	struct btrfs_fs_devices *cur_devices;
5874 
5875 	cur_devices = fs_info->fs_devices;
5876 	while (cur_devices) {
5877 		if (!fsid ||
5878 		    !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5879 			device = __find_device(&cur_devices->devices,
5880 					       devid, uuid);
5881 			if (device)
5882 				return device;
5883 		}
5884 		cur_devices = cur_devices->seed;
5885 	}
5886 	return NULL;
5887 }
5888 
5889 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5890 					    struct btrfs_fs_devices *fs_devices,
5891 					    u64 devid, u8 *dev_uuid)
5892 {
5893 	struct btrfs_device *device;
5894 
5895 	device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5896 	if (IS_ERR(device))
5897 		return NULL;
5898 
5899 	list_add(&device->dev_list, &fs_devices->devices);
5900 	device->fs_devices = fs_devices;
5901 	fs_devices->num_devices++;
5902 
5903 	device->missing = 1;
5904 	fs_devices->missing_devices++;
5905 
5906 	return device;
5907 }
5908 
5909 /**
5910  * btrfs_alloc_device - allocate struct btrfs_device
5911  * @fs_info:	used only for generating a new devid, can be NULL if
5912  *		devid is provided (i.e. @devid != NULL).
5913  * @devid:	a pointer to devid for this device.  If NULL a new devid
5914  *		is generated.
5915  * @uuid:	a pointer to UUID for this device.  If NULL a new UUID
5916  *		is generated.
5917  *
5918  * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5919  * on error.  Returned struct is not linked onto any lists and can be
5920  * destroyed with kfree() right away.
5921  */
5922 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5923 					const u64 *devid,
5924 					const u8 *uuid)
5925 {
5926 	struct btrfs_device *dev;
5927 	u64 tmp;
5928 
5929 	if (WARN_ON(!devid && !fs_info))
5930 		return ERR_PTR(-EINVAL);
5931 
5932 	dev = __alloc_device();
5933 	if (IS_ERR(dev))
5934 		return dev;
5935 
5936 	if (devid)
5937 		tmp = *devid;
5938 	else {
5939 		int ret;
5940 
5941 		ret = find_next_devid(fs_info, &tmp);
5942 		if (ret) {
5943 			kfree(dev);
5944 			return ERR_PTR(ret);
5945 		}
5946 	}
5947 	dev->devid = tmp;
5948 
5949 	if (uuid)
5950 		memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5951 	else
5952 		generate_random_uuid(dev->uuid);
5953 
5954 	btrfs_init_work(&dev->work, btrfs_submit_helper,
5955 			pending_bios_fn, NULL, NULL);
5956 
5957 	return dev;
5958 }
5959 
5960 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5961 			  struct extent_buffer *leaf,
5962 			  struct btrfs_chunk *chunk)
5963 {
5964 	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5965 	struct map_lookup *map;
5966 	struct extent_map *em;
5967 	u64 logical;
5968 	u64 length;
5969 	u64 devid;
5970 	u8 uuid[BTRFS_UUID_SIZE];
5971 	int num_stripes;
5972 	int ret;
5973 	int i;
5974 
5975 	logical = key->offset;
5976 	length = btrfs_chunk_length(leaf, chunk);
5977 
5978 	read_lock(&map_tree->map_tree.lock);
5979 	em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5980 	read_unlock(&map_tree->map_tree.lock);
5981 
5982 	/* already mapped? */
5983 	if (em && em->start <= logical && em->start + em->len > logical) {
5984 		free_extent_map(em);
5985 		return 0;
5986 	} else if (em) {
5987 		free_extent_map(em);
5988 	}
5989 
5990 	em = alloc_extent_map();
5991 	if (!em)
5992 		return -ENOMEM;
5993 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5994 	map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5995 	if (!map) {
5996 		free_extent_map(em);
5997 		return -ENOMEM;
5998 	}
5999 
6000 	set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6001 	em->bdev = (struct block_device *)map;
6002 	em->start = logical;
6003 	em->len = length;
6004 	em->orig_start = 0;
6005 	em->block_start = 0;
6006 	em->block_len = em->len;
6007 
6008 	map->num_stripes = num_stripes;
6009 	map->io_width = btrfs_chunk_io_width(leaf, chunk);
6010 	map->io_align = btrfs_chunk_io_align(leaf, chunk);
6011 	map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6012 	map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6013 	map->type = btrfs_chunk_type(leaf, chunk);
6014 	map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6015 	for (i = 0; i < num_stripes; i++) {
6016 		map->stripes[i].physical =
6017 			btrfs_stripe_offset_nr(leaf, chunk, i);
6018 		devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6019 		read_extent_buffer(leaf, uuid, (unsigned long)
6020 				   btrfs_stripe_dev_uuid_nr(chunk, i),
6021 				   BTRFS_UUID_SIZE);
6022 		map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6023 							uuid, NULL);
6024 		if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6025 			free_extent_map(em);
6026 			return -EIO;
6027 		}
6028 		if (!map->stripes[i].dev) {
6029 			map->stripes[i].dev =
6030 				add_missing_dev(root, root->fs_info->fs_devices,
6031 						devid, uuid);
6032 			if (!map->stripes[i].dev) {
6033 				free_extent_map(em);
6034 				return -EIO;
6035 			}
6036 		}
6037 		map->stripes[i].dev->in_fs_metadata = 1;
6038 	}
6039 
6040 	write_lock(&map_tree->map_tree.lock);
6041 	ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6042 	write_unlock(&map_tree->map_tree.lock);
6043 	BUG_ON(ret); /* Tree corruption */
6044 	free_extent_map(em);
6045 
6046 	return 0;
6047 }
6048 
6049 static void fill_device_from_item(struct extent_buffer *leaf,
6050 				 struct btrfs_dev_item *dev_item,
6051 				 struct btrfs_device *device)
6052 {
6053 	unsigned long ptr;
6054 
6055 	device->devid = btrfs_device_id(leaf, dev_item);
6056 	device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6057 	device->total_bytes = device->disk_total_bytes;
6058 	device->commit_total_bytes = device->disk_total_bytes;
6059 	device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6060 	device->commit_bytes_used = device->bytes_used;
6061 	device->type = btrfs_device_type(leaf, dev_item);
6062 	device->io_align = btrfs_device_io_align(leaf, dev_item);
6063 	device->io_width = btrfs_device_io_width(leaf, dev_item);
6064 	device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6065 	WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6066 	device->is_tgtdev_for_dev_replace = 0;
6067 
6068 	ptr = btrfs_device_uuid(dev_item);
6069 	read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6070 }
6071 
6072 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6073 						  u8 *fsid)
6074 {
6075 	struct btrfs_fs_devices *fs_devices;
6076 	int ret;
6077 
6078 	BUG_ON(!mutex_is_locked(&uuid_mutex));
6079 
6080 	fs_devices = root->fs_info->fs_devices->seed;
6081 	while (fs_devices) {
6082 		if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6083 			return fs_devices;
6084 
6085 		fs_devices = fs_devices->seed;
6086 	}
6087 
6088 	fs_devices = find_fsid(fsid);
6089 	if (!fs_devices) {
6090 		if (!btrfs_test_opt(root, DEGRADED))
6091 			return ERR_PTR(-ENOENT);
6092 
6093 		fs_devices = alloc_fs_devices(fsid);
6094 		if (IS_ERR(fs_devices))
6095 			return fs_devices;
6096 
6097 		fs_devices->seeding = 1;
6098 		fs_devices->opened = 1;
6099 		return fs_devices;
6100 	}
6101 
6102 	fs_devices = clone_fs_devices(fs_devices);
6103 	if (IS_ERR(fs_devices))
6104 		return fs_devices;
6105 
6106 	ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6107 				   root->fs_info->bdev_holder);
6108 	if (ret) {
6109 		free_fs_devices(fs_devices);
6110 		fs_devices = ERR_PTR(ret);
6111 		goto out;
6112 	}
6113 
6114 	if (!fs_devices->seeding) {
6115 		__btrfs_close_devices(fs_devices);
6116 		free_fs_devices(fs_devices);
6117 		fs_devices = ERR_PTR(-EINVAL);
6118 		goto out;
6119 	}
6120 
6121 	fs_devices->seed = root->fs_info->fs_devices->seed;
6122 	root->fs_info->fs_devices->seed = fs_devices;
6123 out:
6124 	return fs_devices;
6125 }
6126 
6127 static int read_one_dev(struct btrfs_root *root,
6128 			struct extent_buffer *leaf,
6129 			struct btrfs_dev_item *dev_item)
6130 {
6131 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6132 	struct btrfs_device *device;
6133 	u64 devid;
6134 	int ret;
6135 	u8 fs_uuid[BTRFS_UUID_SIZE];
6136 	u8 dev_uuid[BTRFS_UUID_SIZE];
6137 
6138 	devid = btrfs_device_id(leaf, dev_item);
6139 	read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6140 			   BTRFS_UUID_SIZE);
6141 	read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6142 			   BTRFS_UUID_SIZE);
6143 
6144 	if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6145 		fs_devices = open_seed_devices(root, fs_uuid);
6146 		if (IS_ERR(fs_devices))
6147 			return PTR_ERR(fs_devices);
6148 	}
6149 
6150 	device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6151 	if (!device) {
6152 		if (!btrfs_test_opt(root, DEGRADED))
6153 			return -EIO;
6154 
6155 		btrfs_warn(root->fs_info, "devid %llu missing", devid);
6156 		device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6157 		if (!device)
6158 			return -ENOMEM;
6159 	} else {
6160 		if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6161 			return -EIO;
6162 
6163 		if(!device->bdev && !device->missing) {
6164 			/*
6165 			 * this happens when a device that was properly setup
6166 			 * in the device info lists suddenly goes bad.
6167 			 * device->bdev is NULL, and so we have to set
6168 			 * device->missing to one here
6169 			 */
6170 			device->fs_devices->missing_devices++;
6171 			device->missing = 1;
6172 		}
6173 
6174 		/* Move the device to its own fs_devices */
6175 		if (device->fs_devices != fs_devices) {
6176 			ASSERT(device->missing);
6177 
6178 			list_move(&device->dev_list, &fs_devices->devices);
6179 			device->fs_devices->num_devices--;
6180 			fs_devices->num_devices++;
6181 
6182 			device->fs_devices->missing_devices--;
6183 			fs_devices->missing_devices++;
6184 
6185 			device->fs_devices = fs_devices;
6186 		}
6187 	}
6188 
6189 	if (device->fs_devices != root->fs_info->fs_devices) {
6190 		BUG_ON(device->writeable);
6191 		if (device->generation !=
6192 		    btrfs_device_generation(leaf, dev_item))
6193 			return -EINVAL;
6194 	}
6195 
6196 	fill_device_from_item(leaf, dev_item, device);
6197 	device->in_fs_metadata = 1;
6198 	if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6199 		device->fs_devices->total_rw_bytes += device->total_bytes;
6200 		spin_lock(&root->fs_info->free_chunk_lock);
6201 		root->fs_info->free_chunk_space += device->total_bytes -
6202 			device->bytes_used;
6203 		spin_unlock(&root->fs_info->free_chunk_lock);
6204 	}
6205 	ret = 0;
6206 	return ret;
6207 }
6208 
6209 int btrfs_read_sys_array(struct btrfs_root *root)
6210 {
6211 	struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6212 	struct extent_buffer *sb;
6213 	struct btrfs_disk_key *disk_key;
6214 	struct btrfs_chunk *chunk;
6215 	u8 *ptr;
6216 	unsigned long sb_ptr;
6217 	int ret = 0;
6218 	u32 num_stripes;
6219 	u32 array_size;
6220 	u32 len = 0;
6221 	u32 cur;
6222 	struct btrfs_key key;
6223 
6224 	sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6225 					  BTRFS_SUPER_INFO_SIZE);
6226 	if (!sb)
6227 		return -ENOMEM;
6228 	btrfs_set_buffer_uptodate(sb);
6229 	btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6230 	/*
6231 	 * The sb extent buffer is artifical and just used to read the system array.
6232 	 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6233 	 * pages up-to-date when the page is larger: extent does not cover the
6234 	 * whole page and consequently check_page_uptodate does not find all
6235 	 * the page's extents up-to-date (the hole beyond sb),
6236 	 * write_extent_buffer then triggers a WARN_ON.
6237 	 *
6238 	 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6239 	 * but sb spans only this function. Add an explicit SetPageUptodate call
6240 	 * to silence the warning eg. on PowerPC 64.
6241 	 */
6242 	if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6243 		SetPageUptodate(sb->pages[0]);
6244 
6245 	write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6246 	array_size = btrfs_super_sys_array_size(super_copy);
6247 
6248 	ptr = super_copy->sys_chunk_array;
6249 	sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6250 	cur = 0;
6251 
6252 	while (cur < array_size) {
6253 		disk_key = (struct btrfs_disk_key *)ptr;
6254 		btrfs_disk_key_to_cpu(&key, disk_key);
6255 
6256 		len = sizeof(*disk_key); ptr += len;
6257 		sb_ptr += len;
6258 		cur += len;
6259 
6260 		if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6261 			chunk = (struct btrfs_chunk *)sb_ptr;
6262 			ret = read_one_chunk(root, &key, sb, chunk);
6263 			if (ret)
6264 				break;
6265 			num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6266 			len = btrfs_chunk_item_size(num_stripes);
6267 		} else {
6268 			ret = -EIO;
6269 			break;
6270 		}
6271 		ptr += len;
6272 		sb_ptr += len;
6273 		cur += len;
6274 	}
6275 	free_extent_buffer(sb);
6276 	return ret;
6277 }
6278 
6279 int btrfs_read_chunk_tree(struct btrfs_root *root)
6280 {
6281 	struct btrfs_path *path;
6282 	struct extent_buffer *leaf;
6283 	struct btrfs_key key;
6284 	struct btrfs_key found_key;
6285 	int ret;
6286 	int slot;
6287 
6288 	root = root->fs_info->chunk_root;
6289 
6290 	path = btrfs_alloc_path();
6291 	if (!path)
6292 		return -ENOMEM;
6293 
6294 	mutex_lock(&uuid_mutex);
6295 	lock_chunks(root);
6296 
6297 	/*
6298 	 * Read all device items, and then all the chunk items. All
6299 	 * device items are found before any chunk item (their object id
6300 	 * is smaller than the lowest possible object id for a chunk
6301 	 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6302 	 */
6303 	key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6304 	key.offset = 0;
6305 	key.type = 0;
6306 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6307 	if (ret < 0)
6308 		goto error;
6309 	while (1) {
6310 		leaf = path->nodes[0];
6311 		slot = path->slots[0];
6312 		if (slot >= btrfs_header_nritems(leaf)) {
6313 			ret = btrfs_next_leaf(root, path);
6314 			if (ret == 0)
6315 				continue;
6316 			if (ret < 0)
6317 				goto error;
6318 			break;
6319 		}
6320 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
6321 		if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6322 			struct btrfs_dev_item *dev_item;
6323 			dev_item = btrfs_item_ptr(leaf, slot,
6324 						  struct btrfs_dev_item);
6325 			ret = read_one_dev(root, leaf, dev_item);
6326 			if (ret)
6327 				goto error;
6328 		} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6329 			struct btrfs_chunk *chunk;
6330 			chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6331 			ret = read_one_chunk(root, &found_key, leaf, chunk);
6332 			if (ret)
6333 				goto error;
6334 		}
6335 		path->slots[0]++;
6336 	}
6337 	ret = 0;
6338 error:
6339 	unlock_chunks(root);
6340 	mutex_unlock(&uuid_mutex);
6341 
6342 	btrfs_free_path(path);
6343 	return ret;
6344 }
6345 
6346 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6347 {
6348 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6349 	struct btrfs_device *device;
6350 
6351 	while (fs_devices) {
6352 		mutex_lock(&fs_devices->device_list_mutex);
6353 		list_for_each_entry(device, &fs_devices->devices, dev_list)
6354 			device->dev_root = fs_info->dev_root;
6355 		mutex_unlock(&fs_devices->device_list_mutex);
6356 
6357 		fs_devices = fs_devices->seed;
6358 	}
6359 }
6360 
6361 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6362 {
6363 	int i;
6364 
6365 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6366 		btrfs_dev_stat_reset(dev, i);
6367 }
6368 
6369 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6370 {
6371 	struct btrfs_key key;
6372 	struct btrfs_key found_key;
6373 	struct btrfs_root *dev_root = fs_info->dev_root;
6374 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6375 	struct extent_buffer *eb;
6376 	int slot;
6377 	int ret = 0;
6378 	struct btrfs_device *device;
6379 	struct btrfs_path *path = NULL;
6380 	int i;
6381 
6382 	path = btrfs_alloc_path();
6383 	if (!path) {
6384 		ret = -ENOMEM;
6385 		goto out;
6386 	}
6387 
6388 	mutex_lock(&fs_devices->device_list_mutex);
6389 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6390 		int item_size;
6391 		struct btrfs_dev_stats_item *ptr;
6392 
6393 		key.objectid = 0;
6394 		key.type = BTRFS_DEV_STATS_KEY;
6395 		key.offset = device->devid;
6396 		ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6397 		if (ret) {
6398 			__btrfs_reset_dev_stats(device);
6399 			device->dev_stats_valid = 1;
6400 			btrfs_release_path(path);
6401 			continue;
6402 		}
6403 		slot = path->slots[0];
6404 		eb = path->nodes[0];
6405 		btrfs_item_key_to_cpu(eb, &found_key, slot);
6406 		item_size = btrfs_item_size_nr(eb, slot);
6407 
6408 		ptr = btrfs_item_ptr(eb, slot,
6409 				     struct btrfs_dev_stats_item);
6410 
6411 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6412 			if (item_size >= (1 + i) * sizeof(__le64))
6413 				btrfs_dev_stat_set(device, i,
6414 					btrfs_dev_stats_value(eb, ptr, i));
6415 			else
6416 				btrfs_dev_stat_reset(device, i);
6417 		}
6418 
6419 		device->dev_stats_valid = 1;
6420 		btrfs_dev_stat_print_on_load(device);
6421 		btrfs_release_path(path);
6422 	}
6423 	mutex_unlock(&fs_devices->device_list_mutex);
6424 
6425 out:
6426 	btrfs_free_path(path);
6427 	return ret < 0 ? ret : 0;
6428 }
6429 
6430 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6431 				struct btrfs_root *dev_root,
6432 				struct btrfs_device *device)
6433 {
6434 	struct btrfs_path *path;
6435 	struct btrfs_key key;
6436 	struct extent_buffer *eb;
6437 	struct btrfs_dev_stats_item *ptr;
6438 	int ret;
6439 	int i;
6440 
6441 	key.objectid = 0;
6442 	key.type = BTRFS_DEV_STATS_KEY;
6443 	key.offset = device->devid;
6444 
6445 	path = btrfs_alloc_path();
6446 	BUG_ON(!path);
6447 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6448 	if (ret < 0) {
6449 		printk_in_rcu(KERN_WARNING "BTRFS: "
6450 			"error %d while searching for dev_stats item for device %s!\n",
6451 			      ret, rcu_str_deref(device->name));
6452 		goto out;
6453 	}
6454 
6455 	if (ret == 0 &&
6456 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6457 		/* need to delete old one and insert a new one */
6458 		ret = btrfs_del_item(trans, dev_root, path);
6459 		if (ret != 0) {
6460 			printk_in_rcu(KERN_WARNING "BTRFS: "
6461 				"delete too small dev_stats item for device %s failed %d!\n",
6462 				      rcu_str_deref(device->name), ret);
6463 			goto out;
6464 		}
6465 		ret = 1;
6466 	}
6467 
6468 	if (ret == 1) {
6469 		/* need to insert a new item */
6470 		btrfs_release_path(path);
6471 		ret = btrfs_insert_empty_item(trans, dev_root, path,
6472 					      &key, sizeof(*ptr));
6473 		if (ret < 0) {
6474 			printk_in_rcu(KERN_WARNING "BTRFS: "
6475 					  "insert dev_stats item for device %s failed %d!\n",
6476 				      rcu_str_deref(device->name), ret);
6477 			goto out;
6478 		}
6479 	}
6480 
6481 	eb = path->nodes[0];
6482 	ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6483 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6484 		btrfs_set_dev_stats_value(eb, ptr, i,
6485 					  btrfs_dev_stat_read(device, i));
6486 	btrfs_mark_buffer_dirty(eb);
6487 
6488 out:
6489 	btrfs_free_path(path);
6490 	return ret;
6491 }
6492 
6493 /*
6494  * called from commit_transaction. Writes all changed device stats to disk.
6495  */
6496 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6497 			struct btrfs_fs_info *fs_info)
6498 {
6499 	struct btrfs_root *dev_root = fs_info->dev_root;
6500 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6501 	struct btrfs_device *device;
6502 	int stats_cnt;
6503 	int ret = 0;
6504 
6505 	mutex_lock(&fs_devices->device_list_mutex);
6506 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
6507 		if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6508 			continue;
6509 
6510 		stats_cnt = atomic_read(&device->dev_stats_ccnt);
6511 		ret = update_dev_stat_item(trans, dev_root, device);
6512 		if (!ret)
6513 			atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6514 	}
6515 	mutex_unlock(&fs_devices->device_list_mutex);
6516 
6517 	return ret;
6518 }
6519 
6520 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6521 {
6522 	btrfs_dev_stat_inc(dev, index);
6523 	btrfs_dev_stat_print_on_error(dev);
6524 }
6525 
6526 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6527 {
6528 	if (!dev->dev_stats_valid)
6529 		return;
6530 	printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6531 			   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6532 			   rcu_str_deref(dev->name),
6533 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6534 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6535 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6536 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6537 			   btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6538 }
6539 
6540 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6541 {
6542 	int i;
6543 
6544 	for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6545 		if (btrfs_dev_stat_read(dev, i) != 0)
6546 			break;
6547 	if (i == BTRFS_DEV_STAT_VALUES_MAX)
6548 		return; /* all values == 0, suppress message */
6549 
6550 	printk_in_rcu(KERN_INFO "BTRFS: "
6551 		   "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6552 	       rcu_str_deref(dev->name),
6553 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6554 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6555 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6556 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6557 	       btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6558 }
6559 
6560 int btrfs_get_dev_stats(struct btrfs_root *root,
6561 			struct btrfs_ioctl_get_dev_stats *stats)
6562 {
6563 	struct btrfs_device *dev;
6564 	struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6565 	int i;
6566 
6567 	mutex_lock(&fs_devices->device_list_mutex);
6568 	dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6569 	mutex_unlock(&fs_devices->device_list_mutex);
6570 
6571 	if (!dev) {
6572 		btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6573 		return -ENODEV;
6574 	} else if (!dev->dev_stats_valid) {
6575 		btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6576 		return -ENODEV;
6577 	} else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6578 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6579 			if (stats->nr_items > i)
6580 				stats->values[i] =
6581 					btrfs_dev_stat_read_and_reset(dev, i);
6582 			else
6583 				btrfs_dev_stat_reset(dev, i);
6584 		}
6585 	} else {
6586 		for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6587 			if (stats->nr_items > i)
6588 				stats->values[i] = btrfs_dev_stat_read(dev, i);
6589 	}
6590 	if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6591 		stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6592 	return 0;
6593 }
6594 
6595 int btrfs_scratch_superblock(struct btrfs_device *device)
6596 {
6597 	struct buffer_head *bh;
6598 	struct btrfs_super_block *disk_super;
6599 
6600 	bh = btrfs_read_dev_super(device->bdev);
6601 	if (!bh)
6602 		return -EINVAL;
6603 	disk_super = (struct btrfs_super_block *)bh->b_data;
6604 
6605 	memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6606 	set_buffer_dirty(bh);
6607 	sync_dirty_buffer(bh);
6608 	brelse(bh);
6609 
6610 	return 0;
6611 }
6612 
6613 /*
6614  * Update the size of all devices, which is used for writing out the
6615  * super blocks.
6616  */
6617 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6618 {
6619 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6620 	struct btrfs_device *curr, *next;
6621 
6622 	if (list_empty(&fs_devices->resized_devices))
6623 		return;
6624 
6625 	mutex_lock(&fs_devices->device_list_mutex);
6626 	lock_chunks(fs_info->dev_root);
6627 	list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6628 				 resized_list) {
6629 		list_del_init(&curr->resized_list);
6630 		curr->commit_total_bytes = curr->disk_total_bytes;
6631 	}
6632 	unlock_chunks(fs_info->dev_root);
6633 	mutex_unlock(&fs_devices->device_list_mutex);
6634 }
6635 
6636 /* Must be invoked during the transaction commit */
6637 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6638 					struct btrfs_transaction *transaction)
6639 {
6640 	struct extent_map *em;
6641 	struct map_lookup *map;
6642 	struct btrfs_device *dev;
6643 	int i;
6644 
6645 	if (list_empty(&transaction->pending_chunks))
6646 		return;
6647 
6648 	/* In order to kick the device replace finish process */
6649 	lock_chunks(root);
6650 	list_for_each_entry(em, &transaction->pending_chunks, list) {
6651 		map = (struct map_lookup *)em->bdev;
6652 
6653 		for (i = 0; i < map->num_stripes; i++) {
6654 			dev = map->stripes[i].dev;
6655 			dev->commit_bytes_used = dev->bytes_used;
6656 		}
6657 	}
6658 	unlock_chunks(root);
6659 }
6660